COURANT VOLUME T <% h-1 DIFFERENTIAL & INTEGRAL CALCULUS /S\ TIAL ALCl ^^ } FFEREN GRAL C VOLUME I HH W © H fc w A8 eau R. COURANT BLACKIE 4 DIFFERENTIAL AND INTEGRAL CALCULUS by R. COURANT Volume I A novel and masterly text- book on the Calculus for students of Mathematics, Physical Science, or Engineering. The author, a mathematical writer of interna- tional repute, has aimed at avoiding the two extremes of slipshodncss and repellent formalism. By skilful arrangement of the material and suitable use of intuition he has succeeded in combining accuracy with unusual attractiveness of presentation. i 40/- NET FROM BLACKIE'S LIST The Mathematical Discoveries of Newton, By H. W. TurnbuU, M.A., F.R.S. 6s, 6d, net. The Absolute Differential Calculus. (Cal- culus of Tensors.) By Tullio Levi- Ci vita, Professor of Rational Mechanics in the Uni- versity of Rome, Fellow of R 'Accademia Nazionale dei Lincei. Edited by Dr. Enrico Persico. Authorized translation by Miss M. Long, late scholar of Girton College, Cam- bridge. 301. net. Theory and Application of Infinite Series . By Dr. Konrad Knopp, Professor of Mathe- matics in the University of Tubingen. Second English Edition. 501. net. The Calculus of Observations. A Treatise on Numerical Mathematics, By E. T. Whittaker, LL.D., Sc.D., F.R.S. ; and George Robinson, M.A., B.Sc, Fourth Edition. 351. net, A Simplified Presentation of Einstein's Unified Field Equations. By Tullio Levi-Civita. Authorized translation by John Dougall, M.A., D.Sc, F.R.S.E. 3j.net Mathematical Notes. By C. F, G. MacDermott, M.A. Intended to help in the preparation and revision for such examina- tions as " Mathematics for Science " at " A " level, the Qualifying Examination for the Mechanical Sciences Tripos, " Higher Mathematics " for the Services Examination, etc 31. 6i, Examples in Applied Mathematics. By F. W. Keliaway. For students at the G.C.E. Advanced Level and O.N.C., &c. Second Edition js. 6d. net. Prlmedin Grist Britain DIFFERENTIAL AND INTEGRAL CALCULUS BLACKIE & SON LIMITED 16/18 William IV Street, Charing Cross, London, W.Ga 17 Stanhope Street, Glascow BLACKIE & SON (INDIA) LIMITED 103/5 Fort Street, BOMBAY BLACKIE & SON (CANADA) LIMITED Toronto DIFFERENTIAL AND INTEGRAL CALCULUS BY R. COURANT Professor of Mathematics in New York University TRANSLATED BY E. J. McSHANE Professor of Mathematics in the University of Virginia VOLUME I fEGQNTf EDITION • . 1 \ j - - > BLACKIE & SON LIMITED LONDON AND GLASGOW First published 1934 Second edition 1937 Reprinted 1940, 194', '942 {twice) 1943, J944, 1945, X946, 1947, K)4S '949, '950, I9S 1 , 195*, "953, 1954 I 9SS, I 9S ( > (twice), 1957 (twice) Z938, 1959, i960, 1961 Printed in Great Britain by Blackie & Son, Ltd., Glasgoa PREFACE TO THE FIRST GERMAN EDITION Although there is no lack of textbooks on the differential and integral calculus, the beginner will have difficulty in finding a book that leads him straight to the heart of the subject and gives him the power to apply it intelligently. He refuses to be bored by diffuseness and general statements which convey nothing to him, and will not tolerate a pedantry which makes no distinction between the essential and the non-essential, and which, for the sake of a systematic set of axioms, deliberately conceals the facts to which the growth of the subject is due. True, it is easier to perceive defects than to remedy them. I make no claim to have presented the beginner with the ideal textbook. Yet I do not consider the publication of my lectures superfluous. In order and choice of material, in fundamental aim, and perhaps also in mode of presentation, they differ con- siderably from the current literature. The reader will notice especially the complete break away from the out-of-date tradition of treating the differential calculus and the integral calculus separately. This separation, a mere result of historical accident, with no good foundation either in theory or in practical convenience in teaching, hinders the student from grasping the central point of the calculus, namely, the connexion between definite integral, indefinite integral, and derivative. With the backing of Felix Klein and others, the simultaneous treatment of differential calculus and integral calculus has steadily gained ground in lecture courses. I here attempt to give it a place in the literature. This first volume deals mainly with the integral and differential calculus for func- tions of one variable; a second volume will be devoted to functions of several variables and some other extensions of the calculus. vi PREFACE TO THE FIRST GERMAN EDITION My aim is to exhibit the close connexion between analysis and its applications and, without loss of rigour and precision, to give due credit to intuition as the source of mathematical truth. The presentation of analysis as a closed system of truths without reference to their origin and purpose has, it is true, an aesthetic charm and satisfies a deep philosophical need. But the attitude of those who consider analysis solely as an abstractly logical, introverted science is not only highly unsuitable for beginners but endangers the future of the subject; for to pursue mathe- matical analysis while at the same time turning one's back on its applications and on intuition is to condemn it to hopeless atrophy. To me it seems extremely important that the student should be warned from the very beginning against a smug and presumptu- ous purism; this is not the least of my purposes in writing this book. The book is intended for anyone who, having passed through an ordinary course of school mathematics, wishes to apply him- self to the study of mathematics or its applications to science and engineering, no matter whether he is a student of a univer- sity or technical college, a teacher, or an engineer. I do not promise to save the reader the trouble of thinking, but I do seek to lead the way straight to useful knowledge, and aim at making the subject easier to grasp, not only by giving proofs step by step, but also by throwing light on the interconnexions and purposes of the whole. The beginner should note that I have avoided blocking the entrance to the concrete facts of the differential and integral calculus by discussions of fundamental matters, for which he is not yet ready. Instead, these are collected in appendices to the chapters, and the student whose main purpose is to acquire the facts rapidly or to proceed to practical applications may post- pone reading these until he feels the need for them. The appen- dices also contain some additions to the subject-matter; they have been made relatively concise. The reader will notice, too, that the general style of presentation, at first detailed, is more condensed towards the end of the book. He should not, however let himself be disheartened by isolated difficulties which he may find in the concluding chapters. Such gaps in understand- ing, if not too frequent, usually fill up of their own accord. PREFACE TO THE ENGLISH EDITION When American colleagues urged me to publish an English edition of my lectures on the differential and integral calculus, I at first hesitated. I felt that owing to the difference between the methods of teaching the calculus in Germany and in Britain and America a simple translation was out of the question, and that fundamental changes would be required in order to meet the needs of English-speaking students. My doubts were not laid to rest until I found a competent colleague in Professor E. J. McShane, of the University of Virginia, who was prepared not only to act as translator but also — after personal consultation with me — to make the improve- ments and alterations necessary for the English edition. Apart from many matters of detail the principal changes are these: (1) the English edition contains a large number of classified examples; (2) the division of material between the two volumes differs somewhat from that in the German text. In addition to a detailed account of the theory of functions of one variable, the present volume contains (in Chapter X) a sketch of the differentiation and integration of functions of several variables. The second volume deals in full with functions of several inde- pendent variables, and includes the elements of vector analysis. There is also a more systematic discussion of differential equa- tions, and an appendix on the foundations of the theory of real numbers. Thus the first volume contains the material for a course in elementary calculus, while the subject-matter of the second volume is more advanced. In the first volume, however, there is much which should be omitted from a first course. These sec- tions, intended for students wishing to penetrate more deeply into the theory, are collected in the appendices to the chapters, '• vii (a 788) viii PREFACE TO THE ENGLISH EDITION so that beginners can study the book without inconvenience, omitting or postponing the reading of these appendices. The publication of this book in English has only been made possible by the generosity of my German publisher, Julius Springer, Berlin, to whom I wish to express my most cordial thanks. I have likewise to thank Blackie and Son, Ltd., who in spite of these difficult times have undertaken to publish this edition. My special thanks are due to the members of their technical staff for the excellent quality of their work, and to their mathematical editors, especially Miss W. M. Deans, who have relieved Prof. McShane and myself of much of the respon- sibility of preparing the manuscript for the press and reading the proofs. I am also indebted to many friends and colleagues, notably to Professor McClenon of Grinnell College, Iowa, to whose encouragement the English edition is due; and to Miss Margaret Kennedy, Newnham College, Cambridge, and Dr. Fritz John, who co-operated with the publisher's staff in the proof-reading. R. COURANT. Cambridge, England, June, 1934. PREFACE TO THE SECOND ENGLISH EDITION This second edition differs from the first chiefly in the improvement and rearrangement of the examples, the addi- tion of many new examples at the end of the book, and the inclusion of some additional material on differential equations. R. COURANT. New Rochellb, N.V.. June, 1937. CONTENTS Iktboduotory Remarks ..... 1 Chapter I INTRODUCTION 1. The Continuum of Numbers ...... 6 2. The Concept of Function 14 3. More Detailed Study of the Elementary Functions - • 22 4. Functions of an Integral Variable. Sequences of Numbers • 27 5. The Concept of the Limit of a Sequence ----- 29 6. Further Discussion of the Concept of Limit .... 38 7. The Concept of limit where the Variable is Continuous - • 46 8. The Concept of Continuity 49 APPENDIX I Preliminary Remarks 56 1. The Principle of the Point of Accumulation and its Applications - 58 2. Theorems on Continuous Functions ----- • 63 3. Some Remarks on the Elementary Functions .... 68 APPENDIX II 1. Polar Co-ordinates 71 2. Remarks on Complex Numbers 73 Chaptbb n THE FUNDAMENTAL DDEAS OF THE INTEGRAL AND DDJFERENTIAL CALCULUS 1. The Definite Integral 76 2. Examples 82 x CONTENTS Page 3. The Derivative 88 4. The Indefinite Integral, the Primitive Function, and the Funda- mental Theorems of the Differential and Integral Calculus - 109 5. Simple Methods of Graphical Integration 119 6. Further Remarks on the Connexion between the Integral and the Derivative - ...... 121 7. The Estimation of Integrals and the Mean Value Theorem of the Integral Calculus 126 APPENDIX 1. The Existence of the Definite Integral of a Continuous Function 131 2. The Relation between the Mean Value Theorem of the Differential Calculus and the Mean Value Theorem of the Integral Cal- culus ....-••••• 134 Chapter HI DIFFERENTIATION AND INTEGRATION OF THE ELEMENTARY FUNCTIONS 1. The Simplest Rules for Differentiation and their Applications - 136 2. The Corresponding Integral Formulae 141 3. The Inverse Function and its Derivative 144 4. Differentiation of a Function of a Function .... 153 5. Maxima and Minima - 158 6. The Logarithm and the Exponential Function .... 167 7. Some Applications of the Exponential Function • • - 178 8. The Hyperbolic Functions 183 9. The Order of Magnitude of Functions 189 APPENDIX 1. Some Special Functions 196 2. Remarks on the Differentiability of Functions • • • -199 3. Some Special Formulae 201 Chapter IV FURTHER DEVELOPMENT OF THE INTEGRAL CALCULUS 1. Elementary Integrals 205 2. The Method of Substitution 207 CONTENTS xi Page 3. Further Examples of the Substitution Method .... 214 4. Integration by Parts 218 6. Integration of Rational Functions ...... 226 6. Integration of Some Other Classes of Functions .... 234 7. Remarks on Functions which are not Integrable in Terms of Elementary Functions .... - 242 8. Extension of the Concept of Integral. Improper Integrals • • 245 APPENDIX The Second Mean Value Theorem of the Integral Calculus • • 258 Chapter V APPLICATIONS 1. Representation of Curves - - - • . 2. Applications to the Theory of Plane Curves 3. Examples 4. Some very Simple Problems in the Mechanics of a Particle 5. Further Applications: Particle sliding down a Curve - 6. Work ......... 258 267 287 292 299 304 APPENDIX 1. Properties of the Evolute 307 2. Areas bounded by Closed Curves 311 Chapter VI TAYLOR'S THEOREM AND THE APPROXIMATE EXPRESSION OF FUNCTIONS BY POLYNOMIALS 1. The Logarithm and the Inverse Tangent .... 315 2. Taylor's Theorem 320 3. Applications. Expansions of the Elementary Functions - - 326 4. Geometrical Applications • • - • ■ - - 331 xii CONTENTS APPENDIX Page 1. Example of a Function which cannot be expanded in a Taylor Series .......... 336 2. Proof that e is Irrational ....... 336 3. Proof that the Binomial Series Converges 337 4. Zeros and Infinities of Functions, and So-called Indeterminate Expressions ......... 338 Chapter VII NUMERICAL METHODS Preliminary Remarks ........ 342 1. Numerical Integration ........ 342 2. Applications of the Mean Value Theorem and of Taylor's Theorem. The Calculus of Errors 349 3. Numerical Solution of Equations 355 APPENDIX Stirling's Formula 361 Chapter Vlll INFINITE SERIES AND OTHER LIMITING PROCESSES Preliminary Remarks ........ 365 1. The Concepts of Convergence and Divergence .... 366 2. Tests for Convergence and Divergence 377 3. Sequences and Series of Functions ...... 383 4. Uniform and Non-uniform Convergence ..... 386 5. Power Series 398 6. Expansion of Given Functions in Power Series. Method of Undetermined Coefficients. Examples .... 404 7. Power Series with Complex Terms 410 APPENDIX 1. Multiplication and Division of Series ..... 415 2. Infinite Series and Improper Integrals ..... 417 3. Infinite Products 419 4. Series involving Bernoulli's Numbers 422 CONTENTS siii Chapter IX FOURIER SERIES Page 1. Periodic Functions ......... 425 2. Use of Complex Notation 433 3. Fourier Series 437 4. Examples of Fourier Series 440 5. The Convergence of Fourier Series ...... 447 APPENDIX Integration of Fourier Series ---.... 455 Chapter X A SKETCH OF THE THEORY OF FUNCTIONS OF SEVERAL VARIABLES 1. The Concept of Function in the Case of Several Variables - • 458 2. Continuity .......... 453 3. The Derivatives of a Function of Several Variables - . . 466 4. The Chain Rule and the Differentiation of Inverse Functions - 472 5. Implicit Functions ......... 4go 6. Multiple and Repeated Integrals 486 Chapter XI THE DIFFERENTIAL EQUATIONS FOR THE SIMPLEST TYPES OF VIBRATION 1. Vibration Problems of Mechanics and Physics .... 502 2. Solution of the Homogeneous Equation. Free Oscillations - 504 3. The Non-homogeneous Equation. Forced Oscillations - - 509 4. Additional Remarks on Differential Equations - . . 519 Summary car Important Theorems and Formulae - . 529 Miscellaneous Examples 549 Answers and Hints ........ 571 Index 611 DIFFERENTIAL AND INTEGRAL CALCULUS Introductory Remarks When the beginner comes in contact with the so-called higher mathematics for the first time, he is apt to be obsessed by the feeling that there is a certain discontinuity between school mathematics and university mathematics. This feeling ulti- mately rests on more than the historical circumstances which have caused university teaching to take a form differing so widely from that of the school. For the very nature of the higher mathematics, or rather, of the modern mathematics, developed during the last three centuries, distinguishes it from the elemen- tary mathematics which wholly dominated the school curri- culum until recently and whose subject-matter was often taken over almost directly from the mathematics of the ancient Greeks. A leading characteristic of elementary mathematics is its intimate association with geometry. Even where the subject passes beyond the realm of geometry into that of arithmetic, the fundamental ideas still remain geometrical. Another feature of ancient mathematics is perhaps its tendency to concentrate on particular cases. Things which to-day we should regard as special cases of a general phenomenon are set down higgledy- piggledy without any visible relationship between them. Its intimate association with geometrical ideas and its stress on individual niceties give the older mathematics a charm of its own. Yet it was a definite advance when at the beginning of the modern age in mathematics quite different tendencies de- a INTRODUCTORY REMARKS veloped, acting as the stimulus for a great expansion of tbe subject, which in spite of many improvements in detail had in a sense stood still for centuries. The fundamental tendency of all modem mathematics is towards the replacement of separate discussions of individual cases by more and more general systematic methods, which perhaps do not always do full justice to the individual features of a particular case, but which, owing to their generality and power, give promise of a wealth of new results. Again, the con- cept of number and the methods of analysis have come to occupy more and more independent positions and now dominate geometry entirely. These new tendencies towards the development of mathematics along a variety of lines are most clearly exhibited in the rise of analytical geometry, whose development is chiefly due to Fermat and Descartes, and of the differential and integral calculus, which is generally regarded as having originated with Newton and Leibnitz. The three hundred years during which modern mathematics has existed have seen such important advances not only in pure mathematics, but in an immense variety of applications to science and engineering, that its fundamental ideas and above all the concept of a function have by degrees become very widely known and have eventually penetrated even into the school curriculum. In this book my aim has been to develop the most important facts in the differential and integral calculus so far that at the close the reader, although he may have had no previous knowledge of higher mathematics, may be well equipped on the one hand for the study of the more advanced branches and of the foundations of the subject, or on the other hand, for the manipulation of the calculus in the varied realms in which it is applied. I should like to warn the reader specially against a danger which arises from the discontinuity mentioned in the opening paragraph. The point of view of school mathematics tempts one to linger over details and to lose one's grasp of general relation- ships and systematic methods. On the other hand, in the " higher " point of view there lurks the opposite danger of getting out of touch with concrete details, so that one is left helpless when faced with the simplest cases of individual difficulty, because in the world of general ideas one has forgotten how to INTRODUCTORY REMARKS 3 come to grips with the concrete. The reader must find his own way of meeting this dilemma. In this he can only succeed by repeatedly thinking out particular cases for himself and acquiring a firm grasp of the application of general principles in particular cases; herein lies the chief task of anyone who wishes to pursue the study of Science. CHAPTER I Introduction The differential and integral calculus ia based upon two concepts of outstanding importance, apart from the concept of number, namely, the concept of function and the concept of limit. These concepts can, it is true, be recognized here and there even in the mathematics of the ancients, but it is only in modern mathematics that their essential character and signi- ficance are fully brought out. In this introductory chapter we shall attempt to explain these concepts as simply and clearly as possible. 1. The Continuum of Numbers The question as to the real nature of numbers is one which concerns philosophers more than mathematicians, and philo- sophers have been much occupied with it. But mathematics must be carefully kept free from conflicting philosophical opinions; preliminary study of the essential nature of the con- cept of number from the point of view of the theory of know- ledge is fortunately not required by the student of mathematics. We shall therefore take the numbers, and in the first place the natural numbers 1, 2, 3, . . ., as given, and we shall likewise take as given the rules* by which we calculate with these numbers; and we shall only briefly recall the way in which the concept of the positive integers (the natural numbers) has had to be extended. * These rules are as follows: (o + 6) + c - o + (6 + 6). That is, if to the sum of two numbers a and b we add a third number c, we obtain the same result as when we add to a the sum of 6 and c. (This is called the associative law of addition.) Secondly, a + b = b + a (the commutative law of addition). Thirdly, (ab)c = a(bc) (the associative law of multiplication). Fourthly, ab = ba (the commutative law of multiplication). Fifthly, a(b + c) - ab + ac (the distributive law of multiplication). 6 6 INTRODUCTION [Chap. 1. The System of Rational Numbers and the Need for its Extension. In the domain of the natural numbers the fundamental operations of addition and multiplication can always be per- formed without restriction; that is, the sum and the product of two natural numbers are themselves always natural numbers. But the inverses of these operations, subtraction and division, cannot invariably be performed within the domain of natural numbers; and because of this mathematicians were long ago obliged to invent the number 0, the negative integers, and positive and negative fractions. The totality of all these numbers is usually called the class of rational numbers, since they are all obtained from unity by using the " rational operations of calcu- lation ", addition, multiplication, jj — J z _j J — j — + — +~- subtraction and division. Fig. ..-The number axU Numbers are usually represented graphically by means of the points of a straight line, the " number axis ", by taking an arbitrary point of the line as the origin or zero point and another arbitrary point as the point 1; the distance between these two points (the length of the unit interval) then serves as a scale by which we can assign a point on the line to every rational number, positive or negative. It is customary to mark off the positive numbers to the right and the negative numbers to the left of the origin (cf. fig. 1). If, as usual, we define the absolute value (also called the numerical value or modulus) | a | of a number a to be a itself when * a Si 0, and to be — a when a < 0, then | a | simply denotes the distance of the corresponding point on the number axis from the origin. The geometrical representation of the rational numbers by points on the number axis suggests an important property which is usually stated as follows: the set of rational numbers is every- where dense. This means that in every interval of the number axis, no matter how small, there are always rational numbers; geometrically, in the segment of the number axis between any two rational points, no matter how close together, there are points corresponding to rational numbers. This density of the rational * By the sign ^ we mean that either the sign > or the sign — shall hold. A corresponding statement holds for the signs ± and T which will be used later. I] THE CONTINUUM OF NUMBERS 7 numbers at once becomes clear if we start from the fact that the numbers -, — , — _, . . . become steadily smaller and approach nearer and nearer to zero as » increases. If we now divide the number axis into equal parts of length 1/2", beginning 12 3 at the origin, the end-points — , _,_,... of these intervals 2" 2" 2" represent rational numbers of the form m/2"; here we still have the number n at our disposal. If now we are given a fixed interval of the number axis, no matter how small, we need only choose n so large that 1/2" is less than the length of the interval; the intervals of the above subdivision are then small enough for us to be sure that at least one of the points of subdivision m/2 n lies in the interval. Yet in spite of this property of density the rational numbers are not sufficient to represent every point on the number axis. Even the Greek mathematicians recognized that when a given line segment of unit length is chosen there are intervals whose lengths cannot be represented by rational numbers; these are the so-called segments incommensurable with the unit. Thus, for example, the hypotenuse of a right-angled isosceles triangle with sides of unit length is not commensurable with the unit of length. For, by the theorem of Pythagoras, the square of this length I must be equal to 2. Therefore, if I were a rational number and consequently equal to p/q, where p and q are integers different from 0, we should have p* = 2q\ We can assume that p and q have no common factors, for such common factors could be cancelled out to begin with. Since, according to the above equation, p 2 is an even number, p itself must be even, say p = 2p'. Substituting this expression for p gives us 4p' 2 = 2q\ or q* = 2p' 2 ; consequently q* is even, and so q is also even. Hence p and q both have the factor 2. But this contradicts our hypothesis that p and q have no common factor. Thus the assumption that the hypotenuse can be represented by a fraction p/q leads to contradiction and is therefore false. The above reasoning, which is a characteristic example of an " indirect proof ", shows that the symbol ^2 cannot corre- spond to any rational number. Thus we see that if we insist that after choice of a unit interval every point of the number axis shall have a number corresponding to it, we are forced to extend 8 INTRODUCTION [Chap the domain of rational numbers by the introduction of new " irrational " numbers. This system of rational and irrational numbers, such that each point on the axis corresponds to just one number and each number corresponds to just one point on the axis, is called the system of real numbers.* 2. Real Numbers and Infinite Decimals. Our requirement that to each point of the axis there shall correspond one real number states nothing a priori about the possibility of calculating with these real numbers in the same way as with rational numbers. We establish our right to do this by showing that our requirement is equivalent to the following fact: the totality of all real numbers is represented by the totality of all finite and infinite decimals. We first recall the fact, familiar from elementary mathe- matics, that every rational number can be represented by a terminating or by a recurring decimal; and conversely, that every such decimal represents a rational number. We shall now show that to every point of the number axis we can assign a uniquely determined decimal (usually infinite), so that we can represent the irrational points or irrational numbers by infinite decimals. (In accordance with the above remark the irrational numbers must be represented by infinite non-recurring decimals, for ex- ample, 0-101101110 . . .). Suppose that the points which correspond to the integers are marked on the number axis. By means of these points the axis is subdivided into intervals or segments of length 1. In what follows, we shall say that a point of the line belongs to an interval if it is an interior point or an end-point of the interval. Now let P be an arbitrary point of the number axis. Then the point belongs to one, or if it is a point of division to two, of the above intervals. If we agree that in the second case the right-hand one of the two intervals meeting at P is to be chosen, we have in all cases an interval with end-points g and g -f- 1 to which P belongs, where g is an integer. This interval we subdivide into ten equal sub-intervals by means of the points 12 9 corresponding to the numbers g + — , g -\- — , . . . , g -\- — , and * Thus named to distinguish it from the system of complex numbers, obtained by yet another extension. I] THE CONTINUUM OF NUMBERS 9 we number these sub-intervals 0, 1, . . . , 9 in the natural order from left to right. The sub-interval with the number a then has the end-points g 4- — and g 4- — 4- — . The point P must be contained in one of these sub-intervals. (If P is one of the new points of division it belongs to two consecutive intervals; as before, we choose the one on the right.) Suppose that the interval thus determined is associated with the number Oj. The end- points of this interval then correspond to the numbers g 4- — and g 4- ^ 4- — . This sub-interval we again divide into ten equal parts and determine that one to which P belongs; as be- fore, if P belongs to two sub-intervals we choose the one on the right. We thus obtain an interval with the end-points 9 + w + w» and 9 + w + W + W where " 2 i8 one of the digits 0, 1, . . . , 9. This sub-interval we again subdivide, and continue to repeat the process. After n steps we arrive at a sub- interval containing P, having length — and with end-points corresponding to the numbers Here each a is one of the numbers 0, 1, . . . , 9. But 10 10 2 " 10" is simply the decimal fraction O-o^ ■ . . a„. The end-points of the interval, therefore, may also be written in the form g 4- 0-0^2 . . . o„ and 4- 0-a x a 2 . . . a„ 4- -—. 10" If we consider the above process repeated indefinitely, we obtain an infinite decimal O-Oja^ . . . , which has the following meaning. If we break off this decimal at any place, say the n-th, the point P will lie in the interval of length — whose end-points (approxi- mating points) are g + O-o^ . . . a„ and g + O-o^ . . . a„ + __. 10 n io INTRODUCTION [Chap. In particular, the point corresponding to the rational number g + O-o^a . . . a n will lie arbitrarily near to the point P if only n is large enough; for this reason the points g + 0-a x a 2 . . . a„ are called approximating points. We say that the infinite decimal g + O-OjOg • • • is the real number corresponding to the point P. Here we would emphasize the fundamental assumption that we can calculate in the usual way with the real numbers, and hence with the decimals. It is possible to prove this using only the properties of the integers as a starting-point. But this is no light task; and rather than allow it to bar our pro- gress at this early stage, we regard the fact that the ordinary rules of calculation apply to the real numbers as an axiom, on which we shall base the whole differential and integral calculus. We here insert a remark concerning the possibility, in certain cases, of choosing the interval in two ways in the above scheme of expansion. From our construction it follows that the points of division arising in our repeated process of subdivision, and such points only, can be represented by finite decimals g + Q-a x a % . . . a n . Let us suppose that such a point P first appears as a point of division at the ra-th stage of the subdivision. Then according to the above process we have chosen at the tt-th stage the interval to the right of P. In the following stages we must choose a sub- interval of this interval. But such an interval must have P as its left end- point. Therefore in all further stages of the subdivision we must choose the first sub-interval, which has the number 0. Thus the infinite decimal corresponding to P is g + O-ajOj . . . o n 000 .... If, on the other hand, we had at the ra-th stage chosen the left-hand interval containing P, then in all later stages of subdivision we should have had to choose the sub- interval farthest to the right, which has P as its right end-point. Such a sub-interval has the number 9. Thus for P we should have obtained a decimal expansion in which all the digits from the (n + l)-th onward are nines. The double possibility of choice in our construction therefore corre- sponds to the fact that for example the number J has the two decimal expansions 0-25000 . . . and 0-24999 .... 3. Expression of Numbers in Scales other than that of 10. In our representation of the real numbers we made the number 10 play a special part, for each interval was subdivided into ten equal parts. The only reason for this is the widespread use of the decimal system. We could just as well have taken p equal sub-intervals, where p is an arbitrary integer greater than 1. We should then have obtained an expression of the form g -f- — -| — | + . . . , where each 6 is one of the numbers P P z I] THE CONTINUUM OF NUMBERS u 0, 1, . . . , p — 1. Here again we find that the rational numbers, and only the rational numbers, have recurring or terminating expansions of this kind. For theoretical purposes it is often convenient to choose p = 2. We then obtain the binary expan- sion of the real numbers, *+!+£+..-. where each b is either * or 1. For numerical calculations it is customary to express the whole number g, which for simplicity we here take to be posi- tive, in the decimal system, that is, in the form a m l(T + a^KT- 1 + ... + a l 10+a v where each a„ is one of the digits 0, 1, . . . , 9. Then for g + O-a^a ... we write simply 0-m a m-l ■ • ■ <h a o ' <h a 2 • • • Similarly, the positive whole number g can be written in one and only one way in the form P*P k + /W* _1 + • . . + AjH- &, where each of the numbers /?„ is one of the numbers 0, 1 p — 1. This, with our previous expression, gives the following result: every positive real number can be represented in the form fi*P k +P*-iP k - 1 + •■. + PiP+Po+ b - i +k+-.., V P 2 where j8„ and o„ are whole numbers between and p — 1. Thus, for example, the binary expansion of the fraction ^ is »-lX*+OX2+l + ?+^. * Even for numerical calculations the decimal system is not the best. The sexagesimal system (p — 60), with which the Babylonians calculated, has the advantage that a comparatively large proportion of the rational numbers whose decimal expansions do not terminate pobsess terminating sexagesimal ex- pansions. 12 INTRODUCTION [Chap. 4. Inequalities. Calculation with inequalities plays a far larger part in higher mathematics than in elementary mathematics. We shall there- fore briefly recall some of the simplest rules concerning them. If a > 6 and c > d it follows that a + c> b + d, but not that a — c> b — d. Moreover, if a > b it follows that ac > be, provided c is positive. On multiplication by a negative number the sense of the inequality is reversed. If o > b > and c > d > 0, it follows that ac > 6d. For the absolute values of numbers the following inequalities aold: \a±b\^\a\ + \b\, |o±6|S=|a|-|*l- The square of any real number is greater than or equal to zero. Therefore, if x and y are arbitrary real numbers (x- yf = x* + y*-2xy^0, or 2xy <^ x 2 + y 2 . 5. Schwarz's Inequality. Let a 1} a 2 , . . . , a n and b v b 2 , . . . , b„ be any real numbers. In the preceding inequality we make the substitutions * Kl y- IM for * = 1, *= 2, ...,»= n successively and add the resulting inequalities. On the right we obtain the sum 2, for / Kl V. i / Kl V- ) W + ... + ".')/ + '"^\V(«i 2 +--- + «»V ' / im y + i ( im Y=i. If we divide both sides of the inequality by 2 we obtain KM + KM + • • • + KM ^ x V(V + • • • + «„ 2 ) V(V + • • • + V) - ' * Here and hereafter the symbol Vx, where x > 0, denotes that positive Dumber whose square is x. I] THE CONTINUUM OF NUMBERS 13 or finally I «A I + I «A I + • • • + 1 a A I ^ V(«i 2 +- • .+a n *W{b* +. . .+6„ 2 ). Since the expressions on both sides of this inequality axe positive, we may square and then omit the modulus signs: (a 1 6 1 + a 2 6 2 + . . . + a n b n f ^ («,»+ . . . +«„*) (V+ ... + 6„ 2 ). This is the Cauchy-Schwarz inequah'ty. Examples * 1. Prove that the following numbers are irrational: (a) VS. (b) Vn, where n is not a perfect square. (c) -y^3. (d)* x = V2 + aV2 (e)* *= V3+^2. 2.* In an ordinary system of rectangular co-ordinates, the points for which both co-ordinates are integers are called lattice points. Prove that a triangle whose vertices are lattice points cannot be equilateral, 3. Prove the inequalities: (a) * -f i ^ 2, x > 0. (6) x + - g -2, x < 0. («) x ^2, a; =)= 0. 4. Show that if a > 0, ax 3 + 2bx + ^ f or all values of x if, and only ff, J'-acgO. 5. Prove the following inequalities: (a) x 1 + xy + y* ^ 0. (6)* a?" + x in -h) -f a^-'V + • . . + y*" ^ 0. (c)* x« — 3s» + 4a? — 3a; + 1 ^ 0. 6. Prove Sohwarz's inequality by considering the expression (a lX + bj' + (o 2 x + h)* + . . . + (a n x + &„)«, collecting terms and applying Ex. 4. 7. Show that the equality sign in Sohwarz's inequality holds if, and only if, the a's and &'s are proportional; that is, ca v + db v = for all v's, where c, d are independent of v and not both zero. 8. For n = 2, 3, state the geometrical interpretation of Sohwarz's inequality. 9. The numbers Yi> Ys are direction cosines of a line; that is, Yi* + Y2 2 = 1. Similarly, tj, 2 + 7] 2 2 = 1. Prove that the equation Yj 7 )] + fsVz — 1 implies the equations y, = r) u y t = 7) 2 . 10.* Prove the inequality V / "(»i-&i) i! +-.. +(«„-&„)* =£ VW+--- + V) + aA 6 i* + - •• + &„"> and state its geometrical interpretation. * The more difficult examples are indicated by an asterisk. 14 INTRODUCTION [Chap. 2. The Concept op Function 1. Examples. (a) If an ideal gas is compressed in a vessel by means of a piston, the temperature being kept constant, the pressure p and the volume v are connected by the relation pv = G, where C is a constant. This formula, called Boyle's Law, states nothing about the quantities v and p themselves, but has the following meaning: if p has a definite value, arbitrarily chosen in a certain range (the range being determined physically and not mathematically), then v can be determined, and conversely: C G «=-, p=-. p v We then say that v is a function of p, or in the converse case that p is a function of v. (b) If we heat a metal rod, which at temperature 0° has length l , to the temperature 8°, then its length I will be given, on the simplest physical assumptions, by the law i=i a + p9), where jS, the " coefficient of expansion ", is a constant. Again we say that Hb a function of 6. (c) In a triangle let the lengths of two sides, say a and b, be given. If for the angle y between these two sides we choose any arbitrary value less than 180° the triangle is completely determined; in particular, the third side c is determined. In this case we say that if a and b are given c is a function of the angle y. As we know from trigonometry, this function is repre- sented by the formula c = \/(a* -\-b 2 — 2ab cos y). 2. Formulation of the Concept of Function. In order to give a general definition of the mathematical concept of function, we fix upon a definite interval of our number scale, say the interval between the numbers a and b, and con- I] THE CONCEPT OF FUNCTION" 15 aider the totality of numbers x which belong to this interval, that is, which satisfy the relation If we consider the symbol x as denoting at will any of the numbers in this interval, we call it a (continuous) variable in the interval. If now to each value of x in this interval there corresponds a single definite value y, where x and y are connected by any iaw whatsoever, we say that y is a function of x, and write sym- bolically y = /(»), V=F{x), y = g(x), or some similar expression. We then call x the independent variable and y the dependent variable, or we call x the argument of the function y. It should be remarked that for certain purposes it makes a difference whether in the interval from a to 6 we include the end-points, as we have done above, or exclude them; in the latter case, the variable x is restricted by the inequalities « < x < b. To avoid misunderstanding we may call the first kind of interval (including end-points) a closed interval, the second kind an open interval. If only one end-point and not the other is included (as for example a<x^b) we speak of an interval open at one end (in this case the end a). Finally, we may also consider open intervals which extend without bound in one direction or both. We then say that the variable x ranges over an infinite (open) interval, and write symbolically a < a; < 00 or — 00 < x < 6 or — 00 < » < oo . In the general definition of a function which is denned in an interval nothing is said about the nature of the relation by which the depen- dent variable is determined when the independent variable is given. This relation may be as complicated as we please, and in theoretical investi- gations this wide generality is an advantage. But in applications, and in particular in the differential and integral calculus, the functions with which we have to deal are not of the widest generality; on the contrary, the laws of correspondence by which a value of y is assigned to each x are subject to certain simplifying restrictions. 16 INTRODUCTION [Chap. 3. Graphical Representation. Continuity. Monotonic Functions. Natural restrictions of the general function-concept are suggested if we consider the connexion with geometry. The fundamental idea of analytical geometry is in fact that of giving a curve denned by some geometrical property a characteristic analytical representation by regarding one of the rectangular co-ordinates, say y, as a function y=f{x) of the other co- ordinate x; for example, a parabola is represented by the func- tion y = a; 2 , the circle with radius 1 about the origin by the two functions y = \/(l — x 2 ) and y = —\/(l — a; 2 ). In the first example we may think of the function as defined in the infinite interval — oo < x < oo ; in the second we must restrict ourselves to the interval — 1 ^ x rgi 1, since outside this interval the function has no meaning (when x and y are real). Conversely, if instead of starting with a curve determined geometrically , we consider a given function y=f(x), we can represent the functional de- Fig. a.— Rectangular axn p e ndence of y on a; graphically by making use of a rectangular co-ordinate system in the usual way (cf. fig. 2). If for each abscissa x we lay off the correspond- ing ordinate y=f(x), we obtain the geometrical representation of the function. The restriction which we now wish to impose on the function-concept is this: the geometrical representation of the function shall take the form of a "reasonable" geo- metrical curve. This, it is true, implies a vague general idea rather than a strict mathematical condition. But we shall soon formulate conditions, such as continuity, differentia- bility, &c, which will ensure that the graph of a function has the character of a curve capable of being visualized geometri- cally. At any rate, we shall exclude a function such as the following: for every rational value of x, the function y has the value 1; for every irrational value of x, the value of y is 0. This assigns a definite value of y to each x; but in every interval of x, no matter how small, the value of y jumps from to 1 and back an infinite number of times. Unless the contrary is expressly stated, it will always be assumed that the law which assigns a value of the function to THE CONCEPT OF FUNCTION 17 each value of x assigns just one value of y to each value of x, as for example y=x 2 or y = sin a;. If we begin with a curve given geometrically it may happen, as in the case of the circle x 2 -f y 2 = 1, that the whole course of the curve is not given by one single (one-valued) function, but requires several functions — in the case of the circle, the two functions y = -\/(l — x 2 ) and y = — V(l — x *)- The same is true for the hyperbola y2 — x 2 = 1, which is represented by the two functions y = <y/(l -\- x 2 ) and y = —- \/{\ + a; 2 ). Such curves therefore do not determine the corresponding functions uniquely. Conse- quently it is sometimes said that the function corresponding to Fig. 3 Fw. 4 Multiple-valued functions the curve is multiple-valued. The separate functions representing the curve are then called the single-valued branches belonging to the curve. For the sake of clearness we shall henceforth use the word " function " to mean a single-valued function. In conformity with this, the symbol \/x (for x S; 0) will always denote the non-negative number whose square is x. If a curve is the geometrical representation of one function it will be cut by any parallel to the y-axis in at most one point, since to each point x in the interval of definition there corre- sponds just one value of y. Otherwise, as for example in the case of the circle which is represented by the two functions y=^(l-x 2 ) and y=-V(l-* 2 ), such parallels to the y-axis may intersect the curve in more than one point. The portions of a curve corresponding to different single-valued branches are sometimes so connected with each 2 (1798) i8 INTRODUCTION [Chap. other that the complete curve is a single figure which can be drawn with one stroke of the pen, e.g. the circle (cf. fig. 3), or, on the other hand, the branches may be completely separated, e.g. the hyperbola (cf. fig. 4). Here follow some further examples of the graphical representation of functions. (a) y = ax. y is proportional to x. The graph (cf. fig. 5) is a straight line through the origin of the co-ordinate system. (6) y = ax+ b. y is a " linear function " of x. The graph is a straight line through the point x = 0, y = b, which, if a #= 0, also passes through the point x = — b/a, y = 0, and if a = runs horizontally. Fig. 5. — Linear functions (C) v = - i ' Fig. 6. — Infinite discontinuities y is inversely proportional to a:. If in particular a = 1, so that 1 y = -, X we find, for example, that y = 1 for * = 1, y = 2 for x = J, y — J for x = 2. I] THE CONCEPT OF FUNCTION 19 The graph (cf. fig. 6) is a curve, a rectangular hyperbola, symmetrical with respect to the bisectors of the angles between the co-ordinate axes. This last function is obviously not defined for the value x — 0, since division by zero has no meaning. The exceptional point x = 0, in whose neighbourhood there occur arbitrarily large values of the function, both positive and negative, is the simplest example of an infinite discontinuity, a subject to which we shall return later (cf. p. 51). (d) 1= 3?. As is well known, this function is represented by a parabola (cf. fig. 7). Similarly, the function y = x* is represented by the so-called cubical parabola (cf. fig. S). V\ Fig. 7. — Parabola Fig. 8. — Cubical parabola The curves just considered and their graphs exhibit a property which is of the greatest importance in the discussion of functions, namely, the property of continuity. We shall later (§ 8, p. 49) analyse this concept in more detail; intuitively it comes to this, that a small change in x causes only a small change in y and not a sudden jump in its value; that is, the graph is not broken off. More exactly, the change in y remains less than any arbitrarily chosen positive bound, provided that the change in x is correspondingly small. A function which for all values of a; in an interval has the same value y = a is called a constant; it is graphically repre- sented by a horizontal straight line. A function y = f(x) such that throughout the interval in which it is defined an increase in the value of x always causes an increase in the value of y is called a monotonic increasing function; if, on the other hand, 20 INTRODUCTION [Chap. an increase in the value of x always causes a decrease in the value of y, the function is called a monotonic decreasing function. Such functions are represented graphically by curves which in the corresponding interval always rise (from left to right) or always fall (cf. fig. 9). If the curve represented by y=f(x) is symmetrical with respect to the y-axis, that is, if x = —a and x = a give the same value for the function, or /(-s) =/(*), we say that the function is an even function. For example, the function y = x 2 is even (cf. fig. 7). If, on the other hand, the Fig. 9. — Monotonic functions curve is symmetrical with respect to the origin, that is, if /(-*) = -f(x), we call the function an odd function; for example, the functions y = x and y = x 3 (cf. fig. 8) and y=l/x are odd. 4. Inverse Functions. Even in our first example on p. 14 it was made evident that a formal relationship between two quantities may be regarded in two different ways, since it is possible either to consider the first variable as a function of the second or to consider the second as a function of the first. If, for example, y = ax -\- b, where we assume that a =)= 0, x is represented as a function of y by the equation x = (y — b) fa. Again, the functional relationship represented by the equation y=x 2 can also be represented by the equation x— + \/y, so that the function y = a; 2 amounts to the same thing as the two functions x = \/y and * = — y/y. THE CONCEPT OF FUNCTION 21 Thus, when an arbitrary function y = f(x) is given we can attempt to determine x as a function of y, or, as we shall say, to replace the function y = f(x) by the inverse function x = <f>(y). Geometrically this has the following meaning: we consider the curve obtained by reflecting the graph of y=f(x) in the line bisecting the angle between the positive ataxia and the positive y-axis * (cf. fig. 10). This at once gives us a graphical representation of a; as a function of y and thus represents the inverse function x = <f>(y). These geometrical ideas, however, show us at once that a /x-(j>(y) y y 3 j& ? Z ~ 7 ! / / 1 y t / >- - _ /l 1 /' 3Cf %2 %3 •% / Fig. io. — Inversion of a function function y = f(x) defined in an interval has not a single- valued inverse function unless certain conditions hold. If the graph of the function is cut by a line y = c parallel to the x-axis in more than one point, the value y=c will correspond to more than one value of x, so that the function cannot have a single-valued inverse function. This case cannot occur if y=f(x) is con- tinuous and monotonic. For then fig. 10 shows us that to each value of y in the interval y^yy z there corresponds just one value of x in the interval x&x%, and from the figure we infer that a func- tion which is continuous and monotonic in an interval always has a single-valued inverse function, and this inverse function is also continuous and monotonic. (For a rigorous proof, see p. 67.) * Instead of reflecting the graph in this way, we oonld first rotate the co- ordinate axes and the curve y — f(x) through a right angle and then reflect the graph in the x-axi*. 22 INTRODUCTION [Chap. 3. More Detailed Study of the Elementary Functions 1. The Rational Functions. We now pass on to a brief review of the elementary functions which the reader has already met with in his previous studies. The simplest types of function are obtained by repeated appli- cation of the elementary operations: addition, multiplication, subtraction. If we apply these operations to an independent Fig. ii. — Powere of * variable x and any real numbers, we obtain the rational integral functions or polynomials: y=a -\-a 1 x + . . . + a n x\ The polynomials are the simplest and, in a sense, the basic functions of analysis. If we now form the quotients of such functions, that is, expressions of the form y _ a o + <h x + • • • + a n x n b + \x + . . . + b m x"' we obtain the general or fractional rational functions, which are defined at all points where the denominator differs from zero. The simplest rational integral function is the linear function y=ax+b. I] THE ELEMENTARY FUNCTIONS 23 It is represented graphically by a straight line. Every quadratic function of the form y = ax 2 + bx ■+- C is represented by a parabola. The ourves which represent rational integral functions of the third degree, y = ax 3 + bx 1 + ex + d, are occasionally called parabolas of the third order, and so on. As examples, we give the graphs of the function y = a;" for the indices n = 1, 2, 3, 4 in fig. 11. We see that for even values of n the func- tion y=x n satisfies the equation /(— £) = j(x), and is therefore an even function, while for odd values of n the function satisfies the condition /(—»)= —f(x), and is therefore an odd function. The simplest example of a rational function which is not a polynomial is the function y = 1/x mentioned on p. 18; its graph is a rectangulaj hyperbola. Another is the function y = \jx* (cf. fig. 12). 2. Algebraic Functions. We are at once led away from the domain of rational func- tions by the problem of forming their inverses. The most impor- tant example of this is the introduction of the function y/x. We start with the function y = x", which for x ^ is mono- tonic. It therefore has a single-valued inverse, which we denote by the symbol *= tyy, or, interchanging the letters used for the dependent and independent variables, y = y/x = x 1 '". In accordance with the definition this root is always non-negative. In the case of odd values of n the function x n is monotonic for all values of x, including negative values. Consequently for odd values of n we can also define y/x uniquely for all values of x; in this case \/x is negative for negative values of z. More generally we may consider y = VR(x), where R(x) is a rational function. We arrive at further functions of similar type by applying rational operations to one or more of these special functions. Thus for example we may form the functions y=V*+V(* 2 +i), y=a + vV+i). These functions are special cases of algebraic functions. (The general concept of an algebraic function cannot be denned here; see Chapter X.) 24 INTRODUCTION [Chap. 3. The Trigonometric Functions. While the rational functions and the algebraic functions just considered are denned directly in terms of the elementary opera tions of calculation, geometry is the source from which we first draw our knowledge of the other functions, the so-called transcendental functions.* We shall here consider the elementary transcendental functions, namely, the trigonometric functions, the exponential function, and the logarithm. In all higher analytical investigations where angles occur it is customary to measure these angles not in degrees, minutes, d and seconds, but in radians. We place the angle to be measured with its vertex at the centre of a circle of radius 1, and measure the size of the angle by the length of the arc of the circumference which the angle cuts out. Thus an angle of 180° is the same as an angle of -n radians (has radian measure n), an angle of 90° has radian measure n/2, an angle of 45° has radian measure ir/1, an angle of 360° has radian measure 2tt. Conversely, an angle of 1 radian expressed in degrees is 180° , or approximately 57° 17' 45". ■n Henceforward, whenever we speak of an angle x, we shall mean an angle whose radian measure is x. After these preliminary remarks we may briefly remind the reader of the meanings of the trigonometric functions sin a;, cosa;, tana;, cota;.f These are shown in fig. 13, in which the angle x is measured from the arm 00 (of length 1), angles being reckoned positive in the counter-clockwise direction. The rectangular * The word " transcendental " does not mean anything particularly deep or mysterious; it merely suggests the fact that definition of these functions by means of the elementary operations of calculations is not possible, " quod algebrae vires transcendit ". fit is sometimes convenient to introduce the functions secx = 1/cosz, cosecj; = 1/sinx. Fig. 13. — The trigonometric functions I] THE ELEMENTARY FUNCTIONS 25 co-ordinates of the point A at once give us the functions cos a; and sin a;. The graphs of the functions sin as, cos a;, tan a;, cot as are given in figs. 14 and 15. -ji Fig. 14 Fig. is 4. The Exponential Function and the Logarithm. In addition to the trigonometric functions, the exponential function with the positive base a, and its inverse, the logarithm to the base a, x = lo&y. are also regarded as elementary transcendental functions. In elementary mathematics it is customary to pass over certain inherent difficulties in the definition of these functions, and we too shall postpone the exact discussion of the functions until we have better methods at our disposal (cf. Chapter III, § 6, pp. 167-177, and also p. 191). We can, however, at least state the basis of the definitions here. If x = p/q is a 2* (1798) 26 INTRODUCTION [Chap. rational number (where p and q are positive integers), then — the number a being assumed positive — we define a x as {/a* = a plq , where the root, according to convention, is to be taken as positive. Since the rational values of x are everywhere dense, it is natural to extend this function a" so as to make it a continuous function denned for irrational values of x also, giving values to a x when x is irrational which are continuous with the values already defined when x is rational. This gives us a con- tinuous function y = a", the " exponential function ", which for all rational values of x gives the value of a x found above. That this extension is actually possible and can be carried out in only one way we meanwhile take for granted; but it must be borne in mind that we still have to prove that this is so.* The function x = log a y can then be defined for y > as the inverse of the exponential function. Examples 1. Plot the graph of y = x*. Prom this, without further calculation, find the graph of y = -^/x. 2. Sketch the following graphs, and state whether the functions are even or odd: (o) y = sin 2*. (6) y = 5 cos a:. (c) y = sina; + cosar. (d) y = 2 sina; + sin2x. (c) y = sin(a; + tz). (/) y=2coa(x+?). (d) V = tan* — x. 3. Sketch the graphs of the following functions, and state whether the functions are (1) monotonic or not, (2) even or odd: (a) y = a? (— oo < x < oo ). (6) y = a* (0 g a: g 1). (c) y=i(-lgigl). (d) y=\x\(-l£x£l). (e) y= Va?(—1 Sigl). \S) y = | * — 1 | (— oo < a; < oo ). (?) y = | *» + 4a; + 2 | (-4 S x S 3). (A) y = [x] (— oo < x < oo), where [a;] means the greatest integer which does not exceed x; that is, [i]^i< [a;] + 1. * Cf. pp. 70 and 173. I] THE ELEMENTARY FUNCTIONS 27 (»') y = x — [x] ( — go < x < 00 ). 0") «/ = Vz— [x] (— 00 < x < 00 ). (&) y = x + Vx — [x] ( — 00 < x < 00 ). (2) y = I x - 1 1 + I x + 1 1 - 2 (-5 g * g 5). (»») y = | a; — 1| — 2 | a; | + | a; + 1 | (— °o < a: < 00). Which two of these functions are identical? 4. A body dropped from rest falls approximately 162* ft. in t see. If a ball falls from a window 25 ft. above ground, plot its height above ground as a function of t for the first 4 sec. after it starts to fall. 4. Functions of an Integral Variable. Sequences of Numbers Hitherto we have considered the independent variable as a continuous variable, that is, as varying over a complete interval. However, numerous cases occur in mathematics in which a quan- tity depends only on an integer, a number n which can take the values 1, 2, 3, . . . . Such a function we call a function of an integral variable. This idea will most easily be grasped by means of examples. 1. The sum of the first n integers, S^n) = 1 + 2+3 + 4+. .. + » = in(n + 1), is a function of ». Similarly, the sum of the first n squares, S,(n) = 1» + 2 a + 3 2 + . . . + n\ is a function * of the integer «. * This last sum may easily be represented as a simple rational expression in n in the following way. We begin with the formula (» + l) 3 - ■/» - 3k 2 + Zv + 1, write down this equation for the values » - 0, 1, 2 n, and add. We thus obtain (n + 1)» - 38, + 3/8, + n + 1; on substituting the formula just given for 8 ly this becomes 38, - (» + l)j(n + l)«-l-|n|-(n+ l){n» + ?n), so that S, - J»(» + 1) (2» + 1). By a similar process the functions 8,(n) - 1* + 2' + . . . + »», S t {n) - 1* + 2* + . . . + B \ can be represented as rational functions of n. z8 INTRODUCTION [Chap. 2. Other simple functions of integers are the expression n! = 1 . 2 . 3 . . . ra and the binomial coefficients l)...(n— fc+1) n! (n\ n(n • k) &1 kl(n — k)\ for a fixed value of k. 3. Every whole number n > 1 which is not a prime number is divisible by more than two positive integers, while the prime numbers are divisible only by themselves and by 1. We can obviously consider the number T(n) of divisors of n as a function of the number n itself. For the first few numbers this function is given by the following table: »=123466789 10 1112 T(n) =122324243 4 2 6 4. A function of this type which is of great importance in the theory of numbers is ir(n), the number of primes which are less than the number n. Its detailed investigation is one of the most interesting and attractive problems in the theory of numbers. Here we merely mention the principal result of these investigations: the number ■n-(ra) is given approximately, for large values of n, by the function * re/log n, where by log 7i we mean the logarithm to the " natural base " e, to be denned later (pp. 168, 174). Functions of an integral variable usually occur in the form of so-called sequences of numbers. By a sequence of numbers we understand an ordered array of infinitely many numbers a^, a 2 , a 3 , . . . , a„, . . . , (not necessarily all different), determined by any law whatever. In other words, we are dealing simply with a function a of the integral variable n; the only difference is that we are using the index notation o„ instead of the symbol a(n). Examples 1. Prove that 1» + 2 s + . . . + r? = (1 + 2 + . . . + n)K 2. From the formula for 1* + 2 2 + . . . + » s , find a formula for l* + 3" + 5* + • • • + (2» + 1) J . 3. Prove the following properties of the binomial coefficients: « G)- C.->** (b) G-i) + 0- (":>">«>• < c)1+ (i) + © + - + (»-i) + C)= 2B - * That is, the quotient of the number n(n) by the number n/logn differs arbitrarily little from 1, provided only that n is large enough. I] FUNCTIONS OF AN INTEGRAL VARIABLE 29 4. Evaluate the following sums: (a) 1.2 -f 2.3 + . . . -f- »(» + 1). 1.2 2.3 m(ra + 1) 1*.2* 2 2 .3 2 n 2 (» + 1)» 5. A sequence is called an arithmetic progression of the first order if the differences of successive terms are constant. It is called an arith- metic progression of the second order if the differences of successive terms form an arithmetic progression of the first order; and in general, it is called an arithmetic progression of order k if the differences of successive terms form an arithmetic progression of order (k — 1). The numbers 4, 6, 13, 27, 50, 84 are the first six terms of an arithmetic progression. What is its order? What is the eighth term? 6. Prove that the n-th term of an arithmetic progression of the second order can be written in the form an* -f bn -f- c, where a, b, e are inde- pendent of n. 7.* Prove that the »-th term of an arithmetic progression of order k can be written in the form an h + ftn* -1 + . . . + pn + q, where a,b,...,p,q are independent of n. Knd the «-th term of the progression in Ex. 6. 5. The Concept op the Limit of a Sequence The fundamental concept on which the whole of analysis ultimately rests is that of the limit, of a sequence. We shall first make the position clear by considering some examples. n We consider the sequence 1 1 1 Oj = 1, o, = -, o, = -,... , o„ = 2 n No number of this sequence is zero; but we see that the larger the number n is, the closer to zero is the number a n . If, therefore, we mark off around the point an interval as small as we please, then from a definite index onward all the numbers a„ will fall in this interval. This state of affairs we express by saying that as n increases the numbers a„ tend to 0, or that they possess the limit 0, or that the sequence a^, a 2 , a 3 . . . converges to 0. If the numbers are represented as points on a line this means that the points l/» crowd closer and closer to the point as » increases. 30 INTRODUCTION [Chap. The situation is similar in the case of the sequence 1 1 1 (-l)"- 1 o, = 1, a 2 =— -, a s=g. o 4 =—-,..., a„= Here, too, the numbers a n tend to zero as n increases; the only difference is that the numbers a n are sometimes greater and sometimes less than the limit 0; as we say, they oscillate about the limit. The convergence of the sequence to is usually expressed symbolically by the equation lim a„ = 0, n — >oo or occasionally by the abbreviation a„-»-0. o n _ 1 . „ _ 1 *• a 2m — — > a 2m-l — X~, • In the preceding examples, the absolute value of the difference be- tween a„ and the limit steadily becomes smaller as n increases. This is not necessarily the case, as is shown by the sequence 1,1111 °i = s» a 2 = 1 ' ° s = 4' a * ~ 2' ° 5 = 6' a " ~ 3' ' " " ' that is, in general, for even values n= 2m, a„= a 2m = 1/m, for odd values n = 2m — 1, a n = a^ n _ 1 = l/2m. This sequence also has a limit, namely, zero; for every interval about the origin, no matter how small, will contain all the numbers o B from a certain value of n onward; but it is not true that every number lies nearer to the limit zero than the preceding one. 3. a„ n + 1 We consider the sequence 1 2 °1 ~ a' a 2 1_ 2' 2_ 3 "~n+l where the integral index n takes all the values 1, 2, 3, . . . . If we write a = 1 we see at once that as n increases the number o„ will " n+1 approach closer and closer to the number 1, in the sense that if we mark off any interval about the point 1 all the numbers a a following a certain a s must fall in that interval. We write lim a_ = 1. I] LIMIT OF A SEQUENCE 31 »*- 1 The sequence n % + n + I behaves in a similar way. This sequence also tends to a limit as n in- creases, to the limit 1, in fact; in symbols, lim a n = 1. We see this most readily if we write *->■» „ 1 «+2 n m + n -+• 1 here we need only show that the numbers r B tend to as n increases. Now for all values of n greater than 2 we have n + 2 < 2» and n 2 + n + 1 > m 2 . Hence for the remainder r n , we have 2» 2 < r « < -2 - - (» > 2 )« from which we see at once that r n tends to as n increases. Our discussion at the same time gives an estimate of the amount by which the number a n (for » > 2) can at most differ from the limit 1; this difference cer- tainly cannot exceed 2/n. The example just considered illustrates the fact, which we should naturally expect, that for large values of n the terms with the highest indices in the numerator and denominator of the fraction for a n pre- dominate and that they determine the limit. 4. a n = y£. Let p be any fixed positive number. We consider the sequence Oj, o^ a s> • • •» o„, . . . , where °n = Vp- We assert that lim a n = lim -\/p = I. n— > 00 n— > «> We can prove this very easily by using a lemma which we shall find useful for other purposes also. If 1 + h is a positive number (that is, if h > —1), and n is an integer greater than 1, then (1 + h) n > 1 + nh (1) Let us suppose that the inequality (1) is already proved for a certain value m > 1; we multiply both sides by (1 + h) and obtain (1 + h)"* 1 > (1 + mh) (1 + h) = I + (m + l)h + mh». If, on the right, we omit the positive term mA 2 the inequality remains valid. We thus obtain (1 + h)™+ x > 1 + (m + l)h. This, however, is our inequality for the index m + 1. It follows therefore that if the inequality holds for the index m it holds for the index m + 1 32 INTRODUCTION [Chap. also. Since it holds for m = 2, it holds also for m = 3, hence for m = 4, and so on; therefore it holds for every index. This is a simple example of a proof by mathematical induction, a type of proof which is often useful. Returning to our sequence, we distinguish between the case p > 1 and the case p < 1 (if p = 1, then \/p is also equal to 1 for every n, and our statement becomes trivial). If p > 1, then ■yjp will also be greater than 1; we put ^/p = 1 + h n , where h n is a positive quantity depending on to, and by the inequality (1) we have p = (1 + h n )» > 1 + nh n , from which it at once follows that p— 1 We therefore see that as n increases the number h n must tend to 0, which proves that the numbers a„ converge to the limit 1, as stated. At the same time we have a means for estimating how close any o„ is to the limit 1; the difference between a„ and 1 is certainly not greater than (p — l)/w. If p < 1, then \/p will likewise be less than 1 and therefore may be taken equal to 1/(1 + h n ), where h n is a positive number. From this it follows, using the inequality (1), that _ 1 1 P ~ (l + h„)" < l + nh n - (By making the denominator smaller we increase the fraction.) It follows that 1 + nh n < ~, P 1 / i and therefore h. < . n From this we see that h n tends to as to increases. As the reciprocal of a quantity tending to 1, yp itself tends to 1. 5. a a = a". We consider the sequence a n = a", where a is fixed and n runs through the sequence of positive integers. First, let a be a positive number less than 1. We may then put a = 1/(1 + h), where h is positive, and inequality (1) gives 1 11 < — -. " (1 + h) n 1 + nh nh Since the number h, and consequently 1/h, depends only on a, and does not change as ra increases, we see that as to increases a™ tends to 0: lim a" = (0 < a < 1). I] LIMIT OF A SEQUENCE 33 The same relationship holds when a is zero, or negative but greater than — 1. This is immediately obvious, since in any case lim | a | n = 0. If a = 1, then a" will obviously be always equal to 1, and we shall have to regard the number 1 as the limit of a™. If a > 1, we put a = 1 + h, where h is positive, and at once see from our inequality that as n increases a" does not tend to any definite limit, but increases beyond all bounds. We express this state of affairs by saying that a" tends to infinity as n increases, or that a n becomes infinite; in symbols, lim a" = oo (a > 1). Nevertheless, as wo must explicitly emphasize, the symbol oo does not denote a number with which we can calculate as with any other number; equations or statements which express that a quantity is or becomes infinite never have the same sense as an equation between definite quanti- ties. In spite of this, such modes of expression and the use of the symbol oo are extremely convenient, as we shall often see in the following pages. If a = — 1, the values of a™ will not tend to any limit, but as n runs through the sequence of positive integers it will take the values -f 1 and — 1 alternately. Similarly, if a < — 1 the value of a" will increase numerically beyond all bounds, but its sign will be alternately positive and negative. 6. Geometrical Illustration of the Limits of a" and v"A If we consider the ourves y = x" and y = x 1 '" = -y/x and restrict ourselves for the sake of con- venience to non-negative values of x, the preceding limits are illus- trated by figs. 16 and 17 respec- tively. In the case of the curves y = x" we see that in the interval from to 1 they approach closer and oloser to the x-axis as n in- creases, while outside that interval they climb more and more steeply and draw in closer and closer to a line parallel to the y-axis. All the curves pass through the point with co-ordinates x = 1, y = 1 and through the origin. y i r* r* . \ti- A \\ i /*» / o " " I 'x Fig. 16.- as n increases In the case of the functions y = x 1 '" = + a/x, the curves approach closer and closer to the line parallel to the x-axis and at a distance 1 above it. On the other hand, all the curves must pass through the origin. Hence in the limit the curves approach the broken line consisting of the part of 34 INTRODUCTION [Chap. the y-axis between the points y = and y = 1 and of the parallel to the a;-axis y = 1. Moreover, it is clear that the two figures are closely related, as one would expect from the fact that the functions y = y /x are actually the inverse functions of the ra-th powers, from which we infer that each figure is transformed into the other on reflection in the line y = «• V •y «-l2-~ / x- Fig. 17. — jr 17 " as n increases 7. The Geometric Series. An example of a limit which is more or less familiar from elementary mathematics is the geometric series 1 + g + ?* + . . . + g"- 1 = S n ; the number q is called the common ratio of the series. The value of this sum may, as is well known, be expressed in the form <S« 1-g" l-<7 provided that q =4= h ^e can derive this expression by multiplying the sum 8„ by q and subtracting the equation thus obtained from the original equation, or we may verify the formula by division. The question now arises, what happens to the sum 8 n when n increases indefinitely? The answer is this: the sum iS„ has a definite limit 8 if q lies between — 1 and +1, these end values being excluded, and it is then true that 1 8 lim S- 1 In order to verify this statement we write the numbers 8 n in the form 1 — q" 1 q" — = z . We have already shown that provided q 1-g 1-g J r S« = I] LIMIT OF A SEQUENCE 35 I g| < 1 the quantity q", and with it — - — , tends to as n increases: q hence with the above assumption the number 8 n tends, as was stated, to the limit as n increases. 1-? _ 1 The passage to the limit lim(l + g'-(-g»+...-f j"-i) = is n—ya 1 — 9 usually expressed by saying that when | q | < 1 the geometric series can be extended to infinity and that the sum of the infinite geometric series is the expression . 1-g The sums S n of the finite geometric series are also called the partial sums of the infinite geometric series 1 + q + g 2 + . . . . (We must draw a sharp distinction between the sequence of numbers S v S s S„, . . . and the geometric series.) The fact that the partial sums S n of the geometric series tend to the limit S = as n increases may also be expressed by saying that the infini te geometric series 1 -f q + q* + .. . converges to the sum S = when I q | < 1. ! — 9 8. a a =tyn. We shall show that the sequence of numbers «i = 1, Oj = V2, a s = {/3 o n = -\/n, . . . tends to 1 as n increases, i.e. that lim y/n = 1. n— *-oo Here we make use of a slight artifice. Instead of the sequence a n = \/n we first consider the sequence b n = Va n = V-y/i = VVn. When n > 1 the term b n is also greater than 1. We can therefore put b„ = 1 + h n , where h„ is positive and depends on n. By inequality (1), p. 31, we therefore have Vn = (6„)» = (I + AJ» Sl + ri, *w 1 ^Vn-1 Vn 1 so that h n 5j <J — = -;-. n n Vn We now have 1 SS a n = 6„» = 1 + 2A„ + V =£ 1 + J- + -. V n n The right-hand side of this inequality obviously tends to 1, and therefore so does a„. n» 36 INTRODUCTION [Chap. 9. a„ = Vn + 1 — \/n. We assert that lim (Vn + 1 — Vn) == 0. To prove this we need only write the expression under discussion in the form .- , (Vn+l—Vn) (Vn+l+Vn) Vn+1— v»= — Vn+l + Vft Vn+l + Vn we see at once that this expression tends to as n increases. 