PROBLEMS IN REAL ANALYSIS Second Edition A Workbook with Solutions CHARALAMBOS D. ALIPRANTIS Departments of Economics and Mathematics Purdue University and OWEN BURKINSHAW Departments of Mathematical Sciences Indiana University, Purdue University, Indianapolis @) ACADEMIC PRESS ‘San Diego London Boston New York Sydney Tokyo Torontooh 300 ALES 1833 ‘This book is printed on acid-free paper. © Copyright © 1999, 1990 by Academic Press All Rights Reserved. [No part of this publication may be reproduced ox ‘wansmitted in any form or by any means, electronic ‘or mechanical, including photocopy, recording, or any information storage and retrieval system, without rtnission in writing from the publisher. ACADEMIC PRESS 4 division of Harcourt Brace & Company 525 B Street, Suite 1900, San Diego, CA 92101-4495, USA. hapu/wwrwspnet.com ‘ACADEMIC PRESS LIMITED 24-28 Oval Road, London NWI 7DX, UK hp: fwwwehbuk.co.ukfapy International Standard Book Number 0-12-050253-4 ‘This book accompanies the following title, catalogued by the Library of Congress: Library of Congress Cataloging-in-Publication Data Alipranis, Charalambos D. Problems in real analysis /Charalambos D. Aliprantis and Owen Burkinshew. Bom. Includes bibliographical references and index. ISBN 0-12-050257-7 (acid-free paper) 1. Mathematical analysis. 2. Functions of real variables. I, Burkinshaw, Owen, I. Tike, QA300.488 1998, 515—de2i . 98-3955 ce Printed in the United States of America 98 99 0 O12 DS 9875654321CONTENTS Foreword (CHAPTER 1. ‘CHAPTER 2. CHAPTER 3. FUNDAMENTALS OF REAL ANALYSIS Elementary Set Theory Countable and Uncountable Sets ‘The Real Numbers ‘Sequences of Real Numbers ‘The Extended Real Numbers Metric Spaces ‘Compaciness in Metric Spaces aawepne ‘TOPOLOGY AND CONTINUITY 8, Topological Spaces 9. Continuous Real-Valued Functions 10. Separation Properties of Continuous Functions LL. The Stone-Weierstrass Approximation Theorem. ‘THE THEORY OF MEASURE 12, Semirings and Algebras of Sets, 13. Measures on Semirings 14, Outer Measures and Measurable Sets 15, The Outer Measure Generated by a Measure 16, Measurable Functions 17. Simple and Step Functions 18. The Lebesgue Measure 19. Convergence in Measure 20, Abstract Measurability u 34 45 54 65 B 92 98 107 lor 2 116 122 133 137 146 187 160CHAPTER 4. CHAPTER 5. CHAPTER 6. CHAPTER 7. ‘CONTENTS ‘THE LEBESGUE INTEGRAL 21. Upper Functions 22. Integrable Functions 23, The Riemann Integral as a Lebesgue Integral 24, Applications of the Lebesgue Integral 25. Approximating Integrable Functions 26, Product Measures and Iterated Integrals NORMED SPACES AND L-SPACES 27. Normed Spaces and Banach Spaces 28. Operators between Banach Spaces 29, Linear Functionals 30. Banach Lattices 31. Lp-Spaces HILBERT SPACES 32. Inner Product Spaces 33. Hilbert Spaces 34, Orthonormal Bases 35, Fourier Analysis SPECIAL TOPICS IN INTEGRATION 36. Signed Measures 37. Comparing Measures and the Radon-Nikodym Theorem 38, ‘The Riesz Representation Theorem 39, Differentiation and Integration 40. ‘The Change of Variables Formula 11 im 174 190 206 20 24 239 239 245 251 259 71 297 310 325 333 345, 345 353, 365, 379 395FOREWORD ‘This book contains complete solutions to the 609 problems in the third edition of Principles of Real Analysis, Academic Press, 1998. The problems have been spread over forty sections which follow the format of the book. ‘All solutions are based on the material covered in the text with frequent refer- ences to the results in the text. For instance, a reference to Theorem 7.3 refers to ‘Theorem 7.3 and a reference to Example 28.4 refers to Example 28.4, both in the third edition of Principles of Real Analysis. This problem book will be beneficial o students only if they use it “properly.” ‘That is to say, if students look at a solution of a problem only after trying very hhard to solve the problem, Students will do themselves great injustice by reading ‘solution without any prior attempt on the problem. It should be a real challenge to students to produce solutions which are different from the ones presented here. We would like to express our most sincere thanks to all the people who’made constructive recommendations and corrections regarding the textand the problems. Special thanks are due to Professor Yuri Abramovich for his contributions and ‘suggestions during the writing of this problem book. C.D, ALIPRANTIS AND 0. BURKINSHAW West Lafayette, Indiana Saly, 1998CHAPTER 1 FUNDAMENTALS OF REAL ANALYSIS 1, ELEMENTARY SET THEORY Probiem 1.41. Establish ihe following set theoretic relations 1, (AUB)NC S(ANC)U(BNC) and (ANB)UC =(AUC)N(BUC); 2 (AUB)\C=(A\ CUB \C) and (ANB)\C =(A\ C)N(B YC); 3. A\B=ANBs; 4 ACB => BPS Aand 5. (AUB = A°N B® and (ANBY = AUB, ‘Also for an arbitrary function f:X — Y, establish the following claims: 6 Fier At) = User FAD: 7 F(a A) S Mer FAD: 8. fre Bi) = ier FB: 9 Fes Bi) = Nig FBI: and 10, f(a) = [fay Solution, (1) We establish the first formula only. We have x@(AUB)NC => re AUB and reC => [xeA orxe Bl andxeC 4} [x€A and x eC] or [xB and xeC] sat EANC orxe BNC =} x €(ANC)UCBNC). (2) Again, we establish the first formula only. Observe that x €(AUB)\C o> x€ AUB andx¢C2 (Chapter 1: FUNDAMENTALS OF REAL ANALYSIS o> [re A orxe BJ and x gC 43 [eA and x ¢C) or [xe B and x €C] ep 1EA\C ore B\C f= x €(A\C)UB\C). (G) Note that XEA\B o> xeAandrgB <> re Aad xe BY es xe ANB. @)Let ACB. Then,x € BY implies x ¢ B and so x € A (ie,x € A*)sothat BE ¢ A®. On the other hand, if BS < A* holds, then (by the preceding case) we have A= (A C (BF = B. (5) Note that XE(ANBY > xEANB ee xg dorsrgB ft AS or xe Be emp x ATURE Moreover, XE(AUBY E> xGAUB E> xgA andxgB exp x EAT and x BS ee re ANS, (© We have yes(UA) ee axeLAr with y= fey ta i $= Ble! with xe Ay and y = fle) => diel with ye f(A) => ye LU f(Ai). (1) From the inclusion f( Mier As) © (Ay) foreach j, we see that (Q ays Aaa) (© We have xe ms) = sore |fa => Bi eT with fa) eB, es Aes with xe (5) o> xe PNB).Section 1: ELEMENTARY SET THEORY 3 (0) Note that xe f"((\Bi) > see Br =e £00 €B, foreach # tel td = re f\(B) foreach fel ee xe (F(R). (10) Observe that sep (B) > fepes = fagee eS re 78) es ve [wy Problem 1.2, For nvo sets A and B show that the following starements are equivalent a AGB; b AUB co ANB B A Solution, (a) = (b) Clearly, 8 AUB holds, Onthe otherhand, fx € AUB, then x € A or x € B, andso ineither case, x € B. This means AUB C B,and hence, AUB = B. (b) => (e) By part (1) of the preceding problem, we have ANB = AD(AUB)=(ANA)U(ANB) = AU(ANB) = A. (6) => (a) Clearly, A= ANB SB. Problem 1.3, Show that (AAB)AC = AA(BAC) holds for every triplet of sets A,B, and C. Solution, Note frst that for any three sets X, Y,and Z we have KAY \Z=1K \WUDIUL \(XUZ)} and Z\ Kar) =[Z\(KUYMUXAY NZ}. For instance, to verify the first identity, note that XEXAY\Z o> [re X\Yorse¥ \X]andx éZ ee eX g¥, and @Z] or [ve¥, x @X, and eZ] ep re (X\ HUZIUM \(XUZ)I4 Chapter I: FUNDAMENTALS OF REAL ANALYSIS ‘Thus, (AAB)AC = [(A4B) \ CIVIC \ (AAB)] [A\(BUCULB\(AUCUIC\(AUBYULAN BNC] A\BUC)IUCANBNCHUIEB\(CUA)UIC\BU ADT) = [A\BAC)IUL(BAC) \ Al = AA(BAC), Problem L4. Givean example ofafunction 1:X —> ¥ and two subsets A and B of X such that f(A NB) # F(A)0 FB). Solution, Define f:{0,1} > (0,1) by f() = f(1) =0. If A = (0) and B= (l}then f(ANB) =o x (0) = f(A)N f(B). Problem 1.5. For a function f:X —> Y, show that the following three state- ‘ments are equivalent: a. f is one-to-one b. FAN B) = f(A)N FCB) holds for all A, Be P(X). &. Porevery pair of disjoint subsers A and B of X, we have f(A) f(B) = 8. Solution, (a) ==> (b) If y € F(A) f(B), then there exist a € A and be B with y = f(a) = flO). Since f is one-to-one, a = b € ANB, and so y € F(AMB). Thus, f(A)M (BY S f(ANB)S f(A) f(B). (6) => (c) Obvious. (6) => (@) Let fla) = fl) If a # b, then the wo sets A = (a) and B = {6} satisfy ANB =, while f(A) fB) = (/@) #d- Problem 16. Let f:X > ¥ be. fumetion. Show that f(f-'(A)) & A for all ACY, and BS f-'(f(B)) forall BCX. Solution. Clearly, x € f!(A) ifand only if f(x) € A. Thus, (f(A) ¢ A. Similarly, x € f-'(/(B)) if and only if f(x) € f(B), and so BS F"(F(B)) holds. Problem 1.7. Showtharafunction f:X— ¥ isontoifandonly if f(f-“B)=B holds for all B CY. Solution, Assume that f is onto and B CY. If b € B, then there exists some ae X with f(a) = bi clearly, a € f-1(B). Thus, b= fla) € f(f-\B)), and so BS f(f-(B) SB holds.Section ts ELEMENTARY SET THEORY 5 Forthe converse note thatthe relation /(J~"((0))) = (0) implies f—\((b)) # @ foreach be ¥ sothat f is onto. Problem 18 Ler X 4+ ¥ $+ Z. if AS Z, show that (go fA) = Fe (AD. Solution, Note that x € (go f(A) a> gf) € A ee f(x) © BNA) ga x © FE AD). Problem 1.9. Show that the composition of functions satisfies the associative law. Thar is, show thar if X Le ¥ > Z +> V, then (hogyo f = holgo fr. Solution, Observe that for each x ¢ X we have [Grogho fix) = hog f@)) = he(FO) = Mego AG) = hols o fIIG). ‘Therefore, (ho g)o f =holgo fi. Problem 1.10. Let f:X — Y. Show that the relation Ron X, defined by xyRex whenever (1) = Fe), 18 an equivalence relation. Solution, We must show that the relation ‘R is reflexive, symmetric, and transi- tive Reflexivity: Note that f(x) = f(x) implies xRx for each x € X. Symmetry: Let x Rexp. Then, f(@r) = f(a) of fn) = Fe), $0 that x97R.xy Transitvity: Hf xyRxp and xjRexy, then f(x) = f(%2) and f(z) = f(s) both hhold. It follows that f(x1) = f(x), and $0 x, Rx, Problem LIL. if X and ¥ are sets, then show that POONPU) = PRAY) and POQUPH)S PUK UY). Solution, (a) Note that AGPI)NPW) ed ACK and ACY = ACKNY 9 AE PIKNY), (b) Clearly, AE PIX)UPU) = ASX of ACY =p ACXUY = AE PIXUY).6 (Chapter 1: FUNDAMENTALS OF REAL ANALYSIS. If X and ¥ are two nonempty disjoint sets, then X UY ¢ P(X)U PCY), and so ‘equality is seldom valid. 2, COUNTABLE AND UNCOUNTABLE SETS Problem 2.1. Show that the set of all rational numbers is countable, Solution. Let Q be the set of rational numbers and let Q* ‘Then the function f:N x N+ Q* defined by f(m,n) conclusion now follows from Theorems 2.7 and 2.5. €Q: r>0} is onto, The Problem 2.2. Show shat the set of ailfinte subsets ofa countable set is countable. Solution. We can assume that A = {p1, po, ...] isthe set of all prime numbers. Let F denote the collection of all finite subsets of A. Define f:F ~» IN by (F) =the product of the elements of F, foreach F ¢ F. Then f is one-to-one, and the conclusion follows from Theorem 2.5. Problem 2.3. Show that a union of an at-most countable collection of sets, each of which is finite, isan at-most countable set Solution. This follows immediately from Theorem 2.6. Problem 24. Let A be an uncountable set and let B be a countable subset of A. Show that A is equivalent to A\ B. Solution, Let B = (bi, 52,-.-}. Since A is uncountable, the set A \ B is also uncountable. Let C = {¢1,€2,...) bea countable subset of 4 \ B. Now define PiA\ BA by x frees FOD= | Cnety if x = capa (= 01,2, by if ey (= 1,2,-.) ‘Then f is one-to-one and onto, proving that A= A \ B. Problem 2.5. Assume that f:A —> B is a surjective (onto) function between wo sets. Establish the fallowing: a card B < card A. ‘ b. IFA iscountable, then B is at-most countable.Section 2: COUNTABLE AND UNCOUNTABLESETS 1 Solution. (a) Consider the family (f~!(b): b € B). Clearly, this is a family of| disjoint subsets of 4. By the Axiom of Choice there exists asubset C of A such that C9 f(b) consists precisely of one element of A for each b € B. The conclusion now follows by observing that #:C —r B is one-to-one and onto. (0) This follows immediately from part (a). Problem 2.6. Show thar wo nonempty sets A and B ave equivalent ifand only if there exists a function from A onto B and a function from B onto-A.. Solution, if A and B are equivalent, then there exists a function f:A —> B ‘which is one-to-one and onto, Clearly, f!: B —» A isa surjective function. For the converse, assume that there exists a function from A onto B and a function from B onto A. A glance at the preceding problem guarantees that card B < card A and card A < card B. Now, use the Schréder-Bemstein theorem to conlude that A and B are equivalent sets. Problem 2.7. Show that if a finite set X has n elements, chen its power set P(X) has 2" elements. Solution, We shall use induction on m, Assume that {1,...,m} has 2" subsets. ‘Then the subsets of the set (I, ...,72, + 1} consist of: a. The subsets of (1, ..., 2}, which are 2" altogether, and b. The subsets of the form AU (n +1), where A isa subset of {1,.... 7), again 2" altogether. ‘Thus, the number of subsets of (1, nal] is 242" = 2M, ‘trot roof gesas follows, Notice thatthe number of subsesf 1.2, having k elements (ohere Ok 2¥ by An fa, where fa(x) = Lif x € A and fx(x) = 0 if x ¢ A, Note that f is one-to-one and onto. ‘Therefore, 2 = PRX). Problem 2.10. Any complex number that is a root of a (nonzero) polynomial with integer coefficients is called an algebraic number. Show that the set of all algebraic numbers is countable. Solution, Let Z = {...,-2,—1,0, 1,2,...J. Fix n> 1. Since every poly rial p(x) = ag-bayx-b- +042" isdetermined uniquely by (ao, ay, da)ei easy to see thatthe nonzero polynomials of degree R. show thatthe set Az [aeR: lim fa) exisisand lim f0) # f(e)} is at-most countable. Solution, Let Z denote the set of all open subintervals of JR with rational endpoints and note that Z is acountable set. Also, let Q denote the countable set of al rational numbers of R. ot each rational real number r, let Ay = (a € A: Bither f(a) 0 such that a-B r. Next, pick an open interval /__ with rational endpoints (i.e. fy €) such that @ € Jy and Ig © (a~8,a+8). Since f(y) > r‘Section 2: COUNTARLE AND UNCOUNTABLE SETS 9 holds for each y € Jy with y #a, we see that y ¢ A, foreach y € Iy\{a). In particular, note that A, 1 fy = {a)) ‘Thus, we have established a mapping a> J, from A, into Z (which in view of 4, "fa = (a) for each a € A,) is also one-to-one, ‘This implies that A, is ‘at-most countable, and hence, A is likewise at-most countable. Problem 2.12, Show that the set of real numbers is uncountable by proving the following: 2) (0,1) * Ry and 2 ) (0,1) is uncountable, Solution. (a) The function f:(0,1) > IR defined by the formula f(x) = tan(rex — 3) is one-to-one and onto. (©) IE (0,1) is countable, then let (x1, 12,...} be one enumeration of (0, 1. For each n write x_ = Odyidea++ in its decimal expansion, where each diy is 0,1,..., of 9. Now, consider the real number y of (0, 1) whose decimal expansion y =0.yiy2~++ satisfies yx = 1 if dyn #% Land Yu = 2 if dyn = 1. An easy argument now shows how?) that y # xy foreach n, which is acontradiction, Hence, the interval (0,1) is an uncountable set. Problem 2.13. Using mathematical induction prove the following a. a> —l,then(+a)" > 1+naforn = 1,2,...(Bernoull’s inequality). bf 0 (+a) form= 1,2... cost) = (1) forn = 1,2, Solution. (a) Let a > —1. Form = | the inequality is trivially true; infact, itis ‘an equality, For the induction step, assume that (1 + a)" = 1 + na holds true for some n, Since 1 +a > Ois assumed to be true, it follows that (tay! = (14a) +a)" = (1 tay +a) = Lbna ba tna Lt (nt Dat na? 214 (n+ 1a, which is the desired inequality when n takes the value + 1. This completes the induction. (b) Assume 0 1 +a, the desired inequality is true for n = 1, For the inductive step assume 1 4+ 3%a > (I +a)". Then, taking into ‘account that 0 < a < J, we see that (tay! = (1 -ayl +a)" <0 +a +3") +8 a-at3'o? =14G'43a4 Da S14 GSH 49a = 143-3a = 143%a,6 (Chapter 1: FUNDAMENTALS OF REAL ANALYSIS which i the desired inequality valid when n is replaced by n+ 1. By the Principle of Mathematical Induction, the inequality is tre for every natural number n. (© For n = 1, we have cos(1 - 2) = cos = —1 = (—1)', Now, assume that cos(nx) = (-)*. Then, using the trigonometric formula cos(x + y) = cos.xcos y ~ sin.x sin y, we see that cos{(n + )n'] = cos(nar +) = costar) cos m ~ sin(nx) sin (1)"(-1) = sini) 0 = 1)", ‘and the induction is complete, Problem 2.14. Show shat the Well-Ordering Principle implies the Principle of Mathematical Induction. Solution, Let $ IN satisfy a Le Sand bo a+1eS whenever n€ 5. ‘We must show that $ = N, or equivalently that N\S'= ob ‘To this end, assume by way of contradiction that we have IN \ S # @. Then, by the Well Ordering Principle, n = min(IN \ S) exists, Clearly, b. Thus, Hatb+\e—b)=Ha+b+a-b)=a=0Vvb, and Na+b—|a~b)= Ha+b- (aD) Problem 3.2. Show that |la} ~ [bl] < la +61 s lal + |b] forall a, 6 ¢ R. Solution. From —jal Oisa rational number. Then V/3 =r —/3 and by squaring, we get 3 =r? — 2r/2 +2. This implies /2 = S=1, a rational number, contrary to our previous conclusion, Hence, v/2 + V3 is an irrational number. Problem 3.4, Show that berween any two distinct real numbers there is an irra~ tonal number, Solution, Leta < b, Choose a rational numbers with 2 < 7 0. 1. Areal number a satisfies a = ~a if and only ifa = 0. ce Una lt Lest | Quere the sum has “n summands" ali equal 101), then show that these elements are al distinct; as usual, we shalt call the collection N ofall these numbers the natural numbers of R. 4. Let Z consist of N together with ts negative elements and zero; we shall call Z, of course, the set of integers of R. Show that Z consists of distinct elements and tha its closed under addition and multiplication.16 (Chapter 1: FUNDAMENTALS OF REAL ANALYSIS: Define the set Q of rational numbers by Q = (2: m,n € Zand n # 0). Show that Q satisfies itself axioms through 10 and that sup(r € Q: r OorO > 1. If > 1, then (by Axiom 9) we have 0+ (=I) > 14(-1) =0or-1 > 0,6, —1 is a positive number, Now, using Axiom 10, we infer that 0- (—1) = 1- (1), or 0 = =I, contrary to ~1 > 0, Hence, 1 > 0. (@) Since 0 +0 = 0, we know that —0 number a satisfies a = —a. This implies a +a 1 > 0, we have 1+12 1-+0= 1 > 0, and so1+1 #0. Consequently, from the zero product rule, (1+ 1)a = 0 implies a = 0. (©) As shown in part (b) above, 1 + 1 # 0 and in fact 1 +1 #1; otherwise 141 = 1 = 1+0 implies (in view of the cancellation law) 1 = 0, which is impossible. Now, by induction, assume that |. Conversely, assume that a real + 1)a = 0, However, since O<1 m= 1-41 +--+ 1 (where the last sum has 7 summands). Indeed, if n +1 < a, then (n + 1)-+ (=n) < n+ (=n) or 1 <0, ‘hich is a contradiction. Hence, #i-+ 1 > m and the induction is complete. (@ By part () we know that the natural numbers together with zero are all distinet real numbers, If —m = —n with m,n © IN, then m = n, which shows that distinct natural numbers have distinct negatives. If m = —n with m, m € JN, thenSection 3: THE REAL NUMBERS w ‘m+n = 0 contradicting (c), and so no natural number can be equal to a negative integer. It now follows that Z consists of distinct elements. (@) Observe that if $ and 2 are two rational numbers, then mi pmg+m om Bmp ni qq nq nq" and if 2 #0, then ()' = 2. Thatis, Q is closed under addition, multiplication and inverses. Since all ral umber satis axioms I through 10, i follows that Q icf eatsies axioms | through 10 nits own right For the second par, fx a € R and lot A = (r © Q: r 0. By Theorem 3.4, there exists some rational number r such thata—e <1 0. This implies a < sup A, and hence sup A. The equality, a= inf{s Q: a 0 in R’. Then there exist rational numbers r.