A Companion to Analysis A Second First and First Second Course in Analysis T. W. Korner Author address Trinrry HALL, CAMBRIDGE E-mail address: twk@dpmms .cam.ac uk URL: nttp: //wwwapmms.cam.ac.uk/~twk/1991 Mathematics Subject Classification. 26-01. ABSTRACT. Th a draft for a book. I would therefore be glad to receive corrections at twk@dpmms.cam.ac.uk. Senders of substantial lists of errors or of lists of substantial errors will receive small rewards and large thanks. General comments are also welcome. Please refer to this as DRAFT G2 (note that Appendix K was reordered between drafts and F). Please do not say ‘I am sure someone else has noticed this’ or “This is too minor to matter’. Everybody notices different things and no error is too small to confuse somebody.[Archimedes] concentrated his ambition exclusively upon those specu lations which are untainted by the claims of necessity. These studies, he believed, are incomparably superior to any others, since here the grandeur and beauty of the subject matter vie for our admiration with the cogency and precision of the methods of proof. Certainly in the whole science of ge- ometry it is impossible to find more difficult and intricate problems handled in simpler and purer terms than in his works. Some writers attribute it to his natural genius. Others maintain that phenomenal industry lay behind the apparently effortless ease with which he obtained his results. The fact is that no amount of mental effort of his own would enable a man to hit upon the proof of one of Archimedes’ theorems, and yet as soon as it is explained to him, he feels he might have discovered it himself, so smooth and rapid is the path by which he leads us to the required conclusion. Plutarch Life of Marcellus (Scott-Kilvert’s translation] It may be observed of mathematicians that they only meddle with such things as are certain, passing by those that are doubtful and unknown. They profess not to know all things, neither do they affect to speak of all things. What they know to be true, and can make good by invincible argument, that they publish and insert among their theorems. Of other things they are silent and pass no judgment at all, choosing rather to acknowledge their ignorance, than affirm anything rashly. Barrow Mathematical Lectures For [A.N.] Kolmogorov mathematics always remained in part a sport. But when ... I compared him with a mountain climber who made first ascents, contrasting him with I.M. Gel ‘fand whose work I compared with the building of highways, both men were offended. ‘ ... Why, you don’t think I am capable of creating general theories?” said Andrei Nikolaevich. ‘Why, you think I can’t solve difficult problems?” added I. M. V.L. Arnol’d in Kolmogorov in PerspectiveContents Introduction Chapter 1. The Real Line §L.1. Why do we bother? §1.2. Limits §1.3. Continuity §l4. The fundamental axiom §1.5. The axiom of Archimedes §1.6. Lion hunting §L.7. The mean value inequality §L8. Full circle §1.9. Are the real numbers unique? Chapter 2. A First Philosophical Interlude 99 §2.1. Is the intermediate value theorem obvious? 99 Chapter 3. Other Versions of the Fundamental Axiom The The Bolzano- Weierstrass theorem Some general remarks Chapter 4, Higher Dimensions 23,vi Contents §4.1. Bolzano Weierstrass in higher dimensions AL §4.2. Open and closed sets 46. §4.3. A central theorem of analysis 54 §4.4, The mean value theorem 57 §4.5. Uniform continuity 62 §4.6. The general principle of convergence 64 Chapter 5. Sums and Suchlike 9 71 §5.1. Comparison tests O 71 §5.2. Conditional convergence 9 73 §5.3. Interchanging limits 0 78 §5.4. The exponential function 0 86 §5.5. The trigonometric functions 9 93 §5.6. The logarithm 9 97 §5.7. Powers 0 103 §5.8. The fundamental theorem of algebra 9 107 Chapter 6. Differentiation 115 §6.1. Preliminaries 115 §6.2.. The operator norm and the chain rule 121 86.3. ‘The mean value inequality in higher dimensions 19 Chapter 7. Local Taylor Theorems 133 §7.1. Some one dimensional Taylor theorems 133 §7.2. Some many dimensional local Taylor theorems 137 § Critical points 145 Chapter 8. The Riemann Integral 159 §8.1. Where is the problem ? 159 §8.2. Riemann integration 162 §8.3. Integrals of continuous functions 171 §8.4. Tirst steps in the calculus of variations 9 179 §8.5. Vector-valued integrals 190Contents vii Chapter 9. Developments and limitations of the Riemann integral 9 193 §9.1. Why go further? 193 §9.2. Improper integrals 9 195 §9.3. Integrals over areas 0 199 §9.4. The Riemann-Stieltjes integral 9 204 §9.5. How long is a piece of string? © 211 Chapter 10. Metric Spaces 219 §10.1. Sphere packing © 219 §10.2. Shannon’s theorem 9 222 §10.3. Metric spaces 227 §10.4. Norms and the interaction of algebra and analysis 232 §10.5. Geodesics 9 239 Chapter 11, Complete Metric Spaces 247 §11.1. Completeness 217 §11.2. ‘The Bolzano-Weierstrass property 255 §11.3. The uniform norm 259 §11.4. Uniform convergence 263 §11.5. Power series 71 §11.6. Fourier series 0 281 Chapter 12. Contraction Mappings and Differential Equations 285 §12.1. Banach’s contraction mapping theorem 285 §12.2. Existence of solutions of differential equations 287 §12.3. Local to global 9 292 §12.4. Green’s function solutions 9 299 Chapter 13. Inverse and Implicit Functions 309 $13.1. ‘Lhe inverse function theorem 309 §13.2. The implicit function theorem © 318 §13.3. Lagrange multipliers 9 Chapter 14. Completion 333viii Contents §14.1. What is the correct question? §14.2. The solution §14.3. Why do we construct the reals? 