10. a„ = — . a" Let a be a number greater than 1. We assert that as n increases the ft sequence of numbers a„ = — tends to the limit 0. As in the case of \/n above we consider the sequence . Vn We put V a = 1 + h. Here h > 0, since a and hence V a is greater than 1. By inequality (1), p. 31, we have vV* = (1 + h) n > 1 + nh, Vn _ Vn , Vn 1 so that Vo_ = (1 + h) n — 1 + nh ~~ nh hVn 1 Hence a. == --z. " nh 1 Since a n is positive and the right-hand side of this equation tends to 0. we see that a_ must also tend to 0. Examples n s 4- n 1 1 1. Prove that lim — — = -. Find an N such that for n > N „_>.„ 3ft 2 + 1 3 w 2 I n I J the difference between ; and - is (a) less than yo, (6) less than l o o o> ( c ) I® 88 than ioooooo- 2. Find the limits of the following expressions as n -*■ oo : I] LIMIT OF A SEQUENCE 37 (a) W ' +3n+1 . (M n'+Sn+l n 8 + 7n» + 2 n« + 7ra» + 2 ( C) 6n« + 2n + 1 g n* + a ^-^ +... + g k » 3 + »* 6 »* + M*"" 1 + • • . + &** n Si" 3. Prove that lim ^/n* = 1. n— >-oo 4. Prove that lim £- = 0. Find an N such that — < 1 . whenever n > N. .->-«2 n 2" 1000 6. Find numbers N u N& N s such that: (a) h < 10 for every " > Nii n 1 (6) 2 s < 100 f ° r 6Very " > N * (e) & < Md f0IeYeiyn>N *- 6. Do the same thing for the sequence a n = Vn +1 — Vn. 7. Prove that lim ( Vm + 1 — Vn) ( Vn + J) = J. 8. Prove that lim (v^+1 — v^w) = 0. n— >« 10" 9. Let a„ = — _. (o) To what limit does o„ converge? (6) Is the sequence monotonic? (c) Is it monotonic from a certain n onwards? (d) Give an estimate of the difference between a„ and the limit, (e) From what value of n onwards is this difference less than jJxf? 10. Prove that lim — = 0. 11. Prove that lim ( I + - + . . . + ±\ = I. 12. Prove that lim (— + + . . . + _L_ ^ = 0. .->- W T (n + I)" T ^ (2«)V 13. Prove that lim (-L- + * + . . . + ^L=\ = 00 . »_».„W»i V»+l V2n/ H.* Prove that «->.i Wn»+ 1 + Vn»+2 ' * ' + Vn 2 + n) = L 15. Prove that if a and 6^0 are positive, the sequence \/ , a n + 6" converges to o. Similarly, for any k fixed positive numbers a : , a 2 , . . . , a 4 prove that -y^ai* + o s B + . . . + a k n converges, and find its limit. 38 INTRODUCTION [Chap. 16. Prove that the Bequenee V2, V2V2, a/2V2\/2 converges. Find its limit. 17.* If v(ra) is the number of distinct prime factors of n, prove that lim *0_0l 6. Fubther Discussion of the Concept oe Limit 1. First Definition of Convergence. From the cases discussed in the last section we are led to form the following general concept of limit: If an infinite sequence of numbers a^ a 2 , a 3 , . . . , a n , . . . is given and if there is a number 1 such that every interval, no matter how small, marked off about the point I contains all the points a n except for a finite number at most, we say that the number I is the limit of the sequence %, a^, . . . , or that the sequence %, &%, . . . converges to I; in symbols, lim a„= J. Here we expressly remark »—>■«> that this includes the trivial case in which all the numbers a n are equal to one another and hence also coincide with the limit. Instead of the above we may use the following equivalent statement: If any positive number e be assigned — no matter how small — a whole number N = N(e) can be found such that from the index N onward (i.e. for n > N(e)) it is always true that | a„ — 1 1 < e. Of course it is as a rule true that the bound N(e) will have to be chosen larger and larger as smaller and smaller values of e are chosen; in other words, N(e) will increase beyond all bounds as e tends to 0. It is important to remember that every convergent sequence is bounded; that is, to every sequence a x , a z , a 3 , . . . for which a limit I exists there corresponds a positive number M, inde- pendent of n, such that for all the terms o„ of the sequence the inequality | a„ | < M is valid. This theorem readily follows from our definition. We choose e equal to 1; then there is an index N such that for n > N it is true that \a n — l\ < 1. Amongst the numbers I oj — I \, I a 2 - 1 1, . . . , \a s —l\ let A be the largest. We can then put M = \ 1 1 -+• A -\- 1 For I] FURTHER DISCUSSION OF LIMITS 39 by the definition of A the inequality | a„ — I | < A -f 1 cer- tainly holds for n = 1, 2, . . . , N, while for n > N \a n -l\<l^A+l. A sequence which does not converge is said to be divergent. If as n increases the numbers a„ increase beyond all bounds we say that the sequence diverges to + a>, and, as we have already done occasionally, we write lim a n = 00. Similarly, we write lim a n = — 00 if as n increases the numbers — a n increase beyond all bounds in the positive direction. But divergence may manifest itself in other ways, as, for example, in the case of the sequence ^ = — 1, a 2 = + 1, a 3 = -1, a 4 = +1, .... whose terms swing to and fro between two different values* In all the examples given above it has happened that the limit of the sequence considered is a known number. If the concept of limit yielded nothing more than the recognition that certain known numbers can be approximated to as closely as we like by certain sequences of other known numbers, we should have gained very little from it. The fruitfulness of the concept of limit in analysis rests essentially on the fact that limits of sequences of known numbers provide a means of dealing with other numbers which are not directly known or expressible. The whole of higher analysis consists of a succession of ex- amples of this fact, which will become steadily clearer to us in the following chapters. The representation of the irrational numbers as limits of rational numbers may be regarded as a first example. In this section we shall become acquainted with further ex- amples. Before we take up this subject, however, we shaD make a few preliminary general remarks. 2. Second (Intrinsic) Definition of Convergence. How can we tell that a given sequence of numbers %, a 2 , °s. • • • j «n» • • • converges to a limit, even when we do not know beforehand what that limit is? This important question is an- swered for us by Cauchy's convergence test.\ * Another useful remark: the behaviour of a sequence as regards conver- gence is unaltered if we omit a finite number of the terms o». In what follows we shall frequently make use of this, speaking of the convergence or divergence of series in which the term a, is undefined for a finite number of values of n. t Sometimes referred to as the general principle of convergence. 40 INTRODUCTION [Chap. We say that a sequence of numbers a ly a 2 , . . . , a n , . . . is intrinsically convergent if to every arbitrarily small positive number e there corresponds a number N = N(e), usually de- pending on e, such that \ a„ — a m \ < e, provided that n and m are both at least equal to N(e). Cauchy's convergence test can then be expressed as follows: Every intrinsically convergent sequence of numbers possesses a limit. The importance of Cauchy's test lies in the fact that it allows us to speak of the limit of a sequence after considering the se- quence itself, without any further information about the limit. The converse of Cauchy's test is very easy to prove. For if the sequence a v a 2 , ... tends to the limit I, then by the definition of convergence we have \l — a n | < ^ and | I — a m | < |, where e is a positive quantity as small as we please, provided only that m and n are both large enough; therefore K — a m | = | {I — a m ) — (I — a n ) | ^ 1 1 — a m \ + 1 1 — a„ | <e. Since e can be chosen as small as we please, this inequality ex- presses our statement. Cauchy's test itself becomes intuitively obvious if we think of the numbers as represented on the number axis. It then states that a sequence certainly has a limit if after a certain point N all the terms of the sequence are restricted to an interval which can be made arbitrarily small by choosing N large enough. In the appendix we shall show how Cauchy's test can be proved by purely analytical methods. For the time being we accept it as a postulate. 3. Monotonia Sequences. The question whether a given sequence converges to a limit is particularly easy to answer when the sequence is a so-called monotonic sequence; that is, if either every number of the sequence is larger than the preceding number (monotonic increasing se- quence) or else every number is smaller than the preceding number (monotonic decreasing sequence). We have the following theorem: I] FURTHER DISCUSSION OF LIMITS 41 Every monotonia increasing sequence whose terms are bounded above {that is, lie below a fixed number) possesses a limit; simi- larly, every monotonia decreasing sequence whose terms never fall below a certain fixed bound possesses a limit. For the present we shall regard these results as obvious, merely referring the student to the rigorous proof in the appendix (p. 61). A con- vergent monotonic increasing sequence must, of course, tend to a limit which is greater than any term of the sequence, while in the case of a convergent monotonic decreasing sequence the numbers tend to a limit which is smaller than any number of the sequence. Thus, for example, the numbers 1/w form a mono- tonic decreasing sequence with the limit 0, while the numbers 1 — 1/w form a monotonic increasing sequence with the limit 1. In many cases it is convenient to replace the condition that a sequence shall increase monotonically by the weaker condition that the terms of the sequence shall never decrease; in other words, to allow successive terms of the sequence to be equal to one another. We then speak of a monotonic non- decreasing sequence, or of a monotonic increasing sequence in the wider sense. Our theorem on limits remains true for such sequences, and also for sequences which are monotonic non- increasing or monotonic decreasing in the wider sense. 4. Operations with Limits. We conclude with a remark concerning calculations with limits. From the definition of limit it follows almost at once that we can perform the elementary operations of addition, multiplication, subtraction, and division according to the follow- ing rules: If a^, Og, ... is a sequence with the limit a and 6 X , b 2 , . . . is a sequence with the limit 6, then the sequence of numbers c n = a n + b„ also has a limit, and lim c„= a-\- b. n— >-«o The sequence of numbers c n = a n b„ likewise converges, and lim c n = ab. Similarly, the sequence c n = a„ — b n converges, and lim c_ = a — 6. 4a INTRODUCTION [Chap. Provided the limit b differs from 0, the numbers c„ = — also converge, and have the limit " lim c„ = * R-»-X In words: we can interchange the rational operations of calcu- lation with the process of forming the limit; that is, we obtain the same result whether we first perform a passage to the limit and then a rational operation or vice versa. For the proof of these simple rules it is sufficient to give one example; using this for a model, the reader can establish the other statements for himself. We consider e.g. the multipli- cation of limits. The relations a n ->a and b„ -> b amount to the following: if we choose any positive number e, we need only take n greater than N, where N = N(e) is a sufficiently large number depending on c, in order to have both | a — a n | < e and | b — b„ | < e. If we write ah — a n b„ = b(a — a„) + a„(b — b„) and recall that there is a positive bound M, independent of n, such that | a„ | <M, we obtain |«6-o.6.|^|6||a-o.| + |o.||6-6 1 ,|<(|6| + Jf)«. Since the quantity ( | b | + M) e can be made arbitrarily small by choosing e small enough, we see that the difference between ah and a n b n actually becomes as small as we please for all suffi- ciently large values of n, which is precisely the statement made in the equation ah= lim a n b n . n—> oo By means of these rules many limits oan be evaluated very easily; for example, we have km — = lun = 1, n ra a since in the second expression the passages to the limit in numerator and denominator can be made directly. Another simple and obvious rule is worth stating. IJ lim a n = a and lim b n = b, and if in addition a n > b n for every n, then a ^ b. However, we are by no means entitled to expect I] FURTHER DISCUSSION OF LIMITS 43 that in general a will be greater than b, as is shown by the case of the sequences o n = 1/n, b n = l/2w, for which a = = b 5. The Number e. As a first example of the generation of a number, which can- not be stated in advance, as the limit of a sequence of known numbers, we consider the sums fl - = - 1 + ri + a + - + a- We assert that as n increases these numbers S n tend to a definite limit ., In order to prove the existence of the limit we observe that as n increases the numbers S„ increase monotonically. For all values of n we also have 1 1 — - 2 The numbers S n therefore have the upper bound 3 and, being a monotonic increasing sequence, they possess a limit, which we denote by e: e = lim S„. n- Further, we assert that the number e defined as the above limit is also the limit of the sequence ■=K)" The proof is simple and at the same time an instructive example of operations with limits. According to the binomial theorem, which we shall here assume, -i+w 1 + n (n-l) 1 j n(n-l)(n-2)..A 1_ n 2! n 2 wl n n n!\ n) \ n) ' "\ ~ n / 44 INTRODUCTION [Chap. From this we see at once (1) that T n 5S S n , and (2) that the T„'s also form a monotonic increasing sequence,* whence the existence of the limit lira T n = T follows. In order to prove that n — > oo T = e, we observe that r . >1+1+ i(,_I) + ...+A(,_iY..(i-!=J\ 2! \ mj n\\ m/ \ m / provided that m > n. If we now keep n fixed and let m increase beyond all bounds, we obtain on the left the number T and on the right the expression S n , so that T 2; S n . We have thus estab- lished the relationship T ^ S n 5g T n , for every value of n. We can now let n increase, so that T n tends to T; from the double inequality it follows that T = lim S n = e. This was the state- ment to be proved. «->» We shall later (Chapter III, § 6, p. 172) reach this number e again from still another point of view. 6. The Number n as a Limit. A limiting process which in essence goes back to classical antiquity (Archimedes) is that by which the number it is defined. Geometrically tt means the area of the circle of radius 1. We therefore accept the existence of this number -n as intuitive, regarding it as obvious that this area can be expressed by a (rational or irrational) number, which we then simply denote by 7T. However, this definition is not of much help to us if we wish to calculate the number with any accuracy. We have then no choice but to represent the number by means of a limiting process, namely, as the limit of a sequence of known and easily calculated numbers. Archimedes himself used this process in his method of exhaustions, where he steadily approximated to the circle by means of regular polygons with an increasing number of sides fitting it more and more closely. If we denote the area of the regular m-gon (polygon of m sides) inscribed in * We obtain T n+ i from T H by replacing the factors 1 — 1/n, 1 — 2/», . . . 1 2 by the larger factors 1 — , 1 — , . . . and finally adding a positive term. " + x n + 1 I] FURTHER DISCUSSION OF LIMITS 45 the circle by/„„ the area of the inscribed 2»i-gon is given by the formula (proved by elementary geometry) ^-sV»-»V'-(£)'- We now let m run, not through the sequence of all positive integers, but through the sequence of powers of 2, that is, m = 2"; in other words, we form those regular polygons whose vertices are obtained by repeated bisection of the circumference. The area of the circle is then given by the limit *r= lim/ 2 .. »— >■« This representation of tt as a limit actually series as a basis for numeri- cal computations; for, starting with the value /„ = 2, we can calculate in order the terms of our sequence tending to 77. An estimate of the ac- curacy with which any term / a « represents 77 can be obtained by construct- ing the lines touching the circle and parallel to the sides of the inscribed 2"-gon. These lines form a circumscribed polygon similar to the inscribed 2"-gon, and having dimensions greater in the ratio I : cos — -. Hence the area F t * of the circumscribed polygon is given by ^" _1 Since the area of the circumscribed polygon is evidently greater than that of the circle, we have 1 < n < F.« = /. 2" I cos . J These are matters with which the reader will be more or less familiar. What we wish to point out here is that the calculation of areas by means of exhaustion by rectilinear figures whose areas can be calculated easily forms the basis for the concept of integral, which will be introduced in the next chapter (p. 76). Examples 1.* (a) Replace the statement "the sequence a„ is not absolutely bounded" by an equivalent statement not involving any form of the words "bounded" or "unbounded". (6) Replace the statement "the sequence a n is divergent" by an equivalent statement not involving any form of the words " convergent " or " divergent ". 46 INTRODUCTION [Chap. 2.* Let o t and 6 t be any two positive numbers, and let a t < 6 V Let a, and b t be denned by the equations aj=Va 1 6„ 6 2 ^ Similarly, let a 3 = Va 2 6 2 , 6 3 = 2 a 2 + 6 2 t + K-i and in general o„ = Va n _ 1 b n _ 1 , b„ = Prove (a) that the sequence a 1? a 2 , . . . , converges, (&) that the sequence &j, & 2 converges, (c) that the two sequences have the same limit. (This limit is called the arithmetic-geometric mean of Oj and b v ) 3.* Prove that if lim a n = %, then lim <j„ = £, where a n is the arith- metic mean (Oj + o 2 -j- . . . + ««)/»• 4. If lim a„ = 5, show that the arithmetic means of the arithmetic n — >-oo means <j„ tend towards £. 6. Find the error involved in using S„ = 1 + — + • • • + — &s an 1! re! approximation to e. Calculate e accurately to 5 decimal places. 7. The Concept of Limit where the Variable is Continuous Hitherto we have considered limits of sequences, that is, of functions of an integral variable n. The notion of limit, however, frequently occurs in connexion with the concepts of a continuous variable x and of a function /(a;). We say that the value of the function f(x) tends to a limit I as x tends to £, or in symbols lim/(z) = I, if all values of the function /(a;) for which x lies near enough to | differ arbitrarily little from I. Expressed more precisely, the condition is as follows: // an arbitrarily small positive quantity e is assigned, we can mark off about £ an interval | x — £ | < S so small that for every point x in this interval different from £ itself the inequality | f(x) — / | < e holds. Here we expressly exclude the equality of x and £. This is Q LIMIT OF A FUNCTION 47 done purely for reasons of expediency, so as to have the definition in a form more convenient for application, e.g. in the case where the fnncfaon/(ar) is undefined at the point £, although it is defined for all other points in a neighbourhood (p. 159) of £ . If our fun ction is defined or considered in a given interval only, e.g. Vl - a; 2 in -l^igl, we shall restrict the values of x to this interval. Thus if £ denotes an end-point of the interval, x is made to approach £ by values on one side of £ only (limit from the interior of the interval or one-sided limit). As an immediate consequence of this definition, we have the following fact: if Jim /(*) = I, and ^ x 2 , x & , . . . , x n , . . . is a sequence of numbers all different from £ but approaching £ as a limit, then lim f(x„) = I. ft — >» For let e be any positive number; we wish to show that for all values of n greater than a certain n the inequality !/(*„)- M<« holds. By definition, there exists a S > such that whenever I x ~ € | < S the inequality \f(*)-l\<* is true. Since x n -* £, the relation | x n - £\ < S is satisfied for all sufficiently large values of n; and for such values it follows that \f{x„) — 1 1 < e , as was to be proved. We shall now attempt to clarify this abstract definition by means of simple examples. Let us first consider the function sin a: f(x) = , x defined for x =t= 0. We state that .. sins km = l. We cannot prove this statement simply by carrying out the passage to the hmit in numerator and denominator separately, for the numerator and denominator vanish when x = 0, and the symbol 0/0 has no meaning. We arrive at the proof in the following way. 4» INTRODUCTION [Chap. From fig. 18 we find by comparing the areas of the triangles OAB and OAO and the sector OAB that if < x < n/2 sin x < x < tan x. Prom this it follows that if < | x \ < tt/2, x 1 1 < -r— < • sin x cos x Hence the quotient lies between the x numbers 1 and cos x. We know that cos x tends to 1 as x -*■ 0, and from this it follows that the quotient can differ only arbit- x rarily little from 1, provided that x is near enough to 0. This is exactly what is meant by the equation which was to be proved. Fig. 1 8 From the result just proved it follows that tan a; ,. sin a; 1 , lim = lim lim = 1» «— »-o x i_>o x »_>.o cos a; and also lim x— >-0 = 0. This last follows from the formula, valid for < | x | < - , 1 — cos x (1 — cos x) (1 + cos x) x x(\ + cosx) sin a; 1 x ' 1 + cos x cos 2 a; x(l + cos a;) . sinx. As x -v the first factor on the right tends to 1, the second to J and the third to 0; the product therefore tends to 0, as was stated. From the same formula, dividing by x, we obtain 1 -c sin x\ a 1 whence lim x I 1 + cos x 1 — cos a; 1 a? = 2" Finally, let us consider the function Vx 2 , defined for all values of x. This function is never negative; it is equal to x for x Si and to —a; for x < 0. In other words, Vx* = | x\. Consequently the function Vx*/x, which is denned for all non-zero values of x, has the value -(-1 when x > and — 1 when x < 0. It is therefore impossible for the limit I] LIMIT OF A FUNCTION 49 lim Vx*/x to exist, since arbitrarily near to we can find values of x for which the quotient is +1 and other values for which it is — 1. In concluding this discussion on limits in connexion with continuous variables we remark that it is of course possible to consider limiting pro- cesses in which the continuous variable x increases beyond all bounds. For example, the meaning of the equation i- x *+ 1 r 1 + «* , hm -j zr r= hm =1 "~ x* is clear without further discussion. It signifies that the function on the left differs arbitrarily little from 1, provided only that x is sufficiently large. In these examples we have proceeded as if operations with limits obeyed the same laws in the case of continuous variables as in the case of sequences. That this is actually true the reader can verify for himself; the proofs are essentially the same as for limits of sequences. Examples 1. Find the following limits, giving at each step the theorem on limits which justifies it: (a) lim 3x. (c) lim — — ^-. *-^a v ,_».i 2x+2 (6) lim 4a; + 3. (d) lim V5+f/2xh 2. Prove that . 3"— 1 sin a: , , , ,. sin(r 3 l (o) hm = n; (&) lim = 1; (c) lim — ' = 0. *— >1 X — 1 *— >-ir 77 — X *— >0 X 3. Find whether or not the following limits exist, and if they do exist find their values: Vl — x Vl + x Vl + z-Vl-x (a) hm ; (6) hm ; (c) lim ■ , *->0 X i_>0 X x->0 X 8. The Concept of Continuity 1. Definitions. We have already illustrated the notion of continuity in § 2 (p. 19) by means of examples. Now, with the help of the idea of a limit, we are in a position to make the concept of continuity precise. 3 (1798) 5° INTRODUCTION [Chap. We thought of the graph of a function which is continuous in an interval as a curve consisting of one unbroken piece; we also stated that the change in the function y must remain arbitrarily small provided only that the change of the independent variable x is restricted to a sufficiently small interval. This state of affairs is usually formulated as follows, with greater prolixity but increased precision. A function f(x) is said to be continuous at the point £ if it possesses the following property: at the point £ the value of the function /(f) is approximated to within an arbitrary pre-assigned degree of accuracy e by all functional values f(x) for which x is f&) +e \ 1 j- yf2^^ near enough to £ . In other words, f(x) is continuous at £ if for every positive number e, no matter how small, there can be de- termined another positive number 8 = 8(e) such that |/(*)-/(0|<€ (cf. fig. 19) for all points x for which | x — £ | < 8. Or again: the condition of continuity requires that for the point £ the limit equation lim /(*)=/(£) shall be true. The value of the function at the point £ is the same as the limit of the functional values f(x„) for any arbitrary sequence x n of numbers converging to £. It is important to observe that our condition involves two different things: (1) the existence of the limit lim f(x), and (2) the coincidence of this limit with /(f), the value of the function at the point £ . Having now defined continuity of a function /(aj) at a point £ , we proceed to state what we mean by the continuity of a function f(x) in an interval. This may be defined simply in the following way: the function f(x) is continuous in an interval if it is continuous at each point of that interval. Stated fully, this requires that if a positive number e be assigned, then for each point x of the interval there is a I] CONTINUITY 51 positive number 8, depending as a rule on e and on x, such that l/(*)-/(*)|<« if |*-*|<8, and x lies in the interval a g i g 4. Closely related to this is another concept, that of uniform continuity. The function f(z) is uniformly continuous in the interval a^x<lb if for every positive number e there is a corresponding positive number 8 such that for every pair of points a^, x 2 in the interval whose distance apart, | cc^ — sc 2 [, is less than 8 the inequality l/fo) — f(x 2 ) | < e holds. This differs from the definition stated above in that the 8 in the definition of uniform continuity does not depend on x, but is equally effec- tive for all values of x — hence the name uniform continuity. It is quite obvious that a uniformly continuous function is necessarily continuous. Conversely, it can be shown that every function f(x) which is continuous in a closed interval a ^x^b is also uniformly continuous. The proof of this we leave for the appendix (p. 64). Even though the reader may not desire to read the proof at present he will find it helpful to study the examples given at the beginning of Appendix I, § 2, No. 2 (p. 65). But until the student has worked through this proof he may assume that whenever a function is said to be continuous in a closed interval uniform continuity is meant. 2. Points of Discontinuity. We can understand the concept of continuity better if we study its opposite, the concept of discontinuity. The simplest type of discontinuity occurs at those points at which the function makes a jump; that is, at which the function has a definite limit as x tends to the point from the right and a definite limit as x tends to the point from the left, but these two limits are different. Whether or how the function is denned at the point of discontinuity itself does not matter. For example, the function f(x) defined by the equations f(x) = for ** > 1, f(x) = 1 for a" < 1, /(*) = | for x* = 1 has discontinuities at the points 5=1 and % = — 1. The limits on approaching these points from the right and from the left differ by I, and the values of the function at these points agree with neither limit, but are equal to the arithmetic mean of the two limits. 5« INTRODUCTION [Chap. It may be noted in passing that our function can be represented, using the idea of a limit, by the expression f(x)= lim 1 + x* For, if x* < 1, that is, if a; lies in the interval — 1 < x < 1, the numbers ie*" will have the limit 0, and the function will have the value 1. If, how- Fig. 20 ever, a; 8 > 1, as n increases a?" will increase beyond all bounds; our function will then have the value 0. Finally f or x* = 1, that is, for x = +1 and x=—l, the value of the function is plainly J (cf. fig. 20). Other curves with jumps are sketched in figs. 21a and 216; they represent functions having obvious discontinuities. In the case of discontinuities of this kind the limit from the right and the limit from the left both exist. We now pass on to the consideration of discontinuities in which this is not the case. The most important of such discontinuities are the infinite discontinuities or infinities. These y\ Fig. 2ia Fig. 2i* are discontinuities such as are exhibited by the functions 1/x or 1/a;* at the point 5=0; as x -> 5 the absolute value | f(x) | of the function in- creases beyond all bounds. In the case of 1/x the function increases numerically beyond all bounds through positive and through nega- tive values respectively as x approaches the origin from the right and from the left. On the other hand, the function 1/3? has for x = an infinite discontinuity at which the value of the function becomes posi- CONTINUITY 53 tively infinite from both sides (of. fig. 6, p. 18, and fig. 12, p. 22). The function y = . . shown in fig. 22, has infinite discontinuities both at x = 1 and at x = — 1. Finally, we shall illustrate by an example another type of discon- tinuity in which no limit from the right or from the left exists. We con- sider the function . y — sin -, x defined for all non-zero values of x. This function takes all values *2z*h- Fig. 22. — Function with infinite discontinuities Fig. 23. — Oscillating function with discontinuity between — 1 and +1 when the number 1/x runs through the values from (2» — %)tt to (2ra + J) 77, no matter what value the integer n has. At the 2 points x = — — — the function will have the value —1, at the points (in — l)ir r x = - — t-tt— it will have the value +1. From this we see that the func- (4n + 1)tt tion swings backwards and forwards more and more rapidly between the values + 1 and — 1 as x approaches nearer and nearer to the point x = 0, and that in the immediate neighbourhood of the point x = an infinite number of such oscillations occur (cf. fig. 23). It is interesting to observe that in contrast to the above example the 54 INTRODUCTION fCHAP. function «/= xsin l/x (of. fig. 24) remains continuous at the point x= if we assign to it the value at that point. This continuity is due to the faot that as the origin is approached the factor x damps the oscillations of the sine. Yet in the neighbourhood of the origin the function y= x sin l/x does not change from monotonic increasing to monotonic decreasing a finite number of times. On the contrary, it oscillates backwards and forwards an infinite number of times, the magnitude of these oscillations becoming as small as we please as the origin is approached. This example shows us that even the simple idea of continuity permits of all sorts of remarkable possibilities foreign to our naive intuitions. There is one important fact which, must be taken into consideration if we are to give our ideas greater precision. It may happen that at a certain point a function is not defined by the original law, as for example at the point x = in the last two examples dis- cussed. We have then the right to extend the definition of the function by assigning to it any desired value at such a point. In the last example, however, we can extend the definition in such a way that the function remains continuous at that point also, namely, by putting y = when x = 0. This can be done whenever the limits from the left and from the right both exist and are equal to one another; then we need only make the value of the function at the point in question equal to these limits in order to make the function continuous there. In the case of the function y = sin l/x this is not possible. 3. Theorems on Continuous Functions. In conclusion we quote the following important general theorems, whose proofs follow immediately from the remarks on operations with limits (p. 41): The sum, difference, and product of two continuous functions are themselves continuous. The quotient of two continuous functions is continuous at every point at which the denominator does not vanish. Fig. 24. — Continuous oscillating function CONTINUITY 55 In particular, it follows that all polynomials and all rational functions are continuous except at the points where the denomi- nator vanishes. The fact that the other elementary functions, such as the trigonometric functions, are continuous will follow naturally from later considerations (cf. pp. 69, 97). 1. Prove that 2. Prove that sin (x — a) Examples x"sin- x (a) lim X s - liiri ■_>.o sinx = 2a'' (6) lim X + cos * x+ 1 = 1; (c) lim cos lfx = 1. 3. (a) Let /(x) be defined by the equation y = 6x. Find a 8, depend- ing on 5, so small that | f(x) — /(5) | < e whenever | x — \ \ < 8, where (1) 6 = t>o; (2) e = xfo; (3) e = x^. Do the same for (6) f(x) = x* — 2x; (c) f(x) = 3** + x 2 - 7; (d) f(x) = Vx, * S: 0; (e) f(x) = Vx». 4. (a) Let /(*) = 6* in the interval g x ^ 10. Find a 8 so small that | f(x t ) — /(x s ) | < e whenever \x 1 — x t \<B, where (1) t = xfo; (2) e is arbitrary, > 0. Do the same for (6) f(x) = x« — 2x, — 1 ^ x g 1; (c) Jf(a;)=3i 4 + i>-7,2gxg4i (rf) /(x) = Vx, ^ x ^ 4; («) f(x) = V**, -2 g x g 2. 5. Determine which of the following functions are continuous, those which are discontinuous, find the points of discontinuity. For (a) x*sinx. (6) a: sin' (a*). (c) -sinx. x (d) sinx Vz' x* + 3x+7 x 1 — 6x + 8' x» + 3a; + 7 X s — 6x + 9' x» + 3a; + 7 x» — 6x + 10* (A) tanx. (/) (?) 1 sinx* (j) cotx. 1 oosar (I) a; cotx. (m) (je — x) tanx. (») (k) 56 INTRODUCTION [Chap. Appendix I to Chapter I Preliminary Remarks In Greek mathematics we find an extensive working-out of the principle that all theorems are to be proved in a logically coherent way by reducing them to a system of axioms, as few in number as possible and not themselves to be proved. This axiomatic method of presentation, which at the same time served as a test for the accuracy of the investigation, was at the be- ginning of the modern era regarded as a model for other branches of knowledge. For example, in philosophy such men as Descartes and Spinoza believed that they had made their investigations more convincing by presenting them axiomatically, or, as they called it, " geometrically ". But it was a different matter with modern mathematics, which began to develop at about the same time as the new philo- sophy. In mathematics the principle of reduction of the material to axioms was frequently abandoned. Intuitive evidence in each separate case became a favourite method of proof. Even in the case of scientists of the first rank we find operations with the new concepts based chiefly on a feeling for the right result and not always free from mystical associations — particularly in the case of the ominous " infinitely small quantities " or " infinitesimals ". Blind faith in the omnipotence of the new methods carried the investigator away along paths which he could never have travelled if subject to the limitations of complete rigour. It is no wonder that only the sure instinct of a great master could guard agair«t gross errors. It is fortunate that this was so, and that the critical counter- currents which sprang up in the eighteenth century and rose to their full strength in the nineteenth century did not come in time to check the development of modern mathematics, but only in time to establish and extend its results. But the need for critical investigation and consolidation of the advances made gradually increased to such an extent that its satisfaction is rightly regarded as one of the most important mathematical achievements of the nineteenth century. I] PRELIMINARY REMARKS 57 In the differential and integral calculus the critical work of Cauchy is particularly important. By formulating the funda- mental concepts in a clear and satisfactory way, Cauchy in many directions rounded off the work, begun in the eighteenth century, of presenting higher analysis in an intelligible manner, free from the vagueness due to the use of infinitesimals. The principal thing which remained to be done was to re- place intuitive considerations in proofs and discussions by con- siderations of pure analysis, depending only on numbers and on the operations which can be performed with numbers — as we say, to " arithmetize " analysis. As a matter of fact, the criti- cally-trained mind feels there is something unsatisfactory about appeals to intuition in proofs in analysis. We need not go into the question of the accuracy or inaccuracy of intuition or of the existence of a " pure a priori intuition " in Kant's sense in order to recognize that naive intuitive thinking includes much vagueness which hinders the approach to completely rigorous proofs in analysis. In the following chapters this will strike us more and more clearly. Even here we may mention, for example, that the concept of a continuous curve is very difficult to grasp in- tuitively. A continuous curve need not by any means possess a definite direction at every point. In fact, there actually exist continuous curves which at no point possess a direction, and continuous curves to which no length can be assigned. In the face of such facts, even the beginner will admit the need for arithmetizing analysis.* Yet we must not allow ourselves to forget that a century of brilliant and fruitful development of mathematics was possible before these requirements were fulfilled. In spite of all its defects intuition still remains the most important driving force for mathematical discovery, and intuition alone can bridge the gap between theory and application. We shall now follow Bolzano and Weierstrass in developing those lines of thought which yield the rigorous and complete proofs of the theorems which we formulated by intuitive means in the first chapter. * Rigorous mathematical concepts are always very highly idealized forma of the ideas which arise intuitively. Hence it is absolutely impossible to dispose of problems relating to the ultimate foundations of mathematics by appealing to naive intuition. S * (1798) 58 INTRODUCTION [Chap. 1. The Principle of the Point of Accumulation and its Applications 1. The Principle of the Point of Accumulation. In the rigorous discussion of the fundamentals of analysis the leading part is played by Weierstrass's principle of the point of accumulation. From the intuitive point of view this principle is merely the statement of a triviality; but just because it summarizes a state of affairs which occurs frequently it is as useful as small change is in daily life. The principle is as follows: If infinitely many numbers are given in a finite interval, these numbers possess at least one point of accumulation; that is, there is at least one point £ such that in every interval, no matter how small, about the point £ there lie infinitely many of the given numbers. In order to prove the principle of the point of accumulation arithmetically, we assume to begin with that the given interval is the interval from to 1. We now divide this into ten equal parts by means of the points 0-1, 0-2, . . . , 0-9. At least one of these sub-intervals must contain infinitely many points. Let us sup- pose that the interval beginning with the number 0-a 1 is that interval (or one of those intervals if there are several). We now subdivide this interval into ten parts by means of the points of division 0-asjl, 0-a t 2, . . ., Oe^J). Again, it is true that at least one of these sub-intervals must contain infinitely many points; let it be the sub-interval beginning with the number 0-o x a 2 . We again subdivide into ten parts — notice that one of these parts must contain infinitely many points — and continue the process. We thus arrive at a sequence of digits, a^, a 2 , a a , . . . , each having one of the values 0, 1, 2, . . . , 9. We now consider the decimal £ = O-a^^dg .... It is clear that this is a point of accumulation of our set of num- bers. For every interval, no matter how small, in whose interior the point £ lies, contains the sub-intervals of our system of sub- division from a certain degree of fineness onward, and these sub-intervals contain infinitely many numbers of the set. If the interval under consideration, instead of being the interval from to 1, is, say, the interval from a to a + ft, nothing essential in the above I] THE POINT OF ACCUMULATION 59 argument is changed. The point of accumulation is then represented simply by a number of the form a + h x O-Oxdja, .... 2. Limits of Sequences. The considerations above throw new light upon the concept of the limit of an infinite sequence of numbers a x , a z , a 3 , . . . , a„, . . . . We first consider the exceptional case in which infinitely many numbers of the sequence are equal to one another, and extend our definition by applying the name " point of accumulation " to this point (or these points) also. If there are infinitely many different numbers in the sequence, and if we assume that the numbers a„ of this sequence are " bounded ", i.e. that there is a number M such that the in- equality I a„ I < M holds for all values of n, the numbers of the sequence form an infinite set of numbers in a finite interval, since they all lie between — M and M. They must, therefore, possess at least one point of accumulation (£). If there is only one point of accumulation, it is easy to show that the sequence converges, and that its limit is the number g. For let us mark off any small interval about the point £. If infinitely many points of the sequence were outside this interval, they would have a limit point other than g, contrary to hypothesis. Hence only a finite number of the numbers of the sequence are exterior to the interval, and thus by definition the sequence approaches £ If, on the other hand, there are several points of accumulation, the sequence approaches no limit. The existence of a limit and the uniqueness of the point of accumulation of a bounded sequence of numbers are therefore equivalent ideas. The case of the non-existence of a limit is to be regarded as the rule rather than the exception. For example, the sequence with the terms Ojn = 1/n, Ojn-! = 1 — 1/n (» = 1, 2, . . .) has the two points of accumu- lation and 1. The aggregate of the positive rational numbers may be re- garded as a sequence of numbers, in which the ordering by magni- tude is, of course, completely destroyed. We arrive most easily at such an arrangement in a sequence if we first write down the rational numbers as shown on p. 60 and then run through this array as shown by the arrows, disregarding those numbers which IK! 60 INTRODUCTION [Chap. have already been encountered (such as 2/4). The system of rational numbers obviously has all rational and irrational points as points of accumulation. It therefore forms a simple example of a sequence with an infinite number of points of accumulation. By means of the concept of convergence we can state the principle of the poinb of accumulation in a remarkable form which is often convenient T~/2 y3~pT /5~T6 /i~*' for applications. 2/ |/ Z-/ 2./ Z-S 2/ . . From every bounded 3 y 3_/ 3/3/ 3./ . . infinite set of numbers it I /* /^ / * / 5 is possible to choose an T {'I s~5 *4 ' ' ' infinite sequence & v a 2 , \ ,z j,\ ' a 3> • • • which converges to *./*./.. a definite limit $. For Enumeration of the rational ttis PU*P<>Se We have numbers on ly to take a point of accumulation £ of the given set of numbers, then to select a number a^ of the set whose distance from | is less than 1/10, then a second number ffl 2 of the set whose distance from £ is less than 1/100, then a third number a 3 whose distance from £ is less than 1/1000, and so on. We see at once that this sequence actually con- verges to the limit |. 3. Proof of Cauchy's Convergence Test. Let us now return to convergent sequences, i.e. to bounded sequences with only one point of accumulation. Cauchy's con- vergence test, stated in § 6 (p. 40), now reduces almost to a triviality. For let us assume that \a m — a n | is arbitrarily small when m and n are sufficiently large. Then the numbers a„ all lie in a finite interval, and therefore possess at least one point of accumulation f . If now there were a second point of accumu- lation 17, the distance of this point from f would be | £ — -q\ = a, a positive quantity. Within an arbitrarily small distance from £, say within a distance less than a/3 from £, there must be infinitely many numbers a„, and hence, in particular, infinitely many numbers a„ for which n> N, however large N is chosen. Simi- larly, within an arbitrarily small distance from the point 77, say within a distance less than a/3 from 17, there are infinitely many I] THE POINT OF ACCUMULATION 61 numbers a m of the sequence; in particular, infinitely many numbers a m for which m > N. For these values a„ and a m it is true that | a m — a n | > a/3, and this relation is incom- patible with the hypothesis that for sufficiently large values of N the difference | a„ — a m | is arbitrarily small provided that n and m are both greater than N. Consequently there are not two distinct points of accumulation, and Cauchy's test is proved. 4. The Existence of Limits of Bounded Monotonic Sequences. It is equally easy to see that a bounded monotonic increasing or monotonic decreasing sequence of numbers must possess a limit. For suppose that the sequence is monotonic increasing, and let | be a point of accumulation of the sequence; such a point of accumulation must certainly exist. Then £ must be greater than any number of the sequence. For if a number a t of the sequence were equal to or greater than £, every number a„ for which n > I + 1 would satisfy the inequality a„ > o, + " > «i sS £. Hence all numbers of the sequence, except the first (I + 1) at most, would lie outside the interval of length 2(a t +i — |) whose mid-point is at the point £. This, however, contradicts the assumption that £ is a point of accumulation. Hence no numbers of the sequence, and a fortiori no points of accumulation, lie above $. So if another point of accumulation ij exists we must have -q < £. But if we repeat the above argument with ij in place of £ we obtain £ < ij, which is a contradiction. Hence only one point of accumulation can exist, and the convergence is proved. An argument exactly analogous to this of course applies to monotonic decreasing sequences. As on p. 41, we can extend our statements about monotonic sequences by including the limiting case in which successive numbers of the sequence are equal to one another. It is in this case better to speak of monotonic non-decreasing and monotonic non-increasing sequences respectively. The theorem about the existence of a limi t, remains valid for such sequences. 5. Upper and Lower Points of Accumulation; Upper and Lower Bounds of a Set of Numbers. In the construction on p. 58 which led us to a point of accumulation £ we had at each step to make the choice of a sub- 6a INTRODUCTION [Chap. interval containing infinitely many points of the set. Had we always chosen the last sub-interval which contained an infinite number of points, we should have been led to a certain definite point of accumulation /?. This point of accumulation )8 is called the upper point of accumulation or upper limit of the set of num- bers, and is designated by the abbreviation lim. It is that point of accumulation of the sequence which lies farthest to the right; i.e. it is quite possible that an infinite number of points of the sequence He above /?, but no matter how small the positive number e may be, there are not an infinite number above /? + e. If, in the construction on p. 58, we had always chosen the first of the intervals containing an infinite number of points of the set, we should again have arrived at a certain definite point of accumulation a. This point a is called the lower point of accumulation or lower limit of the set, and is denoted by lim. There may be infinitely many numbers of the set below o, but no matter how small the positive number e may be there are only a finite number below a — e. The proofs of these facts can be left to the reader. Neither the upper limit /? nor the lower limit a need belong to the set. For example, for the set of numbers a 2n = 1/w, a 2n-i = 2 — l/» these limits are respectively a = and )8 = 2, but the numbers and 2 do not themselves occur in the set. In this example there is no number of the set above /? = 2. In this case we say that j8 = 2 is also the upper bound M of the set, according to the following definition: M is called the least upper bound, or simply the upper bound, of a set of numbers if (1) there is no number of the set greater than M, but (2) for every positive number e there is a number of the set greater than M — e. The upper bound may coincide with the upper limit, as the above example shows. But the set a„ = 1 + l/n (n = 1, 2, . . .) shows that this is not necessarily the case. Here M = 2 and £ = 1. Every bounded set of numbers has a least upper bound. For let /8 be the upper limit of the set. Either there are no numbers of the set greater than /?, or there are such numbers. In the first case ft is the least upper bound, since no numbers are above /3 and there are numbers arbitrarily close to ft below it. In the second case let a be a number of the set greater than /?. There are only a finite number of numbers of the set equal to or greater I] THE POINT OF ACCUMULATION 63 than a, since otherwise there would be an accumulation point above fi, which is impossible. We therefore need only choose the greatest of these numbers; it will be the upper bound of the set. We see that in any case M ^ /}, and we recognize the following fact: If the upper bound of the set does not coincide with the upper limit, it must belong to the set, and is an " isolated " point of the set. Corresponding statements hold for the lower bound m; it is always equal to or less than a, and if m and a do not coincide, m belongs to the set and is an isolated point of the set. 2. Theorems on Continuous Functions 1. Greatest and Least Values of Continuous Functions. A bounded infinite set of numbers must possess a least upper bound M and a greatest lower bound m. But, as we have seen, these numbers M and m do not necessarily belong to the set; as we say, the set does not necessarily have a greatest 01 a least value. In view of this fact the following theorem on continuous functions is by no means so obvious as it appears to simple intuition: Every function f (x) which is continuous in a closed interval a Ss x <: b assumes a greatest value at least once and a least value at least once, or, as we say, it possesses a greatest and a least value. This may easily be proved in the following way. The values assumed by the continuous function f{x) in the interval « Sa a <J b form a bounded set of numbers and therefore possess a least upper bound M. For otherwise a sequence of numbers fi, €2, • • • , f«, ... in our interval would exist for which /(£„) increases beyond all bounds. This sequence would have at least one point of accumulation £ in the interval. Then arbi- trarily near to £ there would always be numbers £„ of our sequence for which the expression |/(|j— /(£„) | exceeds 1 (and in fact is arbitrarily large): that is, the function would be discontinuous at the point f Thus a least upper bound M exists and hence either there is a point £ such that f(£) = M, 64 INTRODUCTION [Chap. which would prove the statement, or there is a sequence of numbers x 1 , x 2 , . . . , x n , . . . in the interval for which ]imf{x n )=M. n— >-oo According to the principle of the point of accumulation as formu- lated on p. 60 we can select a sub-sequence of the numbers x n , which converges to a limit £. Let us call this sub-sequence £i» £» •••>&»••• i so that lim£ B =£. it— >■ 00 It is then certain that lim f(U=M. n—¥-<x> On the other hand, the function has been assumed continuous in the interval, and hence, in particular, at $, so that lim f(L) =/(€). n— >- oo Hence /(£) = M. The value M is therefore assumed by the function at a definite point f in the interior or on the boundary of the interval, as was stated. An exactly similar discussion applies to the least value. The theorem about the greatest and least values of con- tinuous functions does not remain true in general unless we expressly assume the interval to be closed, that is, unless we make the hypothesis of continuity refer to the end-points also. For example, the function y = 1/x is continuous in the open interval 0<sc<oo . It has, however, no greatest value, but has arbitrarily large values near x = 0. Similarly, it has no least value, but becomes arbitrarily near for sufficiently large values of x, without ever assuming the value 0. 2. The Uniformity of Continuity. As we have already seen (cf. p. 54), and as we shall further see, the continuity of a function f(x) in a closed interval a^x^b leaves room for a variety of possibilities which do not suggest themselves intuitively. For this reason we shall give logically rigorous proofs of certain consequences of the idea of continuity which from a naive point of view se«m quite I] CONTINUOUS FUNCTIONS 65 obvious. The definition of continuity simply states that from the relation lim x n = $ the relation lim/(a; n ) =/(f) follows. We can also express this in the following way: for each point f there corresponds to every e > a number S > such that \f( x ) —f(i) I < e whenever | x — £ | < S, provided that all the numbers x considered lie in the interval a^x^b. For example, in the case of the function y= ex (where c 4= 0) such a number S is given by the relation 8 = s/| c (. For the function y = x* we can find such a number in the following way. We assume that a = and 6=1, and ask ourselves how near to % the number x must lie in order that the expression | x 2 — f; 3 j may be less than e. For this purpose we write \a?~V\ = \z-Z\\x+f.\^\x-Z\(l+Z). If, therefore, we choose 8 ^ e/(l -f 5) we can be sure that | x 1 — ? a | < e. We see in this example that the number 8 found in this way depends not only on e, but also on the point of the interval at which we are investigating the continuity of the function. But if we give up the attempt to make the best possible choice of 8 for each |, we can eliminate this dependence of 8 on 5- For we need only replace % on the right by the number 1, thus obtaining for 8 the expression e/2, which is smaller than the previous expression for 8 but serves equally well for all points %,. The question now arises whether something similar does not hold for every function which is continuous in a closed interval. That is, we inquire whether it may not be possible to determine for each e a 8 = 8(e) depending on e only and not on £, such that the inequality I /(*)-/(£)!<* is true, provided | x — f | < S, for all values of £ at the same time (or, better expressed, uniformly with respect to £). As a matter of fact, this is possible merely as a consequence of the general definition of continuity, without any additional hypo- theses. This fact, which first attracted attention late in the nineteenth century, is called the theorem of the uniform continuity of continuous functions. We shall prove the theorem indirectly. That is, we shall show that the assumption that a function /(a;) exists which in a closed interval a 5S x ^ 6 is continuous and yet not uniformly continuous leads us to a contradiction. Uniform continuity means that if we wish to make the difference \f(u) —f{v) \ less than an arbitrarily chosen positive number e, the numbers w and v being chosen in the closed interval a^x^b, we need 66 INTRODUCTION [Chap. only choose u and v near enough to one another, namely, at a distance apart less than 8 = 8(e); it is immaterial where in the interval the pair of numbers u, v is chosen. Now, if f(x) were not uniformly continuous, there would exist a positive (perhaps very small) number a with the following property: to every number S n of an arbitrary sequence 8 V 8 2 , . . . of positive numbers tending to zero there corresponds a pair of values u n , v n of the interval for which | «„ — v„ | < 8„ and |/(«„) — f(v n ) | > o. Accord- ing to the principle of the point of accumulation the numbers w„ must have a point of accumulation f , and so the numbers v n must have the same point of accumulation. If we mark off an arbitrarily small interval | x — $ | < 8 about this point £, an infini t e number of the pairs u„, v n will lie in this interval. But this contradicts the assumed continuity of f(x) at the point £; for that requires, by Cauchy's convergence test, that for points a^ and x 2 near enough to £ l/(*i) -/(**) I <«. The uniformity of the continuity is thus proved. In our proof we have made essential use of the fact that the interval is closed.* In fact, the theorem of the uniformity of continuity does not hold for intervals which are not closed. For example, the function 1/x is continuous in the half -open interval < x 52 1, but it is not uniformly continuous. For no matter how small the length 8 (< 1) of an interval is chosen, the function will take values differing by any fixed number, say 1, in the interval, if only the interval lies near enough to the origin, say 8/2 5S a;5S 38/2. The non-uniformity of continuity is of course due to the fact that in the closed interval <2 x gS 1 the function possesses a discontinuity at the origin. If we had considered the example y = x* in the whole (open) interval — oo < x < oo instead of in a closed interval, it would not have been uniformly continuous. 3. The Intermediate Value Theorem. Another theorem which constantly recurs in analysis is the following: A function f(x), continuous in a closed interval a ^ x <2 b, which is negative for x = a and positive for x = b, (or conversely), assumes the value at least once in the interval. * Otherwise the point of accumulation \ need not belong to the interval. I] CONTINUOUS FUNCTIONS 67 Geometrically this theorem is trivial, since it merely states that a curve which begins below the a;-axis and ends above it must cut the axis somewhere in between. Analytically the theorem is very easily proved. In the interval there are an i nfini te number of points for which f(x) < 0; on account of the continuity of the function, in fact, this is true for a whole interval beginning at the point a. The set consisting of those points x for which f(x) < has a least upper bound g, which is greater than a. Since in every neighbourhood of £ there are points x for which f(x) < 0, we must have /(f) 5S (whence in particular £ 4= b). It is impossible, however, that f(£) < 0, for then f(x) would be negative in a sufficiently small neighbourhood of £, including values x greater than £ , in contradiction to the hypo- thesis that f is the upper bound of the values x for which/(a;) < 0. Therefore /(f) = 0, and our assertion is proved. A slight generalization of our theorem is: If we assume that f(a) = a and f(b) = /}, and if pis any value between a and j8, the continuous function f (x) assumes the value p at least once in the interval. For the continuous function <f> (*)=/(») — f» will have different signs at the two ends of the interval, and will therefore assume the value somewhere in the interval. 4. The Inverse of a Continuous Monotonia Function. If the continuous function y=f(x) is monotonic in the interval a 5S x :£ b, it will assume each value ju. between/(a) and /(&) once and only once; hence if y describes the closed interval between the values a=/(a) and /}=/(&), to each value of y there will correspond exactly one value of x. We can therefore think of a; as a single-valued function of y in this interval; that is, the function y =f(x) has a unique inverse. We assert that this inverse function x = <j>(y) is also a continuous monotonic function of y, as y varies within the interval between a and j8. The monotonic character of the inverse function x = <f>(y) is obvious. In order to prove its continuity we observe that from the monotonic character oif(x) it follows that l/K)-/(*i)| = hfc-yil>o, 68 INTRODUCTION [Chap. provided that a^ and x 2 are distinct numbers of the interval. If A is a positive number less than b — a, the function !/(* + *)-/(*) I is continuous in the closed interval a^x^b— h. At a point £ it therefore has a least value |/(£+ h) — /(£) | = a(h), which by our preceding remark is not zero.* From this we conclude that if x x and x 2 are two points in the interval for which | a^ — x 2 | ^ h, then \f(Xj) — f(x 2 ) | ^ a(h). But this implies the continuity of the inverse function. For if | y x — y 2 \ falls below the positive number a(h), then we must have | x 1 — x 2 1 < h; and hence if a positive number e is given, we need only choose 8 equal to a(e) in order to ensure that for all values y for which \y x — y | < 8 it is also true that | (f>(y x ) — <f>(y) | < e. We have therefore established the following theorem: If the function y = f (x) is continuous and monotonic m the interval a 5j x 5^ b, and f(a) = a, f(b) = /?, then it has a single-valued inverse function x = <f>(y), a ^ y ^ /?, and this inverse function is also continuous and monotonic. 5. Farther Theorems on Continuous Functions. We leave it to the reader to prove the following almost trivial fact: a continuous function of a continuous function is itself continuous. That is, if <j>(x) is a function continuous in the interval a Sg x ^ b, and its functional values lie in the interval a ^ <f>^ /?, and if in addition f(<f>) is a continuous function of ^ in this last interval, then /(<£(a;)) is a continuous function of x in the interval a 5* x ^ b. (Theorem of the continuity of functions of continuous functions.) It has already been mentioned on p. 54 that the sum, differ- ence, and product of continuous functions are themselves continuous, and that the quotient of continuous functions is continuous, pro- vided that the denominator remains different from zero. 3. Some Remarks on the Elementary Functions In Chapter I we tacitly assumed that the elementary func- tions are continuous. The proof of this is very simple. First, * On account of the continuity of f(x), a(h) itself of course tends to as A does. I] THE ELEMENTARY FUNCTIONS 69 the function f(x) = x is continuous; therefore x 2 = x . x is continuous, being the product of two continuous functions, and every power of a; is likewise continuous. Thus every polynomial is continuous, being the sum of continuous functions. Every rational fractional function, being a quotient of continuous functions, is likewise continuous in every interval in which the denominator does not vanish. The function x" is continuous and monotonic. Hence the n-th root, being the inverse function of the n-th power, is con- tinuous. By the theorem of the continuity of functions of con- tinuous functions, the n-th root of a rational function is con- tinuous (except where the denominator vanishes). The continuity of the trigonometric functions, with which the reader is familiar from elementary mathematics, could now readily be proved, using the concepts developed above. The discussion is not given here, since in Chapter II, § 3 (p. 97), this continuity will be seen to follow naturally as a consequence of differentiability. We shall merely make a few remarks about the definition and continuity of the exponential function a x , the general power function a;", and the logarithm. We assume, as in § 3, pp. 25-26, that a is a positive number, say greater than 1, and if r = pfq is a positive rational number (p and q being integers) we take a? = a? 1 " as meaning the positive number whose q-th power is a". If a is any irrational number and r v r % , . . . , r m , . . . is a sequence of rational numbers approaching o, we assert that lim a'm exists; we then call this limit a°. IB— »-°0 In order to prove the existence of this limit, by Cauchy's test we need only show that | a'» — a r m | is arbitrarily small, provided that n and m are sufficiently large. We suppose for example that r„ > r ro , i.e. that r n — r m = S, where 8 > 0. Then o r » — a r »» = a r >»(a s — 1). Since o r "» remains bounded, we need only show that I a" - 1 1 = a 8 - 1 is arbitrarily small when the values of n and m are sufficiently large. But 8 is a rational number, and certainly may be made as small as we please provided the values of n and m are suffi- 70 INTRODUCTION [Chap. ciently large. Hence if I is an arbitrarily large positive integer, 8 < l/l if n and m are large enough. Now the relations S < l/l and a > 1 give * and since a 1 " tends to 1 as I increases (cf. p. 31) our assertion follows immediately. It may be left to the reader to show that the function a" extended to irrational values in this way is also continuous everywhere, and, moreover, that it is a monotonic function. For negative values of x this function is naturally defined by the equation . 1 a" =* — . As x runs from — oo to +oo, a" takes all values between and + oo. Consequently it possesses a continuous and monotonic inverse function, which we call the logarithm to the base a. In like manner we can prove that the general power of is a con- tinuous function of x, where a is any fixed rational or irrational number and x varies over the interval < x < oo; also that sc* is monotonic if a 4= 0. The " elementary " discussion of the exponential function, the logarithm, and the power a? outlined here will later (Chapter HI, § 6, p. 167) be replaced by another discussion which is in principle much simpler. Examples 1. Giv« the upper and lower bounds and upper and lower limits for the following sequences, and state which belong to the sequence: («) — . n= 1, 2 (5) 0, L_ iL, n = 1, 2, »! n! n 2n+ 1 n 2n + 1 («) —.+ —., m, n= 1, 2, . . . . m 3 » a * This statement follows from the fact that when a > 1 the power o"* is greater than 1 if m/n is positive. This is clearly true. For if o m/ " were less than 1, then o"* — (o m ' n )" would be a product of n faotors all less than 1, and would be less than 1. On the contrary, a" is the product of m faotors all greater than 1, and so is greater than 1. I] THE ELEMENTARY FUNCTIONS 71 2.* Prove that if f(x) is continuous for a g x g 6, then for every e > there exists a polygonal function <p(x) (that is, a continuous function whose graph consists of a finite number of rectilinear segments meeting at corners) such that I f(x) — <p(x) | < e f or every x in the interval. 3. Prove that every polygonal function <p{x) can be represented by a sum <p(a;) == a + bx + Sc,- 1 x — x t \ , where the x ( '8 are the abscissas of the corners. Find a formula of this kind for the function /(a;) denned by the equations: f(x) = 2i-l(0ga:g2). f(x) ~ 5 - x (2 g x S 3). f(x) = x — 1 (3 ^ x g 5). f(x) = 4(5^^7). 4. For the following functions f(x) find as in § 1, No. 2, p. 65, a 8(s) such that \f(xj) — f( Xi ) | < s whenever | x^ — x 2 | < 8(e): (a) /(*) = 2* 8 , -1 ^zgl. (b) ftx) = x», -a^x^a. *(«) A*) = \A = *i -1S«S1. 6.* The function y= sinl/a has no discontinuities in the interval < x < 1. Prove that it is rao< uniformly continuous in that open interval. 6. A function /(*) is defined for all values of a; in the following manner: f(x) = for all irrational values of x; Six) = 1/g for rational x = pjq, where p/q is a fraction in its lowest terms; (thus for x = ff , f(x) = ^-). Prove that /(a;) is continuous for all irrational values of * and dis- continuous for all rational values of x. Appendix II to Chapter I 1. Polab Co-ordinates In Chapter I we have set the concept of function in the fore- ground and have represented functions geometrically by means of curves. It is, however, useful to recall * that analytical geo- metry follows the reverse procedure, beginning with a curve given * See also p. 16. 72 INTRODUCTION [Chap. by some geometrical property and representing this curve by a function, for example, by a function which expresses one of the co-ordinates of a point of the curve in terms of the other co-ordinate. This point of view naturally leads us to consider, apart from the rectangular co-ordinates to which we restricted ourselves in Chapter I, other systems of co-ordinates which may be better suited for the repre- sentation of curves given geometri- cally. The most important example is that of polar co-ordinates r, 8, which are connected with the rectangular co-ordinates x, y of a point P by the equations x=rcos0, y=rsin0, r 2 =x 2 +y 2 , tan0=^, Fig. 25. — Polar co-ordinates and whose geometrical interpretation is made clear in fig. 25. Fig. 26. — Lemniscate Let us consider for example the lemniscate. This is geometrically de- fined as the locus of all points P for which the product of the distances r t and r 2 from the fixed points F 1 and F t with the rectangular co-ordinates x = a, y = and x = —a, y = respectively has the constant value a 2 (cf. fig. 26). Since r,« = (x - a)* + y», r 2 » = (x + a)« + y\ a simple calculation gives us the equation of the lemniscate in the form (** + y 2 ) 8 - 2a 2 (a* — y 2 ) = 0. If we now introduce polar co-ordinates, we obtain r* — ZoV^cos'e — sin 2 e) = 0; I] POLAR CO-ORDINATES 73 and if we divide by r a and use a simple trigonometrical formula this becomes „ . . „„ J- 2 = 2a s cos 20. Thus we see that the equation of the lemniscate is simpler in polar co- ordinates than in rectangular. 2. Remarks on Complex Numbers Our studies will be based chiefly on the class of real numbers. Nevertheless, with a view to the discussions in Chapters VIII, IX, and XI we would remind the reader that the problems of algebra have led to a still wider extension of the concept of number, namely, to the introduction of complex numbers. The advance from the natural numbers to the class of all real numbers arose from the desire to eliminate exceptional phenomena and to make certain operations, such as subtraction, division, and correspondence between points and numbers, always possible. Similarly, we are compelled, by the requirement that every quadratic equation and in fact every algebraic equation shall have a solution, to introduce the complex numbers. If, for example, we wish the equation x 2 + 1 = to have roots, we are obliged to introduce new symbols i and — i as the roots of this equation. (As is shown in algebra, this is sufficient to ensure that every algebraic equation shall have a solution.*) If a and b are ordinary real numbers, the complex number c= a-\- ib denotes a pair of numbers (a, b), calculations with such pairs of numbers being performed according to the follow- ing general rule: we add, multiply, and divide complex numbers (among which the real numbers are included as the special case b = 0), treating the symbol i as an undetermined quantity, and then simplify all expressions by using the equation i 2 = — 1 to remove all powers of i higher than the first, thus leaving only an expression of the form a -f- ib. We may assume that the reader already has a certain degree of familiarity with these complex numbers. We shall neverthe- * That every algebraic equation possesses real or complex roots is the statement of the " fundamental theorem " of algebra. 74 INTRODUCTION [Chap. less emphasize a particularly important relationship which we shall explain in connexion with the geometrical or trigonometrical representation of the complex numbers. If c = x -f- iy is such a number, we represent it in a rectangular co-ordinate system by the point P with co-ordinates x and y. By means of the equa- tions x = r cos 6, y= r sin 0, we now introduce the polar co- ordinates r and (cf. p. 72) instead of the rectangular co-ordinates x and y. Then r 2 = Va^ + y 2 is the distance of the point P from the origin, and is the angle between the positive a;-axis and the segment OP. The complex number c is now represented in the form c = r(cos0 + t sin0). The angle is called the amplitude of the complex number c, the quantity r its absolute value or modulus, for which we also write \c\. To the "conjugate" complex number o = x — iy there obviously corresponds the same absolute value, but (except in the case of negative real values of c) the angle — 0. Clearly r* = | c | 2 = cc = a 2 + f. If we use this trigonometrical representation the multipli- cation of complex numbers takes a particularly simple form. For then c.c' = r(cos0 + t sin0) . r'(cos0' + i sin0') == ri (cos 8 cos 8' — sin 8 sin 8') + i (cos 8 sin 8' -f sin 8 cos #'). If we recall the addition theorems for the trigonometric functions this becomes c.c' = rr'(cos(0 + 0') + i sin(0 + 0') ). We therefore multiply complex numbers by multiplying their absolute values and adding their amplitudes. The remarkable formula (cos + i sin 0) (cos 0' + % sin $') = cos (0 + 0') + i sin (0 + 0') is usually called De Moivre's theorem. It leads us at once to the relation (cos + » sin0) n = cosn0 + t sinw0, I] COMPLEX NUMBERS 75 which e.g. at once enables us to solve the equation x n = 1 for positive integers n, the roots (the so-called roots of unity) being «i = « = cos h"in — , e 2 = e 2 = cos \~ % sin — , . . ., n n n n n-l (n— 1)7T . . . (w— 1)tt „ , «„_! = e« l = cos 5 i 1- 1 sin^ '— , e„ = c" = 1. Moreover, if we imagine the expression on the left-hand side of the equation (cos0 + isaxd) n = cosn0+ isaxnd ex- panded by the binomial theorem, we need only separate real and imaginary parts in order to obtain expressions for cos«0 and sinn0 in terms of powers and products of powers of sin0 and cos0. Examples 1. Plot the graphs of the following functions: r = sin 9. r = eos 5<p. r=f. 1 „• a r — — ; ;> * constant. r = sin 69. cos (9 — a) 2. Find the polar equation of (a) the circle of radios a with its centre at the origin; (6) the circle of radius a with centre (a, 9 ); (c) the general straight line. 3. Use De Moivre's theorem to express cos 29 and sin 29 in terms of sin 9 and 00s 9. Similarly, for cos 39, sin 39, cos 59, sin 59. Prove that cos»9 is a polynomial in cos 6, and also that if n is odd sin »6 is a polynomial in sin 9. 4. Work out the following expressions, and state the modulus and amplitude of each of the numbers involved and of the answers: (a) -3.2*. (f)i 11 *. (6) (4 + 4t)(i - i V3»). (g) (1 -f- if. (c) (1 + i)(l - t). (A) (3 - 3») 2/ ». (d) (V3 - »•)«. (fc) li/». (e) l 1/a . (/) (16*) 1 ". 5.* Prove that if s = cos — + » sin — , where n is an integer Greater than 1, n n "* K«»rer e r _|_ 6 ai- _l e s>- _l # # _ 1 s ny = /0 if m is not a factor of v, \n if n is a factor of v. CHAPTER II The Fundamental Ideas of the Integral and Differential Calculus Among the limiting processes of analysis there are two which play an especially important part, not only because they arise in many different connexions, but chiefly because of the very close reciprocal relation between them. Isolated examples of these two limiting processes, differentiation and integration, were considered even in classical times; but it is the recognition of their complementary nature and the resulting development of a new and methodical mathematical procedure that marks the beginning of the real systematic differential and integral calculus. The credit of initiating this development belongs equally to the two great geniuses of the seventeenth century, Newton and Leibnitz, who, as we know to-day, made their discoveries independently of one another. While Newton in his investigations may have succeeded in stating his concepts more clearly, Leibnitz's notation and methods of calculation are more highly developed; even to-day these formal portions of Leibnitz's work form an indispensable element in the theory. 