$ € Qwithr 0. This guarantees /(a)+ (0) = f(a+b), and therefore Fla +b)= fla)+ fe). Problem 3.9. Consider a svo-point set R = (0, 1} equipped with the following ‘operations: a Addition (4) :04+0=0,0+1=1+40=1and1+1=0, b. Multiplication ()):0-1 = 1-0=Oand 1-1 = 1, and Ordering: 0 2 0,12 1 and 1 2 0. Does R with the above operations satisfy all eleven axioms defining the real num- bers? Explain your answer, Solution. It satisfies all axioms except Axiom 9, which states that: © fx z yandz > 0,thenx+22 y +2. ‘To see this, assume that Axiom 9 is valid. We distinguish two cases. CASEE1 > 0. In this case, we must have 0 = 1+ 1 > 0+1=1, which contradicts 1 > 0. CASEM:0>1 ‘This implies 1=0+1>141= Axiom 9 does not hold in this case. {tshould be noticed that Axiom 9 is the one that guarantees that 1+ 1 (je, the ‘umber 2) is distinct from 0 and 1; and, of course, itis the axiom that establishes (as we saw in past (b) of Problem 3.8) the existence of the set of integers. , which again contradicts 0 > 1. Thus, Problem 3.10. Consider the set of rational numbers Q equipped with the usual operations of addition, multiplication, and ordering. Why doesn't Q coincide with the set of real numbers?Section 5: THE REAL NUMBERS ry Solution, The set of rational qumbers satisfies all the axioms of real numbers except the completeness axiom. This was proven in part (e) of Problem 3.8. To see that Q does not satisfy the completeness axiom, assume by way of contradiction that it does. Consider the set Sel0crear<2 “Then Sis nonempty and bounded from above in Q(why?), and so. = sup S exists in Q. Now, repeat the proof of Theorem 3.5 to conclude that ? = 2, ic., that 5 = V2. However, we proved in Problem 3.3 that /2is not rational number, and we have reached a contradiction. Hence, Q does not satisfy the completeness axiom and it cannot coincide with the set of real numbers. Problem 3.11. Thisproblem establishes the familiar rules of “exponents” based ‘on the axiomatic foundation of real numbers. To avoid unnecessary notation, we shall assume that all real numbers encountered here are positive—and so by Theorem 35, all non-negative real numbers have unique roots. As usual, the - “integer” powers are defined by Extending this to rational numbers, for each m,n © IN we define iia eee at = Ya" and a Establish the following properties: a al =(Yay" forallm.neN. b. Ifm,n, pg © IN satisfy & then a’ c. Ifrands are rational numbers, then i fa! =a" and © =a, ii, (@bY =a'’ and (§Y =F, and iii, oy Solution, It should be noticed first that (a")" = (a")" all natural numbers m,n € IN, (a) Notice that (wary = War Way". Wa" = Ya- Ya Ya fear penfasere = Way Way's Way =e-a---8 eefactors msoners Since the n-roots are unique (Theorem 3,5), we infer that (Ya J" = a" =0 (Chapter 1: FUNDAMENTALS OF REAL ANALYSIS, (©) Assume myn, aq € NV satily & = 2, or pn = mg, Using part (a), we see that (off = (Wary = [a] and this shows that a = a (©) The formulas can be established easly ifr and sare integers. Now letr and 4 be rational numbers, We shall assume that rand s are also positive and leave the “negative case” for the reader. By part (b), we can also suppose that r = ands = 2, Since (a - Os given, then note that wh ce eae Ine") =ninx BE inx Problem 4.2. Show that lim.ty = x holds ifand only if every subsequence of (6) has a subsequence that converges tox. Solution. If lim x, =x, then every subsequence must converge to x. So, every ‘subsequence ofa subsequence (es being itself subsequence of (4,}) must converge tor. For the converse, assume that each subsequence of {xq} has a subsequence that converges to x. Now, suppose by way of contradiction that {x,) does not converge to x. Then for some © > 0 we must have |x ~,| 2 € for an infinite number of n, So, there exists a subsequence (4) of (X=) stich that [x— yx] 2 & for each n. However, the later contradicts the fact that (yq) has a subsequence that converges to x. Therefore, lim tq =.Section 4: SEQUENCES OF REAL NUMBERS a Problem 4.3, Consider nwo sequences {ky} and (my) of strictly increasing nat- tural numbers such that for some & & IN we have (EL CED VS (hry PU bamay tag Show that a sequence of real numbers (x4) converges in R if and only if both Subsequences (x) and (n,) of (89) converge in Rand they satisfy tim, Timp, (in which case the Common limit is also the timit ofthe sequenct ‘In particular, show that a sequence of real numbers (%,) converges in R ifand only the "even and “oda” subsequences (az) and {zy 1} bot converge in aed they satisfy Yim xy = lim Xa Solution, If > x, then clearly xy, > x and x, > +. For the converse, assume that x, -» x and Xn, —> x. Let € > 0. Choose some mg € IN such that =a] no. © Put fo = max{E, ky Mag), and we clam that |n—x] <6 forall n> 6, ‘To se thi, len 2 €o. Then the assumption (ERLE dS Urs kaye DU (mys may) guarantees the existence of somer € IN such that ky = 1 orm, =m. Sincer < 7g iimplics ky < ky to and my < tiny < foy we see that r 2 mo. Hence, ether aq =, Of Ny = tm, (With r & no), and so from (+) it follows that fx» — x] <€ This shows that x» —> x. ‘The last part should be immediate from the above conclusion Problem 4.4, Find the limsup and liming for the sequence ((—1)*). Solution, We have liminf{—1)" =—1 and limsup(—1)" = 1 Problem 4.5. Find the lim sup and liminf of the sequence (xy) defined by ays bo xa = beaents and angi =} Xan for n= 1,2, Solution, We claim that 1 . EYE and =} ¥ ms & hold for n = 1,2,.... The validity of the identities can be established by induction. We shail establish the validity of the second identity and lesve the2 (Chapter 1: FUNDAMENTALS OF REAL ANALYSIS verification of the first to the reader, For n = 1, we have em ettesitigai+p=t(r4)) Now, assume that 23.31 suns = fag = | + pe See east +o8 Now, we claim that £ and | are the only limit points of (xq). To see this, let 4 beareal number different from J and 4. Pick some e > 0 such that lim x, Behe} ond @~eateNG—egte=d and (@-e,atend—e,h+e=6. Next, roe that there exists some & such that xp, € (f ~ 6.) +e) and x24) € (—e,} +6) hold forall n > £. Therefor, jr, ~ al > # holds forall n> and this shows that @ cannot be a init point ofthe sequence {25}. Consequently, liminfxy =} and limsupsy = } Problem 4.6. Let (xq) be a bounded sequence. Show that lim sup(—x_) = —liminf x, and lim inf{—x,) = — lim sup xy Solution. We shall use the fact that lim sup x, and liminf x, are the largest and ‘smallest limit points of (x,), respectively. We shall establish the first formula, Choose two subsequences {y,} and {z,)} of (xq) such that lim yq = lim inf z, and tio(—2,) = li im(—yx) Slim sup(~x,) = lim(—2,) = ‘S ~limintx,, lim, and so lim sup(—x,) = ~ lim inf.x,Section 4: SEQUENCES OF REAL NUMBERS 2 Problem 4.7. If {ss} and {yp} are two bounded sequences, then show that a, limsup(t,-+ ys) $ limsup x, + lim sup yp, and b.Timinf(xy + yy) = lim inf.x, + lim inf yy Moreover, show that if one of the sequences converges, then equality holds in both (a) and (b). Solution. (a) By passing to a subsequence, we can assume that lim, + J) = lim sup(ty + yq)- Since (x4) is a bounded sequence, there exists a subsequence (1) that converges. Let x = lima,. By the same reasoning, there exists a subsequence of (yx,} that converges to some y. Thus, there exists a strictly increasing sequence (my} of natural numbers such that x = lim», and y = Jim Yn. Hence, Tis sup(ry + Ye) $y = Hin ajp,+ Tim Yn lim sup.ty + Lim sup yo. Finely, if x = lim, holds, then pick a subsequence (3,) of (yx) such that lim ye, = limsup ye, and note that limsup x, + lim sup y, = x +lim y lim(g, + 94,) & lim suply + J)- {b) It follows from (a) by using the preceding problem. Problem 4.8. Prove that the limsup and lim inf processes “preserve inequali- ties!” That is, show that if two bounded sequences {,) and (Yq) of real numbers satisfy Xy Yu for all n> rg, then limint.x, $liminf yy and limsup x, < limsup yn. Solution, First, we shall show that iftwo sequences of real numbers (55) and (f} converge in R (sey $5 > § and ty > t) and s» < % for each n = no, then s <¢ Indeed if, s > ris true, then let e = 43! > O and note that forall n sufficiently large, we must have se e6-6sO= (FEM) tod Hele on ORE That is, < 242 < Sq ust hold forall n sufficiently large, which is impossible. Bence, <1 ISTANBUL BILGI UNIVERSITY LIBRARYu (Chapter 1: FUNDAMENTALS OP REAL ANALYSIS. ‘Now, assume that two bounded sequences of real numbers {x} and {3} satisfy An Syn for all n = rio. Put f Ye Sy = infx, and t a En If m > no, then notice that for each r > 1 We have Sy = infron Xk Se S Yeu and 80 Sy < ithray Jr = fe foreach n > no. By the discussion ofthe fist pat, we infer that lim infxq = lim, < lim ty = liminF “The lim sup case can be established ina similar manner, orby using the formula lim sup x, = — lim inf(—x,). Problem 4.9. Show that tim./mi = 1 (and conclude from this that Yim 0). Solution. Note that i = (/J)’. An easy inductive argument shows that 9Fa = 1 holds for each n. Thus, we can write Jn = 1+, with x, > 0. ‘Since (I +a)! > 1-+na holds for each m and each a > 0 (see Problem 2.13), wwe get Vii= {ay = (4m) Elta, and so 0 <.% <4, —4, This implies lim, = 0. Therefore, f= (Ja) = (1+ an)? 1 [Analtemate proof goes as follows: By L'Hépitals Role, wehve lito BE = 0, and 80 limy-s99 8 = 0. ‘Therefore, using that the exponential function is continuous, we infer that lim Yn For the parenthetical part, assume first a > 1. Then it is easy to see that 1 < fa < x/m holds true for all n > a. Consequently, by the “Sandwich Theorem,” wwe see that lim a = 1. If0 Tanto ling lim dp = 1, from which it follows that lim 0 holds for ceach m, We shal establish lim sup «/% < lim sup <= and leave the similar proof of the other inequality for the reeder. Put xs timp = AY, Inline Ht and note thatif x = 00, then theres nothing fo prove, So, weean assume x < oo. [Let ¢ > 0 be fixed. Then there exists some k such thar + < x +e holds forall n = k, Now, for n> & wehave Beak oo Meta tee =e teyC, where ¢ = aye +2)" is a constant. Therefore, /iq < (x + s)/E holds for each n > & and so, in view of lim 0 is arbitrary, we infer that lim sup /%y Sx im sup Sh Problem 4.11. The sequence of averages of a sequence of real numbers (x) is the sequence {ay} defined by a, = SS=E=!, If (x5) 8 a bounded sequence of real numbers, then show that ty Sliminfa, x, then show that ay > x. Does the convergence of (ax) imply the convergence of (Xn)?2% (Chapter 1: FUNDAMENTALS OF REAL ANALYSIS Solution. The solution will be based upon the following properties of lim sup and lim inf; © If {uy} is @ bounded sequence of real numbers, then for each « > 0 the ‘inequalities uy > limsupity +e and um < liminfuy, hold for finitely many k and finitely many m. To see this, assume by way of contradiction that 4 > lim sup, + € holds true for infinitely many k. Then there exists a subsequence {vp} of (u) satisfying v% & limsupu, + for each n. Since (vp) is a bounded sequence, there exists ‘a subsequence {w,} of {v,} (and hence of {u,}) satisfying w, > w € IR. By Problem 4.8, we know that w > limsupit, + €, i., w isa limit point of (i) which is greater than the largest limit point (lim sup.) of {ue}, a contradiction. Now, let {4} be a bounded sequence of real numbers and fix « > 0. Put = limsup x, and let K = (K€ Ni x 2 ¢ +). By the above discussion, K is a finite set. Put S,=EN:ieK andisn) and T= (iN ig K andi m. Thats, lim.ty = ~c0. limsupx, € R*. Then Problem $5.2. Let (i) be a sequence of positive real numbers such that 2 = Tim 1 exists ia R, Show that a fb < lythen tim, b> Lithen limay = ) and Solution, (@) Assume ¢ < 1 and fix some 8 such that & < B < 1; for instance, let 8 = 1. Since lim 1 = thre exists some > 1 such hat + < Bholds forall n > k. Now, ifn > k, then note that (ones anid so fe = 3, then O &. Since (in view of 0 < & < 1) 6" + 0, we easly infer that 0. (b) Assume now 2 > 1 and choose some 5 such that 1 <8 < ¢. Since lim S24 = é, there exists some k > 1 such that #4 > 5 holds fr alin 2 . Then, as in the preceding case, there exists some consiant C > 0 satisfying x, > C3" for all n > k. From" -> oo, it easily follows that x4 —> 00. Problem 5.3. Let © < dn» <00,r all m, n, and let o:IN x N—> N x IN be36 (Chapter 1: FUNDAMENTALS OF REAL ANALYSIS ‘one-to-one and onto. Show that Yd en = LS aon mtn metal Solution. It follows immediately from Theorem 5.4, Problem 5.4. Show shat Solution. The convergence or divergence ofthe series is according tothe con- vergence or divergence of the double integral f°" 42, Now, note that PU stp le=t [Fa mim” Gime” wm tm ey implies ea what 2 Cea (ae LS ane Problem 5.5. This problem describes the p-adic representation ofa real num- ber in (0,1). We assume that p is a natural number such that p > 2 and *€,1). a, Divide the interval (0, 1) into the p closed-open intervals ee and number them consecutively from 0 to p — 1, Then x belongs precisely (9 one of these intervals, sa) ky (0 < ky ~< p). Next divide the interval (4s, £22) into p elosed-open intervals (ofthe same legit), number them consecutively from 0 to p ~ 1, and let kz be the subinterval to which x belongs. Proceeding this way, we constractasequence hy] ofnon negative‘Section §: THE EXTENDED REAL NUMBERS a integers such that 0 < ky < p for each n. Show that Ot > b. Apply the same process asin (a) by subdividing each interval now into p open-closed intervals. For example, start with (0 1) and subdivide it into the open-closed intervals (0, 1,5, 2]s..-5 5. U. {As ina), construct a sequence (my) of non-negative integers suck that 0 1, while my =O and m, = 1 form >1 we have ky Problem 5.6. Show thar PON) © IR by establishing the following: i, Ais aninfnite set, and f: A> B is one-to-one such that B\ f(A) is at-most countable, then show that A * B. Ji, Show thatthe set of numbers of (0 1) for which the dyadic (ie., p = 2) representation determined by (a) and (b) of the preceding exercise are differen is a countable set. fii, For each x € (0,1), let x = O.kjky-+- be the dyadic representation determined by part (a) of the preceding exercise; clearly, each kis either Oor 1. Let f(s) = (0 EN: ky = 1). Show that f2(0,1) > PON) is one-to-one such that PON)\ f\(0, 1) is countable, and conclude from part (i) that (0, = PON), Solution, {i} Let $= (a1, a9,...} bea countable subset of A. (a) Assume B \ f(A) = (bi, ba} is a finite set. Then g:A—> B defined by g(x) "(x) if x ¢ S, gla) = bj for 1 si B defined by gx) = C2) if x #5, ldanss) = fle) and g(@zm) = by foreach nis one-to-one and onto,cy (Chapter 1: FUNDAMENTALS OF REAL ANALYSIS Gi) Let D be the set of all numbers of (0, 1) for which the two sequences (t] and {m,} determined by the preceding problem are different. Assume x = Oki ka- O.mym--+ € D and define the natural number r = min{n: ky # re). Wecan assume k, = 1 and m, =O. Then the inequalities cae = godec ds guarantee that x= BAB bet 4, =0 hold foreach n > r. Onithe other hand it is not difficult to see that every x ofthe above type belongs to D. It is mow a routine matter to verify that D is a countable set. (It is also interesting to observe that D consists precisely ofthe endpoints ofthe subintervals appearing during the construction of the expansions.) Gil) Let Ae PQ). Define the sequence {rm,) of (0, 1) by my =1if nA and m, =0 if n ¢ A, and then set Le ot Note that A ¢ f((0,1)) if and only if x € D. Thus, PUN) \ f((O,1)) is countable, and so by part (i) and the fact that f is one-to-one, (0, 1) ® P(N) holds. Problem 5.7. Fora sequence {1,} of real numbers show that the following con- ditions are equivalent: a The series 5°, xy is rearrangement invariant in R b. For every permutation o of NN the series Sx, converges in TR. c. The series D2; |xq| converges in R. For every Sequence {5x} of (=I, 1}, the series S22 sexe converges in R. € For every subsequence {24} of tg) the series $=, x, converges in R, For every > 0, there exists an integer k (depending on e) such that for ‘every finite subset S of N with min § > k, we have | yes %nl <€. (Any series 7%, x, satisfying any one of the above conditions is also referred to as an unconditionally convergent series.)Section $: THE EXTENDED REAL NUMBERS » Solution, (a)=+(b) Obvious. (=H) Assume Fo, bx] = 00. From our hypothesis it follows that x, > O ‘and x» < 0 both hold for infinitely many m. Split (x_) into two subsequences {ie} and {25} such that Jq 20 and 2, band 2 tet Dicer > for n= 12, Then note that Vey seen Tye Bho Wapty sey Thay 2s Waste oe isa permutation of (¢} whose series is not convergent, contrary to our hypothesis, (=H@ Odviows. (=e) Let {xe,) be a subsequence of (x9}- Put sy = —1 if f # ky foreach nvand 5, = Le Then Yeadon] it a convergent sre, (=3(O IF (0 is false, then there exists some ¢ > 0 and asequence (S,} of finite subsets of natural numbers such that max S, IN be a permutation. By our hypothesis, it is readily seen that the partial sums of both series S772 %e and Sets Xo, form Cauchy Sequences, and hence, both series converge in IR. Let = Dien» and y ‘Now. if ¢ > 0 is given, then choose & so large such that b-Sul k. Fixsome r >“0 (Chapter 1: FUNDAMENTALS OF REAL ANALYSIS. such that foreach 1 0, and so x = y. In other word, the series Sy 2p is reavrangemeat invariant Problem 5.8. A series of the form J7m,(—1)""1xq, where xq > O for each n, is called an alternating series. Assume that a sequence (x4) of sriily positive real numbers satisfies x, 4.0. Then establish the following: a. The alternating series ya,(~1)'~'x, converges in R. b. Uf De) Xe = 00, then the alternating series So (—1)''x, is not rear~ rangement invariant. Solution. () Let sy = F(—D! fa, We claim that as +S Sand S Son S Stet Sona $6 S555; holds for each n, The proof is by induction, For n = 1, we have s2 = x) ~ x2 < x1 = 5). So, assume the inequalities to be true for some n. Then taking into account that >, ~ x2n41 > O and x2_41— Xanga = O, We soe that Le Sig $ S044 Gangt — tanga) = Sana = Saas 2. Sunes) = Sansa = Songs ~ Xay47 Sonat = Saale 3. Snyet)-t = San9t = Sanct — fae ~ Tanet) $ Stet and our claim is established. Now, ifsay t's and soq-1 | £ holdin R, then cleanly 5 < 1. Moreover, from San ~ Snot = Xin 0, we obtain s =, But then this implies that (sy) converges tos in R; see Exercise 43 of Section 4. Consequently, the alternating series converges and S722, (—1)'1x, (b) We must have either Sy x SE tax. = oct the other case canbe treated ina similar manner Since D7) x21 = 00, there exist integers 0 = ko < ky < ky < + such that [OSG .22—1] — 22, > 1 holds for each nm = 0,1,.... Consider the rearrangement {y¢} ofthe sequence {(—1)"1x,} given by BU ee abet Hgts WE Aas and note that 22,Section §: THE EXTENDED REAL NUMBERS a Problem 5.9. This problem describes the integral test for the convergence of series. Assume that f:{1, 00) —> (0, 00) is a decreasing function. We define the sequences (0,} and {ta} By n=) and y= f fear Exablsh the folowing: a O05 oy — ty < f(t) forall n. bh the sequence fo, ~ ty) decreasing—and hence, convergent in R. c. Show that the series 5-2, F(k) converges in R if and only if the improper Riemann integral [°° Fx) dx = litt-oeo [" fle) ax exists in R Solution. Since f is decreasing, notice that for each & ¢ NV we have f(x) > FHT) and f(x) < Fk) foreachk 1. The foloing are problems related to the harmonic series S~™ et &. Prove with (a least thee diferent ways that 2%, ! = oo b. Ifa computer starting at 12 midnight on December 31, 1939, adds one million terms of the harmonic series every second, what was the value (within an error of 1) of the sum at 12 midnight on December 31, 1997? (Assume that each year has 365 days.) Show thar 3° mt +2)= 102. ee y= dim (a5 + a Solution. Notice that S Geta [Se [eee beet ‘This limit is finite if p > 1 and infinity if 0 < p <1.