9 §14.4. How do we construct the reals? 9 §14.5. Paradise lost? OO Appendix A. Appendix B. Appendix C Appendix D. Appendix E. Appendix F. Appendix G. Appendix H. Appendix I. Appendix J. Appendix K. Bibliography Index Ordered Fields Countability The care and treatment of counterexamples A More General View of Limits Traditional Partial Derivatives Another approach to the inverse function theorem Completing Ordered Fields Constructive Analysis Miscellany Executive Summary Exercises 333, 339 342 346 352 355 359 363 369 375 381 389 395 401Introduction In his autobiography [12], Winston Churchill remembered his struggles with Latin at school. ‘... even as a schoolboy I questioned the aptness of the Classics for the prime structure of our education, So they told me how Mr Gladstone read Homer for fun, which I thought served him right.’ ‘Naturally’ he says ‘I am in favour of boys learning English. I would make them all learn English; and then I would let the clever ones learn Latin as an honour, and Greek as a treat.’ This book is intended for those students who might find rigorous anal- ysis a treat. The content of this book is summarised in Appendix J and corresponds more or less (more rather than less) to a recap at a higher level of the first course in analysis followed by the second course in analysis al Cambridge in 2003 gether with some material from various methods courses (and thus corresponds to about 60 to 70 hours of lectures). Like those cout it aims to provide a foundation for later courses in functional analysis, differential geometry and measure theory. Like those courses also, it assumes complementary courses such as those in mathematical methods and in elementary probability to show the practical uses of calculus and strengthen computational and manipulative skills. In theory, it starts more or less from scratch but the reader who finds the discussion of section 1.1 baflling or the e, 5 arguments of section 1.2 novel will probably find this book unrewarding. great but empty Opera House. The guide is enthusiastic but interested only i , Tighling and stage machinery. If you wish lu see Uhe you must come at another time and with a different guide. ixx Introduction Although I hope this book may be useful to others, I wrote it for students to read either before or after attending the appropriate lectures. For this reason, I have tried to move as rapidly as possible to the points of difficulty, show why they are points of difficulty and explain clearly how they are overcome. If you understand the hardest part of a course then, almost automatically, you will understand the easiest. The converse is not true. In order to concentrate on the main matter in hand, some of the sim- pler arguments have been relegated to exercises. The student reading this book before taking the appropriate course may take these results on trust and concentrate on the central arguments which are given in detail. The student reading this book after taking the appropriate course should have no difficulty with these minor matters and can also concentrate on the cen- tral arguments. I think that doing at least some of the exereises will help students to ‘internalise’ the material but I hope that even students who skip most of the exercises can profit from the rest of the book. T have included further exercises in Appendix K. Some are standard. some form commentaries on the main text and others have been taken or adapted from the Cambridge mathematics exams. None are just ‘makeweights’, they are all intended to have some point of interest. I have tried to keep to standard notations but a couple of notational points are mentioned in the index under the heading notation. T have not tried to strip the subject down to its bare bones. A skeleton is meaningless unless one has some idea of the being it supp: and that being in turn gains much of its significance from its interaction with other beings, both of ils own species and of ulher species. Fur this reason, I have included several sections marked by a Y. These contain material which is not necessary to the main argument but which sheds light on it. Ideally, the student should read them but not study them with anything like the same attention which she devotes to the unmarked sections. There are two sections marked © which contain some, very simple, philosophical discussion. It is entirely intentional that removing the appendices and the sectious marked with a 9 more than halves the length of the book. Tlearned calculus from the excellent Calculus [18] of D. R. Dickinson and its inspiring author. My first glimpse of analysis was in Hardy’s Pure Math- ¢ (2. y understand it. I nm Low: elementary analysis from Ferrar’s A Textbook of Convergence [18] (an ex- cellent book for those making (he tau ly, now, unfortunately, out of print) and Burl A First Course in Mathematical Analysis [10]. The books of Kolmogorov and Fomin (31 and, particularly, Dieudonné [14] showed me that analysis is not a collection of theorems but a single coherent theory. Stromberg’s book An Introduction to Classical Real ion from school to univerIntroduction