1. The Definite Integral We first encounter the integral in the problem of measuring the area of a plane region bounded by curved lines. More refined considerations then permit us to separate the notion of integral from the naive intuitive idea of area, and to express it analyti- cally in terms of the notion of number only. This analytical definition of the integral we shall find to be of great significance, not only because it alone enables us to attain complete clarity Chap. II] THE DEFINITE INTEGRAL 77 in our concepts, but also because its applications extend far beyond the calculation of areas. We begin by considering the matter intuitively. 1. The Integral as an Area. Let us suppose that we are given a function f(x) which is continuous and positive in an interval, and that a and b (a<b) are two values in that interval. We think of the function as re- presented by a curve, and consider the area of the region which is bounded above by the curve, at the sides by the straight lines x = a and x = b, and below by the portion of the ataxia be- tween the points a and b (fig. 1). That there is a definite meaning in speaking of the area of this region is an assumption inspired by intuition, which we Fig. i a, b ~x Fig. a. — Upper sum and lower sum here state expressly as a hypothesis. We call this area F a *> the definite integral of the function i(x) between the limits a and b. When we actually seek to assign a numerical value to this area, we find that we are in general unable to measure areas with curved boundaries; but we can measure polygons with straight sides by dividing them into rectangles and triangles. Such a subdivision of our area is usually impossible. It is, however, only a short step to conceive of the area as the limiting value of a sum of areas of rectangles, in the following manner. We subdivide the part of the z-axis between a and b into n equal parts, and at each point of division we erect the ordinate up to the curve; the area is thus divided into n strips. We can no more calculate the area of such strips than we could that of the original surface; but if, as shown in fig. 2, we find first the least and then the greatest value of the function f(x) in each sub- interval, and then replace the corresponding strip (1) by a rectangle 7« FUNDAMENTAL IDEAS [Chap. whose height is equal to the least value of the function, and (2) by a rectangle whose height is equal to the greatest value of the function, we obtain two step-shaped figures. (In fig. 2 the first of these is drawn with a solid line, the second being shown by dotted lines.) The first step-shaped figure obviously has an area which is at most equal to the area F a b which we are trying to determine; the second has an area which is at least as large as F a ". If we denote the sum of the areas of the first set of rect- angles by F n (lower sum), and the sum of the areas of the second set by F n (upper sum), we have the relation F\^ F a "^T n . If we now make the subdivision finer and finer, i.e. let n increase without limit, intuition tells us that the quantities F n and F„ approach closer and closer to each other and tend to the same limit F a b . We may therefore consider our integral as the limiting value F a " = Km F n = lim T„. n " Intuition also shows us the possibility of an immediate generalization. It is by no means necessary that the n sub- intervals should all be of the same length. They may, on the contrary, have different lengths, provided only that as w increases the length of the longest sub-interval tends to 0. 2. The Analytical Definition of the Integral. In the above section we have considered the definite integral as a number given by an area, and hence to a certain extent pieviously known, and have subsequently represented it as a limiting value. We shall now reverse the procedure. We no longer take the point of view that we know by intuition how an area can be assigned to the region under a continuous curve, or, indeed, that this is possible; we shall, on the contrary, begin with sums formed in a purely analytical way, like the upper and lower sums defined previously, and shall then prove that these sums tend to a definite limit. We take this limiting value as the definition of the integral and of the area. We are naturally led to adopt the formal symbols which have been used in the integral calculus since Leibnitz's time. n] THE DEFINITE INTEGRAL 79 ke* f( x ) De a function which is positive and continuous in the interval agigi (of length b — a). We think of the in- terval as divided by (n — 1) points x 1 , x 2 , . . . , x n _ ± into » equal or unequal sub-intervals, and in addition we put x = a, x n = b. In each interval we choose a perfectly arbitrary point, which may be within the interval or at either end; suppose that in the first interval we choose the point fj, in the second the point £ 2 , • • • > in the n-th the point £„. Instead of the continuous function f(x) we now consider a discontinuous function (step- function) which has the constant value /(&) in the first sub- interval, the constant value /(f 2 ) in the second sub-interval, . . . , Fig. 3. — To illustrate the analytical definition of integral the constant value /(£„) in the n-th sub-interval. As is shown in fig. 3, the graph of this step-function defines a series of rectangles, the sum of whose areas is given by the expression F n = (x, - *„)/(&) + (x 2 - *,)/(£,) +... + («.- *„-!)/(£„). This expression is usually shortened by using the summation sign 2: *■„=£(*„-*„_!)/(£); K-l by introducing the symbol Aa;,, = x v — x r _ t we can abbreviate it still further: v-l 80 FUNDAMENTAL IDEAS [Chap. (Here the symbol A is not a factor, but denotes " difference ". The whole inseparable symbol Ax„, by definition, means the length of the y-th sub-interval.) Our basic assertion may now be stated as follows: If we let the number of points of division increase without limit and at the same time let the length of the longest sub-interval tend to 0, then the above sum tends to a limit. This limit is independent of the particular manner in which the points of division x 1; x 2 , . . . , x n _ 1 and the intermediate points £ v £ 2 , . ■ . , £„ are chosen. This limiting value we shall call the definite integral of the function/(a:), the integrand, between the limits a and b; as we have already mentioned, we shall consider it as the definition * of the area under the curve y=f{x), for a^x^b. Our basic assertion may then be re-worded thus: If f(x) is continuous in a ^ x ^ b its definite integral between the limits a and b exists. This theorem on the existence of the definite integral of a continuous function can be proved by purely analytical methods, without appealing to intuition. We shall nevertheless pass it over for the present and return to it in the Appendix to this chapter (p. 131), after the use of the concept of integral has stimulated the reader's interest in constructing a firm foundation for it. For the moment we shall content ourselves with the fact that the intuitive considerations on pp. 77-78 have made the theorem appear extremely plausible. 3. Extensions, Notation, Fundamental Rules. The above definition of the integral as the limit of a sum led Leibnitz to express the integral by the following symbol: £f(x)dx. The integral sign is a modification of a summation sign which had the shape of a long S. The passage to the limit from a sub- division of the interval into finite portions Ax v is suggested by the use of the letter d in place of A. We must, however, guard ourselves against thinking of dx as an " infinitely small quan- tity " or " infinitesimal ", or of the integral as the " sum of * Of course we may also define the notion of area in a purely geometrical way, and then prove that snch a definition is equivalent to the above limit- definition (ci. Chap. V, § 2, No. 1 (p. 268) ). II] THE DEFINITE INTEGRAL 81 & Fig. 4 an infinite number of infinitely small quantities ". Such a con- ception would be devoid of any clear meaning; it is only a naive befogging of what we have previously carried out with precision. In the above figures we have assumed (1) that the function f(x) is positive throughout the interval, and (2) that b > a. The formula which defines the integral as the limit of a sum is, how- ever, independent of any such assumptions. For if f(x) is negative in all or part of our interval, the only effect is to make the corre- sponding factors /(£,) in our sum negative instead of positive. To the region bounded by the part of the curve below the a;-axis we shall naturally assign a negative area, which is in agreement with the familiar convention of sign in analytical geometry. The total area bounded by a curve will thus in general be the sum of positive and negative terms, corresponding respectively to the portions of the curve above and below the as-axis.* If we also omit the condition a < b and assume that a > 6, we can still retain our arithmetical definition of integral; the only change is that when we traverse the interval from a to b the differences Ax p are negative. We are thus led to the relation £f(x)dx=—£f(x)dx, which holds for all values of a and b (a =)= b). In conformity with this we define / f(x)dx as equal to zero. Our definition immediately gives the basic relation (see fig. 4): fj(x)dx + f b f(x)dx = fj{x)dx for a < b < c. By means of the preceding relations we at once find that this equation is also true for any position of the points a, b, c relative to one another. We obtain a simple but important fundamental rule by • For the area of regions bounded by arbitrary closed curves see Chap. V, § 2, p. 269. 4 (1708) 8» FUNDAMENTAL IDEAS [Chap. considering the function cf(x), where c is a constant. From the definition of the integral we immediately obtain J cf(x) dx — c J f(x) dx. Further, we assert the following addition rule: If then j^f(x)dx — J <f>(x)dx + J tfi(x)dx. The proof is quite simple. We add a final remark, which is perfectly obvious, but very important in applications, about the " variable of integration ". We have written our integral in the form / f(x)dx. For evaluating the integral it does not •'a matter whether we use the letter x or any other letter to denote the abscissae of the co-ordinate system, i.e. the independent variable. The particular symbol we use for the variable of integration is therefore a matter of complete indifference; instead of / f(x)dx we could equally well write pb pb J a I f(t)dt or / f(u)du or any similar expression. 2. Examples We are now in a position to carry out the limiting process prescribed by our definition of the integral, and thus actually to calculate the area in question in a number of special cases; this we shall do in a series of examples, where (except in No. 5, p. 86) we shall make use of the upper or lower sum alone.* 1. Integration of a Linear Function. We first consider the function f(x) = x", where n is an integer greater than or equal to 0. For n = 0, i.e. for f(x) = 1, the result is so obvious that we simply write it down: rldx— I dx= b — a. For the function f(x) = x the integration is again a triviality from the geometrical point of view. The integral of the function f(x) = x. £ xdx, * We leave it as a useful exeroise for the reader to prove that in the follow- ing examples we actually do arrive at the same result, whether we use the upper sum or the lower. m EXAMPLES OF INTEGRATION 83 is simply the area of the trapezoid shown in fig. 6, which by an elemen- tary formula has the value J(6-o)(6 + a)=i(6 s -o»). We shall now verify that our limiting process leads to exactly the same result. In calculating the limit we can restriot ourselves to the discussion of upper sums or lower sums. We subdivide the interval from a to 6 into n equal parts by means of the points of division a + h, a + 2h, . . . , a + (» — l)h, where h — (6 — a)/n. The integral must then be the limit of the foDow- ing sum, which is an upper sum if 6 < a and a lower sum if 6 > a: h{a + (a + h) + (a + 2h) + . . . + (a + n— lh)} = h{na + h + Zh + . . . + (m - l)h}. ^> Fig. 6 By an elementary formula we have 1 + 2 +... + (»_ l) = in(n - 1), and our expression may therefore be written in the form As n increases the right-hand side obviously tends to the limit (6 -a){a+ i(b - a)} = J(6 2 - a"), which was to be proved. 2. Integration of jc 2 . A problem not quite so simple is that of integrating the function f(x) = a; 2 , or, in geometrical language, of determining the area of the region bounded by a segment of a parabola, a segment of the x-axis and two ordinates. We consider e.g. the integral •'n 3? dx, where 6^0 (see fig. 6), and divide the interval ^ x ^ 6 into » equal 84 FUNDAMENTAL IDEAS [Chap. parts of length h = b/n; the area which we wish to find is then the limit of the following expression (upper sum): h(h* + 2W + 3% 2 + • • • + » 2 A 2 ) = A 3 (l 2 + 2 2 + . . . + »») = fc^l 2 + 2 2 + . . . + n 2 )/» 3 . The sum enclosed in brackets, however, has already been found (see p. 27, footnote): 1» + 2* + • • • + » 2 = in(n + 1) (2ra + 1). If we substitute this expression and rewrite the result in a slightly different form, our sum becomes 6 s i !('+D(*+D- As n increases beyond all bounds this tends to the limit ^6 3 , and we obtain the required integral formula •'n x 2 dx = i& 3 . From this, using the general relationships given above, we immediately derive the general formula fx* dx = \x*dx- fx* dx = %(b* - a 3 ). J a Ja •'0 3. Integration of x a , where a is any Positive Integer. As a third example we consider the integration of the function V = /(*) = * a . where a is any positive integer. For the computation of the integral ■b xPdx "a (where we assume < a < b) it would be inconvenient to divide the interval into n equal parts.* The passage to the limit may, however, be accomplished very easily if we effect a subdivision in " geometric pro- gression " in the following manner. We put \/b\a = q and subdivide the interval by the points a, aq, aq\ .a, ag"" 1 , ag" = 6. * We should then be obliged to base the evaluation of the integral upon the calculation of the limit of — L^ (l a + 2 a + . . . + n a ) as n -»■ °o ; the reader may work this out for himself as indicated in the footnote on d. 27. II] EXAMPLES OF INTEGRATION 85 The required integral is then the limit of the following sum: a« (aq-a)+ (aqf (ag 2 - aq) + (aq*) a (a« 3 - og 2 ) + . . . + (tup- 1 )* (aq n — aq"- 1 ) = «+l(g_l){l-)- 3 a+l_j_ 3 2(a+l)_^_ g 3(a+l)+ . . . + 3 (n-l)(a+l) }, The terms in the last bracket form a geometric progression with com- mon ratio q a+1 4= 1. If we sum this progression, we obtain for the whole expression the value "" +1 (g-D g ga+1 _ I • We now replace q by its value (b/a) 1 !"; our sum then takes the form q-l (6«+l _ n «+l) S*+i— 1* If we now let n increase without limit, the first factor retains its value. Since q 4= 1 we can use the formula for the sum of a geometric progression and write the second factor in the form r + r- 1 + • ■ ■ + I' and as the equation q = (b/a) 1 !" shows that q tends to 1 as » -> 00 , the second factor will have the limit l/(a + 1). Thus finally the value of our integral is given by the formula "a xP-dx— (6*+ 1 — a a+1 ). a -f- 1 The above calculation is simple in principle, but somewhat compli- cated in detail. We shall later find that it can be entirely avoided, once we are better acquainted with integration theory. 4. Integration of x a , where a is any Rational Number other than — 1. The result obtained above may be generalized considerably without essentially complicating the method. Let a = r/« be a positive rational number, r and s being positive integers; then in the evaluation of the integral given above nothing is changed except the evaluation of the limit -^rj as q approaches 1. This expression is now simply . T_ ., . Let us put q 1 !* = t (t 4= 1): then as q tends to 1, t will also tend f» 1 to 1. We have therefore to find the limiting value of as t T r+ * — 1 approaches 1. If we divide both numerator and denominator by t — 1 and transform them as before the limit simply becomes T"- 1 + f-» + • • • + 1 hm ; — ; ; . - — - +1 8< 5 FUNDAMENTAL IDEAS [Chap. Since both numerator and denominator are continuous in t this limit is at once determined if we substitute t = 1. We thus arrive at the limit s 1 ~, — = — r^i ; anc * so f° r every positive rational value of a we obtain the integral formula •J a x"dx = (b a + 1 — a a + 1 ). oc -f- 1 This formula remains valid for negative rational values of a, provided we exclude the value a = — 1 for which the formula used above for the sum of the geometric progression loses its meaning. We have now to investigate the limit of the expression — vr , for negative values of a, say a = —r/s. To do this we put q~ x l' = t; we obtain q = t - *, q a+1 = g-fr— «)/» = T r_ ». We accordingly seek to determine the limiting value of t-« — 1 _ 1 — t* T r-« _ i _ T r _ T »* We leave it to the reader to prove that this limit is again equal to — —— ; that is, that we have the integral formula * a x°-dx = (6 a + 1 — o a+1 ) a+ 1 for the general case of rational values of a either positive or negative, with the exception of a = — 1. The form of the right-hand side of this equation shows that the ex- pression is not valid for a = — 1, since both numerator and denominator would then be zero. It is natural to suppose that the range of validity of our last formula extends also to irrational values of a. We shall aotually establish this in § 7 (p. 129) by a simple passage to the limit. 5. Integration of sin a; and cos a;. As a last example we consider the function f(x) = sin as. This too we shall treat by means of a special device. We express the integral J a saxxdx i as the limit of the following sum: 8 h = *{sin(a + h) + sin(a + 2h) + . . . + sin(a -f nh)}, where h = . We multiply the right-hand bracket by 2 sin _, and II] EXAMPLES OF INTEGRATION 87 recall the well-known trigonometrical formula 2 sinu sine = cos(u — ») — cos(« + v); provided h is not a multiple of 2n, we thus obtain the formula A_ j cos ( a + - J — cos ( a + - h J + cos f a + - h J — cos (a + - h J *»- 2 sin - h [ / , h\ I , 2n 4- 1 2 sin- (.+»)-„(. + t+i»)j. 2 Since a + «* = 6, the integral becomes the limit of A Jcos(a+*)-cos(6 + *)| M *^0. 2 sin- 2 Now we know from Chapter I (p. 47) that as h tends to 0, the expres- sion r/sin- approaches the limit I. The desired limit is then simply 00s a — cos 6, and we thus arrive at the integral formula sinx dx = — (cosi — cos a). r- In the same way, as the reader m*y verify for himself, we obtain the formula •'o cosx dx = sin 6 — sin a. a Almost every one of these examples has been attacked by means of some special method or particular device. The essential point of the systematic integral and differential calculus is the very fact that instead of such special devices we utilize considerations of a general character which lead us directly to the desired result. In order to arrive at these methods we must now turn our attention to the other fundamental con- cept of higher analysis, the derivative. Examples 1. Find the area bounded by the parabola y = 2a? + x + 1, the ordinates « — 1 and x= 3, and the z-axis. 2. Find the area bounded by the parabola y = %x* -+- 1 and the straight line y = 3 + x. 88 FUNDAMENTAL IDEAS [Chap. 3. Find the area bounded by the parabola y 2 = 5x and the straight line y = 1 + x. 4. Find the area bounded by the parabola y = x 2 and the straight line y = ax + 6. 5. Using the methods in the text, evaluate the integrals rb rb rb (a) I (x + l) a dx, (6) j sinax&e, (c) f co&axdx, Ja Ja Ja where a is an arbitrary integer. 6. Use the formulae obtained in Ex. 5, along with the identities sin 2 * = I — b cos 2b, cos 2 x = J + £ cos 2*, to prove that r b „ , b — a , sin26 — sin2a / COS Z XdX= ;;— + Ja rb I sva i xdx = Ja 2 4 6 — a sin26 — sin2a 7. By use of Ex. 1, p. 28, evaluate / x*dx, using division into equal J a -intervals. 8. Evaluate / (1 — x) n dx (where n is an integer) by expanding the sub-intervals, bracket. 3. The Derivative The concept of the derivative, like that of the integral, is of intuitive origin. Its sources are (1) the problem of constructing the tangent to a given curve at a given point and (2) the problem of finding a precise definition for the velocity in an arbitrary motion. 1. The Derivative and the Tangent. We shall first take up the tangent problem. If P is a point on a given curve (see fig. 7), we shall, in conformity with naive intuition, define the tangent to the curve at the point P by means of the following geometrical limiting process. In addition to the point P, we consider a second point P x on the curve. Through the two points P, P x we draw a straight line, a secant of the curve. If we now let the point P ± move along the curve towards the point P, this secant will tend to a limiting position which is independent of the side from which P 1 approaches P. This II] THE DERIVATIVE 89 limiting position of the secant is the tangent, and the statement that such a limiting position of the secant exists is equivalent to the assumption that the curve has a definite tangent or a definite direction at the point P. (We have used the word " as- sumption " because we have actually made one. The hypothesis that the tangent exists is valid for most simple curves, but is by no means true for all curves, or even for all continuous curves.) Once we have represented our curve by means of a function y=f(%) the problem arises of representing our geometrical limiting process analytically, using the function f(x). We take the angle which a straight line I makes with the sc-axis as being the angle through which the positive a>axis must be turned in the positive direction* in order to become for the first time parallel to the line I. Let % be the angle which the secant PP X forms with the positive x-axis (cf. fig. 7) and a the angle which the tangent forms with the positive aj-axis. Then if we disregard the case of a perpendicular tangent we obviously have lim cij = a, where the meaning of the symbols is perfectly clear. If x, y(=f(x)) and a^, 2/i(=/( a; i)) are the co-ordinates of the points P and P 2 respectively, we immediately havef _ vi — y _ /fa) —/(*) . 9 and thus our limiting process is represented by the equation 1- /fa)-/(z) X lim J -±-±± — ±±-! = tan a. 6* ■^y-f(x) p/a ■^\<ti ¥ Fig. 7. — Chord and Tangent tan aj : *i" X, X * That is, in such a direction that a rotation of ir/2 brings it into coincidence with the positive y-axis; in other words, counter-clockwise. t In order that this equation may have a meaning, we must assume that < I x — x y I < 8, 8 being chosen sufficiently small. In what follows, corre- sponding assumptions will often be made tacitly in the steps leading up to limiting processes. 90 FUNDAMENTAL IDEAS [Chap. The expression /fo)— /(») ^ th — V = ty a^ — x Xy — x Ax we call the difference quotient of the function y = f(x), since the symbols Ay and Aa; denote the differences of the function y=f(x) and of the independent variable x. (Here, as on p. 79, the symbol A is an abbreviation for difference, and is not a factor.) The tangent of a, the direction angle of the curve,* is therefore equal to the limit to which the difference quotient of our function tends when a^ tends to x. We call this limit the derivative \ of the function y=f(x) at the point x and, as Lagrange did, use the symbol y' =/'(#) to denote it, or, as Leibnitz did, the symbol t -^ or -^i-J or — f(x). dx dx dx On p. 100 we shall discuss the meaning of Leibnitz's notation in more detail; here we would point out that the notation f'(x) expresses the fact that the derivative is itself a function of x, since it has a definite value for each value of a; in the interval which we are considering. This fact is sometimes emphasized by the use of the terms derived function, derived curve (see p. 99). We again quote the definition of the derivative: *i — ► * X-y — x or f=& =f{x)= lj m /(*l)-/W = Um ^ dx dx *,->* x^ — x a*-=>-oAa; - lim /(»+*)-/(*) > *->o h where in the last expression we have replaced Xj by x + h. It is impossible to find the derivative merely by putting a^ = x in the expression for the difference quotient, for then the numerator and denominator would both be equal to and we * The slope or gradient of the curve is given by tana, and hence the term gradient is occasionally used for the derivative of the function represented by the curve. t The term differential coefficient is also used, particularly in the older text- books. J Cauchy's notation Df(x) is also occasionally found in the literature. II] THE DERIVATIVE 91 should be led to the meaningless expression 0/0. On the con- trary, the actual performance of the passage to the limit in each individual case depends upon certain preliminary steps (trans- formation of the difference qaotient). For example, for the function f(x) = x s we have * J^-f- X. X^ — X Xy — X This function x 1 + x is not the same function as Xl ~ — , for the function Kj — x x t + x is defined at one point where the quotient Xl ~ x is undefined, x 1 — x namely, the point x 1 = x. For all other values of x 1 the two functions are equal to one another; hence in the above passage to the limit, in which we specifically required that x, =|= x, we obtain the same value for lim as for lim fe + x). But since the function x 1 + x is de- *,-».* X 1 — X x l -)-x fined and continuous at the point x 1 = x,we can do with it what we could not do with the quotient, namely, pass to the limit by simply putting x 1 = x. For the derivative we then obtain the expression ax The carrying out of such a process, i.e. the actual formation of the derivative, is called the differentiation of the function f(x). We shall see later how this process of differentiation can actually be carried out in all important cases. Now the fact that the problem of differentiating a given function has a definite meaning apart from the geometrical intuition of the tangent is of great significance. The reader will recall that in the case of the integral we freed ourselves from the geometrical intuition of area, and on the contrary based the notion of area on the definition of the integral. Now, inde- pendently of the geometrical representation of a function V=f( x ) by means of a curve, we shall define the derivative of the function y =f(x) as being the new function y' —f'(x) given by the equation above, provided always that the limit of the difference quotient exists. If this limit exists we say that the function f(x) is differentiable. From now on we shall assume that every function dealt with is differentiable unless specific men- * Cf. p. 89, second footnote. 02 .FUNDAMENTAL IDEAS [Chap. fcion is made to the contrary.* It should be observed that if the function f(x) is to be differentiable at the point x the limit as h -> of the quotient l~ must exist independently of the manner in which h tends to 0, whether it be through positive values or through negative values or without restriction as to sign. Once we have found the derivative/'^), we take the direction which makes an angle a with the positive a-axis given by the equation tana =f'(x) as the direction of the tangent to the curve at the point (x, y). We thus avoid the difficulties which arise out of the indefiniteness of the geometrical view, since we ^ Fig. 8. — Tangents to graphs of increasing and decreasing functions base the geometrical definition on the analytical and not vice versa. Nevertheless, the visualization of the derivative as the tangent to the curve is an important aid to understanding, even in purely analytical discussions. Accordingly we shall at once accept the following statement based on geometrical intuition: If f'(x) is positive and the curve is traversed in the direction of increasing x, then the tangent slants upwards, and therefore at the point in question the curve rises as x increases; if, on the other hand, f'(x) is negative, the tangent slants downwards and the curve falls as x increases (see fig. 8). Analytically this follows from the fix -\- h) fix) remark that the limit of ' ^— cannot be positive unless the function is increasing at the point x, by which we * Examples of cases In which this assumption is not satisfied will be given later (see p. 97). n J THE DERIVATIVE 93 mean that for all values of h sufficiently close to the value of f(x + h) is greater or smaller than/(x) according as h is positive or negative. We can, of course, make a corresponding statement for the case where f'(x) is negative. 2. The Derivative as a Velocity. Just as naive intuition led us to the notion of the direction of the tangent to a curve, so it causes us to assign a velocity to a motion. The definition of velocity leads us once again to exactly the same limiting process as we have already called differentia- tion. Let us consider, for example, the motion of a point on a straight line, the position of the point being determined by a single co-ordinate y. This co-ordinate y is the distance, with its proper sign, of our moving point from a fixed point on the line. The motion is given if we know y as a function of the time t, y =f(t). If this function is a linear function f(t) = ct -f b, we call the motion a uniform motion with the velocity c, and for every pair of values t and ^ which are not equal to one another we can write „- /fc)-/(Q h-t The velocity is therefore the difference quotient of the function ct+b, and this difference quotient is completely independent of the particular pair of instants which we fix upon. But what are we to understand by the velocity of motion at an instant t if the motion is no longer uniform? In order to arrive at this definition we consider the difference quotient ■'MizM, wn ich we shall call the average velocity in the time interval between ^ and t. If now this average velocity tends to a definite limit when we let the instant ^ come closer and closer to t, we shall naturally define this limit as the velocity at the time *. In other words: the velocity at the time t is the deri- vative «,-»-« ty t From this new meaning of the derivative, which in itself has nothing to do with the tangent problem, we see that it really is 94 FUNDAMENTAL IDEAS [Chap. appropriate to define the limiting process of differentiation as a purely analytical operation independent of geometrical intuitions. Here again the differentiability of the position-function is an assumption which we shall always tacitly make, and which, in fact, is absolutely necessary if the notion of velocity is to have any meaning. As a simple example of the connexion between motion and velocity we consider the case of a freely falling body. We begin with the experi- mentally established law that the distance traversed in time t by a freely falling body is proportional to t 3 , and therefore can be represented by a function of the form V = /(«) = «*. As on p. 91, we find immediately that the velocity is given by the ex- pression /'(<) = 2at, which shows us that the velocity of a freely falling body increases in proportion to the time. 3. Examples. We now proceed to work out a number of examples of the actual dif- ferentiation of functions. We begin with the function y — f(x) = c, where c is a constant. It is then always true that f(x + A) — f(x) = c — c = 0, fix + h) — fix) so that lira — - = 0; that is, the derivative of a constant is zero. h— >o A For a linear function y = f(x) = ex -f- b, we find that Further, we shall differentiate the function y = /(*) = »". at first assuming that a is a positive integer. Provided x l =t= x, we have /(*i) ~ /(*) = *i a ~ *' . X 1 — X X l — X ' the right-hand side of this equation is equal to as," -1 + a^ a- *a; -{-••• + x a ~ 1 , as we see either by direct division or by using the formula for the sum of a geometric progression. The new expression for the right-hand side of the equation is a continuous function, and so we can carry out the passage to the limit (x 1 — ¥■ x) by simply replacing x 1 everywhere by x. Each term is then a;" -1 , and since the number of terms is exactly a, we obtain II] THE DERIVATIVE 95 We arrive at the same result if a is a negative integer — 0; we must, however, assume that x is not zero. We then find that /to)- *1- -/(*) - X 1 1 x l — X aP— xf X — Xj - 2 *i + ... 1 ' a^a;^ x^xfi Once again we can cany out the passage to the limit simply by replacing x-, everywhere by x. Then just as above we obtain for the limit the expression Hence for negative integral values of a the derivative is again given by / = ax*- 1 . Finally, we shall prove the same formula where * is positive and a. any rational number. We suppose that a. = p/q, where p and q are both in- tegers and, moreover, positive. (If one of them were negative no essential changes in the proof would be needed; for a = the result is already known, since x" is then constant.) We now have /to) ~ /(*) _ «!** - x*H X 1 — X X 1 — X If we now put x 11 ' = I and * 1 1 '» = 5 lf we obtain /to) -/(«) = 5i»- V> = 51*-* + ^p-'g+.-.-f gp-i *i-* 5i«-5« 5i 0_1 + 5i*- a ^ + . - - + 5*- 1 * After this last transformation we can immediately perform the passage to the limit (x t -> x or, what amounts to the same thing, §, -> §), and thus obtain for the limiting value the expression or finally ? 5*" 3 2 ? which is formally the same result as before. We leave it for the reader to prove for himself that the same differentiation formula holds for negative rational indices also. We shall come back (p. 130) to the differentiation of powers once we have developed the theory in a more connected form. As a further example we finally consider the differentiation of the 96 FUNDAMENTAL IDEAS [Chap. trigonometric functions, sina; and cosa:. We use the elementary trigono- metrical formula ain(a: + a) — sina ; sina; cosh + cosa; sink — sina; h h cosa — 1 , sinA = sina; 1- cosa; . A A Now from Chapter I, § 7 (pp. 47-48), we know that ,. sinA , ,. cosh — 1 hm — — = 1, lim = 0. »— >-0 A 4— >0 A For the required derivative we thus immediately obtain , dlsinx) y = — ^ - = cosx. dx The function y = cosa: can be differentiated in exactly the same way. Starting with cos(a; -f A) — cosa; cosa — 1 . sinA : — — — = cosa; sina; , and taking the limit as a -*■ 0, we at once obtain the derivative , d(cosa;) y = — ' = —sina;. dx 4. Some Fundamental Rules for Differentiation. Just as in the case of the integral, certain simple but funda- mental rules for forming the derivative follow immediately from the definition. If (f> (x)=f(x) + g(x), then <f>' {x) =- /' (x) -)- g' '(x); again, if ip(x) = cf(x) (where c is a constant), then ifi'(x) = cf (x). For we have <j>(x+h)-<f>(x) = f(x+h)-f(x) g(x +h)-g(x) h h ^ h and ^x+JQ-tix) _ J(x+h)-f(x) h ° h ' and our statements follow directly by passage to the limit. According to these rules, for example, the derivative of the function <f>(x) =f(x) -f ox -f b (where a and b are constants) is given by the equation f(s) =/'(*) + a. II] THE DERIVATIVE 97 5. Differentiability and Continuity of Functions. It is useful to note that if we know that a function can be differentiated we need not give any special proof of its continuity. If a function is differentiable, then it is necessarily continuous. For if the difference quotient -^^ ~^~ '~J^ X ' approaches a h definite limit as h tends to zero, the numerator of the fraction, that is, f(x + h) — f(x), must tend to zero with h; and this fact expresses the continuity of the function f(x) at the point x. The converse of this, however, is absolutely false; it is not true that every con- tinuous function has a de- rivative at every point. The simplest example disprov- ing this assumption is the function /(x) = | x \, i.e. f{x) ^—xioxx-g.0 and f{x) = x for x ^ 0; its graph is shown in fig. 9. At the point x = this function is continuous, but has no derivative. The limit of f(x -\- h) fix) 7 — — is equal to 1 if A tends to through positive values, and is equal to —1 if A tends to zero through negative values; if we do not restrict the sign of ft, no limit exists. We say that our function has different right-hand and left-hand derivatives at the point x, where by right-hand derivative and left-hand derivative we mean respectively the limiting values of f(x+h)—f(x) , , „ , y as ft approaches through positive values only and negative values only. The differentiability of a function thus requires not merely that the right-hand and left-hand derivatives exist, but that they are equal. Geometrically the inequality of the two derivatives means that the curve has a sharp corner. As further examples of points where a continuous function is not differentiable we consider the points where the derivative becomes infinite, i.e. the points at which there exists neither a 9 8 FUNDAMENTAL IDEAS [Chap. right-hand nor a left-hand derivative, the difference quotient j — ^-^ increasing beyond all bounds as h -*■ 0. For h example, the function y = f(x) = \/x = x* is denned and con- tinuous for all values of x. For all non-zero values of x its deri- vative is given (p. 95) by the formula y 1 = 5 ar*. At the point x = we have • * - — j — ^— - = — = A - *, and we see at once h A that as h -»■ the expression has no limiting value, but, on the contrary, tends to oo. This state of affairs is often briefly de- scribed by saying that the function possesses an infinite deri- Fig. io Fig. II vative, or the derivative oo, at the point in question; we should remember, however, that this merely means that as h tends to the difference quotient increases beyond all bounds, and that the derivative in the sense in which we have denned it really does not exist. The geometrical meaning of an infinite derivative is that the tangent to the curve is vertical (cf. fig. 10). The function y = f(x) = -y/x, which is defined and con- tinuous for x 2^ 0, is also non-differentiable at the point x = 0. Since y is undefined for negative values of x, we here consider the right-hand derivative only. The equation •LLi — ■LL-l = h \/h shows us that this derivative is infinite; the curve touches the y-axis at the origin (fig. 11). Finally, in the function y = y/x 2 = a;* we have a case in which the right-hand derivative at the point x = is positive II] THE DERIVATIVE 99 and infinite, while the left-hand derivative is negative and infinite, as follows from the relation h fyh' As a matter of fact, the continuous curve y = a;*, the so-called semi-cubical parabola or Neil's para- bola, has at the origin a cusp perpendicular to the x-axis (cf . fig. 12). Fig _ lfc _ Coip 6. Higher Derivatives and their Significance. The derivative f'(x) of a function is itself a function of x, the graph of which we call the derived curve of the given curve. For example, the derived curve of the parabola y = jc 2 is a Fig. 13. — Derived curves of sin* and Cos* straight line, represented by the function y = 2x. The derived curve of the sine curve y = sina; is the cosine curve y = cosa;; similarly, the derived curve of the curve y = cosa; is the curve y = —sina;. (Any of these latter curves can be obtained from the others by translation in the direction of the a:-axis, as is shown in fig. 13.) loo FUNDAMENTAL IDEAS [Chap. It is now quite a natural step to form the derived curves of the derived curves, i.e. to form the derivative of the function f'(x) = <f>(x). This derivative m=^ f{x+h) ~ f ' {x \ A-»-0 h provided that it really exists, we shall call the second derivative of the function f(x), and we shall denote it by f"(x). Similarly, we may attempt to form the derivative of f"(x), the so-called third derivative of f(x), which we then denote by f'"(x). In the case of most functions of importance there is noth- ing to hinder us from imagining this process repeated as many times as we like, and from thus defining an n-th derivative f^"\x). Occasionally it will be convenient to call the function f(x) its own 0-th derivative.* If the independent variable is interpreted as the time t and the motion of a point is represented by means of the function f(t), the physical meaning of the second derivative is found to be the velocity with which the velocity f'(t) changes, or, as it is usually called, the acceleration. Later (pp. 158-159) we shall discuss the geometrical interpretation of the second derivative in detail. Here, however, we may note the following facts: at a point where f"{x) is positive, f'(x) increases as x increases; if, on the other hand, f"(x) is negative, f'(x) decreases as x increases. 7. The Derivative and the Difference Quotient. The fact that in the limiting process which defines the deri- vative the difference Ax tends to is sometimes expressed by saying that the quantity Ax becomes infinitely small. This expres- sion indicates that the passage to the limit is regarded as a pro- cess during which the quantity Aa; is never zero, yet approaches zero as closely as we please. In Leibnitz's notation the passage to the limit in the process of differentiation is symbolically ex- pressed by replacing the symbol A by the symbol d, so that we can define Leibnitz's symbol for the derivative by the equation ^=lim^. dx A*-»-oAa; * The terms second, third, . . . , n-th differential coefficient are also used; of. second footnote, p. 90. II] THE DERIVATIVE IOI If, however, we wish to obtain a clear grasp of the meaning of the differential calculus we must beware of regarding the deri- vative as the quotient of two quantities which are actually " infinitely small ". The difference quotient p. absolutely must be formed with differences Aa; which are not equal to 0. After the forming of this difference quotient we must imagine the passage to the limit carried out by means of a transformation or some other device. We have no right to suppose that first Ax goes through something like a limiting process and reaches a value which is infinitesimally small but still not 0, so that Ax and Ay are replaced by "infinitely small quantities" or "in- finitesimals " dx and dy, and that the quotient of these quanti- ties is then formed. Such a conception of the derivative is in- compatible with the clarity of ideas demanded in mathematics; in fact, it is entirely meaningless. For a great many simple- minded people it undoubtedly has a certain charm, the charm of mystery which is always associated with the word " infinite "; and in the early days of the differential calculus even Leibnitz himself was capable of combining these vague mystical ideas with a thoroughly clear understanding of the limiting process. It is true that this fog which hung round the foundations of the new science did not prevent Leibnitz or his great successors from finding the right path. But this does not release us from the duty of avoiding every such hazy idea in our building-up of the differential and integral calculus. The notation of Leibnitz, however, is not merely attractive in itself, but is actually of great flexibility and the utmost usefulness. The reason is that in many calculations and formal transformations we can deal with the symbols dy and dx in exactly the same way as if they were ordinary numbers. They enable us to give neater expression to many calculations which can be carried out without their use. In the following pages we shall see this fact verified over and over again, and shall find ourselves justified in making free and repeated use of it, provided we do not lose sight of the symbolical character of the signs dy and dx. For the second and higher derivatives also Leibnitz has devised a notation of great suggestiveness and practical utility. He thinks of the second derivative as the limit of the " second toa FUNDAMENTAL IDEAS TChap. difference quotient " in the following manner. In addition to the variable x we consider a^ = x -\- h and x 2 — x + 2fe. We then take the second difference quotient as meaning the first difference quotient of the first difference quotient, i.e. the ex- pression where y=f(%), y\=f{%-i), and y 2 =f(x 2 ). If we also write h = Aa; and y 2 — y t = Ay lt y 1 — y= Ay, we may appropriately call the expression in the last bracket the difference of the differ- ence of y or the second difference of y and write symbolically * V% — 2 Vi + V = A Vi — % = A( A 30 = Afy In this symbolic notation the second difference quotient is then A 2 w written — — , where the denominator is really the square of Ax, (Aa;) 2 while in the numerator the numbei 2 symbolically denotes the repetition of the difference process. This symbolism for the difference quotient f led Leibnitz to introduce the notation y ' ,==/ " (a!)= S' y, "= r{x)== % &o -' for the second and higher derivatives, and we shall find that this notation also stands the test of use. 8. The Mean Value Theorem. Between the derivative -^ =f'(x) and the difference quotient ax there exists a simple relation which is important for many purposes. This relation is known as the mean value tficorem, * Here AA ~ A* is not a square, but merely a symbol for " difference of difference " or " seoond difference ". f We must emphasize that the statement that the second derivative may be represented as the limit of the second difference quotient requires proof. For we previously defined the seoond derivative, not in this way, but as the limit of the first difference quotient of the first derivative. In actual fact, the two definitions are equivalent, provided the second derivative is continuous; the proof, however, is not given, as we have no particular need of it here. II] THE DERIVATIVE 103 Fig. 14. — To illustrate the mean value theorem and is obtained in the following way. We consider the difference quotient /fa) -/fa) _ A/ x 1 — x 2 Ax of a function f{x), and assume that the derivative exists every- where in the interval x 1 g x ^ x 2 , so that the graph of the curve has a tangent everywhere. The difference quotient will be re- presented by the direction of the secant (see fig. 14); it is, in fact, the tangent of the angle a shown in the figure. Let us imagine this secant shifted parallel to itself. At least once it will reach a position in which it is a tangent to the curve at a point between x 1 and Og, namely, at the point of the curve which is at the greatest distance from the secant. Hence there will be an inter- mediate value £ such that /fa) ~/fa) ==_/•'(£). x^ x 2 This statement is called the mean value theorem of the differential calculus. We can also express it somewhat differently by noticing that the number £ may be written in the form i = x 1 + d(x 2 — x 1 ), where is a certain number between and 1. In applications of the mean value theorem we shall often find that 6 cannot be more accurately determined than this, but it will usually turn out that a more accurate value is not needed. When accurately formulated, then, the mean value theorem runs as follows: If f(x) is continuous in the closed interval x x <i x gj x 2 and differentidble at every point of the open interval x x <x <x 2 , then there is at least one value 8, where <. 6 < 1, such that x » — "i 104 FUNDAMENTAL IDEAS [Chap. If we replace Xj by x and x 2 by x + h, we can express the mean value theorem by the formula f(x+h)-f(x}_ h = /'(!) =/'(*+«*), s<£<z+ft. Fig. 15. — To illustrate the mean value theorem We wish to emphasize that while it is essential that f(x) should be continuous for all points of the interval, including the end-points, we need not assume that the derivative exists at the end-points. This apparently trivial remark is actually useful in many applications. If at any point in the interior of the interval the derivative fails to exist, the mean value theorem is not necessarily true. This is shown by the example f(x) = \x\, p. 97. We can complete our intuitive argument by the following con- siderations. There is at least one point P on the curve which has the greatest possible distance from the chord joining the points on the curve whose abscissas are x 1 and x z (see fig. 15). At this point the curve by hypothesis has a definite tangent. We shall now prove that this tangent must be parallel to the chord. By definition the tangent is the limiting position of the secant and is obtained by joining P to a point Q on the curve and letting the point Q move towards P. Since by hypothesis Q is not farther from the chord than P, the line PQ produced in the direction P to Q must either out the chord or run parallel to it; and this must be the case, no matter on which side of P the point Q lies. This, however, is only possible if the limiting position is parallel to the chord. If we denote the abscissa of the point P by the letter £, the slope /'(£) of the tangent at P is then tt„ \ f/ x \ equal to the slope of the chord, -Li-Si — JK 2I - hence for the number a^ x 2 I in the theorem we may simply take the abscissa of P. The rigorous proof of the mean value theorem is usually developed in the following way. We first establish Rolle's theorem, which is a special case of the mean value theorem: If a function ^(x) is continuous in the closed interval Xj ^ x ^ x 2 II] THE DERIVATIVE io 5 and differentiable in the open interval x 1 <x<x 2 , and if in addition ^(x,) =- and <f>(x 2 ) = 0, then there exists at least one point £ in the interior of the interval at which <£'(£) = 0. In fact, there must be at least one point £ interior to the interval, at which the function $(x) takes on its greatest or its least value (cf. Chap. I, Appendix I, § 2, p. 63); to be specific, we assume that g is a point where </>(£) is a maximum, so that for every x in the interval <f>(x) <; <£(£). Then for every number h whose absolute value | h | is small enough it is certainly true that <f>(£) — <f>($ + h)^0. If A is positive, h we now let h tend to zero through positive values and obtain $'(£) ^ 0. If, on the other hand, h is negative, t 2S 0, and thus by letting h tend to zero through negative values we obtain <£'(£) ^ 0; comparing this with the preceding inequality, we see that <j>'{g) = 0, which establishes our theorem. We now apply Rolle's theorem to the function * m =m -fix,) - ^1 {f(x 2 ) -f (Xl )}. This function obviously satisfies the condition <j>{xj = ^(a^) = 0, and is of the form <£ (x) = f(x) + ax + b with constant coefficients a = _ /fa) -/fa) and 6 By p 96) we know that <f>'(x) =/'(«) + a, and thus by Rolle's theorem we have 0= ?(€)=*/'(£) + a for a suitably chosen intermediate value £ hence /'(^-^/faO-Zfa), a^ x, and the mean value theorem is proved. • * T^" 8 ,*" 111011011 " a I»rt from a factor independent of *, is the distance of the point (x, /(a:)) of the curve from the secant; the reader can easily verify this for himself. * io6 FUNDAMENTAL IDEAS [Chap. As the first of many applications of the mean value theorem we shall prove the foEowing. Let the function f(x) be continuous in the closed interval a ^ x ^ b and have a derivative f '(x) at every point of the open interval a < x < b. Then if f'(x) is positive everywhere in a < x < b, the function f(x) m monotonia increasing in the interval a^ x5S b; and likeunse if f'(x) w negative in a < x < b, then f(x) is monotonia decreasing. We shall prove the first statement; the second can be proved in a similar way. Suppose that f'(x) > 0, and let x t and x 2 > x^ be any two values of a; in the closed interval. Then by the mean value theorem /ta>-/(*i)-(*-*i)/m where x x <\ < x 2 . Since both factors on the right are positive, this proves that f(x t ) > /(a^); hence f(x) is monotonic increasing. 9. The Approximate Representation of Arbitrary Functions by Linear Functions. Differentials. fix -+- h) f (x) The equation Yao. JK ' — =^— =/'( a; ) defining the deri- vative is equivalent to the equations Xx+h)-m=hf(x) + ch or y + Ay = fix + Ax) = /(*) + f'(x) Ace + e Aa;, where e is a quantity which tends to zero with h = Ax. If for the moment we think of the point x as fixed and the incre- ment Ace as variable, then by this formula the increment of the function, that is, the quantity Ay, consists of two terms, namely, a part hf'{x) which is proportional to h, and an " error " which can be made as small as we please relative to h by making A itself small enough. Thus the smaller the interval about the point x which we consider, the more accurately is the function f[ x _|_ %} (which is a function of h) represented by its linear part f(x) + hf'{x). This approximate representation of the function f(x -f- h) by a linear function of h is expressed geometrically by the substitution for the curve of its tangent at the point x. Later, in Chapter VII, we shall consider the practical application of these ideas to the performance of approximate calculations. Here we merely remark in passing that it is possible to use this approximate representation of the increment Ay by the linear expression hf'(x) to construct a logically satisfactory definition of the notion of a " differential ", as was done by Cauchy in particular. II] THE DERIVATIVE 107 While the idea of the differential as an infinitely small quan- tity has no meaning, and it is accordingly futile to define the derivative as the quotient of two such quantities, we may still try to assign a sense to the equation /'(a;) = dyjdx in such a way that the expression dyjdx need not be thought of as purely symbolic, but as the actual quotient of two quantities dy and dx. For this purpose we first define the derivative f\x) by our limit- ing process, then think of a; as fixed and consider the increment h =s Aa: as the independent variable. This quantity h we call the differential of x, and write h — dx. We now define the ex- pression dy—y'dx — hf'(x) as the differential of the function y; dy is therefore a number which has nothing to do with infinitely small quantities. So the derivative «j l/ =/'(a ; ) is now really the quotient of the differentials dy and dx; but in this statement there is nothing remarkable; it is, in fact, merely a tautology, a restatement of the verbal definition. The differential dy is accordingly the linear part of the increment Ay (see fig. 16). We shall not make any im- mediate use of these differentials. Nevertheless, it may be pointed out for the sake of completeness that we may also form second and higher differentials. For if we think of h as chosen in any manner, but always the same for every value of x, then dy — hf'(x) is a function of x, of which we can again form the differential. The result will be called the second differential of y, and will be denoted by the symbol dh] = dJf(x). The increment of hf'{x) being h{f'(x + h) —f'(x)}, the second differential is obtained by replacing the quantity in brackets by its linear part hf"(x), so that d?y = hj"(x). We may naturally proceed further along the same lines, obtaining third, fourth, — differentials of y, &c, which can be defined by the expressions Wf'"{x), Wf iv (x), and so on. 10. Remarks on Applications to the Natural Sciences. In the applications of mathematics to natural phenomena we never have to deal with sharply defined quantities. Whether a length is exactly a metre is a question which cannot be decided x x+h Fig. 16. — The differential dy 108 FUNDAMENTAL IDEAS [Chap. by any experiment and which consequently has no "physical meaning ". Again, there is no immediate physical meaning in say- ing that the length of a material rod is rational or irrational; we can always measure it with any desired degree of accuracy in rational numbers, and the real matter of interest is whether or not we can manage to perform such a measurement using rational numbers with relatively small denominators. Just as the ques- tion of rationality or irrationality in the rigorous sense of " exact mathematics " has no physical meaning, so the actual carrying out of limiting processes in applications will usually be nothing more than a mathematical idealization. The practical significance of such idealizations lies chiefly in the fact that if they are used all analytical expressions become essentially simpler and more manageable. For example, it is vastly simpler and more convenient to work with the notion of instantaneous velocity, which is a function of only one definite time-instant, than with the notion of average velocity between two different instants. Without such idealiza- tion every rational investigation of nature would be condemned to hopeless complications and would break down at the very outset. We do not intend, however, to enter into a discussion of the relationship of mathematics to reality. We merely wish to emphasize, for the sake of our better understanding of the theory, that in applications we have the right to replace a derivative by a difference quotient and vice versa, provided only that the differences are small enough to guarantee a sufficiently close approximation. The physicist, the biologist, the engineer, or anyone else who has to deal with these ideas in practice, will therefore have the right to identify the difference quotient with the derivative within his limits of accuracy. The smaller the increment h = dx of the independent variable, the more accurately can he represent the increment Ay =f(x + h) —f(x) by the differential dy = hf'(x). So long as he keeps within the limits of accuracy required by the problem, he is accustomed to speak of the quantities dx=h and dy = hf'(x) as " infinitesi- mals ". These " physically infinitesimal " quantities have a precise meaning. They are finite quantities, not equal to zero, which are chosen small enough for the given investigation, e.g. smaller than a fractional part of a wave-length or smaller II] THE DERIVATIVE 109 than the distance between two electrons in an atom; in general, smaller than the degree of accuracy required. Examples 1.* Replace the statement " At the point x = Z, the function f(x) is not differentiable " by an equivalent statement not using any form of the word "differentiable". 2. Differentiate the following functions directly by using the de- finition of the derivative: ( '».TT m +T* W^TT W^-,. (e) sin 3a;. (/) cosaz. (0) sin 2 a;. (A) cos 2 *. 3. Find the intermediate value £ of the mean value theorem for the following functions, and illustrate graphically: (a) 2x. (6) *». (c) 5a? + 2x. (d) l/(z*+ 1). (e) x 1 ' 3 . 4. Show that the mean value theorem fails for the following functions when the two points are taken with opposite signs, e.g. x 1 = —I, x =1: (a) l/x. (b) |*|. (c) ar" 3 . Illustrate graphically, and compare with the previous example. 4. The Indefinite Integral, the Peimitive Function, and the Fundamental Theorems of the Differential and Integral Calculus. As we have already mentioned above, the connexion between the problem of integration and the problem of differentiation is the corner-stone of the differential and integral calculus. This connexion we will now study. 1. The Integral as a Function of the Upper Limit. The value of the definite integral of a function /(a;) depends on the choice of the two limits of integration a and b. It is a func- tion of the lower limit a as well as of the upper limi t, b. In order to study this dependence more closely we imagine the lower limit a to be a definite fixed number, denote the variable of in- tegration no longer by x but by u (cf. p. 82), and denote the upper limit by x instead of b in order to suggest that we shall let the no FUNDAMENTAL IDEAS [Chap. upper limit vary and that we wish to investigate the value of the integral as a function of the upper limit. Accordingly, we write f*f(u)du = <!>(x). V ^%' m l/////////////, Fig. 17 We call this function <3>(x) an indefinite integral of the function f(x). When we speak of an and not of the indefinite integral, we suggest that instead of the lower limit a any other could be chosen, in which case we should ordi- narily obtain a different value for the integral. Geometrically the indefinite integral for each value of x will be given by the H area (shown by shading in fig. 17) under the curve y=f{u) and bounded by the ordinates u = a and u = x, the sign being determined by rules given earlier (p. 81). If we choose another lower limit a in place of the lower limit a, we obtain the indefinite integral T(x)=£f(u)du. The difference T(a;) — Q>(x) will obviously be given by J f(u)du, which is a constant, since a and a are each taken as fixed given numbers. Therefore y F(x) = 0(x) + const.; Different indefinite integrals of the same function differ only by an additive constant. We may likewise regard the integral as a function of the lower limit, and introduce the function <t>{x) = ff(u)du, in which b is a fixed number. Here again two such in- II] THE INDEFINITE INTEGRAL m tegrals with different upper limits b and j8 differ only by an additive constant / f(u)du. 2. The Derivative of the Indefinite Integral. We will now differentiate the indefinite integral <&{x) with respect to the variable x. The result is the following theorem: The indefinite integral ^(x) = ff{u)du J a of a continuous function f(x) always possesses a derivative <D'(x), and, moreover, thai is, differentiation of the indefinite integral of a given con- tinuous function always gives us back that same function. a xx, XgX+h. x Fig. 1 8. — Differentiation of the indefinite integral This is the root idea of the whole of the differential and integral calculus. The proof follows extremely simply from the inter- pretation of the integral as an area. We form the difference quotient Q(x+h) — Q(x) h and observe that the numerator <S>(x+h) - «(x) =f* +h f{u)du-£f(u)du=f +h f{u)du is the area between the ordinate corresponding to x and the ordinate corresponding to x + h. Now let x be a point in the interval between x and x + h at which the function f(x) takes its greatest value, and x 1 a point at which it takes its least value in that interval (cf. fig. 18). ri 2 FUNDAMENTAL IDEAS [Chap. Then the area in question will lie between the values hf(x ) and hfixj), which represent the areas of rectangles with the interval from xtoic + Aas base and/(a: ) and/fo) respectively as alti- tudes. Expressed analytically, / W £ »<« + *>-» ( 'W (»j. This can also be proved directly from the definition of the in- tegral without appealing to the geometrical interpretation.* To do this we write x+h « /(«) du = lim 2/(w„) Aw„, r where u = x, u v u 2 , . . . , u„ = x + h are points of division of the interval from x to x -f h, and the greatest of the absolute values of the differences Aw„ = w„ — u v _ x tends to zero as n increases. Then A.u v /h is certainly positive, no matter whether h is positive or negative. Since we know that/(a; ) ^/(m„) ^ /(%) and since the sum of the quantities Am„ is equal to h, it follows that and thus if we let n tend to infinity we obtain the inequalities stated above for r x+h ,, v , ®(x+h) — <X>(x) 1 f '" A iu or — - 1 . * f f(u)dv If A now tends to zero, both f(x ) and f{xj must tend to the limit f(x), owing to the continuity of the function. We therefore see at once that ... . ,. <S>(x+ h) — ®(x) ,, . «D'(a;) = hm v ^ ' y -L=f{x), h->0 "■ as stated by our theorem. Owing to the differentiability of <&(x), we have the following theorem, by § 3, No. 5, p. 97: The integral of a continuous function f(x) is itself a continuous function of the upper limit. * Compare also the later discussion on p. 127. II] THE INDEFINITE INTEGRAL 113 For the sake of completeness we would point out that if we regard the definite integral not as a function of its upper limit but as a function of its lower limit, the derivative is not equal to f(x), but is instead equal to —f(x). In symbols: if we put H x )=\ f(u)du, then <j>'(x)= — f(x). The proof follows immediately from the remark that ff(u)du = — lf(u)du. 3. The Primitive Function; General Definition of the Indefinite Integral. The theorem which we have just proved shows us that the indefinite integral O(x) at once gives the solution of the follow- ing problem: given a function f (x), to determine a junction F(x) such that F'(x)=f(x). This problem requires us to reverse the process of differentiation. It is a typical inverse problem such as occurs in many parts of mathematics and such as we have already found to be a fruitful mathematical method for generating new functions. (For ex- ample, the first extension of the idea of natural numbers was made under the pressure of the necessity for reversing certain elementary processes of calculation. The formation of inverse functions has led and will lead us to new kinds of functions.) A function F(x) such that F'(x) =f{x) is called a primitive function of f(x), or simply a primitive of f(x); this terminology suggests that the function /(a;) arises from F(x) by differentiation. This problem of the inversion of differentiation or of the finding of a primitive function is at first sight of quite a different character from the problem of integration- From p. Ill, however, we know that: Every indefinite integral O(x) of the function f(x) is a primitive o/f(x). Yet this result does not completely solve the problem of find- ing primitive functions. For we do not yet know if we have found all the solutions of the problem. The question about the s (E798) "4 FUNDAMENTAL IDEAS [Chap. group of all primitive functions is answered by the following theorem, sometimes referred to as the fundamental theorem of the differential and integral calculus: The difference of two primitives F x (x) and F 2 (x) of the same func- tion f (x) is always a constant: F^x) - F 2 {x) = c. Thus from any one primitive function ~E(x) we can obtain all the others in the form F(x) + c by suitable choice of the constant c. Conversely, for every value of the constant o the expression F x (x) = F(x) + c represents a primi- tive function ofi(x). It is clear that for any value of the constant c the function F(x) + c is a primitive, provided that F(x) itself is one. For (cf. p. 96) we have { F(x + h) + c> - {F{x) + c} _ F(x +h)- F(x) h h and since by hypothesis the right-hand side tends to/(a;) as h -*■ 0, so does the left-hand side, and therefore l{F(x) + c}=f(x) = F'(x). Thus to complete the proof of our theorem it only remains to show that the difference of two primitive functions F x (x) and F 2 (x) is always a constant. For this purpose we consider the difference FM-FJM^Glx) and form the derivative G'(x) = lim f *i(g +*)-*!(*) _ F 2 (x + h)-F 2 (x) \ h-+o\ h h y Both the expressions on the right-hand side, by hypothesis, have the same limit /(x) as h -> 0; thus for every value of x we have G'(x) = 0. But a function whose derivative is everywhere zero must have a graph whose tangent is everywhere parallel to the x-axis, i.e. must be a constant; and therefore we have G(x) = c, II] THE INDEFINITE INTEGRAL 115 as we stated above. We can prove this last fact without relying upon intuition, by using the mean value theorem. Applying the mean value theorem to G(x), we have G(x 2 ) - CHxJ = (x 2 - xJG'd); x 1 <$<x a . But we have seen that the derivative G'(x) is equal to for every value of x, and hence in particular for the value £; hence it follows immediately that G(x 1 ) = G(x 2 ). Since x l and x 3 are arbitrary values of a; in the given interval, G(x) must be a constant. Combining the theorem just proved with the result of No. 2 (p. Ill), we can now make the following statement: Every primitive function F(x) of a given function f (x) can be represented in the form F(x) = c + <D (*) = c + £f{u)du, where c and a are constants, and conversely, for any constant values of a and c chosen arbitrarily this expression always represents a primitive function. It may readily be guessed that the constant c can as a rule be omitted, since by changing the lower limit a we change the primitive function by an additive constant. In many cases, however, we should not obtain all the primitive functions if we omitted the c, as the example f(x) = shows. For this function the indefinite integral of No. 1 (p. 110) is always 0, inde- pendently of the lower limit; yet any arbitrary constant is a primitive function of f(x) = 0. A second example is the func- tion f(x) = -y/jc, which is defined for non-negative values of x only. The indefinite integral is ^(x) = fx m -§a sl \ and we see that no matter how we choose the lower limit a the indefinite integral <D(a;) is always obtained from § (a;) 3 ' 2 by addition of a constant which is less than or equal to zero, namely, the con- stant — §a s/2 ; yet such a function as §a; 3/2 + 1 is also a primi- tive function for y/x. Thus in the general expression for the primitive function we cannot dispense with the additive constant. The relationship which we have found suggests an extension of the idea of the indefinite integral. We shall henceforth call every expression of the form c + Q>(x) = c -f- / f(u)du an hide- n6 FUNDAMENTAL IDEAS [Chap. finite integral of f(x). In other words, we shall no longer make any distinction between the primitive function and the indefinite in- tegral. Nevertheless, if the reader is to have a proper under- standing of the interrelations of these concepts it is absolutely necessary that he should clearly bear in mind that in the first instance integration and inversion of differentiation are two entirely different things, and that it is only the knowledge of the relationship between them that gives us the right to apply the term " indefinite integral " to the primitive function also. It is customary to represent the indefinite integral by a notation which in itself is perhaps not perfectly clear. We write F(x) = c + fj(u)du=ff(x)dx; that is, we omit the upper limit x and the lower limit a and also the additive constant c and use the letter x for the variable of integration. It would really be more consistent to avoid this last change, in order to prevent confusion with the upper limit x which is the independent variable in F(x). In using the notation ff(x)dx we must never lose sight of the indeterminacy con- nected with it, i.e. the fact that that symbol always denotes an indefinite integral only. 4. The Use of the Primitive Function in the Evaluation of Definite Integrals. Suppose that we know any one primitive function F(x) = ff(x)dx for the function /(a;) and that we wish to evaluate the definite integral Jf(u)du. "We know that the indefinite integral Ja being also a primitive of f{x), can only differ from F(x) by an additive constant. Consequently <b(z) = F(x) + e, and the additive constant c is at once determined if we recollect rx that the indefinite integral <J>(aj)= / f(u)du must take the value when x = a. We thus obtain = (a) = F(a) + c, II] THE DEFINITE INTEGRAL 117 whence c=—F(a) and O (a;) = F(x) — F(a). In particular, for the value x= b we have f b f(u)du=F(b)-F(a), which gives us the important rule: If F(x) is any -primitive of the function i(x) whatsoever, the definite integral of f (x) between the limits a and b is equal to the difference F(b) — F(a). If we use the relation F'(x) =f{x) this may be written in the form Fib) - F(a) = f F'{x)dx = f* d -^ dx. Ja Ja dx This formula can easily be proved and understood directly. We divide the interval a <S x fS b into sub-intervals Ax 1 , Ax 2 , . . . , Ax n , A EI and consider the sum S — Aa;„. On the one hand, this sum is Ax v simply S AF = F(b) — F(a), independently of the particular subdivision; hence its limit is F(b) — F(a). On the other hand, its limit is also equal to / F'(x) dx, as follows from the mean value theorem. For AF/Ax„ = F'(£ v ), where f„ is a point intermediate between the ends x v _ x and x v of the interval Ax„. The sum is therefore equal to S Aa;„ F'(i y ); and by the definition of integral this tends to the limit / F'(x)dx as the subdivision is made finer, "a which establishes our formula. In applying our rule we often use the symbol | to denote the difference F(b) — F (a); i.e. we write f(x)dx = F(b) - F(a) = F(x) jO meaning by the vertical line that in the preceding expression first the value b, and then the value a, is to be substituted for x, and finally the difference of the resulting numbers is to be found. 5. Examples. We are now in a position to illustrate by a series of simple examples the relationships between the definite integral, the indefinite integral, and the derivative, which we have just in- n8 FUNDAMENTAL IDEAS [Chap. vestigated. In virtue of the theorem on p. Ill, from each of the integration formulae directly proved in § 2 (p. 82) we can derive a differentiation formula. On p. 86 we obtained the integration formula £ x*dx = (6»+ 1 — a** 1 ) a+ 1 for every rational number a 4= 1 and all positive values of a and 6; if we replace the variable of integration by « and the upper limit by x, this may be written £' u" du = (a;^ 1 — a"* 1 ). oc -f~ 1 Prom this it follows by the fundamental theorem that the right-hand side is a primitive function of the integrand, i.e. the differentiation formula d — z«+i = (« + l)a?» ax is valid for every rational value of a 4= — 1 and all positive values of x. By direct substitution we find that this last formula is also true for a = — 1, if x > 0. The result obtained exactly agrees with what we have already found (p. 95) by direct differentiation. Thus by using the fun- damental theorem after we had carried out the integration, we could have saved ourselves the trouble of that differentiation. Further, from the integration formula "a cos u du = sin x — sin a given on p. 87 it follows that — sin x = cos x, in agreement with the result found on p. 96. *" Conversely, however, we may regard every directly-proved differen- tiation formula F'(x) = f(x) as a connexion between a primitive function F(x) and a derived function f(x), that is, we may regard it as a formula for indefinite integration and then obtain from it the definite integral of f(x) as on p. 117. This very method is frequently made use of, as we shall see in Chapter IV (p. 205). In particular, we may start from the results of § 3 (p. 94) and obtain the integral formulae of § 2, p. 82, in virtue of the fundamental theorem. For example, from p. 95 we know that d x*+ 1 — 1'+» = (a + l)a; a . Therefore — — - is a primitive function or indefinite ax a. + 1 integral of x a , provided that a #= — 1, and thus by p. 117 we again arrive at the integral formula above. II] GRAPHICAL INTEGRATION Examples "9 1. From the differentiations performed in Examples 2, 3 on p. 109, set np the corresponding integrations. 2. Evaluate (a) /* ** - (6) /* **** . 3. Using Example 2, prove from the definition of the definite integral that (a) Mm nf - + I + . . . + —1 = -. .-».