‘Section §: THE EXTENDED REAL NUMBERS. 8 (@) Weletoy = Ct; }. Here are four proofs ofthe divergence ofthe harmonic mi tanto +a =" 1, Notice that omy ~ 0 she = med = | foreach This shows that {o,} isnot a Catchy Sequence, and hence, divergent, 2. Asshown atthe beginning of te solution ofthe problem, /-" 4 = 00, and so Det 3, Wecloim'that op > 1-44 for each n, (If this inequality is established, then clearly SZ, | = limoy = 00.) The proof of the inequality is by induction, Forn = 1, we have oy) = oz = 1+ 4. Now, if we assume the ‘inequality true for some n, then fr Hee Dei Bitte 4. Note that [ede sHlelpes L 1 1 DOG +O: Tr +900- saaG + (b) From Problem 5.9, we know that the harmonic series is associated with the function’ f(x) = + end that 0-< 0, —Inn-<-f(1) = 1 for each n.-So, Inn approximates o,, within an error of one. Ifthe computer started adding the terms of the harmonic series at 12 midnight on December 31, 1939, then up to 12 midnight ‘on December 31, 1997, there are ‘ST(years) x 365(days) x 24(hours) x 60(minutes) x 60(seconds) = 1,797,552 x 10? seconds. So, if the computer adds 1,000,000 = 10° terms per second of the harmonic series, the last number V added a second before midnight on December 31 of 1997 is N = I, 797, 552 x 10°, Therefore, 5.12520213.. “which shows that the harmonic series is a “very slow" divergent series.“ (Chapter 1: FUNDAMENTALS OF REAL ANALYSIS (© From Problem 5.8, we know thatthe altemating series SO, [=Y" is con- ‘vergent in R. Also, from Problem 5.9, we know that lim(o, — Inn) = y € R. So, ‘if we let x, = y —(o, — Inn), then x, > O anda, = y +1nn + xy for each n. Now, note that wet teat tae = Om ~ Oy = [y +In(2n) + x24] — [y + Inn 4x4] = n2 +30) — foreach n. This implies P22, SP" = in. Problem 5.11 (Foeplitz). Let (ay) bea sequence of postive real numbers (Le. Gq > 0 for each n) and put by at 9. Assume that by t S724 a; = 00. If (ie) is @ sequence of real numbers such that xy —» x in R, then show that i ans, ye Solution. Lete > 0. Choose some k such that |x, —x] < ¢ foreach n > i: Put M = max{lxj— xl: 1 ‘},and then select some é > k such that M2 < for all n= €. Now, notice that ifn > ¢, then le Sas| sp bY am - alte ala Moy sop tesetente land the conclusion follows, (Note that this problem is a substantial generalization of Problem 4.11.) Problem 5.12 (Kronecker). Assume that a sequence of positive numbers (Ds) satisfies 0 < by < by < by <-+-andby t 0. Ifaseries S-™ xq of eal numbersSection 6: METRIC SPACES 4“ converges in R, then show that m > obs =O. In particular, show that f(y #4 sequence ofreal numbers such that the series SEE converges in R, then SHS —> 0, Solution. Let x = DiEpxy. Put by = 0, 0 each n > 1, Now, notice tat Shay = ode — sd ase = Posy = bo 0, and S_ = 4 #-1+ by for Therefore, Tfay busi = Se — ff Dhan siealbs — Bea), Since by — ber > O for each i and Y°%_(b; ~;1) = By 7 09, it follows from the preceding problem that £ hg si-ilb; — by-1) = =. Hence, ag Doe in 5, Leh] For the second part, notice that if 2%, % is convergent in R, then let by for each n and notice tha (by the above) i mitt yn bee —o, ine as desired. 6. METRICSPACES Problem 6.1. For subsets A and B of a metric space (X, d) show that: a (ANY = Ang", b ASU BSC (AU BY. c AUB=AUB. 4 anB cane. _ FB isopen, hen AB SAAB Solution, (@) From (ANB) © A and (ANB)? & B°, itfollows that (ANB)? S AMBP. Onthe other hand, since ACNB® S ANB holds and A°N.B° is open, it easly follows that APM B® < (AN BY.46 ‘Chapter 1: FUNDAMENTALS OF REAL ANALYSIS. (b)From 4 © AUB, itfollows that A° ¢ (AUB). Similarly, B° < (AUB), and the desired inclusion follows. _ (©)From SC 5 and the fact that 3 isa closed set for any subset S, we see that AuBc AUBUAUB= UBS (@ Since ANB CANE, we have ANB ¢ (If x © AMB and r > 0, then choose some 0 <8 x. Consequently, x € BG@,7). and thus C(@,r) S BG,7) also holds. For a counterexample, consider X = (0, 1} with the discrete distance, and note that X is a complete metric space. Also, observe that B(0, 1) = {0}, while CO, 1) = (0,1). Problem 63. if A tsa nonempty subser of IR, then show that the set B= (ae A: There exists some ¢ > 0 with (a,a+e)NA=G} is at-most countable. Solution. Foreacha € B pickarational numberr, > a sothat(a, ra) A=. We claim that if a, b € B satisfy a # b, then ry ry. Indeed, if a (fp) be a fiancton. Show that f is continuous fend only f 71(B°) & [J-!CBNT for every subset B of Y. Solution. Assume f continuous, and BS Y. Since B® is open, the set 77(B°) is likewise open. Thus, in view of BY ¢ B, we have P78) = [FB s Lap. In the opposite direction, assume that the condition is satisfied. If BC ¥ is oper (ie, if B = B° holds), then [ray ¢ f= “ey c [f-ayy? shows that ~!(B) is open, Therefore, f is co NOUS. Problem 6.5. Show that the boundary of a closed or open set ina metric space is nowhere dense, Is this statement true for an arbitrary subset? Solution. Since 3A = 3A‘ Thus, ANA holds, we can assume that A is closed. Gay = ANF = (Ane) =aney can = an (ay ad. Since 9A is closed, this shows that 3A is nowhere dense ‘An altemate proof goes as follows: Ifx e (0), then there exists some r > 0 such that B(x, r) © 9A = ANF ¢ A, This implies B(x, 1) A® = @, contrary wxe a “The boundary of an arbitrary set need not be nowhere dense. An example: Let X =R withthe Euclidean distance, and let A = (the setof rational numbers) Note that 34 = R Problem 6.6. Show that the set of irrational numbers is not a countable union of closed subsets of R. Solution. Let / denote the set of all irrational numbers, and let {7,72 ...) be an enumeration of the rational numbers of8 Chapter 1: FUNDAMENTALS OF REAL ANALYSIS. Assume by way of contradiction that there exists a sequence of closed sets (A,) of R such that J = 2, Ay. Then r= (a.)u (Cra), and by the Baire Category Theorem (Theorem 6.18), we must have (A,)° # for some n. Thus, some A, contains an interval. However, since A, CJ holds and cach interval contains rational numbers, this is impossible, and the conclusion follows, Problem 6.7. Let(X, d)bea metric space. Show that if (x,) and (yq) are Cauchy sequences of X, then (d(xn, ¥n)} converges in R, Solution. Use the inequality [dGin. Ye) ~ Am Jn) S An Xm) + Fn, In) ‘Also, see the discussion before Theorem 6,19.) Problem 6.8. Show that in a metric space a Cauchy sequence converges if and only ifit has a convergent subsequence. Solution, Let (2,} be a Cauchy sequence in a metric space (Xd). If.x_—+ x holds in X, then every subsequence of {5} converges tox. For the converse, assume that there exists a subsequence (xz,) of (xq) such that x, > x holds in X. Let ¢ > 0. Choose rg such that d(x,,x) <€ and Ua. %m) <€ for n, m > no. Now, ifn > no, then ky =n > no, and so Gn, x) SU(m Xe,) +d (24.2) <€ +e = 2e. This shows that, im.x, = x holds in X. Problem 6.9. Prove that the closed interval (0, 1] isan uncountable set: &. by using Cantor's Theorem 6.14, and b, byusing Baire’s Theorem 6.17. Solution. (a) Assume by way of contradiction that (0, 1] is a countable set, say (0, 1} = (21,22,...). We consider (0, 1] equipped with the usual distance d(x, y) = [x — y| 50 that [0, 1} is @ complete metric space.‘Section 6: MIETRIC SPACES. * Subdivide (0, 1] into three closed subintervals (asin the construction of the Cantor se) of equal length, Remove from (0, 1] the middle open subinterval and consider the remaining two closed subintervals (here the subintervals [0, #] and [3.1] and then select one of them, say /), such that x, ¢ /;. Next, repeat this process with fin place of [0,1] and select a closed subinterval /a of fof length equal to one-third off, such tha x ¢ /. Inductively, assume that we have chosen fr closed intervals fi,» fy such that: Lo InGhaSeShoh, 2 meh fork =1, vn, and 3, the length ofeach fs [As above, there exists a closed subinterval /,41 off, of length equal to one-third fof fy such that put € Inst ‘Thus, there exists a sequence (J,} of closed subintevals of (0, 1} such that Ina © Ins a € fy and d(J,) = & for each n, By Theorem 6.14, we infer that 7%, 7, consists exactly of one point, But, since x, ¢ fy for each n, we see that AE, 1, =G, a contradiction. Hence, [0,1] must be uncountable ‘Again, assume by way of contradiction that (0, 1] = (x1. 25...) and again we consider{(0, 1] asacomplete metric space. If Ay = (tp) then each A, is closed ‘and has no interior points. However, Theorem 6.17 (or Theorem 6.18) applied to the equality (0, 1] = Uy An implies that some Ay must have an interior point which is impossible. This shows that (0, 1] cannot be countable, Problem 6.10. Let (71,72, ..] be an enumeration of all rational numbers in the interval (0, {] and for each x € [0,1] let Ay = {nm © IN: ty R by the formula f= # Show that f restricted tothe set of irrational numbers of (0, 1] is continuous. Solution. Fix an irrational number a ¢ [0, 1} and let > 0. Pick a natural number k such that 7%, 2 0. ‘We claim that if x € (0, 1] is an irrational number, then a — x| < 5 implies LF() — f(@)| < ¢ (which tells us that f is continuous when restricted to the irrational numbers). ‘To see this, let x €[0, 1] be an arbitrary irrational number satisfying |x — al <6. Let Jy denote the half-open subinterval of {0, 1] which is open at left and closed50 (Chapter 1: FUNDAMENTALS OF REAL ANALYSIS. at right having endpoints a and x. If B, = {n € IN: ry € Jz}, then note that Be Sk E+ 1 E+2,...} (hy?) and s0 [feo - f@|= T= Vk (0, 1} is constant, where the two point set (0, 1} is considered to bbe a metric space under the discrete metric b. Hina metric space (X,d) we have B.S A © X, then the set B is a ‘connected subset of (A, d) if and only ifB is a connected subset of (X, d) ©. If f(X,d) > WF, p)isa continuous function and A isa connected subset Of, then f(A) is a connected subset of ¥. Uf (Ailies 18 a family of connected subsets of a metric space such that Aer Ar ¥G. ther Use Aris likewise a connected set. If A isa subset of a metric space and a € A, then there exists a largest (with respect to inclusion) connected subset Cg of A that contains a. (The Connected set C. is called the component of a with respect 10 A.) £. I.a,b belong to a subset A of a metric space and Cq and Cy are the components of a and b in A, then either Ca = Cy or else Ce Cp = @. Hence, the identity A = yea Ca shows that A can be written asa disjoint union of connected sets. B.A nonempry subset of IR with at least ovo elements is a connected set if and only if iis an interval. Use this and the conclusion of (f)10 infer that every ‘open subset of R can be written as an at-most countable union of disjoint open intervals. Solution. (a) Let (X,d) be a connected space and let f:X —> (0,1) bea continuous function, Then the set A = f~!(0) is an open and closed subset of X. Since X is connected either A = ¢ (in which case f(x) = I holds for each x €X)or A=X (in which case f(£) =0 holds for each x € X). For the converse, assume that every continuous function from X into (0,1) is constant and let A be a closed and open subset of X. Then the function‘Section 6: METRIC SPACES st f:X o> IR defined by . ixeA; rom | tes is continuous (why?). By our hypothesis, | must bea constant function, and this implies that either A = or A= X, ie., X is a connected metric space. {b) It follows immediatay from (a). {@) Assume that f and A satisfy the stated properties and consider the contin- nous functions (A,d) + (f(A), e) > (0, 1). By (@), the continuous function ‘go f must be a constant function and from this, we see that g is also a constant function, By (a), (f(A), ) is a connected metric space. (a) Assume the family (Ai: i € 1) satisfies the stated properties. Put A= UserAi and let f:(A,4) —> (0, 1) be a continuous function, Then the function 7:4, 4)-—+(0, 1) 18. continuous Function and sof restricted to each A is constant. Since MyerAi # G, we see that f is constant on A, and so—by (@)—the set A isconnected. (@)Fix a € A and let A= BC A: B isconnected and @ B), Note that {a} € A and that \gexB # @. By (@), the set Co = UperB isa connected subset of A that satisfies the desired propertics. (If Cy Cy # @ then by (A) we infer that Cy U Cy is a connected set containing a. Hence, C, © Cy UCy © Ca. Similarly, Ca © Cy and so Ca = Co. (@) Let A bea connected subset of IR and let a, be A satisly @ < b. IF (0,1). Pick a, beT such that f(a) =0 and f(b) = 1; we can suppose that a Lx, defined by f() = rx, satisties |f()— (8)! < d(x, O|s—7|, and so f is (uniformly) continuous. From parts (g) and (c) of Problem 6.11, we see that Ly is a connected set. Now, use part (4) ofthe preceding problem and the identity R” = User Ln t0 infer that Ris itself 2 connected metric space. For the last part of the problem, let U and V be two open subsets of IR” such that K = UAV is aclosed set. Then K is both open and closed (and since X is a proper subset of R) it must be the empty set. Problem 6.13. Let C be a nonempty closed subset of R, Show that a function F:C — Ris continuous if and only if it can be extended to a continuous real- valued function on R. Solution. Let C be @ nonempty closed subset of R and let f:C > R bea function. If f can be extended to a continuous real-valued function on WR, then F:C = R is obviously continuous. For the converse, assume that f:C > R is a continuous function. Start by observing that the complement C* of C is an open set and so (by part (g) of Problem 6.11) C® ean be written as an at-most countable union of pairwi disjoint open intervals; say C* = Ujej(ai,b,), where I is at most countable. Since the open intervals ((q), bi): # € 1) are pairwise disjoint, i follows that all the endpoints a, and by belong to C. Hence, f(a;) and f(b,) are defined for each i. Now, extend the domain of 7 by defining the graph of the function f on the interval (a;, bs) to be the straight line segment joining the points (a;, f(aj)) and (by, (oD). In other words, foreach ay < x fla) is triviallySection 6: METRIC SPACES 3 ‘nue, Also, we shall assume that a is not one ofthe endpoints a; or by. For each 1 pick (the unigue) i, € J with aj, < &, < bj,. Note that in this case, we must have lima, = limb, =a (why?). From [ren — Fa] = [LF + REE ay — zoe [ae ]- 7 S[Fb,) ~ Fla.) fi) (4a, — If(@)- fla)| =0, swe sce that lim f(t,) = f(@). A similar conclusion holds tru if a is one of the endpoints a; of by. This shows that f is continuous at a, as claimed. Foran altemate proof see Problem 10.11. Problem 6.14. Show that a metvic space is a Baire space if and only if the complement of every meager set is dense. Solution. Let X be a metric space. Assume firs that X isa Baire space and Tet A be a meager set, Pick a sequence (Ag) of nowhere dense sets such tha =U, Aa. To show that A® is dense, i sties to show that V1 A® # o for cach nonempty open set V. To see this, lt V bea nonempty open se and assume by way of contradiction that V 91 AS =. This implies ¥ SA, and so veUvndn m= Hence, is @ nonempty open meager set, a contradiction, Hence, Ai «dense = For the converse, assume thatthe complement of every meager set is dense, and let V be an open meager set. Then ¥* is dese, So, if V is nonempty, then Vin Ve oo eich is impossible, Thus, te empty sti the only open meager se and Hence, X isa Baie space Problem 6.15. A .subset of a metric space fs called co-meager ifits complement isa meager set. For a subset A of a Baire space show that: Ais co-meager if and only ifit contains a dense Ge-set. b. Als meager ifand only ifit is contained in an F,-set whose complement is dense. Solution. Notice that if A is a nowhere dense set in a metric space X, then from Lemma 6.8 we see that b= Gy =A =Ey Chapter: INDAMENTALS OF REAL ANALYSI ‘This implies that a subset A is nowhere dense if and only if the open set (A) is dense Now, assume that X is a Baire space and let A be a subset of X. (@) Suppose first that A isa co-meager set. Then there exists a sequence (Ay) of nowhere dense sets such that A = (\%>., An)°. This implies o) By the above discussion, each set (Hp) is an open dense set, and since X is a Baire space, the Gy-set E = (, (Ap)® is also dense (see Theorem 6.16). Now, a glance at (») shows that ES A FFor the converse, assume that A contains a dense Gi-set B,ie., BS A. So, there exists « sequence (Vf open sets such that B = (2, Vy. From B & Vq, ‘we see that each Vj is also dense. This implies [WFP = Ea) I = Fa = =, and s0 each (Vs) is nowhere dense closed set. Now, use the inclusion Ac B= vy to conclude that A* is a meager set, ie. A is a co-meager set, () Assume first that A is a meager set, ie., A® is a co-meager set. By part (®), there exists adense Gy-set E such that E S A°. This implies A C E*, where now E isan F-set whose complement (E*)¥ = E is dense. For the