xi Analysis [46] lies permanently on my desk for browsing. The expert will easily be able to trace the influence of these books on the pages that follow. If, in turn, my book gives any student half the pleasure that the ones just cited gave me, I will feel well repaid. Cauchy began the journey that led to the modern analysis course in his lectures at the Ecole Polytechnique in the 182 The times were not propitious. A reactionary government was determined to keep close control over the school. The faculty was divided along fault lines of politics, religion and age whilst physicists, engineers and mathematicians fought over the contents of the courses. The student body arrived insufficiently prepared and then divided its time between radical polities and worrying about the job market (grim for both staff and students). Cauchy’s course was not popular!, Everybody can sympathise with Cauchy’s students who just wanted to pass their exams and with his colleagues who just wanted the standard material taught in the standard way. Most people neither need nor want to know about rigorous analysis. But there remains a small group for whom the ideas and methods of rigorous analysis represent one of the most splendid triumphs of the human intellect. We echo Cauchy's defiant preface to his printed lecture notes. As to the methods [used here], I have sought to endow them with all the rigour that is required in geometry and in such a way that I have not had to have recourse to formal ma- nipulations. Such arguments, although commonly accepted . cannot be considered, it seems to me, as anything other than [suggestive] to be used sometimes in guessing the truth. Such reasons [moreover] ill agree with the mathematical sci- ences’ much vaunted claims of exactitude. It should also be observed that they tend to attribute an indefinite extent to algebraic formulas when, in fact, these formulas hold under certain conditions and for only certain values of the variables involved. In determining these conditions and these values and in settling in a precise manner the sense of the notation and the symbols I use, I eliminate all uncertainty. ... It is sometimes find myself forced to depend on several propos tions that perhaps seem a little hard on first encounter ... . But, those who will read them will find, I hope, that such propositions, implying the pleasant necessity of endowing the theorems with a greater degree of precision and restricting 1Bethoste’s splendid biography [4] gives the fascinating, details,xii Introduction statements which have become too broadly extended, will actually benefit analysis and will also provide a number of topics for research, which are surely not without importance.Chapter 1 The Real Line 1.1. Why do we bother? It is surprising how many people think that analysis consists in the difficult proofs of obvious theorems. All we need know, they say, is what a. limit is, the definition of continuity and the definition of the derivative. All the rest is ‘intuitively clear’! If pressed they will agree that the definition of continuity and the defini- tion of the derivative apply as much to the rationals Q as to the real numbers IR. If you disagree, take your favorite definitions and examine them to see where they require us to use & rather than Q. Let us examine the workings of our ‘clear intuition’ in a particular case. What is the integral of t?? More precisely, what is the general solution of the equation () We know that f°/3 is a solution but, if we have been well taught, we know that this is not the general solution since eB ©) IO-zrs with ¢ any constant is also a solution. Is (#*) the most general solution of (#)? If the reader thinks it is the most general solution then she should ask herself why she thinks it is. Who told her and how did they explain it? 1A good example of this view is given in the book [9]. The author cannot understand the problems involved in proving results like the intermediate value theorem and has written his book to share his lack of understanding with a wider audience,2 1. The Real Line Figure 1.1. A continuous and differentiable function over the rationals If the reader thinks it is not the most general solution, then can she find another sulution?: After a little thought she may observe that if g(t) is a solution of (+) and we set f= 90-5 then f/(t) = 0 and the statement that (**) is the most general solution of (+) reduces to the following theorem. Theorem 1.1. (Constant value theorem.) If f : 2 — B is differentiable and f'(t) =0 for allt €R, then f is constant. If this theorem is ‘intuitively clear” over R it ought to be intuitively clear over @. The same remark applies to another ‘intuitively clear’ theorem. Theorem 1.2. (The intermediate value theorem.) If f : R + E is continuous, b >a and f(a) >0> f(b), then there exists ac with b > c>a. such that f . However, if we work over @ both putative theorems vanish in a puff of smoke. Example 1.3. If f :Q— Q is given by f(@)=-1 ife’? <2, f()=1 otherwise, then. (i) f és a continuous function with f(0) = —1, f(2) =1, yet there does not exist ac with f(c) =0, (ii) f is a differentiable function with f'(c) =0 for all «, yet f is not constant.1.2. Limits 3 Sketch proof. We have not yet formally defined what continuity and dif ferentiability are to mean. However, if the reader