- L(w + 1)* ^ (n + 2) 2 T T (2»)*J 2 (6) lim n*\ l - + ? + . . . + - 1 = J 5. Simple Methods of Graphical Integration An indefinite integral or primitive function of f(x) is a func- tion y = F{x) which not only can be visualized as an area, but like any other function can be represented graphically by a curve. Our definition immediately suggests the possibility of constructing this curve approxi- mately and thus obtaining a graph of the integral function. To begin with we must remember that this last curve is not unique, but on account of the additive constant can be shifted parallel to itself in the direction of the #-axis. We can therefore require that the in- tegral curve shall pass through an arbitrarily selected point, e.g. if x = 1 belongs to the interval of definition of f(x), through the point with the co-ordinates x = 1, y = 0. The curve is thereafter determined by the requirement that for each value of x its direction is given by the corresponding value of f(x). To obtain an approximate construction of a curve which satisfies these conditions, we seek to construct not the curve y = F(x) itself, but a polygonal path (broken line) whose corners lie vertically above previously as- signed points of division of the z-axis and whose segments have approximately the same direction as the portion of the integral curve between the same points of division. For this purpose we Fig. 19. — Graphical integration 120 FUNDAMENTAL IDEAS [Chap. divide our interval of the 2>axis by means of the points x = 1, x l , aj 2 > • • • m to a certain number of parts, not necessarily all Fig. 20. — Graphical integration of of the same length, and at each point of division we erect the parallel to the y-axis. Then (fig. 19) we draw through the point x = 1, y = the straight line whose slope is equal to /(l); through the intersection of this line with the line x=x 1 we draw the line with the slope /(a^); through the in- tersection of this line with x = a? 2 we draw the line with slope f(x 2 ), and so on. In the actual practical con- struction of these lines, we erect at each point of divi- sion the ordinate to the curve y=f(x), and project these ordinates on to any parallel to the y-axis; to be specific, let us suppose that they are projected on to the y-axis itself. We then obtain the direction of the integral curve by joining the point with co-ordinates x= and y—f(x) to the point x = — 1, y = 0. By carrying over these directions y, y- x / 1 / / 1 1 y 1 1 / "71 ' / 1 1 1 1 1 1 1 1. -/{■ 1 1 1 1 yt ~x 1 / \ V Fig. ai. — Graphical integration of * II] GRAPHICAL INTEGRATION 121 parallel to themselves we obtain a polygonal path whose corners lie vertically above the given points of division of the a;-axis and whose direction agrees with the direction of the integral curve at the initial point of each interval. This polygonal path can be made to represent the integral curve with any desired degree of accuracy by making the subdivision of the interval fine enough. We can frequently improve the accuracy of the construction by choosing for the direction of each segment of the polygon that direction which belongs not to the beginning but to the mid-point of the corresponding interval (cf. figs. 20 and 21).* In fig. 21 the construction described above is carried out for the func- tion /(x) = x. By graphical integration we obtain an approximation to the integral curve, which is the parabola y = $x* — J. In addition, fig. 20 shows an approximation to the integral function of the function /(a;) = 1/x. We shall study this integral later in greater detail — it will turn out to be the logarithmic function. Finally, the reader would be well advised to work out some other examples for himself, e.g. the graphical integration of the functions sin x and cos x. Examples 1. By graphical integration with the interval h = -fe construct the following integral curves: (a) fa 2 dx (0 <^ x <^ 2). (6) J* jj dx (1 <^ x <£ 2). W F T^dx (0^x^,1). ■'0 * ~> * r 1 1 In particular, evaluate / — - — - dx. J n 1 -f- x 6. Further Eemarks on the Connexion between the Integral and the Derivative Before we begin to follow up the relationships found in § 4 (p. 109) systematically we shall illustrate them from another point of view, which is closely related to the intuitive idea of density and to other physical concepts. * We may mention in passing that graphical integration (that is, the find- ing of the graph of a primitive F(x) of a function f(x) which is itself given by a graph) can also be performed by means of a mechanical device, the so-called integraph. In this mechanism a pointer is moved along the given curve and a pen automatically traces one of the curves y = F(x) for which F'(x)=f(x). The indeterminacy of the constant of integration is expressed by a certain arbi- trariness in the initial position of the instrument. For integrating devices generally see B. Williamson, Integral Calculus, pp. 214-217 (Longmans); Dic- tionary of Applied Physics, Vol. Ill, pp. 450^457 (Maemillan, 1923). 6* (1798) i22 FUNDAMENTAL IDEAS [Chap. 1. Mass Distribution and Density; Total Quantity and Specific Quantity. We suppose that any mass whatsoever is distributed along a straight line, the a;-axis, the distribution being continuous but not necessarily uniform. We may, for example, think of a vertical column of air standing on a surface of area 1; as a>axis we take a line pointing vertically upwards and as origin the point on the earth's surface. The total mass between two abscissae x 1 and x 2 is then determined in the following way by means of a so-called sum-function F(x). We measure distance along the sine from the initial point of the mass-distribution, x — 0, and by F(x) we mean the total mass between the abscissa and the abscissa x. The increment of mass from the abscissa Xj to the abscissa x z is then given simply by F(x 2 ) - F(zJ; a sign is thus assigned to the increment, and this sign changes if x 1 and x 2 are interchanged. The average mass per unit length in the interval x 1 to x 2 is F(x 2 )-F(xJ ^ If we assume that the function F(x) is differentiable, then as x 2 -^x 1 this value tends to the derivative F'lxj). This quantity is precisely what is usually called the specific mass or density of the distribution at the point a^; as a rule, of course, its value depends on the particular point chosen. Between the density f(x) and the sum-function F(x) there accordingly exists the relation F(x) =£f(u)du; f(x) = F'(x). The sum-function is a primitive function of the density, or, what amounts to the same thing, the mass is the integral of the density; conversely, the density is the derivative of the sum-function. Exactly the same relation is very frequently encountered in physics. For example, if by Q(t) we denote the total amount of heat needed to raise unit mass of a substance from tempera- II] INTEGRAL AND DERIVATIVE 123 tare t to temperature t, then to raise the temperature from t^ to t 2 an amount of heat equal to is needed. Between ^ and t 2 the average amount of heat used per unit increase in temperature is then If we once again assume the differentiability of the function Q (t) we obtain in the limit a function which we call the specific heat of the substance. This specific heat is in general to be regarded as a function of the tem- perature. Here again, between specific heat and total quantity of heat there exists the characteristic relation of integral and derivative, f b q(t)dt=Q{b)-Q(a). We shall meet with the same relations in all cases where total quantity and specific quantity are contrasted, e.g. electric charge as contrasted with density of charge, or the total force on a surface as contrasted with the force-density or pressure. In nature it usually happens that what we know directly is not the density or specific quantity, but the total quantity; thus it is the integral which is primary (as the name " primitive " suggests) and the specific quantity is only arrived at after a limiting process, namely, differentiation. Incidentally it may be noted that if the masses considered are by their nature positive, the sum-function F(x) must be a monotonic increasing function of x, and consequently the specific quantity, the density /(as), must be non-negative. Nothing hinders us, however, from considering negative quantities also (e.g. negative electricity); then our sum-functions F{x) need no longer be monotonic. i24 FUNDAMENTAL IDEAS [Chap. 2. The Question of Applications. The relation of the primitive sum-function to the density of distribution perhaps becomes clearer when it is realized that from the point of view of physical facts the limiting processes of integration and differentiation represent an idealization, and that they do not express anything exact in nature. On the con- trary, in the realm of physical actuality we can form in place of the integral only a sum of very many small quantities and in place of the derivative only a difference quotient of very small quantities. The quantities Ax remain different from 0; the passage to the limit Ace -> is merely a mathematical simplifi- cation, in which the accuracy of the mathematical representation of the reality is not essentially impaired. As an example we return to the vertical column of air. Ac- cording to the atomic theory we find that we cannot think of the distribution of mass as a continuous function of ce. On the contrary, we will assume (and this, too, is a simplifying ideali- zation) that the mass is distributed along the cc-axis in the form of a large number of point-molecules lying very close to one another. Then the sum-function F(x) will not be a continuous function, but will have a constant value in the interval between two molecules and will take a sudden jump as the variable x passes the point occupied by a molecule. The amount of this jump will be equal to the mass of the molecule, while the average distance between molecules, according to results established in atomic theory, is of the order of 10 -8 cm. If now we are per- forming upon this air column some measurement in which masses of the order 10* molecules are to be considered negligible, our function cannot be distinguished from a continuous function. For if we choose two values x and x + Ax whose difference Ace is less than 10 -4 cm., then the difference between F{x) and F(x + Ace) will be the mass of the molecules in the interval; since the number of these molecules is of the order of 10 4 , the values of F{x) and F{x -f Ax) are, so far as our experiment is concerned, equal. As density of distribution we consider simply ., ,.- ... AF(x) F(x+Ax)-F(x) . x . the difference quotient — = — i — ■ '- ^— '; it is an Asc Ax important physical assumption that we do not obtain measurably different values for this quotient when Ace is allowed to vary II] APPLICATIONS "5 between certain bounds, say between 10~ 4 and 10~ 5 cm. Now let us imagine that F{x) is measured and plotted for a large number of points about 10 -4 cm. apart, and that the points thus found are joined by straight lines; we obtain a polygon, and by rounding off the corners we finally obtain a curve with a continuously turning tangent. This curve is the graph of some function, say F^x). This new function F x (x) cannot within the limits of experimental accuracy be distinguished from F(x), and its derivative is within the same limits equal to AF/Ax; we thus have found a continuous differentiable function which for the purposes of physics is the function F(x). It is perhaps appropriate to discuss yet another example of the concepts of sum-function and density of distribution. In statistics, e.g. in the kinetic theory of matter or in statistical biology, these concepts frequently occur in a form in which the nature of the mathematical idealization is particularly clear. Let us consider e.g. the molecules of a gas confined in a vessel and observe their velocities at a given instant of time. Let the number of molecules be N, and let the number of those with velocities less than x be N<t>(x). Then <J>(z) denotes the ratio of the number of molecules moving with velocities between and a; to the total number of molecules. This sum-function is, of course, not continuous, but is sectionally * constant and suddenly increases by 1/2V when x as it increases passes a value which is equal to the velocity of some molecule. The idealization which we shall make here is that we shall think of the number N as increasing beyond all bounds. We assume that in this passage to the limit N -> 00 the sum-function $(2:) tends to a definite continuous limit function F(x). That this is really the case (i.e. that we can with sufficient accuracy replace <J>(a;) by this continuous function F(x)) is obviously an important physical assumption; and it is another such assump- tion to suppose that this sum-function F(x) possesses a deri- vative F'(x)=f(x), which we then call the density of distri- bution. The sum-function is connected with the density of dis- tribution by the equations *(*) = fmdu; F(b) - F(a) = f f{x)dx. • Got. stUclctoeiie; of. Chap. IX, § 3. 126 FUNDAMENTAL IDEAS [Chap. This density of distribution is occasionally referred to as the specific probability that a molecule possesses the velocity x. The idealization we have just carried out plays a great part in the kinetic theory of gases originated by Maxwell; in exactly the same mathematical form it appears in many problems of mathematical statistics. 7. The Estimation of Integrals and the Mean Value Theorem of the Integral Calculus We close this chapter with some considerations about a matter of general significance, the whole importance of which will not appear until somewhat later. The point in question is the estimation of integrals. 1. The Mean Value Theorem of the Integral Calculus. The first and simplest of these estimation rules runs as fol- lows: if in an interval a ^ x ^ b the continuous function /(as) is everywhere non-negative (is either positive or zero), then the definite integral f(x)dx f.> is also non-negative. Similarly, the integral is not positive if the function is positive nowhere in the interval. The proof of this theorem follows directly from the definition of the integral. From this the following theorem arises: if f(x)^g(x) everywhere in the interval a ^x<^b, then rf(x)dx Si / g(x)dx •'a also. For by our first remark the integral of the difference f(x) — g (x) is non-negative and by our addition rule (cf . p. 82) / (f(x)—g(x))dx= I f{x)dx— f g{x)dx. Let M be the greatest and m the least value of the function f(x) in the interval ab. The function M — f(x) is non-negative in the interval, and the same is true for the function f(x) — m. II] THE ESTIMATION OF INTEGRALS 127 From the above remark we immediately obtain the double inequality rmdx^ J f(x)dx^ j Mdx. But / mdx= ml dx= m(b — a) and likewise / M dx=M(b — a), whence m(b — a) ^ jf(x)dx ^M(b — a). The integral under •'a consideration can therefore be represented as the product of (6 — a) and some number p between m and M : f f(x)dx-- J a fi(b — a), m^fi^M. As a rule there is no need to state the exact value of this mean value fj.. We may, however, state that it will be assumed by the function at one point f of the interval a ^ £ ^ 6 at least, since in its interval of definition a continuous function assumes all values between its greatest value and its least. As in the case of the mean value theorem of the differential calculus, the exact statement of the value $ is in many cases unimportant. We may therefore put p = /(£), where £ is an intermediate value of x, and we then have ff(x)dx=(b-a)f(Z), a^i^b. This last formula is called the mean value theorem of the integral calculus. We can generalize the theorem somewhat by considering, instead of the integrand /(a;), an integrand of the form f(x) p (x), where p (x) is an arbitrary non-negative function, which, like f(x), is to be assumed continuous. Since mp(x) ^f(x)p(x) ^ Mp{x), we immediately obtain the relation m j p(x)dx <; j f(x)p(x)dx ^ M f p(x)dx, or, in a single equation, j" f{x)p{x)dx=f{£) f p(x)dx, J a Ja where £ is again a number intermediate between a and b. 128 FUNDAMENTAL IDEAS [Chap. W f f-e W© have thus proved the following theorem: If f(x) and p(x) are continuous junctions in a ^ x ^ b, amd p(x) ^ 0, <Aew f f(x)p{x)dx=m f V {x)dx, J a J a where a :g f 5S b. 2. Applications. The Integration of x" for any Irrational Value of a. The mean value theorem and the equivalent integral estimates immediately afford us an insight into an intuitive and easily apprehended fact: the value of an integral changes very little if the function itself is everywhere changed very little. In precise language: if in the whole interval a ^ x ^ b the absolute value of the difference of two functions f(x) and g(x) is less than a number e, then the difference of their inte- grals is in absolute value less than e(b— a). In symbols: if throughout the interval a ^ x ^ b we have \f(x) — g(x) | < e, then f f(x)dx —J g(x)dx < e(b — a) or, otherwise expressed, — e(b — a) + fg{x)dx < jf(x)dx < lg(x)dx + e(b — a). Ja •* a Ja Fig. 22 illustrates this theorem very clearly. For the curve y=f{x) we draw the "parallel curves" y—f(x)-\-e and y = j(x) — e. By hypothesis the function g(x) keeps within the strip bounded by these " parallel curves ". It is clear from this that the areas which are bounded by the curves f(x) and g(x) differ from one another by less than half the area of the strip, and the area of the strip is just Fig. 23. — To illustrate the continuity of an integral II] THE ESTIMATION OF INTEGRALS 129 f a {/(*) + e}d* -£{/(*>) - e ) dx = 2«(6 - a). No appeal to intuition is needed. Since -e + g(x)<f(x)<e + g(x)', it follows, by considerations analogous to those on p. 126, that £{-<■+ g(x)}dx <£f{x)dx <£{g(x) + e }dx, which as the result of the fundamental rules of integration takes the form — e(b — a) +£g(x) dx <Jf(x)dx < fg{x)dx + e(6 - a); here we have merely replaced the integral of a sum by the cor- responding sum of integrals, and have noticed that redx= e(b — a). _ As an indication of the importance of this theorem, we shall show that with its help we are able to integrate the function x°- for any irrational value of a, or, more exactly, to calculate the definite integral /x"- dx. Here we assume that < o < 6. _ We represent the index a as the limit of a sequence of rational numbers "l* a o^, . . . , so that a = lima„; here we can assume that none of the values a n is equal to — 1, since a itself is different from — 1. For the power a: a we then use the definition x"- = lim x a n and notice the following: no matter how small a positive number e we choose, we can always find an n so large that in the whole interval * i^igJwe have j x* — «"*» I < e. * This can be proved quite simply as follows. (Cf. Appendix I, « 3 n 69) Remembering that x°- is monotonic, and putting 8„ = a» - o, we have | a* - x*. | -a*|l-a*.|s(a» + 6a)(|l-a»»| + | 1 - W» | ); for a» lies between o". and 6«, so that a*» s a* + fta, a nd likewise 1 - a*, lies between 1 - <»s» and 1 - 68», so that |l-a&.|£(ll-aM + ll-M.l\ From Urn a*.- lim M. - 1, it follows that ' * ' * W * 1) - *» — >w n— >.<» lim | 1 - a*» | = lim II - 6*» I - 0; so if n is chosen large enough the right-hand side of the inequality is less than c This gives us I as*, -a* I < € simultaneously for all values of x in the interval a £ x £ 0. 130 FUNDAMENTAL IDEAS [Chap. Now we need only apply the relationship mentioned above to the functions f(x) = x°- and g(x) = a;"*, obtaining rb sb pb — e(6 — a) + J x°™dx < / x a dx < I x°*dx + e(6 — a). J a Ja "a The integrals on the right and on the left, however, may be evaluated in accordance with the result on p. 85, giving — e(6 - a) H — (tr* +1 - W+ 1 ) < /V dx < — - (6«»+ 1 — o»«+ 1 ) + e(6 — a). If we now let the number s steadily decrease and tend to 0, the cor- responding values of n increase beyond all bounds; the numbers <x„, a"*, and &<"« must then converge to a, a a , and 6" respectively, and we immediately obtain the result x"- dx = (6«+ 1 — o^ 1 ). a + 1 In other words, the integration formula that holds for rational values of a holds also for irrational values of a. From this it follows in virtue of the fundamental theorem of p. Ill that for positive values of x the differentiation formula d — x «+i = (a + 1)3° dx already obtained for rational values of a remains valid for irrational values of a also. Examples 1. Find the intermediate value 5 of the mean value theorem of the integral calculus for the following, and interpret geometrically: xdx. (a) / Idx. (b) \ rb H (c) J x n dx. (d) / Ja Ja b dx x*' 2. Let f(x) be continuous. Prove, from the mean value theorem of the integral calculus, that the derivative of the indefinite integral of f(x) is equal to f(x). 3. (a) Evaluate l n = jx L l"dx. What is lim /„? Interpret geometri- cally. (6) Do the same for J n = j x n dx. Jo II] THE ESTIMATION OF INTEGRALS 131 4.* Let the function /(£) be continuous for all values of 5, and let F(x) be denned by tie equation F(x) = i/* /(a: + t)dt ' where 8 is an arbitrary positive number. Prove that: (a) the function F(x) possesses a continuous derivative for all values of x; (6) in any fixed interval ogigi we can make | F{x) — j(x) \ < e, where e is an arbitrary pre-assigned positive number, by choosing 8 small enough. 5.* Schivarz's inequality for integrals. Prove that for all continuous functions f(x), g(x) f (S^)fdxj\g(x)fdx ^ (f b f(x)g(x)dxy. Appendix to Chapter II 1. The Existence of the Definite Integral of a Continuous Function We have still to give a proof of the fact that the definite integral of a continuous function between the limits a and b (a < b) always exists. For this purpose we recall the notation °f § 1 (P- 79), and consider the sum F n = i f(£ v )bx„. It is certainly true that £. = i f(v v )Ax v g F n 5S S /(«,) As„ = T n , v = 1 v =■ 1 where f(v v ) denotes the least and/(w„) the greatest value of the function in the v-th sub-interval. The problem is to prove that F„ tends to a definite limit independent of the particular manner of subdivision and of the particular choice of the quantities £„, provided that as n increases the length of the longest sub-interval tends to zero. To establish this it is obviously necessary and sufficient to show that the two expressions F n and F~ n converge to one and the same limit. 13a FUNDAMENTAL IDEAS [Chap No matter how small the positive number e is chosen, we know by the uniform continuity of f(x) that in every suffi- ciently small interval the "oscillation" | /(«„)— /(«„) | is less than e; so that if the subdivision is fine enough we cer- tainly must have ^ T n - F» = i Ax r {f(u v ) -/(«,)} < e(6 - a). V ■= 1 We therefore see that as n increases this difference must tend to zero, and so we can content ourselves with proving that one of the sums, say F„, converges. This convergence will be proved as soon as we show that | F n — F m | can be made as small as de- sired by requiring that the corresponding subdivisions (which we shall refer to as " subdivision n " and " subdivision m " respectively) go beyond a certain degree of fineness. This degree of fineness is characterized by the property that for both sub- divisions the oscillation of the function in each sub-interval is less than e (e > 0). We pass to a third subdivision whose points of division consist of all the points of subdivision n and of subdivision m taken together. This new subdivision, which has say I points of division, we denote by the suffix I, and we con- sider the corresponding upper sum Ft. We shall now estimate the value of \F n — F m \ by first obtaining estimates for the expressions \F~ n — F t \ and \F m — F t \. We assert that the following two relationships hold: F^T t ^T n and F^^It^F^. The proof follows at once from the meaning of our expressions. Let us consider say the v-th sub-interval of the subdivision n. This sub-interval will consist of one or several sub-intervals of the subdivision I; the terms corresponding to these intervals will each consist of two factors, one of which is a difference Ax and the other of which is certainly not greater than /(«„) and not less than /(»„). The sum of the lengths Ax of those intervals of the subdivision I which lie in the v-th sub-interval of the coarser subdivision n is, however, exactly Aa;„. Wo therefore see that the corresponding contribution to the sum F t must lie between the limits f(u y )Ax v aa.df(v v )Ax v . If we now sum over all the n sub-intervals we obtain the first of the above inequali- II] EXISTENCE OF DEFINITE INTEGRAL 133 ties; the second is obtained in exactly the same way if we con- sider the subdivision m instead of the subdivision n. We have already seen that Y n — F\, < e(6 — a); it is like- wise true that F m — F^ < e(6 — a). From the inequalities for F t proved above, it therefore follows that 0^T„-¥ l <e(b—a) and 0<, F^ t —¥ l < € (b- a). Thus it is also certain that I K- f^\ = I (FZ~F\)~ (F m - F\) I < 2e(6- a). Since e can be chosen as small as we please, this relation shows us by Cauchy's convergence test (p. 40) that the sequence of numbers F n actually converges. At the same time we see at once from our argument that the limiting value is completely independent of the manner of subdivision. The proof of the existence of the definite integral of a con- tinuous function is thus complete. Our method of proof teaches us still more. It shows us that in many cases we are also led to the integral by a somewhat more general limiting process. If, for example, /(a;) = 4>(x)r]t(x) and the interval from a to 6 is subdivided into n parts by the points of division x v , we consider instead of the sum £/(£,) Ax r the more general sum Z*(f/W(£,")Aav, where £/ and £„" are two not necessarily coincident points of the v-th sub-interval. This sum will also tend to the integral Jf(x)dx = J <f>(x)<Ji(x)dx as n increases, provided that the length of the longest sub-interval tends to zero. A corresponding statement holds for all sums formed in an analogous way; for example, the sum tends to the integral fv{<l>(x)* + 4>(x) 2 }dx. 134 FUNDAMENTAL IDEAS [Chap. The proof of these facts follows lines exactly similar to the above and hence need not be worked out in detail. 2. The Relation between the Mean Value Theorem op the Differential Calculus and the Mean Value Theorem of the Integral Calculus. Between the mean value theorem of the differential calculus and that of the integral calculus there exists a simple relation which is arrived at by way of the fundamental theorem (p. Ill) and which we give as an instructive example of the use of that theorem. We take the mean value theorem of the integral calculus in its more special form, ff(x)dx=(b-a)f(£). If we put ff(x)dx—F(x), so that f(x) = F'(x), the theorem just written takes the form F(b)- F(a)=(b-a)F'(t) b — a Here we can obviously choose for F(x) any function whose first derivative F'(x)=f(x) is continuous, and thus for such functions the mean value theorem of the differential calculus is proved. If we consider the more general form of the mean value theorem of the integral calculus, rf(x)p(x)dx=f(g) I p(x)dx, where p(x) is a function which in our interval is continuous and positive and f(x) is an arbitrary continuous function, we are led to a correspondingly more general mean value theorem of the differential calculus. We put Jf(x) p(x)dx = F(x), i.e. f(x)p (x) = F'(x), and fp(x)dx=G{x), i.e. p{x) = G'(x); II] THE MEAN VALUE THEOREMS 135 the above mean value formula then takes the form F(b)-F(a) = {G(b)-G(a)}f(g), ,, . Fix) or, since /(a;) = ^LJ F(b)-F(a) _F(j) 0(6) -0(a) G'd)' where a =f= b. This formula, in which £ once again denotes a number inter- mediate between a and b, is called the generalized mean value theorem of the differential calculus. For this to be valid it is obviously sufficient to assume that F(x) and G{x) are continuous functions with continuous first derivatives and that in addition G'(x) is everywhere positive (or everywhere negative). For with these assumptions the whole process can be reversed. Finally, it should be observed that in the present discussion of the mean value theorem of the differential calculus we have had to make assumptions more stringent than the theorems in themselves require. (Cf. § 3, No. 8, p. 103, and later p. 203.) Example 1. Show that if f(x) haa a continuous derivative in the interval * ^ x ^ b, then /(a) can be represented as the difference of two monotonia functions. CHAPTER III Differentiation and Integration of the Elementary Functions 1. The Simplest Rules for Differentiation and their Applications In higher analysis and its applications it is usually the case that the problems of integration are more important than those of differentiation, but that differentiation offers less difficulty than integration. Consequently the natural method of building up the integral and differential calculus is first to learn to dif- ferentiate the widest possible classes of functions and then by virtue of the fundamental theorem (Chap. II, § 4, p. 116) to make the results thus obtained available for the solution of integration problems. In the following sections it will be our task to carry out this programme. To a certain extent we shall make a fresh start, since we shall work out the most important differen- tiations and integrations systematically without calling upon the results of last chapter. In this development of the subject certain rules for differentiation, with the first of which we are already acquainted (p. 96), will play an important part. 1. Rules for Differentiation. We assume that in the interval which we are considering the functions f(x) and g (x) are differentiable; our rules then run as follows: Rule 1. Multiplication by a constant. If c is a constant and <f>(x) = cf{x), then <f>(x) is differentiable, and <f>\x) = cf'(x). Chap. Ill] RULES FOR DIFFERENTIATION 137 This follows immediately from the relation 4(x + h)-4( X ) /(» + *)-/(») A A if we take the limits as h -> 0. Rule 2. Derivative of a sum. If <ji{x) =f(x) + g(x), then <f>(x) is differentiable, and f (*)=/'(*) + flr'(a;); that is, the processes of differentiation and addition are inter- changeable. The same holds for the sum of any finite number (n) of terms #(*)=S /,(*), for which we obtain +'(z) = £/„'(*). We may pass over the proof, which after Chap. II, § 3 (p. 88) is fairly obvious. Rule 3. Derivative of a product. If <f>(x) = f(x)g(x), then <f>(x) is differentiable, and <l>'(x)=f(x)g'(x) + g(x)f'(x). The proof follows from the equation <f>(x+h)~<f,(x) __ f(x+h)g(x+h)-f(x)g(x) h h _ f^+h)g(x+h)-f{x+h)g(x)+f(x+h)g{x)-f{x)g(x) h = f(x+h) 9{x+h) -9W+g(x) f {x+h) -K X \ h h In this last expression the passage to the limit h -*■ can be directly carried out, yielding the formula stated. This formula takes a still more elegant form if we divide * throughout by <f>(x) = f{x)g{x). We then obtain $'{x) _f'{x) g'(x) 4>(x) f{x) g(x)' * We most, of course, assume that 4>{x) is nowhere equal to zero. 138 DIFFERENTIATION AND INTEGRATION [Chap. By repeated application of this product formula we obtain by induction for the derivative of a product of n factors an expression consisting of n terms, each of which consists of the derivative of one factor multiplied by all the other factors of the original product. In symbols: 4>'(*) = ~{h(*)Ux)-..Ux)} or on division * by <J>(x) = Mx)f 2 (x) . . .f n (x) #*) /i(«) /,(*) /.(*) ,-ifJxj' Rule 4. Derivative of a quotient. For a quotient the following rule holds: the function <f>(x) is differentiable at every point at which g(x) does not vanish, and * u {9W 2 • If <f> (x) =j= 0, this can be written 4>i x ) f(x) g{x)' If we accept the differentiability of <f>(x) as a hypothesis, we can apply the product rule to f(x) = <f>(x)g(x) and conclude that f'{x)^4>{x)g\x) + g{x) < j>'{x). By substituting-^ for tf>{x) on the right and solving for <f>'{x) we obtain the rule stated above. In order to prove the differen- * We most, of course, assume that <j>(x) is nowhere equal to zero. HI] RULES FOR DIFFERENTIATION 139 tiability of <f>{x) as well as the rule we use the following method. We write /(»+*) f{x)_ 4>(x+h)—<f>(x) _ g(x+h) g(x) h h jH /(*+A)-/(aQ _ 9(* + h)-g(x) =x » h g(x)g(x + h) If we now let h tend to 0, we arrive at the result stated; for by hypothesis the two terms obtained by performing the division on the right have definite limits, which are respectively ?WW AS'Wffr) rm- {SWY {q(x)\ 2 ' at ° nCe P roves botl1 tne existence of the limit on the left-hand side and the differentiation formula. 2. Differentiation of the Rational Functions. To begin with, we shall again deduce the differentiation formula d — x n = na;" -1 ax for every positive integer n, basing the proof on the rule for differentiating a product. We think of x n as a product of w factors, x n = x ... x, and thence obtain d .»-i : The second derivative of the function x» results if we use the above formula and the first rule of differentiation: d? — x n —n{n~ \)x n -\ Continuing the process, we obtain dx 3 d? ~x n =n(n—l)(n— 2)x nr ~ 3 d n -3— x n = 1 . 2 . . . n = n! ax" i 4 o DIFFERENTIATION AND INTEGRATION [Chap. From the last of these it is clear that the (n+ l)-th derivative of x n vanishes everywhere. In virtue of our first two rules, a knowledge of the differ- entiation of powers at once enables us to differentiate any polynomial y = a + <H X + °» r? + • • • + a n x "- We have simply y' = ay + ^a^ + 3a 3 a? + • • • + na n x tt ~ 1 , and further y" = 2a 2 + 3 . 2a z x + 4 . Za^x 2 +... + «(«— l)a„a: B - 2 , and so on. The differentiation of any rational function now follows with the help of the quotient rule. In particular, we shall again deduce the differentiation formula for the function x n , where n = — m is a negative integer. The application of the quotient rule, together with the fact that the derivative of a constant is equal to zero, gives us the result mx™^ 1 m -(-) = dx vc"*/ j2m jpw»+l' or, if we take m = — n, d — x n = wa;" -1 , dx which agrees formally with the result for positive values of n and with the results given earlier (p. 95). 3. Differentiation of the Trigonometric Functions. For the trigonometric functions sin a; and cos a; we have already (p. 96) obtained the differentiation formulae d . -,d — sina; = cosa; and — cosa; = — Binsr. dx dx The quotient rule now enables us to differentiate the functions sina; , . cosa; y = tana; = and y — cota; = — — . cosa; sin x Ill] RULES FOR DIFFERENTIATION i 4I According to the rale, the derivative of the first of these func- tions is . _ cos 2 a; + sin 2 a; 1 cos 2 a; cos 2 a;' and we obtain the result <*.___ _1 cos k a: Similarly, we obtain — tana; = —- J - = S ec 2 a; = 1 + tan 2 x. d „._ 1 dx cota; = — -__ = — cosec 2 a! = —(1 + cot 2 a;). sin* a; 2. The Corresponding Integral Formula 1. General Boles for Integration. The fundamental theorem of p. 116 and the definition of the indefinite integral reveal to us the possibility of writing down an integral formula corresponding to each differentiation formula. The following rules of integration (of which the first two have already been mentioned oh p. 82) are completely equivalent to the first three rules of differentiation. Multiplication by a constant: If c is a constant, then fcf(x) dx = cff(x) dx. Integration of a sum: It is always true that /{/(*) + g(x)}dx =ff(x) dx+fg (x) dx. To the third rule of differentiation corresponds the rule for the integration of a product, or, as it is usually called, the rule for integration by parts. On integration the product rule gives j{f(*>)9(x)}'dx =Jf(x)g'(x)dx +fg(x)f'(x)dx. The indefinite integral on the left is obviously f(x)g(x) (except possibly for an additive constant), and we can therefore write the rule for integration by parts in the following form: Jf(x)9'(x)dx =f(x)g(x) -fg(x)f'(x)dx. 142 DIFFERENTIATION AND INTEGRATION [Chap. This last integration, formula, the counterpart of the rule for the differentiation of a product, has been given here only for the sake of completeness; it will not become important for us until the next chapter (p. 218). 2. Integration of the Simplest Functions. Corresponding to the differentiation formulae for special functions which we have recently found, we now set down the equivalent integration formulae. The formula d — x" = nx ,n-l dx when expressed as an integration formula becomes / x n ~ 1 dx = — , n 4= 0. For this formula merely means that the derivative of the right- hand side is equal to the expression under the integral sign on the left. If we replace n by n + 1, we obtain the integral formula fx n dx = x n+1 , n 4= — 1. J n-\- 1 This formula holds for every integral index n (where n < it of course holds only if x 4= 0) with the exception of n = — 1, for which the denominator ft + 1 would vanish. Later (p. 167) this exceptional case will be studied in detail. The fundamental theorem of the integral calculus at once permits us to use our integral formulas for the determination of areas, that is, of definite integrals. By p. 117 we immediately obtain "'O x"dx = (b n + 1 — a n+1 ), »4=- 1, n -f- 1 where if n is negative we assume that o and b are of the same sign, since otherwise the integrand would be discontinuous in the interval of integration. To the differentiation formulae for sin a;, cos a;, tana;, and cot a; correspond the following integration formulae: HI] RULES FOR INTEGRATION 143 Jcoaxdx= sina:, fainxdx = — cosa;, / — — dx = tana;, /"-— - dx = — cotx. J cos 2 a; ./ sin 2 a; From these formulae we obtain by way of the fundamental rule of Chap. II, § 4 (p. 117) the value of the definite integral between any limits, the only restriction being that when the last two formulas are used the interval of integration must not contain any point of discontinuity of the integrand. For example, r- coaxdx = sina; = sin& — sina. It scarcely needs to be emphasized that with the help of the first two rules of integration we are now in a position to integrate any polynomial in x, and, in fact, any linear combina- tion with arbitrary constant coefficients of the functions inte- grated here. The following point, however, should be noted. Rules of integration and rules of differentiation must according to the fundamental theorem be equivalent to one another; it is therefore possible first to prove the general integration rules of this section and then to read off the differentiation rules of the preceding section. The reader would be well advised to carry out this suggestion for himself. Examples 1. Find the numerical values of all the derivatives of x s — x* at x = 1. 2. What is the numerical value of the eleventh derivative of 317a; 9 — 202a;' + 76 at x = 13J? 3. Differentiate the following functions and write down the correspond- ing integral formulae: (a) ox+b. . aa? + 26a; + c (6) 25ca;'. «** + 2 P* + Y* 1 1 (c) + 2bx + cx\ if) j^^ j-j^j. (d) ax + b , . , (a?-V8a;« + 4)(s»+ VSx* + 4) cx + d Ky > *"+16 • 4. Let P(x) = ffl + a x x + a^ + ... + a n x n . (a) Calculate the polynomial F(x) from the equation F(x) — F'(x) = P(x). (&)* Calculate F(x) from the equation c^x) + o^'ix) + c a F"(x) = P(x). 144 DIFFERENTIATION AND INTEGRATION TChap. 6. Differentiate the following functions and write down the corre- sponding integral formulae: (a) 2 sin x cos x. (c) x tan x. . . sin a ; ... 1 .,. sin x + cos x x ifi) . (a) — 1 + tan x sm x — cos x Recalling that sec x = , coseo x = , find the derivatives indicated in Ex. 6-9: ooax Bma: 6. — sec x. 8. — coseo x. dx 3 dx* J* »4 7. — seo x tan X. 9. ■— tan x sin x. tfct» dx* 10. Find the limit as n -> 00 of the absolute value of the m-th deri- vative of — at the point x — 2. x Evaluate: 11. yW + b)dx. 16. y(x* + A) dx. 12. /"(ox" + 26x + c)dx. 16. /Ya cos x + -rrj dx - 13. f(6x a +7x t +Sx*+3x t +l)dx. 17. f( 3a: + 7Bina: +^ _ ^)* ,: - 14. /Yi + _ + — )dx. 18. F seo x tan x dx. 3. The Inverse Function and its Derivative 1. The General Formula for Differentiation. We have seen earlier (pp. 21 and 67) that a continuous func- tion y=f(x) has a continuous inverse in every interval in which it is monotonic. More exactly: If a ^ x ^ b is an interval in which the continuous function y = f (x) is monotonic, and if i (a) = a and f (b) = /?, then x is a function of y which in the interval between a and )S is one-valued, continuous, and monotonic. As we have already shown on p. 92, the concept of the deri- vative gives us a simple means for recognizing that a function is monotonic and therefore has an inverse. For a difEerentiable function is certainly always monotonic increasing if f'(x) is greater than zero throughout the corresponding interval, and HI] INVERSE FUNCTIONS i 4S similarly is monotonic decreasing if f'(x) is everywhere less than zero in the interval. We shall now prove the following theorem: // in the interval a < x < b the function y = f (x) is differen- tiable, and in that interval either f(x) > everywhere or eke i'(x) < everywhere, then the inverse function x = <f>(y) also possesses a derivative at every point of its interval of definition, and between the derivative of the given function y = f (x) and that of the inverse function x = (f>(y) there exists for corresponding values of x and y the relationship f'(x) . ^'(y) = 1, which we can also write in the form dx dx' dy In this last formula we again observe the flexibility of Leibnitz's notation. It is just as if the symbols dy and dx were quantities which could be operated with like actual numbers. The proof of this formula is correspondingly simple if we regard the derivative as the limit of the difference quotient, &y_ n m yx-y y'=f'(x)= lim =»= lim Ax- OASB Xt^-xXj^ — X where x and y=f(x), and x 1 and y 1 =f(x 1 ), respectively denote pairs of corresponding values. By hypothesis the first of these limiting values is not equal to zero. On account of the continuity °£ V —f( x ) an d x = j>(y) the equation lim Ax = is equivalent to lim Ay = 0, and consequently the relations y x ->■ y and x x ->- x are also equivalent. Therefore the limiting value lim*LZL5= fon^LZf *t->*yi — y y^yyi — y exists and is equal to — — v . On the other hand, the limiting value is by definition the derivative <f>'(y) of the inverse function <f)(y), and thus our formula is proved. This formula has a simple geometrical meaning, which is clearly shown in fig. 1. The tangent to the curve y = J(x) or x = <f(y) forms with the ' (B798) 146 DIFFERENTIATION AND INTEGRATION [Chap. positive x-axis an angle a, with the positive «/-axis an angle P, and from the geometrical meaning of the derivative y yC ^1C f'(x) == tan a, <p'(y) = tan p. Since, however, the sum of the angles a and (3 is rc/2. tan a tan p = 1, and this relationship is exactly equivalent to our differentiation formula. We have hitherto expressly assumed that either f'(x) > or f'(x) < 0, i.e. that f'{x) is never zero. What, then, happens if /'(*)=(» If /» = every- where in an interval the function is constant there, and consequently has no inverse, since the same value of y must correspond to all values of x in the interval. If the equation f'(x) = is true only at isolated points, "' and if for the sake of sim- plicity f'(x) is assumed con- tinuous, then we must dis- tinguish whether on passing through these points f(x) Fig. i. — Differentiation of the inverse function o Fig. 2. — Parabola ~? Fig. 3. — Cubical parabola changes sign or not. In the first case this point separates a point where the function is monotonic increasing from another where it is monotonic decreasing. In the neighbourhood of such a point there can be no single-valued inverse function. In the second case the vanishing of the derivative does not destroy the monotonic character of the function y=f(x), so that a single- HI] INVERSE FUNCTIONS 147 valued inverse exists. But the inverse function will no longer be differentiable at the corresponding point; in fact, its derivative will be infinite there. The functions y=x 2 and y = a^ at the point x = offer examples of the two types. Figs. 2 and 3 illustrate the behaviour of the two functions where they pass through the origin and at the same time show that one of the functions, namely y = as 8 , has a single-valued inverse, but that the other function, y = x% has not. 2. The Inverse of the Power Function. The simplest example of an inverse function is offered by the functions y = x n for positive integers n and, as we at first assume, positive values of x. Under these conditions y 1 is always positive, so that for all positive values of y we can form a unique positive inverse function *= \/y=y 1,n - The derivative of this inverse function is immediately obtained in accor- dance with the above general rule by the following calculations: <% 1/w ) = dx = \_ = _J 1 1 = 1 1/n _ a dy dy dy nx"- 1 n y(«-D/» » y ' dx and if we now denote the independent variable by x, we may finally write dx dx n which agrees with the result obtained directly on p. 94. The point x = requires special consideration. If x approaches through positive values, dfaV^/dx, where n > 1, will obviously increase beyond all bounds; this corresponds to the fact that f or n > 1 the deri- vative of the n-th power f(x) = a;" vanishes at the origin. Geometrically this means that the curves y = a; 1 /", n > 1, touch the j/-axis at the origin (cf. fig. 17, p. 34). For the sake of completeness it should be noted that for odd values of n the assumption that x > can be omitted and the function y=x n can be considered for all values of x without loss of its monotonic character or of the uniqueness of its inverse. The differentiation formula — (w 1 '") = - » l '" —1 dy n d(x n ) still holds for negative values of y; for x = 0, n > 1, we have — — = 0, dx which corresponds to an infinite derivative (dx/dy) of the inverse function at the point y =»= 0. 148 DIFFERENTIATION AND INTEGRATION [Chap. 3. The Inverse Trigonometric Functions. In order to form the inverses of the trigonometric functions we once again consider the graphs of sin a;, cos a;, tana;, and cot a;. We at once see from figs. 14 and 15, p. 25, that for each of these functions it is necessary to select a definite interval if we are to speak of a unique inverse; , for the lines y = c parallel to the x-axis cut the curves in an infinite number of points, if at all. ■ sinx X-arcsiny Fig. 4. — The inverse sine function For the function y = sin a; the derivative y' = cos a; will e.g. be positive in the interval — n-/2 < x < it/2. In this interval the sine accordingly has an inverse function; we write the in- verse function of the sine in the form * x = arc sin^ (read arc sine y; this means the angle whose sine has the value y). This function runs monotonically from — tt/2 to +7r/2 as y tra- verses the interval — 1 to +1. If we especially wish to emphasize that we are considering the inverse function of the sine for this very interval, we speak of the principal value of the arc sine. If we form the inverse function for some other interval in which sina; is monotonic, e.g. the interval +tt/2 < x < 3tt/2, we obtain " another branch " of the arc sine; without the exact statement of the interval in which the values of the function must lie the arc sine is a multiple-valued function, and in fact has an infinite number of values. In general, the fact that arc siny is multiple- valued is expressed by the statement that to any one value y of the sine there corresponds not only the angle x but also the angle 2kir + x, as well as the angle (2k + l)ir — x, where k is any integer (cf. fig. 4). * The notation x = waT'y is also used in English books. Ill] INVERSE FUNCTIONS 149 The differentiation of the function x = arc siny is performed in virtue of our general rule by the following short calcula- tion: dx dy 1_ V' 1 cos a; ± V(l - sin 2 a;) + V(l - V 2 )' where the square root is to be taken as positive if we confine ourselves to the first interval mentioned.* If the independent variable is finally changed back from y to x, the differentiation formula for the function arc sin a: is obtained in the following form: d 1 — arc smx = — — -. dx V(l — a; 2 ) Here it is assumed that the arc sine lies between — w/2 and +w/2, and the square root sign is chosen positive. Ly-cosx x-arccosy Fig. 5. — The inverse cosine function For the inverse function of y= cos a;, denoted by arc cos z, we obtain the differentiation formula d 1 — arc cosa; = — dx v(l x 2 ) in exactly the same way. Here we take the positive sign of the root if the value of arc cos a; is taken in the interval between and tr (not, as in the case of arc sinx, between — tt/2 and +w/2); cf. fig. 5. A word remains to be said about the end-points x = — 1 and x= +1. The derivatives become infinite on approaching these end-points, corresponding to the fact that the graphs of the » If instead of this we had chosen the interval w/2 < x < 3ir/2, correspond- ing to the substitution of x + it for *, we Bhould have had to use the negative square root, since coax is negative in this interval. 150 DIFFERENTIATION AND INTEGRATION [Chap. inverse sine and inverse cosine must possess vertical tangents at these points. We can deal with the inverse functions of the tangent and cotangent in an analogous way. The function y = tan x, whose de- rivative 1 /cos 2 a; for x 4= n-/2 -f- /for is everywhere positive, has a unique inverse in the interval — ir/2 < x < it 12. We call this inverse function x = arc tan y or (by interchange of the letters x and y)y= arc tan x. We see at once from fig. 6 that the original many-valuedness of the inverse — i.e. the many-valuedness which occurs if the interval of the values of the function is not fixed — is expressed by the fact that for each x we could have chosen instead of y any of the values y -\-1cit (where k is an integer). For the function y = cota; the inverse x = arc coty, or (by interchange of x and y) y = arc cot a;, is uniquely determined if we require that its value shall lie in the interval from to tr; the many-valuedness of arc cot a; is otherwise the same as for arc tan a;. The differentiation formulae may be found as follows: dx 1 li x = arc tany, — = _ = cos 2 a; = = _ • dy dy 1 -f tan a 3 1 + y*' dx . dx . . 1 1 x = arc cotw, — = — sm 2 a; = — = _ x . dy 1 + cot 2 a; 1 -f y*' or finally, if we denote the independent variable by x, d . 1 — arc tana; = — , dx 1 + x* Fig. 6. — The inverse tangent function d 1 — - arc cota; = — . dx l-\-x* HI] INVERSE FUNCTIONS 151 4. The Corresponding Integral Formula. Expressed in the language of the indefinite integral, the formulae which we have just derived read as follows: J -. , » dx = arc tanas, J dx = —arc cotsc. Between the pair of formulae on the left and that on the right, which express each indefinite integral in the form of two functions which appear entirely different, no contradiction exists. We must remember that in the case of the indefinite integral an arbitrary additive constant remains at our disposal. If we choose these constants so that they differ by tt/2 and recall that it/2 — arc cos a; = arc sin a; and likewise w/2 — arc cot a; = arc tana: this formal disagreement is immediately cleared up. The in- definiteness simply depends on the fact that the indefinite integral is not a single definite function, but a whole family of functions which differ from one another by arbitrary additive constants. The equation for an indefinite integral specifies not the value, but only o value, of it. As we have already remarked, it would be more correct to express this fact by always including the undetermined constant, thus writing, not ff(x)dx=F(x), but ff(x)dx=F(x) + c For convenience, however, it is usual to avoid this more detailed form; the reader should therefore be all the more careful to bear in mind the indefiniteness which is always associated with the shorter form (see also p. 116). From the formulae for indefinite integration there immedi- ately follow formulae for definite integration, as on p. 117. In particular, r dx ■ = arc tana: 1 + a* = arc tan 6 — arc tana. i5« DIFFERENTIATION AND INTEGRATION [Chap. If we put a = 0, b = 1 and recall that tanO = and tanir/4 = I, we obtain the remarkable formula tt_ r 1 1 4 Jo 1 + 4-z 2 dx. The number it, which originally arose from the consideration of the circle, is by this formula brought into a very simple relation- ship with the rational function defined as shown in fig. 7. Fig. 7. — ir/2 illustrated by an area 1 \ + X , and is expressed by the area Examples x* dy 1. If « = — , y = 16 corresponds to x = 8. Find — for x = 8; solve a; 3 * dx u = — for x and find — for y = 16, and show that the values of these a 4 dy derivatives are consistent with the rule for inverse functions. 2. Prove that (a) arc sin oc+ arc sin p = arc sin (a V 1 — p 2 -f- P V 1 — a 2 ); (6)arc since + arc sin(3=arc cos(Vl— a 2 Vl — p 2 — aP); a+P (c) arc tan a + arc tan P = arc tan 1-aG Differentiate the expressions in Ex. 3-10 and write down the corre- sponding integral formulae: 6. \/x 1 — tana; 9. arc sini arc tan a: 4. Va 7. arc sinar.aro oosa;. 10. B arc cotse -f- 1 + Vx 8. 1 + arc tana; 1 — arc tana; Ill] INVERSE FUNCTIONS r S3 II. Using graph paper, plot y = on a large scale. By oounting ' — — 5 dx, thus obtaining an estimate for - (of. Ex 1 F . x*x,. o 1 + & 4 • ' 4. Differentiation of a Function of a Function 1. The Chain Rule. The preceding rules for differentiation enable us to differen- tiate every function which can be expressed as a rational expres- sion whose terms are functions with known derivatives. We can, however, take yet another important step forward and differentiate all those functions obtained by compounding func- tions with known derivatives. Let <f>(x) be a function which is differentiable in an interval a^Lx^b and assumes all values in the interval a^<f>^ /?. We now wish to consider a second differentiable function g(<f>) of the independent variable <j>, in which the variable <f> ranges over the interval from a to /?. We can now regard the function g(<j>) = g{<f>(x)} =f(x) as a function of a; in the interval a^.x<Lb. The function f(x) = g{<f>(x)} will then be called a function of x compounded from the functions g and <f>, or a function of a function. If, for example, <p(x) = 1 — x* and gr(<p) = V<p, this compound func- tion is simply f(x) = V(l — a; 2 ). For the interval ogigJwe here take the interval ^ x ^ 1. The values of the function <j>(x) exactly fill up the interval ^ ip g 1; the compound function f(x) = V(l — x 2 ) is there- fore denned in the interval ^ x ^ 1. Another example of the compounding of functions is the function f(x) — V(l +X 2 }, where the compounding process may be indicated by the equations y(x)=l + x*, g( 9 ) = V9 and where the value of the function <p(ar) runs through all positive numbers ^ 1, so that the function f(x) = g{<?(x)} can be formed for all values of x. In compounding functions in this way we must naturally be careful to restrict ourselves to intervals a gas^i for which the compound function is defined. For example, the compound function V(l — x 2 ) is defined only for values of a; in the region -lgj;^l, and not in the region 1 < x ^ 2, for when a; is in this last interval the values of the function tp(x) consist of negative numbers, for which the function g(<f) is not defined. Just as we can compound two functions with one another, we can and • • (B798) 154 DIFFERENTIATION AND INTEGRATION [Chap. must consider functions in which the compounding process is performed more than once. Such a function is V (1 + aro tan a?) which can be built up by the compounding process <p(x) = x*, <\>(<p) = 1 + aro tan <p, g(ty) = V<K<p) = f(x). For the differentiation of compound functions we have the following fundamental theorem, the chain rule of the differential calculus: The function f(x) — g{<^(x)} is differentiable, and its derivative is given by the equation or, in Leibnitz's notation, dy _dy d<j> dx d<f> dx' In words: the derivative of the compound function is the product of the derivatives of the constituent functions. The proof of this formula follows very easily if we recall the meaning of the derivative. For any arbitrary Ace =)= and cor- responding values of A<f> and Ag there exist two quantities e and 7], tending to with Ax, such that &9 = 9'(<! > )&<l> + «A<£ an d A<£ = <f>'(x)Ax + tjAx; we have only to calculate ij from the second equation and, where A<f> =j= 0, e from the first equation, while if A<f> = 0, we put e = 0. If in the first of these equations we now substitute the value of Acj) from the second equation, we obtain A<7 = 9'{<t>)4>'{x)Ax + {rf(4>) + ef (x) + erj) Ax, or jf = *W(*) + {r,g'(<f>) + ef (*) + «,}. In this equation, however, we can let Ace tend to 0, and at once obtain the result stated, since the bracket on the right tends to zero with Ax. Consequently the left-hand side of our equation has a limit f'(x), and this limit is equal to the first term on the right-hand side, as was stated.* * We could also have proved the rule by carrying out the passage to the limit As -*■ 0, and consequently A<j> -*• 0, in the equation =£ — QL . ^r. The Ax A# Ax Ill] FUNCTIONS OF A FUNCTION 155 By successive application of our formula we can immediately extend it to functions which arise from the compounding of more than two functions. If, for example, y = g(u), u = <f>(v), v = if,{x), we can think of y=f(x) as a function of x\ its derivative is given by the rule ax du dv ax The case of a function compounded of an arbitrary number of functions is essentially similar. The proof may be left to the reader. 2. Examples. As a very simple example we consider the function y—nf-, where we put a = p/q, q being a positive integer and p a positive or negative integer, so that a is an arbitrary positive or negative rational number. Let x be positive. By the chain rule with y = 9*, 9 = x l 'i we have the formula y' = pqp-l . I a;(l-9)/« _ P x v lt~ 1 , ? 9 so that for arbitrary rational values of a we obtain the differentiation formula j — x a = cue" -1 , dx in agreement with the result already found in another way in Chap. II, § 3 (p. 94). As a second example, we consider y=V(l — a?) or y=V<f, where 9 = 1 — x 2 and — 1 < x < 1. The chain rule gives us j,'= _L .(_2z) = - x 2V9 V(l — x 2 )' Further examples are given in the following brief calculations: 1. y = arc sin V(l — x 2 ), dy = 1 dV(l — x*) dx V{1 — (1-x 2 )}' dx = _L ~ x ~ =c * ~° \x\'y/(l-x*) "Wa-a?)' method in the text is, however, to be preferred, since it avoids the necessity for considering the case <j>'(x) — specially. 156 DIFFERENTIATION AND INTEGRATION [Chap. dy 1 \\ — x/ B=2 V(K)' * = V(l - x) 2 1 = 2 V(l + *)'(!- a;) 2 (1 + *) 1/2 (1 - s) 3/2 ' The chain rule for differentiation can also be expressed in the form of an integration formula, in agreement with the fact that to each differen- tiation formula there corresponds a completely equivalent integration formula. Nevertheless, we will pass over this formula for the present, since we have no immediate need of it here and, moreover, it is discussed in detail later (Chap. IV, § 2, p. 207). 3. Further Remarks on the Integration and Differentiation of x a when a is Irrational. In view of the elementary definition of the power x" by the equation s" = lim x Tn , where the numbers r n form a sequence of rational numbers with the limit a, we might be tempted to effect the differentiation of x°- by direct passage to the limit in the differentiation formula dx x " = r n x ' We are not entitled to do this unless we have the right to conclude that d d from the relation x r * -»■ x* there follows the relation — x n -> jt x*. There is, however, a very serious objection to such a passage to the limit. For in any arbitrarily small neighbour- hood of a given curve other curves may be drawn whose direction at arbitrarily selected points differs from the direction of the original — „ . . „. . . . . ,. curve by any desired amount; for Fig. 8. — Approximation to a straight line , by wavy curves example, we may approximate to a straight line by a wave lying ar- bitrarily near it, the angle between the wave and the line reaching a value as high as 45° (see fig. 8). In other words, the above example shows us that from the fact that two functions differ only very little from one another, we cannot immediately conclude that their derivatives also are everywhere nearly equal to one another. This objection forbids us to Ill] FUNCTIONS OF A FUNCTION 157 perform the apparently obvious passage to the limit, in the absence of further justification. In this respect, however, the integral behaves quite differently from the derivative. We have already observed on p. 128 that if two functions differ by less than e throughout the interval from o to 6, their integrals must differ by less than e(6 — a). We there used this result to establish the validity of the differentiation formula x a + 1 = x*. a. + 1 dx or, replacing a + 1 by a, d — x* = ax" -1 . dx In this indirect way, therefore, the relation — x r n -* — a;" given above is verified. " ,x " ,x The above discussion is a characteristic example of the interrelations of the differential calculus and the integral calculus. Yet in principle it is preferable to replace (as we shall do on p. 173 et seq.) the elementary definition of *" by another, essentially simpler, definition which will lead us once more to the same result, and this time directly. Examples Differentiate the following functions: 1. (x + l) s . 11. sin (a; 2 ). 2. (3a; + 6)*. 12. V(l + sin"*). 3. (x» - 3x* - x 3 ) 1 . 1 13. a^sin-L 4. —L-. ** 1 + s ,. . 1 + a: 14. tan — ' — . 1 1-a: 1 — ** 15. sin(a: 2 + 3a; + 2). 6. (ax + b) n (ra an integer). 16. arc sin (3 + x 3 ). 7. 17. arc sin (cos a;). 3-l V(a^— 1) v ' 18. sin (arc cos V(l — a; 2 )). c l/ax* + bx+c\ ,\ 8 -yl W + mx+ n } 19. *'»_.-*». 9. ( V(l - x) 2/3 ) 5 . 20. [sin (a; + 7) ] ^*. 10. sin 2 a;. 21. [arc sin (a cos a; + b) ]". 158 DIFFERENTIATION AND INTEGRATION [Chap. 5. Maxima and Minima Now that we have attained a certain mastery of the problem of differentiating the elementary functions and the functions compounded from them, we are in a position to make a variety of applications. Here we shall consider the simplest of these applications, the theory of maxima and minima of a function, in conjunction with a geometrical discussion of the second deri- vative, and then in the next section we shall again take up the thread of the general theory. 1. Convexity or Concavity of Curves. By definition the derivative j~f(x) of a function /(x) gives dx the slope of the curve y =/(%)■ This slope can itself be repre- y, / 1 / 1 / X *x Fig. ga.— /"(*) > Fie. gb.~ r(x) < sented by a curve y = — f(x) = f'(x), the derived curve of dx the given curve. The slope of this last curve will be given by the d d? derivative —f'(x) = —-f(x)=f"(x), the second derivative of dx dx 2 f(x), and so on. If the second derivative f"(x) is positive at a point x — so that owing to continuity (which we here assume) it is positive in a certain neighbourhood of the point x— then the derivative /'(x) must increase as it passes this point in the direc- tion of increasing values of x. Hence the curve y = f(x) turns its convex side towards the direction of decreasing values of y. The opposite is true if f"(x) is negative. In the first case, there- Ill] MAXIMA AND MINIMA »59 fore, the curve in the neighbourhood of the point lies above the tangent, in the second case below the tangent (see figs. 9a and b). Special consideration is required only in the case of points where f"(x) = 0. On passing through such a point the second IV Fig. 10. — Point of inflection -5 derivative f"(x) will, as a rule, change its sign. Such a point will then be a point of transition between the two cases indicated above; that is, the tangent will on one side be above the curve, and on the other side below it, so that besides touching the curve it will also cross it (see fig. 10). Such a point is called a paint of inflection of the curve, and the corresponding tangent is called an inflectional tangent. The simplest example is given by the function y — x*, the cubical parabola, for which the ai-axis itself is an inflectional tangent at the point x = 0. Another example is given by the function j(x) = sin x, for which /'(«) = d(sinx)/dx= cos x and f"(x)~d*(smx)/dx>=—Bmx. Conse- quently /'(0) = 1 and /"(0) = 0; since the sign of f"(x) changes at x = 0, the sine curve has at the origin an inflectional tangent inclined at an angle of 45° to the s-axis. It must, however, be noted that points can exist where f"(x) = although the tangent does not cut the curve, but remains entirely on one side of it. For example, the curve y «*= aj* lies entirely above the z-axis, although the second derivative f"{x) vanishes for x = 0. 2. Maxima and Minima. We say that a continuous function or a curve y = f(x) has a maximum (minimum) at a point £ if in at least some neighbour- hood of the point x = £ the values of the function /(x) for x 4= £ are all less than/(£) (greater than /(£)), By a neighbourhood of a point we mean an interval o 5S x ^ /3 which contains the 160 DIFFERENTIATION AND INTEGRATION [Chap. point £ in its interior. Geometrically speaking, such maxima and minima are respectively the wave-crests and wave-troughs of the curve. A glance at fig. 11 shows us that the value of the maximum at one point P s may very well be less than the value of the minimum \ at another point; P 2 ; thus \^^-t4 the concept of maximum I | and minimum is always to x some extent relative, on Fig. it.— Maxima and minima account of the restriction to a certain neighbourhood. If we wish to fix upon the actual greatest or least value of the function we must employ special means for deciding how this value is to be selected from among the maxima or minima. The point for us at present is to find the (relative) maxima or minima, or, to use a word that covers both maxima and ininima, the relative extreme values (extrema) * of a given func- tion or curve. This problem, which is very frequently en- countered in geometry, mechanics, and physics and which occurs in many other applications, formed one of the principal incentives for the development of the differential and integral calculus in the seventeenth century. We see at once that if the function is assumed to be differen- tiable, the tangent to the curve at an extreme value £ must be horizontal. Hence the condition /'(£) = is a necessary condition for an extreme value; by solving this equation for the unknown f we obtain the points at which an extreme value may possibly occur. Our condition, however, is by no means a sufficient condition for an extreme value; there may be points at which the derivative vanishes, i.e. at which the tangent is horizontal, although the curve has neither a maxi- mum nor a minimum there. This occurs if at the given point the curve has a horizontal inflectional tangent cutting it, as in the above example of the function y = x 3 at the point x = 0. * The expressions turning value, turning point, are also used. On the other hand, the terms stationary value, stationary point, include inflections as well as maxima and minima. in J MAXIMA AND MINIMA 161 If, however, we have found a point at which f'(x) vanishes, we may immediately conclude that the function has a maximum at that point if /"(£) < 0, a minimum if/"(f) > 0. For in the first case the curve in the neighbourhood of this point lies com- pletely below the tangent, in the second case completely above the tangent. Instead of basing the deduction of our necessary condition on intuition we could, of course, have given an easy proof by ( purely analytical methods (cf. the exactly analogous considera- tions for Rolle's theorem, p. 105). If the function f{x) has a maximum at the point £, then for all sufficiently small values of h different from the expression /(f) -/(£ + h) must be posi- tive. Therefore the quotient ?~^ ( ^ will be positive or negative, according as A is negative or positive. Thus if h tends to zero through negative values the limit of this quotient cannot be negative, while if h tends to zero through positive values the limit cannot be positive. But since we have assumed that the derivative exists these two limits must be equal to one another and, in fact, to /'(£), which therefore can only have the value zero; we must have f'(£) = 0. A similar proof holds for the case of a mi nimum We can also formulate, and prove analytically, conditions which are necessary and sufficient for the occurrence of a maximum or a minimum, without involving the second derivative. We suppose that the function /(x) is continuous and has a continuous derivative f'(x) which vanishes only at a finite number of points. Then f (x) has a maximum or a minimum at the point x = gif, and only if, the derivative f'(x) changes sign on passing through this point; in particular, the function has a minimum if the deri- vative is negative to the left of £ and positive to the right, while in the contrary case it has a maximum. We prove this by using the mean value theorem. First, we observe that to the left and right of £ there exist intervals f i < x < £ and £ <x<£ 2 (extending to the nearest points at which f'(x) = 0) in each of which f'(x) has only one sign. If the signs of f'(x) in these two intervals are different, then f(f + A) — /(f) = A/'(f + 0h) has the same sign for all numeri- cally small values of h, whether h is positive or negative, so that 162 DIFFERENTIATION AND INTEGRATION [Chap. /(£) is an extreme value. Iif'(x) has the same sign in both in- tervals, then A/'(| + 6h) changes sign when A does, so that /(!+ A) is greater than/(|) on one side and less than/(£) on the other side, and there is no extreme value. Our theorem is thus proved. At the same time we see that the value f(i) is the greatest or least value of the function in every interval, containing the point f , in which the only change of sign of f'(x) occurs at | itself. The mean value theorem on which this proof is based can still be used even if f(x) is not differentiable at an end-point of the interval in which it is applied, provided that f(x) is differentiable at all the other points of the interval; for example, the above proof still holds if fix) does not exist at x = f . This leads us to the following more general result: if the function /(x) is continuous in an interval containing the point $, and everywhere in this interval, with the possible exception of | itself, has a derivative /'(a;) which vanishes at not more than a finite number of points, then/(x) has an extreme value at the point x = £ if, and only if, the point g separates two intervals in which f'(x) has different signs. For example, the function y = | x | has a minimum at x = 0, since y' > for x > and y' < for x < (cf. fig. 9, p. 97). The function y = tyx* likewise has a minimum at the point x = 0, even though its derivative § ar* is infinite there (cf. fig. 12, p. 99). In addition we make the following remark on the general theory of maxima and minima: the finding of maxima and minima, is not directly equivalent to the finding of the greatest and least values of a function in a closed interval. In the case of a monotonic function these greatest and least values will be assumed at the ends of the interval and are therefore not maxima and minima in our sense; for this latter ooncept refers to a complete neighbourhood of the place in question. Thus for ex- ample the function f(x) == x in the interval <£ x <| 1 assumes its greatest value at the point x s= 1, and its least value at x = 0, and a corresponding statement holds for every monotonic func- tion. The function y = arc tan x, whose derivative is 1/(1 + x % ), is monotonic for — oo < x < +» , and in that open interval possesses neither a maximum nor a minimum, nor a greatest or a least value. in J MAXIMA AND MINIMA ^63 If after finding the zeros of f\x) we wish to make sure that we have thereby found the points at which the function has its greatest or least values, we can often make use of the following criterion: A point i at which f(x) vanishes gives the least or greatest value of the function f(x) in a whole interval, if throughout that interval i"(x) > or f"(x) < respectively. For if I and £ + h both belong to the interval f'tf + h) =/'(£ + h) -/'(£) ^ hf"(g + eh), by the mean value theorem. Hence at the point x = $ + h the derivative/'^) has the same sign as h or the opposite sign, accord- ing as/"(x) > oxf"(x) < 0; the statement then follows from the remark following the theorem at the top of p. 162. 3. Examples 01 Maxima and Minima. Ex. 1. Of all rectangles of given area, to find that with the least peri- Let a* be the area of the rectangle and x the length of one side (here we must cons l der x as ranging over the interval < x < «,); then the length of the other side is a a /*, and half the perimeter is given by /(*) = *+?!. x We have f(x) = 1 - * /"(*) = 2o_ a . The equation /'(?) «, has the single positive root I = a. For this value / (x) is positive (as it is for any positive value of x); it therefore gives the required least value, and we obtain the very plausible result that of all rectangles of given area the square has the smallest perimeter. Ex. 2. Of all triangles with given base and given area, to find that with the least perimeter. To solve this problem, we take the *-axis along the given base AB and the middle point of AB as the origin. If C is the vertex of the triangle, ft its altitude (which is fixed), and (x, ft) are the co-ordinates of the vertex, then the sum of the two sides of the triangle AG and BG which are to be determined will be given by f(x) = V{(z + a)" + ft")} + V{(x - a)* + ft*} where 2a is the length of the base. From this we obtain V{(* + a) 2 + h>] ^V{( X - o)» + ft*}' 164 DIFFERENTIATION AND INTEGRATION [Chap. -(x+af 1 , -(x-af V{(* + a) 2 + A 2 } 3 1 + V{(x + a)* + h*} + V{(x - a) 2 + A 2 } 3 + V{(x- A 2 -a) 2 + A 2 } -. + ■ A 2 V {{x + af + ft 2 } 8 V {(x - af + A 2 } 8 We see at once (1) that /'(0) vanishes, (2) that f"(x) is always positive; hence at x = there is a least value. For since f"(x) > the first derivative f'(x) always increases and therefore cannot be equal to zero at any other point, so that the point x = must really give the least value of /(*). This least value is accordingly given by the isosceles triangle. Similarly, we find that of all triangles with given perimeter and given base the isosceles triangle has the greatest area. Ex. 3. To find a point on a given straight line such that the sum of its distances from two given fixed points is a minimum. Let there be given a straight line and two fixed points A and B on the same side of the line. We wish to find a point P on the straight line such that the distance PA + PB has the least possible value. We take the given line as the x-axis and use the notation of fig. 12. Then the distance in question is given by f(x)*= V(x* + A 2 ) + V{(x - a) 2 + h x *}, and we obtain t'(x\ = X 4- a—a ' V ' vV-f A 2 ) V{(x - af + A! 2 }' —3? , 1 , -(x—af Fig. 1 a. — Law of reflection /"(*) = y/(3? + A 2 ) 3 + + V(x* + A 2 ) 1 + V^x-af + h^f A 2 V{(x~- if + A x 2 } A, 2 VV + A 2 ) 3 V{(a; - o) 2 + A x 2 } 3 ' The equation/' (?) = accordingly gives us I _ _ a-l V{(5-a) 2 +A, 2 }* or V(5 2 + A 2 ) ' cosa= cosp, Ill] MAXIMA AND MINIMA i65 which means that the two lines PA and PB must form equal angles with the given line. The positive sign of f"(x) shows us that we really have a least value. The solution of this prohlem is olosely connected with the optical law of reflection. By an important principle of optics, known as Fermat's principle of least time, the path of a light ray is determined by the property that the time that the light takes to go from a point A to a point B under known conditions must be the least possible. If the condition is imposed that a ray of light shall on its way from A to B pass through some point on a given straight line (say on a mirror), we see that the shortest time will be taken along the ray for which the " angle of incidence " is equal to the " angle of reflection ". Ex. 4. The Law of Refraction.