converse, assume that A & F holds, where F is an Fe-set with dense complement. I follows that F® C A®, where now F° is a dense Gs-set. By part (@), A® is co-meager set, which means that A is a meager set. 7. COMPACTNESS IN METRIC SPACES Problem 7.1. Let f:(X,d) -» (¥, p) be a function. Show that f is continuous and only if f restricted to the compact subsets of X is continuous. Solution, Assume that f is continuous on every compact set. Let xq > x. ‘Then the set A = (21, x2, ..] U(x) is compact (note that every open cover of A can be reduced to a finite cover), and x, > x holds in A. Since f restricted to A is continuous, lim (x4) = f(x) holds, which shows that f is continuous,‘Section 7: COMPACTNESS IN MICTRIC SPACES 5s Problem 7.2. A metric space is said to be separable if it contains a countable subset that is dense in the space. Show that every compact space (X,d) is sepa rable. Solution. Foreach n choosea finitesubset F, of X suchthat X= yep, BC. Let F = US, Fay and note that F is at-most countable. ‘Now, let 2 X and r > 0. Picksome n with ! S, from X into PON) has been established that is clearly one-to-one. Consequently, ‘card X < card PON) = ¢. (See also Problem 5.6.) Problem 7.4, Let (Kj.di)s-+++(Xndy) Be arbitrary metric spaces, and let Ke Ky oe x Xp PES (hye tn) ard y= (Pty oe Yale define Silden yak)! ‘ Dye, 9) = Yo dnlins Ym) and D(x, y= Show that Dy and Dz are distances on X. Show that Dy is equivalent ¢o Da. Show that (X, D,) is complete ifand onty if each (X;, di) 1s complete ‘Show that (X, D1) is compact if and only ifeach (Xi, di) is compact. Solution. (a) Routine. (b) Use the inequalities 1 Dix, y) ¢ Dale, y) S nde, 9). (© Assume that each Xq(m = 1y--.5n) is a complete metric space. Let {a} be a Dy-Cauchy sequence of X, where ry = (xf,...,x4). Cleary, each ik} is a Cauchy sequence of X,, and thus there exists 2y € Xq such that Timesoo g(x,y) = 0. Hence, if x = (xys-+.y%q) € X, then we have Timysren Dy (2a, -t) = O, so thatthe metrie space X is Dy-complete86 ‘Chapter 1: FUNDAMENTALS OF REAL ANALYSIS. Now, let X be Di-complete, Fix an element (yy,..., y.) € X. Let (x4) be a Cauchy sequence of Xn. If x; € X is the element whose j® component equals 1 for jm and equals xf if j =m, then (x4) is a Cauchy sequence of X. If € X is its limit, them itis easy to see that limy_sa0 dm (4, %,,) = 0, 60 that each Xp, is complete. (2) Assume first that each X,, is compact. Then following the proof of the second part of Theorem 7.4, we can see thatevery sequence of X has a convergent subsequence, and so X must be a compact metric space. ‘On the other hand, if X is acompact metric space, then the function SniX —> Xp defined by fn (iy «+» %n) = Xm, is continuous and onto for each 1 Smen Hence, by Theorem 75, each Xp = fn(X) is compact, Problem 75. Let (Xn, d,)} be a sequence of metric spaces, and let X = IP. Xn. For each x = (x5) and y = (yn) in X, define SL dab Jn) N= De Teno ft 8. Show thar dis. distance on X. b. Show thar Qf, d) isa complete metric space ifand only if each (Kn. dy) is complete. © Show that (X, 4) is a compact metric space if and only if each (Xr, dy) is compact, Solution. (@) Note firstthatif d is distance on aset X,then atx, 9) = 2, is likewise a distance on X, which is equivalent to d. From this observation it easily follows that SL dali yn) dey Le T#d,Gn. yn) isadistance on X = []",X,. (6) Let {x} bea sequence of X, where x! = (xt, x4,...). The proof follows from the following two properties (whose verifications are straightforward), 1. xh —+ x holds in X if and only if of + xy holds in Xj foreach fs and 2 (4) isa Cauchy sequence in X if and ony if {xt} is a Cauchy sequence in X; foreach i. (©) Assume that (X, ) is acompactmetric space, Then the function fi: X —> Xi, defined by f(x) = x, for each x = (x1,13, ...) € X, is continuous and onto. By Theorem 7.5, cach X; is a compact metric space.‘Section 7: COMPACTNESS IN METRIC SPACES 37 For the converse, assume that each X; is a compact metric space. By (b), X is a complete metric space, and so by Theorem 7.8 it suffices to show that X is totally bounded, To this end, let © > 0. Choose » such that 2-" < e, and note that ee cel Plt = LO Ted defines a distance on [[fuy Xi. It should be clear that py is equivalent to the distances of the preceding problem, and ([Yfy X,, Px) isa compact metric space. Choose a fnite subset F of [fay X; such that the p,-balls with centers at the points of F and radii ¢ cover [fy X;. Next, extend each x € F toan clement of X Ge, add to each x € F arbitrary components pet, n42y--.)- Now, if Y= (nen o.-) € X, then pick some x € F with (x,y) <&, and note that SL dix ys a dea atin Do oe Petes cetrt sete aan ‘(Sis We 26. Thus, X = UcerB(s, 22) holds, and therefore, X is totally bounded. Problem 7.6. A family of set F is said to have the finite intersection property if every finite intersection of sets of F is nonempty. Show that a metric space is compact if and only if every family of closed sets with the finite intersection ‘property has a nonempty intersection. Solution, Let X becompact, and et (Ai: f € 1) bea family of closed sets with the finite intersection property. If (jer A: =, then X = Uj_r AF holds, and by the compactness of X, there exist f,.-.yi4 € 4 such that X= U2, Aj, Thus, \1-1 As, = @, contrary to our hypothesis. Hence, es Ai #@. ‘For the converse, assume that every family of closed sets with the finite in- tersection property has a nonempty intersection. Let X = Uj X is called a ‘contraction if there exists some 0 < a < 1 such that d(f(2), F(Q)) < ad(x. y) forall x, y € X;« is called a contraction constant. ‘Show that every contraction f on a complete metric space (X.d) has a unique faxed point; that is, show that there exists a unique point x € X such that f(z) = Solution, Note first that if f(x) =x and f()) = y hold, then the inequality x, y) = d( F(a), FO) = ad(x, y) easily implies that d(x, y) = 0, and so x = y. Thats, f has at-most one fixed point. To see that f has a fixed point, choose some a € X, and then define the sequence (4) inductively by xis and xn41 = fq) for m= 1,2, From our condition, it follows that Gnas He) = d(F CH), Fra) Sodas Feat) holds for nm = 2,3,.... Thus, as in Problem 4.15, we can show that {xy} is Cauchy sequence. Since X is complete, {1} is a convergent sequence. Let x = lim.x,. Now, by observing that 7 is (uniformly) continuous, we obtain that = i, sees = Jim, fm) Ww, and so x is a (unique) fixed point for f.Section 7: COMPACTNESS IN METRIC SPACES 2 Problem 7.9. A property of a metric space is called a topological property ifit is preserved in a homeomorphic metric space. ‘a. Show that compactness isa topological property. b. Show that completeness, boundedness, and total boundedness are not topo- logical properties. Solution, (a) [follows from Theorem 7.5. {(b) Consider (0, 1} and (1,00) as metric spaces under the usual Euclidean distance dx, ») = | — y}- Clearly, 0,1] is not complete but it is bounded and totally bounded, Also, [1, 00) is complete (because it is a closed subset of IR), but is neither bounded nor totally bounded. On the other hand, f:(0, 1} -» (1,00), defined by f(x) = 4, 8a homeomorphism, and the claims in () follow. Problem 7.10. Let(X,d)beametric space. Define the distance of vo nonemply subsets A and B of X by dA, B) = infld(x, ): x € A and y = Bl a. Give an example of two closed sets A and B of some metric space with ANB = Gand such that d(A, B) = 0. b. FANB =G, Ais closed, and B is compact (and, of course, both are nonempty), then show that d(A, B) > 0. Solution. (a) Let X = IR? with the Buclidean distance, and consider the closed subsets of X Aa{(d x2t) and B= {(x,0): x= 1}. Note that ANB =, while d(A, B) =0. (b) Let A and B be as stated in the problem. If d(A, B) = 0, then pick two sequences {%,) © A and (yx) © B with d(te, yx) —> 0. Since B is compact, by passing to a subsequence (if necessary), we can assume that y, —> y holds for some y ¢ B. The inequality Ginn Y) SMe Ye) + a YD shows that dtp, y) —+ 0. Since A is closed, y € A, and hence ANB # 3, contrary to our hypothesis. Therefore, d(A, B) > 0 must hold. Problem 7.11. Let (X,d) be a compact metric space and f: X -+ X an isome: try; that is, d( f(x). f()) = dl, y) holds for all x, y € X. Then show that f is onto. Does the conclusion remain true if X is not assumed to be compact?° ‘Chapter 1: FUNDAMENTALS OF REAL ANALYSIS Solution. Let y € X. Define the sequence (x,) of F(X) by = FQ) and xn) = fq) for n=1,2,...- Note that d(x, Xnep) = dQ), 4p) holds for all n and all p. Since f(X) is ‘compact, (1) must have a limit point in f(X). Let a be a limit point of {x4} Now, let ¢ > 0, Pick n > I and p such that d(x, a) 0, and so d(y, f(X)) = 0. Thus, y ¢ #(X) = f(X), so that SO) = X holds. IE X is not supposed to be compact, then the conclusion is no longer true. A counterexample: Take X = IN with d(x, m) = nm! and consider the function SAN IN defined by f(x) =n-+1. Problem 7.12. Show that a metric space X is compact if and only if every con- tinuous real-valued function on X attains its maximum value. Solution. Let (X,d) be a metric space. Assume that X is compact and that f:X — RR isa continuous function, By Theorem 7.5, we know that f(X) is a compact subset of R, and so (by Theorem 7.4) F(X) is closed and bounded. The maximum of F(X) is the maximum value of f on X, For the converse, assume that every coniinuous real-valued function on X attains 2 maximum value. Cleary, every continuous real-valued function on X autains also a minimum valve. ‘We shall establish first that X is a complete metric space. Let (X.d) denote the completion of (X, d) and let ¥ © X. The function f:X —+ R, defined by FCs) = d(&, x), satisfies inf f(x): x € X} =O. So, there exists some ay € X satisfying (xq) = d(2, xo) = O. It follows that # = xp € X and so R= X. This means that X isa complete metric space. Next, we shal] show that X is totally bounded. To establish this, assume by way of contradiction that X is not totally bounded, Then an easy inductive argument shows that there exist some # > 0 anda sequence {%,) of X such that Gn, Xn) & 3e holds for n + m. For each n consider the nonempty closed set Cy = (By, OF (x EX: dx, x) Ze), and then define the function j,:X + R by Fala) = d(x, Cy) = inflde, ye y € ChSection 7: COMPACTNESS IN METRIC SPACES So, fa isa bounded function, fq(s) = 0 holds foreach x € Cy and fin) > 0 Muitipiying by a constant cy, we can assume that supl a(x): x € X} > 1 holds foreach 1. Now, define the function f:X —+ R by | fil), if x € BGs) ro={f if x € UR: Bone), and weclaimthat j isacontinuous function. Clearly, f is continuous atthe points of the balls BGiy, 8). IF x0 ¢ Uy Bes 6) note that B(x, £)0 Bla, 6) # hholds for at-most one 1 (why). If B(xo, $) 9 BG.) = @ for each m, then FG) =0 for each x in B(xp, §), and sof is continuous at xp. Thus, we can ‘assume that B(zp, §) Bq, 6) # G for some mn. We distinguish two cases. CASE I: dx, x0) > & In this case, there exists some 0 xo; we can assume that ze belongs to B(to, §) for each k. Note that if zz ¢ Bln.) then f(z«) = 0. On the other hand, if 2¢ € BGsn,£), then OS fl2u) = Onde, Ce) S CuCzes Xo). Thus, 0S f(z) S cud(2e, x0) holds for each &. In view of fim dee) = wwe see that lim f(21) ‘Gzo) and so f is continuous at xp in this case too. ‘To contradict our hypothesis, note that f does not attain a maximum value. ‘Thus, X mustalso be totally bounded, By Theorem 7.8, wesee that X is acompact metric space. Problem 743, This exercise presents a converse of Theorem 7.7. Assume that (Xd) is a metric space such that every real-valued continuous furction on X is uniformly continuous. a Show that X is a complete metric space. b. Give an example of a noncompact metric space with the above property. ©. IFX has a fite number of isolated points (an element a ¢ X is said t0 be an isolated point whenever there exists some positive r > 0 such that B(a,r) A(X \ {a}) = @), then show that X is a compact metric space.@ (Chapter 1: FUNDAMENTALS OF REAL ANALYSIS. Solution. Let (X,d) be a metric space such that every continuous real-valued function on X is uniformly continuous. (@)IE 2 € & (the completion of X) isan element that does not belong to X, then the function f:X — IR defined by f(s) = a5, x ¢ X, is a continuous real- valued function on X that fails to be uniformly continuous (why?),a contradiction. Hence, X = X holds, which means that X is a complete metric space. (byLet x -] equipped with the discrete distance d. Then every sets ‘open and so every real-valued function f on X iscontinuous. Since d(x, y) < 1 implies x = y (and so f(x) ~ f(Q) = 0), we see that every real-valued function on X is uniformly continuous. Now, note that X is not a compact metric space. {) In view of (a), we need to establish that X is totally bounded. “To this fend, assume that X is not totally bounded. Then, there exist some © > 0 and a sequence of elements (x4} of X such that d(xy,xq) > 3e for n zm. Brom ‘our hypothesis, we can suppose that each x, is an accumulation point of X. For ‘each n pick an element y, such that 0 < d(xp, ya) < £ and let ry = de, Yo) Pat Cy le eX: dex) EM) and define the functions f, and fas in the solution of Problem 7.12 (the open ball Ba.) is now replaced by B(x,,r_))- Then f is a continuous function and satisfies f(y.) = 0 for each nm. Pick 2 € BQ.) such that f(%q) >”, ‘and note that [F0n)- FG] > 2 and Jim dlyp, 20) = 0. ‘This shows that the continuous function f is not uniformly continuous, contrary to our hypothesis. Hence, X is totally bounded, as desired, Problem 7.14. Consider a function f:(X,d) —» (¥, p) between two metric spaces. The graph G of f is the subset of X x ¥ defined by G=(@eXx¥: y= sO). IF, p) isa compact metric space, then show that f is continuous if and only if G is a closed subset of X x ¥, where X x ¥ is considered 10 be a metric space under the distance D((x, y), (u, v}) = d(x, u) + p(y, v); see Problem 7.4. Does the result hold true if (Y,) is not assumed t0 be compact? Solution. Observe that an arbitrary sequence {(x», ¥n)} of X x ¥_ satisfies (as Ye) > Oy) in X x ¥ if'and only if x > x and ye > y both hold.Section 7: COMPACTNESS BN METRIC SPACES 6 ‘Assume (Y, p) compactand G closed, If f isnot continuous, then there exists sequence {x,) of X andsome ¢ > 0 such that x, —> x and p{ f(a), f() = ¢ forall n (why?). Since (Y, p) is compact, by passing to a subsequence, we can assume that f()q) —> 9 holds in ¥. Now, observe that (tq, /(%)) € G holds foreach n and (xq, f(t,)) > (x,y) holds in X x Y. Since G is closed, it follows that (x, 9) € G and so y = f(x). This implies (Fn). £00) > (FCO, flx)) = 0, which contradicts p( f(s). f(2)) 2 © for all n, Hence, fis a continuous function, If (Y, p) isnot compact, then 2 function with closed graph need not be contin- uous. Foran example, consider the function f:R —> R defined by galt ifx#0, won () E228 Problem 7.15, A cover (Vilier ofa set X is said 10 bea pointwise finite cover whenever each x € X belongs atmos 0. fnite number ofthe V; Shove that a metric space is compact if and only if every pointwise finite open cover ofthe space contains a finite subcover. Solution. Clearly, if X is compact, then every pointwise finite open cover of 1X contains «finite subcover. For the converse, assume that every pointwise finite ‘open cover of X contains finite subcover. To establish thatthe metic space X compact, it suffices to show that every sequence in X contains a convergence subsequence. Let {&,} be a sequence in X. We can suppose (why?) that the sequence consists of distinct elements. Suppose by way of contradiction that [,) has no convergence subsequence. Then x; isnotinthe closure of the set (xy: #1) and thus, there exists an open ball Vj = B(x), 4)) about x; with radius 0 <8) <1 and satisfying x» ¢ Vi for all n # I. Also, x is not in the closure of the set (cg: #2} and thus, there exists an open ball V2 = B(x, 62) about x2 with radius 0 < 8; < } and such that xy ¢ Ve forall n-# 2, Procecding inductively, wwe see that for each k there exists an open ball Vi = B(e, és) with radius Oe by < 4 satisfying x, ¢ Vi forall n #E. Since the set F = [x,%2,..} contains no convergent subsequences, the set F mst contain all ofits closure points. Thus, F isa closed set, and fence, the set G =X \ F isan openset. Then, the collection C = (G,Vi, Va...) isan ‘open cover of X. In fact, the collection C is a pointwise finite open cover of X ‘because fa point x belongs to an infinite number of sets in C, then. x belongs 10 an infinite number of the sets V,.. However, this would imply that a subsequence6 (Chapter 1: FUNDAMENTALS OF REAL ANALYSIS: of (39) converges othe point x. Since the sequence (3) contains no convergent subsequences, we infer that C is a pointwise fnite open cover. Therefore, C contains a finite subcover of X, say Vj,...,¥g,G. Since G does not intersect (x1, x2, ..), it follows that (x;,22,...) © UM, Vj. However, this contradicts the fact x, ¢ Vj for n # k. Conclusion: The sequence {x,} must have « convergent subsequence—and hence, the metric space X is compact.CHAPTER 2_ TOPOLOGY AND CONTINUITY 8, TOPOLOGICAL SPACES Problem 8.1. For any subset A of a topological space show the following: a At = (APY. b aaa Aer ce (A\A =, Solution. (a) Note that x €A® s=> there exists a neighborhood V of x with VC A = there exists a neighborhood V of x with VMAS = em x eH me re (FY. (b) Using a), we see that 0A = ANA =A \ (HF (c) If x € (A\ A°Y, then for some open set V we have A\ a xeVCA\ASA ‘This implies x € A \ A® and x € A®, acontradiction. Hence, (A \ A°)* = ¢. Problem 8.2. if A and B are two arbitrary subsets of a topological space, then show the following: a AUB=AUB. b (AUBY = AUB" Solution. (a) See Problem 6.1 (0) Clearly, A CB implies A’ ¢ BY, andso 4’ UB’ ¢ (AU BY. For the reverse inclusion, let x € (AU BY. If x ¢ A’U B’, then there exist two neighborhoods —_“ (Chapter 2: TOPOLOGY AND CONTINUTTY V and W of x such that AMV \ 2) =B0W \ ED=d. Now, note thatthe neighborhood U = V AW of the point x satisfies (AUB)NU \ [xp= proving that x ¢ (AU BY, a contradiction. Problem 8.3. If A is an arbitrary subset of a Hausdorff topological space, then show that its derived set A’ is a closed set, Solution, Let A be an arbitrary subset of a Hausdorff topological space X. We shall establish that (4’)¢ isan open set (and this will guarantee that A’ isa closed set). To this end, let x € (A, ie, let x g A". This means that there exists @ neighborhood V of x such that VA(A\ Gh =o. @ We claim that V ¢ (A holds. To see this, let y ¢ V with y xx. Since X 2 Hausdorff topological space, there exist neighborhoods U and W of y and 2, respectively, such that UW = @. Now, note that V "1 U isa neighborhood of y with x ¢ VU and so from (x), we see that (V NU)NA = @. The later shows that y ¢ A’. Hence, V & (A'S holds proving that every point of (4’)® is ‘an interior point, as desired, Problem 84. LerX = IR, and lett be the topology on X defined in Example 8 4. In other words, A € « if and only if for each x € A there exist € > 0 and an ‘at-most countable set B (both depending on x) such that(x—e,x-+e)\ BS A. a. Show that r is a topology on X. b. Verify that 0 € ©. 