believes that f is discon- tinuous, she must find a point « € @ at which f is discontinuous. Similarly, if she believes that f is not everywhere differentiable with derivative zer she must find a point © € Q for which this statement is false. The reader will be invited to give a full proof in Exercise 1.16 after continuity has been formally defined. a The question ‘Is («*) the most general solution of (*)?” now takes on a more urgent nole. Of course, we work in B and wut in Q but we are tempted to echo Acton ([1], end of Chapter 7) This example is horrifying indeed. For if we have actually seen une Liger, is nol Uke jungle immediately filled with ligers, and who knows where the next one lurks. Here is another closely related tiger Exercise 1.4. Continuing with Example 1.3, set g(t) =t + f(t) for all t. Show that g!(t)—1>0 for all t but that g(-8/5) > 9(-6/5). Thus, if we work in Q, a function with strictly positive derivative need not be increasing. Any proof that there are no tigers in R must start by identifying the dif- ference between R and Q which makes calculus work on one even though it fails on the other. Both are ‘ordered fields’, that is, both support operations of ‘addition’ and ‘multiplication’ together with a relation ‘greater than’ (‘or der’) with the properties that we expect. I have listed the properties in the appendix on page 355 but only to reassure the reader. We are not interested in the properties of general ordered fields but only in that particular prop- erty (whatever it may be) which enables us to avoid the problems outlined in Example 1.3 and so permits us to do analysis. 1.2. Limits Many ways have been tried to make calculus rigorous and several have been successful. We choose the first and most widely used path via the notion of a limit. In theory, my account of this notion is complete in itself. However, my treatment is unsuitable for beginners and I expect my readers to have substantial experience with the use and manipulation of limits. Throughout this section F will be an ordered field. The reader will miss nothing if she simply considers the two cases F = R and F = Q. She will, however, miss something if she fails to check that everything we say applies to both cases equally.4 1. The Real Line Definition 1.5. We work in an ordered field F. We say that a sequence a1, ag, ... tends to a limit a as n tends to infinity, or more briefly Gm > a as 0O if, given any > 0, we can find an integer no(e) [read ‘no depending on €’] such, that Jan — a| <€ for all n > nol). The following properties of the limit are probably familiar to the reader. Lemma 1.6. We work in an ordered field ¥. (i) The limit is unique. That is, if an > a and ay +b as n= 00, then a=b. (ii) If an > a as n> 00 and 1 < n(1) 4 (iv) If an > a and by > b as n > 0, then an + bp + atd. (v) If an > a and by > b as n = 00, then anbp > ab. (vi) Suppose that ay > a asn— co. If an #0 for each n and a #0, then a} + a}. (vii) If ay B for each n and by, + b, as n — 00 then b > B. Proof. I shall give the proofs in detail but the reader is warned that similar proofs will be left to her in the remainder of the book. (i) By definition: Given e > 0 we can find an nj(e) such that |an — a| <€ for all n > m4(€). Given € > 0 we can find an no(¢) such that Jan — 6] < € for all n > no(e). Suppose, if possible, that @ # 6. Then setting ¢ — |a — b|/3 we have ¢ > 0. If N = max(nj(e), na(e)) then a—b| < lay —a| + lay —b| 0 we can find an nj(e) such that Jan — a| <€ for all n > n4(¢), Let € > 0. Since n(j) > j (proof by induction, if the reader demands a proof) we have |ay(j) — a] <€ for all j > mi(e). The result follows.1.2. Limits 5 (iii) Let € > 0. Taking ni(e) = 1 we have la, —cl =O nj(e). The result follows. (iv) By definition, %* holds as does (ae) Given € > 0 we can find an n2(¢) such that |by — b| < for all n > no(e) Observe that I(@n + Bn) — (4 + 5)) = [Can — @) + On — 5)| S [an — a] + [On — 8). Thus if € > 0 and ng(c) = max(ns(c/2), na(€/2)) we have (dn + bn) — (a +B)| < Jan — al + [bp — 5] < €/2+€/2=€ for all n > ng(c). The result follows. (v) By definition, % and %% hold. Let « > 0. The key observation is that (1) [auby = ab] < [abn dnb + [and — ab] = Janl|bn — Ol + [lay — a If n > mi(1) then lay — a| <1 80 |an| < Ja| +1 and (1) gives @) andy ~ ab] < (la| + 1)]Pn — | + [Ollan — a. Thus setting? ns(c) = max(n,(1), ri(c/(2(/8| + 1)), m2(¢/(2{a| + 1))) we see from (2) that |anbn — ab] <€/2+€/2=€ for all n > ng(e). The result follows. (vi) By definition, # holds. Let € > 0. We observe that 14) _la~ aul a a] [allan| Since a # 0 we have |a|/2 > 0. If n > ni(lal/2) then |a,, — lan| > |a|/2 and (3) gives @) 1 1. la @) a al~ ae Thus setting ng(e) = max(n1(lal/2), r1(a2e/2)) we see from (4) that 11 —--| ng(e). The result follows. 2The reader may ask why we use m1(¢/(2({b| + 1)) rather than nx (¢/(2[b))). Observe first that we have not exchided the possibility that 6 — 0. More importantly, observe that all we are required to do is to find an na(e) that works and is futile to seek a. ‘best? na(e) in these or similar circumstances.6 1. The Real Line (vii) The proof of the first sentence in the statement is rather similar to that of (i). By definition, % holds. Suppose, if possible, that a > A, that is, a— A> 0. Setting N = nj(a — A) we have ay = (ay —a)+a>a-|ay -—a| >a (aA) =A, contradi ing our hypothesis. 