— -Let there be given two points A and B on opposite sides of the s-axis. Which path from A to B corresponds to the shortest possible time if the velocity on one side of the z-axis is c, and on the other side c 2 ? Fig. 13. — Law of refraction It is clear that this shortest path must lie along two portions of straight lines meeting one another at a point P on the z-axis. Using the notation of fig. 13, we obtain the two expressions V(A a + x*) and V{A,» + (o — x) 1 } for the lengths PA, PB respectively, and we find the time of passage along this path by dividing the lengths of the two segments by the corresponding velocities and adding. This gives us f(x) = 1 V(h* + x*) + I V {V + (a - *)*}, for the time taken. By differentiation, we obtain a— x c, V(#» + x») c 8 V {V + (a - x)*}' 1 A« .1 A,' , + i c, V(A* + x»)» ^ c 2 V{V + (<* -*)*}»' As we readily see from the figure, the equation f'(x) = 0, i.e. e, V(A" + *») e, V {V + (a - *)» }' 166 DIFFERENTIATION AND INTEGRATION [Chap. is equivalent to the condition — sin a = — sin (3, or sin a Cj sin P c 2 We leave it to the reader to prove that there is only one point which satisfies this condition and that this point actually yields the required least value. The physical meanii-g of our example is again given by the optical principle of least time. A ray of light travelling between two points describes the path of shortest time. If c, and c 2 are the velocities of light on either side of the boundary of two optical media, the path of the light will be that given by our result, which accordingly gives Snell's law of refraction. Examples 1. Find the maxima, minima, and points of inflection of the following functions. Graph them, and determine the regions of increase and de- crease, and of convexity and concavity: (a) a* — 6x + 2. (6) s 2/3 (l - x). (o) Zcftl + a 2 ). (d) x s /(x i + 1). (e) sin 2 *. 2. Determine the maxima, minima, and points of inflection of a? + $P X + ?• Discuss the nature of the roots of a? -f- Zpx + q = 0. 3. Which point of the hyperbola y 2 — %x* = 1 is nearest to the point x = 0, y = 3? 4. Let P be a fixed point with co-ordinates x , y in the first quadrant of a rectangular co-ordinate system. Find the equation of the line through P such that the length intercepted between the axes is a minimum. 5. A statue 12 ft. high stands on a pillar 15 ft. high. At what distance must a man 6 ft. high stand in order that the statue may subtend the greatest possible angle at his eye? 6. Two sources of light, of intensities a and 6, are at a distance d apart. At which point of the line joining them is the illumination least? (Assume that the illumination is proportional to the intensity and inversely pro- portional to the square of the distance.) 7. Of all rectangles with a given area, find (a) the one with the smallest perimeter; (6) the one with the shortest diagonal. x* v 2 8. In the ellipse - — (- — = 1 inscribe the rectangle of greatest area. o ! 6 2 9. Two sides of a triangle are a and 6. Determine the third side so that the area is a maximum. 10. A circle of radius r is divided into two segments by a line g at a distance h from the centre. In the smaller of these segments inscribe the rectangle of greatest possible area. 11. Of all circular cylinders with a given volume, find the one with the least area. Ill] MAXIMA AND MINIMA 167 12. Given the parabola y a = 2px, p > 0, and a point P(x = £, y = -n) within it (r? < 2p5), find the shortest path (consisting of two line seg- ments) leading from P to a point Q on the parabola and then to the focus F{x = \p,y = 0) of the parabola. Show that the angle FQP is bisected by the normal to the parabola, and that QP is parallel to the axis of the parabola. (Principle of the parabolic mirror.) 13.* A prism deflects a beam of light travelling in a plane perpendicular to the edge of the prism. What must the relative position of prism and beam be for the deflection to be a minimum? 14. Given n fixed numbers a lt . . . , a n , determine x so that S (a t — x)» is a minimum. < -1 15. Prove that if p > 1 and x > 0, x* — 1 |> p(x — 1). 16. Prove the inequality 1 ^ 5E? S - , ^ x <, n x 71 2 17. Prove that (a) tana: §i,0gj^-. * 2 (0) cosa: ;> 1 — — . - 2 18.* Given Oj > 0, a 2 > o„ > 0, determine the minimum of at + ■ . . + <!„_! + z -V^a^ . . . a n _,x for * > 0. Use the result to prove by mathematical induction that 1 + . . . + o. V^OiO, . «i 6. The Logarithm and the Exponential Function The systematic relations between the differential calculus and the integral calculus lead naturally to a convenient method of approach to the exponential function and the logarithm. Although we have already (pp. 25, 69) investigated these functions, we now define them afresh and develop their theory again without making any use of our previous definition and the results based on it. We begin with the logarithm, and then obtain the exponential function as its inverse. 1. Definition of the Logarithm. The Differentiation Formula. We have seen that indefinite integration of the power x n for integral indices n in general leads to a power of x. The only exception is the function 1/x, which does not appear as the 168 DIFFERENTIATION AND INTEGRATION [Chap. derivative of any of the functions which we have dealt with so far. It is natural to suppose that the indefinite integral of the function 1/x represents a new sort of function; so, following up this idea, we will proceed to investigate the function for x > 0. We call it the logarithm of x, or, more accurately, the natural logarithm of x, and write it y — log a; or y = nat log a;. We have denoted the variable of integration by £ in order to avoid confusion with the upper limit <c. The choice of the number 1 as lower limit is an arbitrary one, which, however, will soon prove its convenience. In the course of the following argument it will appear that the logarithm denned here is the same as the logarithm which we previously (p. 70) denned in an "elementary way". But, as we once more emphasize, the results of the following investigations are independent of those obtained earlier. Geometrically our logarithmic func- tion means the area shown shaded Fig. 14. — Log* illustrated by an area in fig. 14, which is bounded above by the rectangular hyperbola y = 1/5, below by the £-axis, and at the sides by the lines 5=1 and % = x. This area is to be reckoned positive if x > 1, negative if x < 1. For x = 1 the area vanishes, and we therefore have log 1 = 0. According to the above definition the derivative of the logarithm is given by the formula d(logx) _ 1 dx x Here let us expressly emphasize that we assume through- out that the argument x is positive; the logarithm of or of any negative value cannot be formed in accordance with the formula above, for the integrand 1/f becomes infinite when f = 0. On the other hand, if we choose some negative number, say —1, for the lower limit, we can form the integral ill] LOGARITHM AND EXPONENTIAL FUNCTION 169 with a negative upper limit x, i.e. we can consider the expres- sion r*dj (x < 0). Owing to the significance of the integral as the limit of a sum or as an area, we see that for x < f*d£ r-*d£ rMdg . . , In conformity with this we can in general write the formula for indefinite integration as /dx — = log I X [. The logarithm can, of course, be represented by means of a graph. This graph, the loga- rithmio curve, is shown in fig. 15. We have already seen (p. 119 et seq.) how to construct it. 2. The Addition Theorem. The logarithm defined as above obeys the following fundamental law: log (oft) = log a + log 6. yj 1 1 /I 2 3 * X Fig. 15 The proof of this addition theorem follows directly from the differentiation formula. For, writing z = log (ax), and applying the chain rule, we have dz 1 1 _ = _.a=-. dx ax x But d , 1 te logx== x' and since the functions z and logx have the same derivative they differ only by a constant, so that 2 = log a; -f cor log ox = log* + c. 170 DIFFERENTIATION AND INTEGRATION [Chap. This being true for all positive values of x, we first put x = 1 to find c; since logl = 0, this yields log a = c. Substituting this value for c, we have log ax = log a; + loga, whence, for x=b, logab = logos + log&, which was to be proved. For arbitrary positive numbers o^ o 2 , . . . , a n the equation logKa 2 . . • o n ) = logo! + loga 2 + . . . -f logo„ follows from the addition theorem for the logarithm. In particular, if all the numbers a l5 a 2 , . . . , a„ are equal to one and the same number a, we have loga" = n logo. Similarly, it follows that logo + log- = logl = 0, a so that logo = — log -. a If, further, we put \/a = a it follows that logo = n log a, or log-y/a = logo 1 '" = - logo. n From this by repeated use of the addition theorem we find that, when m is a positive integer, - logo = log a/o™ = logo™'". n The equation logo r = r logo is thus proved for all positive rational values of r , and for r = it is obviously correct. For negative rational values of r it is also valid, for then logo r = log-—- = — logo -r = r logo. Ill] LOGARITHM AND EXPONENTIAL FUNCTION 171 3. Monotonic Character and Values of the Logarithm. The value of the logarithm obviously increases when x in- creases, and decreases when x decreases; the logarithm is there- fore a monotonic function. Since the derivative 1/x becomes smaller and smaller as x increases, the function increases more and more slowly as x increases. Nevertheless, as x increases beyond all bounds the function log a: does not tend to a positive limit, but becomes infinite; that is to say, for every positive number A, no matter how large, there are values of x for which logs > A. This fact follows very readily from the addition theorem. For log 2" == rc log2, and since log 2 is a positive number, by taking x = 2" with sufficiently large values of n we can make logo; as large as we please. Since log(l/2«) = — n log 2, we see that as x tends to zero through positive values loga: is negative and increases numeri- cally beyond all bounds. Summing up these results: The function loga; is a monotonic function which assumes all values between -00 and +00 as the independent variable x ranges over the oontinuum of positive numbers. 4. The Inverse Function of the Logarithm (the Exponential Function). Since the function y = loga; (x > 0) is a monotonic function of x which assumes all real values, its inverse function, which we shall at first denote by x = E(y), must be a single-valued mono- tome function denned for every real value of y; it is difEerentiable, since loga; itself is differentiable. We interchange the notation for the dependent and independent variables, and proceed to study the function E(x) in detail. In the first place, it must clearly be positive for every value of x. Further, we must have E(0)=1; for this equation is equivalent to the statement that log 1 = 0. Secondly, from the addition theorem for the logarithm there immediately follows the multiplication theorem E(a)E(p)=E(a+p). 172 DIFFERENTIATION AND INTEGRATION [Chap. To prove this we need only notice that the equations E{a) = a, E(P)=b, E{a+P) = c are equivalent to o = loga, /? = logfc, a + fi = logc. Since by the addition theorem for the logarithm a + j3 = logafc, it must be true that c = ab, which proves the multiplication theorem. From this theorem we derive a fundamental property of the function y = E(x), which gives us the right to call our function the exponential function and to write it symbolically in the form y=e*. In order to obtain this property we observe that there must be a number — which we shall call * e — for which loge = 1. This is equivalent to the definition E(l) = e. Using the multiplication theorem for the function E(x), we have E(n) = e n , and, in the same way, for positive integers m and n, *(?) = ** which we could also have found directly from the addition theorem for the logarithm. The equation E(r) = e r thus proved for positive rational numbers r holds also for negative rational numbers in virtue of the equation E{r)E(—r) = E(0) = 1. The function E{x) is therefore a function which is continuous for all values of x, and which for rational values of x coincides with e x . These facts give us the right to call our function e" for * Its identity with the number e considered on p. 43 will be proved in No. 6 (p. 175). Ill] LOGARITHM AND EXPONENTIAL FUNCTION 173 arbitrary irrational values of x also.* (It should be noticed that here the continuity of e* is an immediate consequence of its definition as the inverse function of a continuous monotonic function, while if the elementary definition is adopted the con- tinuity must be proved.) The exponential function is differentiated according to the formula d „* * — e" = e* or y = y. This formula expresses the important fact that the derivative of the exponential function is the function itself. The proof is extremely simple. For we have x = logy, whence, by the formula for the differentiation of the loga- rithm, we have — = -, dy y and then by the rule for inverse functions dy dx y=e*, as was stated. Fig. 16. — The exponential function The graph of the exponential function &, the so-called exponential curve, is obtained by reflection of the logarithmic curve in the line which bisects the first quadrant. It is shown in fig. 16. 5. The General Exponential Function a" and the General Power x". The exponential function a" for an arbitrary positive base a is now simply defined by the equation y=a*= e* l08a , * If we anticipate the fact, which will be proved on p. 175, that our number e is identical with the number so denoted previously, we have now proved that the definition given here yields the same exponential function with base « as was formerly defined by the process of raising to powers. For, according to that elementary definition, we defined the values of e x for irrational z's as the limit of the expressions e*n, where x n takes on a sequence of rational values with the limit x. 174 DIFFERENTIATION AND INTEGRATION [Chap which agrees with the earlier definition in virtue of the relation e loga ==a. Using the chain rule we immediately obtain — a* = — e* loga = e* 1<>Ba . loga, dx dx = a" loga. The inverse function of the exponential function y = a" is called the logarithm to the base a and is written x=log a y, while the logarithmic function previously introduced, when a distinction is necessary, is spoken of as the natural logarithm, or logarithm to the base e. From the definition it follows immediately that logy = x loga = log a i/ . loga, which shows us that the logarithm of y to an arbitrary positive base a 4= 1 is obtained by multiplying the natural logarithm of y by the reciprocal of the natural logarithm of a, the modulus of the system of logarithms to the base * a. Instead of our previous definition of the general power x*(x > 0) we shall now define this power by the equation x* =e° log *. The rule for differentiating the power af follows immediately from the definition, using the chain rule; for Lx°- = e alosx .-=ax a - 1 , dx x in agreement with our previous result (cf. p. 155). * If we take = 10 we obtain the ordinary " Briggian " logarithms, which have already been met with in elementary mathematics and which are advan- tageous for use in numerical calculations. Ill] LOGARITHM AND EXPONENTIAL FUNCTION 175 6. The Exponential Function and the Logarithm represented as Limits. We are now in a position to state important limiting relations referring to the quantities introduced above. We begin with the formula for differentiating the function f(x) = log 3, -=/'(*)= k™ /<* + *> -/(*) - 1^ l°g(« + ft)-log» x a-^o A A _^ h = limllog(l + ^). A->Oft \ X/ If we put - = 2, this becomes x lim - log(l + zh) = 2 . A->qA Since the function ef is continuous for all values of x, this implies that e*=Iime' ,0 ^+*WA'=li m (l + ^)i/A > . # (o) If in particular we give h the sequence of values 1, -, -, , , ., I, . . ., we have 2 3 n lim ( 1 + B = e " W If to z we assign the value 1, formula (a) gives the following important fact: As h tends to zero, the expression (1 + h) 1/h tends to the number e: lim (1 + hj Llh = e. A->0 Formula (6) gives lim (l + *Y = e, which proves that the number e is the same as the number denoted by the symbol e on p. 43. From the differentiation formula for a x , a x log a = lim — ~» *->q h 176 DIFFERENTIATION AND INTEGRATION [Chap. it follows for x = that i ,. «» - 1 loga = iim — - — > a-*-o h a formula which expresses the logarithm of a directly as a limit. To this equation we append the remark that by its means we can complete the relation [" x «dx = -i— (6* +1 -a* +1 ) J a a + 1 established earlier. We have always been obliged to exclude the case a = — 1. Now, however, we can trace what happens when the number a tends to the limit —1. If we put a= 1 the left-hand side will by our definition of the logarithm have the limit * t dx , , — = log 6; 1 x the right-hand side therefore has the same limit when o -»• — 1. This fact, moreover, is in accordance with the formula b h — 1 log6=hm— _— ; A-»o n we need only write a + 1 = h. We have thus cleared up the exceptional case a = — 1 in the integration formula which we have so often used. The formula above is still meaningless when a = — 1, but as a limit formula it retains its significance as a -> — 1. 7. Final Remarks. Here we briefly review the train of thought followed out in this section. We first defined the natural logarithm y = loga; for x > by means of an integral, whence we immediately deduced the differentiation formula, the addition theorem, and the existence of an inverse. We then investigated the inverse function y = ^", where the number e was seen to be the number * We have here carried out the passage to the limit a — »• — 1 under the integral sign without further investigation; cf. the discussion on p. 128 et seq. Ill] LOGARITHM AND EXPONENTIAL FUNCTION 177 whose logarithm is 1, and we derived its differentiation formula, as well as limit expressions for it and for the logarithm. The introduction of the functions y = of = e" log * and y=a x =e x lo « a followed naturally. In the discussion given here, as contrasted with the " elemen- tary " treatment, the question of continuity causes no difficulty, since the logarithm is defined as an integral and therefore as a continuous and differentiable function, whose inverse function is also continuous. Examples 1. Sketch the function y = - (1 ^ x ^2) on a large scale, using x graph paper, and find log, 2 by counting squares. Differentiate the functions in Ex. 2-6: 2. x(logx — 1). 4. log{x + V(l + x 2 )}. 3. log logx. 5. log{ V(l + logx) — sinx). i/(x 2 + 1) 6. Differentiate log — y — — -; (a) by using the chain rule and the V (2 + x) quotient rule, without preliminary simplification; (6) first simplifying by means of the theorems on logarithms. 7. (a) Differentiate y = , ^^ + 1} V(x-2)V(x*+l) (b) Differentiate the same function, first taking logarithms and sim- plifying. / x\ n 8.* Given lim s„ = 0, prove that lim ( 1 + e„ . - 1 =1. n— >-«> n— >«A W' 9. Show that the function y = er aX (a cosx-|- 6 sinx) satisfies the equation y" + 2<xy' + (a a + l)y=0 for all values of a and 6. 10.* Show that — («- 1 ' x *) = P "J a,) e - 1/l ', when x =t= 0, where PJx) is dx n x 3 " a polynomial of degree 2n — 2. Establish the " recurrence formula " P n+ A*) = (2 - 3nx*) P n (x) + x»P n '(x). 11. Find the maximum of y — x\ a e~ **, where X and a are constants. Find the locus of this maximum when X is allowed to vary. 12. Differentiate aM (a > 0). 13. Differentiate a****™*'. 7 1*798) 178 DIFFERENTIATION AND INTEGRATION [Chap. 7. Some Applications of the Exponential Function In this section we shall consider some miscellaneous problems involving the exponential function, and we shall thus gain an insight into the fundamental importance of this function in all sorts of applications. 1. Definition of the Exponential Function by Means of a Dif- ferential Equation. We can define the exponential function by a simple theorem, whose use will save us many detailed investigations of particular cases. If a function y = f (x) satisfies an equation of the form y'=ay where a is a constant other than zero, then y has the form y=f(x) = ce**, where c is also a constant; and conversely, every function of the form ce°* satisfies the equation y' = ay. The latter is usually briefly referred to as a differential equation, since it expresses a relation between the function and its derivative. In order to make the theorem clear, we notice first of all that in the simplest case a = 1 the above equation becomes y' = y. We know that y=e x satisfies this equation, and it is clear that the same is also true of y = ce x , if c is an arbitrary constant. Conversely, we can easily see that no other function satisfies the differential equation. For if y is such a function, we consider the function u = ye~*. We must then have u' = y'e~* — ye'" = e-%' — y). But the right-hand side vanishes, since we have assumed that y' = y; hence u' = 0, so that by p. 114 et seq. u is a constant c and y = ce x , as we wished to prove. The case of any non-zero value of a can be treated in exactly the same way as the special case a = 1. If we introduce the function u = ye~°*, we obtain the equation u' = y'e~° x — aye~"*- Hence from the assumed differential equation we find that u' = 0, so that « = c and y = ce°*. The converse is clear. Ill] THE EXPONENTIAL FUNCTION 179 We will now apply this theorem to a number of examples and thus make it more intelligible. 2. Interest Compounded Continuously. Radioactive Disintegra- tion. A capital sum, or principal, which has its interest added to it at regular periods of time, increases by jumps at these interest periods in the following manner. If 100a is the rate of interest per cent, and if further the interest accrued is added to the principal at the end of each year, then after x years the accumulated amount of an original principal of 1 will be (1 + a)«. If, however, the principal had the interest added to it not at the end of each year, but at the end of each »-th part of a year, then after x years the principal would amount to R)' Taking x = 1 for the sake of simplicity, i.e. reckoning the interest at 100a per cent for one year, we find that if the interest is computed in this latter way the principal I amounts after one year to (-!)"■ If we now let n increase beyond all bounds, i.e. if we let the interest be calculated at shorter and shorter intervals, the limiting case will signify in a sense that the interest is compounded continuously, at each instant; and we see that the total amount after one year will be e a times the original principal. Similarly, if the interest is calculated in this manner, an original principal of 1 will have grown after x years to an amount e"*; here x may be any number, integral or otherwise. The discussion in No. 1 (p. 178) forms a framework within which examples of this type are readily understood. We consider a quantity, given by the number y, which increases (or decreases) with the time. Let the rate at which this quantity increases or decreases be proportional to the total quantity. Then if we take the time as the independent variable x, we obtain a law of the form y' = ay for the rate of increase, where a, the factor of proportionality, is positive or negative according as the quantity is increasing or decreasing. Then in accordance with No. 1 the quantity y itself will be given by a formula y = ce« x , where the meaning of the constant is immediately obvious if we con- sider the instant x = 0. At that instant e"* = 1, and we find that c = y is the quantity at the beginning of the time considered, so that we may write 180 DIFFERENTIATION AND INTEGRATION [Chap. A characteristic example of the use of these ideas is the case of radio- active disintegration. The rate at which the total quantity y of the radio- active substance is diminishing at any instant is proportional to the total quantity present at tha* instant; this is a priori plausible, as each portion of the substance decreases as rapidly as every other portion. Therefore the quantity y of the substance expressed as a function of the time satis- fies a relation of the form y" = — ky, where k is to be taken as positive since we are dealing with a diminishing quantity. The quantity of sub- stance is thus expressed as a function of the time by y = y e~ ka , where y is the amount of the substance at the beginning of the time considered (time x = 0). After a certain time t the radioactive substance will have diminished to half its original quantity. This so-called half-value period is given by the equation to. = ifor*. log 2 whence we immediately obtain t = -f— . k 3. Cooling or Heating of a Body by a Surrounding Medium. Another typical example of the occurrence of the exponential function is offered by the cooling of a body, e.g. a metal plate, which is immersed in a very large bath of given temperature. In considering this cooling we assume that the surrounding bath is so large that its temperature ia un- affected by the cooling process. We further assume that at each instant all parts of the immersed body are at the same temperature, and that the rate at which the temperature changes is proportional to the difference between the temperature of the body and that of the surrounding medium (Newton's law of cooling). If we denote the time by x and the temperature difference by y = y(x), this law of cooling is expressed by the equation y' = — ley, where k is a positive constant whose value depends on the body itself. From this instantaneous relationship, which expresses the effect of the cooling process at a given instant, we now wish to derive an " integral law " which will allow us to find the temperature at an arbitrary time x from the temperature at an initial time x = 0. The theorem of No. 1 (p. 178) immediately gives us this integral law in the form y = ce _fel! , where k is the above-mentioned constant depending on the body. This shows that the temperature decreases " exponentially " and tends to become equal to the external temperature. The rapidity with which this happens is expressed by the number k. As before, we find the meaning of the constant c by considering the instant x = 0; this gives us y a = e, so that our law of cooling can finally be written in the form y = y^r hx . Ill] THE EXPONENTIAL FUNCTION 181 It is obvious that the same discussion will also apply to the heating of a body. The only difference is that the initial difference of temperature y„ is in this case negative instead of positive. 4. Variation of the Atmospheric Pressure with the Height above the Surface of the Earth. As a further example of the occurrence of the exponential formula we shall deduce the law according to which the atmospheric pressure varies with height. We here make use (1) of the physical fact that the atmospheric pressure is equal to the weight of the column of air vertically above a surface of area 1, and (2) of Boyle's law, according to which the pressure of the air (p) at a given constant temperature is proportional to the density of the air (a). Boyle's law, expressed in symbols, is p = aa, where a is a constant which depends on a specific physical property of the air and in addition is proportional to the absolute temperature — here we are not concerned with this, as we shall assume that the temperature is constant. Our problem is to determine p = /(A) as a function of the height (A) above the surface of the earth. If by p we denote the atmospheric pressure at the surface of the earth, i.e. the total weight of the air column supported by a unit area, and by a (X) the density of the air at the height X above the earth, the weight of the f* column up to the height h will be given by the integral / a(X)<2X. The pressure at height h will therefore be ° p = f(h) = p - [ o(X)iX. Jo By differentiation this yields the following relation between the pressure p = f(h) and the density <j(h): °(A) = -/W = -¥'■ We now use Boyle's law to eliminate the quantity a from this equation, thus obtaining an equation 1 p'= — -p a which involves the unknown pressure function only. From p. 178 it follows that V = /(*) = ce _A/a . If as above we denote the pressure at the earth's surface, i.e. /(0), by p w it follows immediately that o = p , and consequently p = f(h) = j» e-*/«. Changing to logarithms, we obtain A=alog?' . P i8a DIFFERENTIATION AND INTEGRATION [Chap. These two formulae find frequent application. For example, if the oonstant a is known they enable us to find the height of a place from the barometric pressure, or to find the difference in height of two places by measuring the atmospheric pressure at each place. Again, if the atmospheric pressure and the height h are known we can determine the constant a, which is of great importance in gas theory. 5. Advance of a Chemical Reaction. We now consider an example from chemistry, namely, the so-called unimolecidar reaction. We suppose that a substance is dissolved in a rela- tively large amount of solvent, say a quantity of cane sugar in water. If a chemical reaction takes place, the chemical law of mass action in this simple case states that the rate of reaction is proportional to the quantity of reacting substance present. If we suppose that the cane sugar is being transformed by catalytic action into invert sugar, and if by u(x) we denote the quantity of cane sugar which at time x is still unchanged, the velocity of reaction will be — du/dx, and in accordance with the law of mass action an equation of the form du , -- = — ku Ax holds, where k is a constant depending on the substance reacting. From this instantaneous law we immediately obtain, as on p. 178, an integral law, which gives us the amount of cane sugar as a function of the time: u(x) = ae~ kx . This formula clearly shows us how the chemical reaction tends asymptoti- cally to its final state u = 0, that is, complete transformation of the re- acting substance. The constant a is obviously the quantity present at time x = 0. 6. Making and Breaking an Electric Circuit. As a final example we consider the growth of a (direct) electric current when a circuit is completed (or its decay when the circuit is broken). If R is the resistance of the circuit and E the impressed electromotive force (voltage), the current / will gradually increase from its original value to the steady final value E/JR. We have therefore to consider / as a func- tion of the time. The growth of the ourrent depends on the self-induction of the circuit; the circuit has a characteristic constant L, the coefficient of self-induction, of such a nature that as the ourrent increases an electro- motive force of magnitude Ldljdx, opposed to the external electromotive force E, is developed. From Ohm's law, according to which the product of the resistance and the current is at each instant equal to the actual effeotive voltage, we obtain the relation IR=E-L A 1. ax Ill] THE EXPONENTIAL FUNCTION 183 Here we put we immediately find that f'(x) w — -- f(x), so that by the theorem Li on p. 178 /(a;) = /(0)e-**/£ Recalling that 2(0) = 0, we see that /(0) «i — — , and thus we obtain the expression it '-/(*) + §- fa -•-**> for the current as a function of the time. IVom this expression we see that when the circuit is closed the current tends asymptotically to its steady value E/B. Examples 1. The function f(x) satisfies the equation /(* + 2/) =/(*)/&). (o) If f(x) is differentiable, either f(x) = or else /(a;) = e°*. (6)* If f(x) is continuous, either /(*) s or else f(x) = e". 2. If a differentiable function f(x) satisfies the equation then f(x) = a logs. 3. A quantity of radium weighs 1 gm. at time t = 0. At time t = 10 (years) it has diminished to -997 gm. After what time will it have diminished to -5 gm.? 4. Solve the following differential equations: (a) y' = a(y — 0). (e) y' — a y = p«« (6) y' — a.y— p. (d) y' _ ay = p e i*. 8. The Hyperbolic Functions 1. Analytical Definition. In many applications the exponential function does not enter alone, but in combinations of the form l(e» + <r«) or |(e*-<r«). It is convenient to introduce these and similar combinations as special functions; we denote them as follows: 184 DIFFERENTIATION AND INTEGRATION [Chap. sinha; = cosh a: = e x _|_ e -x tanha;: e x -\- e~ x ' cotha: = e x + er» and we call them the hyperbolic sine, hyperbolic cosine, hyper- bolic tangent, and hyperbolic cotan- gent respectively. The functions sinha;, cosh a;, and tanha; are de- fined for all values of x, while in the case of cotha; the point x = must be excluded. This notation is designed to express a certain analogy with the trigonometric functions; it is this analogy, which we are about to study in detail, that justifies special consideration of our new functions. In figs. 17, 18, and 19 the graphs of the hy- perbolic functions are shown; the dotted lines in fig. 17 are the graphs of y = \e x and y = \e~ x , from which the graphs of sinha; and cosh a; may easily be constructed. Fig. 17 Fig. 1 8 We see that cosh a; is an even function, i.e. a function which remains unchanged when x is replaced by — x, while sinha; is an odd function, i.e. a function that changes sign when x is replaced by -x. (Cf. p. 20.) Ill] THE HYPERBOLIC FUNCTIONS 185 The function cosha; = ! 2 v-coth: -2 X is, by its definition, positive for all values of x. It has its least value when x = 0; coshO = 1. Between cosh a; and sinha; there exists the fundamental re- lation cosh 2 a; — sinh 2 a; = 1, which follows immediately from the definitions of these functions. If we now denote the independent variable by t instead of x and write x = cosh*, y = sinhf, we have x 2 — f = 1; that is, the point with the co- ordinates x = cosht, y = waht moves along the rectangular hyperbola x 2 — y 2 = 1 as t runs through the whole scale of values from — 00 to +00. According to the defining equation, x ^ 1, and we may easily con- vince ourselves that y runs through the whole scale of values — 00 to + co as t does; for if t tends to infinity so does e', while e~* tends to zero. We may therefore state more exactly that as t runs from — 00 to +00 , the equations x = cosh(, y = sinhtf give us one branch, namely, the right-hand one, of the rectangular hyperbola. 2. Addition Theorems and Formulae for Differentiation. From the definitions of our functions there follow the formulae, known as addition theorems: cosh (a -f 6) = cosha cosh 6 -f sinha sinh&, sinh(a + b) = sinha coshft + cosha sinh&. The proofs are obtained at once if we write e°e* + e~"e~ h Fig. 19 cosh (oft): 7» -, sinh(a + 6): e°e» e~ a e~ 2 (1798) 186 DIFFERENTIATION AND INTEGRATION [Chap. and in these equations put e° = cosha -f- sinha, e~" = cosha — sinha, e" = coshJ + sinhi, e~ 6 = cosh& — sinhi. The analogy between these formulae and the corresponding trigonometrical formulae is clear. The only difference in the addition theorems is one sign in the first formula. A corresponding analogy holds for the differentiation for- mulae. Eemembering that d(e x )ldx = e*, we readily find that * — cosh a; = sinha;, — sinha; = cosha;, ax dx d * v. 1 d ^ 1 — tannaj = — — -, — - cotha; = — . ax cosh 2 a; dx smh 2 aj 3. The Inverse Hyperbolic Functions. To the hyperbolic functions a;=cosh«, y=sinh«, there correspond inverse functions, which we denote f by t = ar coshx, t = ar sinhy. Since the function sinhi is monotonic increasing throughout the interval — oo < t < -f- oo , its inverse function is uniquely de- termined for all values of y; on the other hand, we learn from a glance at the graph (cf. fig. 17, p. 184) that t = ar cosha; is not uniquely determined, but has an ambiguity of sign, for to a given value of x corresponds not only the number t but also the number —t. Since coshZ ^ 1 for all values of t, its inverse ar cosha; is defined only for x ^ 1. We can express these inverse functions very easily in terms of the logarithm, by regarding the quantity e* = u in the de- finitions e* + e~* e* — er* 2 " 2 as unknown and solving these (quadratic) equations for u. Then «-=*±V(a^-l), «=y+V(y 2 +l); since w = e* can have only positive values the square root in *It is sometimes convenient to introduce the functions secha; = 1/cosha:, cosechz = 1/sinha;. t The notation cosh- 1 *, &c, is also used; cf. footnote, p. 148. Ill] THE HYPERBOLIC FUNCTIONS 187 the second equation must be taken with the positive sign, while in the first either sign is possible. In the logarithmic form, t = log(a; +-\/(a? — 1)) = ar coshaj, t = \ g(y -f -v/(y 2 + 1)) = ar sinhy. In the case of ar cosh a; the variable x is restricted to the interval x ^ 1, while ar sinhy is defined for all values of y. The formula gives us two values, log {a; + */(x* — 1)} and log {a; — VO" 2 — !)}> f° r arcosha;, corresponding to the two branches of ar cosh a;. Since {x + V(^ - 1)} {x - V(^ - 1)} = 1 the sum of these two values of ar cosh a; is zero, which agrees with a remark made above. The inverses of the hyperbolic tangent and hyperbolic co- tangent can be defined analogously, and can also be expressed in terms of logarithms. These functions we denote by ar tanha; and ar cotha;; and, expressing the independent variable every- where by x, we readily obtain 1 1 + x ar tanha; = - log - in the interval — 1 < x < 1, 2 1 — a; 1 x4- 1 . ar cotha; = - log in the intervals x < — 1, x > 1. 2 x — 1 The differentiation of these inverse functions may be carried out by the reader himself; here he may make use of either the rule for differentiating an inverse function or the chain rule in conjunction with the above expressions for the inverse functions in terms of logarithms. If a; is the independent variable, the results are d , , 1 d . . i — ar cosha; = + — — , — arsinha;: dx - v^z 2 — 1)' dx Vfx 2 + 1)' — - ar tanha; = -, — ar cotha; = dx 1 — x 2 ' dx 1 — x 2 The last two formulae do not contradict each other, since the first holds only for — 1 < x < 1 and the second only for x < — 1 and 1 < x. The two values of -j- ar coshx, expressed by the dx 188 DIFFERENTIATION AND INTEGRATION TChap. sig 11 i i* 1 * ne first formula, correspond to the two different branches of the curve y = ar cosh a; = log {a; +-\/(x 2 — 1)}. 4. Further Analogies. In the above representation of the rectangular hyperbola by the quan- tity t we did not attempt to bring out any geometrical meaning of the "parameter" t itself. We shall now return to this matter, thus gaining still more insight into the analogy between the trigonometric functions and the hyperbolic functions. If we represent the circle with equation a? -\- y a = 1 by means of a parameter t in the form x = cost, y = sint, we can interpret the quantity t as an angle or as a length of arc measured along the cir- cumference; we may, however, also regard t as twice the area of the cir- cular sector corresponding to that angle, the area being reckoned positive or negative according as the angle is positive or negative. y\ , *?-f->/ ■t-cotkt-^4 4j MM '1 4-COSflt \ H *x Fig. ao. — Parametric representation of the hyperbola Fig. 21. — To illustrate the hyperbolic functions We now make the analogous statement that for the hyperbolic func- tions the quantity t is twice the area oi the hyperbolic sector* shown shaded in fig. 20. The proof is obtained without difficulty if we refer the hyperbola to its asymptotes as axes by means of the transformation of co-ordinates x—y=V2l, a;+y=V'2Yi, or with these new co-ordinates the equation of the hyperbola is %t\ = £. We thus see immediately that the area in question is equal to the area of the figure ABQP; for the two right-angled triangles OPQ and OAB * Just as the notation I = arc cos a; recalls that t is an are of a circle of reference, so t = arcoshz recalls that ( is a certain area connected with a rectangular hyperbola. III1 THE HYPERBOLIC FUNCTIONS 189 have the same area, according to the equation of the hyperbola. The two points A and P obviously have the co-ordinates V2 V2 V2 ' V2 respectively, and for double the area of our figure we thus obtain f U+y)lV2 2 / (l/2r)) dyj = log(a: + y) = log { x ± V(x» - 1)}. •'l/v'2 Comparison of this expression with the formula of p. 187 for the inverse function <= arcosha; shows us that our statement about the quantity t is true. In conclusion, it may be pointed out that, as shown in fig. 21, the hyperbolic functions can be diagrammatically represented on the hyper- bola, just as the trigonometric functions can be represented on the circle.* Examples 1. Prove the formula sinha + sinhft = 2 sinh( a+6 ) cosh( a ~ 6 ). Obtain similar formulae for sinha — sinhft, cosha + cosh 6, cosh a — cosh 6. 2. Express tanh(a ± 6) in terms of tanha and tanhfi. Express coth(a ± 6) in terms of cotha and cotkfc. Express sinhfa and cosh Ja in terms of cosha. 3. Differentiate (a) cosha; + sinha:; (6) e tanh*+coth*. ^ logg^},^ + cosh'x); {d) ar cosha: -f ar sinha;; (e) arsinh(acoshx); (/)artanh : 1-fa: 2 4. Calculate the area bounded by the catenary y = cosha;, the ordinates x — a and x = b, and the z-axis. 9. The Obdee of Magnitude of Functions The various functions that we have met in this chapter exhibit very important differences as regards their behaviour for large values of the argument or, as we also say, in the order * The numerical values of the hyperbolic functions, which are useful in a variety of calculations, are to be found in many tables. We may mention the following: J. B. Dale, Five-figure Tables of Mathematical Functions (Arnold, 1918); K. Hayashi, FUnfstellige Tafeln der Kreis- und Byperbelfunktionen (Berlin, 1930); E. Jahnke and F. Emde, FunUionentafeln mit Formeln und Kurven (German and English, Leipzig, 1933). 190 DIFFERENTIATION AND INTEGRATION [Chap. of magnitude of their increase. On account of the great importance of this we shall here discuss the matter briefly, even although it is not directly connected with the idea of the integral or of the derivative. 1. The Concept of Order of Magnitude. The Simplest Cases. If the variable x increases beyond all bounds, then when a > the functions x*, log a;, e x , e°* will also increase beyond all bounds. As regards the manner of this increase, however, we can immediately point out an essential difference between them. For example, the function x 3 will become infinite to a higher order than a; 2 ; we mean thereby that as x increases the quotient aP/x* itself increases beyond all bounds. Similarly, we shall say that the function af becomes infinite to a higher order than a^ if a > /? > 0, and so on. Quite generally, we shall say of two functions /(a;) and g(x) whose absolute values increfie with x beyond all bounds that f(x) becomes infinite of a higher order than g(x), if as x increases the quotient 9(x) increases beyond all bounds; we shall say that f(x) becomes infinite of a lower order than g{x) if the quotient 9( x ) tends to zero as x increases; and we shall say that the two functions become infinite of the same order of magnitude if as a; increases the quotient /(*) 9( x ) possesses a limit different from or at least remains between two fixed positive bounds. For example, the function ax 3 -\- bx 2 + c =f(x), where a 4= 0, will be of the same order of magnitude as the function 9( x ) as? -\-bx 2 + c ~aT has the the function x 3 + x -\- 1 becomes x z = g(x); for the quotient limit \a\. On the other hanc infinite of a higher order of magnitude than the function x*+z+l. A sum of two functions f(x) and <f> (x), where f(x) is of higher order of magnitude than <f> (x), has the same order of magnitude as f{x). For f(x)+<f,(x) /(*) and by hypothesis this expression tends to 1 as a: increases. Ill] ORDER OF MAGNITUDE 191 We might be tempted to measure the order of magnitude of functions by a scale, assigning to the quantity x the order of magnitude 1 and to the power x* (a > 0) the order of magnitude a. A polynomial of the n-th degree then obviously has the order of magnitude n; a rational function, the degree of whose numerator is higher by h than that of the denominator, has the order of magnitude h. 2. The Order of Magnitude of the Exponential Function and of the Logarithm. It turns out, however, that any attempt to fix the order of magnitude of arbitrary functions by the above scale must end in failure. For there are functions that become infinite of higher order than the power of of x, no matter how large o is chosen; again, there are functions which become infinite of lower order than the power x*, no matter how small the positive number a is chosen. These functions therefore will not fit in anywhere in our scale. Without entering into a detailed theory of the order of magni- tude we shall prove the following theorem: If & is an, arbitrary number greater than 1, then the quotient a x — tends to infinity as x increases. To prove this we construct the function qX (f>{x) — log — = x logo — logo; x it is obviously sufficient to show that this increases beyond all bounds if as tends to -f°°- For this purpose we consider the derivative <f>'(x) = loga — - x 2 and notice that for x ^ = - — - this is not less than the 1 g0 positive number - loga. Hence it follows that f or x ^ <l>(x) — <f>(c) =f<f>'(t)dt ^J\ logo dt ^ £(3 — c) loga, Hz) ^<f>(c) + %(x — c) loga, and the right-hand side becomes infinite as x increases. 192 DIFFERENTIATION AND INTEGRATION [Chap. We shall give a second proof of this important theorem. If we write \/a = b = 1 + h, we have b > 1 and h > 0. Let n be the integer such that n ^ a; < n + 1; we may take a; > 1, so that n ^ 1. Applying the lemma of p. 31, we have l/a x \_ 6* (1+A)* (1 + A)» 1 + wA ^ wA h , so that V* V^+i) V( n + l ) V2n V 2 a" A 2 x > 2- W ' and therefore tends to infinity with x. From the fact just proved many others follow. For example, for every positive index a and every number o > 1 the quotient a'/x* tends to infinity as x increases; that is: The exponential function becomes infinite of a higher order of magnitude than any power of x. In order to see this, we need only show that the a-th root of the expression, that is, a*'° _ 1 a"i a _ 1 a* / _ A x a x/a ay \ a/' tends to infinity. This, however, follows immediately from the preceding theorem, when x is replaced by y = x/a. We can prove the following theorem in a similar fashion. For every positive value of a the quotient (logajj/x" tends to zero when x tends to infinity; that is: The logarithm becomes infinite of a lower order of magnitude than any arbitrarily small positive power of x. The proof follows immediately if we put log a; = y, by which our quotient is transformed into y/e ay . We then put e* = a; then a is a number > 1, and our quotient y/a y approaches as y increases. Since y approaches infinity as x does our theorem is proved.* * Another very simple proof may be suggested: for x > 1 and e > log* -J"^ <f*e- 1 d£ -£(*•- 1); if we choose e smaller than a and divide both members of this inequality by x", then as x — *■ oo it follows that (logx)jx°- -> 0. Ill] ORDER OF MAGNITUDE 193 With these results as a basis we can construct functions of an order of magnitude far higher than that of the exponential function and other functions of an order of magnitude far lower than that of the logarithm. For example, the function e** is of a higher order than the exponential function, and the function log log a; of a lower order than the logarithm; and we can obvi- ously repeat these iteration processes as often as we like, piling up the symbols e or log to any extent we please. 3. General Remarks. These considerations show that it is not possible by systematic reasoning to assign to all functions definite numbers as orders of magnitude in such a way that when two functions are com- pared the function of the higher order of magnitude has the higher number. If, for example, the function a; is of the order of magnitude 1 and the function a^+* of the order of magnitude 1 + e, then the function x logic must be of an order of magnitude that is greater than 1 and less than 1 + e no matter how small e is chosen. But there is no such number. Apart from this, however, it is easy to see that functions need not possess a clearly denned order of magnitude. For example, the function a^sina;) 2 -f x + I ,„.,.. ~- — approaches no definite limits as x increase; on ar(cosa;) <! -\- x ' the contrary, for x = rnr (where n is an integer) the value is — , while for x — ( n + - W it is (n + - W + 1 + 1 _. n " \ V \ V T ^ („+£),/ Although the numerator and denominator both become infinite, the quotient neither remains between positive bounds, nor tends to zero, nor tends to infinity. The numerator, therefore, is neither of the same order as the denominator, nor of lower order, nor of higher order. This apparently startling situation merely means that our definitions are not designed in such a way that we can compare any pair of functions. This is not a defect; we have no desire to compare the orders of such functions as the numerator and denominator above, since knowledge of the value of one of them gives us no useful information about the other. 194 DIFFERENTIATION AND INTEGRATION [Chap. 4. The Order of Magnitude of a Function in the Neighbourhood of an Arbitrary Point. Just as we can inquire into the behaviour of a function when x increases without limit, we may also ask ourselves whether and how functions that become infinite at the point x = £ may be distinguished as regards their behaviour at that point. We further state that the function f(x) = , becomes infinite \ x ~€\ of the first order at the point x = f , and correspondingly that the function ; -;- becomes infinite of the order a, provided that \x—g\« a is positive. We then recognize that the function e 1 "* -11 becomes infinite of higher order, and the function log | x — f | infinite of lower order, than all these powers; i.e. that the limiting relations Iim(|as— £|«.e*fl"-*l)=oo and lim(|as— £|» . log| x— f |) = hold. In order to see this we merely put j j = y; our state- ments then reduce to the known theorem on p. 192, since |x- f |".^l-« = eft and \x- £|-Mog| x- || = -(logy)/^ and y increases beyond all bounds as x tends to £ . The method of reducing the behaviour at a finite point to the behaviour at infinit y by the substitution j = y frequently proves useful. I*— fl 5. The Order of Magnitude of a Function tending to Zero. Just as we seek to describe the approach of a function to infinity more definitely by means of the concept of order of magnitude, we may also specify the way in which a function approaches zero. We say that as sc-»-oo the quantity 1/x vanishes to the first order, the quantity x~", where a is positive, to the order a. We find once again that the function 1/logx vanishes to a lower order than an arbitrary power x~ a , that is, for every positive a the relation lim(a; -cl . log a;) = holds. «- >0 HI] ORDER OF MAGNITUDE 195 In the same way we say that for x — $ the quantity x — f vanishes to the first order, the quantity | x — £ | a to the order a. With the above results it is easy to prove the relations lim( I x |« . log I x I ) = 0, Iim( I x I- . *-vm» = *->o x ^. ' which are usually expressed as follows: The function ■ — vanishes as x -► to a lower order than log I x I any power ofx; the exponential function e" 1/w vanishes to a higher order than any power of x. Examples 1. Compare the following functions with powers of x as regards their Older of magnitude as x -> 00: (a) e^- 1. . I* cos's (6) (log*)". 1/B + (c) sinx. to) j _- r . 1/ «- (i) sinha;. ^ x* — 1. (e) x 1 /2 sin x . arc tan *. (j ) log (a; log x). 2. Compare the functions of Ex. 1 with e aX , e 1 "", (log a;)". 3. Compare the functions of Ex. 1 with powers of x as x -»■ 0. 4. Does the limit lim e*"e( — **> exist? 5. What are the limits, as a: -> 00, of e( _ «*> and «(*"*>? 6. Let /(a;) be a continuous function vanishing, together with its first derivative, for x = 0. Show that f(x) vanishes to a higher order than x aa * -*■ 0. 7. show that /(„ = ff+;;tl+;;; -^ *« - ^^ of the same order of magnitude as a:" -m , when x -*■ 00. 8.* Prove that e* is not a rational function. 9.* Prove that e* cannot satisfy an algebraic equation with poly- nomials in x as coefficients. 196 DIFFERENTIATION AND INTEGRATION [Chap. Appendix to Chapter III 1. Some Special Functions From time to time we have made it clear by examples that the general concept of function contains many possibilities foreign to naive intuition. As a rule these examples were not given in terms of single analytical expressions. Here, therefore, we wish to show that it is possible to represent various typical discon- tinuities and abnormal phenomena by means of very simple expressions built up from the elementary functions. We begin, however, with an example in which no discontinuity is present. 1. The Function y = e~ Vx '. This function (cf. fig. 22), which is defined in the first instance only for values of x other than zero, obviously has the limit zero as x -> 0. For by the transformation 1/ir* = \ our function becomes y = erf and lim e~f = 0. Hence in order to extend our function so that it is continuous for x = we define the value of the function at the point x = by the equation 1,(0) = 0. By the chain rule the derivative of our function for x 4= is 1/ = — er 1 /"'. 3? If x tends to 0, this derivative will also have the limit 0, as we find imme- diately from p. 194 et seq. At the point x = itself the derivative *— >o h *— >o h is also zero. If we form the higher derivatives for x #= 0, we shall obviously always obtain the product of the function e -1 !"' and a polynomial in 1/x, and the passage to the limit * -*• will always yield the limit 0. All the higher derivatives will likewise vanish, like y", at the point x = 0. Ill] SOME SPECIAL FUNCTIONS 197 Thus we see that our function is continuous everywhere and differen- tiable as many times as we please, and yet at the point x = it vanishes with all its derivatives. We shall later realize (Chap. VI, Appendix, p. 336) how remarkable this behaviour really is. 2. The Function y = e~ l,x . We may readily convince ourselves that for positive values of x this function behaves in the same way as the function just dealt with; if x tends to through positive values the function tends to 0, and the same is true of all its derivatives. If we define the value of the function at x = as y(0) = 0, all the right-hand derivatives at the point x = will have the value 0. It is quite another matter when x tends to through negative values; for then the function and all its derivatives become infinite, and left-hand derivatives at the point x = do not exist. At the point x = 0, therefore, the function has a remarkable sort of discontinuity, quite unlike the infinite discontinuities of rational functions considered on pp. 22, 53 (of. fig. 23). 3. The Function V = tanh — . X We have already seen on pp. 33, 52 that functions with " jump " dis- continuities can be obtained from simple functions by a passage to the limit. The exponential function denned on p. 171 and the principle of compounding of functions give us another method for constructing functions with such discontinuities from elementary functions, without any further limiting process. An example of this is the function . , 1 e 1 '* — e- 1 /* y = tanh «!/• + e' -1/a and its behaviour at the point x = 0. The function is in the first instance 198 DIFFERENTIATION AND INTEGRATION [Chap. not defined at this point. If we approach the point x = through posi- tive values of x, we obviously obtain the limit 1; if , on the other hand, we approach the point x = through negative values, we obtain the limit — 1. The point x = is therefore a point of discontinuity; as x increases through the value of the function jumps by 2 (cf. fig. 24). On the other hand, the ^_ derivative y, , / ^^tanhji -/ —•a y' = - cosh 2 (l/a;) x* Fig. 24 a? ( e i/« _|_ e -i/»)a approaches the limit from both sides, as follows readily from * § 9, p. 194. 4. The Function y = x tanb. — . x In the case of the function 1 e l/* _ e -i/« y = xtann- = x —, TT - * x e i/« _|_ e -l/» the above discontinuity is removed by the factor x. This function has the limit as x -*■ from either side, so that we can again appropriately Fig. 25 define y(0) as equal to 0. Our function is then continuous at x = 0, but its first derivative y = tanh - — x x cosh* (1 /a;) * Another example of the occurrence of a " jump " discontinuity is given by the function y — arc tan - as x -* 0. Ill] SOME SPECIAL FUNCTIONS 199 has just the same kind of discontinuity as the preceding example. The graph of the function is a curve with a corner (cf. fig. 25); at the point x = the function has no actual derivative, but a right-hand derivative with the value +1 and a left-hand derivative with the value —1. 5. The Function v = x sin -, ,y(0) = 0. x We have already seen that this function is not composed of a finite number of monotonia pieces — as we may say, it is not " sectionally * monotonio " — but that it is nevertheless continuous (p. 54). Its first derivative y" = sin cos -, tx 4= 0) XXX on the contrary, has a discontinuity at x = 0; for as x tends to this deri- vative oscillates oontinually between bounding curves, one positive and one negative, which themselves tend to + 00 and — 00 respectively. At the actual point * = the difference quotient is "^ ' ~ M ' — gju _. Bmce h h as A-»-0 this swings backwards and forwards between 1 and —1 an infinite number of times, the function possesses neither a right-hand derivative nor a left-hand derivative. 2. Remarks on the Differentiability op Functions The derivative of a function which is continuous and has a derivative at every point need not be continuous. As the simplest example of this we consider the function 1 y — f( x ) = a 2 sin x This function is in the first instance not defined at x = 0; we shall define /(0), its value there, as 0, so that the function is now defined and con- tinuous everywhere. For all values of x different from zero the derivative is given by the expression /'(*) = —x* cos + 2x sin - = —cos - + 2x sin-. x 3? x x x When * tends to 0, f'(x) has no limit. If, on the other hand, we form the difference quotient J( ' ^ = ( h* sin - ) /A = h sin -, we see at once that ft \ ft' ft this tends to as a does. The derivative therefore exists for x = 0, and * Ger. stllckweise; of. p. 438, footnote. 200 DIFFERENTIATION AND INTEGRATION [Chap. has the value 0. In order to grasp intuitively the reason for this para- doxical behaviour we represent the function graphically (cf. fig. 26). It swings backwards and forwards between the curves y = a 2 and y = — x 2 , which it touches alternately. Thus the ratio of the heights of the wave- crests of our curve and their distances from the origin steadily becomes smaller. Yet these waves do not become flatter; for their slope is given 2x sin cos -; at the points x = where X X 2«7t cos - = 1 this is equal to — 1, and at the points x = — where x (2» + 1)71 cos = — 1 it is equal to + 1. x 3. ,.'''/\ by the derivative /'(*) = Fif?. 26 In contrast to the possibility illustrated here, that a deri- vative may exist everywhere and yet not be continuous, we state the following simple theorem, which throws light on a whole series of earlier examples and discussions: if we know that in a neighbourhood of a point x = a the function f(x) is continuous and has a derivative f'{x) everywhere, except that we do not know whether /'(a) exists, and if in addition the equa- tion lim f'(x) — b holds, then the derivative f\x) exists at the point a also, and f'(a) = b. The proof follows immediately from the mean value theorem. For we have — '- — i±-' =/'(!), n where $ is a value intermediate between a and a + h. If A now tends to 0, by hypothesis /'(£) tends to 6, and our statement follows at once. A companion theorem to this is the following, which may be proved in a similar way: if the function f(x) is continuous Ill] DIFFERENTIABILITY OF FUNCTIONS 201 in a^x^b and for a < x < b possesses a derivative which increases beyond all bounds as x tends to a, then the right-hand difference quotient -^i- X— £ — =^i-J also increases beyond all h bounds as h tends to 0, so that no finite right-hand derivative exists at x = a. The geometrical meaning of this state of affairs is that at the point with the (finite) co-ordinates (a, /(a)) the curve has a vertical tangent. 3. Some Special Formula 1. Proof of the Binomial Theorem. Our rules for differentiation enable us to give a simple proof of the binomial theorem; this proof will be introduced here as an example of the method of undetermined coefficients which we shall find important later. We wish to expand the quantity (1 + x) n in powers of x for all positive integral values of n. We see at once that the function (1 + »)" must be a polynomial of degree n, i.e. it must be of the form (1 + x) n = a + OjX + a 2 x 2 + ... + a n x n , and the problem now is to determine the coefficients a„. If we put x = 0, we at once obtain a = 1. If we differentiate both sides of the equation once, twice, three times, &c, we obtain the equations n(l + x)"- 1 =<h+ % a 2 x + • • • + ncbnX"- 1 , n(n - 1) (1 + x) n ~ z = 2a a + 3 . 1a z x + . . . + n{n — l)a n x n ~\ Since these equations hold for all values of x, we can put x = in each of them and thus obtain for the coefficients %, a 2 , . . . the expressions _ n(n — 1) _ n(n — 1) {n — 2) <h 1.2 3 1.2.3 n(n — 1) (n — 2) . . . (n — k + 1) _ (n\ a * = i! W" We thus finally obtain the binomial theorem in the form (l + x)»=l + nx+(^x*+...+ (fyx>°+... + x" 202 DIFFERENTIATION AND INTEGRATION [Chap. 2. Successive Differentiation. Leibnitz's Role. In connexion with the above we leave the reader to prove as an exercise that the successive differentiation of a product may be performed according to the following rule (Leibnitz's rule): dn i fa) _ *'/ a ■ m dn -v fr i m dn -y d2 9 + dx« KJ9) ~dx» 9 ^V/ dx"- 1 dx ^ \y dx»~* dx*^'" (.-0 dfd»- 1 g + ,d»g dx dx"- 1 dx n ' The repeated differentiation of a compound function y = /{<£(«)}, however, follows no such easily remembered law. From the rules for differentiation in last chapter (the product rule and the chain rule) we have dy = df_d4 = ,,^ dx d<f> dx g=/"<P +/'<£", g=/»'fi + 3/"W"+/'f". 3. Farther Examples of the Use of the Chain Rule. The Dif- ferentiation of fix)* 1 **. The Generalized Mean Value Theorem. To form the derivative of the function x°° we write x x = e xl °z x , whence we obtain _ X" =!B"(l0gX -f 1) dx by the chain rule. Similarly, we can carry out the differentiation of the more general expression f(x)K x) — e'( z > l0 * /(a,) by means of the chain rule, in the following way: ^{/(*)*-»} =/(*)*<> .f'(x){logf(x) + 1}. As a further application of the chain rule we here give a proof of the theorem which we have already called the generalized Ill] SOME SPECIAL FORMULA 203 mean value theorem of the differential calculus (p. 135), the theorem being established here under less stringent conditions. Let G(x) — u be a function which in the closed interval a ^ x ^ b is continuous and monotonic, and which in the open interval a < x < b has a derivative which is nowhere equal to 0, and let F(x) be a function which is also continuous for a <S x ^ b and differentiable for a < x < b. By means of the inverse function x = $>(u) of G(x) we introduce the new independent variable u instead of a; in F(x), thus obtaining the compound function f(u) = F(<$>(u)); according to the chain rule, /'( M )=^'(s)<D'( M )=J|| The ordinary mean value theorem, applied to the function /(«) and to the interval between v^ = G(a) and Mg = G(b), shows us that for an intermediate value a> f(u 2 )-f(u 1 ) _ f , F(b)-F(a) _F'(£) M,-^ JK } G(b)-G(a) G'(£)' where £ = $(«) is a value intermediate between a and b. Examples 1. Find the 3econd derivative of f[g{h(x)}]. 2. Differentiate the following functions: (a) x*»*. (6) (cosa;)""*. (C) log,^, »(*)» (that is, the logarithm of u(x) to the base v(x); v(x) > 0). 3. Prove Leibnitz's rule. 4. Find the »-th derivative of: (a) x*t?*. (d) cos mx sin kx. (6) (loga:) 2 . (e) e*oos2a;. (c) sina; sin2x. (/) (1 + x)°e*. 6.* Find the n-th derivative of arc sina: at x = 0, and then that of (arc sina;) 2 at x = 0. 6. Prove that S k(k - 1) (^) = n(n - l)2"-\ CHAPTER IV Further Development of the Integral Calculus The rules for differentiation which we formulated in the pre- ceding chapter have given us extensive powers over the problem of differentiating given functions. Almost always, however, the inverse problem of integration greatly exceeds it in importance. Hence we must now study the art of integrating given functions. The results attained by means of our differentiation formulae may be summed up as follows: Every function which is formed from the elementary functions by means of a " closed expression " * can be differentiated, and its derivative is also a closed expression formed from the elementary functions. On the other hand, we have not met with any exactly corre- sponding fact applying to the integration of elementary functions. We do know that every elementary function, and, in fact, every continuous function, can be integrated, and we have integrated a large number of elementary functions either directly or by inversion of differentiation formulae and have found their in- tegrals to be expressions involving elementary functions only. But we are still far from being able to find a general solution of the following problem: given a function /(x) which is expressed in terms of the elementary functions by any closed expression, to find an expression for its indefinite integral, ^"(a;) = lf(x)dx, which is itself a closed expression in terms of the elementary functions. * By this we mean a function which can be built up from the elementary functions by repeated application of the rational operations and the processes of compounding and inversion. In this connexion it should, however, be emphasized that the distinction between " elementary " functions and others is in itself quite arbitrary. 204 [Chap. IV] ELEMENTARY INTEGRALS 205 The fact is that this problem is in general insoluble; it is by no means true that every elementary function has an integral which itself is an elementary function. In spite of this, it is extremely important that we should be able actually to carry out such integrations when they are possible, and that we should acquire a certain amount of technical skill in the integration of given functions. The first part of this chapter will be devoted to the develop- ment of devices useful for this purpose. In this connexion we would expressly warn the beginner against merely memorizing the many formulae obtained by using these technical devices. The student should instead direct his efforts towards gaining a clear understanding of the methods of integration and learning how to apply them. Moreover, he should remember that even when integration by these devices is impossible the integral does exist (at least for all continuous functions), and can actually be calculated to as high a degree of accuracy as is desired by means of numerical methods which will be developed later (Chap. VII, p. 342). In the latter part of the chapter we shall endeavour to deepen and extend our conceptions of integration and integral, quite apart from the problem of the technique of integration. 1. Elementary Integrals First of all we repeat that to each of the differentiation formulas proved earlier there corresponds an equivalent integra- tion formula. Since these elementary integrals are used time and again as materials for the art of integration, we collect them in a table (p. 206). The right-hand column contains a number of elementary functions, the left-hand column the corresponding derivatives. If we read the table from left to right we obtain in the right-hand column an indefinite integral of the function in the left-hand column. We would also remind the reader of the fundamental theorems of the differential and integral calculus, proved in Chap. II, § 4 (p. 117), in particular, of the fact that the definite integral is obtained from the indefinite integral F(x) by the formula ff(x)dx=F(x) = F(b) - F(a). 206 THE INTEGRAL CALCULUS [Chap. J"(») = /(*) 1. x° (a *-l). 2. 1 X 3. d". 4. a x (a *!)• 5. sing. 6. cos a; F{x) = lf(x)dx 7. — — (s cosec 2 x). sin 2 a; 8. -L- (==sec 2 x). p,na a a: 9. 10. 11. 12. 13. 14. 15. 16. 17. oos'x sinhx. coshx. 1 sinh 2 x 1 cosh 2 a; (s cosech'x). (= sech 2 x). — k(|-I<D. V(l-x 2 ) 1 1 + x 2 ' 1 V(l + x 2 )' 1 ±V(x 2 — 1) | x | < 1 1*1 >1 ( I * | > 1). 1 S" xM a+ 1* log | x \. a" logo* — cosx. sinx. —cot*. tanx. coshx. sinhx. — cothas. tanhx. {arc sinx. — arc cosx. aro tanx. arc cot a:. far< ar sinhx = log{x + V(l -f- x 2 ) }. ar coshx = log{x ± Vfx 2 — 1) }. ar tanhx = - log — —. 2 1 — x , , 1 . X + 1 ar cothx = - log — ■ 2 x- 1 IV] ELEMENTARY INTEGRALS 207 Finally, for the technique of integration the reader should have the elementary rules of integration collected in Chap. II, § 1 (pp. 81-82) at his finger-tips. In the following sections we shall attempt to reduce the calculation of integrals of given functions in some way or other to the elementary integrals collected in this table. Apart from devices which the beginner certainly could not acquire syste- matically, but which, on the contrary, occur only to those with long experience, this reduction is based essentially on two useful methods. Each of these methods enables us to transform a given integral in many ways; the object of such transformations is to reduce the given integral, in one step or in a sequence of steps, to one or more of the elementary integration formulae given above. 2. The Method op Substitution The first of these useful methods for attacking integration problems is the introduction of a new variable (i.e. the method of substitution or transformation). The corresponding integral formula is just the chain rule of the differential calculus expressed in the integral form. 1. The Substitution Formula. We suppose that a new variable u is introduced into a func- tion F(x) by means of the equation x = <f>(u), so that F(x) becomes a function of u: F(x) = F(</>(u)) = G(u). By the chain rule of the differential calculus dG dF Ji^H + M' If we now write F'(x)=f(x) and G'(u) = g(u), or the equivalent expressions F(x) = f f( x )dx and G(u) = j ' g{u)du, ao8 THE INTEGRAL CALCULUS [Chap. then on the one hand the chain rule takes the form and on the other hand G(u) — F(x) by definition, that is, J g{u)du = jf{x)dx, and we obtain the integral formula equivalent to the chain rule, //(#«)) f («) du = ff(x) dx, {x = flu)}. This is the basic formula for the substitution of a new variable in an integral. It means that if we wish to find an indefinite integral of a function of u, which is given in the special form /(<£(«))<£'(«), then we can instead find the indefinite integral of the function f(x) as a function of x and after integration return to the variable m by putting x = <f>{u). ,,,. If, for example, we apply the formula to the integrand £-— we obtain T f*Mdu= /"*?=]og|*Hlog|-«u)| J <p(u) J X or, replacing u by x, [&dx=log\<l>(x)\. J <}>{x) If in this important formula we substitute particular functions, such as <f(x) = log a or <f(x) = sin x or <p(x) = cos a;, then we obtain* f —JL-= log | log x |, J a; log a; / cot xdx = log | sin x\, J tan xdx = —log | cos x \. A further example is / <p(u)<?'(u)du = / xAx= 2 a;2 = gl/PMP. where f(x) = *. This yields for <p(u) = logw J u 2 * These and the following formulae are verified by showing that differentia- tion of the result gives us back the integrand. The formulae, moreover, are of course only asserted as true in so far as the expressions occurring in them have a meaning. IV] THE METHOD OF SUBSTITUTION 209 We finally consider the example / sin n « cos udu. Here x = sin w = 9 («), and hence /r x"* 1 sin™* 1 11 sin»« cos«d« = J x»dx = —^ = -^p-f • In many cases, however, we shall use the above formula in the reverse direction, starting with the right-hand side, the integral f f(x)dx. We now have to evaluate or simplify a pre- scribed indefinite integral F(x) = ff(x)dx by introducing the new variable of integration u by means of the transformation formula x = <f>{u), then working out the indefinite integral G(u)=fM(u))4>'{u)du, and finally replacing the variable u in this integral by x. In order to carry out this last step we must be certain that a definite value m does actually correspond to the value x, i.e. that the function x = </>(u) has an inverse. Accordingly we now make the following assumption, in which we regard x as the primary variable. In the interval under consideration u = tjj(x) is a mono- tonic differentiable function whose derivative ^'(z) does not vanish anywhere in the interval. The inverse function — which under these conditions is definite and single-valued — we denote by x = 4>(u); its derivative is then given by <f>'(u) = l/0'(a;). As the basic formula for the substitution of a new variable u in an integral, we obtain ff(x)dx= ff(+(u)) $ (u)du (u = f(x)). The indefinite integral J i(x)dx can be obtained by calculating the indefinite integral f f(^(u))^'(u)du and finally introducing x instead of u for the independent variable by means of the equation U=i/r(x). It is therefore not sufficient merely to express the old variable x in terms of the new one u, and then to integrate with respect to this new variable; before integrating we must multiply by g (E798) 2io THE INTEGRAL CALCULUS [Chap. the derivative of the original variable x with respect to the new variable u. The corresponding formula for definite integration between two limits is Jj(x)dx = l (a) f(<Hu))<f>'(u)<lu. In the new integral we have to choose those limits of integration which are obtained by subjecting the old integration limits to the transformation x = <f> (u), u = ifi (x). In most applications the integrand f(x) will appear at the outset as a function of a function, say/(a;) = h(u), where u = ^(a;). It is then more convenient to write our integral formula in a slightly different form by identifying the expression /{^(m)} with the expression h{u). If for u we make the substitution u = >ji(x), x = 4>{u), then our transformation formula is simply / h{if>(x)}dx = J h(u) — du. As a first example we consider the integration of the function f(x) = sin 2x, taking u = i/ (x) = 2x and h(u) = sin u. We have ax If we now introduce u = 2x into the integral as the new variable, then it is transformed, not into / sin u du, but into = / smudu= — - cos « = — - cos 2s; this may of course be verified at once by differentiating the right-hand side. If we integrate for x between the limits and rc/4, the corresponding limits for u are and rc/2, and we obtain ■ . . 1 |»/ 2 1 sm udu = cos it = -. 2 | 2 r/4 J /rr/2 sin 2xdx = - I si 2 •'o rdx —r-. Here we take VI u= ty(x)=* Vx, whence x = <p(«) = « a . Since <p'(w) = 2u, we have f* dx a f 2 udu „ r 2 IV] THE METHOD OF SUBSTITUTION an 2. Another Proof of the Substitution Formula. Our integration formula can also be explained in another and more direct manner, by aiming at the formula for definite integration and basing the proof on the meaning of the definite integral as a limit of a sum. To calculate the integral h(ift(x))dx J (for the case a < 6), we begin with an arbitrary subdivision of the interval a^x^b, and then make the subdivision finer and finer. We choose these subdivisions in the following way. If the function w = ifi(x) is assumed to be monotonic increasing, there is a (1, 1) correspondence between the interval a^x^Lb on the x-axis, and an interval a ^ u ^ /3 of the values of u = tli(x), where a = i/i(a), /} = i[i(b). We divide up this u- interval into n parts of length * Am; there is a corresponding subdivision of the cc-interval into sub-intervals which in general are not all of the same length. We denote the points of division of the a;-interval by x = a, x^ x 2 , . . . , x n = b and the lengths of the corresponding sub-intervals by Aast, Aa*j, .... Ax„. The integral we are considering is then the limit \ of the sum i h{f(t v )}Ax„, V— 1 where the value f„ is arbitrarily selected from the v-th sub- interval of the a;-subdivision. This sum we now write in the " Ax form S h(u v ) — -?