1). c. Show that there is no sequence {xq} of 1) with lim xy = 0. Solution, (a) Straightforward, (b) Since for each ¢ > 0 and each countable set B the set (—e, ) \ B is uncountable, we must have ((—e, £) \ B) (0,1) # @. This easily implies that 0e GD. (DIE (44) is a sequence of (0, 1),then V = (1,1) \ tai, x2...) is ancigh- bothood of zero, and 2x, ¢ V forall n. This shows that no sequence of (0, 1) can converge to 0. Problem 8.5. If A is a dense subset of a topological space, then show that OS ANG holds for every open set O. Generalize this conclusion as follows: If Aisopen, then ANB © ATB for each set B.Section 8: TOPOLOGICAL SPACES o Solution. Let x € © and let V be a neighborhood of x. Since © is open, V0 is aneighbothood of +, and so the denseness of A implies VAAN) =UNO)NAFD, which means that x ¢ AAO. For the general case, assume A is an open set and letx ¢ ANB. If V isa neighborhood of x, then V 1A is also a neighborhood of x. Since x ¢ B, it follows that V0.(A1B) = (V9 A)NB # @. This shows that x ¢ ANB, and hence, ANB ANB. Problem 8.6. If {O,)ier is an open cover for @ topological space X, then show that a subset A of X is closed if and only if A 0 O; is closed in O; for each i € I (where O; is considered equipped with the relative topology). Solution. If A is closed, then clearly 4; is closed in ©; for each i. For the converse, assume that A” Q; is closed in O; for each i. Put Vi =O \ ANZ =ONA, and note that—by our hypothesis—each ¥; is open in 0). Since each O; is an ‘open subset of X, it follows that each Vj is likewise an open subset of X. Now, rote that x\ d= (Yo) \4=Ueo\a=Us is an open subset of X, and so A is a closed set. Problem 8.7. If (Sr) is a Hausdorff topological space, then show the follow- ing: ‘a. Bvery finite subset of X is closed. b. Every sequence of X converges to at-most one point Solution, (a) Let A=(x) be a one-point set. If y €A, then (since X is a Hausdorff space) there exists a neighborhood V of y with x ¢ V,andso VS A‘. ‘Thus, A® is open, and hence A, is closed, Now, observe that every finite set is finite union of one-point sets. (b) If x # y, then there exist neighborhoods V, and ¥; of x and y respec tively, such that VV, = @. Now, a sequence of X cannot converge to x and 1 atthe same time simply because its terms cannot be eventually in both V_ and V5.“ (Chapter 2: TOPOLOGY AND CONTINUTTY Problem 8.8, Fora function f:(X, 1) —> (¥, 1) show the following: a. If vis the discrete topology, then f is continuous. b. If tis the indiscrete topology and 7, is a Hausdorff topology, then f is continuous if and only if f is a constant function, Solution, (a) Note that every subset of X is open. Thus, f~!(A) is an open set for every subset A of ¥, and so fis continuous. () Recall thatthe indiscrete topology is the topology ¥ = (G,X). If f isa constant function, then f~"(A) is either @ or X, and so f is continuous. For the converse, let f be a continuous function. If for some x, y € X we have £2) # 0), then there exists a neighborhood V of f(x) such that f(y) ¢ V. Now note that f-1(V) is neither equal to ¢ nor equal to X, and so f-X(V) is not open, contradiction. Thus, must be a constant function. Problem 8.9. Let j and g be nwo continuous functions from (Xt) into a Haus- ddorff topological space (¥, %). Assume that there exists a dense subset A of X such that f(x) = g(x) for allx € A. Show that f(x) = g(x) holds for all x € X Solution. Suppose that for some x € X we have f(x) # g(x). Pick a neigh- borhood V of f(x) and another W of g(x) such that VW = @. Since F-V)Ng-(W) is aneighborhood of x and A is dense in X, there exists some ¥ € FW) Ag) OA, Now, note that f(y) = (9) € VW =o must hold, which is absurd, Thus, f(x) = g(x) holds for each x € X, Problem 8.10. Ler f:(X,t) —r (Y, t1) bea function. Show thar f is continuous ‘ifand only if f-"(B°) ¢ [f-*(B)P? holds for every subser B of Y. Solution. Repeat the solution of Problem 6.4 Problem 8.11. If f:(%, 2) — (¥, 11) and g:(¥, 11) > (2, 12) are continuous Junctions, show that their composition g o f:(X,t) > (Z, %) is also continuous. Solution, Use the identity (go )"'(V) = f-!(e-*(V)). (See Problem 1.8.) Problem 8.12. Let X be a topological space, let a € X, and let N denote the collection ofall neighborhoods axa. The oslllation of a function f:X —» R at the point a is the extended non-negative real number ota) = ing | sup (72) ~ Foil).‘Section: TOPOLOGICAL SPACES. “ Establish the following properties regarding the oscillation: a. The function fis continuous at a ifand only if f(a) = b. IF X isan open interval of R and f:X —> R is a monotone function, then eog{a) =| lim, FG) ~ Jim FG] Solution. (a) Assume that f is continuous at a. Fix ¢ > 0. Then there exists some W € Ny (ie., some neighborhood W of a) such that x ¢ W implies [fl@) — fO@)| <«. So, ifx, y € W, then 170) = FOS 1F@)~ F@I+ fla) ~ FOI sete =e, ‘and thus Os er(a)e sup |f(2)- 09] <2 for each € > 0. This implies (a) For the converse, assume «/(a) = 0. Lot e > 0. Then from the definition ‘of the oscillation, we see that there exists some neighborhood V of a such that ups yev [f(2) — F(9)| < ¢. In particular, we have |f(x) ~ f(@)| < € for all x € V, and this shows that f is continuous at a. (b) We can assume that f is an inereasing function. Note that we can consider neighborhoods of a of the form (c, d) with a € (c,d). Consider first a neigh- bothood (c,d) of a and assume that x, 9 € (c, d) satisfy x 0 is given, then there exists some & > 0 such that the ‘open interval J = (a ~ 4,4 +4) satisfies J & X and 0,2 < sup 179 ~ F091 < [im, 70 Jim Fo] +e se) This implies 0a) < limo FO — -2- £(0), and so enfla) = lim, FQ) ~ him £0 holds tru. Problem 8.13. Show thata finite union of nowhere dense sets is againa nowhere dense set, Is this statement true for a countable union of nowhere dense sets?0 ‘Chapter 2: TOPOLOGY AND CONTINUITY Solution. Let 4 and B betwonowheredensesets. Using theidentity $° = 5 (see Problem 8.1), we have ony = Guay = Aus (Aen Bye (a 8 G@yuG@)y =us=6. ‘An easy induction argument can now complete the proof. ‘The countable union of nowhere dense sets need not be nowhere dense, An ‘example: Take X = IR, and let Ey = (ra), where (ri...) is an enumeration of the rational numbers. Clearly, each £, is nowhere dense, while US Ea = r1.ro..24) isnot nowhere dense. Problem 8.14. Show that the boundary of an open or closed set is nowhere dense. Solution. Repeat the solution of Problem 6.5. Problem 8.15. Let f:(X, +) > R, and let D be the set of all points of X where J is discontinuous. If D* is dense in X, then show that D is a meager set. Solution, From D® = X, itfollows thar D® = (BE)* = g. Now the proof can be completed by observing that D is an F,-set (Theorem 8.10), Problem 8.16. Show that there is no function f:]R —> R having the irrational mumbers as the set ofits discontinuities. Solution, Let / denote the set of all irrational numbers of R. If J is the set of discontinuities of «function f:1R —> R, then (by Theorem 8.10) f isan Fy-set. However, this is impossible by Problem 6.6, Problem 8.17. Show that every closed subset of a metric space is a Gy-set and every open set is an Fe-set. Solution. Let A be a nonempty closed subset of a meitic space X. Then the function f:X —> R, defined by Fox) is continuous (see the proof of Lemma 10.4) and satisfies Soy =f AC oe) Arch a). de, A) infld(x, y): y € A},‘Section 8: TOPOLOGICAL SPACES n {See the discussion at the end of Section 6 of the text.) Thus, A is a Gy-set. By ‘Theorem 8.9 every open set is an F,-set Problem 8.18. Ler B be a collection of open sets in a topological space (X, x) {for each x inan arbitrary open set V there exists some B € Bwithx €B SV, then B is called a base fort. In general, a collection B of subsets of a nonempty set X is said to be a base if 1. User 8 =X and 2. for every pair A,B € Band x € ANB, there exists some C € B with xeCCANB. Show that if B isa base for a set X. then the collection V CX: Vx eV there exists Be Bwith xe BV) isa topology on X having Bas a base. Solution. Obviously, B © t holds. Clearly, @ € t, and from condition (1) it follows that X € t. Also, it should be clear that r is closed under arbitrary Now, let V,W et and x ¢ VW. Choose two ses A, B € B with x € ACV and x © B CW, By condition (2), there exists some C ¢ B with TEC CANBCV AW, thatis, VOW ex. Thus, r isa topology. ‘The verification that B is a base for r is straightforward Problem 8.19. Let (X, x) be a topological space, and let B be a base for the topology t (see the preceding exercise for the definition). Show that there exists a dense subset A of X such that card A < card B. Solution, If B ¢ B and B # G, then fix some xg ¢ B and consider the set A= (xe: BEB \ (9). We claim thet: 1. A isdense in X, and 2. card A = cardB. To sec (1) let V be anonempty open set. If x € V, then there exists some B eB with x ¢B CV. Itfollows that xp ¢ V, and so VA #¢. This shows that A is dense in X For (2) note that the function f:B \ {¢} —-» A, defined by /(B) = xa, is conto. By the Axiom of Choice there exists a subset C of B such that C9 f-!(x)) ‘consists precisely of one point foreach x € A. Then f:C — A is one-to-one and onto, proving that card A = card < card Bn (Chapter 2: TOPOLOGY AND CONTINUITY Problem 8.20. Let f:X — ¥ bea function. if isatopologyon X, then the quo tient topology ty determined by f onY isdefinedby cy =(OCY: f-(O)er} X,) > Ys ty) is a, Show shar ty is indeed a topology on ¥ and that continuous. b. if: (Yt) + ©, e1)isafunetion, then show thatthe composition function 0 f:(X, 0) > (Z, 1) is continuous if and only if gis continuous. ©. Assume that f:X —> ¥ is onto and that c* isa topology on Y such that 20%, ©) + (Y, 2") is anopen mapping (Le, it carries open set of X onto ‘open sets of ¥ and continuous. Show that c* = ty. Solution. a, (1) Since f"(G) = et and f-(Y) = X et, we see that BY ery. Q)If V, W e xy, then the identity f-1(V MW) = f-MV)N f-"W) implies that VINW € tp. G)IE (Vj: (© 1) isa family of ty, then in view of the identity f-(U¥i) = Us-'W, we see that UV; € ty. b. Assume go iscontinuous. If V isan open subsetof Z,then f-1(g~"(V)) = (go f"W) ex shows that g-'(V) € ry. That is, ¢ is continuous. ¢. Since f is continuous, its easy to see that r* C ty holds. On the other hand, let Ve ay, Then f1(V) € x, and moreover, since f isan open mapping and ‘onto, we have V = f(f-"()) € r* (see Problem 1.7). That is, ty © t* also holds, and so ty = 1*. Problem 8.21. This exercise presents an example of a compact set whose closure isnot compact. Start by considering the interval (0, 1] with the topology t gener ated by the metric d(x, y) = (xy. It should be clear that ((0, 1}, 2)i8@ compact topological space. Next, put X = [0, 1] UIN = [0, 1} U(2,3,4,..-} and define ri =TU(I0, UA: ACN). a. Show that x” is a non-Hausdorff topology on X and that x* induces ton (0.11, Show that (X, =*) is not a compact topological space. Show that (0, 1] is @ compact subset of (X.t"). Show that (0, 1] is dense in X (and hence, its closure is not compact. Why doesn't this contradict Theorem 8.12(1)? Solution. a. (1) Clearly, 4, X € e*. (Q)Let V, W et, Then we have the following cases: CASEL V, W er. Inthis case, VOW et Sr". CASEILV er and W ¢ x (andvice versa), Note that VW = Ver Sr",‘Section 9: CONTINUOUS REAL-VALUED FUNCTIONS, 8 CASEI V ¢r and W ¢r. Inthis case, wehave YW = (0, 1]UA for some ACN, That is, VW €r* (3) Let {Viz f € J) be a family of c*. If Vj € + holds for each i, then clearly Uv; eC c* holds. On the other hand, if some V; is of the form {0, 1]U A, then JV; is of the same type, and hence, it belongs to 1° ‘Thus, x* is. topology on X that induces t on [0, 1) b. The cover X = USa({0, 11U (n}) cannot be reduced toa finite cover. «. Since ({0, 1), £) i# 8 compact topological space and r* induces t on {0, 1} ie follows that (0, 1} isa compact subset of X. 4. If x © X \ (0,1), then every neighborhood V of x is of the form V = {0,1]UA forsome A CN. Thus, V.1{0, 1} = [0,1] # @ holds for every neighborhood ¥ of x. Therefore, 10,1] = X holds. c. This does not contradict Theorem 8.12(1) because (X, t*) is not a Hausdorft topological space. Problem 8.22. A topological space (X, x) is said to be connected ifa subset of X that is simultaneousty closed and open (called a clopen set) is either empty or else equal 10 X. a. Shove that (X, £) is connected if and only if he only continuous functions {fros(X, t) into {0,1} (with the diserete topology) are the constant ones. b Let fil, 2) > (W,2") be onto and continuous. If (X, x) is connected, then show thar (¥,*°) is also connected. Solution. (a) If f:X —> (0, 1} is a nonconstant continuous function, then 77"(O) is a nonempty clopen set which is different from X, and so X is not connected. For the converse, assume that every continuous function from X into (0, 1} is constant. If A isaclopen subset of X different from @ and. X, then the function #:X —> (0,1), defined by fQ) = 1 if x 0; lu—v] =uvv-Kay; uy (ut v + fe = vl); fut v— wv. Solution, We use the identities (a), (b), and (d) in Section 9 of the text. (a) Replace w by (w+) in uAv+w=(ut+w)AW+w) toger WAv= (uv) = (2) A(-u) = uv», (0) uv wu UAC) =~ v) AW =e) (©) u-wAw=u+(-v) Vw) =v uw). (@)If @ > 0, then au Av) = af=(-u) v (W)] = ~a(—n) v (-2}} (Hau) v (av) = (au) A (av). () Using (@), we see that uvu-uav=uvetfuvy-(+v)] =Auvv)-(Mtv) = Qu) v Qo) (u+0) = (=v) v (=H) = [a= oh (O Using (2) and (a), we get aut vl= (ev vbw nn) + (uveurv) =2uv0). (@) Asin (9, we get utvu-|e—vj=uvutuAv—(MVvu—wAL) = Au Ad). Problem 9.2. [fu and v are elements in a vector lattice, then show that: a feb ol fue vl = lal + fel, and be te vl A vl = ffl — Ie Solution. (a) Note that Web Ut Y= vl = Ee a) var WNL We = BV e+ On = (ut) ut vv Iu) vu = [wv (-u) + ¥] v (-«) vu — vy) = uv ud] + evo) = lal + fol‘Section 9: CONTINUOUS REAL-VALUED FUNCTIONS 8 (b) Using the distributive law, we see that let vl Ate =v} (ube) V UWA Mu = Y(t OI) [luto) 6 WW V nv) A Gea] v Coe) A (uted) v View ACu to 2 eV A Uy) HH) A= OY EA 0) + V)V fo a0) = a) (4 = fo) v (=u = Jol) v (v= Ha) v (=v ~ fa) Ca) = [lv fo v C9) — fall = Cel ~ fav (ol — Had) ll = tot] Problem 9.3. Show that \ul A |v] = 0 holds if and only if + v) holds: Solution, If lu|Alv| = 0, then using parts (a) and (b) and part (c) of Problem 9.1, wwe get Je wf mae vl = [lel — fot] = bel v fod bah A Teh = fel v jo tu fed = fe) fv = fel Fo) = fet vl YL = ‘This easily implies that ju + ul = | ~ v] holds. For the converse, assume that {u-+ vj = Ju ~ vl, Then by parts (a) and (b) of Problem 9.2, we have fut ~ fot] (ul + fol — fe» fu) — La 4 fot = La + fol ~ 2a 4 Jods fee) v Jot = Lal a fot from which it follows that jul A [ul = 0. Problem 9.4. Show that the vector space consisting of ail polynomials (with real coeficients) on R is not a function space. Prove a similar result for the vector space ofall real-valued differentiable functions on R Solution, If p is the polynomial defined by p(x) = x, then |p|(x) = |p| ix| holds, Clearly, |p| is not differentiable (and hence, it is not a polynomial either). Problem 9.5, Let X be a topological space. Consider the collection L of all real-valued functions on X defined by La (Ff @R% 3 (fa) SCC) sue thar tim fuox) = fo) Ve € X}. Show that L isa function space.