'The result follows by reduction ad absurdum. To prove the second sentence in the statement we can either give a similar argument or set an = —Bn, a= —b and A= —B and use the first sentence, [Your attention is drawn to part (ii) of Exercise 1.81] . Exercise 1.7. Prove that the first few terms of a sequence do not affect convergence. Formally, show that if there exists an N such that an = bn for nN then, an > a as n> 00 tmplics by > a as'n > 00. Exercise 1.8. In this rcise we work within Q. (The reason for this will appear in Section 1.5 which deals with the aciom of Archimedes.) (i) Observe that if e€ Q and € > 0, then e = m/N for some strictly pos- itive integers m and N. Use this fact to show, directly from Definition 1.5, that (if we work in Q) 1/n = 0 as n= 00 (ii) Show, by means of an example, that, if ay + a and ay, > b for all rn, it does not follow that a > b. (In other words, taking limits may destroy strict inequality.) Does it follow that a > b? Gine reasons. Exercise 1.9. A more natural way of proving Lemma 1.6 (i) is to split the argument in two. (i) Show that if |a — 6| <¢ for alle > 0, then a=b. (ii) Show that if an + @ and an > b as n > 00, then |a — | <€ for all e>0. (iii) Deduce Lemma 1.6 (i). (iv) Give a similar ‘split proof’ for Lemma 1.6 (vii). Exercise 1.10. Here is another way of proving Lemma 1.6 (v). I do not claim that it is any simpler, but it introduces a useful idea fi 7 4) Sh (i) Shs that, if a that, if an th n (ii) Show from first principles that, if an + a as n > 00, then a2 = a”. (iii) Use the relation xy = ((« + y)? —(«—y))/A together with (ii), (i) and Lemma 1.6 (iv) to prove Lemma 1.6 (v). The next result is sometimes called the sandwich lemma or the squeeze lemma.1.3. Continuity 7 Exercise 1.11. Suppose am = cn = bm for all m. Then, if an > ¢ and by = ¢, it follows that ey > ¢ as n+ co Suppose |am| > cm) > \bm| for all m and that an + ¢ and by + ¢ as n — oo. Does it follow that en + c? Give a proof or counterecample as appropriate. 1.3. Continuity Our definition of continuity follows the same line of thought. Definition 1.12. We work in an ordered field F. Suppose that E is a subset of F and that f is some function from E to F. We say that f is continuous at x € F if given any e > 0 we can find So(e, x) fread Sq depending one and with 5o(€,2) > 0 such that |f(2) — F(y)| < for all y € E such that |x — y| < do(e, 2). If f is continuous at every point of B we say that f : E+ F is a continuous function. ‘The reader, who, I expect, has seen this definition before, and is, in any case, a mathematician, will be anxious to move on to see some theorems and proof. Nou-mathermaticians might object that our definition does not correspond to their idea of what continuous should mean. If we consult the dictionary we find the definition ‘connected, unbroken; uninterrupted in time or sequence: not discrete’. A mathematician would object that this merely defines one vague concept in terms of other equally vague concepts. However, if we rephrase our own definition in words we see that it becomes “f is continuous if f(y) is close to f(c) whenever y is sufficiently close to w and this clearly belongs to a different circle of ideas from the dictionary definition. This will not be a problem when we come to define differentiability since there ig no ‘common sense’ notion of differentiability. In the same way the existence of a ‘common sense’ notion of continuity need not trouble us provided that whenever we use the word ‘continuous’ we add under our breath ‘in the mathematical sense’ and we make sure our arguments make no appeal (open or disguised) to ‘common sense’ ideas of continuity. Here are some simple properties of continuity. Lemma 1.13. We work in an ordered field F. Suppose that E is a subset of F, that « € E, and that f and g are functions from E to F. @ If f(x) =e for all x € E, then fis continuous on EB. (ii) If f and g are continuous at x, then so is f +9.8 1. The Real Line (iii) Let us define f x gi E =F by f x g(t) = f(Og(t) for allt e B. Then if f and g are continuous at x, so is f xg. (iv) Suppose that f(t) #0 for all t € E. If f is continuous at x so is 1/f. Proof. Follow the proofs of parts (iii) to (vi) of Lemma 1.6. a By repeated use of parts (ii) and (iii) of Lemma 1.13 it is easy to show that polynomials P(t) = Sc". art" are continuous. The details are spelled out in the next exercise. Exercise 1.14. We work in an ordered field F. Prove the following results. (i) Suppose that E is a subset of F and that f : E > F is continuous at «e¢E. Ife E' CE then the restriction f\g of f to E’ is also continuous at (ii) If J: F F is defined by J(x) = x for all x € F, then J is continuous on F. (iii) Every polynomial P is continuous on F. (iv) Suppose that P and @ are polynomials and that () is never zero on some subset E of F. Then the rational function P/Q is continuous on E (or, more precisely, the restriction of P/Q to E is continuous.) The following result is little more than an observation but will be very useful. Lemma 1.15. We work in an ordered field F. Suppose that E is a subset of I, that « € H, and that f is continuous ate. If ty € B for all n and aq > asm — 00, then f(tn) > f(x) asn = 00 Proof. Left to reader. a We have now done quite a lot of what is called ¢, 6 analysis but all we have done is sharpened our proof of Example 1.3. The next exercise gives the details. Exercise 1.16. We work in Q. The function f is that defined in Exam- ple 1.3. fi) Show that the equation. x? = 9 has no solution (See any elementary text on number theory or Exercise K.1.) (i) Af |x| <2 and |n| < 1 show that \(a +n)? (iii) If x2 <2 and 5 = (2—27)/6 show that y? <2 whenever |x—yl <6. Conclude that f is continuous at x. (iv) Show that if x? > 2 then f is continuous at x. 2") <5|n| (v) Conclude that f is a continuous function.1.4, The fundamental axiom 9 Unless we can isolate the property that distinguishes the rationals from the reals we can make no progress. 