- Am, where m„=^(£„). By the mean value K-i Aw Ace,, theorem of the differential calculus -jr 1 = fii 7 )*)' where ij„ is a suitably chosen intermediate value of the variable u in the v-th sub-interval of the w-subdivision and x=<f>(u) denotes the * The assumption that these sub-intervals are all equal is by no means essential for the proof. t This limit exists (for Att — »■ 0) and is the integral, since on account of the uniform continuity of x — </>(u) the greatest of the lengths Ax tends to with Am. 212 THE INTEGRAL CALCULUS [Chap. inverse function of u = tfi(x). If we now select the value £„ in such a way that f„ and tj„ coincide, i.e. £„ = ^(ij„), i\ ¥ — 0(|„). then our sum takes the form „ = i If we here make the passage to the limit we immediately obtain the expression f h{u) — du, J a dU as the limiting value, that is, as the value of the integral we are considering, in agreement with the formula given above. We have therefore proved the following theorem: Let h(u) be a continuous function ofnin the interval a ^ u ^ /?. Then if the function u = ^r(x) is continuous and monotonic and has a continuous non-vanishing derivative — in a ^ x ^ b, and ^(a)=a,<A(b) = & «** J h{tji{x) }dx =Jh(u)dx =J h(u) — du. This formula exhibits the advantage of Leibnitz's notation. In order to carry out the substitution u = tfi(x), we need only write — du in place of dx, changing the limits from the original values du of a; to the corresponding values of u. 3. Examples. Integration Formulae. With the help of the substitution rule we can in many cases evaluate a given integral / f(x)dx if we reduce it by means of a suitable substitution x = <f>(u) to one of the elementary in- tegrals in our table. Whether such substitutions exist and how to find them are questions to which no general answer can be given; this is rather a matter in which practice and ingenuity, in contrast to systematic method, come into their own. As an example, we shall work out the integral / * 8 J V{<^-^) IV] THE METHOD OF SUBSTITUTION 213 by means of the substitution * x = <f>(u) = au, u = ^(x) — x/a, dx = adu, by which, using No. 13 of the table on p. 206, we obtain f ^_ = f adu = arc sinw= arc sin-,for | x\ < | a\. By the same substitution we likewise obtain /dx f adu 1 , 1 . x = / = _ arc tanw = - arc tan , a 2 + x 2 Ja?(l + tt?) a a a /dx . , x . = ar sinh -, ■x/(a 2 + x 2 ) a f ar cosh -, for | x \ > | a |, V(a; 2 — a 2 ) a dx r dx J a 2 — x 2 ' - ar tanh - for | x | < | a \ a a - ar coth - for | x | > | a | a a formulae which occur very frequently and which can easily be verified by differentiating the right-hand side. In conclusion, we again emphasize the following point. In our substitution process we have made the assumption that the substitution has a unique inverse a;= </>(«), and indeed that iji'(x) is nowhere equal to zero in the interval under consideration. If our assumption is not fulfilled, application of the substitution formula may easily lead to wrong conclusions. If </>'(a;) = at isolated points of the interval of integration only, we can avoid these difficulties by subdividing this interval in such a way that ifi'(x) vanishes only at the ends of a sub-interval; we can then apply the substitution to each sub-interval separately .f * For the sake of brevity we take the liberty of writing the symbols dx and du separately, i.e. dx - <j/(u)du instead of dx/du = <j>'{u) (cf. pp. 106-107). t An application of this method at once leads to the following result, which applies to many special cases: if the derivative <p'(x) vanishes at a finite number of points, but the function ip(x) remains monotonia, then the substitution for- mula remains valid. 3i4 THE INTEGRAL CALCULUS [Chap. 3. Further Examples op the Substitution Method In this section we bring together a number of examples which the reader may consider carefully by way of practice. By the substitution u — 1 ± x*, du = ±2x dx, we deduce that In these formulae we must take either the sign + in all three places or the sign — in all three places. By the substitution u= ax+ b, du = a dx (a 4= 0), we obtain / r-s = - log | ax + b |, J ax+ b a f(ax+ 6)« dx = \ (ax + 6J-+ 1 (a 4= - 1), J a(a + 1) /sin(aa; + b)dx = cos(aa; + 6); a similarly, by means of the substitution tt= cosx, du = —mixdx, we obtain / t&nxdx= — log|oosx|, and by means of the substitution u = sin a;, du = cos as dx, I cotx dx = log | sinx | (of. p. 208). Using the analogous substitutions u = cosh a;, du = sinhx dx and u = sinhx, du = cosh a; dx, we obtain the formulae / tanha: dx = log | cosha; |, / cotha; dx = log | sinha: |. In virtue of the substitution u — - tana;, du— - sec 2 a; dx, we arrive at o 6 the two formulae dx 1 r 1 da; /■ <fo = 1^ r 1 J o» sin a a;+ 6 2 cos'* 6 2 o a . , , ' I — tan a a: + I J 6' = — • arc tan ( - tana; ) cos 2 as IV] THE METHOD OF SUBSTITUTION 215 and artanhf - tanx) ab \o / /dx a 1 sin 2 * — b 1 We evaluate the integral ar ooth ( - tana;). ab xb / f dx J sina; US 4R Hf X 3S by writing sina: = 2 sin- cos- = 2 tan- cos*-, and putting « = tan-, so that du=\ sec 9 - Ax; the integral then becomes /dx f du , — = / — =log sin a; •> u tan- 2 1 If we replaoe * by x + jt/2, this formula becomes r_*L-iog|t«g+j)|. J oosa: I \2 4/ 1 The substitution u = 2x yields, if we also apply the known trigono- metrical formulae 2 cos* a; = 1 + cos 2a; and 2 sin 2 a; =1 — cos2x, the frequently used formulae and / cos 2 x dx = J (x + sina; cos a;) / sin* a; dx = J (x — sina: cosx). By the substitution x = cosm, equivalent to «= arc cos a;, or, more generally, x = a cosm (a =f= 0), we can reduce J^/il — x^dx and f\/( a2 — *?) dx respectively to these formulae. We thus obtain ly/ifl* — a?)dx = — — arc cos - + - \/(a 2 — a: 2 ). J 2 a 2 Similarly, by the substitution x = a coshw we obtain the formula /Vfx 2 - a*)dx = - 5 ar cosh - + 5 V(^ ~ « 2 ) and by the substitution x= a sinhu /V(« 2 + a 2 )*" = ? ar sinh - + ? v> 2 + * 2 )- 216 THE INTEGRAL CALCULUS [Chap. The substitution u= -, dx = -du leads to the formulae x M 2 r dx 1 . a / // 2 s\ = — - arc sin -, J xy \x i — er) a x r dx _ _ 1 a Jx^(x 2 +a 2 ) a x' r dx 1 , a / // 2 s; = — - ar cosh -. J xy(cr — x 2 ) a x Finally, we consider the three integrals / sinma; sinwa; dx, I ainmx oosnx dx, F cos mx cos nx dx, where m and n are positive integers. By well-known trigo- nometrical formulae we can divide each of these integrals into two parts, writing sinma; sinwa: = - {cos(m — n)x — cos(m + n)x}, sinma; cos tub = - {sin(m -f- n)x + sin(m — w)a;}, cosma; coswa; = - {cos(m + n)x + cos(m — n)x}. If we now make use of the substitutions w= (m + n)x and w=(m — n)x respectively, we directly obtain the following system of formulae: sinma; sinwa:<fa;= /si / sin mx cos nx dx = ;\ fsin(m — n)x sin(m+m)x1 ., j — i — — L. ij m ^ n 2 I m—n m+n J 1 / sin2ma:\., _ i x I if m = n: -2 \ 2m / r 1 fcos(m-fra)a: . cos(m— w)a;"| ., 21 m + n m — n J 1 /cos2ma;\ ., IV] THE METHOD OF SUBSTITUTION 217 ,1 f sin(m+w)x ^ sin(w— n)x \ ^ m + n ^ 2 I m + n m— n f - ( + a; I 11 to = n. ^2 \ 2m, J I cosmx cos nxdx= If in particular we now integrate from — n to -\-it, we obtain from these formulae the extremely important relations / +w . , f0ifm4=w. smma;smna;aa;= < ., lw if to = w, / sin ma; cosnxdx = 0, •' — IT r +w j 1 if to 4= n, / cosma;cosna;aa;= -! ., J_ w [it if to = n. These are the " orthogonality relations " of the trigonometric functions, which we shall meet with again in Chap. IX (p. 438). Examples Evaluate the following integrals and verify the results by differen- tiation: 1. fxe^dx. 9. f , X+1 dx. V(l — x*) dx I xe x *dx. 9. f 2. [x*er*'dx. 10. f -, _. J J V(5 + 2x + x 1 ) 3. fx*VT+&dx. 11. f , dx -. J J V(3— 2x— x*) 5. ' X SMdx. 12. f xdx X J X*— X+ ] dx ,„ /" a;<fcr f dx . 13. /" J »(loga;) n J (log a;)"' J V(x*—4x+ 1) 6 r 3(fa 14 f ( a + V**. ' J 9z 2 — 6a; + 2 "J V(2 + 2x— 2x*)' „ r dx . f dx " J V(z 2 - 2a;+T)' * J x* + x + 1* 8. /"-*!-& 16. /"* ./ 2 + 3* Ji'-i+l 8* (B798) 18, 19. 20. 21. zi8 THE INTEGRAL CALCULUS [Chap. l-j C &x /^arotana: "" J * + 2a X + b- 23 - 1 T+*r dx - J j _ dx. 24. / cos" x sin x dx. I ain*xcoa'xdx. 25. f , . J J Vfl + Sx 2 ) . I BuPxcoaixdx. 26. / - dx. J I (1 + z 2 ) 2 { x*(y/l-a?Ydx. 27. /" <fc(l<a<6). ;■■_ ds- 28. / K8in2a^(ia!. 29. Evaluate / (1 — a;)" da: (where n is a positive integer) by sub- stitution, •'o 4. Integration by Pabts The second useful method for dealing with integration problems is given by the formula for differentiating a product: U9Y=f'9+ff- 1. General Remarks. If we write this formula as an integral formula, we obtain (cf. p. 141) f(x)g(x) =fg(x)f'(x)dx +ff(x)g'(x)dx <* ff(x)g' {x)dx =f(x)g(x) -fg(x)f'(x)dx. This formula will be referred to as the formula for integration by parts. The calculation of one integral is thereby reduced to the calculation of another integral. For if we split up the inte- grand of an integral jo> (x) dx into a product <a (x) = f(x) <f> (x), and if we can find the indefinite integral g{x) = J</>(x)dx of the one factor <f> (x), so that <f> (x) = g'(x), then by our formula the integral f<o(x)dx = J f{x)<f>{x)dx = J f(x)g' (x)dx is reduced *° J 9(. x )f'( x )dx> which in some cases can be found more readily IV] INTEGRATION BY PARTS 219 than the original form. Since a given function <o (x) which occurs as an integrand can be regarded as a product/(a;) <f> (x) = f(x)g' (x) in a great many different ways, this formula provides us with a very effective tool for the transformation of integrals. Written as a formula for definite integration, the formula for integration by parts is f a b f(x)9'(x)dx=f(x)g(x) [- f*g(x)f'(x)dx =/(%(&) -f(a)g(a) - f g(x)f (x)dx. For in order to obtain the formula for definite integration from the formula for indefinite integration (Chap. II, § 4, p. 117) we have only to replace the z > variable appearing on both sides in the formula for the indefinite integral (1) by the value x = b, (2) by the value x = a, and write down the difference of these two ex- pressions. A simple interpretation of this formula, at least with suitable restrictions on the fupctions involved, can be given. Let us suppose that y=f(x) and z — g(x) are monotonic, and that f(a) = A, f(b) = B, g(a) = a, g(b) = j3; we can then form the inverse of the first function and substitute in the equation, thus ob- taining 2 as a function of y. We assume that this function is monotonic increasing. Since dy =f'(x)dx and dz = g'(x)dx the formula for integration by parts can be written Fig. 1 Jzdy -\-Jydz = Bfi — Aa, in agreement with the relation made clear by fig. 1, area NQLK + area PMLQ = area OMLK — area OPQN. The following example may serve as a first illustration: / log a: dx= I logx . 1 . dx. 220 THE INTEGRAL CALCULUS [Chap. We write the integrand in this way in order to indicate that we intend to put /(*) = log* and g'{x) = 1, so that we have /'(*) = 1/* and g(x) = x. Our formula then becomes / log* dx = x log* — / - dx = x log* — x. This last expression is therefore the integral of the logarithm, as may be verified at once by differentiation. 2. Examples. The following further examples may help the reader to grasp this method. If we put /(*) = x, g'(x) = e*, we have f'(x) = 1, g(x) = e*, and f x^dx = ef"(x— 1). In a similar way we obtain / x sin* dx = —x cos* + sin* / xooaxdx= xsm.x+ aoax. For /(*) = log*, g'{x) = x", we have the relation /x" log* dx = I log* — ). Here we must assume a 4= — 1. For a = — 1 we obtain (of. p. 208) /- log* dx = (log*) 3 — / log* . — ; * J x transferring the integral on the right-hand side over to the left, we have f\ i / - log* dx = - (log*) 2 . J x 2 We calculate the integral / arc sinxdx by taking /(*) = aro sin*, gr'(*) = 1. From this we obtain /arc sin* dx = * aro sin* — / — , J V(l - **) The integration on the right-hand side can be performed as in § 3 (p. 214); we thus find that / arc sin* dx = x are sin* + V(l — X s ). IV] INTEGRATION BY PARTS 221 In the same way we calculate the integral / aro tana; dx = x arc tanx log (1 + x*) and many others of a similar type. The following examples are of a somewhat different nature; here a double application of the method of integration by parts brings us back to the original integral, for which we thus obtain an equation. Integrating by parts twice, we obtain /e ox sin&x dx = — - e * cos 6a; + ? / e * cosfcc dx <= e*" cos bx + — e?* sin bx — - I eP" sin&x dx, b & a 6 2 J and, solving this equation for the integral / e"" sin bx dx, /e°* ainbxdx— — ef^ia sin&x — & cos bx). a* + 6 a v ' In a similar way it follows that /e ax cosbx dx = e^la coabx + 6 sin&x). o a + 6 s ' 3. Recurrence Formulae. In many cases the integrand is a function not only of the independent variable, but also of an integral index n, and on integrating by parts we obtain, instead of the value of the in- tegral, another similar expression in which the index n has a smaller value. We thus arrive after a number of steps at an integral which we can deal with by means of our table of integrals. Such a process is called a recurrence process. The following examples illustrate this: by repeated integration by parts we can calculate the trigonometrical integrals JcoB n xdx, jBm n xdx, j8m m xcos n xdx, provided that m and n are integers. For we find that Jcos n xdx = cos n_1 a; sina; -)- (n — l)fcoa n ~ i x sm 2 xdx; we can write the right-hand side in the form cos n_1 a; sina; + (n — 1) Jcos n ~ z xdx — (n — 1) icos n xdx, 222 THE INTEGRAL CALCULUS [Chap. thus obtaining the recurrence relation /cos"xdx = - cos n_1 a; sina: -\ [coa n ~ 2 xdx. n n J This formula enables us to keep on diminishing the index in the integrand until we finally arrive at the integral lcoaxdx=smx or ldx = x > according as n is odd or even. In a similar way we obtain the analogous recurrence formulae jsia n xdx = sin n-1 a; cosa; ^ lsin. n ~ 2 xdx J n n J f ■ ™ « j sin m+1 a; cos" -1 a: . n — 1 r . _ „_, , / sin" 1 ! cos n a; ax = 1- / sm m x cos" z x ax. J m 4- n m + n J In particular, these formulas enable us to calculate the integrals r 1 / sin 2 a; dx = -(x — sina; cosa;) /I cos 2 x dx = - (x + sina; cosa;), 2 as we have already done by the method of substitution (p. 215). It need hardly be mentioned that the corresponding integrals for the hyperbolic functions can be calculated in exactly the same way. Further recurrence formulae are given by the following transformations: f (log a:)™ (fas = a; (log a;)™ — ml (loga:)™- 1 ^, j x m e*dx = x m e?° — m j x m ^ 1 e!"dx, I x m wax dx = — x m cosa: + tn J x™- 1 cosa; dx, I x m cosx dx = x m sina: —ml a;™"" 1 sina; dx, f x" (loga;)™ dx = xa+1 < lo 8 g r - ™ /^"(Ioga;)™- 1 ^ (a* -1). IV] INTEGRATION BY PARTS 223 4. Wallis's Product. The recurrence formula for the integral lsax fi xdx leads in an elementary way to a most remarkable expression for the number it as an infinite product. We suppose that ft > 1, and in the formula sin"a; dx = sin n-1 aj cosa; -\ / sin n-2 a; dx n n J we insert the limits and it/2, thus obtaining /•f/2 n \ /.W2 / wn n xdx = / ain n ~ 2 xdx for« > 1. •M) ft •'0 If we again apply the recurrence formula to the right-hand side and continue the process, we obtain, distinguishing between the cases n = 2m and n = 2m + 1» r' 2 - ^ j 2m — 1 2m — 3 / aw^xdx — ■'0 2m 2m — 2 1 r'% J sin 2m+1 a!(fa; — m . m ~ . Jo 2m + 1 2m — 1 2 /"•. , . . - . / sinasoa;, 3 -/o whence f *'*• 2™ j 2m — 1 2m — 3 J 2m 2m— 2 1 it ' 2 " 2* 1 wi? m + 1 xdx — J o 2m + 1 2m — 1 2 '3' By division this yields ■n _ 2.2 4.4 6.6 2m. 2m 2 1.3"3.5'5.7"'(2m— l).(2m4 y.V2 / sin^aidas •'0 - 1} r"' 2 The quotient of the two integrals on the right-hand side con- verges to 1 as m increases, as we recognize from the following considerations. In the interval < x < 77/2 we have < sin 2 ™- 1 " 1 * ^ sin 2 ™ a; ^ sin 2 ™- 1 a;: 324 THE INTEGRAL CALCULUS [Chap. consequently J /.if/2 /.ir/2 ^ir/2 o »o •'o If we here divide each term by / sin 2n,+1 a; dx and notice that by the first formula proved above /.t/2 / sin 2 "* -1 xdx „ . . , ^ _ 2m + 1 _ j , 2_ / sin 2 ™* 1 ^^; m m •'o / sin 2 " we have 1 ^ -2-™ ^ 1 + -7T/2 / sin^xda; /' Bin im+1 xdx 2m from which the above statement follows. The relation it ,.224466 2m 2m - = Mm 2 m-xol 3 3 5 5 7 2m— 12m +1 consequently holds. This product formula (due to Wallis), with its simple law of formation, gives a remarkable relation between the number it and the integers. If we observe that lim = 1, we can write m _>»2m+ 1 .. 2 2 .4 2 ...(2«-2) 2 77 lim - 2m = -, ffl _*oo3 2 .5 2 ...(2m-l) 2 2' and if we take the square root and then multiply mumerator and denominator by 2 . 4 . . . (2m — 2) we find that „ =]hn 2.4...(2m-2)^ =lim 2 2 .4 2 ...(2m-2) 2 ^ m 2 m->oo3.5...(2m— 1) „,->«, (2m— 1)1 = j. 2 2 .4 2 . . .(2m) 2 V2m m-»-ot> (2m)! 2m 4 IV] INTEGRATION BY PARTS 225 From this we finally obtain ,. (m!) 2 2 2m , lim / , . = Vt. a form of Wallis's product which will be of use to us later (cf. Chap. VII, Appendix, p. 363). Examples Evaluate the integrals in Ex. 1-14: I. [ x S2^dx. 2. f x ' dx. 3. fx^oosxdx. J sin 8 * J (1 — a; 4 ) 2 J 4. / x^e-B'dx. 6. J aPcoanxdx (n a positive integer). 6. J x^sinnxdx (n a positive integer). 7. /a? cos a? dx, 8. f sin* x dx. 9. fcoa"xdx. 10. fx i V\ — x t dx. II. fx*<*dx. 12. f l ^dx (nM=l). 13. jx m logxdx (m=#l). 14. J x* (logx)* dx. 15. Prove the formula f<*p{x)dx = e*{p(z) - j>'(z) + !>"(*) - + ...}, where #(2:) is any polynomial. 16. Show that for all odd positive values of » the integral / ir^x^dx can be evaluated in terms of elementary functions. 17. Show that if n is even the integral / e~ x 'x n dx can be evaluated in terms of elementary functions and the integral / e-*"'dx (for which tables have been constructed). 18. Prove that •'0 ^0 ' •'0 19.* Ex. 18 gives a formula for the second iterated integral. Prove that the ra-th iterated integral of f(x) is given by _ I— f X f(u)(x-u)»-idu. (n— 1)\J 226 THE INTEGRAL CALCULUS [Chap. 5. Integration op Rational Functions The most important general class of functions integrable in terms of elementary functions consists of the rational functions where /(a;) and g(x) are polynomials: /(*) = <V" + am-ia™ -1 + . . . + a , g(x) = b n x" + b n _ x x»- 1 + . . . + b (b n #= 0). We recall that every polynomial can be integrated at once and that the integral is itself a polynomial. We therefore need only consider those rational functions for which the deno- minator is not a constant. Moreover, we can always assume that the degree of the numerator is less than the degree (ri) of the denominator. For otherwise we can divide the poly- nomial f(x) by the polynomial g(x) and obtain a remainder of degree less than n\ in other words, we can write f(x) = q(x)g(x) + r(x), where q{x) and r{x) are also poly- nomials and r(x) is of lower degree than n. The integration of f( x ) — — is then reduced to the integration of the polynomial q(x) y*" 1 ' r(x) and of the " proper " fraction -y-r. We further notice that the f(x) 9^' function -^-~ can be represented as the sum of the functions —t~z, so that we need only consider integrands of the form ~ r-r. g(x) J 8 g(x) 1. The Fundamental Types. We shall not at once proceed to the integration of the most general rational function of the above type, but shall instead study only those functions in which the denominator g(x) is of a particularly simple type, namely, g(x) = x, g(x)=l + x i , or, more generally, g(x) = x», g(x)=(l + x*) n where n is any positive integer. IV] INTEGRATION OF RATIONAL FUNCTIONS 22J To this case we can reduce the somewhat more general case in which g(x) == (ax + j3) n , a power of a linear expression ax + /? (a 4= 0), or g(x) = (ax 2 -\- Ibx + c) B , a power of a definite * quadratic expression. In the first case we introduce a new variable £ = ax + £. Then d£/dx — a, and x = ($ — fl)/a is also a linear function of £. Each numerator /(x) becomes a polynomial </>(() of the same degree, and consequently J (ax+ B) n a J £ » (ax+jS) In the second case, we write ax*+2bx+c = -(ax+b) 2 + - (<P = ac - 6 2 , d > 0), a a observing that, since we have assumed our expression to be definite, ac— b 2 must be positive and a =4= 0. By introducing the new variable £_ ax-\-b ? d~ r^2 "In we arrive at an integral with the denominator — (1 + {?)\ . Hence in order to integrate rational functions whose de- nominators are powers of a linear expression or of a definite quadratic expression it is sufficient to be able to integrate the following types of functions: 1 a?" x 2 ^ 1 X"' (x 2 + 1)"' (x* + 1)»' We shall, in fact, see that even these types need not be treated in general, for we can reduce the integration of every rational function to the integration of the very special forms of these three functions obtained by taking v = 0. Accordingly we now consider the integration of the three expressions 2. 1 x x n ' (x 2 + 1)"' (x 2 + 1)»' * A quadratic expression Q(x) = ax 1 + 2bx + c is said to be definite if for all real values of x it takes values having one and the same sign, i.e. if the equation Q(x) - has no real roots. For this it is necessary and sufficient that ac — b 2 should be positive. 228 THE INTEGRAL CALCULUS [Chap. 2. Integration of the Fundamental Types. Integration of the first type of function, — , immediately yields the expression log \ x\ if n = 1, and the expression — if » > 1, so that in both cases the integral is again an elementary function. Func- tions of the third type can be integrated immediately by introducing the new variable $ = X s + h whence we obtain 2xdx = d% and Jp+ir iJv i- t (.-i)^. Mr » " n Finally, in order to calculate the integral I = f dx " J (s* + 1)»' where m has any value exceeding 1, we make use of a recurrence method. For if we put 1 = 1 x* (a? + 1)» ~ (a? + I)"" 1 (*» + 1)"' da; r dx f X s dx "^ I vhr- ! IFTvF*- ! (*" + 1)» we can transform the right-hand side by integrating by parts, using the formula on p. 218 with Then, as we have just found above, 1 1 g(x) = - 2 (ra — l){a?+ l)"- 1 and consequently we obtain T _ r dx _ x 2m— 3 r dx * ~~ J (x> + 1)» ~ 2(n- l)(a; 2 + l)"" 1 2(n-l)J (x* + 1) The calculation of the integral I n is thus reduced to that of the integral /„_j. If n — 1 > ,1 we apply the same process to the latter integral, and continue the process until we finally arrive at the expression / dx . = arc tanx. (* a + 1) IV] INTEGRATION OF RATIONAL FUNCTIONS 229 We thus see that the integral * I n can be explicitly expressed in terms of rational functions and the function arc tana;. Incidentally, we could also have integrated the function directly using the substitution x = tant; we should then have obtained dx = sec 2 1 dt and 1/(1 + x 8 ) = cos* t, so that f — — = f cos 8 "- 2 tdt, J (**+ 1)» J and we have already learned (p. 222) how to evaluate this integral. 3. Partial Fractions. We are now in a position to integrate the most general rational functions, in virtue of the fact that every such function can be represented as the sum of so-called partial fractions, i.e. as the sum of a polynomial and a finite number of rational functions, each one of which has either a power of a linear expression for its denominator and a constant for its numerator, or else a power of a definite quadratic expression for its denominator and a linear function for its numerator. If the degree of the numerator /(x) is less than that of the denominator g(x) the polynomial does not occur. We are now in a position to integrate each partial fraction. For according to p. 226 the denominator can be reduced to one of the special forms x n and (x 2 + 1)", and the fraction is then a combination of the fundamental types integrated on p. 228. We shall not give the general proof of the possibility of this resolution into partial fractions. On the contrary, we shall confine ourselves to making the statement of the theorem in- telligible to the reader and to showing by examples how the resolution into partial fractions can be carried out in typical cases. In actual practice only comparatively simple functions are dealt with, for otherwise the computations become far too complicated. * The integral of the function r -= ttt; can be calculated in the same way; (%* — l) n by the corresponding recurrence method we reduce it to the integral / artanhx (orarcothz). 1 - x' 330 THE INTEGRAL CALCULUS [Chap. As we know from elementary algebra, every polynomial g(x) can be written in the form flr(x) = a(a;-a 1 ) l '(x-a a ) , '...(a?+26 1 a; + c 1 )"(ic 2 +26 a a; + c 2 )'''... Here the numbers a v a 2 , . . . are the real and distinct roots of the equation g(x) = 0, and the positive integers 1^1^ .. ■ indicate the numbers of times they are repeated; the factors x 2 + 2b v x + c„ indicate definite quadratic expressions, of which no two are the same, with conjugate complex roots, and the positive integers r v r 2 , . . . give the numbers of times that these roots are repeated. We assume that the denominator is either given to us in this form or else that we have brought it to this form by calcu- lating the real and imaginary roots. Let us further suppose that the numerator f(x) is of lower degree than the denominator (cf. p. 226). Then the theorem on resolution into partial fractions can be stated as follows. For each factor (x — a)', where a is any one of the real roots and I is the number of times it is repeated, we can determine an expression of the form (X- ay (x- af^ " (x- af and for each quadratic factor Q(x) == x 2 + 26a; + e m our product which is raised to the power r we can determine an expression of the form Bj, + C lX B 2 + G& B r +C,x Q "*" Q 2 " t ~-"~ t ~ Qr > fix) in such a way that the function ^— ' is the sum of all these ex- S( x ) f(x) pressions. In other words, the quotient -r-r can be represented as a sum of fractions each of which belongs to one or other of the types integrated on p. 228.* • Here we give a brief sketch of the method by which the possibility of this decomposition into partial fractions is proved. If g(x) — (x — a)* k(x) and A (a) + 0, then on the right-hand side of the equation M - /(") _L /(*)*(«)-/(«)*(») g(x) h(a)(x - a)* A(o) (Z - a)*A(z) the numerator obviously vanishes for x — a; it is therefore of the form IV] INTEGRATION OF RATIONAL FUNCTIONS 231 In particular oases the splitting up into partial fractions can be done easily by inspection. If, for example, g(x) = a? — 1, we see at once that 1 1111 x*— 1 2 x— 1 2 x + 1* f dx _ 1 i„„ | * — 1 J a? -I 2°°|a!+l • so that More generally, if g(x) = (x— a)(x — P), that is, if g (x) is a non-definite quadratic expression with two real zeros a and (3, we have 1 1111 so that (* — a)(x — P) « — pa;— a a — p * — P /dx _ 1 , x — a. (x-a){x- p) ~ a- p ° g a- p 4. Example. The Bimolecular Reaction. A simple example of the application of this easy reduction to partial fractions is given by the so-called bimolecular reaction. Let us suppose that we have two reagents whose original concentrations in mols per unit volume are a and b, where we assume that a < b, and let us suppose that in time t there is formed in the unit volume a quantity x (mols) of the product of reaction. Then, according to the law of mass action (cf. p. 182), in the simplest case — reaction between one molecule of each of the re- agents — the rate of increase of the quantity x is given by the equation dx rr = k(a — x)(b — x). The problem is to determine the function x(t). If, inversely, we think of the time t as a function of x, we have dt = 1 = 1 / 1 _ 1 \ dx k(a — a;)(6 — x) k(b — a) \a — x b — x' hence by integration kt = log + e, for x < a < b. h{a)(x — a) m f x (x), where f x (x) is also a polynomial, the integer m 2: 1, and /^a) 4= 0. Writing {^) - fl, this gives us IM g fi(x) g(x) (x - a) k (x - o)*-«»A(z)' Continuing the process, we can keep on diminishing the degree of the power of (x — a) occurring in the denominator until finally no such factor is left. On the remaining fraction we repeat the process for some other root of g(x), and do this as many times as g(x) has distinct factors. This being done not only for the real but also for the complex roots, we eventually arrive at the oomplete analysis into partial fractions. 232 THE INTEGRAL CALCULUS [Chap. We determine the constant of integration c by the condition that at time t = no product of reaction has yet been formed, so that 1 log? + c=0. a—b °b We thus obtain finally 1 l ~a la = log , a—b _ x b and if we solve for x this gives the required function x(t): _ ab(\ — e( a - fc >**) 6 _ aeia-ftkt 5. Farther Examples of Resolution into Partial Fractions. The Method of Undetermined Coefficients. II g(x) = (x — ai) (a; — c^) ... (x — o„), where a t 4= o-k if » 4= k, i.e. if the equation g(x) = has only single real roots, the expression in terms of partial fractions has the simple form 1 _ <h _|_ g 2 + # _ j_ a n ^(a;) a; — c^ x — a% x — a„ We obtain explicit expressions for the coefficients %, a 2 , . . . if we multiply both sides of this equation by (x — o^), cancel the common factor (x — dj) in the numerator and denominator on the left and in the first term on the right, and then put x= a^. This gives* 1 Oj = . («1 — a 2 ) (ci! — a 3 ) . . . (dj — a„) As a typical example of a denominator g(x) with multiple roots, we consider the function . The preliminary statement x*(z — 1) 1 =_?_ + *+£ a?(a;— 1) x — 1 x x* in accordance with p. 230 leads us to the required result. If we multiply both sides of this equation by a; 2 (a; — 1) we obtain the equation 1 = (a + b)x* — (6 — e)x — c, * The reader will observe that the denominator on the right is (^(a,), i.e. the derivative of the function g(x) at the point x — a,. IV] INTEGRATION OF RATIONAL FUNCTIONS a 33 true for all values of x, from which we have to determine the coefficients a, b, c. This condition cannot hold unless all the coefficients of the poly- nomial (a+ ft)* 2 — (6— c)x — c— 1 are zero, i.e. we must have a+6=&-c=c+l = 0orc=-l, 6=-l, a=l. We thus obtain the resolution 1 111 * 2 (*-l) x-1 and consequently /. ; = log|x- 1|- log |a;|+ l We shall now split up the function — (which is an example of x(x a +1) r the case where the zeros of the denominator are complex) in accordance with the equation 1 a bx -f c /, *(a; 2 +1) x x*+l For the coefficients we obtain a + b = c= a — 1 = 0, so that 1 _ 1 x_ X(x 2 + 1) X X*+T and consequently *,l^T) = log l a! l-^ lo ^ 8+1 )- As a third example we consider the function . Even Leibnitz x* + 1 found this a troublesome integration. We can represent the denominator as the product of two quadratic factors: * 4 + 1 = (a? + l) 2 - 2x* = (x* + I + V2x) (x* + 1 -V2x). We know, therefore, that the resolution into partial fractions will have the form 1 _ ax + b ex + d x i +l x*+V2x+l x*— V2x+l' To determine the coefficients a, 6, c, d, we have the equation (a + c)x* + (6 + d - aV2 + cV2)x* +{a + c-bV2 + dV2)x + (b + d - 1) = 0, which is satisfied by the values a=_L 6=1 C =__L d= 1 2V2' 2' " 2V2' 2* We therefore have 1 1 x+V2 1 x—V2 x*+l 2V2*z 2 + V2x+ 1 2V2"ie»— V2x + l' 234 THE INTEGRAL CALCULUS [Chap. and, applying the method given on p. 227, we obtain /?TT = 4V2 l0g|a!a+V ' 2a,+ 1| -4V2 l0g|a:, - V ' 2a!+11 J L_ arc tan (V2x + 1) + _L_ arc tan (V2x — 1), 2V2 2V2 which may easily be verified by differentiation. Examples Integrate: 1 f ** 8 f Ax ' J 2x — Za?' 'Jl + J?' 2. f_*L. 9. f , (a; - 4) dx. J x*-x J (x*+ l)(s-2) 3. f 3dx . 10. f !L±i «fa. /• x* + x+l ^ f— A d*. J 3a; 2 — 2x — 6 J 1-*' — (2x. 2 f dx 12*[^L. J (x- lf(x* + 1) J * 6 + 1 /a^efce ., /• a 8 (a:— l) s (a»+l)' ' J x* + x* — /dx ,. r dx l^~x*' ' J a^+l) 2 ' 6. Integration of Some Other Classes of Functions 1. Preliminary Remarks on the Rational Representation of the Trigonometric and Hyperbolic Functions. The integration of some other general classes of functions can be reduced to the integration of rational functions. We shall be better able to understand this reduction if we begin by stating certain elementary facts about the trigonometric and hyperbolic functions. If we put t = tan -, elementary trigonometry gives 2t us the simple formulae It 1 — t* bulx = „ , cosa; = ; 1 + < 2 1 + t 2 IV] for OTHER CLASSES OF FUNCTIONS 235 = cos 2 - and t* l + t* 2 1 + fi ' and hence from the elementary formulae = sin 2 2' sinx = 2 cos 2 ~ tan ? and cosx = cos 2 - — sin 2 - 2 2 2 2 we obtain the above equations. These equations show that sin a; and cos a; can be expressed rationally in terms of the quantity t — tan -. From t = tan - we have by differentiation 2i 2 dt _ 1 __ 1 + fi dx _ 2 dx 2cos 2 a;/2 2~* S ° ttot A ~ lTP 5 hence the derivative — is also a rational expression in t. at The geometrical representation of our formulae and their geometrical meaning are given in fig. 2. Here the circle u* + v 1 = 1 in a ttt>-plane ia shown. If x denotes the angle POT in the figure, then u = cosx and v = sinx. The angle OSP with its vertex at the point u = — 1, v = is equal to x/2, by a theorem in elementary geometry, and we can read off the geometrical meaning of the parameter t from the figure; t = tan \x = OB. If the point P starts from S and runs once round the circle in the positive direction, i.e. if x runs through the interval from — 7c to +7r, the quantity t will run through the whole range of values from — oo to + oo exactly once. Fig. 2. — Parametric representation of the trigonometric functions We may correspondingly express the hyperbolic functions cosh a; = £(e" -f e -<°) and sinha; = £(e* — «r") as rational func- tions of a third quantity. The most obvious way is to put e x = t, so that we have cosha ! =l(r+^, smhx=l(r-iy which are rational expressions for sinha; and cosh a;. Here again dxjdt = 1/t is rational in t. But we obtain a closer analogy 236 THE INTEGRAL CALCULUS [Chap. with the trigonometric functions by introducing the quantity t = tanh-t we then arrive at the formulae 2 Binha; = 2f cosh as = 1 + 1- By differentiating t = tanh- we obtain, as on p. 235, the rational expression dx_ 2 dt l — t 2 for the derivative dx/dt. Here again the quantity t has a geometri- cal meaning similar to that which it has in the case of the trigo- nometric functions, as we see at once from fig. 3. V, HtM 1* R 1 +ox" —coskx- X rb it pig. 3. — Parametric representation of the hyperbolic functions But whereas in the case of the trigonometric functions t must run through the whole range of values from — °o to + oo in order to give all pairs of values of cos a; and sin a;, in the case of the hyperbolic functions t is limited to the interval — Kt<l. Having made these preliminary remarks, we proceed to our integration problem. 2. Integration of R(coax, sinar). Let 22 (cos as, sin a;) denote an expression which is rational in the two functions sin a; and cos a;, i.e. an expression which is IV] OTHER CLASSES OF FUNCTIONS 237 formed rationally from these two functions and constants, such as 3 sin 2 a; + cosa; 3 cos 2 a; + sin a;' If we apply the substitution t = tan - the integral I R{cosx, sina:)efa: is transformed into the integral and under the integral sign we now have a rational function of t. Thus we have in theory obtained the integral of our expression, since we can now perform the integration by the methods of the preceding section. 3. Integration of i? (cosh a;, smhx). In the same way, if .R (cosh a;, sinha;) is an expression which is rational in terms of the hyperbolic functions cosh a; and sinha:, we can effect its integration by means of the substitution t = tanh -. Recalling that dx 2 dt l-* 2 we have fR(coahx, sinhx)dx= fRJ^tL, yzTJ%) f^r -dt. t 2 (According to a previous remark we could also have introduced t = e x as a new variable and expressed cosh a; and sinha; in terms of t.) The integration is once again reduced to that of a rational function. 4. Integration of R{x, -\/(l — x 2 )}. The integral J R{x, -\/(l — x 2 )}dx can be reduced to the type treated in No. 2 by using the substitution x = cosm, \/(l — a^) = sinu, dx = — sinw du; 238 THE INTEGRAL CALCULUS [Chap. froin this stage the transformation t = tan - brings us to the 2i integration of a rational function. Incidentally, we could have carried out the reduction in one step instead of two by using the substitution dx — it dt~ (1 + fi/' that is, we could have introduced t = tan - directly as the new variable and thereby obtained a rational function to integrate. 5. Integration of R{x, \/(x 2 — 1)}. The integral jR {x, \/(x 2 — l)}dx is transformed by the sub- stitution x = coshw into the type treated in No. 3 (p. 237). Here again we can arrive at our goal directly by introducing t = aII . t ) = tanh -. ! "VV+i)' 6. Integration of R{x, V(**+ *)}• The integral jR{x, -\/(x 2 + l)}dx is reduced by the trans- formation a;=sinhM to the type considered in No. 3 (p. 237), and can therefore be integrated in terms of elementary functions. Instead of the further reduction to the integral of a rational u function by the substitution e u = r or tanh - = t, we could have 2* reached the integral of a rational function at a single stroke by either of the substitutions T-.+ V<*+l>. f-- 1 + ^*+ 1 >. X 7. Integration of R{x, i/(ax* -f Zbx + c)}. The integral jR {x, y/(ax*-\- 2bx-\- c)}dx of an expression which is rational in terms of x and the square root of an arbitrary IV] OTHER CLASSES OF FUNCTIONS 339 polynomial of the second degree in x can immediately be re- duced to one of the types just treated. We write (cf. p. 227) 1 , , ,>, , ac — 6 2 ax 2 + 2bx + c = ~ (ax + bf + a, a If ac — b 2 > we introduce a new variable f by means of the transformation f = f*"*" whereupon the surd takes ■\/(ac — 6 2 ) the form J j (£2+ 1) I. Hence our integral when expressed in terms of $ is of the type of No. 6. The constant a must here be positive in order that the square root may have real values. If ac — b 2 = 0, a > 0, then by way of the formula A/(az 2 + 2bx + c) = ■y/a (x + -\ we see that the integrand was rational in x to begin with. If, finally, ac - b* < 0, we put £ = ™ +b and obtain for V (0 — "jC) the surd the expression A\ ~ ac (p _ 1)1 If a j 8 positive, our integral is thus reduced to the type of No. 5 (p. 238); if, on the other hand, a is negative, we write the surd in the form yj( —3 — ) V0- — €*) and see that the integral is thus reduced to the type of No. 4 (p. 237). 8. Further Examples of Reduction to Integrals of Rational Functions. Of other types of functions which can be integrated by re- duction to rational functions we shall briefly mention two: (1) rational expressions involving two different surds of linear expressions, R{x, ^/(ax+b), ■y/{ax+ 0)}; (2) expressions of the form R\x, W/ ^ U, where a, b, a, fi are constants. In the first case we introduce the new variable $ = -\/(ax + /}), so that ax + /? = £* and consequently x^^zl and i* = 2J. a dg a 24 o THE INTEGRAL CALCULUS [Chap. then JR{x, V(oa; + b), ^(ax + j8)} dx = fB {£=£. </!{«*• -<«/»-*-)}, *} ^ % which is of the type discussed in No. 7. If in the second case we introduce the new variable "j/ax+b\ we have ox + 6 -fig" + & & = <*£ — fa B «■-» f "ai+T a*"-a' d* (a£»-af and we immediately arrive at the formula /«(*V(=T?))* which is the integral of a rational function. 9. Remarks on the Examples. The preceding discussions are chiefly of theoretical interest. In the case of complicated expressions the actual calculations would be far too involved. It is therefore expedient to make use, when possible, of the special form of the integrand to sim- plify the work. For example, in order to integrate the expression _ it is better to use the substitution t = tana; a 2 sin 2 a;+ & 2 cos 2 a; instead of that given on p. 237, for sin 2 a; and cos 2 a: can be ex- pressed rationally in terms of tana;, and it is therefore unneces- sary to go back to t = tan-. The same is true for every expres- sion formed rationally from* sin 2 a;, cos 2 as, and sina; cosa;. More- over, for the calculation of many integrals a trigonometrical form is to be preferred to a rational one, provided that the trigono- metrical form can be evaluated by some simple recurrence method. * For sin x cos x — tan x cos* x can of course be expressed rationally in terms of tan x. IV] OTHER CLASSES OF FUNCTIONS 241 For example, although the integrand in fx n {\/(l — a?)} m dx can be reduced to a rational form, it is better to write x = sin u and bring it to the form fsin B ucos m+1 udu, since this can easily be treated by the recurrence method on p. 222 (or by using the addition theorems to reduce the powers of the sine and cosine to sines and cosines of multiple angles). For the evaluation of the integral / ** , ■ (a 2 + h % >Q), •> a cos* + 6 sin* instead of referring to the general theory we determine a number A and an angle in such a way that a = .4 sinG, 6=4cos0; that is, we write A = Vo> + 6 a , sin G = j, cos 6=2- The integral then takes the form I f dx A J sin(* + 6)' and on introducing the new variable * + 6 we find (cf. p. 215) that the value of the integral is log tan Integrate: J 1 + ainx J 1 A e | 2 Examples dx + cos 2 a; f dX 8. f dx J 1 + cos*" J 3 -)- sin 2 *" /dx r 2 + sin*" 9 - J t* 113 *'**- /^- */- J sin 3 * J sir dx sin 3 * J sin* 4- cos* K f dx f sin 2 * 4- cos a * . / Z^T' 11 - / s „ : Bmxdx. J cos* J 3 C0S 2* -)- su^x 6- / r- . 12. /V(* 2 -4)i*. Jo o + cos* J IK798) 242 THE INTEGRAL CALCULUS [Chap. 13. fv(l+9x*)dx. 16. J. dx 14 f dx 17 f^SL J (x— 2)V(x i — ix + 3)' J V(l Vas + V(l- x) V(l + a;) + V(l — x) (x- 2)V{x*— ix+ 3) -/ V(l +*)— V(l — x) 15. / xV(x* + 4x)dx. 18. /— ,, ■ , ;, fe dx. 5. fxV(x* + 4x)dx. l8 ' I i + V(x—a+ 1) dx V(x—a) + V(x—b) 19. /"-, J V(x— a 7. Remarks on Functions which are not Integrable in Terms of Elementary Functions 1. Definition of Functions by means of Integrals. Elliptic Integrals. With the above examples of types of functions which can be integrated by reduction to rational functions, we have practi- cally exhausted the list of functions which are integrable in terms of elementary functions. Attempts to express general integrals such as f-T, i r 1 ^' f V(°o + a i x + • • • + a n x n )dx J VK + a i x + • • • + a«x n ) J or I —dx in. terms of elementary functions have always ended J x in failure; and in the nineteenth century it was finally proved that it is actually impossible to carry out these integrations in terms of elementary functions. If, therefore, the object of the integral calculus were to inte- grate functions in terms of elementary functions, we should have come to a definite halt. But such a restricted object has no intrinsic justification; indeed, it is of a somewhat artificial nature. We know that the integral of every continuous function exists and is itself a continuous function of the upper limit, and this fact has nothing to do with the question whether the integral can be expressed in terms of elementary functions or not. The distinguishing features of the elementary functions are based on the fact that their properties are easily recognized, that their ap- plication to numerical problems is often facilitated by convenient tables or, as in the case of the rational functions, that they can easily be calculated with as great a degree of accuracy as we please. IV] NON-INTEGRABLE FUNCTIONS 343 Where the integral of a function cannot be expressed by means of functions with which we are already acquainted, there is nothing to hinder us from introducing this integral as a new " higher " function in analysis, which really means no more than giving it a name. Whether the introduction of such a new func- tion is convenient or not depends on the properties whioh it possesses, the frequency with which it occurs, and the ease with which it can be manipulated in theory and in practice. In this sense the process of integration therefore forms a basis for the generation of new functions. After all, we are already acquainted with this principle from our dealings with the elementary functions. Thus we found ourselves obliged (p. 167) to introduce the previously unknown integral of 1 /z as a new function, which we called the logarithm and whose properties we could easily determine. We could have introduced the trigonometric functions in a similar way, making use only of the rational functions, the process of integration, and the process of inversion. For this purpose we need only take one or other of the equations . f x dt r* dt arc tanx = / or arc sina; = / ■'ol + C Jo V(l - ^ as the definition of the function arc tana; or arc sin* respectively, in order to arrive at the trigonometric functions by inversion. By this process the definition of these functions is separated from geometry, but we are naturally left with the task of develop- ing their properties, also independently of geometry.* The first and most important example which leads us beyond the region of elementary functions is given by the elliptic integrals. These are integrals in which the integrand is formed in a rational way from the variable of integration and the square root of an expression of the third or fourth degree. Among these integrals the function U(s)= I ; turns out to be of particular importance. Its inverse function s(u) plays a correspondingly important part. In particular, for * We shall not go into the development of these ideas here. The essential step is to prove the addition theorems for the inverse functions, i.e. for the sine and the tangent. 244 THE INTEGRAL CALCULUS [Chap. k = we obtain m(s) = arc sin a; and s(u) = sinw respectively. The function s(u) has been as thoroughly examined and tabu- lated as the elementary functions. This, however, leads us away from the line of the present discussion and into the realm of the so-called elliptic functions, which occupy a central position in the theory of functions of a complex variable. Here we shall merely remark that the name " elliptic inte- gral " arises from the fact that such integrals enter into the problem of determining the length of an arc of an ellipse. (Cf. Chap. V, p. 289.) We may further point out that integrals which at first sight have quite a different appearance turn out after a simple substitution to be elliptic integrals. As an example, the integral dx I V (cos a — cos x) is transformed by means of the substitution u = cos - into the integral du , 1 ifc : -kV2 f du J V(l-u 2 )(l- V(l- u 2 )(l- JbV) cos a/2 dx V (cos 2x) the integral / by means of the substitution u = sin x becomes /du V(l - u 2 )(l - 2m 2 )' and finally the integral / -; ; — J V(l — A 2 sin 2 a:) is transformed by the substitution u = sin x into du I V(l — M 2 )(l - fc 2 « 2 ) 2. On Differentiation and Integration. Another remark on the relation between differentiation and integration may be inserted here. Differentiation may be considered a more elementary process than integration, since it does not lead us away from the domain of known functions. On the other hand, we must remember that the differentiability of an arbitrary continuous function is by no means a foregone conclusion, but a very stringent additional assumption. We have, in fact, seen that there are continuous functions which IV] NON-INTEGRABLE FUNCTIONS 245 are non-differentiable at isolated points, and we may mention without proof that since Weierstrass' time many examples have been constructed of continuous functions which do not possess a derivative anywhere at all.* (There is therefore much less in the mathematical definition of continuity than simple intuition would lead us to suppose.) In contrast to this, even though in- tegration in terms of elementary functions is not always possible, in all circumstances we are certain at least that the integral of a continuous function exists. Taken all in all, we see that integration and differentiation cannot be simply classified as more elementary and less elementary, but that from some points of view the one and from other points of view the other should be thought of as the more elementary. In so far as the concept of integral is concerned, we shall see in the next section that it is not closely bound up with the assumption that the integrand is continuous, but that it may be extended to wide classes of functions with discontinuities. 8. Extension op the Concept of Integral. Improper Integrals 1. Functions with Jump Discontinuities. In the first instance we see that there is no difficulty in ex- tending the concept of integral to the „. case where the function to be in- tegrated has jump discontinuities at one or more points in the interval of integration. For we need only take the integral of the function as the sum of the integrals over the Fig- 4— The integral of a _ . - . 1-1.1 discontinuous function separate sub-intervals in which the function is continuous, f The integral then retains its intuitive meaning as an area (cf. fig. 4). *Cf. Titchmarsh, The Theory of Function* (Oxford, 1932), §§ 11-21-11-23 (pp. 350-354). f We should really observe that in our previous definition of integral we took the interval as closed and the function as continuous in the closed interval. This gives us no trouble, since in each closed sub-interval we can extend the function so that it is continuous by taking for the value of the function at the end-point the limit of the function as x approaches the end-point from the interior of the interval. 246 THE INTEGRAL CALCULUS [Chap. 2. Functions with Infinite Discontinuities. It is quite a different matter when the function has an infinite discontinuity in the interior of the interval or at one of its ends. In order even to formulate the notion of integral in this case we must introduce a further limiting process. Before stating the general definition we shall illustrate some of the possibilities by means of examples. dx We begin with the integral / — , J x* where a is a positive number. The integrand ljx* becomes infinite as x -*• 0, and we therefore cannot extend the integral to the lower limit 0. We can, however, try to find what happens when we take the integral from the positive limit e to the limit 1, say, and finally let z tend to 0. According to the elementary rules of integration, we obtain, provided a* 1, /• t ^=_l_(l_ el -« ) . J, x* 1 - a We immediately recognize that the following possibilities occur: (1) a is greater than 1; then as e -»• the right-hand side tends to 00 : (2) a is less than 1; then the right-hand side tends to the limit 1/(1 — a). In the second case, therefore, we shall simply take this limiting value as the integral between the limits and 1. In the first case we shall say that the integral from to 1 does not exist. (3) In the third case, where a = 1, the integral will be equal to — logs and therefore as e -»• it approaches no limit, but tends to 00 ; that is, the integral from to I does not exist. Another example of the extension of the integral of a function up to an infinite discontinuity is given by the integrand — ; ; . We find that V (1 — a?) r 1 -' dx . ,. . aro sin (1 — e). I '0 v a - **) If we let e tend to 0, the right-hand side converges to a definite limit, tc/2; /•l tint we therefore call this the value of the integral / -7 , even Jo v (1 — x 3 ) though the integrand becomes infinite at the point x = 1. In order to extract a perfectly general concept from these examples, we notice in the first place that it clearly makes no essential difference whether the discontinuity of the integrand lies at the upper end or the lower end of the interval of in- tegration. We now make the following statement: IV] IMPROPER INTEGRALS 247 // in an interval a ^ x ^ b the function f(x) is continuous ' f(x)dx as a the limit lim / f(x) dx r — ^.0 •* n -O^a — where the point b — e approaches the end-point b from the in- terior of the interval — provided that such a limit exists. In this case we say that the improper integral I f(x)dxccm- " a verges. If, however, no such limit exists, we say that the integral / f(x)dx does not exist or does not converge or that it diverges. Fig. 5. — To illustrate the convergence or divergence of improper integrals An analogous definition holds for the case where the lower limit of the interval of integration, and not the upper, is the exceptional point. Even improper integrals can be interpreted as areas. In the first instance, of course, there is no sense in speaking of the area of a region which extends to infinity; yet one may attempt to define such an area by means of a passage to the limit from a bounded region with a finite area. For example, the above results for the function \jx°- imply that the area bounded by the sr-axis, the line x = 1, the line x — e, and the curve y = lfx* tends to a finite limit as e -> 0, provided that a < 1, and that it tends to infinity if a Si 1. This fact may be simply expressed as follows: the area between the a;-axis, the y-axis, the curve, and the line x = 1 is finite or infinite according as a < 1 or a Ji 1. Intuition can, of course, give us no precise information about the finiteness or infiniteness of the area of a region stretching to infinity. Of such a region we can only say that the more closely its sides approach one another the more likely it is to have a finite area. In this sense fig. 5 illus- 248 THE INTEGRAL CALCULUS [Chap. trates the fact that for a < 1 the area under our curve remains finite, while for a ^ 1 it is infinite. In order to find out whether a function /(a;) which has an in- finite discontinuity at the point x= b can be integrated up to b, we can often save ourselves a special investigation by using the following criterion: Let the function f(x) be positive* in the interval a ^ x ^ b, and let lim f(x) = 00 . Then the integral ff(x) dx converges if there exist both a positive number ju. less than 1, and a fixed number M independent of x, such that everywhere in the interval M a<Lx<b the inequality /(a;) ^ is true; in other words, (6 — xf if at the point x = b the function f (x) becomes infinite of a lower order than the first. On the other hand, the integral diverges, if there exist both a number v 22 1 and a fixed number N, such that everywhere in the interval a ^ x < b the inequality f(x) ;> is true; in other words, if at the point x = b J K ' - (j _ x y J r the function f(x) becomes infinite of the first order at least. The proof follows almost immediately by comparison with the very simple special case discussed above. In order to prove the first part of the theorem we observe that for < e < b — a we have dx. As e -> the integral on the right, which is obtained from the integral j— (p. 128) by a simple change of notation, has a limit, and therefore remains bounded. Moreover, the values of [ f(x)dx increase monotonically as e-»-0; since they are also Ja -b bounded, they must possess a limit, and the integral J f(x) dx therefore converges. The parallel proof of the second part of the theorem is left as an exercise for the reader. * In the Appendix to Chap. VIII (p. 418) we shall see that this restriction of sign can easily be removed. IV] IMPROPER INTEGRALS 249 We likewise see at once that exactly analogous theorems hold where the lower limit of the integral is a point of infinite dis- continuity. If a point of infinite discontinuity lies in the interior of the interval of integration, we merely use this point to divide the interval into two sub-intervals and then apply the above considerations to each of these separately. As a further example we consider the elliptic integral \\ dX (P<1). Jo V(l — x^il — kV) From the identity 1 — x* = (1 — x)(l + x) we see at once that as x -*■ 1 the integrand becomes infinite only of order \, whence it follows that the improper integral exists. 3. Infinite Interval of Integration. Another important extension of the concept of integral consists in taking one of the limits of integration as infinite. In order to make this extension precise, we introduce the following notation: if the integral f(x)dx, J where a is fixed, tends to a definite limit when A increases posi- tively beyond all bounds, we denote the limit by / f(x)dx, Ja and call it the integral from a to 00 of the function f{x). Of course such an integral does not necessarily exist or, as we often say, converge. Simple examples of the various possibilities are again yielded by the functions f(x) = 1/a", /"J-_L. M w_ 1} . Jl x* 1 — a Here we see that, if we again exclude the case a = 1, the integral to infinity exists for the case a > 1, and in fact Jf dx on the contrary, when a < 1 the integral no longer exists. For the case a = 1 the integral again clearly fails to exist, since log* tends to in- finity as x does. We see, therefore, that with regard to integration over an 9* (1798) 250 THE INTEGRAL CALCULUS [Chap. infinite interval the functions l/x* do not behave in the same way as for integration up to the origin. This statement also is made plausible by a glance at fig. 5, For we see that the larger a is, the more closely do the curves draw in towards the r-axis when x is large, so that we can readily suppose that the area under consideration tends to a definite limit for sufficiently large values of a. The following criterion for the existence of an integral with an infinite limit is often useful. We again assume that for suffi- ciently large values of x, say for x S: a, the integrand has always the same sign, which without loss of generality we can choose to be positive.* Then we have the following statement: /•CO The integral / f (x)dx converges if the function f (x) vanishes at J a infinity to a higher order than the first, that is, if there is a number v > 1 such that for all values of x, no matter how large, the M relation <.f(x) ^ — is true, where M is a fixed number inde- x" pendent of x. Again, the integral diverges if the function remains positive and vanishes at infinity to an order not higher than the first, that is, if there is a fixed number N > such that xf(x) S; N. The proof of these criteria, which runs exactly parallel to the previous argument, can be left to the reader. 1 — dx(a > 0). The integrand a «* vanishes at infinity to the second order. As a matter of fact, we see at C A 1 11 once that the integral does converge, for / — dx = , and therefore J a x 2 a A f Ja — dx = -. 3? a Another equally simple example is f Jo -dx= lim (arc tan.4 — arc tan 0) = Jo 1 + a; 2 a->* 2 i. The Gamma Function. A further example of particular importance in analysis is offered by the so-called gamma function l dx (n>0). • As we shall see in the Appendix to Chap. VIII (p. 418), this restriction of Bign can easily be removed. IV] IMPROPER INTEGRALS 251 Here also the criterion of convergence is satisfied; e.g. if we choose v = 2, we have lima;'' . e~ <c x n ~ x = 0, since the expo- se— >-» nential function e~" tends to zero to a higher order than any power 1/a;"* (m > 0). This gamma function, which we can think of as a function of the number n (not necessarily an integer), satisfies a remarkable relation, which we can arrive at in the following way by integration by parts. To begin with, we have f(T*x n - 1 dx = — e-*x n ~ x + (» — l)fe- x x n - 2 dx. If we take this formula between the limits and A and then let A increase beyond all bounds, we immediately obtain r» = (n — 1) Ce-*x n -*&x = (n — l)F(n — 1), Jo and by this recurrence formula, provided /x is an integer and < n < n, T(n) = (n — 1) (n — 2) . . . (w — p.) f e-'x n -> L - 1 dx. Jo In particular, if w is a positive integer, we have r(w) = (n — l)(n - 2) . . . 3 . 2 . 1 f e~*dx, Jo r°° and since / e~*dx = 1, •'0 it follows finally that r» = (n — l)(ra — 2) . . . 2 . 1 = (» — 1)! This expression of a factorial by an integral is of importance in many applications. The integrals / e-*' dx, J a"*-** dx Jo Jo also converge, as we may easily convince ourselves by means of our criterion. 5. The Dirichlet Integral. A convergent integral, important in many applications, whose convergence does not follow directly from our criterion, and which is a simple case of a type investigated by Dirichlet, is i=r s ™dx. Jo X 252 THE INTEGRAL CALCULUS [Chap. This integral is easily seen to be convergent if the upper limit is finite, for -*■ 1 as x -*■ 0. Its convergence in the infinite interval is due to the periodic change of sign of the integrand, which causes the contributions to the integral from neighbour- ing intervals of length -n almost to cancel one another. In order to make use of this fact we write the expression in the form „ psinas. Dab = / dx •>A X ^"•""sina:, f B+,r sina; 7 , ^'siat = fSmx dx __ fSmx dx fSmt Ja x Jb x Ja+k t introduce in the last of the three integrals on the right the new variable x=t — v, whence sini = — since, and obtain r^+'g^x r B+ "smXj r B sina; , D AB = / ax — i ax — / ax. J A X J B X J A X-\- IT Addition of this to the original expression for D AB gives us /-*+«• sin a, r B+ *smx, , r B sina; , 2JJ AB = I ax — / ax-\- tt\ ax. Ja x j b x Ja x(x-\- it) s(a; + ff) Prom this it follows, if we assume that B > A > 0, that 2tt , r B dx x*'' \2DAB\< 2 i+«f A E for we may nse the method of p. 127, observing that 1 < sina; < 1 x x ~ x , 1 sina; and = S> x 2 x (x + 77) x 2 for positive values of x. The integral on the right is conver- gent, by our criterion, and our formula shows that | D AB I -*■ as A and B both tend to infinity. Now I A)B Doa I = I Dab |> and it follows from Cauchy's convergence test that D 0B tends to a definite limit as B -> 00. In other words, the integral I 253 IV J IMPROPER INTEGRALS exists. Another proof of this is given in the Appendix to Chap. VIII (p. 418), and on p. 450 we shall further show that I has the value 7t/2. 6. Substitution. It is obvious that all rules for the substitution of new variables, &c, remain valid for convergent improper integrals. As an example, in order to calculate J xe~**dx we introduce the new variable u = x 2 and obtain ° / xe-^dx = ^ / e -"du = lim - (1 — e~ A ) = - Another example of the use of substitution in the investigation of improper integrals is given by the Fresnel integrals, which occur in the theory of diffraction of light: Fi=J o sm(a5 2 )<fc, F 2 =fcoa(a?)dx. The substitution x 2 = u yields Fl =\r^ du , F,=ir™*. Integrating by parts, we have r B sinw du = cos .4 _ cosg _ 1 /- B cosw , As A and B tend to <» the first two terms on the right tend to 0, and by the criterion of p. 250 the integral also tends to 0. Hence by the same argument as for the Dirichlet integral we see that the integral F t converges. The convergence of the integral F 2 is proved in exactly the same way. These Fresnel integrals show that an improper integral may exist even although the integrand does not tend to zero as x -+ oo . In fact, an improper integral can exist even when the integrand is unbounded, as is shown by the example .00 / 2u coa (u*)du. When u l = nn, i.e. when w = Vw, n = 0, 1, 2, . . . , the in- a 5 4 THE INTEGRAL CALCULUS [Chap. fcegrand becomes 2V ntt cos rnr = +2Vn77, so that the integrand is unbounded. By the substitution w 2 — x, however, the integral is reduced to / cos(a; 2 )&, Jo yhich we have just shown to be convergent. By means of a substitution an improper integral may often be transformed into a proper one. For example, the transfor- mation x = sin u gives r 1 dx r'j * •A) ^(1 — x 2 ) J o 2 On the other hand, integrals of continuous functions may be transformed into improper integrals; this occurs if the trans- formation u = <f> (x) is such that at the end of the interval of integration the derivative <£'(«) vanishes, so that dxjdu is infinite. Examples Test the convergence of the improper integrals in Ex. 1-11: i. r 3 ^ 2 . f 1 *l % 3- /•• _&_ J-B x" J-i -tyx ./_„ 1 + x* 1"° dx r" dx J (1 + x)Vx' J 1 — cos x f B dx 6. / =, where Oj, o 2 , a a , a t are J a V(x— a^(x — a 2 )(x — a z ){x — a 4 ) all difierent and lie between A and B. „ r m arc tan x , „ r x arc tan x , 7. / — dx. 8. / -ax. i 1+J^ J 1-x* ' — dx. 10. / — dx. 11. / log t&nxdx. 1 1 - e* J e* - 1 Jo 12.* Prove that / sin* it ( x + - ) \dx does not exist. 13.* Prove that lim f° — ^— = 0. *->■ J 1 + hx w X* X^ — ^ /*00 gjji 3. rfa;, (6) / dx conver- se, - * + * •'O ** 15.* Does / dt converge? Jo 1 + t IV] IMPROPER INTEGRALS 355 16.* (a) If a is a fixed positive number, prove that lim / dx = 7t. (6) If f(x) is continuous in the interval -lgigl, prove that h lim f 1 h* + x* f(x) dx = tc/(0). Miscellaneous Examples Evaluate the integrals in Ex. 1-7: 1. I e* lcsiax dx. 2. I Bin 3 xcoa*xdx. (By a shorter method than that of the text, using trigonometrical identities.) 3. [(logxfdx. 4. [ ***?" . 5. [Vl-e-^dx. 6. f xe-*'*™'*dx. 7. r 2 - sin (a; - -) dx. 8.* Prove that lim <r*' f*e*'dt = 0. *— >*> Jo 9. Assuming that | a | 4= | P l> prove that 1 r T lim — / sin otx sin $x dx — 0. r.7 /o 10. Evaluate / ~ a?er<* cos2a:<fo;. 1 f x*er-»'< 11.* Prove that the substitution x = — i-?, where cc8 — yp 4= 0, transforms the integral ^ ./ Vaa: 4 - <fa Voa; 4 + bx 3 + ex 2 + dx + e into an integral of similar type; and that if the biquadratic ax* + bx* + ex* + dx + e has no repeated factors, neither has the new biquadratic in t which takes its place. Prove that the same statements are true for fB(x, Vox* + bx 3 + ex 1 + dx + e)dx, where ij is a rational function. 256 THE INTEGRAL CALCULUS [Chap. 12. Find the limit as n -*■ co of a n = — — -\ — - +... + —. n + 1 » + 2 2m 13* Find the limit of V»» — Vn 1 — 1 Vn* — 4 V» s — (» — l) 3 14.* Prove that lim "Inl 1 15.* If a is any real number greater than — 1, evaluate .. 1" + 2« + 3* + . . . + W liro »+i Appendix to Chapter IV The Second Mean Value Theorem of the Integral Calculus The method of integration by parts affords us an easy method for proving an important theorem on the estimation of integrals, usually called the second mean value theorem of the integral calculus. Let us suppose that the function <f>(x) is monotonic and con- tinuous in the interval a^x^b, and that the derivative <f>'(x) is continuous; and let us further suppose that/(x) is an arbitrary function continuous in the same interval. Then the second mean value theorem of the integral calculus is expressed as follows. There exists a number £ , such that a ^ | ^ b, for which fj(x) <j>(x)dx=<f> (a) jj{x) dx+<f> (b) f ( "f(x) dx. To prove this we notice first that we can assume that <f>(b) = 0; for replacing </>(x) by <f>(x) — <f>(b) changes both sides of the equation by the same amount, and gives us a function which vanishes at x = b. Moreover, we can assume that <j>(a) > 0; for if <f>(a) < we need only replace ^>(x) by — (f>[x), which changes the sign of both sides of the equation. (The case ^(a) = is trivial; for if both (f>(a) and <f>(b) vanish, <f>(x) must be identically zero, and our equation becomes = 0.) We therefore IV] SECOND MEAN VALUE THEOREM 257 need only prove that if <f>(x) is continuous and monotonic de- creasing, and <£(&) = 0, then J f(x)<j} (x) dx= <f> (a) f f(x) dx. We now put F(x) = / f(x)dx and apply the formula foi integration by parts to the left-hand side of the last equation; we then have \f(x) <j>(x)dx=F(x)<f>(x)\ + f F (x) {—<f>'(x) } dx. Ja \ a Ja The integrated part vanishes, since F(a) and <£(&) are zero. The expression — <f>'(x) is everywhere positive, so that we can apply the first mean value theorem of the integral calculus. We thus find that the integral on the right has the value F(£) £{-*' (*)} dx > a^i^b. But F{$) = f f(x)dx and f {— <f>' {x)}dx = 0(a) — 0(6) = <j>(a) and our theorem is established. This theorem can be extended (although we shall not carry out the proof) to more general classes of functions. For the theorem remains true for all continuous monotonic functions <£(:r), whether they have derivatives or not. In fact, it is true for any discon- tinuous monotonic function for which we are in a position to integrate f(x) <f> (x). CHAPTER V Applications In this chapter, after disposing of a few preliminaries, we shall illustrate how what we have now learned may be applied in a great variety of ways in geometry and physics. 1. Eepresentation of Curves 1. Parametric Representation. As we saw in Chap. I (p. 17), in representing a curve by means of an equation y = f(x) we must always restrict ourselves to a single- valued branch. Hence it is often more convenient — when we are dealing with a closed curve, in particular — to introduce other analytical methods of representation. The most general and at the same time the most useful representation of a curve is parametric representation. Instead of considering one of the rectangular co-ordinates as a function of the other, we think of both the co-ordinates x and y as functions of a third independent variable, a so-called parameter; the point with the co-ordinates x and y then describes the curve as t traverses a definite interval. Such parametric representations have already been encountered. For example, for the circle x 2 + y 2 = a 2 we obtain a parametric representation in the form x = a cost, y=asin.t. Here, as we already know, t has the geometrical meaning of an angle at the centre of the circle. For the ellipse a; 2 /a 2 + y 2 /b 2 == 1 we like- wise have the parametric representation x = a cost, y= b smt, where t is the so-called eccentric angle, that is, the angle at the centre corresponding to the point of the circumscribed circle lying vertically above or below the point P(a cost, bsint) of the ellipse (fig. 1). In both these cases the point with the co- ordinates x, y describes the complete circle or ellipse as the parameter t traverses the interval from to 2tt. 258 [Chap. V] PARAMETRIC REPRESENTATION 259 In general, we can seek to represent a curve parametrically by taking that is, by considering two functions of a parameter t ; the shorter notation x(t) and y(t) will henceforth be used where there is no danger of confusion. For a given curve these two functions <j>(t) and tfi(t) must be determined in such a way that the totality of pairs of functional values x(t) and y(t) corresponding to a given interval of values of t gives all the points on the curve and no points that are not on the curve. If a curve is in the first instance given in the form y—f(x), we can arrive at a representation of this kind by first writing x = <f>(t), where <f>(t) is any continuous mono- tonic function which in a definite interval passes exactly once through each of the values of a; in question; it then follows that y=f{<f>(t)}, that is, the second function ip(t) is deter- mined by compounding/ and <f>. We thus see that owing to the arbitrariness in the choice of the function cf> we have a great deal of freedom in representing a given curve parametrically; in particular, we may actually take t = x and may thus think of the original representation y=f(x) as a parametric repre- sentation with the parameter t = x. The advantage of the parametric representation is that this arbitrari- ness may be utilized for purposes of simplification. For example, we repre- sent the curve y = -y/i 8 by taking x = P, y — t l , so that ip(t) = f, ty(t) = t s . The point with the co-ordinates x, y will then describe the whole curve (semicubical parabola) as t varies from — 00 to +00. If, on the other hand, a curve is originally given by a para- metric representation x = <f>(t), y = tfi(t), and we wish to obtain the equation of the curve in non-parametric form, that is, in the form y = f(x), we have only to eliminate the parameter t from the two equations. In the case of the parametric representations of the circle and ellipse given above we can do this at once h\ Fig. I 260 APPLICATIONS [Chap. squaring and using the equation sin 2 * + cos 2 * = 1. (For a further example see below.) In general, we should have to find an expres- sion for t from the equation x = <f>(t) by means of the inverse function t = <J>(a;) and substitute this in y = tfi(t), in order to obtain the representation* y = ^t{0(a;)}=/(a;). In such an elimination, of course, we must ordinarily restrict ourselves to a portion of the curve; in fact, to a portion which is not cut twice by any line parallel to the y-axis. The parametric representation has associated with it a de- finite sense in which the curve is described, corresponding to the direction in which the values of the parameter increase; this direction we shall call the positive sense. If, for example, the point x = x(t), y — y(t) describes a curve G as t traverses an interval t ^ t ^ ^ and the end-points P and P 1 of the curve correspond respectively to * and t v then the curve is traversed positively in the direction from P to P v If we introduce t = — t as a new parameter, the curve C will correspond to the values — ^ ^ t ^ — t of the variable t, and the points P and P x will correspond to t = — 1 and t = — ty respectively. If we now traverse the curve from P to P ± we proceed in the direction in which the values of the parameter t decrease, that is, in the negative sense. In general, a change of parameter t = t(r) pre- serves the sense in which the curve is described if the function t(r) is monotonic increasing, but reverses it if the function t[r) is monotonic decreasing. 2. Interpretation of the Parameter. Change of Parameter. In many cases we can give an immediate physical interpre- tation to the parameter t, namely, time. Any motion of a point in the plane may be expressed mathematically by the fact that the co-ordinates x and y appear as functions of the time. These two functions therefore determine the motion along a path or trajectory in parametric form. As an example of this we have the cycloids which arise when a circle rolls along a straight line or another circle. Here we limit ourselves to the simplest case, in which a circle of radius a rolls along the ai-axis, and * It may happen, however, that the equation y — f(x) obtained in this way represents more than the original parametric representation. Thus for example the equations x = a sin t, y ■= b sin t represent only the finite portion of the line y — bx/a lying between the points x — —a, y — —b and x = o, y — b, whereas the equation y — bxla represents the whole of the line. V] PARAMETRIC REPRESENTATION 261 we consider a point on its circumference. This point then describes a " common " cycloid. If we choose the origin of the co-ordinate system and the initial time in such a way that for time t = the corresponding point of the curve coincides with the origin, we obtain (cf. fig. 2) the parametric representation x = a(t — sin<), y = o(l — cos*) for the cycloid; here t denotes the angle through which the circle has turned from its original position; in the case where the velocity of rolling is uniform it is proportional to the time. Fig. 2. — Cycloid By eliminating the parameter t we can obtain the equation of the curve in non-parametrio form, at the cost, however, of neatness of expression. We have a — « . cost = -, t = arc cos «^-.,_ ± ^i-(-iE}, and hence x = a arc cos - =F V {{2a — y)y), a thus obtaining x as a function of y. In the parametric representation of a given curve we have a great deal of freedom in the choice of parameter (p. 259). For example, instead of the time t we could take the quantity r = ft as parameter, or indeed any arbitrary quantity t which is related to the original parameter t by an arbitrary equation of the form t = w(t), where we assume that for the whole interval of values of t considered this function has a unique inverse t = k(t). If increasing values of t correspond to increasing values of t, the positive sense of description remains the same; otherwise it is reversed. Parametric representation is, of course, not limited to rect- angular co-ordinates, e.g. it can just as well be used with the polar co-ordinates r and 6, which are connected with the rect- angular co-ordinates by the well-known equations a;=rcos#, 262 APPLICATIONS [Chap. y= r sin 5, or r = \/(3? + y z ), sin<?