% (Chapter 2: TOPOLOGY AND CONTINUITY Solution. Clearly, L is a vector space. Now, let f, g € L. Choose two se- quences {fn} and {ge} of CCX) with lim f,(x) = f(x) and lim gaGx) = g(x) forall x. Then f, V gx €C(X) foreach n and 1 Tim fa V BoC) = Him 5 [faC2) + 8n(2) + Ife) ~ ax] 1 [7@) + g@) + F@)- 8@l] = fv 8), 2 sothat fv g ¢ L. Similarly, f Ag € L, so that L isa function space. Problem 9.6. Let L be a vector space of real-valued functions defined on a set X. If for every function f © L the function |f| (defined by |f x) = |f(2)| for each x € X) belongs to L, then show that L is a function space Solution, Use the identities 1 1 fvgnzitetifmsd and fAg= af te~if—asl. Problem 9.7. Consider eack rational number written inthe form %, where n > 0, and m and n are integers without any common factors other than'-1. Clearly, such a representation is unique. Now, define f:R > R by f(x) = 0 ifx is irrational and f(x) = } if.x = % as above. Show that f is continuous at every irrational number and discontinuous at every rational number Solution. The proof will be based upon the following property: Let (ra) be a bounded sequence of distinct rational numbers. If tm = 3 (where ky > 0, and mq, and ky do not have common factors), then lim ke = 03 "To see this, pick some number M > 0 such that |r—| 0 we have [k,| C for infinitely many 1, then [mig] < MC must also hold for the same infinitely many n. However, this contradicis the fact that there isa finite number of rational numbers with |m| = MC and |nj x, then 0 = f(%,) —+ 0 = f(2). Thus, if fis not continuous at x, then there exists a sequence (rq) of rational numbers with ry, —> x and F(a) #0. Since x is irrational, we can assume 75 3a whenever n zm. Write r, = $,and note that f(r.) = 7 0 implies &, 7 00, 2 contradiction. ‘Therefore, f ‘is continuous at every irrational number. Now, let r be a rational number. Choose a sequence {ra} of. numbers with ry = 2 —+ r. Now, note that lim f(r) = lim £ holds, which shows that f is mot continuous at r. That is, f is discontinuous at every rational number.Section 9: CONTINUOUS REAL-VALUED FUNCTIONS n Problem 9.8. Let f:{a, 6] > Rbe increasing, ie.,x < yimplies f(x) = f() ‘Show that the set of points where f is discontinuous is at-most countable. Solution, Let f:{a,b] —+ IR be increasing, and let D be the set of dis- continuities of f. For each x € D choose a rational number r, such that Timyye f(0) Risa ‘continuous open mapping, then f is a strictly monotone function—and hence, @ homeomorphism. Solution, Let (a, b) be a finite open interval of IR, Since f atains a maximum value on [a,b] and f ((a,6)) isan open set, it is easy to see that the extrema of 7F on {a,b} take place at the end points. In particular, this implies f(a) # f(b). Ue fla) = f(0). then f(a, b)) must be a one-point set, contradicting the fact that f is an open mapping.) Next, we claim that fis strictly monotone on (@,b). Toscethis, assume f(@) < f(B),and a f(®) holds, ten fis strictly decreasing on (a,b) Now, assume that fis strictly increasing on (0, 1), and let x < y. Choose some with (0, 1) ¢ (—n,n) and x, y € (—n,n). Since f is strictly monotone con (n,m), and strictly increasing on (0,1, itis easy to see that f- must be strictly increasing on (—n, n). Thus, f(x) < ft) holds, and this shows that f is strictly increasing on R. (We remark thatthe function f need not be onto. However, the mapping f:IR —+ f(R) is a homeomorphism.) Problem 9.11, Let X be a nonempty set, and for any two functions f,g € R¥ let 1f@) COL we TF1F@)— ee" dfig) Establish the following: a, (RY, d) isa metric space. b. A sequence { fy) © RX satises d(fy, f) > Ofor some f € R* if and only if {fa} converges uniformly to f. Solution, (a) Clearly, d(f, 2) = Ofer all f, g € R¥ and d( f, g) = Oif and only if f = g. Moreover, it should be clear that d(f, g) = d(g, f) forall f.g € R® ‘What needs verification isthe triangle inequality. To do this, weneed the following ‘wo properties: 1 O O for each ¢ > —1. For (2) fix x, y 2 0, and note that (+d tal + ya HN +y) $y EN +9) < [tod ty) tad +] + hd +0 +) +0049] x+y tay) + y+) +x+y). Dividing across by (1-+2X1-+y)(1 ++), the validity of 2) can be established.Section 9: CONTINUOUS REAL-VALUED FUNCTIONS » Now, let f,g,J € IR and x € X, From LAG) — 01 < LF) = AGL + ACD) ~ BOL and (1) and (2), we get LAG) = g@01_ ¢ _ LF) = Aen + |G) ~ 00) THTFG)— gl ~ THI PC— AO + The — 8} << Lf) = AL \ieex) = a0 S Tee) — AG TFG) 6) 0, So, there exists ny such at | fx) ~ fla) < e bods for dlls > hy and all x © X, and hence, ql < | f(a) — fla)| < ¢ forall fh> no and all x € X. Iefollows that 7" = gap = FO Mo D= SP TE Lf) — FOO = for all n > 19, This shows that d(fe, f) > 0. For the converse, assume d(f,, f) ~> 0, and let ¢ > 0. Then there exists some rng such that sup a= LO Me D= SE THC) Fool ~ Tre for alln & no, and hence, AILINL < o£- forall m > mo and all x € X. This implies |f,C0 — fG01- F(a) ia R. If f is ‘Continuous, then show thatthe sequence offuctions{f, converges uniformiyto f Solution, Assume that the functions f, fi, fa, --. satisfy the stated proper- ties and thet the function f:X ——r Ris continuous. Also, assume by way ofw ‘Chapter 2: TOPOLOGY AND CONTINUITY contradiction that the sequence {f,) does not converge uniformly to f. Then an ‘easy argument shows (how?) that there exist e > 0, a subsequence {g,} of (fxl+ and a sequence {xq} of X such that [eaCse) — F%9)| Be foreach n. oy Since X is compact, the sequence {x,} has a convergent subsequence in X, say 11, —> x. By the continuity of we see that f(xi,) —> f(@). Also, from our hypothesis, it follows that gi,(x4,) —> f(2), and s0 [es.04,) ~ £G,)| + [F@)— FOI] = 0, ‘contrary o (4). Therefore, the sequence (fq) converges uniformly tof. Problem 9.13. For a sequence (fa) of real-valued functions defined on a topo- Logical space X that converges uniformly toa real-valued function f onX establish the following. 8 If Xx and f iscontinuous at x, then fyl%) + FO. b. Teach fy is continuous at some point xe € X, then f is also continuous at the point xq and ing, im, foe) = Jim, Jim fala) = Fo). Solution. (2) Assume fis continuous at x, x, > x andlet ¢ > 0. Choose some k with |fx(»)—f()] k andall y € X. By the continuity of f at x, there exists some m > k with [f(q)— fG)| <¢ for all n > m. Thus, [falsn) = F009] < [Falte) ~ Fn) + [FO = fo] <26 holds for all m > m, so that lim f(t.) = f(x). (b) Assume that each f, is continuous at xy € X and let € > 0. Since {fa} converges uniformly to fon X, there exists some k satisfying [/x(x) — f(2)] < € for all x € X. Now, the continuity of jf at x guarantees the exisience of a neighborhood V of xo such that |fi(x) — fi(%o)| < € forall x ¢ V. Then IF) ~ Flee ¢ 1F)~ ALDI + Lfele) — AG + Lid ~ F@0)| R be defined by fx(x) = x" for x € [0,1] ‘Show that { f,] converges pointwise and find its limit function. Is the convergence uniform? Solution. Clearly, 0, f0sx 0. Choose some 0 < 6 0 with Ig(s)| k. Thus, foreach n > k we have [x%g(x)| < MB" < 6 for 0 0. Pick some k such that forall > & we haves 1. aya O satisfying M forall x {a,b}. Thus, [[rerar~ ['seoas| = [[seoae— [soar [ forex [ roas| w fx(x) — f(x)|dx + " (x)| dx ‘ (x)[ dx [\n00- seolacs [ipeoiacs [ir e(b — a) + M(a, — a) +M(b — by) < e(2M +b ~a) w holds for all n > &, and our conclusion follows, Problem 9.17. Let {fa} be a sequence of continuous real-valued functions on a ‘metric space X such that {f,] converges uniformly to some function f on every compact subset of X. Show that f is a continuous function Solution, Let x, —> x in X. Put K = (x;,x2,...} U {x], and note that K is a compact set—every open cover of K can be reduced toa finite cover. Since Up) isa sequence of continuous Functions that converges uniformly to f on K, it follows from Theorem 9.2 that f is continuous on K. Since x, —> x holds. in K, we get f(q) —> f(a). Thatis, f isa continuous function Problem 9.18. Let {fa} and (gn) be to uniformly bounded sequences of real- valued functions on a set X. If both {fa} and (gq) converge uniformly on X, then show that {fu} also converges uniformly on X. Solution, Assume that (fg) and (gq) converge uniformly to f and g, respec- tively. Let @ > 0. Choose some & with |fuG)—f (| < e and [ga(2)—gC2)| <€ forall n > k and all x © X. Also, pick some M > 0 so that [f,G2)| = M and lanGs)| & and all x ¢ X. Problem 9.19. Suppose that (fa) is a sequence of monotone real-valued func- tions defined on (a, 6) and not necessarily all increasing or decreasing. Show‘Section 9: CONTINUOUS REAL-VALUED FUNCTIONS. ® that if (fa) converges pointwise to a continuous function fon (a, bl, then (fu) converges uniformly to f on {a,b}. Solution. Let e > 0. Since f is uniformly continuous (Theorem 7.7), there ‘exists some 6 > 0 so that [f(x) - f0)| <2 holds whenever |x ~ yl < 5. Fix a finite number of points a = x9 < 1 <7 < xq = B with x; ~ x11 <6 for 1 $i Sk, and then pick some mt such that |f-(x)) ~ f(4:)| < # holds for each 0 m. Now, let > m. Assume that fy is decreasing. If x € [a,b], then 2/1 < x $m holds for some 1. m, Thatis, {fx] converges uniformly to f. Problem 9.20, Let X be a topological space and let {fq} be a sequence of real- valued continuous functions defined on X. Suppose that there is afunction f:X > R such that f(x) = lim fa(x) holds for all. x ¢ X. Show that f is continuous at ‘a point a if and only if for each ¢ > O-and each m there exist a neighborhood V of and some k > m suck that | f(x) ~ fel2)| < € holds for all x € Solution. Assume that fis continuous at some point a. Let e > 0 and an integer m be piven. Pick a neighborhood U of a such that |f(s) ~ /(@)| <¢ holds forall x € U. Since lim fa(a) = f(a) holds, there exists an integer r > m such that |f(a) ~ f,(a)| < € holds for all n > r, Fix any integer > r and note that £ > m. Since fi is a continuous function, there exists a neighborhood W of a such that |fu(a) — fe(x)| < © holds for all x © W. Now, note that if xeV =U NW, then [ye — Aol s [fe — Fa] + LF(@) ~ fe@)| + [fala ~ fled} < 3 For the converse, assume that f satisfies the stated condition at the point a and let ¢ > 0. Since f(a) = lim fy(a) holds, there exists an integer m_ sucha (Chapter 2: TOPOLOGY AND CONTINUITY that |(@) ~ fy(@)] < € holds for all m > m. By the hypothesis, there exist a neighborhood V of « and an integer k > m such that I(x) — fe(x)I 0 such that |f,(2)| 0 and put 6 = e/M. Now, note that x, y € [a,b] and be=yl <8 imply [dec ~ 6.09] =| if * falt)dt f "parrat| < [fisseolae| <|f”mae| = anx= 31 R be a monotone (either increasing or decreasing) function. If there exists a dense subset A of R such that lim f(x) exists in R for each x € A, then show that lim fa(x) exists in at-most for all but countably many x.Section 9: CONTINUOUS REAL-VALUED FUNCTIONS 85 Solution. Assume that the functions f, and the dense subset A of satisfy the properties of the problem. Also, assume at the beginning that all but a finite umber of the fy are increasing functions. Define the function f:R —> R by FG) = limsup fel), ER. Note that f(e) is a real number for each x € R Indeed, if x ¢ R, then there exist a, b¢ A with a O. Pick some 5 > 0 such that x —8 mo, we have (00) ~ fal) S fl00) — fala) L/Ga) - f@) + F@— fla < e+e = 26 and alo) ~ F€%0) S fulb~ F€60) Lin) = FON + FO) ~ fGo0)] <6 +8 = 22.86 (Chapter 2: TOPOLOGY AND CONTINUITY ‘Thus, [fol&0) — f(&o)| <2e holds for all n> no, proving that lim f, (0) = Ste0). For the general case, assume that there are infinitely many increasing and in- finitely many decreasing f,. Splitthe sequence {,f) into wo subsequences (gq) ‘and (ha) such that each gy is increasing and each fy is decreasing. Put (x) = lim sup go(x) and x) = lim in h(x) = lim sopl—ha(2)) ‘and note that g(a) = f(a) holds foreach a € A. By the above conclusion, ¢ and ‘A ate continuous except possibly at the points of an at-most countable subset C ‘of IR, and for each point x ¢ C we have Tim ga(x) = g(x) and lime) = h(x). Now let ¢C and fix ¢ > 0. Pick some § > 0 such that [x—cl < & implies lg) — g(@)| 0 is arbitrary, we see that g(c) = f(e) holds for each ¢ ¢ C. This implies (how?) that lim f,(c) exists in R for each c ¢ C. Problem 9.23. Consider a continuous function f:{0,c0) > R. For eack n define the continuous function fy:{0, 00) — RR by falx) = f(x"). Show that the ‘set of continuous functions {f,, fr...) is equicontinuous at x = 1 if and only if f isa constant function. Solution. Let f € C{0,o0), let f,i[0,00) —> R be defined by fal) = F2"), and let E = (fi, fa, --.). HF fis a constant function, then it should be ‘lear thatthe set E is equicontinuous at x = 1. For the converse, assume that the set E is equicontinuous at x = 1, Fix a > 0 andlet ¢ > 0, The equicontinuity of E at x = 1 guarantees the existence of some 0 <6 <1 such that [x~ I] <4 implies |f,(x) ~ f,(1)I <6 foreach n, From lim /@ = I (why?), we see that there exists some np such that |y@~ 1] < 8 holds for each = mo. Thus, if n 2 no, then we have IfG@)~ FO) = [F(WAY) ~ FA|= fala) ~ ful) 0 is arbitrary it follows that f(a) = f(1) holds for each a > 0. By continuity, we see that f(a) = f(1) for esch a > 0, and so f is a constant function,Section 9: CONTINUOUS REAL-VALUED FUNCTIONS 8 Problem 9.24, Let (X,d) be a compact metric space and let A be an equicon- tinuous subset of C(X). Show that A is uniformly equicontinuous, ie, show that {for each ¢ > Oshere exists some & > O suck that x, y € X and d(x, y) < 8 imply LF) — FO) < ¢ forall f A. Solution, Let (X,d) be a compact metric space, let A be an equicontinuous subset of C(X), and let ¢ > 0. Foreach x € X there exists (by the equicomtinuity of A) some 8 > O such that d(x, y) < implies [f(x)— FO) <& forall $ GA. Brom X = yey B(x, $) and the compactness of X, we se that there exist x1,--+0te @X such that ¥ =U, Bx, 4). Let min[Sy,,--++8x,) > O-and let x, y € X satisfy d(x, y) < 8. Now, pick some 1 0 such that [(2)| $M holds for8 ‘Chapter 2: TOROLOGY AND CONTINUTTY each f A. In particular, we have |f0)| = £0) - F@|+|f@] <1 + forall f € A. This means that y € E, and so E = E. ie., E isa closed set. Next, we shall establish that E is an open set. To this end, let y € E. Pick some C > 0 such that {f(9) <= C holds for each f € A. By the equicontinuity of A, there exists a neighborhood W of y such that [f(2)— f(Q)| <1 holds foreach x €W andaall f € A. In particular, if x € W, then [feo] s [f@)- FO] +|F0)| <1 46 holds forall f © A, and so x € E. Thatis, WS E holds, which shows that y is an interior point of . Therefore, E is also an open set. Problem 9.26. Let (be an equicontinuous sequence in C(X),where X is not necessarily compact. If for some function f:X > R we have lim f(x) = f(8) for each x ¢ X, then show that f € C(X). Solution. Let x ¢ X andlet ¢ > 0. Since {f,} is an equicontinuous sequence, there exists aneighborhood V of the point x such that |fx(y)— fa(x)| < & holds forall n and.each y € V. Now, let y € V. Pick some k with |fi¢x) ~ f(@)] < # and |fi(y) - FON <2 and note that [£@)— FO] s [FO ~ AD] + [ACD — AO)| + [A — FO] <3e. That is, f is continuous at the arbitrary point x. Problem 9.27. LetX beacompacttopological space, and let f,) bean equicon- tinuous sequence of C(X). Assume thar there exists some f € C(X) and some dense subset A of X such that lim fu(x) == f(x) holds for each x € A. Then show that { fy} converges uniformly to f. Solution. Let ¢ > 0. By the equicontinuity of (f,) andthe continuity of f, for each x € X, there exists some neighborhood V, of x such that 1, [fa() ~ Jy(2)] mj. Cleasly, | fc) ~ FQ)! <2e. Thus, [far FO) [fod ~ fold | +1) fa] + LA) — FOO|+[F) — FO] <5e holds for ally € Vs, and all» > mj. Finally, put m = max(m: 1 m Problem 9.8. Show that for any five integer n > 1 the set of functions Fin C(O, t} such that there is some x € {0,1 ~ 1) for which [fe+A)— FG) snk whenever O 0. Consider the function g < C{0, 1] whose ‘raph is shown in Figure 2.1. Note that for each x € {0, 1) wehave [g(x +) — 20)| = 3rh for all sufficiently small A > 0. Put f = f +g, and note that0 (Chapter 2: TOPOLOGY AND CONTINUTTY FIGURE 2.1. ‘The Construction of a Nowhere Differentiable Function DF, fi) = [igloo = @. On the other hand, if x © {0, 1) is fixed, then for all sufficiently small # > 0 we have nh < Ink =3nh— nh < |g¢e+h)~ @@)|— [f+ — F0] S [eG +i) g) - 170) - For + I] = [Ale +H) ~ AQ] Thus, fr ¢ A, and so B(f,2e) Z Ay forall ¢ > 0. This shows that (4,)” =. Now, foreach n > 2 let By =(FeCW,1k 3x ELLA) with |f@—A)— fel sah whenever 0 < hk < £ By the same arguments, cach By is closed and nowhere dense, Consequently, ‘rom Baire's Theorem 6.17, we have cous © An) (Qa). In particular, note that every f € C10, 1} \ (U2 An) U (UR: have any one-sided derivative at any point of [0,1] By) does not Problem 9.29. Establish the following result regarding differentiability and uni- form convergence. Let (f,) be a sequence of differentiable real-valued functions defined on a bounded open interval (a, b) such that: a. for some xy € (a, b) the sequence of real numbers {f,(x0)} converges in R, ‘andSection 9: CONTINUOUS REAL-VALUED FUNCTIONS 1 b. the sequence of derivatives (f)} converges uniformly to a function g:(@,b) > R. Then the sequence | fx) converges uniformly to a function f: (a, b) > R which is differentiable at xo and satisfies f'(x0) = gto). Solution, First, we shall show that {isa uniformly Cauchy sequence, To this end, let € > 0 and pick some M > O such that |x — xo] < M for each x € (a, 6. "Next, choose some & such that [AG@) - f@)| <€ forall m,n > k andall x € (a,b) @ and [fals0) ~ fn(20)] < © for all m,n 2 &. (rs) ‘Using the Mean Value Theorem, (+) and (+#), we see that for each x & (a,b) and each pair n,m > k there exists some ¢ & (a, b) such that [sacs = fabsdl ss [[FoC) ~ fn] ~ [folao) ~ foGso!] | + [faGoa) ~ fated] [FsC0) ~ FC Ol be — ol + | fal) — SonC20)] R. ‘Next, for each n we consider the continuous function gy:(a,b) ~* R defined by dg(z) = BER Ebe) ifx A xo and gu(xa) = fi(to). Using the Mean Value ‘Theorem and (#), we sec that for each x € (a,) there exists some cx € (2,8) such that Lala \falesd ~ fale] <& Joa) — ts = Hei foralin, m 2 &. ‘This shows that (f] is a uniformly Cauchy sequence, and nence, itconverges uniformly tothe function (a,b) > Redefined by g(x) = Lead ite # xy and dea) = glo). Finally, from Problem 9.13, we obtain (sa) = Jim, £240) = Jim, fim @4(2) = Jim. Tim, 46 Li) 6) _ 5, LEd= Fle) ‘This shows that f is differentiable at xp and that f"(x0) = g(t0).2 (Chapter 2: TOPOLOGY AND CONTINUITY 10, SEPARATION PROPERTIES OF CONTINUOUS FUNCTIONS Problem 10.1, Ler (X,d) be a metric space and let A be a nonempty subset of X. The distance function of A is the function d(-, A): X > R defined by (x, A) = infld(e, a): a € A) Show that d(x, A) = 0 ifand only ifx € F. Solution, Cleatly, d(x, A) 2 Oforesch x € X. Assumethatx € Aandlete > 0. ‘Then B(x, €)MA # G,and sothere existssome y € A such that d(x, y) < ¢. From the definition of the distance function, we see that 0 < d(x, A) < d(x,y) <€. Since € > 0 is arbitrary, this implies d(x, A) = 0. For the converse, assume that d(x, A) = 0. If € > 0, then it follows from d(x, A) = infld(x, a): @ € A) < that there exists some a € A with d(x,a) <€. Hence, B(x, €) 9 A # db for each > 0, and this implies that x € A, Problem 10.2. Let (X,d) be'a metric space, let A and B be two nonempty disjoint closed sets and consider the function f:X —> {0.1} defined by FO) = a Sheow. Show that: a. f isa continuous function, b. f"(O)) = Aand f-"((1)) = B, and ©. if d(A, B) = infld(a, by: a € A and b € B) > O, then f is uniformly continuous. Solution, Let C be an arbitrary nonempty subset of X, We shall show first that the funetion x +» @(x,C) is uniformly continuous. To see this, fix x. y € X. Choosing some ¢ € C, we see that d(x, C) d(x, C) follows. (@) Observe that since A and B are disjoint closed sets, it follows from the Problem 10.1 that d(x, A) + d(x, B) > 0 for each x © X. This, in connection With the (aniform) continuity of the functions d¢-, 4) and d(-, B), guarantees that F isa continuous function. (b) Note that f(x) = 0 if and only if d(x, A) = 0. Now, by Problem 10.1, we have d(x, A) = O if and only if x € A = A. In other words, we have f(x) = Oif and only if x € A. This means f~'({0)) = A.‘Section 10; SEPARATION PROPERTIES OF CONTINUOUS FUNCTIONS 93 Similarly, notice that f(2) = qpeti-gy = 1 if and only if d(x, B) = 0. As above, this shows that f~"({1}} = c) Fix some € > O such that d(u, ») > ¢ for allu ¢ Aandve B. Ia e A ‘and b € B are arbitcary, then for each z © X'we have € £dla,b) IR defined by Fe) = dle, A) = dx, B). By the above, f isacontinuous function, From ANB = and Problem 10.1, we sce that f(x) = —d(x, B) <0 holds for each x © A. Similarly, f(x) > 0 holds for each x € B. Consequently, the two disjoint open sets U = f-'((—02. 