1.4. The fundamental axiom The key property of the reals, the fundamental aciom which makes every- thing work, can be stated as follows: The fundamental axiom of analysis. If ay € R for eachn > 1, AER and ay < a2 < a3 <... and ay 0 > f(b) then there exists a ¢ C [a,b] such that flo) =0 (The proof will be given as Theorem 1.35.) Exercise 1.18. Assuming Theorem 1.17 prove the apparently more general result: If f : {a,b] > R is continuous and f(a) >t > f(b) then there exists a c€ [a,b] such that f(c) =t. How might our programme of obtaining the intermediate value theorem from the fundamental axiom fail? (1) The reader has higher standards of rigour than I do but can fill in the gaps herself. For example, in the statement of Theorem 1.17, I do not explicitly say that 6 > a. Again, I talk about the ‘algebraic’ properties of R when, strictly speaking, a set cannot have algebraic properties and I should refer instead to the algebraic properties of (, +. x.>). Such problems may annoy the reader but do not cause the programme to fail. (2) As is almost certainly the case, my proofs contain errors or have gaps which do not occur in other accounts of the material and which can thus be corrected. In such a case, I apologise but it is I who have failed and not the programme.10 1. The Real Line (3) My proofs contain a serious error or have a serious gap which occurs in all accounts of this material. If this can be corrected then the programme survives but looks a great deal more shaky. If a serious error has survived for a century who knows what other errors may lurk. (4) All accounts contain an error which cannot be corrected or a gap that cannot be filled. The programme has failed. We start our attempt with a simple consequence of the fundamental axiom, Lemma 1.19. in R every decreasing sequence bounded below tends to a limit. Proof. Observe that if a1, a2, a3, ... is a decreasing sequence bounded below then —a1, —a2, —a3 n increasing sequence bounded above. We leave the details to the reader as an exercise . Exercise 1.20. (i) If mi, mo, ... is an increasing sequence of integers bounded above, show that there exists an N such that mj; my for all g2N (ii) Show that every non-empty set A C Z bounded above has a maa imum. More formally, show that if A © Z, A # 0 and there exists a K such that K > a whenever a € A then there exists an ag € A with ag > a whenever a € A, 1.5. The axiom of Archimedes Our first genuinely ‘analysis Theorem 1.21. (Axiom of Archimedes.) result may strike the reader as rather odd. 1 ——0asn— 00 n Proof. Observe that the 1/n form a decreasing sequence bounded below. Thus, by the fundamental axiom (in the form of Lemma 1.19), 1/n tends to some limit |. To identify this limit we observe that since the limit of a product is a product of the limits (Lemma 1.6 (v)) 1 iit >... ubsequence is the limit of the sequence (Lemma 1.6 (ii)) i sol Qn as n — oo. Thus, by the uniqueness of limits (Lemma. 1.6 (i)), 1 = 1/2 so 1=0 and 1/n = 0 as required. and since the limit of a1.5, The axiom of Archimedes 11 Exercise 1.22. [Exercise 1.8 (ii) concerned Q. We repeat that exercise but this time we work in R.] Show, by means of an example, that, if an > a and ay, > b for all n, it does not follow that a > b. (In other words, taking limits may destroy strict inequality.) Does it follow that a > b? Give reasons. Theorem 1.21 shows that there is no ‘exotic’ real number J say (to choose an exotic symbol) with the property that 1/n > J for all integers n > 1 and yet J > 0 (that is ] is smaller than all strictly positive rationals and yet strictly positive). There exist number systems with numbers (the famous ‘non-standard analysis’ of Abraham Robins: ‘surreal numbers’ of Conway constitute two such systems) bu rationals are, in some sense, too small a system for the standard theorems of analysis to hold so these non-Archimedean systems are, in some sense, too big. Eudoxus and Archimedes? realised the need for an axiom to show that there is no exotic number “1 bigge: than any integer T>n n for all 1 /. However, in spite of its name, what was an axiom for Eudoxus and Archimedes is a theorem for us. Exercise 1.23. (i) Show that there does not exist a K € 2 with K > n for all n € Z by using Theorem 1.21. (ii) Show the same result directly from the fundamental axiom. Exercise 1.24. (i) Show that if a is real and 0 < a <1 then a” — 0 as n— 0. Deduce that if a is real and \a| <1 then a” +0. (ii) Suppose that a is real and a # —1. Discuss the behaviour of ita as n—+ 00 for the various possible values of a [Hint (1 = a®)/(14a") = (a = 1) f(a" +.) Here is an important consequence of the axiom of Archimedes. Exercise 1.25. (i) Use