= y/r, cos#= xjr; the equations of the curve would then be r = r(t), 8 = 6(t). As an example, the straight line may be represented parametrically (see fig. 3) in the form cos< 6= a + t (p and a. being constants), from which we immediately obtain the equation of the line in polar co-ordinates, P Fig. 3 cos (9 — a) by eliminating the parameter t. 3. The Derivatives for a Curve Represented Parametrically. If on the one hand a curve is given by an equation y =/(#), and on the other hand it is given parametrically by x = x(t), V = y(0> * nen we must have y(t) = f{x(t)}. By the chain rule for differentiation it follows that or dy dt' dy dx dx dt y' = ,<h = y dx x where as an abbreviation for differentiation with respect to the parameter t we use a dot over the variable (Newton's notation), instead of the dash '; the latter we shall reserve for differen- tiation with respect to x. For the oyoloid, for example, we have t A ■= a(l — coat) = 2a sin 8 -, i) = a shut = 2a sin - cos -. 2 2 These formnlse show that the cycloid has a cusp with a vertical tangent at the points t = 0, ±27t, ±4rt, ... at which it meets the z-axis, for on approaching these points the derivative y 1 = #/as — cot(</2) becomes infinite. At these points y is equal to 0; everywhere else y > 0. V] PARAMETRIC REPRESENTATION 263 The equation of the tangent to the curve is (£ — x)y — (v — y)* = 0, where g and rj are the "current" co-ordinates, that is, the variable co-ordinates corresponding to an arbitrary point on the tangent. For the equation of the normal, i.e. the straight line through a point of the curve perpendicular to the tangent at that point, we likewise obtain (£— x)x + ( v — y)y=0. The direction cosines of the tangent, that is, the cosines of the angles a, /? which the tangent makes with the x and y axes respectively, are given by the expressions cos a = ±V(* 2 + y z ) cos)8 = y ±V(* 2 + f)' as we may verify by elementary methods. The corresponding direction cosines of the normal are given by cos a = -y ™ff- x (See fig. 4.) These formulae show us that at every point at which & and y are continuous and x 2 + y 2 4= the direction of the tangent varies continuously with t. This is the most important case for us; it is interesting, however, to illustrate by examples the various possibilities that arise when our assumptions are not fulfilled and we cannot state directly that the tangent keeps on turning continuously. At a point at which x = y = the tangent may or may not turn continuously. As one example we have the curve x = t s , y = t 2 discussed on pp. 99, 259, which has a cusp at the origin even though x and y are continuous everywhere. As another example we consider the curve x = 1?, y = i 8 , which is the straight line y = x. This curve has the same tangent direc- Fig. 4. — Direction cosines of the tangent and the normal 364 APPLICATIONS [Chap- tion everywhere; the latter is therefore continuous, although the derivatives x and y both vanish for t = 0. Moreover, at a point at which x and y are discontinuous the direction of the tangent may or may not be continuous. For let </>(t) be any continuous monotonic increasing function, denned for t x ^ t ^ t % , which has a sharp corner at t — t z , t^ <t z <.t 2 . Then the curve * = t> y = 4(t)> which is the same curve as y — <f>(x), has a sharp corner at x = t z ; while the curve x = <f>(t), y = <f>(t), which is a segment of the straight line y = a;, has a constant tangent direction, even though the derivatives x and y do not exist at t = t 3 . This indicates that if we wish to investigate the behaviour of the tangent at a point where our theorem does not apply, we should first use the formulae to find cos a or cosjS as functions of t and then investigate these direction cosines themselves. From a well-known formula in trigonometry or analytical geometry we find that the angle between the two curves repre- sented parametrically by x = x 1 (t), y=y\ (t) and x== x 2 (t), y—Vi (0 respectively (that is, the angle between their tangents or normals) is given by the expression cosS = - *A + ft& . ±V(*l 2 +y 1 2 )V(*2 2 +2/2 2 ) The indeterminacy of the signs of the square roots in the last few formulae suggests that the angles are not completely determined, since we can still specify either sense of direction on the tangent or normal as " positive ". Taking the square root as positive, as is usually done, corresponds to choosing for the positive direction on the tangent the direction in which the parameter increases, and for the positive direction on the normal the direction obtained by rotating the tangent through an angle 7r/2 in the positive * sense. d?v The second derivative y" = — " is obtained in the following CLX way by means of the chain rule and the rule for differentiating a quotient: v " = ^L = ^ — = - (t\ \ = *$ ~ y* I dx dt dx dt \x/ x a? £ * I.e. in the counter-clockwise sense. V] PARAMETRIC REPRESENTATION 265 whence «" = *k = £ y—y £ y dx* x* ' 4. Change of Axes for Curves Represented Parametrically. If we rotate the axes through an angle a in the positive direc- tion, the new rectangular co-ordinates |, rj and the old ones x, y are related by the equations x= |cosa — 17 sina, £= a;cosa-{- y sina, y = |sina + -q cosa, 17 = — a; sina + y cosa. Thus the new co-ordinates £ and 17 are specified along with x and y as functions of the parameter t. By differentiation we at once obtain £ = £ cosa — 1) sina, £ = a cosa -f- y sina, y = £ sina -f- 17 cosa, r;= — jc sina + y cosa. Let us suppose that the curve is given in polar co-ordinates and that both polar co-ordinates and rectangular co-ordinates are given as functions of a parameter t. Then by differentiation with respect to t we obtain from the equations a;=rcos0, y = r sin 6 the formulae a5=rcos0 — rsin0.0, ) . s . f . . . (o) y = f sin<?-}- r cos0 . 8, > which are frequently used in passing from rectangular co-or- dinates to polar. As an example we consider the polar equation of a curve, r = /(#), which might, for example, arise from a para- metric representation r = r(t), d = 6(t) by elimination of the parameter t. The angle ift between the radius vector to a point on the curve and the tangent to the curve at that point is then given by We can convince ourselves of this in the following way. If we think of the curve as given by an equation y = F(x) and use 6 as parameter, so that 0=1 and r =f'(0), we have , v f tan 6 + r tana = w = - = -— x f — rtan0 266 APPLICATIONS [Chap. (cf. fig. 5 and equations (a) above). In addition, ifi — a — 0, and hence _ y 1 — tan 9 _ r + f tan 2 6 _ r ~~ 1 + y'tsmO ~ r+ rtan 2 ~~ r H tan^r ■■ Fig. s This formula can also be established by geometrical methods. 5. General Remarks. In discussing given curves we sometimes consider properties which do not assert anything about the form of the curve itself, but merely something about the position of the curve with respect to the co-ordinate system; for example, the occur- rence of a horizontal tangent, expressed by the equation y = 0, or the occurrence of a vertical tangent, expressed by x = 0. Such properties do not persist when the axes are rotated. In contrast to this, a point of inflection will still be a point of inflection after the axes have been rotated. According to the formula on p. 265 the condition for a point of inflection is xy — xy = 0. If on the left we replace the expressions A, y, if, y' by their values in terms of the new co-ordinates f , rj, we readily obtain xy — xy = £j — £rj . Hence from the equation xy — xy = it follows that £77 — £17 = 0, so that our equation expresses a property of the point of the ourve which is independent of the co-ordinate system. We shall often see later that properties which are truly geometrical are expressed by formulae which are unaltered in form by rotation of the axes. V] PARAMETRIC REPRESENTATION 267 Examples 1. Find the equation, in non-parametric form, of the curve x = acos28 cos 6 y = a cos 26 sin 6. 2. A oirole 0, of radius r, rolls on the outside of a fixed circle of radius B. The point P on the circumference of e moves with c, and describes a curve called the epicycloid. Find the parametrio representation of the epicycloid (consider c to rotate with constant velocity, and measure time so that at t = the point P is in contact with the circle O). 3. Sketch the epicycloid for the special case r = B, and find its para- metric equations. (This particular epicycloid is called the cardioid.) 4. If in Ex. 2 the radius r is less than JJ and c rolls inside O, the point P describes a hypocycloid. Find its parametric equations. 5. Sketch the hypocycloid (1) for B — 2r, (2) for B = Zr. 6. Sketch the hypocycloid for B = Ar (the astroid) and find its non- parametric equation. 7. Find the parametrio equations for the curve X s + y* = 3axy (the folium of Descartes), choosing as parameter t the tangent of the angle between the z-axis and the radius vector from the origin to the point (*. y)- 8. Find the formula for the angle a between two curves r = /(0) and r = 0(8) in polar co-ordinates. 9. Find the equation of the curves which everywhere intersect the straight lines through the origin at the same angle a. 10. Let G be a fixed curve and P a fixed point with co-ordinates x , y . The pedal curve of O with respect to P is defined to be the locus of the foot of the perpendioular from P on the tangent to C. Find the parametric representation of the pedal of C if is itself given parametrically by x = f(t),y^g(f). 11. Find the pedal curve of the circle 0, (a) with respect to its centre M , (6) with respect to a point P on its circumference. 12. Find the pedal curve of the ellipse x = a cos 6, y = b sin 8 with respect to the origin. 2. Applications to the Theory ov Plane Curves We shall consider two different kinds of geometrical proper- ties or quantities associated with curves. The first type consists of properties or quantities which depend only on the behaviour of the curve in the small, i.e. in the immediate neighbourhood of a point, and which can be expressed analytically by means of the derivative at the point. Properties of the second type depend on 268 APPLICATIONS [Chap. the whole course of the curve or of a portion of the curve, and are expressed analytically by means of the concept of integral. We shall begin by considering properties of the second type. 1. Orientation of Areas. The idea of area was our starting-point for the definition of the integral; but the connexion between definite integral and area is still somewhat incomplete. The areas with which we are concerned in geometry are bounded by given closed curves; on f{x)dx is bounded only in part by the given curve y = /(»), the rest of the boundary consisting of lines which depend on the choice of the co-ordinate system. If we wished to determine the area interior to a closed curve, such as a circle or ellipse, by means of integrals of this type, we should have to use some such device as breaking up the area into several parts, each of which is bounded by a single- valued branch of the curve and also by the as-axis and the corresponding ordinates. For the discussion of this general case it is convenient first to make some remarks on the determination of the sign of the area considered. For any surface bounded by an arbitrary closed curve which does not intersect itself, we can relate the sign of the area to the purely geometrical idea of the sense in which the curve is described, according to the following con- vention. We say that the boundary of a region is described in the positive sense if we go round the boundary in such a direction that the interior of the region is on the left;* the opposite sense we call negative. If then we consider a region whose boundary is traversed in an assigned sense, a so-called oriented region, we reckon the area as positive if this sense is positive, and negative if this sense is negative (cf. fig. 6). Suppose, in particular, that in the interval a ^ x ^ b the function f{x) is everywhere positive. We consider the closed curve obtained by starting at the point x = b = x lt y = 0, traversing the a>axis back to the point x = a = x , y = 0, then * If we wish to avoid the words " right " and " left " in such a context, we say that the triangle, whose vertices in order are the origin, the point x — 1, y = 0, and the point x = 0, y ** 1, is described in the positive sense if the vertices are passed in the order mentioned. For every other region, we say that the boundary is positively described if it ia described in the same sense as this triangle; otherwise it is negatively described. V] THEORY OF PLANE CURVES 269 proceeding along the ordinate to the curve y = f{x), then along this curve to the ordinate x = b, and finally along this ordinate to the x-axis (cf. fig. 7). The absolute value of the area interior to this curve — the number of square units contained in it — is, as we know, / f(x)dx. Hence, denoting by A^ the area with its sign as determined above, the integral gives us the value A 01 except for sign. To determine the sign we need only observe n n X, Fig. 6. — A positive area Fig. 1 that the boundary of the region is traversed in the negative sense, so that A 01 is negative; hence we have A 01 = — / f(x)dx. "a Similarly, if a > b, we find that according to our convention A 01 is positive, while the integral / f(x) dx is negative; hence in either case A^ is given by the above equation. 2. The General Formula for the Area as an Integral. After these preliminaries, the difficulties mentioned at the beginning can now be avoided in a simple way by representing our curve parametrically. If we introduce t formally as a new independent variable in the above integral, writing x = x (t), V = y W =f{ x (0}» w e have 4>i =-/V (*)*(«)*, where t and ^ are the values of the parameter corresponding to the abscissse x = a and a^ = b respectively. Here we suppose that the branch in question of the curve y —f(x) is related to 370 APPLICATIONS [Chap. an interval t g t ^ t± by a(l, 1) correspondence,* that f(x) is everywhere positive, and that x (t) never vanishes in this interval. As we have seen, our expression then gives us the area of the region bounded by the curve, the lines x =*= a and x = b, and the a;-axis. It is, of course, still subject to the disadvantages mentioned above. We shall now show that if the curve x = x (t)> H=y (*)> *o ^ ' Sa h> i s a closed curve bounding a region of area A 01 , the area A 01 is given by an integral which in form is exactly the same as the preceding. Let us then consider a closed curve which is represented para- metrically by the equations x== x(t), y = y{t), the curve being B A B Fig. 8. — Area of a closed curve A described just once as t describes the interval t ^ t 5g t v In order that the curve may be closed it is essential that x(t ) = x(t±) and y(t ) = y{t-j). We shall assume that the deri- vatives are continuous except for a finite number of jump-dis- continuities at most, and that x 2 + y z is different from zero except perhaps at a finite number of points which may be corners f of the curve. We shall first consider a closed curve which has no corners and is convex and of such a type that no straight line intersects it in more than two points. We denote by P x and P 2 the points at which the curve possesses a vertical tangent; these tangents are said to be " lines of support " at P x and P 2 respectively, because the points of the curve in the neighbourhood of P 2 and P 2 lie entirely on one side of the line. We can then (cf. fig. 8) * I.e. is such that every point of it corresponds to a single value of t in the interval < £ t S t v and conversely. t A continuous curve x — x(t), y — y(t) is said to have a corner at t ■= t if the positive direction of the tangent approaches a limit as (t — <„)-* through positive values, and approaches a limit as (( — t ) -*■ through negative values, but the two limits are not the same. V] THEORY OF PLANE CURVES 271 regard the area bounded by the curve as the sum of the area A 12 bounded by the closed curve P X MP 2 ABP X , formed as in the pre- ceding section, and the area A 21 bounded by the closed curve P^NP^AP^. Here we assume that the curve is described in the positive sense, as in the figure; by our sign convention A n is then positive and A a negative. We suppose that the point x (t), y{t) describes the upper part of the curve from P 1 to P 2 as t goes from t to t, and the lower part from P 2 to P x as t goes from t to t v We then immediately obtain A ia =-f y(t)x(t)dt and A il =-f' l y(t)x(t)dt; hence, for the total area bounded by the convex curve, we have Mi A=— y(t)x(t)dt. J t, If we denote by " absolute area " of a region the number of square units contained in it— which is, of course, never negative — then the above expression always gives us the absolute area bounded by the curve, except perhaps for sign. In order to see what happens when we reverse the sense in which the curve is described, we simply take the same integral from ^ to t instead of from t to < x ; our integral becomes — / yxdr, which is equal to —A. We thus recognize the truth of the follow- ing statement: The area represented by our formula is positive or negative, according as the sense in which the boundary is described is positive or negative.* * In drawing the figure we have assumed that y > f or all points of the curve. This really does not restrict the generality of the result. For if we dis- place the curve through a distance a parallel to the y-axis, without rotating it, in other words, replace y by y + a, the area is unchanged; the value of the integral is likewise unaltered, for the above integral is replaced by - f "' (y + a)x(t)dt, •>t, and since the curve is closed f'axdt = a{x(t x ) - x(t„)\ = 0. 272 APPLICATIONS [Chap. Two simple observations enable us to extend our results. Firstly, our formula remains valid for closed curves which do not intersect themselves, even when they are not convex, but have a more general form as illustrated in fig. 9. Secondly, the derivatives may have jump discontinuities or may both vanish at a finite number of points, which may represent corners; according to Chap. IV, § 8, p. 245, the function yx remains integrable. (The ordinate to a corner-point is considered to be a line of support if the curve in the neighbourhood of the point lies entirely to one side of the ordinate). We assume that the curve has only a finite number of lines of support, corre- sponding to the points P x , P 2 , . . . , P„, and we subdivide the Fig. 9 curve into the single-valued branches P X P 2 , ■ . ■ , P n -iP n > P n Pi- Then as in fig. 9 we obtain the area bounded by the curve in the form A = A Xi + A 23 + . . . + A n - X , „ + A nl . (See fig. 9, which illustrates this for the case n = 6.) If we express each of these portions of area parametrically and combine the expressions into a single integral, we find that the area bounded by the curve is given by the expression — yxdt, which as before has the same sign as the sense in which the boundary curve is traversed. Our formula even gives us the area, in a certain sense, in the case where the curve intersects itself. But we shall not enter into such a discussion here; the reader may if he wishes turn to § 2 of the appendix to this chapter (p. 311). V] THEORY OF PLANE CURVES 273 We can express our formula for the area in a more elegant symmetrical form if we first transform the integral by integration by parts: / yxdt= — xydt+xy Jt, Jt, Since the curve is closed, x(t ) = x(t 1 ), y(t ) = «/&), and therefore * A — — I yxdt = I xydt. Jt, Jt, If we form the arithmetic mean of the two expressions we obtain the symmetrical form 1 f* 1 A = — - (yx — xy)dt. 3. Remarks and an Example. In connexion with these expressions we must make a remark of a fundamental nature. Both the proof and the statement of the formulae depend on a particular system of rectangular co-ordinates. But the value of the area, a purely geometrical quantity, cannot depend on the particular co-ordinate system chosen. It is therefore important to show that our integrals are unaltered in value by a change of co-ordinates. If the axes are merely displaced without rotation the integrals are clearly unaltered (see the footnote on p. 271). Let us then suppose that the axes are rotated through an angle a; instead of x and y we now have new variables £ and 17, denned by the equations x = $ cos a — 17 sin a, y = g sin a + i) cos a, the new variables being also functions of the parameter t. If we recall that x — £ cos a — r\ sin a and y = £ sin a + i] cos a, a short calculation gives us yx — xy = ij£ — &, so that A = ~ I /"V - ay)* = - 5 Phi ~ ft)*- * Instead of finding the second expression for the area by integration by parts, we could have derived it by using the fact that as regards the definition of area the K-axis and the y-axis are interchangeable, except that the sense of rotation which brings the a;-axis into the y-axis in the shortest way is opposite to the sense -which brings the j/-axis into the a-axis in the shortest way. 10 (B798) *74 APPLICATIONS [Chap. This equation expresses the fact that the area is independent of the co-ordinate system. Our integral expression for the area is also independent of the choice of parameter. For suppose that we introduce a new parameter r by the equation t = r(t); we have dx dx dr dy _ dy dr dt dr dt' dt dr dt' so that -jC(»5-s)*-r('s-s)s* where t and r x are the initial and final values of the new parameter, corresponding to the parametric values t and ^ respectively.* As an example of the application of our formula for the area we con- sider the ellipse y = -V(o a — a: a ). In order to find its area we take a the tipper and lower halves of the ellipse separately and in this way express the area by the integral 6 r +a b r +a , 2 - / V(a*-x?)dx. a J - a If, however, we use the parametric representation x= acoet, y = b sin<, we find immediately that the area is given by the expression i-2ir ah I sin 2 1 dt. ■f This can be integrated as on p. 216; it has the value dbit. * In this section we have based the definition of the area on the concept of integral and have shown that this analytical definition has a truly geometrical character, since it yields a quantity independent of the co-ordinate system. It is, however, easy to give a direct geometrical definition of the area bounded by a closed curve which does not intersect itself, as follows: the area is the upper bound of the areas of all polygons lying interior to the curve. The proof that the two definitions are equivalent is quite simple, but will not be given here. V] THEORY OF PLANE CURVES *75 4. Areas in Polar Co-ordinates. For many purposes it is important to be able to calculate areas using polar co-ordinates. Let r =f(0) be the equation of a curve in polar co-ordinates. Let A{0) be the area of the region which is bounded by the x-axis (that is, the line 8 = 0), the line through the origin making an angle 8 with the z-axis, and the portion of the curve between these two lines. Then A\8) = \r\ For if we consider the radius vector corresponding to the angle 6 and that correspond- ing to the angle 8 + A0, and denote the smallest radius vector in this angular interval (cf. fig. 10) by r and the greatest by r l5 the sector lying between the radius vector 8 and the radius vector 8 + A0 will have an area AA which lies between the bounds £r o 2 A0 and \r^A8. Consequently 1 r 2<A4 1 , 2 ° -A0-2 ri ' and on passing to the limit as A8 -*■ 0, we obtain the relation given above. By the fundamental theorem of the integral cal- culus, the area of the sector between the polar angles a and /? is then given by the expression 1 rP 2 I 0< Fig. io. — Element of area in polar co-ordinates . r*d8. 2. If /? > a, this expression cannot be less than zero. Since we readily see that as 8 increases the point with co-ordinates (r, 8) describes the boundary of the region in the positive sense, this is in agreement with our previous convention for sign. As an example, let us consider the area bounded by one loop of a lemnia- cate. The equation of the lemniacate (of. p. 73) is r % = 2a a cos26, and we obtain one loop by letting 6 vary from — - to +-. This gives us the expression J-tH cob 20 (ft 276 APPLICATIONS [Chap. for the area. This can be integrated at once by introducing the new vari- able u = 29; we find the value of the integral to be a 2 . 5. Length of a Curve. Another important geometrical concept associated with a curve leads to an integration. This is the length of arc. We shall first explain geometrically how we are led to a definition of the length of an arbitrary curve. The elementary process of measuring a length consists of comparing the length to be measured with rectilinear standards of length. The sim- plest method is to apply our standard length to the curve, with its ends on the curve, and count the number of times that we have to repeat the process in order to pass from the beginning to the end of the curve; we can refine the method as required by using smaller and smaller standards of length. By analogy with this elementary intuitive idea, we set up the definition of the length of a curve in the following manner. We suppose that our curve is given by the equations a; = x(t), y = y(t), a^ti== ft. (This includes curves in the form y =f(x), since these can be written y=f(t), x= t.) In the interval between a and fi we choose points t = a, ^, t 2 , . . . ,t„= j8, in that order. The points on the curve corresponding to these values t„ we join in order by line segments, thus obtaining part of a polygon inscribed in the curve; we now measure the perimeter of this polygon. This length will depend on the way in which the points t v , or, as we may also say, the vertices of the polygon, are chosen. We now let the number of the points t v increase beyond all bounds, in such a way that the length of the longest sub-interval in the interval a ^ t 5S ft at the same time tends to 0; this makes the number of sides of our polygon increase without limit, while the length of the longest side tends to 0. The length of the curve is then denned to be the limit of the perimeters of these inscribed polygons, provided that such a limit does exist and is independent of the particular way in which the polygons are chosen. It is only when this assumption that the limit exists (assumption of rectifiability) is fulfilled that we can speak of the length of the curve. We shall soon see that very wide classes of curves can be proved to be rectifiable. To express the length analytically by an integral, in fact, we think of the curve as represented in the first instance by a function V] THEORY OF PLANE CURVES 277 y = f(x) with, a continuous derivative y 1 . By the points a = a^, x 2 , . . . , x n = 6 we divide up the interval a 5j x ^ 6 of the x-axis, over which our curve lies, into (n — 1) sub-intervals of lengths Aa^, . . . , ^x„- v In the curve we inscribe a polygon whose vertices lie vertically above these points. The total length of this inscribed polygon is given according to Pythagoras' theorem (cf. fig. 11) by the expression 'Ay; kAa;,, : u^(m Ax„. But by the mean value theorem of the differential calculus the difference quotient Ay y /Ax v is equal to /'(£,), where g y is an Fig. zx. — Rectification of curves intermediate value in the interval Ax„. If we now let n increase beyond all bounds and at the same time let the length of the longest sub-interval Aa;„ tend to zero, then by the definition of integral our expression will tend to the limit Since this passage to the limit always leads us to the same result, namely, the integral, no matter how the subdivision of the interval is made, we have established the following theorem: Every curve y = f (x) for which the derivative f'(x) is con- tinuous is a rectifiable curve, and its length between x = a and x = b (b Sg a) is given by the formula s(o,6) = /V(l + y' 2 )^. a7» APPLICATIONS [Chap. If by s we denote tie length of arc measured from an arbitrary fixed point to the point with abscissa x, the above equation gives us the following expression for the derivative of the length of arc with respect to x: g-Vfl + rt Our expression for the length of arc is still subject to the special and artificial assumption that the curve consists of one single-valued branch above the x-axis. Parametric represen- tation frees us from this restriction. If a curve of the kind which we have been considering is given in parametric form by the equations x = x(t), y — y(t), then by introducing the parameter t in the above expression we obtain the parametric form of the length of arc ,(«,/?) =/"V(* 2 +y 2 )<&, •'a where a and /S are the values of t which correspond respectively to the points of the curve x = a and x = b. This parametric expression for the length of a curve has a considerable advantage over the previous form in that it is not restricted to single-valued branches of curves, represented by the equation y —f(x), but instead holds for any arbitrary arcs of curves, including closed curves, provided that the derivatives x and y are continuous along the arcs. We recognize this most easily if we go back again to the formula for the length of the inscribed polygon. We suppose that along the arc x and y are continuous. As in the definition, we subdivide the interval a^-t^ft by points t = a, <!,..., t n = /?, with the differences At v and use the corresponding points on the curve as vertices of an inscribed polygon; in the passage to the limit w-><» we assume that the greatest difference Ai„ tends to 0. If we now write the length of the polygon in the form .!/'--'+«=5,V{(©'+(t)>- we see at once that this sum tends to the integral / \/(x 2 -\- y 2 ) dt; we need only recall the generalized method of formation of an V] THEORY OF PLANE CURVES 279 integral (cf. p. 133). If the curve is composed of several arcs of this type, which may join one another at corners, the expression for the length of the curve is simply the sum of the corresponding integrals. Collecting the results, we have the following statement: If in the interval a 5S t ^ /? the functions x(t) and y(t) are continuous and their derivatives ±(t), y(t) are also continuous, except perhaps for a finite number of jump discontinuities, the arc of x = x (t), y = y (t) has a length given by the expression f V(* 2 + f)dl, where this integral, if necessary, is to be taken as an improper integral in the sense of Chap. IV (p. 245). In virtue of this for- mula, in which a must be less than ft, there is a meaning in ascribing a negative length, given by the same formula, to an arc of a curve traversed in the direction in which the value of the parameter t decreases. The sign of the length of arc therefore depends on the choice of the parameter. If we introduce a new parametric expression for the same curve which does not re- verse the sense of description, that is, if we introduce a new para- meter by the equation r = r(t), where dr/dt > 0, we see a priori that our integral formula should give the same value no matter whether t or t is used as parameter; for the two integrals give the length of the same curve and must therefore be equal. This, however, may also be verified directly, for We now give the expression for the length of arc when the curve is expressed in polar co-ordinates. In the last expression we have only to substitute for * and y their values as given in formula (a) on p. 265 in order to obtain & + f = f 2 + rW, whence »(a,j8) = fy/{r*+rW)dt. 280 APPLICATIONS [Chap. If we now change over from the parametric expression to the equation in the form r=/(0), by introducing as parameter t = 8 itself, so that 6 = 1, we have the expression for the length of arc. A simple example of the explicit calculation of the length of an arc is given by the parabola y = - x 2 ; for its length of arc we immediately obtain r b 2 the integral / V(l + x 2 )dx, which with the substitution x = sinhtt becomes a Xar sinh b \ par sinh b ] cosh a wefot=- / (l + cosh2tt)cfot=- (tt + sinhu coshu) _ * sinh a " ^ar sinh a ^ ar sinho arsinha so that the length of arc of the parabola between the abscissae x= a and x = 6 is given by the expression «(o, 6) = - {ar sinh6 + b V(l + 6 a ) — ar sinho — aV(l + a 2 )}. For the catenary y = cosh a; we find that /*b pb a(a,b)= I V(l + ainh i x)dx~ I coshiia;, or s (a, 5)= sinh 6 — sinho. Finally, let it be noted that in many cases it is convenient to introduce as parameter the length of arc reckoned from some fixed point P on the curve, that is, to take x= x(s) and y=y(s). Points of the curve on opposite sides of P will cor- respond to values of s with opposite signs. In this case we have x z + f -<*}'-'• whence by differentiation xx + yy = 0; these two relations find frequent application. 6. Curvature of a Curve. The area and the length of arc of a curve depend on the complete course of the curve. We now insert a discus- sion of a concept which has reference only to the behaviour of a curve in the neighbourhood of a point, namely, the curvature. V] THEORY OF PLANE CURVES 281 If we think of the curve as described uniformly in the posi- tive sense, in such a way that equal lengths of arc are passed over in equal periods of time, the direction of the curve will vary at a definite rate, which we take as a measure of the curva- ture of the curve. If, therefore, we denote the angle between the positive direction of the tangent (p. 264) and the positive .c-axis by a, and if we think of a as a function of the length of arc s, we shall define the curvature k at the point corresponding to the length of arc s by the equation k = da/ds. We know that a = arc tan y', and hence by the chain rule da da . ds y" 1 ds~ dx~dx~ l + y' z ' V(l + y' 2 ) (where the positive sign of the square root means that increasing values of x correspond to increasing values of s). The curvature is consequently given by the expression (l + y' 2 ) s ' r Using the parametric formulae for y 1 and y" we obtain the following simple expression for the curvature of a curve repre- sented parametrically: , x y — yx which, of course, can also be found directly from the equation a = arc tan - = arc cot -. x y In contrast with the previous expression, which is dependent on the equation y=f(x) and consequently involves a special assumption about the position of the arc with respect to the £-axis, the parametric expression for the curvature holds for all arcs along which x, y, x, and y are continuous functions of t and £ 2 + y 2 4= 0. In particular, it holds for points where x = 0, i.e. where dyjdx becomes infinite. If we introduce the length of arc s as parameter and recall that x 2 + y 2 = 1 and xx -f- yy = 0, we have k = xij — yx = y (x + y \ \ = y - = —%. 10* (1788) 382 APPLICATIONS [Chap We thus obtain a particularly simple expression for the curva- ture. The sign of the curvature is changed if we reverse the sense of description of the curve, that is, if we replace the parameter t or s by the new parameter t = — £ or <r = — s. For then x and y change sign, but not x, y, x 2 or y 2 , as the following simple calculation shows: £.{,„}-**_«,_« *.{.<„>-£[-«{.(„}]— §*-H,<-» (A similar calculation can be made for y.) In the case of the v" expression k = ■■ ■ ? first found, this fact is concealed, since it is natural and customary to think of the curve as de- scribed from left to right, in which case the square root can only be positive. As an example we consider the curvature of a positively de- scribed circle with radius a. If we start from the parametric representation a; = a cost, y — a&m.t, we immediately obtain k = l a The curvature of a positively described circle is therefore the re- ciprocal of its radius. This result assures us that our definition of curvature is really a suitable one; for in the case of a circle we naturally think of the reciprocal of the radius as a measure of the curvature. Let us put />= -. The quantity | p | == ,— -, is generally called k \h\ the radius of curvature of the curve at the point in question. For a given point on the curve, that circle which touches the curve at the point and there has the same sense of descrip- tion and the same curvature as the curve, and, more- over, has its centre on the positive or negative side of the normal according as h is positive or negative, is called the circle of curva- ture corresponding to the point. Let us think of the equation of the circle (or an arc of the circle containing the point in ques- V] THEORY OF PLANE CURVES 283 tion) as written in the form y — g(x). Then at the point in ques- tion we have not only f(x) = g(x) and f'(x) = g'(x), as follows from the fact that the circle and curve touch, but in virtue of the relation f"(x) _, g" (x) V{1+/'W 2 } 3 V{l + 0» 2 } 3 we also have f"(x) = g"(x). The centre of the circle of curvature is called the centre of cwrvatwre corresponding to the given point. Its co-ordinates are expressed parametrically by To prove this we need only make use of the formulae for the direction cosines of the normal, on which the centre of curvature lies at a distance 1/| h\ = | p\ from the tangent. These for- mulae give us an expression for the centre of curvature in terms of the parameter t. As t describes its range the centre of curva- ture describes a curve, the so-called evolute of the given curve; and since, with x and y, we have to regard x, y, and p as known functions of t, the formulae above give parametric equations for this evolute. For special examples the reader may be referred to § 3 (p. 287 et seq.) and to the appendix (p. 307 et seq.). 7. Centre of Mass and Moment of a Curve. We now come to some applications which bring us into the realm of mechanics. We consider a system of n particles lying in a plane. Let m^, m 2 , . . . , m n be the masses of these particles, and let y v y 2 , . . . , y n be their respective ordinates. We then call n T = S m v y p = m 1 y 1 + m^y* + . . . + m n y n i- = 1 the moment of the system of particles with respect to the x-axis. The expression 77 = TjM, where M denotes the total mass m 1 + m 2 + . . . + m„ of the system, gives us the height of the centre of mass of the system of particles above the as-axis. We 284 APPLICATIONS [Chap. define the moment with respect to the y-axis and the abscissa of the centre of mass in a corresponding way. We shall now see that this idea can easily be extended to give us a definition of the moment of a curve along which a mass is uniformly distributed, and of the co-ordinates £ and 77 of the centre of mass of such a curve. Merely for the sake of brevity we assume that the density has a constant value, say /x, along the curve; any continuous distribution could equally well be discussed. To arrive at this extension we go back to the consideration of a system of a finite number of particles and then pass to the limit. For this purpose we suppose that the length of arc s is introduced as a parameter on the curve, and that the curve is subdivided by (w — 1) points of division into arcs of lengths As x , As 2 , . . . , As n . The mass pAs { of each arc As< we represent as concentrated at an arbitrary point of the arc, say that with the ordinate y t . By definition the moment of this system of particles with respect to the x-axis has the value T = pXytAst. If now the greatest of the quantities As t tends to 0, this sum tends to a definite limit given by the expression T = ^f'yds = ix f'yVO- + V' 2 ) dx, which we shall therefore naturally accept as the definition of the moment of the curve with respect to the a-axis. Since the total mass of the curve is equal to its length multiplied by ju. : M («i — s ), pi 1 we are immediately led to the following expressions for the co- ordinates of the centre of mass of the curve: / yds I xds » *• t '• V= > S= • These statements are actually definitions of the moment and centre of mass of a curve; but they are such straightforward V) THEORY OF PLANE CURVES 285 extensions of the simpler case of a number of particles that we naturally expect that — as is actually the case — any statement in mechanics which involves the centre of mass or the moment of a system of particles will be valid for curves also. In particular, the position of the centre of mass with respect to the curve is independent of the system of co-ordinates. 8. Area and Volume of a Surface of Revolution. If we rotate the curve y =f(x), for which f(x) ^ 0, about the x-axis, it describes a so-called surface of revolution. The area of this surface, whose abscissae we suppose to lie between the bounds x and a^ > x , can be obtained by a discussion analogous to the preceding. For if we replace the curve by an inscribed polygon, instead of the curved surface we shall have a figure composed of a number of thin truncated cones. Fol- lowing the suggestions of intuition, we define the area of the surface of revolution as the limit of the areas of these conical surfaces when the length of the longest side of the inscribed polygon tends to zero. We know from elementary geometry that the area of each truncated cone is equal to its slant height multiplied by the circumference of the circular section of mean radius. If we add these expressions and then carry out the passage to the limit, we obtain the expression A = 2t7 rVV(! + V' 2 ) dx = 2tt Py ds J x, Ji a for the area. Expressed in words, this result states that the area of a surface of revolution is equal to the length of the curve generating it multiplied by the distance traversed by the centre of mass (Guldin's rule). In the same way we find that the volume interior to the surface of revolution and bounded at the ends by the planes x = x and x = a^ > x is given by the expression V = 77 / y 2 dx. This formula is obtained by following the suggestion of intuition that the volume in question is the limit of the volumes of the above-mentioned figures consisting of truncated cones. The rest of the proof is left to the reader. z86 APPLICATIONS [Chap. 9. Moment of Inertia. In the study of rotatory motion in mechanics an important part is played by certain quantities called moments of inertia. These expressions will be briefly mentioned here. We suppose that a particle m at a distance y from the as-axis rotates uniformly about that axis with angular velocity ca (that is, in unit time it rotates through an angle co). The kinetic energy of the particle, expressed by half the product of the mass and the square of the velocity, is obviously We call the coefficient of Jco 2 , that is, the quantity my 1 , the moment of inertia of the particle about the x-axis. Similarly, if we have n particles with masses m^, m 2 , . . . , m„ and ordinates y v y 2 , . . . , y a we call the expression T^'Zm.y? ■ the moment of inertia of the system of masses about the a;-axis. The moment of inertia is a quantity which belongs to the system of masses itself, without reference to its state of motion. Its importance lies in the fact that if the whole system is set in rigid rotation about an axis, without change of the distances between pairs of particles, the kinetic energy is obtained by multi- plying the moment of inertia about that axis by half the square of the angular velocity. Thus the moment of inertia about an axis plays the same part in rotation about an axis as is played by the mass in rectilinear motion. Suppose now that we have an arbitrary curve y =/(») lying between the abscissae x and x± (> x ), along which a mass is uniformly distributed with unit density. In order to define the moment of inertia of this curve we proceed just as we did in the sub-section 7 (p. 284); as before, we arrive at an expression for the moment of inertia about the :c-axis, namely, t x = /■>«*» = f W(i + y' 2 ) dx. For the moment of inertia about the y-axis we have the corre- sponding expression T v = fV ds = f 'zV(l + y'^dx V] THE CYCLOID 387 3. Examples The theory of plane curves with its great variety of special forms and properties offers us a rich store of examples of these abstract concepts. But to avoid being lost in a mass of detail we must limit ourselves to a few typical applications. 1. The Common Cycloid. From the equations (of. p. 261) x = a(t — sin*), y = ffl(l — cost), we at once obtain x = o(l — cos*), y = a sin*, whence the length of arc is «= f V(x* + y*)dt = /"V {2a 2 (1 — cost)}*. Jq Jo But since 1 — cost = 2 sin 2 - the integrand is equal to 2a sin-, and hence for ^ a g 2tu 2 2 X a t t a / a\ a sin- dt = —4a cos - = 4a 11 — cos - 1 = 8a sin*-. 2 2 2 4 If, in particular, we consider the length of arc between two successive cuspa we must put a = 2tc, since the interval ^ t ^ 2jt of values of the parameter corresponds to one revolution of the rolling circle. We thus obtain the value 8a; that is, the length of arc of the cycloid between suc- cessive cusps is equal to four times the diameter of the rolling circle. Similarly, we calculate the area bounded by one arch of the oycloid and the x-axis: / «= / yx dt = a 2 / (1 — cost) 2 * •'0 »0 /.2ir = a 2 / (1 — 2 cost + cos 2 *) dt Jo «( ~ . < sin 2<\ = a 2 [t - 2 sin* + - + ——) 2ir = 3a 2 *. This area is therefore three times the area of the rolling circle. For the radius of curvature p = 1/iwe have (*' + f) 3 2 = _2aV{2(l-cos*)}= -4a xy — yx . t sin- 2 at the points * = 0, t = ±27t, . . . this expression has the value zero. These are actually the cusps, where the cycloid meets the x-axis at right angles. 288 APPLICATIONS [Chap. The area of the surface of revolution formed by rotating an arch of the cycloid about the x-axis is given according to our formula (p. 285) b y f8a /-2.T t A= 2n I yds = 2n / o(l — cost) . 2a sin - dt J J 2 /.2ir f pit = 8o 2 7i/ sin 3 „ <ft = 16a a rc / sin 3 a aw J 2 J = 16a a 7t / (1 — cos a it) sinudu. Jo The last integral can be evaluated by means of the substitution cosm = v; we find that A = 16a a 7t( — cosm + - cos'm) = . 3 lo 3 As an exercise the reader may calculate for himself the height 7) of the centre of mass of the cycloid above the x-axis, and also the moment of inertia T x . The results are 4 A a m 256 s •n == _ o = — and T m = — a 3 . 1 3 2tm * 15 2. The Catenary. The length of aro of the catenary has already been calculated as an example in the preceding section (p. 280), and we found its value to be » = / cosh x (fa = sink 6 — sinha. Ja For the area of the surface of revolution obtained by rotating the catenary about the x-axis, the so-called catenoid, we find 6 1.1 j o /" 6 l + cosh2x A = 2tc f cosh a xax = 2n f Ja Ja dx = 7t(6 — a + sinh22> — - sinh2a). 2 ^ From this we further obtain the height of the centre of mass of the aro from a to 6: m 6 — o + ~ sinh26 — - sinh2a tj = - — = _ __. Zns 2(sinh6 — sinha) Finally, for the curvature we have y" cosh a; 1 fc = (1 _|_ y '2)3/2 C osh 3 a; cosh a x" V] THE ELLIPSE AND THE LEMNISCATE 289 3. The Ellipse and the Lemniscate. The lengths of arc of these two ourves cannot be reduced to elementary functions, but belong to the class of " elliptic integrals " mentioned on p. 243. For the ellipse y = V (o a — x 8 ) we obtain a aJ V \ a? — 3? ) J V(l— £ 2 )(l-x 2 £>) where we have put x/a = %, 1 — 6 2 /a 2 = x 2 . By the substitution i; = sin <p this integral can be expressed in the form «= fV{a*— (a 3 — & 2 )sin 2 <p}(29 = a f V(l — x"sin 2 <p)d<p. Here, to obtain the semi-perimeter of the ellipse, we must let x traverse the interval from — a to +a, which corresponds to the interval — 1^5^+1 or — re/2 g <p ^ +7t/2. For the lemniscate, whose equation in polar co-ordinates is r 2 = 2a a cos 2t, we similarly obtain « = /V(r> + f*)dt = [Jfa oos2< + 2a 2 !^£f ) <ft ■* •/ Tf \ COS .&{ / = aV2f J dt =aV2f- / * J V(cos2i) J V(l — 2 sin 2 <) If we introduce u = tan* as independent variable in the last integral, we have • i. «* j, du sin a ( = , at ■■ 1 + u a ' 1 + u 1 ' and consequently ,= aV2f , dw . J V(i — « 4 ) In a complete loop of the lemniscate u runs from — 1 to + 1. and the length of arc is therefore equal to a V2 , r+ 1 du Li V(l - u 1 )' a special elliptic integral which played a great part in the researches of Gauss. «90 APPLICATIONS [Chap. Examples 1. Calculate the area bounded by the semicubioal parabola y = **/*. the x-axis, and the lines x= a and x= b. 2. Calculate the area of the region bounded by the line y = x and the lower half of the loop of the folium of Descartes. (Use the parametric representation found in Ex. 7, p. 267.) 3. Calculate the area of a sector of the Archimedean spiral r = o9 (a > 0). 4. Calculate the area of the cardioid (Ex. 3, p. 267), using polar co- ordinates. 5. Calculate the area of the astroid (Ex. 6, p. 267). 6. Calculate the area of the pedal curve of the circle x* 4- y 2 = 1 with respect to a point P (x , 0) on the x-axis. Show that this area is least when P is at the origin. 7. Do the same for the ellipse — + - = 1. a 2 6 2 8. Find the parametric representation of the cardioid when the length of arc is used as parameter. 9. Do the same for the cycloid. 10. Calculate the length of aro of the semicubioal parabola y = a^' 2 . 11. Calculate the length of the astroid. 12. Calculate the length of arc of: (a) The Archimedean spiral r = a9 (a > 0). (6) The logarithmic spiral r = e m6 . (c) The cardioid (Ex. 3, p. 267). (d) The curve r=a(0 i — 1). 13. Find the radius of curvature of (a) the parabola y = a; 2 ; (6) the ellipse x= a cos 9, y = b sin 9, as a function of x and of 9 respectively. Find the maxima and minima of the radius of curvature and the points at which these maxima and minima occur. 14. Sketch the curve rf* ' fins 11. _ /•' s /" cos u , t sm u , x = I — = du, y = I — = du Jo Vu Jq Vu and determine its radius of curvature ( p). 15. Show that the expression for the curvature of a curve x = x(t), y = y (t) is unaltered by rotation of axes and also by change of parameter given by t = 9(f), where 9'(t) > 0. V] THE ELLIPSE AND THE LEMNISCATE 291 16. Let r = /(0) be the equation of a curve in polar co-ordinates. Prove that the curvature is given by the formula k — 2r™ — rr" + r* (r* + r a\3/a ' r' = - d f r", dO 2 ' where 17. Find the volume and surfaoe area of a zone of a sphere of radius r, i.e. of the portion of the sphere cut off by two parallel planes distant A a , ftj respectively from the oentre. 18. Find the volume and surfaoe area of the torus or anchor ring obtained by rotating a circle about a line which does not intersect it. 19. Find the area of the catenoid, the surface obtained by rotating an arc of the catenary y = coshx about the x-axis. 20. Sketch the curve defined by the equations *= fcos(in^)dt, y«= / sin($7rf 2 )<tt. Jo Jo What is the behaviour of the curve as t runs from — 00 to +00T Cak culate the curvature k as a function of the length of arc. 21. The curve for which the length of the tangent intercepted between the point of contact and the y-axis is always equal to 1 is called the tractrix. Find its equation. Show that the radius of curvature at each point of the curve is inversely proportional to the length of the normal intercepted between the point on the curve and the y-axis. Calculate the length of arc of the tractrix and find the parametric equations in terms of the length of arc. 22. Let x — x(t), y = y(t) be a closed curve. A constant length p is measured off along the normal to the curve. The extremity of this seg- ment describes a curve which is called a parallel curve to the original curve. Find the area, the length of arc, and the radius of curvature of the parallel curve. 23. Find the centre of mass of an arbitrary aro (o) of a circle of radius r, (6) of a catenary. 24. Calculate the moment of inertia about the x-axis of the boundary of the rectangle a ^x f^b, a ^ y ^ p. 25. Calculate the moment of inertia of an arc of the catenary y = coshx (a) about the x-axis, (6) about the y-axis. 26. The equation y — f(x) + a, a £x ^ &, represents a family of curves, one for each value of the parameter a. Prove that in this family the curve with the least moment of inertia about the x-axis is that which has its centre of mass on the x-axis. z 9 2 APPLICATIONS [Chap. 4. Some very Simple Problems in the Mechanics of a Particle Next to geometry the differential and integral caleulus are especially indebted to the science of mechanics for their early development. Mechanics rests upon certain basic principles which were first laid down by Newton; the statement of these principles involves the concept of the derivative, and their application requires the theory of integration. Without analysing these basic principles in detail, we shall illustrate by some simple examples how the integral and differential calculus are applied in mechanics. 1. The Fundamental Hypotheses of Mechanics. Here we shall restrict ourselves to the consideration of a single particle, that is, of a point at which a mass m is imagined to be concentrated. We shall further assume that motion can only take place along a certain fixed curve, on which the position of the particle is specified by the length of arc s measured from a fixed point on the curve; in particular, the curve may be a straight line, in which case we use the abscissa x as the co-ordinate of the point instead of s. The motion of the point is determined by expressing the co-ordinate s= <f>(t) as a function of the time. By the velocity of motion we shall mean the derivative <f>'{t), or, as we shall also write, dt <f>'(t) = i. The second derivative, we call the acceleration. In mechanics we start from the assumption that the motion of a point can be explained by means of forces of definite direc- tion and magnitude. Newton's second fundamental law of mechanics may, in the case of motion on our given curve, be expressed as follows: The mass multiplied by the acceleration is equal to the force acting on the particle in the direction of the curve; in symbols ms = F. V] MECHANICS OF A PARTICLE 293 Thus the direction of the force is always the same as that of the acceleration; its direction is that of increasing values of s if the velocity in that direction is increasing, otherwise it is opposed to the direction of increasing values of s. The law of Newton is in the first instance nothing more than a definition of the concept of force. The left-hand side of our equation is a quantity which can be determined by observation of the motion, by means of which we measure the force. But this equation has a far deeper meaning. As a matter of fact, it turns out that in many cases we can determine the acting force from other physical assumptions, without any consideration of the corresponding motion. The above fundamental law of Newton is then no longer a definition of force, but is instead a relation from which we can draw important conclusions about the motion. The most important example of a known force is given us by gravity. From direct measurements we know that the force of gravity acting on a mass m is directed vertically down- v * wards and is of magnitude mg, where "*£?££?££?- the constant g, the so-called gravita- tional acceleration, is approximately equal to 981 if the time is measured in seconds and the lengths in centimetres. If a mass moves along a given curve, we learn by experiment that the force of gravity in the direction of this curve is equal to mg cos a, where a denotes the angle between the vertical and the tangent to the curve at the point under consideration (cf. fig. 12). In the case of motion on our given curve the basic problem of mechanics is as follows: if we know the force acting on the particle (e.g. the force of gravity), we have to determine the position of the point, that is, its co-ordinate s or x, as a function of the time. If we restrict ourselves to the simplest case, in which this force * mf(s) is known at the outset as a function of the length of arc — so that the force is independent of the time — we shall * The separation of the factor m in the expression for the given force is not essential, but makes the formulae simpler. 294 APPLICATIONS [Chap. show how the course of the motion along the curve can be found from the equation m Here we have to deal with a differential equation, that is, an equation from which an unknown function — here s(t) — is to be determined and in which the derivative of this function occurs as well as the function itself (cf. Chap. Ill, § 7, p. 178). 2. Body Falling Freely. Resistance of the Air. In the case of the free fall of a particle along the vertical jr-axis. New- ton's law gives us the differential equation x=g. From this follows x(t) = gt + » , where v is a constant of integration. Its meaning is easily found by putting t = 0. We then find x(Q) = v ; that is, i> is the velocity of the particle at the instant from which the time is reckoned, the initial velocity. By another integration, we obtain x(t) = i£rf 2 + w « + No- where x is also a constant of integration, whose value is again found by putting t = 0; we thus find that x is the initial position, that is, the co- ordinate of the point at the beginning of the motion. Conversely, we can choose the initial position x and the initial velocity » arbitrarily, and then obtain the complete representation of the motion from the equation x = \gt* + v t + x . If we wish to take account of the effect of the friction or air resistance acting on the particle, we have to consider this as a force whose direction is opposite to the direction of motion and concerning which we must make definite physical assumptions.* We shall work out the results of different physical assumptions: (a) the resistance is proportional to the velocity, being given by an expression of the form — rx, where r is a positive con- stant; (6) the resistance is proportional to the square of the velocity, being of the form — rx 1 . In accordance with Newton's law we obtain for the equations of motion (a) mx= mg — r&, (6) mx = mg — rx*. If we at first consider x = u(t) as the function sought, we have x(l) — u(t), so that . . „, , (a) mu — mg — ru, (o) mu = mg — ru a . * These assumptions must be chosen to suit the particular system under consideration; for example, the law of resistance for low speeds is not the same as that for high (e.g. bullet velocities). V] MECHANICS OF A PARTICLE 295 Instead of determining u as a function of t by these equations, we deter- mine t as a function of u, writing our differential equations in the form i„\ dt l ,m * 1 dm, g — ru/m du g — ruf/m With the help of the methods given in the preceding chapter we can imme- diately carry out the integrations and obtain (o) t(u) - - ? log (l - -I u) + «„, r \ mg I (&)«(«)=- J*iog5LzJf + ^ where we have put V(m/rg) = k and where t„ is a constant of integration. Solving these equations for u, we have (6) «(<)= — gk mg f e -2 («-«,)/*_ 1 (O) U(t) = - ^? (e-r(«-«.)/m _ 1)( r e -2(»-«.)/ft + !' These equations at once reveal an important property of the motion. The velocity does not increase with time beyond all bounds, but tends to a definite limit depending on the mass m. For (a) lim u(t) = ^?, (6) lim «(«) = J^i. A second integration, performed on our expressions for u(t) = x with the help of the methods of the preceding chapter, gives the results (which may be verified by differentiation) (a) x(t) = — ge-nt-t.)lm +^£4 + 0, r* r (6) x(t)=- log cosh J r l {t - < ) + c, where c is a new constant of integration. The two constants of integration t and c are readily determined if we know the initial position a(0) = x and the initial velocity x(0) = m(0) = »„ of the falling particle. 3. The Simplest Type of Elastic Vibration. As a second example we consider the motion of a particle which moves along the jr-axis and is pulled back towards the origin by an elastic force. As regards the elastio force we assume that it is always directed towards the origin and that its magnitude is proportional to the distance from the origin. In other words, we take the force as equal to — tec, where the coefficient k is a measure of the stiffness of the elastic connexion. Since 296 APPLICATIONS [Chap. k is assumed positive, the force is negative when x is positive and positive when x is negative. Newton's law now tells ua that mx = — kx. We cannot expect that this differential equation will determine the motion completely, but it is plausible to suppose that for a given instant of time, gay t = 0, we can arbitrarily assign the initial position a;(0) = x„ and the initial velocity x(0) = v ; that is, in physical language, that we can start off the particle from an arbitrary position with an arbitrary velocity and that thereafter the motion is determined by the differential equation. Mathematically this is expressed by the fact that the general solution of our differential equation contains two constants of integration, at first undetermined, whose values we find by means of the initial conditions. This fact we shall prove immediately. We can easily state such a solution directly. If we put to = V(k/m), we may at once verify by differentiation that our differential equation is satisfied by all the functions x(t) = Cj cosui + c 2 sincot, where e t and c 2 denote constants chosen arbitrarily. On p. 297 we shall see that there are no other solutions of our differential equation and hence that every such motion under the influence of an elastic force is given by the above expression. This expression can easily be put in the form x{t) = a sinco(i — 8) = — a sincoS cos cut + a coscoS sin cot; we need only write — a sincoS = c^ and a coscoS = Cj, thus introducing in- stead of Cj and c 2 the new constants a and 8. Motions of this type are said to be sinusoidal or simple harmonic. They are periodic; any state (i.e. position x(t) and velocity x(f)) is repeated after the time T = 2n/a, which is called the period, since the functions sincoi and coscoi have the period T. The number a is called the maximum displacement or amplitude of the oscillation. The number 1/T = co/27t is called the frequency of the oscillation; it measures the number of oscillations per unit time. We shall return to the theory of oscillations in Chap. XI (p. 501). 4. Motion on a Given Curve. Finally, we shall discuss the most general form of the problem stated above, namely, the problem of motion along a given curve under an arbi- trary pre-assigned force mf(s). The point in question here is the determination of the function s(t) as a function of t by means of the differential equation ■*• = /(*), where /(«) is a given function. This differential equation in s can be solved completely by the following device. We begin by considering any primitive function F(s) of /(«), so that V] MECHANICS OF A PARTICLE 297 F '(a) = f(s), and multiply both sides of the equation * = /(«) = F'(a) by a. We can then write the left-hand side in the form — ( - a 2 J, as we see dt \& ' at onoe by differentiating the expression a 2 ; the right side F'{a)a, however, is the derivative of F(a) with respect to the time t, if in F{s) we regard the quantity £ as a function of t. Hence we immediately have dt \2 / dt l ' or by integration h* = F(s) + c, where c denotes a constant yet to be determined. da . Let us write this equation in the form -37 = V2(F(a) + c). We see that from this we cannot immediately find a as a function of t by integration. But we arrive at a solution of the problem if we at first content ourselves with finding the inverse function t(a), that is, the time taken by the particle to reach a definite position a. For this we have the equation dt 1 / da \>2{F(a) + 0} thus the derivative of the function t(a) is known, and we have da V2{F(a) + c} where o x is another constant of integration. As soon as we have performed this last integration we have solved the problem, for while we have not determined the position a as a function of t, we have inversely found the time t as a function of the position a. The fact that the two constants of integration c and c± are still available enables us to make the general solution fit special initial conditions. In the above example of elastic motion we have to identify x with a; we have f{a) = — to 2 s and correspondingly, say, F(a) = — $ to 2 * 2 . We therefore obtain *= 1 da V(2c-o>V)' and further ■-; da V(2c— &)V) This integral, however, can easily be evaluated by introducing o>«/V2c as a new variable; we thus obtain t = — arc sin — ^= + Cj, <a V2c 298 APPLICATIONS [Chap. or, forming the inverse function, a = smco(< — Cj). CO We are thus led to exactly the same statement of the solution as before. From this example we also see what the constants of integration mean and how they are to be determined. If, for example, we require that at the time t = the particle shall be at the point a = and at that instant shall have the velocity «(0) = 1, we obtain the two equations ■\/2c = sincoCj, 1 = V2c coscoc,, CO from which we find that the constants have the values Cj = 0, c = J. The constants of integration and c^ can be determined in exactly the same way when the initial position « and the initial velocity « (at time t = 0) are prescribed arbitrarily. Examples 1. A point A moves with constant velocity 1 on a circle with radius r and centre the origin. The point A is connected to a point B by a line of constant length l(>r); B is constrained to move on the x-axis (cf. the orank, connecting-rod, and piston of a steam engine). Calculate the velocity and acceleration of B as functions of the time. 2. A particle starts from the origin with velocity 4, and under the influence of gravity slides down a straight wire until it reaches the vertical line x = 2. What must the slope of the path be in order that the point may reach the vertical line in the shortest time? 3. A particle moves in a straight line subject to a resistance producing the retardation ku\ where u is the velocity and k a constant. Find ex- pressions for the velocity (w) and the time (<) in terms of a, the distance from the initial position, and v , the initial velocity. 4. A particle of unit mass moves along the x-axis and is acted upon by a force f(x) = — sinx. (a) Determine the motion of the point if at time t = it is at the point x = and has velocity v = 2. Show that as t -*■ 00 the particle approaches a limiting position, and find this limiting position. (6) If the conditions are the same, except that v may have any value, show that if v B > 2 the point moves to an infinite distance as t ->■ 00, and that if v < 2 the point oscillates about the origin. 5. Choose axes with their origin at the centre of the earth, whose radius we shall denote by R. According to Newton's law of gravitation, a partiole of unit mass lying on the y-axis is attracted by the earth with a force — =— jpt where jx is the "gravitational constant" and M is the mass of the earth. V] MECHANICS OF A PARTICLE 299 (a) Calculate the motion of the partiole after it is released at the point y (> B); that is, if at time t = it is at the point y=y a and has the velocity v = 0. (b) Find the velocity with which the particle in (a) strikes the earth. (c) Using the result of (6), calculate the velocity of a particle falling to the earth from infinity, f 6.* A particle of mass m moves along the ellipse r = k/(l — e cos 0). The force on the particle is cm/r 2 directed towards the origin. Describe the motion of the particle, find its period, and show that the radius vector to the particle sweeps out equal areas in equal times. 5. Further Applications: Particle sliding down a Curve 1. General Remarks. The case of a particle sliding along a frictionless curve under the in- fluence of gravity can be treated very simply by the method just described. We shall first discuss this motion in general, and then with special ¥1. reference to the cases of the ordinary pendulum and the oy- cloidal pendulum. We ohoose axes in such a way that the j/-axis points vertically upwards, that is, opposite to the direction of the force of gravity, and consider the curve as given in terms of a parameter 8 by the para- metric equations x = 9(6) = a;(0), y = (^(9) = y(0). A portion of the curve, for which the motion will be studied, is shown in fig. 13. At every point of the curve the force of gravity acts downwards (that is, in the direction of decreasing y) on the particle with magnitude mg. If we denote the angle between the negative y-axis and the tangent to the curve by a, according to the hypothesis stated on p. 293 the force acting along the direotion of the curve is Fig. 13 mg cosa = —mg —. " , where dip ■ _ &ty _ '-I-«w. '= I = +'«»• (Note that here the dash denotes the derivative with respeot to 6, and not with respect to as.) If in particular we introduce the length of arc a t This is the same as the least velocity with which a projectile would have to be fired in order that it should leave the earth and never return. 300 APPLICATIONS [Chap. as parameter in place of 6, we obtain the expression — mg-E for the force ds along the curve. By Newton's law, therefore, the function s{t) satisfies the differential equation dy ds The right-hand side of this equation is a known function of s, since we know the curve and must therefore regard the quantities x and y as known functions of s. As in the last section, we multiply both sides of this equation by i. The left-hand side then becomes the derivative of - s a with respect to t. If in the function y(s) we regard s as a function of t, the right-hand side of our equation is the derivative of — gy with respect to t. On integrating, we therefore have where c is a constant of integration. In order to fix the meaning of this constant, we suppose that at the time t = our particle is at the point of the curve for which the value of the parameter is 6 and the co- ordinates are x = <p(0 o ), y = <l>{%), and that at this instant its velocity is zero, that is, «(0) = 0. Then putting t = we immediately have — 9Va + c = 0, so that -i' = -p(y-y„). Now instead of regarding s as a function of t we shall consider the inverse function t{s). For this we at once obtain dt_ = 1 ds V{2g(y ll -y)}' which is equivalent to /• ds "' J V{2g(y -y)}' where c^ is a new constant of integration. As regards the sign of the square root, which is the same as the sign of i, we notice that if the particle moves along an arc which is lower than y e everywhere except at the ends, the sign cannot change. For the sign of s can change only where i = 0, that is, where y — y = 0. The integrand on the right is known in terms of the parameter 0, since the curve is known. Introducing G as inde- pendent variable, we obtain . fds dQ , T If ** + lf % \ ik 1 " W A V { 2g { y -y )} " * ±H (i^)) ««' where the functions x" = <p'(0), y' = <1>'(Q), y = 4*(9) are known. In order to determine the constant of integration c, we note that for t = the V] PARTICLE SLIDING DOWN CURVE 301 value of the parameter must be 6 . This immediately gives us our solution in the form When integrated this equation represents the time taken by the particle to move from the parameter value 6 to the parameter value 6. The in- verse function 6(() of this function t(6) enables us to describe the motion completely; for at each instant t we can determine the point x = 9{6(t)}, y = <|>{6(f)} which the particle is then passing. 2. Discussion of the Motion. From the equations just found, without an explicit expression for the result of the integration we can deduce the general nature of the motion by simple intuitive reasoning. We suppose that our curve is If •< of the type shown in fig. 14, that is, that it consists of an arc convex downwards; we take « as increasing from left to right. If we initially re- lease the particle at the point A with co-ordinates x = x , y=y<n corresponding to 6= 6 , the velocity increases, for the acceleration * is positive. The particle travels from A to the lowest point with ever-increasing velocity. After the lowest point is passed, however, the acceleration is negative, since the right-hand side —g-f-oi the equation of motion is negative. as The velocity therefore decreases. From the equation i* = — 2g(y — y ) we see at once that the velocity reaches the value when the particle reaches the point B whose height is the same as that of the initial position A. Since the acceleration is still negative, the motion of the particle must be reversed at this point, so that the particle will swing back to the point A; this action will repeat itself indefinitely. (The reader will recall that friction has been disregarded.) In this oscillatory motion the time which the point takes to return from B to A must clearly be the same as the time taken to move from A to B. If we denote the time required for a complete journey from A to B and back again by T, the motion will obviously be periodic with period T. If 6 and 4 are the values of the parameter corresponding to the points A and B respectively, the half-period is given by the expression 2 V2g\Je V \Vo-y I I Fig. 14 J_J f e > // y' a (9)+<]/s(6) \ I V2g\J eB VI <K9o)-<K9)/ r 302 APPLICATIONS [Chap. If 6j is the value of the parameter corresponding to the lowest point of the curve, the time which the particle takes to fall from A to this lowest point is -jLI/*Jp^±*>| 3. The Ordinary Pendulum. The simplest example is given by the so-called simple pendulum. Here the curve under consideration is a circle of radius 2: x = I sin 6, y — — I cos 6, where the angle 6 is measured in the positive sense from the position of rest. From the general expression above we at once obtain t = B. r dQ = F r rfe V g L a V (cosG - cos*) V? J_^ («in«! - sin^)' where a(0 < a < 7t) denotes the amplitude of oscillation of the pendulum, that is, the angular position from which the particle is released at time t = with velocity 0. By the substitution _ sin (6/2) du = cos ( 6/2) sin Ox/2)' 56 _ 2 sin (a/2) our expression for the period of oscillation of the pendulum becomes du r==2 I 1 r 1 VyJ_,V(l-«»: )(1 — tt a sin 2 (a/2)) We have therefore expressed the period of oscillation of the pendulum by an elliptic integral. If we assume that the amplitude of the oscillation is small, so that we may with sufficient accuracy replace the second factor under the square root sign by 1, we obtain the expression 2 a r 1 du NgJ^Vil-v?) as an approximation for the period of oscillation. We oan evaluate this last integral by formula 13 in our table of integrals (p. 206), and obtain the expression 2n ,. /- as an approximate value for T. '9 I. The Cycloidal Pendulum. The fact that the period of oscillation of the ordinary pendulum is not strictly independent of the amplitude of oscillation caused Christian Huygens, in his prolonged efforts to construct accurate clocks, to seek for a curve such that the period of osoillation is strictly independent of V] PARTICLE SLIDING DOWN CURVE 3°3 the particular position on the curve at which the oscillating particle begins its motion.