0)) and V = f-'((0,c0)) satisfy ACU and BEY. Problem 10.4. Show hata closed set of a normal space is itselfa normal space. Solution, Let C be closed subset of anormal space X.. Weconsider C equipped ‘with the topology induced by X. Now, assume that A and B are two disjoint closed subsets of C. Since C is closed, itis easy to see that A and B are also closed subsetsos (Chapter 2: TOPOLOGY AND CONTINUITY of X. Pick two open subsets Vj and W; of X satisfying A ¢ Vi, B.S Wy and ViNW; = @. Now if ¥ = COV and W = COW, then V and W are two disjoint open subsets of C satisfying A C V and B CW. This shows that C equipped with the relative topology is a normal space. Problem 10.8. Let X be a normal space and let A and B be two disjoint closed subsets of X. Show that there exist open sets V and W such that A CV, B SW and VW = ¢. Solution. Assume that A and B are wo disjoint closed subsets ofa normal space X. Pick two disjoint open sets V and W) satisfying A C V and BC Wi. We claim that VW; = @. Indeed, ifx © VW), then on one hand Wi is a neighborhood of x, and on the other hand, x belongs tothe closure of V, which imply W, NV # @, acontradiction. Now, since VA B =o and X is normal, there exist two disjoint open sets Vi ‘and W such that VV, and B CW. As before, V, AW = @, and clearly the ‘open sexs V and W satisfy the desired properties. ‘Alternatively: If continuous function f: X —> (0, 1] satisfies AS f-1(0)) and BC f7!({1), then the open sets V = f-"({0, $)) and W = f-"((2, 1)) satisfy ACV, B CW, and VW = 9. Problem 10.6. Show thai a topological space is norma if and only iffor each closed set A and each open set V with A S V, there exists an open set W such that ACW CWC. Solution, Let X be 2 topological space. Assume frst that X is a normal space and let a closed set A and an open set V satisfy A CV. Then AN VE = G and V° is a closed set. Pick two disjoint open sets W and U such that A © W and V¢ © U. In particular, WU = @. This implies WAV" = @, and so Wey. For the converse, assume that the property is satisfied and let A and B be two (0, t] such that f~'({0}) then the identity Fo) = °F.) =A Fo. 2) shows that A is a Gy-set. For the converse, assume that A is a Gerset. Pick a sequence {V,) of open sets such that A = ()f2; Va. Since A MVE = G, it follows from Uryson’s lemma that there exists @ continuous function fy:X — (0, 1] satisfying f(a) = 0 for each a A and f(x) = | forall x € Vz. Now, consider the function f:X —> (0. 1] defined by fl) =did Ke) From the Weierstrass’ M-test (Theorem 9.5) and Theorem 9.2, itis easy to see that is. continuous function, and we claim that f~!((0}) = A. Clearly, f(x) = 0 Toreach + € A. Now, assume f(x) = 0. Then f(x) = 0 forall n, and so (in view of fu(u) = I foreach v € Vs) we have x € Vy foreach nie. € (\R Ya = 4 ‘Therefore, f~"((0}) = A. (6) Assume now that Ais a closed Gy-set and B is another closed set such that ANB =@. So, there exist two disjoins open set V and W such that A & V and 'B CW. This implies thatthe sequence {V,} introduced in pat (a) can be assumed to satisfy Vy € V for each n. In particular, each fy satisies f,(b) = 1 for each be B, Now, its easy to see that the continuous function f constructed in the preceding part satisfies the desired propery Problem 10.8. Show that a compact subset A of a Hausdorff locally compact topological space is a Gy-ser if and only if there exists a continuous function f:X — [0, I] such that A= f-"(0)). Solution, If A = f~1({0}),then—as in the solution of part (a) of the preceding problem-—theset A isa G,-set, Fortheconverse, assume thet A = (ny Vav Where tach Vy isan open set, By Theorem 1038, for each n there exists @ continuous96 (Chapter 2: TOPOLOGY AND CONTINUITY function fa: X —> (0, 1] such that fy(x) = 1 for each x € A and fy(x) = 0 for each x ¢ V,. Now, as in the the solution of part (a) ofthe preceding problem, notice that the function f:X > {0, 1] defined by f(x) = SS, + fa(x) satisfies the desired properties. Problem 10.9. A topological space X is said to be perfectly normal ior every pair of disjoint closed sets A and B, there is a continuous function f:X —»r (0,1) such that A = f—'({0}) and B = f~\({1}). (Part (6) of Problem 10.2 shows that every metric space is perfectly normal.) Show that « Hausdorff normal topological space is perfectly normal if and only if every closed ser isa Gy-se. Solution. Let X be « Hausdorff normal topological space. Assume first that X is perfectly normal and let A be a proper closed subset of X. Ifa € X satisfies a ¢ A, then AN (a} = G and (a) is a closed set. So, there exists a continuous function f:X —> (0, 1] with f~!{(0}) = A. This implies (as in the solution of Problem 10.7), that A is a Gy-set. For the converse, assume that every closed set is a Gy-set. Let A and B be two closed disjoint seis. By Problem 10.7 there exist two continuous functions ‘gh: X — [0, 1] such that: i. g7({0)) =A and g(b) = 1 foreach b ¢ B, and ii, A-'((0}) = B and h(a) = 1 foreacha € A. Now, let f = 4g + 41h), and note that f:X + (0.1). A = f7"((0}) and B= fy, Problem 10.10, Show that a nonempty connected normal space is either a sin- ‘gleton or uncountable. Solution. Let X be a (nonempty) Hausdorff connected normal space. If X is nota singleton, then there exist a,b € X with a # b. Since X is Hausdorf, singletons are closed sets, and we have (a) (1{} = @. Now, pick a continuous function f:X — {0, 1] such that f(a) =O and f(b) = 1. The assumption that X isconnected guarantees (according to Problem 6,1(g) that f() isan interval and so f(X) = [0, 1]. This easily implies that X is uncountable—in fact, it has cardinality greater than or equal othe cardinality of the continuum. Problem 10.11. Let X be anormal space, let C be a closed subset of X, and let 1 be a nonempty imerval—with the possibility I = (—90,c0). If fC > Iisa continuous function, then show that f has a continuous extension to all of X with values in 1‘Section 10: SEPARATION PROPERTIES OF CONTINUOUS FUNCTIONS 97 Solution. Assume that C is aclosed subset of a normal space X and that f:X —> is acontinuous function, where / isan interval. The interval ! must be one of the following type: (2, ), {a, 6], [a, 5), (a, b]. So, we shall establish the continuous extension of f by steps. STEP I: I is either of the form la, b) or (b, a. In this case, there exists a homeomorphism h: 1 —> [0,1). For instance, if moo (0,1) S (0, 1]. By Tietze's extension theorem, there exists a ‘continuous function g:X —> [0, 1] satisfying g(x) = h( f(x)) for all x € C. The continuity of g guarantees that the set A = g~'((1}) is wclosed subset of X. Also, since for each x € C, we have g(x) = h(/(e)) € [0, 1), we see that CA =o. By Uryson’s lemma, there exists a continuous function 8: X ~» (0, 1] such that 6(a) =O for each a € A and 9(c) = 1 foreach ¢ € C. ‘Now, consider the function @: X + {0, 1] defined by (x) = O(x)gt). We claim that p(X) & {0, 1). Tosee this, etx € X. If € A,then g(x) = 80) = 0-1 =O,andifx € Athen < g(x) < Land so g(r) = O(x)g(x) < Lisalsotrue, Next, define the function f:X — 7 by defines ahomeomorphism between rt odin = ~H(acsds). If €C, then 6(x)g(x) = g(a) = h(/(2)), and hence, (h( F000) = £00. “This shows that f:X ~> Iisa continuous extension of to al of X. STEP UL { = [2,6] with -90 [0, 1} satisfying g(x) = (to fYGx) for each x « C. Then the continuous funetion Fon! op:X -> a,b] satisfies fc) = f(e) foreach c € C. STEP II: Assume | = (a, 6) with ~c0 ¢ a (1, 1). (For instance, for —c0 (1, 1] satisfying g(c) = (I o f)(c) for each c € C.98 ‘Chapter 2: TOPOLOGY AND CONTINUITY Next let B= g~!((1,1)). Thon B is closed and BAC = g. By Uryson’s Jemma, there exists a continuous function 8: X ~» (0 1 satisfying 6(5) = 0 for each b € B and 9(c) = 1 foreache € C. As before, define the continuous function 2X ~ [-1, 1] by d(x) = x)g(4). Then itis eaay to see that (X) ¢ (1,1) and the function f:X > (a, b) defined by f = h-!0¢, isacontinuous extension of f. 11, THE STONE-WEIERSTRASS APPROXIMATION THEOREM Problem 11.1. Let X be a compact topological space. For a subset L of COX), {et L denate the uniform closure of L in C(X). Show the following a. If Lis a function space, then so isT. b. IfL is analgebra, then so is L. Solution. Let f, g € EL. Picktwo sequences {f,} and (gx) of L that converge my to f and g, respectively. Also, pick some M > 0 sothat ll fnlleo 0. Bytbe uniform continuity of there exists some § >0 such that [x ~ y| <8 implies [7«) — f(9)] <6. Let 0 = xp TR és the uniform limit of a sequence of polynomials on (0,1) if and only if it admits a continuous extension t0 (0, 1)- Solution, Let f:(0, 1) —> R be a continuous function. Assume first that f has a continuous extension to [0, I}-which we denote by f, Then, by Cor: ollary 11.6, the function f is the uniform limit of a sequence of polynomials on {0,1}, and consequently :(0, 1) —> IR is likewise the uniform limit of « sequence of polynomials on (0,1). For the converse, assume that there exists a sequence of polynomials {p,) that ‘converges uniformly to fon (0, 1). Let e > 0 and then pick some rig such that pax) — FCS)| <6 holds for all x € (0,1) and all w 2 1p. From the tiangle inequality, we see that [pad — pao] S pal) — F60)] + [pated — FO] < e+e = 20 for all x € (0, 1) and all n 2 no, By continuity, we infer that pats) ~ Pm) $28 holds for all x € (0, 1] and all 1 > no, The above show that (pp) is a Cauchy sequence of C[0, 1], and so (by Theorem 9.3) the sequence (p,} converges in C0. 11, say to g € C10, 1]. Hfollows that fx) = g(x) forall x € (0, 1, and so ¢ is acontinuous extension of f:(0, 1) > R to (0,1) Problem 11 forn = 0,1, ff ivacontimousfanctionon{0,t}suckthat fx" fodx = iter sho that (2) = O forall x € (0.1). Solution. By Corollary 11.6, there exists a sequence of polynomials (ps} that converges uniformly to f. It easily follows that (p, f} also converges uniformly to f?, and by our hypothesis we see that {j pa(x) f(x) dx = 0 holds for each n. ‘Now, invoke Problem 9.16 to infer that fp f%(x)dx = lim fp pal) f(x) dx = 0. “The latter easily implies that /(x) = 0 holds foreach x € [0, 1.100 (Chapter 2: TOPOLOGY AND CONTINUTY Problem 11.5. Show shat the algebra generated by the set {1,x?} is dense in C10, 1 bu fais tobe dense in C1, 1, Solution, Since the function f(x) = x? separates the points of (0, 1}, the algebra generated by (1, x?) also separates the points of [0,1]. Thus, by the Stone-Weierstrass, this algebra must be dense in C{0, 1]. ‘To see that the algebra generated by (1, *) isnot dense in C[—1, 1}, note that for every f in the closure of this algebra, we have f(—1) = f(1). Thus, this algebra is not dense in C[~1, 1] Problem 11.6. Let us say that @ polynomial is odd (resp. even) whenever it does not contain any monomial of even (resp. odd) degree. Show that a continuous function f:{0, 1] > R vanishes ar zero (ce, f(0) = ‘fand onty if itis the uniform limit ofa sequence of odd polynomials on (0,1). Solution. If / is the uniform limit of a sequence of odd polynomials, then it should be clear that f vanishes at zero, For the converse, assume that f © C10, 1} satisfies #(0) = 0 and let ¢ > 0, Define the function g:[—1, 1] + R by fo. if0sx fim, if-1<%<0, at) = and note that g € C[-1, 1]. By the Stone-Weierstrass theorem there exists a polynomial p such that [g() — p(x)| Risa continuons fueion such that fé0°Y%) dx = Ofor 0,1,2,..., then show that f(x) = 0 for all x € (0, 1], Does the «ern conclusion hold rue ifthe buerval (0, 1} replaced bythe interval (=? Solution. Assume that a continuous function f € C10, 1 satisfies fy FYE) (for each = 0,1,2,... Then the change of variable = ™4/% [ tevoar scarey [errand , and so [x2 fle)de = 0 holds forall = 0.1, 2,-... The concusion now follows immediately from Problem 11.5. "The conclusion is not valid if we replace the interval [0,1] by the inte- val [1.1]. For instance, if f(x) = x for all x € [=1, 1], then note that PL FOME) ax = 0 holds for all 1 = 0, 1.2... Problem 118. Assume that a function f:[0, 00) > R is either a polynomial or flsea continuous bounded function. Then show that fis identically equal to zero Ge, show that f = 0)ifand only if f° Flxde™ dx = Oforalln = 1,2,3 Solution. Let f:{0,co) —+ R be a continuous bounded function. If f then cleanly /2°f(x)e"™ dx =O holds forall x = 1,2,3,- For the converse, assume that [poner as = 0 holds forall m= 1,2,3, rey bo Using the change of variable w = e°*, it follows from («) that 0 q [ive ace [’p-mane'n 0, n=1.2, bo) in paul, fg(de" du = O holds foreach w= 0,1,.0., vere g(t) = tift-Inu). Since f is bounded, note that lim,-or gu) = 0 holds, and so g defines a continuous function on {0, 1]. From (a), we see that Jy g(x)x" dx = 0 holds forall n= 0, 1,2... Problem 11.4 implies that ¢ = 0, and consequently f=0. 'A closer look at the above arguments reveals that we have actually proven the following result1m ‘Chapter 2: TOPOLOGY AND CONTINUITY © Assunte that f:{0, 00) > R isa continuous function such that [lrover ern frat nate btn [ where k is a positive integer. If tim,.o+ u" f(—Inu) = 0 for some natural number m, then the function f is identically equal ro zero. Indeed, replacing by n +k +m + 1 in (os), we get 1 [ot etna di 0, 2012s oe ‘which implies (as above) that f = 0. The reader can verify easily that any function F that satisfies 17(0)| = Ce** for some C > Oand a > 0 and all x 2 2p also Satisfies Limy..o¢ «" f(= In) = 0 for some natural number m. In particular, the reader should notice that every polynomial p satisfies an estimate of the form Ip(a)| = Cem One more comment regarding the above discussion is in order. Recall that if {:(0,00) > RR is “nice” function, then the formula L(fys) -f er pat is called the Laplace transform of f. The Laplace transform is a linear operator and plays an important role in a wide range of applications. The reader should notice that in actuality property (e) asserts that the Laplace transform is a one-to- ‘one operator when defined on an appropriate linear space of functions. (See also Example 30 of Chapter 5 in the text.) Problem 11.9, Show that a continuous bounded function f:{1,00) > Ris identically equal to zero if and only if {7° x-" f(x)dx = 0 for each n = 8,9, 10, Solution. The “if” part only needs verification. Therefore, assume tha the func tion f:[1,00) —+ R satisfies ff"x-" fa)dx = 0 for each n= 8,9, 10... Using the change of variable w © x“, we see that [roe fore) dus f wr Fg(u) du (er) where g(t) = uf s(2), Since is bounded, we see tht Timy-»o- a(t) = 0, nd so g defines a continuous function on [0, 1]. In addition, from (« * *), we sec thatSection 11: THE STONE-WEIERSTRASS APPROXIMATION THEOREM 103 {gx"g@)dx = 0 holds for each n = 0,1,2,..-. By Problem 11.4, it follows ihat ¢ =0,and consequently, f = 0. Problem 11.10, Ler Abe analgebraof continuous real-valued functions defined ‘ona compact topological space X and separating the points of X. Show that the closure A of Ain COX) with respect tothe uniform metric is either all of C(X) or else that there exists a € X such thar A = (f € COX): f(a) = 0} Solution, Let AC C(X) be an algebra, where X is compact. Now, consider the sequence of polynomials {P,()) on (0, 1] defined by Pi) =O and Pravlx) ~ (Pj(s))"] for n= 1, An easy inductive argument shows that each polynomial P,(x) has a constant term equal to zero. This guarantees that if f € A, then P,(f) € A for each n Also, by Lemma I1.4, we know that the sequence {P,(x)} converges uniformly to VE on (0, 1]. Thus, if f € A is non-zero, then put ¢ = ff fo, and note tha: 1. The sequence | P,()] © A converges uniformly to IL. Hence, |f| ¢ A 2. Since (P,(2)} € A converges uniformly to /iZl,we seechat J/l eA, ‘Thus, iff € A, then both [f1 and /IFT betong to algebra and a function space, Now, suppose that A is not of the form (f € C(X): f(a) = 0} for some a € X. This implies that for each x € X, there exists some f ¢ A with “fla) #0. Thus, for each x € X, there exists some f, ¢ A and a neighborhood V; of x with f,(9) #0 forall y e V,. By the compactness of X, there exist Kjgsveodn €X with X= Uf Va. Note thatthe function g-= 72 bo 72, of A satisfies g(x) > 0 foreach x ¢ X. Multiplying by an appropriate constant, wwe can assume that g(x) > 1 holds for all x. Put hy = 2/@, and note that hig €A and that fg(x) 41 for each x € X. By Dini’s theorem, {fy} converges uniformly tothe constant function 1,andso 1 € A. Theorem 11.5 now guarantees that A= COX) must hold In particular, A is an Problem 11.11. Let A be the vector space generated by the functions Asi x,sin® x, defined on{0, 1}, Thatis, f € Aifand only ifthere isa non-negative imegerkand real numbers...» (all depending on f)such hat f(3) = TS i x104 ‘Chapter 2: TOPOLOGY AND CONTINUITY {for each x € [0,1}. Show that A is an algebra and that A is dense in C(O, 1) with respect to the uniform metric. Solution, Clearly, A isan algebra of functions that contains the constant function 1. Also, since the function f(x) = sin.x separates the points of (0, 1, the algebra A likewise separates the points of (0, 1]- By the Stone-Weierstrass theorem, is dense in C10, 11 Problem 11.12. Let X bea compact subset of R. Show that C(X) is a separable metric space (with respect to the uniform metric), Solution, The polynomials with rational coefficients form a countable set why”). By Corollary 11.6, this set is dense in C(X). Problem 11.13. Generalize the previous exercise as follows: Show thar if (X, d) is.a compact metric space, then C(X) is a separable metric space. Solution, By Problem 7.2, we know that X is a separable metric space. Fix a ‘countable dense subset (x1,%2,...) of X and for each m let jy:X —> R be the function defined by fa(0) = d(t, x») for each # € X Now, let x, y © X satisfy x # y. Put d(x, y) = 26 > 0. Choose some n with d(x.) < 3, and note that Sol) = dye) & U(x, y) — dG He) = 25 5 > dx, an) Sule) so that f(x) fel»). This implies thatthe algebra.generated by (1, fi, fas} separates the points of X. By the Stone—Weierstrass theorem (Theorem 11.5), this algebra must be dense in C(X), ‘Next, consider the collection C of all finite products of the countable colicc- tion (1, fy, fa...-} and note that C is a countable set, say C = (815 821-+-}- To ‘complete the proof note thatthe finite linear combinations of (4, g1, gay.) with rational coefficients form a countable dense subset of C(X). Problem 1114, Ler X and Y be two compact metric spaces. Consider the Cartesian product X x Y equipped with the distance Dy given in Problem 74, $0 that X x ¥ isa compact metric space. Show that if f € C(X xY) and ¢ > 0, then there exist functions (fy.-+++ fa) S C(X) and (3)... 