the fact that every non-empty set of integers bounded above has a maximum. (see Exercise 1.20) to show that, if x € R, then there exists an integer m such that m 0, there exists ay € Q such that |x —y| 00 as n = 00 if, given any real K, we can find an no(K) such that ty >K for alln > no(K) Exercise 1.28. Using Exercise 1.23 show that n > 00 as n — 00. Exercise 1.28 shows that two uses of the words ‘n tends to infinity’ are consistent. It is embarrassing to state the result but it would be still more embarrassing if it were false. Ilere is a another simple exercise on Definition 1.27. Exercise 1.29. Let a1, a, ... be a sequence of non-zero real numbers. Show hai, if un — 00, then i/an > 0. Is the converse irue? Give a proof or counterecample. Exercise 1.30. It is worth noting explicitly that ordered fields may satisfy the aziom of Archimedes but not the fundamental axiom. Show in particular that the rationals satisfy the axiom of Archimedes. (This is genuinely easy s0 do not worry if your answer és brief.) Exercise 1.31. The reader may be interested to see an ordered field con- taining Z which does not satisfy the axiom of Archimedes. We start by considering polynomials P(X) = Ng anX" with real coefficients ay and form the set K of rational functions P(X)/Q(X) where P and Q are poly- nomials and Q is not the zero polynomial (that is Q(X) = YM gb X™ with byy #0 for some M). Convince yourself that, if 2 use the usual stan- dard formal algebraic rules for manipulating rational functions, then IK is a field (that is, it satisfies conditions (A1) to (D) as set out in the axioms on page 355) To produce an order on K we define the set P to consist of all quotients of the form with ay, bar #0 and aby > 0. Convince yourself that this is a consis- tent definition (remember that the same quotient will have many different representations; P(X)/Q(X) = R(X)P(X)/R(X)Q(X) whenever R(X) is anon-zero polynomial) and that P satisfies conditions (P1) to (P3). If we de- fine P(X)/QuU(X) > Pa(X)/Qo(X) whenever Pi(X)/Qu(X)—Po(X)/Qo(X) €1.5, The axiom of Archimedes 13 P condition (P4) is automatically satisfied and we have indeed got an ordered field. We note that the elements of K of the form a/1 with a € R can be identified in a natural way with B. If we make this natural identification, K. contains Z. To see that the aviom of Archimedes fails, observe that 1/n > 1/X > 0 for alln€Z,n>1. Of course, since the axiom of Archimedes fails, the fundamental axiom fails. By examining the proof of Theorem 1.21, show that the 1/n form a decreasing sequence bounded below but not tending to any limit. If the reader knows some modern algebra she will see that our presen- tation can be sharpened in various ways. (It would be better to define K using equivalence classes. We should take greater care over checking consis- tency. The words ‘identified in a natural way’ should be replaced by ‘there is an isomorphism of (R,+,X,>) with a subfield of K’.) Such readers should underiuke the shurpeniny us un instructive ewercise. Exercise 1.32. We shall not make any essential use of the decimal expan- sion of the real numbers but it is interesting to see how it can be obtained. Let us write D={n€Z:9>n>0} (i) If © € 1D show that (ii) If 23 € D show that YM, xj10~ converges to a limit x, say, as N + 00. Show that U bar. If Dp wid — a and DY 8 i the precise necessary and sufficient condition for equality and prove it. bus N > &w show thai a Give Exercise 1.33. It seems, at first sight, that decimal expansion gives a nat- ural way of treating real numbers. Il is nol impossible to do Uiings in this way, but there are problems. Here is one of them. Let 0 N but a #1 we set co £2) = SF rag.) 1109 a Otherwise we set f(x) = 0. Show that given any y € [0,1], any € > 0 and any t € (0,1) we can find an x € (0,1) with |x — y| <€ such that f() = In other words f takes every value in [0,1] arbitrarily close to every point. (ii) Show that if 0 < a 0 and any #€ [0,1], we can find an « with |x —y| <€ with f(x) =t (iv) (This may require a little thought.) Find ag: B + B such that given any y EIR, any €> 0 and any t € R we can find an x with |x —y| <€ such that f(x) = t. Although decimal expansion is a very useful way of representing numbers it is not the only one. In Exercises K.13 and K.14 we discuss representation by continued fractions. 1.6. Lion hunting Having dealt with the axiom of Archimedes, we can go on at once to prove the intermediate value theorem. Theorem 1.35. (The intermediate value theorem.) We work in R. If f : [a,b] — B ts continuous and f(a) > 0 > f(b), then there exists a c€ [a,b] such that f(c) =0. Proof. Since the method of proof is important to us, I shall label its three main parts.1.6. Lion hunting 1 Part A Set ag = @ and by = b. We observe that f(ao) > 0 > f(bo). Now set co = (ag + bo)/2. There are two possibilities. Either f(cg) > 0, in which case we set a; = cy and 6) = bo, or f(co) < 0, in which case we set a) = ap and b, = cp. In either case, we have f(a1) 20> f(b); ag Sa Si < bo, and by — a1 = (bo — a9) /2. Continuing inductively we can find @ sequence of pairs of points uy, and by, such that F(an) = 0> f(bn); Ap—1 Sdn S Bp S Pars and by — an — (ba—1 — @n—1)/2, Part B We have ap < a1 < +++ < ay < bp 80 that the a, form an increasing sequence bounded above. By the fundamental axiom there is real number c, say, such that ay cas n = 00. Since @ = ao < dn < by we havea 0 for all n, it follows that f(c) > 0. A similar argument applied to the 6, shows that f(c) < 0. Since 0 < f(e) < 0, it follows that f(c) = 0 and we are done. / Exercise 1.36. (i) Give the complete details in the inductive argument in Part A of the proof of Theorem 1.95 above (ii) Give the details of the ‘similar argument applied to the by,’ which shows that f(c) <0. (iii) We use narious parts of Lemma 1.6 in our Theorem. 