48) S CY) suet that [fe D Aiea] <€ folds for all (x, y) © X x ¥.Section 11; THE STONE-WEIERSTRASS APPROXIMATION THEOREM 105 Solution, Consider the set heC KY) lfc fal SCO (Biveee Se SCO) with AG 9) =o fia) ¥ (ey) eX x ¥ | “Then, A isan algebra of functions of C(X x ¥) and 1 € A. On the other hand, if Gera 9) # Gen, then either xy xe oF 94 #2 Yo. HE xy 7 xa, then select some F OC(X) with fC) # Ga), ander F(x, y) = FO) forall (c, y) € Xx ¥. TE ‘y1 # Ja then pick some g € C(V) with g(y1) # a(2), and put F(x, y) = 86) neither case, F € A and F(%i, i) # Fz, Yo) holds, so that A separates the points of X x ¥. Now, by the Stone-Weierstrass theorem (Theorem 11.5), we have A = C(X x ¥), and the desired conclusion follows.CHAPTER 3. THE THEORY OF MEASURE 12. SEMIRINGS AND ALGEBRAS OF SETS Problem 121. if X isa topological space, then show that the collection $= (C0: C closed and O open] = {Cr \ Cx: Cr, Cx closed ses} isa semiring of subsets of X Solution, From @ = GAG and X = X NX, we see that , X € S. Next, notice that C, 1.01, C02 € S imply (iN OYNCrN On) = (C1 1C2)(0, 901) € S. Now if C190), C20 02 € S, then ELMO, \ C22 = (C191.02) (C2 Oa C104) N(CHUOD crn onnfesulognca] crn (O, NED] U[(E, E203) 1) AUB, where A= C, (01 NC%) € S and B= (C1 NC.NOHNO €S satisfy AnB=d. Problem 12.2. Let S be a semiring of subsets of a set X, and let Y CX. Show that Sy = (Y 1A A € S) is asemiring of ¥ (called the restriction semiring of Sto¥). 107108 ‘Chapter 3: THE THEORY OF MEASURE Solution. The conclusion follows from the identities: a YNG=4; BM AAW NB) =¥ NAN), and c. YMA\YNB=YNCA\ B) Problem 12.3. Let S be the collection ofall subsets of (0,1) that can be written as finite unions of subsets of (0, 1) ofthe form (a,b). Show that Sis an algebra of sets but not a ¢-algebra Solution, Let A =Ujylai.b)) and B= Jf.;[¢j.d;). Then, we have a AUBES, ANB = Sins Ujurlar, 61) fe), d)) € Ss and 10, 1) \ A = ((0, 1) \ [a;, 5s) © S, where the last membership holds since each [0, 1) \ [a;, bj) can be written as a finite union of sets of the form (a, b). ‘To see that S isnot a o-algebra note that 2,(0, 1) = (0) ¢ 5. Problem 12.4. Prove that the o-sets of the semiving S= (lab): be R} form a topotogy for the real numbers. Solution. Let r be the collection of all o-sets of S. Clearly, ¢ t and R= USi(-n.n) € t. It should be clear that x is closed under finite intersections. ‘Thus, in order to establish that x isa topology, we need to show that z is closed under arbitrary unions. That is, if (fa,,bj): 1 € J) is a collection of nonempty members of 5, then we must show that A = UJj 0, and hence s.(E) = 0. Problem 14.4. If E isameasurable subset of X, then show that for every subset A of X the following equality holds MEO A) + WE UA) = w(E) + lA), Solution, ‘The measurability of £ gives MEUAY (EU AN B) + a(EU AYN E*) = w(E) + WANES). ‘Consequently, we have MEU A) + WENA) (BE) + BAO ES) + (ANE) = ME) + HCA). Problem 145. Let jz be an outer measure on a set X. If A isa nonmeasur~ ‘able subset of X and E is a measurable set such that A © E, then show that ME \ A) > 0. Solution, If 2(E \ A) =O holds,then £ \ A ¢ A, Thus,A = E \(E \ Aye ‘A. which is a contradiction. Therefore, w( \ A) > 0.48 Problem 146. Ler A be a subset of X, and let {Ey} be a disjoint sequence of measurable sets. Show that w(Uangs) = Soman Solution, From the o-subadditivity of 1, we see that (an [U)e]) =«(Cane,) = Suan en. On the oterhan, Lemma 145 implies : Ya nen=x(an[Ye)) =u(ao[Cle) for each band so S224 1(A 0 Ey) < u(A A[UR, Bo]) also hoes Problem 14.7. Let (Aq) be a sequence of subsets of X. Assume that there exists a disjoint sequence {B,) of measurable sets such that Ay © B, holds for each n. Shove chat (Can) =o mao. Pa} ot Solution, Put A= US; Ax and note that AB, = Ay holds for each m, ‘Thus, using the preceding problem, we see that (Qs) =o(ao(Oe) Problem 148. Let be an outer measure on a set X. Show that a subset E of X is measurable if and only iffor each € > O there exisis a measurable set F such that F © E, and w(B\F) <¢. Yodo sn = Suan. Solution, If E ismeasurable,then F = E satisfies thecondition foreach e > 0. For the converse, assume that the condition is satisfied, Start by choosing foreach x ameasurableset F, with Fy SE andu(E \ F,)< b. Pu F = USF, < E, and note that F is measurable. Consequently, ME\F) < mE \F,) < } for each n implies w(E \ F) = 0, and so E.\ P is measurable. The measurability of E now follows from the identity E=FUE\F).‘Section 14: OUTER MEASURES AND MEASURABLE SEIS. ny ‘An alternate proof of the preceding part goes as follows. Let A be a subset of X with u(A) < 00, Ife > Os given, pick a measurable subset F with F CE and w(E\ PF) < ¢. Then. m(ANE) (AOE UE\ FD) < ANF) +u(AN(E\F)) ~€,and so BA) = WAN F)+ ANF) = MANA) + MANE) = MANE) + WANE) + [WAN F)~ BANE] > MANE) + MANE) —€ for all « = 0. This implies (A) = (4 £) + (AO E*), which shows that E is a measurable set Problem 149. Ler 4 be an outer measure on a set X. Assume that a subset E of X has the property that for each ¢ > 0, there exists a measurable set F such that WEAF) < €. Show that E is a measurable set. Solution, “Let e > 0. According to the preceding problem, it suffices to show that 2(E \ G) < € holds for some measurable set G with G ¢ E. For each n choose Fy A with 4(EAF,) <2-*e, Put F = [1% Since F \ ES F, \ E holds, we have Fen CF \ £) SMF, \ E) <2 for each n, and so w(F \ £) = 0. Thus, F\ E € A, and hence FOE = F \ (E \ B)isalsoa measurable set. Now, note that FE & E holds and ME \ BOF) = wb \ P= «(LIE \ Fo) s SOME \ FD <6 Problem 14.10. Let X = (1,2,3), ¥ = (Qs (1), (1,2}) and consider the set {function zF > [0,00] defined by (GB) = 0, w(1)) = 2 and go({1,2)) = 1. a. Describe the outer measure jx" generated by the set function 1. b. Describe the o-algebra of all s*-measurable subseis of X (and conclude that the set {1} © F is not @ measurable sei),120 ‘Chapter 3: THE THEORY OF MEASURE Solution. (@) The outer measure 4*: POX) — [0, oo] is given by H@) =O, w(U)) = H"(12)) = 1, H*((31) = 00, w*((1.2)) = 1, w*((1,3)) = 2°((2,3)) = (11.2, 3) = () The o-algebra ofall measurable sets is A = {¢, (3), (1,2) X]- Problem 14.11. Ler v: P(X) -» (0, 00] be asetfunction. Show that v isan outer ‘measure ifandonly if there exist acollection F of subsets of X containing the empty set and a set function p:-F > (0, 00} with (G3) = 0 satisfying v(A) = *(A) forall A @ PX). Solution. Assume first that v: POX) + (0, oo] is an outer measure. Let F = P(X) and = v, We claim that v(A) = s.*(A) holds for each A € P(X), where w(A) = ini | J aCAsy: (Aa) SF and ASL Ae}. and inf @ = oo. To see this, let A € P(X). From A = AUGUGUG:---, we see that °(A) < (A) = v(A). On the other hand, if A ¢ UJ; Ay holds true, then from the «-subadditivity of v, we see that WA) s Dv An) = Sacer, and so v(A) < °(A)is also true, Hence, »(A) = 1A) for each subset A of X. For the converse, assume that the outer measure 4.” generated by a set function uF —+ (0, 00] satisfies u() = 0 and v(A) = u*(A) for cach A € P(X). We shall show that wis an outer measure by verifying the three properties required to be satisfied by v in order to be a measure. (1) From 0 < 1) = uw") < WE) + HOD) + WD) + +++ = O, we see that v@) = 0. (2) (Monotonicity) Let A © B and let {A,} be a sequence of F with BC Jrat Ane Then, AS Uso Ans and so 4°(A) = Foy M(An). Therefore, (A) = CA) s intl) (An): (Aq) SF and BS J Ay | = WB) = v(B). m= (rthereis no sequence (4_) of F with B.S US, Ap, then 1°(B) = 00, and v(A) = v(B) = °(B) is trivially tue.)Section 14: OUTER MEASURES AND MBASURABLESETS mt @) (o-Subaaltivity) Let (Eq) be a sequence of subsets of X and let E US) En If DE) H"(Eq) = 00, then vi) = WE) Ss Day eEs) = SS CEs) is tively tre. So, assume Y%, 4°(E,) < 00 and let € > 0. Foreach n pick a sequence {At} of F with Ey ¢ Uj, AS and Me BAY) < Wi En) +2 Ey) + 2%. Clearly, £ © UR UE: AS holds, and so ME) = Wt) s OMA < SIME +2] = ME +6 mict tat Since ¢ > 0 is arbitrary, v(E) < LG, MCE»). and we are done, Problem 14.12, Consider an outer measure 2 on a set X and let A be the collection of all measurable subsets of X of finite measure. That is, consider the family A= (A € A: (A) < 00} a. Show that Ais a semiring. b. Define a relation = on A by A= B if w(AMB) = 0. Show that = is an equivalence relation on A. c. Let D denote the set of all equivatence classes of A. For Ae A let A denote the equivalence class of A in D. Now, for A, 8 & D define (A, B) = w(AAB). Show that d is well defined and that (D,d) is @ complete metric space. Solution, Note that if A, B,and C are three bitrary sets, thei AAC © (ABB)U (BAC), (@) Straightforward, (Note that in actuality A is a ing of ses.) (IEA, Byand C in A satisly A= B and B ~ C, then the relation BAC) < 4((ASB)U BAC) < w(AAB) + H(BAC) = 0 shows that A= C. (If A~ A; and B = By, then M(AAB) < (AAA) U(ALAB)U (BAB) << WASAL) + M(ALABy) + 4B) AB) (A, 6B).mm (Chapter 3: THE THEORY OF MEASURE. Similarly, (A) AB)) < 4(AAB), and so u(AAB) = 4(4,AB)). This shows that d(A, B) = n(AAB) is well defined. For the triangle inequality, note that (A, B) = WAMB) < WAAC) + (CAB) = dA, C) + d(C, B), ‘Thus, (D,d) is a metric space. What remains to be shown is that (D.d) is a complete metric space. Tothis end, let (Ay) be a Cauchy sequence of D. By passing to a subsequence, ‘we can assume that Ant, Ap) = Wl Ang AAy) <2 holds foreach n, Set A= (2, UfEs Ai € A. Now, let 1 be fixed and note that AS Ula, Ai = Aa U (Uz (Acer VA) holds, Thus, HOA) < ulAs) + D> wlAias \ AD < H(Ag) +2 < 00, td andso A € D. Moreover, we have MAY Ad) = (Chern \ An) = Satan \an<2%, On the other hand, if x € Ay \ Athen x € Ay and x ¢ A= (RUE, Ar Consequently, there exists some k > n with x ¢ A; foreach i > k. This implies An \ AS UA: \ Aist)s and 80 (Ay \ A) = Dhan HAL \ Aig) < 2% also holds. Therefore, (Ag, A) = H(A AA) = HlAn \ A) + MCA \ Ay) < 20" holds for each n. This shows that fimd(dy, A) = 0, and so (D, d) isa complete metric space. (For an alternate proof of this part, see Problem 31.3.) 15. THE OUTER MEASURE GENERATED BY A MEASURE, Problem 15.1. Let (X, S, 4) be a measure space, and let E be a measurable subser of X. Put Sp = (EM A: A € S), the restriction of S to E. Show that (E, Se, u") is a measure space.Section 1S: THE OUTER MEASURE GENERATED BY A MEASURE, 1s Solution. Let E be a measurable subset of X and let (Aq} be a sequence of A such that a. (An ME} is pairwise disjoint sequence; and thee exists some A ¢S such that ANE = Us, An VE Using the fact that ju": A —+ {0,00} isa measure, we see that wane) =u"(taen Bd) = wane), and so jc" is a measure when restricted to the semiring Se Problem 15.2. Ler(X, S,j) be a measure space. Show that wa influ"(B): B is ac-set such that AC BY holds for every subset A of X. Solution. Let (Aq) GS and let B = US Ax. By Theorem 12.213), there exists a pairwise disjoint sequence (B,) of S such that B = U2, By. Thus, for AC X, there exists a sequence (A,) of S with AC US An if and only if there exists a g-set B with A ¢ B. The desired equality now follows from the relation HAS Suoo = = u®) s ua. Problem 15.3. Show that every interval I of Ris Lebesgue measurable and AAD) = All (the length of 2). Solution. InExample 15.5, weestablished thatthe intervals 7 of the form (a, b} ‘and {a, 00) are Lebesgue measurable and that 4°(7) = | holds for these cases ‘We shall consider the other cases separately. Assume —20 0 and so 2°((a}) = M*([a —6,a-+6)) = 2e for all © > 0. Therefore, 2*({a}) = 0 holds forall aR. Hf A= (ay, 22, -..} = Use laa) isa countable set, then note that AMA) = De, A(la}) = 0 80 that 2°(A) = 0. Problem 15.5. For @ subset A of Rand real numbers a and b, define the set aA+b= (ax +b: x € Al. Show that a AGA +b) = falA*(A), and b. if Ais Lebesgue measurable, then so is aA +b. Solution. Let A ¢ Rand fixtworeal numbersa andb. Since A © U% lay, Bs) holds if and only if A +b © UP jay + 6, by +5) holds, it is easy to see that At(A +b) = 2°(A). The identities EN(@+A)=b+(E-b)NA and EN(b+ AS =b+(E-b)NAT imply AEN +A) EX (ENO + AY) =A (E-b)NA) 4A(E- BAY), which shows that A is measurable if and only if 6 + A is measurable for each beR. ‘Next, note that A*(c(s, 1)) = |clA*((s, 1) = Iel(¢—s) holds. On the other hand, since A CUS (aq.5,) holds if and only if aA © L%; aay, by) holds for cachSection 15: THE OUTER MEASURE GENERATED BY A MEASURE bs a ER and since 2*([a9, be) = A* (ay, by) it follows that 2*(aA) = |ala*(A) foreach a € R. Now, the identities Enad=a(e'£)NA) and EN (aay =a(a ZN At) (a xO, imply AME Maa) + A(E Ma AY) = falfX" (aE) 1.4) + °(@rtEDN AY), which shows thet A is measurable if and only if aA is measurable for cach aeR ‘Now, (a) and (b) follow from the preceding discussion, Problem 15.6, Ler S bea semiring of subsets ofa set X, and let u5 —» [0,00] be a finively additive measure that is not a measure. For each A & X define (as usual) BA) = int Seca (ane 5 and ac Ae] Show by a counterexample that itis possible to have ye 7 1° on S. Why does this not contradict Theorent 15.1? Solution. Consider the finitely additive measure j2 of Problem 13:7. Cleary, AN) = 00. Since N= U2 [i] € 5, we have 4°0N) = Ye, w((n}) = 0, and 50.0 = A*(N) < AON) = 00. “This conclusion does not contradict Theorem 15.1, since the o additivity ofthe ‘measure was essential for is proof. Problem 15.7. Let E be an arbitrary measurable subset of a measure space (X, S, and consider the measure space (E, Sg. v),whereSs = (ENA: & € 8) and VUE 1A) = p°(EN A); see Problem 15.1. Establish the following properties regarding the measure space (E, Se. ¥): ‘a. The outer measure v" is the restriction of u* on E, i.e., v"(B) = 1B) for each BCE, b. The v-measurable sets of she measure space (E, Se, v) are precisely the sets ofthe form EA where A is a u-measurable subset of X,1.2., FOE: Fe Ay),126 (Chapter 3: THE THEORY OF MEASURE Solution, Let (X, 8,1), E,and v be as defined in the problem. (2) Let B be an arbitrary subset of E. If {4p} is a sequence of S satisfying BUR, Ans then note that BC US, ENA, and so vO) s Poe OA) = UME Ay Sato This implies v*(B) < *(B). On the other hand, if {A,] is a sequence of S satisfying B.C Uff £9 Ap, then we have w(B)s s BME NM Ag) = WEN An). ‘Thus, u"(B) < v"(B) also holds, and so v*(B) = 4.%(B) for each subset B of E (b) Let F be a subset of E. Assume first that F is v-measurable, If A € S, then note that, HA) = BANE) + u"(ANE \ B)) = VANE) + HANK) (AN E)NF)+9°((ANE)NE\ F))+u(AN(K\E)) BMAP) + EAN \ FYULANG \ ED) (ANF) + uM (AN \ FI), which shows that F is j-measurable. For the converse, assume that F is «measurable. If A is an arbitrary subset of , then note that vA) = aA) = AAD FP) + u(ANUX \ F)) = W(ANF)+ w(AN{E \ F)) = (ANF) +u(ANE \ F)). Which means that F is also v-measurable, Problem 18.8. Show thata subset E ofa measure space (X, S, :)is measurable ‘and only iffor each ¢ > O there exists a measurable set A, and two subsets Be and C. satisfying Es(AUB)\C. WB) 0 is given, then let A, = E and Be , and note that these sets satisfy the desired properties. For the converse, assume tht for each ¢ >, 0 there exist a measurable set Ay and subsets Bi, and C, satisfying E=(AUB)\ Ce, a(B) <2, and BC) <8. co) Replacing Cy by (4¢UB,) Ce, weean assume that C, isa subset of Ay U Be. From (4), we see that BUC, =A UB, Oo) Now, by Theorem 15.11, there exists a measurable set D, such that By © Ds and ju"(D,) = 2*(Be). Using (+4), we get EUC,U(Dz \ By) = Ac U Be U(Dy \ Be) = As U De: Clearly, Ay U Dy is a measurable set and BCU. \ BD) 5 uCe) + WD») < 26, In other words, the preceding show that for each ¢ > 0 there exist a measurable set F, and asubset G, such that EUG =F and uG) <6. Now, foreach n pick a measurable set Fy anda subset Gy with 2*(G,) < 4 and EUG, = Fy. Clearly the set F = (2 Fy is meastrable, Also, the set G = (2 Gr isamull set—-and hence G \E is also measurable, In view of ar F. Bug =[\eva, ! we see that £ UG is a measurable set. immediately from the identity ally, the measurability of £ follows E=(EUG)\(G\ £)=F\G\®). Problem 15.9. Let (X,S, 2) be a measure space, and let A be a subset of X. Show that ifthere exists a measurable subset E of X such that A & B, w'(E) < 00, land w*(E) = p%(A) + n%(E\ A), then A is measurable128 Chapter: “THE THEORY OF MEASURE Solution. By Problem 15.7, we know that the outer measure generated by the ‘measure space (E, Se, 1") coincides with * and the o-algebra ofall measurable sets of the measure space (E, Sp, u") is {A € sty: AC E). Now, tocomplete the proof, assume E € A,, andthatasubset A of E satisfies H(A) + WE \ A) = wi). If aE) < 00 holds, then it follows from ‘Theorem 15.8 that A isa measurable set for (E, Sp, 1°). Thus, by the preceding discussion, A € Ay. Problem 1510. Let A be a subset of with 2*(A) > 0. Show thar there exists a nonmeasurable subset B of R such that BS A. Solution, If A is nonmeasurable, then there is nothing to prove. So, assume that ‘A ismeasurable. Since some {n,n-+ 1] A musthave nonzero measure (why?), by translating appropriately (and using Problem 15.5), we can also assume that AS {0,1} As in Example 15.13 define an equivalence relation ~ on A by saying that xy whenever x—y isarational number. By the Axiom of Choice, there exists subset B of A containing precisely one member from each equivalence clas. Let (ry, ra...) beanenumeration ofthe rationals of (~1, 1} and let By = r9-+B. ‘Then: The sequence (By) is pairwise disjoint b. 2°(B,) = 4°(B) holds (by Problem 15.5) for each mand ASUS: Be SI-1,21 Novr,note that if B is a measurable set, then each B,, is likewise a measurable set (see Problem 15.5 again). Thus, from (c), it follows that o 0 holds, and so 2*(E,) = A*(E) > 0. In particular, 772, A"(E,} = 90. On the other band, UR, Ey © [1,2] implies 2°(UE, B,) $3 <00 Problem 15.12. Let (X,S, 4) be a measure space, and let (Ay) be a sequence oftsubsets of X such that Ay S Anat holds for all n. If A= Us Any then show that p*(Aq) ¢ HCA), Solution. Choose some E € A with AC E and y*(A) = u°(E). (This is possible by Theorem 15.11.) By the same theorem, for each n there exists some E, € A with A, © E, © E and u*(Aq) = u*(E,) Now, for each m put Fea ( (Eq Ex € Avand then let F = Uy Fx € A. Then, we have: a Ag S Fy and u*(An) = (Fp) foreach mj and bo Fy t F and u*(A) = w"(F). By Theorem 15.4, it follows that An) = Fy) PF) = HA), Problem 15.13. For subsets of a measure space (XS, w) let us define the fol- lowing almost everywhere (ae.) relations: a AC Bae. if uA\B) b. A= Bae. if x°(ASB) =0; c. Ay t Aae. if An © Anni ae forall n and A= U2 Ay ae, (The meaning of Ay 4 A ae. is similar) Generalize Theorem 15.4 by establishing the following properties for a sequence (E,) of measurable sets: i FE, t Bae, then (Es) t w'CE). ii, FE, | Bae. and (Ey) < 00 for some k, then u*(En) 4 w*(E), Is (i) true without assuming measurability for the sets Ex? Solution. (i) Assume that (Z,} is sequence of measurable sets such that Ey E ac, holds, Let a=[(Ce)oe)ufCle\#)] and note that (A) =0. Now, define F = EUA and Fy = EyUA foreach n.