1.95. Tdentify the points where we use Lemma 1.6. Exercise 1.37. (i) Think how the argument used to prove Theorem 1.35 applies to [a,b] = [0,1], f(#) =2—4«?. (You are not asked to write anything though you may well choose to draw a diagram.) (ii) Think also how the argument used to prove Theorem 1.35 applies to (a, 6] = [0,1], f(@) = (1 — 52) (2 — 5e)(3 — 5a).16 1. The Real Line The method used to prove Theorem 1.35 is called ‘Lion hunting’. The method is also called ‘successive bisection’, ‘bisection search’ or simply ‘bi- section’. Let us summarise the proof. In Part A we have evidence of a lion in the interval [@n—1, ni]. We split the interval into two halves [an—1,¢n—1] and [en—1;n—1] and show that, since there is evidence of a lion in the interval [an-1, bn—1), either there is evidence of a lion in [an—1, ¢n—1}, in which case we take [an, a] = [an—1 Cai]; OF, if there is no evidence of a lion in [@n—1,€n—1] (this does not mean that there are no lions in [an—1,¢n—1], simply that we do not have evidence of one), then it follows that there must be evidence of a lion in [é,—1,bn—1] and we take [an, bn] = [en—1,dn—1)- In Part B we use the fundamental axiom of analysis to show that these successive bisections ‘close in’ on a point © which we strongly suspect of being a lion. Finally in Part C we examine the point c to make sure that it really is a lion. (It might be a wolf or a left handed corkscrew.) orem 1.35. If we omit the condition f(a) > 0 > f(b), then we cannot even start Part A of the argument. The example [a,b] = (0,1), [() = 1 shows that the conclusion may indeed be false. If we have f(a) > 0 > f(b) but replace R by another ordered field for which the fundamental axiom does not hold, then Part A goes through perfectly but Part B fails. Example 1.3 with which we started shows that the conclusion may indeed be false. (Take [a,0] = [-2, 0].) If we have f(a) > 0 > f(b) and we work over R but we do not demand f continuous then Part C fails. Working over R we may take {a,6] = [0,1] and define f(x) = 1 for « < 1/3 and f(r) = —1 for ¢ > 1/3. Parts A and B work perfectly to produce c = 1/3 but there is no lion (that is, no zero of Pale. Exercises 1.38 to 1.40 are applications of the intermediate value theorem. Exercise 1.38. Show that any real polynomial of odd degree has at least one root. Is the result true for polynomials of even degree? Give a proof or counterexample. considering f(«e) = g(x) —, or otherwise, show that there evists ac € [0,1] with g(c) = ¢. (Thus every continuous map of |0,1| into itself has a fived point.) ise 1.88. Suppose that g . [0,1] — [0,1] is a continuous function. By “The name probably comes from A Contribution to the Mathematical Theory of Big Game Hunting by H. Pétard. This sauib is reprinted in (8)1.7. The mean value inequality 17 Give an example of a bijective continuous function k : (0,1) + (0,1) such that k(x) # x for all x € (0,1). Give an ecample of a bijective contin- uous function L: RR such that U(c) 4 # for alle € R. Give an ecample of a bijective (but, necessarily, non-continuous) func- tion h: [0,1] + [0,1] such that h(x) # x for all « € (0, 1). [Hint: First find « function H : [0,1] \ {0, 1, 1/2} — [0,1] \ {0, 1, 1/2} such that H(2) # ¢.] Exercise 1.40. Every mid-summer day at sic o’clock in the morning, the youngest monk from the monastery of Damt starts to climb the narrow path up Mount Dipmes. At six in the evening he reaches the small temple at the peak where he spends the night in meditation. At six o'clock in the morning on the following day he starts downwards, arriving back at the monastery at sic in the evening. Of course, he does not always walk at the same speed. Show that, none the less, there will be some time of day when he will be at the same place on the path on both his upward and downward journeys. Finally we give an example of lion hunting based on trisecting the interval rather than bisecting it. Exercise 1.41. Suppose that we have a sequence «1, #2, ... of real num- bers. Let [ao, bo] be any closed interval. Show that we can find a sequence of pairs of points a, and bj, such that either apy > by + (bn—1 — @n—1)/3 Or fp S dn — (bn—1 — @n—1)/3, Gy—1 Sdn S bn S baa, and by — dy = (bn—1 — an-1)/3; for alln > 1. Show that a, and b, tend to some limit c € {ap, bo]. Show further that, for each n > 1, either an > 6+ (bn—1 — @n—1)/3 Or tn < E= (bn—1 = dn—1)/3 and so in particular tn # ¢. Thus we cannot write the points of [an.bo] as a sequence. (We say that [a0,bo] is uncountable. The reader may know a proof of this via decimal expansions.) Appendix B and Exerci B. 7 within that pad 1.7. The mean value inequality Having disposed of one of the three tigers with which we started, by proving the intermediate value theorem, we now dispose of the other two by using the ‘mean value inequality’.