* Huygens recognized that the cycloid is such a curve. In order that a particle may actually be able to oscillate on a cycloid the cusps of the cycloid must point in the direction opposite to that of the force of gravity; that is, we must rotate the cycloid considered pre- viously (p. 261) about the s-axis (cf. fig. 15). We therefore write the equa- tions of the cycloid in the form x — as(6— sinO), y= o(l-f cos0), which also involves a translation of the curve through a distance 2a in the positive ^-direction. The time which the particle takes to travel from a point at the height y = a(l + cosa) (0<a<7t) Fig. 15. — Path described by a cycloidal pendulum down to the lowest point, by the formula worked out on p. 301, is We now use the equation 00s a — cos9 = 2 1 cos 2 - ■ V 2 this gives T _ la VH- cos 1) dO. We then work out the definite integral, making use of the substitution 6 a. . 8 ,„ cos - = u cos - , sin — ao : 2 2 2 * The oscillations are then said to be isochronous. 304 APPLICATIONS [Chap. This gives J . e sin dQ = — 2 / -j-^ = — 2 arc sinu, V( cos 1 _cos 1) V(l — u") and we therefore obtain COS - 2 are sin a eos- 2 -v£ The period of oscillation T, therefore, is actually independent of the ampli- tude a. 6. Work 1. General Remarks. The concept of work throws new light on the considerations of the last section and on many other questions of mechanics and physics. Let us again think of the particle as moving on a curve under the influence of a force acting along the curve, and let us suppose that its position is specified by the length of arc measured from any fixed initial point. The force itself will then, as a rule, be a function of s. We assume that it is a continuous function f(s) of the length of arc. This function will have positive values where the direction of the force is the same as the direction of increasing values of s, and negative values where the direction of the force is opposite to that of increasing values of a. If the magnitude of the force is constant along the path, by the worlt done by the force we mean the product of the force by the distance («!—«„) traversed, where «j denotes the final point and s the initial point of the motion. If the force is not constant we define the work by means of a limiting process. We subdivide the interval from «„ to s 1 into n equal or unequal sub-intervals and notice that if the sub-intervals are small, the force in each one is nearly constant; if <r v is a point chosen arbitrarily in the v-th sub-interval, then throughout this sub-interval the force will be approximately /(c„). If the force throughout the v-th sub-interval were exactly /(a,,), the work done by our force would be exactly S/(a„)As,,, >•- 1 where A«„ as usual denotes the length of the v-th sub-interval. If we now pass to the limit, letting n increase beyond all bounds while the length of the longest sub-interval tends to zero, then by the definition of an integral our sum will tend to W=f H f(8)da, which we naturally call the work done by the force. V] WORK 305 If the direction of the force and that of the motion are the same, the work done by the force is positive; we then say that the force does work. On the other hand, if the direction of the force and that of the motion are opposed, the work done by the force is negative; we then say that work is done against the force* If we regard the co-ordinate of position « as a function of the time t, so that the force /(«) = p is also a function of t, then in a plane with rect- angular co-ordinates s and p we can plot the point with co-ordinates s = s(t), p = p(t) as a function of the time. This point will describe a curve, which may be called the work diagram of the motion. If we are dealing with a periodic motion, as in the case of any machine, then after a certain time T (one period) the moving point s(t), p(t) will return to the same point; that is, the work diagram will be a closed curve. In this case the curve may consist simply of one and the same arc, traversed first forwards and then backwards; this happens, for instance, in elastic oscillations. But it is also possible for the curve to be a more general closed curve, enclosing an area; this is the case e.g. with machines in which the pressure on a piston is not the same during the forward stroke as during the backward stroke. The work done in one cycle, that is, in time T, will then be given simply by the negative area of the work dia- gram, or in other words, by the integral l o+ T ... ds ,, p(t)~dt, at where the interval of time from t„ to t B + T represents exactly one period of the motion. If the boundary of the area is positively traversed the work done is negative, if negatively traversed the work done is positive. If the curve consists of several loops, some traversed positively and some traversed negatively, the work done is given by the sum of the areas of the loops, each with its sign changed. These considerations are illustrated in practice by the indicator diagram of a steam engine. By a suitably-designed mechanical device a pencil is made to move over a sheet of paper; the horizontal motion of the pencil relative to the paper is proportional to the distance s of the piston from its extreme position, while the vertical motion is proportional to the steam pressure, and hence proportional to the total force p of the steam on the piston. The piston therefore describes the work diagram for the engine on a known scale. The area of this diagram is measured (usually by means of a planimeter) and the work done by the steam on the piston is thus found. Here we also see that our convention for the sign of an area, as discussed in § 2, No. 1 of this chapter (p. 271), is not of exclusively theoretical interest. For it sometimes happens when an engine is running light, that the highly expanded steam at the end of the stroke has a pres- * Note that here we must carefully distinguish the force of which we are speaking. For example, in lifting a weight the work done by the force of gravity is negative: work is done against gravity. But from the point of view of the person doing the lifting the work done is positive, for the person must exert a force opposed to gravity. 11 («798) 3©6 APPLICATIONS [Chap. sure lower than that required to expel it on the return stroke; on the diagram this is shown by a positively traversed loop; the engine itself is drawing energy from the flywheel instead of furnishing energy. 2. The Mutual Attraction of Two Masses. Let us suppose that a particle attracts another particle according to Newton's law of attraction; as a first example we shall consider the work done by the force of attraction as the second particle moves along the line joining the two particles. According to Newton's law of gravitation, the attracting force is inversely proportional to the square of the distance. If we imagine the first particle at rest at the origin and the second particle at the distance r from the origin, the attracting force is given by the expression where (i is a positive constant. The work done by this force when the particle moves from the distance r to the distance rj(<r) is therefore positive and equal to the integral rnds /l IN If, by means of an opposing force, the particle is moved farther away from the origin, going from the distance r to a distance r x > r, the work done by the force of attraction will, of course, still be given by this integral (now negative). The work done by the opposing force has the same numeri- cal value, but the opposite sign; it is therefore equal to ix ( - — -). If we \r rj think of the final position as being chosen farther and farther away, this approaches the limiting value ", which we may call the work which must r be done against the force of attraction in order to move the particle from the distance r to " infinity ". This important expression is called the mutual potential of the two particles. Here, therefore, the potential is denned as the work required to separate two attracting masses completely; for example, the work required in order to tear an electron completely away from an atom (ionization potential). 3. The Stretching of a Spring. As a second example we consider the work done in stretching a spring. As is usual in the theory of elasticity, we assume (cf. p. 295 also) that the force needed to stretch the spring is proportional to x, the increase in the length of the spring, that is, p = lex, where k is a constant. The work which must be done in order to stretch the spring from the unstressed position x = to the final position x = x t is therefore given by the integral kx dx = £ kx t % . V] WORK 307 4. The Charging of a Condenser. The concept of work in other branches of physios can be treated in a similar way. For example, we may consider the charging of a condenser. If we denote the quantity of electricity in the condenser by Q, its capacity by G, and the difference of potential (voltage) across the condenser by V, then we know from physics that Q = GV. Moreover, the work done in moving a charge Q through a difference of potential V is equal to QV. Since in the charging of the condenser the difference of potential V is not constant but increases with Q, we perform a passage to the .limit exactly analogous to that on p. 304 and as the expression for the work done in charging the condenser we obtain where Qj is the total quantity of electricity passed into the condenser and Fj is the difference of potential across the condenser at the end of the charging process. Appendix to Chapter V 1. Properties of the Evolute The parametric equations ^ x ~ p VW+f)' v==y+p VWW)' for the evolute of a given curve x = x(t), y = y{t) (cf . p. 283) enable us to deduce some interesting geometrical relations between it and the given curve. For convenience we use the length of arc s as parameter, so that x 2 + y z = 1 and xx + yy = 0, p x y or py = x and px = — y. We thus have i—x—py, tj = y+ px; on differentiation these give i = x— py— 'py=—py, y = y+ p*'+ p*= p&, and therefore £x + i# = 0. 308 APPLICATIONS [Chap. Since the direction cosines of the normal to the curve are given by — y and x, it follows that the normal to the curve is tangent to the evolute at the centre of curvature; or, the tangents to the evolute are the normals of the given curve; or, the evolute is the " envelope " of the normals (cf. fig. 16). If further we denote the length of arc of the evolute, measured from an arbitrary fixed point, by a, we have ©'-'- £ 2 + From the above formulae, since x 2 -f- y 2 = 1, we obtain so that if we choose the direction in which a is measured in a suitable way, 6= p, provided that a 4= 0, or on integration °i — 0b = Pi — Po- That is, the length of arc of the evolute between two points is equal to the difference of the corresponding radii of curvature, provided that p remains different from zero for the arc under consideration. This last condition is not superfluous. For if p changes sign, then from the formula a = p we see that on passing the corre- sponding point of the evolute the length of arc a has a maximum or minimum; that is, on passing this point we do not simply continue to reckon a onward, but must reverse the sense in which a is measured. If we wish to avoid this, on passing such a point we must change the sign in the above formula, i.e. put a = — p. It may also be noted that the centres of curvature which correspond to maxima or minima of the radius of curvature are cusps of the evolute. (The proof will not be given here.) The geometrical relationship which we have just found can be expressed in yet another way. If we imagine a flexible inex- tensible thread laid along an arc of the evolute and stretched so that a part of it extends away from the curve tangentially to it, and if in addition the end-point Q of this thread lies on the original curve C, then as we unwind the thread the point Q will describe the curve C. This accounts for the name evolute V] PROPERTIES OF THE EVOLUTE 309 (eoolvere, to unwind). The curve C is called an involute of the evolute E. On the other hand we may start with an arbitrary curve E and construct its involute C by this unwinding process. We then see that E conversely is the evolute of C. To prove this we consider the curve E, which is now the given curve, as given in the form £ = f (cr), V = V ( a )> where the current rectangular co-ordinates are denoted by £ and 77 and the parameter a is the length of arc. The winding is done as indicated in fig. 17; when the thread is completely wound on to the evolute E, its end Q coincides with the point A of E corre- Fig. 16. — Evolute (B) Fig. 17. — Involute (O sponding to the length of arc o. If the thread is now unwound until it is tangent to the evolute at the point P, corresponding to the length of arc a ^ a, the length of the segment PQ will be (a — a) and its direction cosines will be £ and t), where the dot denotes differentiation with respect to a. Thus for the co- ordinates x, y of the point Q we obtain the expressions x = £+ (a — ct)|, y = t\ + (o — a)i), which give the equations for the involute described by the point Q in terms of the parameter a. By differentiation with respect to a it follows that *=£-£+(«- *)£= («- °)l y = v — v + (° — aYv = (« — a )v- 3io APPLICATIONS [Chap. And since £| + ijij = 0, we at once find that fz -f yy = 0, which shows that the line PQ is normal to the involute C. We can therefore state that the normals to the curve C are tangent to the curve E. But this is the characteristic property of E, the evolute of C. Hence every curve is the evolute of all its involutes. Fig. 18. — The cycloid as evolute and involute As a particular case we consider the evolute of the cycloid x = t — sin*, y-'l — cos*. By pp. 281, 283, % = x- . i a + y* y ...._ , xy — yx y + x * ** + 9\ xy— yx we therefore obtain the evolute in the form $ = < + sin*, yj = — 1 + cos*. If we put < = t + 7t, then 5 — jt = t — sinx and i) + 2 = 1 — cost, and these equations show that the evolute is itself a cycloid which is similar to the original curve, and can be obtained from it by translation, as indicated in fig. 18. As a further example we shall work out the equa- tion for the involute of the circle. We begin with the circle 5= cost, rj = sin* and unwind the tangent, as indicated in fig. 19. The involute of the circle is then given in the form x ■■ cos* + * sin*, y = — sin* + *cos*. „. lit Finally, we shall determine the evolute of the ellipse of the circle x = a cos*, y = 6 sin*. We at once have , . tf a + t a 2 5 = x— $ — " — V xy — yx cos 8 * and 1 = y + *- Jl = - o s -6 s sin 8 *, xy—yx b which is a parametric representation of the evolute. If from these equations V] PROPERTIES OF THE EVOLUTE 3" we eliminate t in the usual way, we obtain the equation of the evolute in non-parametric form: (a£)2/8 + (6t)) 2/3 = (a 2 - 6 2 ) 2/s - This curve is called an astroid. Its graph is given in fig. 20. By means of the parametric equations we may readily convince ourselves that the centres of curvature corresponding to the vertices of the ellipse are actually the cusps of the astroid. Fig. ao. — Evolute of the ellipse Examples 1. Show that the evolute of an epicycloid (Ex. 2, p. 267) is another epicycloid similar to the first, which can be obtained from the first by rotation and contraction. 2. Show that the evolute of a hypocycloid (Ex. 4, p. 267) is another hypocycloid, which can be obtained from the first by rotation and ex- pansion. 2. Abbas bounded by Closed Curves We saw in § 2 (p. 271) that the area bounded by a closed curve x = x (t), y—y (t), t ^ t ^ t^ which nowhere intersects itself (a so-called simple closed curve) is given by the integral -f'y{t)x(t)dt, where the value obtained is positive or negative according as the sense in which the boundary is described is positive or 312 APPLICATIONS [Chap. negative. We shall now extend this result to more general curves. Suppose that the curve C, given by the equation x=x(t), y=y (t), intersects itself in a finite number of points, thus divid- ing the plane into a finite number of portions R 1 , R 2 , .... Sup- pose further that the derivatives are continuous, except perhaps for a finite number of jump discontinuities, and that x 2 -j- y* =j= 0, except perhaps at a finite number of values of t which may correspond to corners. Finally, it is assumed that the curve has a finite number of lines of support (p. 270). To each region R { we then assign an index fi ( denned in the following way: we choose an arbitrary point Q in R iy not lying on any line of support, and erect the line extending from Q y\ -*-x Fig. 21 upwards, in the direction of the positive y-axia. We count the number of times the curve C crosses the half-line from right to left, and subtract the number of times the curve C crosses from left to right; the difference is the index /*<. For example, the interior of the curve illustrated in fig. 6 (p. 269) has the index fi = +1; and in fig. 21 the regions R 1 , . . . , R & have the indices ju.j = — 1, [i 2 — +1> H-3 = +2> p-i = — 2, /x 5 = — 1. This number ^< actually does depend on the region R t and not on the particular point Q chosen in R ( , as we readily see in the following manner. We choose any other point Q' in R t , not on a line of support, and join Q to Q' by a broken line lying entirely in the region R { . As we proceed along this broken line from Q to Q' the number of right-to-left crossings minus the number of left- to-right crossings is constant; for between lines of support the number of crossings of either type is unchanged, while on crossing a line of support either the number of crossings of both types increases by one, or else both the numbers decrease by one; in either case, the difference is unaltered. In the case where the V] AREAS BOUNDED BY CLOSED CURVES 313 line of support meets the curve at several different points, say A,B,...,H,we consider it as several different lines of support, FA, FB, . . . , FH, where F is the point of the a;-axis vertically below all the points A, B, ... , H. Our argument then applies to each of these lines. Hence the number //.< has the same value whether we use Q or Q' in determining it. In particular, if our curve does not intersect itself, the interior of the curve consists of a single region R whose index is +1 or — 1 according as the sense in which the boundary is described is positive or negative. To see this we draw any vertical line (not a line of support) intersecting the curve; on this line we find the highest point of intersection (P) with the curve, and in R we choose a point Q below P and so near it that no point of intersection lies between P and Q. Then above Q there lies one crossing of the curve, which if the curve is traversed positively must be a right-to-left crossing, so that /* = -{-1; otherwise fi = — 1. As we have just seen, this same value of p holds for every other point of R. For such a curve, and in fact for all closed curves, one of the regions, the " outside " of the curve, extends unboundedly in all directions; we see immediately that this region has index 0, and henceforth neglect it. Our theorem about the area is now as follows: the value of the integral — / yx dt is equal to the sum of the absolute areas of the regions R f , each area R ( being counted /x< times; in symbols — / yxdt = S/i 4 1 area R { |. The proof is simple. We assume, as we are entitled to do, that the whole of the curve lies above the x-axis (cf . footnote, p. 271). The lines of support cut R f into a finite number of por- tions; let r be one of these portions. Then on taking the integral — I yxdt for each single- valued branch of the curve, we find that the absolute area of r is counted +1 times for each right-to-left branch over r and — 1 times for each left-to-right branch over r, in all fj. f times. The same is true for every other portion of R ( ; hence R t is counted /*,< times. Thus the integral round the com- plete curve has the value S/ij | area R ( |, as stated. This for- mula agrees with what we have found for simple closed curves, 11* (B798) 3H APPLICATIONS [Chap. V as we recognize from the discussion of the values of ju. for such curves. The definition given for the index fj, t has the disadvantage of being stated in terms of a particular co-ordinate system. As a matter of fact, however, it can be shown that the value of fi t is independent of the co-ordinate system and depends solely on the curve; but we shall not prove this here. CHAPTER VI Taylor's Theorem and the Approximate Expression of Functions by Polynomials In many respects the rational functions are the simplest functions of analysis. They are formed by a finite number of applications of the rational operations of calculation, while in the last resort the formation of every other function involves a more or less concealed passage to the limit from rational functions. The questions whether and how a given func- tion can be expressed approximately by rational functions, in particular by polynomials, are therefore of great importance both in theory and in practice. 1. The Logarithm and the Inverse Tangent 1. The Logarithm. We begin by considering some special cases in which the integration of the geometrical progression leads almost at oncp to the desired approximations. We first remind the reader of the following fact. For q =)= 1 and for positive integers n, we have 1 *,_.■_«, i _„-i . 1-q where r„ = = 1 + q + f + . . . + J"" 1 + r„ If | q | < 1 the remainder r„ tends to as n increases, and we then obtain (pp. 34-35) the infinite geometric series 1 4- a + a 2 + . . . with the sum , . 31 1 — q 316 TAYLOR'S THEOREM [Chap. As our starting-point we take the formula it lo g{ l + x)=f oT + t and expand the integrand in accordance with the above formula, putting q = — t. Then by integration we at once obtain «.2 <r<3 ™i »n iog(i + *) = »- 1 + 1_ *+_... + (_ iri + b., where R n = fr„dt = (—1)" /" n - < n «ft Hence for any positive integer n we have expressed the function log(l + x) approximately by a polynomial of the n-th degree, namely, 2 3 v ' n at the same time the quantity R„, the remainder, specifies the amount of the error made in this approximation. In order to estimate the accuracy of this approximation, we need only have an estimate for the remainder R n ; and such an estimate is given us immediately by the integral estimates on p. 126. If at first we suppose that x ^ 0, then in the whole interval of integration the integrand is nowhere negative and nowhere exceeds t n . Consequently 1 •'0 M + 1 and we therefore see that for every value of x in the interval ^ x £J 1 this remainder can be made as small as we like by choosing n large enough (cf. p. 32). If, on the other hand, the quantity a; is in the interval — 1 < x £J 0, the integrand will not change sign and its absolute value will not exceed 1 1 | n /(l + x), and we thus obtain the estimate for the remainder l E -lSr^X"^-(TT^ n+l + xJo (l + a! )( n+ l) We see, therefore, that here again the remainder is arbitrarily small when n is sufficiently large. Of course, our estimate has no meaning when we put x = — 1. V1J LOGARITHM AND INVERSE TANGENT 317 Summing up, we can say that l0g{l + z)=X-^+^-+... + (-l)r>-l X - + R n , 2 3 n where the remainder R n tends to zero as n increases, provided that x lies in the interval * — 1 < x j£ 1. From the above inequalities we can, in fact, find an estimate for the remainder, independent of x, which is valid for all values of a; in the interval — 1 + h ^ x ^ 1, where h is a number such that < h ^ 1. For then 1 1 l#„ 'hn+l' and this formula shows us that in the whole interval the function log(l + a;) is expressed approximately by our polynomial of the n-th degree, the error being nowhere greater than . h n + 1 We leave it to the reader to convince himself that for all values of x for which | x | > 1 the remainder not only fails to approach zero, but, in fact, increases numerically beyond all bounds as n increases, so that for such values of x our polynomial does not give us an approximation to the logarithm. The fact that in the above interval the remainder R n tends to zero may be expressed by saying that in this interval we have the infinite series f ffZ /y3 rf\ 108(1+*)=*-!+!-!+-... for the logarithm. If in this series we insert the particular value x = 1, we obtain the remarkable formula log 2=1-1+---+— ... 8 2^3 4 This is one of the relations whose discovery made a deep impression on the minds of the first pioneers of the differential and integral calculus. * It is to be noted that this interval is open on the left and closed on the right. I Infinite series will be considered in detail in Chap. VIII (p. 366). 318 TAYLOR'S THEOREM [Chap. The above approximation for the logarithm leads us to another formula which is useful for many purposes, particularly in numerical calculations. Provided that — 1 < x < 1, we have only to write — x in place of a; in the above formula to obtain sp2 /*t3 nA log(l- a3 )=- a! -|-|-|-..._S . Taking n as even and subtracting, we have 1 , 1 + x , x 3 , x 5 . , x n ~ x , « Z 1 — x 3 5 w — 1 where R„ is given by the expression .dt. ~~ Jo 1 — On account of the relation n+ 1 1 — x 2 the remainder tends to zero as n increases, a fact which we again express by writing the expansion as an infinite series: 1, 1 + x . , . x 3 . x 5 . x 1 . - log = ar tanh x= x+ — + — [-—+..., 2 5 1 — x ^3 5 7 for all values of x such that \x\ < 1. An advantage of this formula is that as x traverses the in- 1 -4- x terval from — 1 to 1, the expression ranges over all positive 1 — x numbers. Hence if the value of x is suitably chosen this series enables us to calculate the value of the logarithm of any positive number, with an error not exceeding the above estimate for R n . 2. The Inverse Tangent. We can treat the inverse tangent in a similar way if we begin with the formula, true for every positive integer n, 1 1 + t* 1 — t 2 + t l — + . . . + (— l)n-lf2— 8 + Tn VI] LOGARITHM AND INVERSE TANGENT 319 t 2n where r n = ( — 1)™ By integration, we obtain arc tana; = x— _ + _ + ...+ (— l)"" 1 + R n , 3 5 In — 1 Jrx fin ol + P and we see at once that in the interval — 1 ^ x ^ 1 the re- mainder R„ tends to zero as n increases, for by the mean value theorem of the integral calculus r\x\ I _ I 2n+l \R n \^f t 2n dt = LH 1 " ' —Jo 2n 4- 1 + From the formula for the remainder we can also show fairly easily that for | x | > 1 the absolute value of the remainder increases beyond all bounds as n increases. We have accord- ingly deduced the infinite series arc tana; = x— - + — !-•••» 3 5 valid for I x I ^ 1. For x = 1, since arc tan 1 = - , we have 4 ' 7 =1- 1 + -- + 4 3 + 5 + '--' as remarkable a formula as that previously found for log 2. Examples 1. Prove that x-% + f < log(l + x) < x- * + ** (x > 0). 4 2 3(1 + a;) 2 3 Hence find log- to 2 places. 2. Calculate log - to 3 places, using the series 5 log(l + a;) = x — g + -—.... Prove that the result is accurate to 3 places. 3. How many terms of the series for log(l + x) must be used in order to obtain log(l + x) to within 10 per cent if 30 ^ x ^ 31? 320 TAYLOR'S THEOREM [Chap. 2. Taylor's Theorem An approximate representation by rational functions, as in the special cases above, can also be obtained in the case of an arbitrary function /(a;), about which we assume only that for all values of the independent variable in an assigned closed interval the function possesses continuous derivatives up to the (n + l)-th order at least. In most of the cases which actually occur the existence and continuity of all the derivatives of the function is known to begin with, so that for n we can choose any arbitrary integer. The approximation formula which we shall now derive was discovered in the early days of the differential and integral cal- culus by Taylor, a student of Newton's, and is known as Taylor's theorem *. 1. Taylor's Theorem for Polynomials. In order to get a clear idea of the problem, we shall begin by considering the case where f(x) = a + <h x + t hf^ + • • ■ + °n a! " is itself a polynomial of the ra-th degree. We can then easily express the coefficients of this polynomial in terms of the deri- vatives of f(x) at the point x = 0. For if we differentiate both sides of the equation once, twice, &c, with respect to x and then put x = 0, we at once find that the coefficients are «o =/(0). <h =/'(0), a 2 = I /"(0), .... «„ = I /« (0). Any polynomial f(x) of the n-th degree can therefore be written in the form /(*) =/(0) + */'(0) + |j /"(0) + ^/'"(0) + . . . + ^>>(0). This formula merely states that the coefficients a„ can be ex- pressed in terms of the derivatives at x = and gives the expressions for them. We can generalize this " Taylor series " for the poly- nomial slightly if we replace x by f = x -\- h and consider the function /(£) =/(«+ h) = g(h) as a function of h, for * A special case of this theorem is often referred to, without historical justification, as Maclaurin's theorem. We shall not follow this usage. VI] TAYLOR'S THEOREM 321 the moment thinking of x as fixed and h as the independent variable. It then follows that and hence, if we put A = 0, <7'(0) =/'(*), ..., <? n) (0)=f\x). If we apply the previous formula to the function/(a; + h) = g(h), which is itself a polynomial of the w-th degree in h, we im- mediately obtain the Taylor series h 2 h 3 fit) =/(* + h) =/(*) + */'(*) + I /"(*) + !/"'(*) + . . . 2. Taylor's Theorem for an Arbitrary Function. These formulae suggest that we should seek a similar formula in the case of an arbitrary function f{x), not necessarily a polynomial; in this case, however, the formula can lead only to an approximation to the function by a polynomial. We wish to compare the values of the function/ at the point x and at the point £ = x + h, so that h = $ — x. If now n is any positive integer whatever, the expression /(*) + ($- x)f(x) + ... + (i^LV'to will not, as a rule, be an exact expression for the functional value /(I). We must therefore put m =/(*) + a - *)/'(*) + ( -^p^/"(*) + . • . + ( i^!/W(*) + R n , where the expression R„ denotes the remainder when /(£) is replaced by the expression f(x) +f'(x) (f — x) + . . . . In the first instance this equation is nothing but a formal definition of the expression R n . Its significance lies in the fact that we can easily find a neat and useful expression for this remainder R n . For this purpose we think of the quantity £ as fixed and the 332 TAYLOR'S THEOREM [Chap. quantity x as the independent variable. The remainder is then a function R n (x). By the above equation this function vanishes for x = £: R n (t) = o. Further, by differentiation, we obtain n! For if we differentiate the equation defining the remainder with respect to x, we obtain on the left, since /(£) does not depend on x and is therefore to be regarded as a constant. On the right we differentiate each term by the rule for products, and find that all the terms cancel out except the last one, which is written above with a minus sign. Now by the fundamental theorem of the integral calculus R n (x) = R n (x) - R n (i) = fR n '(t)dt= fR„'(t)dt, so that we obtain the formula RM = f +h {x + h ~ t)n f +1 Ht)dt. Jx n! If we introduce a new variable of integration r by means of the equation t = t — x, this becomes R„=^f o \h- t)»/ (b+1) (s + r)dr. Collecting these results, we have the following statement: If the function f(x) has continuous derivatives up to the (n -f- l)-th order in the interval under consideration, then /(* + h) =/(*) + hf\x) + *J/"(s) + £/'"(*) + . . . + ljf (n Hx) + R«, or (the equivalent expression for h = £ — x) /(€) =/(*) + (£- x)f(x) + <l^lf"(x) + . . . 2! + { -^f^f M (x) + R« VI] TAYLOR'S THEOREM 323 where the remainder Rn is given by the formula R « = ~, C ( A - T) n f ln+1 Hx + r)dr. If in particular we put x = and then replace h by x, we obtain the formula /(*) =/(0) + J/'(0) + p"{0) +... + *j>>(0) + i?» with the remainder «-=-. f"(x-TW +1 >(r)&r. These formula are known as Taylor's theorem. They give expressions for the functions f(x + h) and f(x) respectively in terms of polynomials of degree n in h and in x respectively (the so-called polynomial of approximation), and a remainder. The polynomial of approximation is characterized by the fact that when h = (or x = 0, as the case may be) its value and that of its first n derivatives are the same as those of the given function and its first n derivatives. In contrast with the Taylor series for polynomials * the remainder and the expression for it are essential here. The significance of the formula lies in the fact that the remainder, even though it has a more complicated form than the other terms of the formula, nevertheless affords us a useful means for estimating the accuracy with which the sum of the first n + 1 terms, /(0) + J/'(0) + ~f"(0) +... + J/ (B) (0), represents the function/(x). 3. Estimation of the Remainder. Whether the first n + 1 terms of Taylor's series actually give a sufficiently good approximation to the function naturally depends on whether the remainder is sufficiently small. We therefore now turn our attention to the estimation of this re- mainder. Such an estimate can most easily be made by means of the mean value theorem of the integral calculus (Chap. II, § 7, p. 127). * Whose representation requires no remainder. 334 TAYLOR'S THEOREM [Chap. We use this theorem in the form f"p(r) <f>{r) dr = <f>{6h) f" p(r) dr, JQ Jq where p(r) is a continuous function which is nowhere negative in the interval of integration, and ^(t) is merely a continuous function there, while 6 is a number in the interval * 5S 6 5S 1. If in the formula for the remainder we take (h — t) b to be p(r) we obtain *-(^V' + "<»+<»); while if instead we put p(r) = 1 we obtain the expression £» = ^ r (l-0)"/ (n+1) (s+0A) J which is less important for us and is stated here only for the sake of completeness. In these formulae 6 denotes a certain number in the interval ^ 6 ^ 1, whose value we cannot in general specify more accurately; as a rule, of course, this value is dif- ferent in the two formulae for the remainder, and in addition depends on n, x, and h. The first form of the remainder was given by Lagrange, the second by Cauchy, and they are correspondingly named.j- Our interest will be directed chiefly towards finding out whether the remainder R„ tends to zero as n increases; if this is the case, the larger we choose n the more accurately is f(x + h) represented by the corresponding polynomial in h. In this case * We may in fact assume that < d < 1, but this is of no importance here. t These expressions for the remainder, as well as others, can be derived from the mean value theorem of the differential calculus and from the general- ized mean value theorem (p. 203) respectively. We apply these theorems to the function R M (z) = S„(x) — B n (£) and to the pair of functions B n (x) and (x — £ )" +1 respectively, where we consider ( as fixed and make use of the formula R»'X*) - - {i ~ , !t) " / ( » +1> (!g). These methods of deriving the formulas for the remainder throw more stress on the fact that Taylor's theorem is a generalization of the mean value theorem; they also offer the advantage, which for many theoretical purposes is important, that we need only assume the existence and not the continuity of the (» + l)-th derivative. On the other hand, however, we lose the advantage of having an exact expression for the remainder in the form of an integral. VI] TAYLOR'S THEOREM 325 we say that we have expanded the function in an infinite Taylor series /(* + *) =/(*) + *,/'(*) + |/"(*) + ?f"(z) +..., or, in particular, if we first put x = and then write x in place of h, /(*) =/(0) + £/' (0) + p" (0) + p'" (0) + . . . . We shall meet with examples of this in the next section. First, however, we wish to point out the second important point of view arising from consideration of Taylor's series. If in the first formula we think of the quantity k as becoming smaller and tending to zero, then in the terminology of Chap. Ill, § 9 (p. 195), the various terms of the series will tend to zero with different orders of magnitude; we accordingly call the expression f(x) the term of zero order in Taylor's series, the 12 expression hf'(x) the term of first order, the expression — /" (x) 2i\ the term of second order, and so on. From the form of our remainder we see the following fact: In expanding a function as far as the term of n-th order we make an error which tends to zero with order (n + 1) as h -»■ 0. On this fact many important applications depend. It shows us that the nearer the point x + h lies to the point x, the better is the representation of the function f(x + h) by the polynomial of approximation, and that in a given case the approximation in the immediate neighbourhood of the point x can be improved by increasing the value of n. Examples 1. Let /(*) have a continuous derivative in the interval agi^l, and let J"(x) ^ for every value of x. Then if 5 is any point in the inter- val, the curve nowhere falls below its tangent at the point x = Z,,y = /(!;)• (Use the Taylor expansion to three terms.) 2. Find the value of in Lagrange's form of the remainder B n for the expansions of and in powers of x. 1 — x 1 + x 336 TAYLOR'S THEOREM [Chap. 3. Applications. Expansions of the Elementary Functions We shall now use the general results of the preceding section in order to express the elementary functions approximately by means of polynomials and to expand them in Taylor series. We shall, however, restrict ourselves to those functions for which the coefficients of the expansion in series are given by simple laws of formation. The series for certain other functions will be discussed in Chap. VIII (p. 405 et seq.). 1. The Exponential Function. The simplest example is offered by the exponential function, f(x) = e*. Here all the derivatives are identical with f{x) and therefore have the value 1 for x = 0. Hence by using Lagrange's form for the remainder we at once obtain the formula 1! 2! ' 3! ' n! (n+l)l in accordance with § 2 (p. 320 et seq.). If we now let n increase beyond all bounds the remainder will tend to zero, no matter what fixed value of x we have chosen. For | e e * | ^ e 1 *' to begin with. We now choose a fixed integer m greater than 2 | x |. Then f or n ~2i m we have n < 2' a-n+i (n + 1)! 35™ m\ m-\- 1 ^ \x m \ 1 ^ |2a;| m 1 — m \ 2"+ 1 -™ — m! 2 B ' so that \R n \ ^ I ^ I" el*! 1. 1 ' _ m! 2" Since the first two factors on the right are independent of n, while the number 1/2" tends to zero as n increases, our state- ment is proved. If we think of the number x not as fixed, but as free to vary in the interval — a ^ x ^ a, where a is a fixed positive number, it follows from the above that if we choose m > 2a the estimate IT? |<£l 2a l"«- 1 VI] SERIES FOR ELEMENTARY FUNCTIONS 327 is valid provided n^m. For the remainder we have therefore specified a bound which holds for all values of a; in the interval — a^x^a, and which tends to zero as n -*■ 00. For the func- tion e x we can therefore write the expansion as an infinite series V n.% /».3 » — V 1! 2! 3! >.ov! the last expression being merely an abbreviated expression for the series. This expansion is valid for all values of x. Thus we have again proved that the number e considered in Chap. I (cf. p. 43) is the same as the base of the natural logarithms (cf. Chap. Ill, § 6). For numerical calculations we must, of course, make use of the finite form of Taylor's theorem with the re- mainder; for x = 1, for example, this gives c = 1 + l + 1 + 1+ . . . + 1 + — il— 2! 3! »! (n+ 1)! If we wish to calculate e with an error of at most 1/10,000, we need only choose n so large that the remainder is certainly less than 1/10,000; and since this remainder is certainly less * than 3/(n + 1)!, it is sufficient to choose n = 7, since 8! > 30,000. We thus obtain the approximate value e = 2-71822 with an error less than 0-0001. Here we do not take any account of the error due to neglecting the figures in the sixth decimal place. 2. Sin:*;, cos a;, sinhx, cosh a:. For the functions sin a;, cos a;, sinha;, cosh a; we find the follow- ing formulae: t /(*) = sin a; cos a; sinha; cosh a;, /'(*) == cos a; — sin a; cosh a; sinha;, /"(*) = — sin a; — cos a; sinha; cosh a;, /'"(*) = — cos a; sin a; cosh a; sinha;, f v W = sin a; cos a; sinha; cosh a;. * Here we have made use of the fact that e < 3. This follows immediately (of. p. 43 also) from our series for e; for it is always true that _ ^ T , and therefore *■'■ 2 "~ 1 e<l + l+i+i+ ... = 1 + 1/(1 - i) - 3. •f If f(x) — sina; or f(x) — cos*, the n-th derivative can always be repre- sented by the expression /<»'(X) - f(X + i«77). 328 TAYLOR'S THEOREM [Chap. Hence in the polynomials of approximation for sin a; and sinha: the coefficients of the even powers of x vanish, while in the poly- nomials of approximation for cos a; and cosh a; the coefficients of the odd powers vanish. Thus in the first case the (2w + l)-th and the (2w + 2)-th polynomials are identical, while in the second case the 2n-th and the (2n + l)-th are identical. If in each case we use the higher of these polynomials we at once obtain, using Lagrange's form of the remainder, a? . x 5 . , . _. n x 2n+1 sinx = x — —. + ~-. (-••• + (—1)" 3! 5! ■ x / ( 2w+ i)l T 2n+3 + ( - 1)n+1 ^+l)! C08( ^ cosa;=l -%+^-+ ... + (_i)- * 2! 4! ' (2m)! ~2n+2 + (— l)"* 1 — . cos (to), ^ v ' (2n+2)! v * X 3 X 5 X 2n+1 sinha; = x+ — + -- + .. .+ 3! 5! (2w+l)! x 2n+3 4- cosh (to), ^(2n+3)! v * , . . a; 2 x* x 2 » cosha;=l + - + - + . .. + _ /g2n+2 + cosh (to), (2w+2)! v * where in each of the four formulae, of course, 9 denotes a different number in the interval ^ 6 ^ 1, a number which in addition depends on n and on x. In these formulas we can also make the approximation as exact as we please for each value of x, since the remainder tends to as n increases. We thus obtain the four series x 3 , x? , " ..„ a; 2 "* 1 sina; = x — - + ]-... = !• (— 1) 3! ' 5! r - (2v + 1)! cos* = l_* +£_+.. . = S (-1)'* 2! 4! „=o (2v)! VI] SERIES FOR ELEMENTARY FUNCTIONS 329 4*3 4*5 sinha; =x+ — + — + 3! 5! V ^ ,_o(2v+l)!' X 2 X* cosh a; = 1 + — + — + 2! 4! S a 2 " • • • — ^ ,„ ..' ,_o(2v)l The last two can also be obtained formally from the series for e* in accordance with the definitions of the hyperbolic functions. 3. The Binomial Series. We may pass over the Taylor series for the functions log (1 -f- a;) and arc tana;, which have already been treated directly in § 1 (p. 315). We must, however, take up the generalization of the binomial theorem for arbitrary indices, which is one of the most fruitful of Newton's mathematical discoveries, and which repre- sents one of the most important cases of expansion in Taylor series. Our object is the expansion of the function f(x) = (1 + xT in a Taylor series, where x > — 1 and a is an arbitrary number, positive or negative, rational or irrational. We have chosen the function (1 + x) a instead of af because at the point x = it is not true that all the derivatives of a;* are continuous, except in the trivial case of non-negative integral values of a. We first calculate the derivatives of f(x), obtaining fix) = 0(1 + *rs /"(*) = 0(0 - 1) (i + xy-\ . . . , /'>(*) = a(a - 1) . . . (a - v + 1) (1 + xf~\ In particular, for x = we have /'(0) = a, /"(0) = a(a - 1), . . . , /«(0) = a(a - 1) (a - v + 1). Taylor's theorem then gives (l + *)«=l + a»+^jli ) ^-r-... a(a-l)(a-2)...(a-n+l) ^ R ^ We have still to discuss the remainder. This problem is not very difficult, but nevertheless is not quite so simple as in the cases treated previously. Here we shall pass over the estimation of the remainder, since the general binomial theorem will be proved 33° TAYLOR'S THEOREM [Chap. completely in a somewhat different and simpler way in Chap. VIII (p. 406 et seq.; cf. also p. 336). The result, which we mention here in advance, is that whenever | x | < 1 the re- mainder tends to and therefore the expression (1 4- x) a can be expanded in the infinite binomial series » + ">'- I + D'+ :S T i, *+--!.(*V' where for brevity we have introduced the general binomial coefficients Q = <*(*- *)• • >~ "+ D (for „ > 0)> M = L Examples 1. Expand (1 + x)l to two terms plus remainder. Estimate the re- mainder. 2. Use the expansion of Ex. 1 (discarding the remainder) to calculate V2. What is the degree of accuracy of the approximation? 3. What linear function best approximates to <y/(l + x) in the neigh- bourhood of x = 0? Between what values of x is the error of the approxi- mation less than -01? 4. What quadratic function best approximates to -^(l + x) in the neighbourhood of x = 0? What is the greatest error in the interval —0-1 ^ x g 0-1? 5. (a) What linear function, (6) what quadratic function, best approxi- mates to -\/(l + x) in the neighbourhood of x = 0? What are the greatest errors when — -1 g x 5j -If 6. Calculate sin (-01) to 4 places. 7. Do the same for (a) cos (01), (6) -^126, (c) \/97. 8. Expand sin (x + h) in a Taylor series in h. Use this to find sin 31° (= sin (30° + 1°)) to 3 places. Expand the functions in Ex. 9-18 in the neighbourhood of x = to three terms plus remainder (writing the remainder in Lagrange's form). 9. sin 2 !. 14. «-*'. 15. -L. cos a: 10. cos 8 *. 11. log cos*. 12. tan i. 13. log oosz 16. cot a; ■ 17. -L- sinz lg log(l + x) 1 + x VI] SERIES FOR ELEMENTARY FUNCTIONS 331 19. (a) Expand e sln * to five terms plus remainder; (5) in the power series for e* substitute for z the power series for sin a;, taking enough terms to secure that the coefficient of x* is correct. Compare with (a). 20. Find the polynomial of fourth degree which best approximates to tanx in the neighbourhood of x = 0. In what interval does this polynomial represent tana; to within 5 per cent? 21. Find the first 6 terms of the Taylor series for y in powers of x for the functions defined by (a) x* + y* = y, y(0) = 0; (6) 7? + y* = y, y(0) = 1; (c) r> + y» = y, y(0) = 0. 4. Geometrical Applications The behaviour of a function f(x) in the neighbourhood of a point x = a, or the behaviour of a given curve in the neigh- bourhood of a point, can be studied with increased accuracy by means of Taylor's theorem, for this theorem resolves the increment of the function on passing to a neighbouring point x = a-\-h into a sum of quantities of the first order, second order, &o. 1. Contact of Curves. We shall now make use of this method in order to investi- gate the concept of contact of two curves. If at a point, say the point x = a, two curves y = f(x) and y=g(x) not only intersect, but also have a common tan- gent, we shall say that at this point the curves touch one another or have contact of the first order. The Taylor expansions of the functions f(a-\-h) and g(a-\-h) then have the same terms of zero order and of first order in h. If at the point x = a the second derivatives of f(x) and g(x) are also equal to one another, we say that the curves have contact of the second order. In the Taylor expansions the terms of second order are then also the same, and if we assume that both functions have con- tinuous derivatives of the third order at least, the difference D(x) =f{x) — g{x) can be expressed in the form D(a + h ) =/(o + h) - g(a + h) = | D'" (a + 9h)= ^F(h), where the expression F(h) tends to /'"(a) — g'"{a) as h tends to zero. The difference D(a-{-h) therefore vanishes to at least the third order with h. 33* TAYLOR'S THEOREM [Chap. We can proceed in this way and consider the general case, where the Taylor series for f(x) and g(x) are the same up to terms of the w-th order, that is, /(«) = 9 («),/'(«) = <f («),/"(«) = *"(«), • • • ./ (B) («) = (?*(»)■ We here assume that the (» + l)-*h derivatives are also continuous. Under these conditions we say that at this point the curves have contact of the n-th order. The difference of the two functions will then be of the form /(a + h) - g(a + h) = ^-^ F(h), Fig. i. — Osculating parabolas of e* where since ^ B ^ 1 the quantity F (h) = & n+ti (a + 0h) tends to f n+1) (a) — g (n+1) (a) as h tends to 0. We recognize from this formula that at the point of contact the difference f(x) — g{x) vanishes to the (n + l)-th order at least. The Taylor polynomials are simply defined geometrically by the fact that they are those parabolas of the w-th order which at the given point have contact of the highest possible order with the graph of the given function. Hence they are sometimes called osculating parabolas. For the case y=e x fig. 1 gives us the first three osculating parabolas at the point a;=0. If two curves y=f{x) and y = g(x) have contact of the ra-th order, the definition does not exclude the possibility that VI] GEOMETRICAL APPLICATIONS 333 the contact may be of still higher order, i.e. that the equation /<" +1 >(a) = $r< B+1) (a) is also true. If this is not the case, i.e. if / (B+1) (o) =j= fif (B+1) (a), we speak of a contact of exactly the n-th order or say that the order of the contact * is exactly n. From our formulae as well as from our figures we can at once state a remarkable fact which is often unnoticed by beginners. If the contact of two curves is exactly of an even order, that is, if an even number ra of derivatives of the two functions have the same value at the point in question, while the (w + l)-th deri- vatives differ, then in conformity with the above formulae the difference f(a + h) — g(a -\- h) will have different signs for small positive values of h and for numerically small negative values of h. The two curves will then cross at the point of contact. This case occurs e.g. in contact of the second order if the third derivatives have different values. If, however, we consider the case of contact exactly of an odd order, e.g. the case of an or- dinary contact of the first order, the difference f(a + h) — g (a + h) will have the same sign for all numerically small values of h, whether positive or negative; the two curves therefore will not cross in the neighbourhood of the point of contact. The simplest example of this is the contact of a curve with its tan- gent. The tangent can cross the curve only at points where the contact is of the second order at least; it will actually cross the curve at points where the order of contact is even, e.g. at an ordinary point of inflection, where /" (a;) = but /'" (x) 4= 0. At points where the order of contact is odd it will not cross the curve; as examples we may take an ordinary point of the curve where the second derivative is not zero, or the curve y = sb 4 at the origin. 2. The Circle of Curvature as Osculating Circle. When looked at from this point of view the concept of the curvature of a curve y =f(x) gains a new intuitive significance. Through a definite point of the curve with the co-ordinates x = a, y = 6 there pass an infinite number of circles which touch the curve at the point. The centres of these circles lie on the normal to the curve, and to each point of this normal * That the order of contact of two curves is a genuine geometrical relation which is unaffected by change of axes is a fact which can easily be established by means of the formula for change of axes. 334 TAYLOR'S THEOREM [Chap. there corresponds just one such tangent circle. We may expect that by proper choice of the centre of the circle we can bring about a contact of the second order between the curve and the circle. As a matter of fact, we know from Chap. V (p. 283) that for the circle of curvature at the point x = a, whose equation is, say, y = g(x), we not only have g («)=/(») and g'(a)=f'{a), but also g"(a)=f"(a). Hence the circle of curvature is at the same time the osculating circle at the point of the curve under discussion; that is, it is the circle which at that point has contact of the second order '^ /^" \ with the curve. In the limiting case of a point of inflection, or in general of a point at which the cur- vature is zero and the radius of curvature is in- finite, the circle of curva- ture degenerates into the tangent. In ordinary cases, that is, when the contact at the point in question does not happen to be of an order higher than the second, the circle of curvature will not merely touch the curve, but will also cross it (cf. fig. 2). 3. On the Theory of Maxima and Minima. As we have already seen in Chap. Ill (p. 161), a point x = a at which/' (a) = gives a maximum of the function/(a;) if/" (a) is negative, a minimum if/" (a) is positive. These last conditions, therefore, are sufficient conditions for the occurrence of a maxi- mum or minimum. They are by no means necessary; for in the case where /" (a) = there are three possibilities open; at the point in question the function may have a maximum or a mini- mum or neither. Examples of the three possibilities are given by the functions y = — cc 4 , y = a; 4 , and y = x 3 at the point x = 0. Taylor's theorem at once enables us to make a general statement of sufficient conditions for a maximum or a minimum. We need only expand the function /(a + h) in powers of A; the essential point is then to find whether the first non-vanishing sr Fig. a. — Osculating circle VI] GEOMETRICAL APPLICATIONS 335 term contains an even power of h or an odd power. In the first case we have a maximum or a mini mum according as the co- efficient of h is negative or positive; in the second case we have a horizontal inflectional tangent and neither maximum nor minim um. The reader may complete the argument for himself by using the formula for the remainder.* Examples 1. What is the order of contact of the curves y = e* and y = 1 + * + i sin'a: at x = 0? 2. What is the order of contact of y = sin 4 a; and y = tan 4 a; at x = 0? 3. Determine the constants a, b, c, d in such a way that the curves y = e** and y = a cosa; + 6 sina; + c cos2a; + d sin2a; have contact of order 3 at x = 0. 4. What is the order of contact of the curves x* + y* = xy, a? + y a = a: at their points of intersection? Plot the curves. 5. What i» the order of contact of the curves s* + Z/ a = y x*=y at their points of intersection? 6. The curve y = f(x) passes through the origin O and touches the a;-axis at O. Show that the radius of curvature of the curve at O is given x 3 by p = Iim — . *->o2y 7.* K is a circle which touches a given curve at a point P and passes through a neighbouring point Q of the curve. Show that the limit of the circle K as Q -*■ P is the circle of curvature of the curve at P. 8.* R is the point of intersection of the two normals to a given curve at the neighbouring points P, Q of the curve. Show that, as Q -». P, R tends to the centre of curvature of the curve for the point P. (The centre of curvature is the intersection of neighbouring normals.) 9.* Show that the order of contact of a curve and its osculating circle is at least three at points where the radius of curvature is a maximum or minimum. 10. Determine the maxima and minima of the function y = er 11 *'. (See p. 336.) * The necessary and sufficient condition given previously (p. 161), how- ever, is more general and more convenient in applications, namely: provided the first derivative f'(x) vanishes at only a finite number of points, a necessary and sufficient condition for the occurrence of a maximum or minimum at one of these points is that the first derivative f'(x) changes sign as it passes through the point. 336 TAYLOR'S THEOREM [Chap. Appendix to Chapter VI 1. Example of a Function which cannot be expanded in a Taylor Series The possibility of expressing a function by means of a Taylor series with a remainder of the (n + l)-th order depends essen- tially on the differentiability of the function at the point in ques- tion. For this reason the function log a; cannot be represented by a Taylor series in powers of x, and the same is true of the function y/x, whose derivative is infinite at x = 0. In order that a function may be capable of being expanded in an infinite Taylor series, all its derivatives must necessarily exist at the point in question; this condition, however, is by no means sufficient. A function for which all the derivatives exist and are continuous throughout an interval still need not neces- sarily be capable of expansion in a Taylor series; that is, the remainder R„ in Taylor's theorem may fail to tend to zero as w increases, no matter how small the interval in which we wish to expand the function. The simplest, example of this phenomenon is offered by the function y = f(x) = e -1 '"* for a; =t= 0, /(0) = 0, which we have already considered in the appendix to Chap. Ill (p. 196). This function, with all its derivatives, is continuous in every interval, even at x = 0, and we have seen that at this point all the derivatives vanish, i.e. that / <n) (0) = for every value of ». Hence in Taylor's theorem all the coefficients of the polynomials of approximation vanish, no matter what value we choose for n. In other words, the remainder is and remains equal to the function itself, and therefore, except when % = 0, does not approach as n increases, since the function is positive for every non-zero value of x. 2. Proof that e is Irrational From the formula e=2+- + ...-| + — eP, we immediately 2! n! (n+ 1)! deduce that the number e is irrational. For if the contrary is true, that is, if e = p/q, where p and q are integers, we can certainly choose » larger than q. Then n!e = n\ - must be an integer. On the other hand, ? »!e = 2»! + — + ...+ — + e e , and since e" < e < 3, we must 21 nl n + 1 VI] CONVERGENCE OF BINOMIAL SERIES 337 have < < 1. Hence the integer re!e = the integer 2re! + — + n + 1 * . . . + 1 plus a non- vanishing proper fraction, which is impossible. 3. Proof that the Binomial Series Converges In § 3 (p. 329) we postponed the estimation of the remainder R n in the expansion of f(x) = (1 + xf for | x \ < 1. This estimation we shall carry out here. It is most convenient to separate the cases x > and x < 0. For/ ( " +1) (a;) we have the expression y^»(,) = o(«-l)...(a-»)^±^ i . If x > 0, we write the remainder in Lagrange's form, u / x x " +1 , i\ i \ ( l + 6x )' R n (x) = - — 1 — 7ri a(o — 1) . . . (a — n) so that *.(*) I ^ (n+l)I a(a — 1) ... (a — n) (n + 1)! (1 + Ox)"* 1 x n +\l + x) a ]n+l " Writing b = [ | a | ] + 1, where [ | a | ] means the greatest integer which does not exceed | a |, we have \R n (x)\^W±^p^} x » + i \n -\- L)i <Z 2 " l-2...(n+l)(n+2)...(»+6) %n+ i -(6-1)! (n+1)! 2" — (n + 6)»-V+ x , (6-l)! v ; and since b is fixed, if < x < 1, this approaches as n increases. For the case —1 < x < we write the remainder in Cauchy's form, -* M1 a-«>-«<--i)...<«-»)< 1 + tora R„(X) : so that nl (1 + dx) n Iwli^pW a(a— l)...(a — n) 12 ia+ for 1 1 (E798) 338 TAYLOR'S THEOREM [Chap. Since \x\ < 1, the last factor cannot exceed a constant E, independent of n. Also (1 — d) / (1 — 8 1 x | ) < 1. As before, writing b = [ | a \ ] 4- 1, we have \R n (x)\ ^ K\x\">> —1— (n+ 2) (n+ 3) . . . (n + 6) (6 — 1)! ^(T~Iji (n + 6)W|af,M1 » which approaches as n increases. Thus in either case when | x | < 1 the remainder tends to zero as n increases, justifying the expansion in § 3 (p. 330). 4. Zekos and Infinities op Functions and So-called Indetebminate Expeessions The Taylor series for a function in the neighbourhood of a point x = a enables us to characterize the behaviour of the function in the neighbourhood of this point in the following way. We say that a function f(x) at x = a has an exactly n-twple zero or it vanishes there exactly to the order n, if f(a) = 0, J (a) = 0, f"(a) = 0, . . . , /'-"(a) = 0, and /<»>(<*) #= 0. Here we assume that in the neighbourhood of the point the function possesses continuous derivatives to the w-th order at least. By our de- finition we seek to indicate that the Taylor series for the function in the neighbourhood of the point can be written in the form in which as h ->- the factor F(h) tends to a limit different from 0, namely, the value /"(a). If a function <f>(x) is defined at all points in the neighbour- hood of a point x = a, except perhaps at x = a itself, and if At \ f( x ) where at the point x = a the numerator does not vanish, but the denominator possesses a p-tuple zero, we say that the func- tion ^(aj) becomes infinite of the v-th order at the point x = a. If at the point x = a the numerator also possesses a /x-tuple VI] INDETERMINATE EXPRESSIONS 339 zero and if /* > v, we say that the function has a (/* — v)-tuple zero there; while if (*.<.v we say that the function has a (v — /i)-tuple infinity. All these definitions are in agreement with the conventions already laid down (cf. Chap. Ill, § 9, p. 194) regarding the be- haviour of a function. In order to make these relations precise, we expand numerator and denominator by Taylor's theorem, using Lagrange's form of the remainder; the function then has the form Ma 4- h\- ^ a + *> - " ! ^ M(a + 6h) in which 8 and 8 1 are two numbers between and 1 and the factors by which — and — are multiplied do not tend to zero fil vl as h does, since they approach the limits / (,l) (a) and g^ v) (a) respectively, which differ from zero. If /x > v, we then have lim #> + A) = lim v l h»~ J ^ = 0. 0/*! The expression cf>(x) accordingly vanishes to the order /i — v. If v> fi, we see that the expression <f>(a -f- h) becomes infinite of the order v — /u. as h -*■ 0. If jj. = v, we obtain the equation h-yO 9 {<*) We can express the content of the last equations in the fol- lowing way: if the numerator and denominator of a function Mx) = 1^1 both vanish at x = o, we can determine the limiting value as x -*■ a by differentiating the numerator and denominator an equal number of times until one at least of the derivatives is other than zero. If this happens for numerator and denominator simultaneously, the limit which we are seeking is equal to the quotient of these two derivatives. If we encounter a non-vanish- ing derivative in the denominator earlier than in the numerator, the fraction tends to zero. If we encounter a non-vanishing derivative in the numerator earlier than in the denominator, the absolute value of the fraction increases beyond all bounds. 340 TAYLOR'S THEOREM [Chap. We thus have a rule for evaluating the so-called indeter- minate expression 0/0 — a subject that is discussed at exaggerated length in many textbooks on the differential and integral cal- culus. In reality the point in question is merely the very simple determination of the limiting value of a quotient in which the numerator and the denominator tend to zero. The name " indeterminate expression " usually found in the literature is misleading and vague. We can arrive at our results in a somewhat different way by basing the proof on the generalized mean value theorem * instead of on Taylor's theorem (cf. p. 135). According to this, if gf(x) 4= 0, we have f(a+h)-f(a) _ f'(a+eh) g(a+h)-g(a) g , (a+6h)' where 8 is the same in both numerator and denominator. Hence, in particular, when f(a) = = g(a), f(a+h) _ f'(a+6h) g(a + h) g'(a+6h)' Here is a value in the interval < 6 < 1, and if we put k = 6h, we obtain Una f^t+3 = lim ££±*>, h^-og(a + h) k^o g'(a + k) it being assumed that the limit on the right exists. If f'(a) = = g'(a), we can proceed in the same manner until we come to the first index for which it is no longer true that / w (a) =0=g w (a). Then lim /(^±i) =lim /^ + ^ h -+og(a + h) ,-+ g M (a + l) in which we also include the case in which both sides have the limit infinity. * This method of deriving our rule has the advantage that in it no use is made of the existence of the derivative at the point x = a itself: further, it includes the case in which 4>(x) is defined for x^2l a only, so that the passage to the limit x -*■ a or ft -»• is made from one side only. VI] INDETERMINATE EXPRESSIONS 341 As examples we consider wax 1 — oosa; «*" — 1 x* tans T' x * log(l + x)' V(l - x>) - 1 as x -» 0. We have .. sina: oosO . .. 1 — cosa: sinO „ lim = «=» 1; nm = — — = u; X 1 *-H> x 1 dm = lim — ■ = 2; _.„ log(l + *) —9-0 1/(1 + x) x* tana; ,. 2x tana + a^/oos»i lim = lim - — - — ._^„ </,'] - **) - 1 _h> -a;/V(l - a; 2 ) = -lim (2 tanas + -JL V(l - x*) = 0. ^_».o V cos* a;/ We further note that other so-called indeterminate forms can also be reduced to the case we have considered; for example, the limit of - r — sinx x as x -»■ 0, being the limit of the difference of two expressions which both become infinite, is an " indeterminate form "00 — 00, By the trans- formation 1 1 x — sina: we at once arrive at an expression whose limit as x -> is determined by our rule to be 1 — cosx .. sina; n lim = hm : — = 0. «->o x coss -f sins «-h> 2 cosa: — x sm* Examples Evaluate the limits in Ex. 1-12: l.lim*^l n . 7. lim(-l__ 1\ ,_ > . x — a *->i vb* — 1 x — 1/ 2. lim *- sina: . 8. lim (4- - Y\. ^.0 3? x-+o\am*x x*l n ,. 24 — 12a? + x* — 24 cosa; . .. . _ 3. lim - 5 . 9. lima: 8 ' 11 *. (sina;) 6 *-»-o 4. lim e °'~ er< '. 10. lim (1 + *)!/*. 1— »-o sina; *— »-o 6. lim?£2i^. "• Mm *" ~ * *_>o log(l + x) 6. lim *5*L 12. lim xtmx «->.*« tanas" *-»-o V(l — 3?) — 1 13. Prove that the function y — (a?)*, y(0) = 1 is continuous at 1 = 0. CHAPTER VII Numerical Methods Preliminary Kemarks Anyone who has to use analysis as an instrument for investigating physical or technical phenomena is faced by the question whether and how the theory can be adapted to yield useful practical methods for actual numerical calculations. Yet even from the point of view of the theorist, who desires only to recognize the connexions between natural phenomena, not to conquer them, these questions are of no trifling interest. For a systematic treatment of numerical methods we refer the reader to special textbooks on the subject.* Here we can only discuss some particularly important points which are more or less closely connected with the preceding ideas. We wish to direct special attention to the fundamental fact that the mean- ing of an approximate calculation is not precise unless it is supplemented by an estimate of the errors occurring, i.e. unless it is accompanied by definite knowledge of the degree of accuracy attained. 1. Numerical Integration We have seen that even relatively simple functions cannot be integrated in terms of elementary functions and that it is futile to make this unattainable goal the aim of the integral calculus. On the other hand, the definite integral of a continuous function does exist, and this fact raises the problem of finding methods for calculating it numerically. Here we shall discuss the simplest and most obvious of these methods with * Cf. e.g. Whittaker and Robinson, The Calculus of Observations (Blackie & Son, Ltd., 1929). M2 Chap. VII] NUMERICAL INTEGRATION 343 the aid of geometrical intuition, and we shall then consider the estimation of errors. Our object is to calculate the integral I = / f(x)dx, where •'o a is less than 6. We imagine the interval of integration divided into n equal parts of length h = (b — a)ln, and we denote the points of subdivision by x = a, x^ — a + h, . . . , x„ = b, the values of the function at the points of division by /q, f v •••>/«> and similarly the values of the function at the mid- points of the intervals by / 1/2 , / 3/2 , . . . , / (2n _ 1)/2 . We interpret our integral as an area and cut up the region under the curve into strips of breadth h in the usual manner. We must now obtain an approximation for each such strip of surface, that is, for the integral f(x)dx. 1. The Rectangle Rule. The crudest and most obvious method of approximating to I is directly connected with the definition of integral; we replace the area of the strip Z„ by the rectangle of area/, h and obtain for the integral I the approximate expression * I*>h(f + &+... +/„_!). 2. The Trapezoid and Tangent Formulae. We obtain a closer approximation with no greater trouble if we replace the area of the strip Z„, not by the above rectangular area, but by the trapezoid of area - (/, -f f p+1 )h shown in fig. 1. For Z the whole integral this gives the approximate expression I * *(/l +/.+ ... +/»-!) + *(/,+/.) (the trapezoid formula), since, when the areas of the trapezoids are added, each value of the function except the first and the last occurs twice. * Here and hereafter the symbol «tf means " is approximately equal to ". 344 NUMERICAL METHODS [Chap. As a rule the approximation becomes even better if instead of choosing the trapezoid under the chord AB as an approxi- mation to the area of I we choose the trapezoid under the tan- gent to the curve at the point with the abscissa x = x„ -\- h/2. The area of this trapezoid is simply hf v+i , and the approxi- mation for the entire integral is I fin h(f m +f m + ... +/( 2n -i)/ 2 )» which is called the tangent formula. Fig. I. — The trapezoid formula 3. Simpson's Bale. By means of Simpson's rule we arrive, with very little more trouble, at a numerical result which is generally much more exact. This rule depends on estimating the area I v -\- I v+1 of the double strip between the abscissse x = x v and x = x f + 2h = a5 v+ 2 by considering the upper boundary to be no longer a straight line but a parabola; to be specific, that parabola which passes through the three points of the curve with abscissae x *> x v+i = x v +h, and x v+2 = x y +2h (cf. fig. 2). The equation of this parabola is ^ (x - x v ) (x-x v - h) f v+2 - 2/ r+1 +/, ^ 2 # VII] NUMERICAL INTEGRATION 345 (The student may verify by direct substitution that for the three values of x in question this equation gives the proper values of y, namely, /„, /„ +1 , and/„ +2 respectively.) If we integrate this polynomial of the second degree between the limits x v and x v + 2 A, we obtain, after a brief calculation, the following ex- pression for the area under the parabola: j x ydx=2hf v +2h(f v+1 -/ r ) + ^|A-2AJ(/ r+a -2/, +1 +/ r ) = *(/,+4/H-i + /r+i). Fig. 2. — Simpson's rule This represents the required approximation to the area of our strip, /„ + I v+1 . If we now assume that w = 2m, i.e. that n is an even number, by the addition of the areas of such strips we obtain Simpson's rule: Ar}) '«* |(/i+/a + + .../ 2m - 1 ) + -Q (/a +/» + +/2m-2) + 2 (/o +/am)- 12 • (1708) 346 NUMERICAL METHODS [Chap. 4. Examples. — If 1 x we divide the integral from 1 to 2 into ten equal parts, h will be equal to i^y, and by the trapezoid formula we obtain x 1 = 11 /, = 0-90909 x 2 = 1-2 f 2 = 0-83333 x 3 = 1-3 /„ = 0-76923 x t = 1-4 U = 0-71429 x 6 = 1-5 ft = 0-66667 x e = 1-6 /„ = 0-62500 x 7 = 1-7 /, = 0-58824 x s = 1-8 f a = 0-55556 x, = 1-9 /„ = 0-52632 Sum 618773 x = 10 hfo = 0-5 x 10 = 2-0 J/io = 0-25 6-93773 X ^g log,2 ss 0-69377 This value, as was to be expected, is too large, since the curve has its convex side turned towards the a;-axis. By the tangent rule, we have av, + lh = 1-05 /,/, = 0-95238 Xl + \h = 1-15 /s /2 = 0-86957 x s + $h = 1-25 / 5 / 2 = 0-80000 x 3 + \h = 1-35 /,/, = 0-74074 a; 4 + | A = 1-45 /„/, = 0-68966 x, + \h = 1-55 /„/, = 0-64516 *„ + \h = 1-65 jf ls /, = 0-60606 x,+ lh= 1-75 / 15/l = 0-57143 x g + lh = 1-85 / 17 / s = 0-54054 *» + lh = 1-95 / 19 /ij =. 0-51282 6-92836 X ^g log, 2 <sn 0-69284 Owing to the convexity of the curve, this value is too small. VIIJ NUMERICAL INTEGRATION 347 For the same set of subdivisions we obtain the most exaot result by means of Simpson's rule. We have *, = 11 *, = 1-3 x t = 1-5 x ? = 1-7 x„ = 1-9 /, = 0-90909 / 3 = 0-76923 / 5 = 0-66667 / 7 = 0-58824 /, = 0-52632 Sum 3-45955 X 4 13-83820 *,= 1-2 x, = 1-4 x 6 = 1-6 z a = 1-8 x,„ = 2-0 /, = 0-83333 / 4 = 0-71429 /, = 0-62500 /„ = 0-55556 Sum 2-72818 X 2 5-45636 13-83820 20-79456 X 3*3 As a matter of fact log. 2 f%) 0-69315 log.2 = 0-693147 . . . 5. Estimation of the Error. It is easy to give an estimate of the error in each of our methods of integration if the derivatives of the function f(x) are known throughout the interval of integration. We take M v M % , ... as the upper bounds of the absolute values of the first, second, . . . derivatives, respectively; that is, we assume that throughout the interval |/ w (x) | < M r . Then the estimation formulae are as follows: For the rectangle rule l^-A/J^M^or For the tangent rule K-V, +i |<§A3 or For the trapezoid rule I-AS/, <-M 1 nA2=-M 1 (6-a)A. 2i 2i v=0 < Mi 24 (6 -a) A 2 . j,-;;(/v+/h-i) < M 2 12 A 3 . For Simpson's rule ■*» + *v+i — o (/" ~^~ V»+i +A+2) < 90 A 5 348 NUMERICAL METHODS [Chap. From the last two estimates there also follow estimates for the entire integral I. We see that Simpson's rule has an error of much higher order in the small quantity h than the other rules, so that where M i is not too large it is very advantageous for practical calculations. To avoid wearying the reader with the details of the proofs of these estimates, which are fundamentally quite simple, we shall content ourselves with the proof for the tangent formula. For this purpose we expand the function f(x) in the (v + l)-th strip by Taylor's theorem: /<*>-/,+*+ (* - * - *)/'(** + i) + -J(—*- s)>(A where £ is a certain intermediate value in the strip. If we integrate the right-hand side over the interval x v ^ x £S x y -\- h, the integral of the middle term is zero. Since 1 /■*+*/ A\» h 3 as is easily verified, it follows immediately that ry+ll f(x)dx-Kf r . which proves our assertion. <*.£. Examples 7r r 1 1 1. From the formula - — I ; dx, calculate it 4 J 1 + x 2 (a) using the trapezoid formula with h = 0*1; (6) using Simpson's rule with h = 0-1. /•00 2. Calculate / e-^'dx numerically to within y£ c (of. p. 496). r 1 1 3. Calculate / , numerically with an error less than (H. J Vl + z 4 VII] THE CALCULUS OF ERRORS 349 2. Applications op the Mean Value Theorem and op Taylor's Theorem. The Calculus op Errors 1. The " Calculus of Errors ". We now come to quite a different type of numerical calcu- lations. These are applications of the mean value theorem, or more generally of Taylor's theorem with remainder, or finally of the infinite Taylor series. As an application which, though simple, is quite important in practice we shall consider the cal- culus of errors. This rests upon the idea — which lies at the root of the whole of the differential calculus — that a function f(x) which is differentiable a sufficient number of times can be re- presented in the neighbourhood of a point by a linear function with an error of order less than the first, by a quadratic func- tion with an error of order less than the second, and so on. Let us consider the linear approximation to a function y=f(x). If y + Ay =f(x-\- Ax)=f(x-\- A), by Taylor's theorem we have Ay=hf'(x) + ^f"(i), where £ = x + dh(Q < <1) is an intermediate value, which need not be more precisely known. If h = Ax is small we obtain as a practical approximation Ay ps hf'{x). In other words, we replace the difference quotient by the deri- vative to which it is approximately equal and the increment of y by the approximately equal linear expression in A. We use this fairly obvious fact for practical purposes in the following way. Suppose two physical quantities x and y are connected by the relation y= f(x). The question then arises, what effect an inaccuracy in the measurement of x has on the determination of y. If instead of the " true " value x we happen to use the inaccurate value x -J- h, then the corresponding value of y differs from the true value y=f{x) by the amount Ay= f(x + A) — f(x). The error is therefore given approximately by the above relation. 35° NUMERICAL METHODS [Chap. We shall understand the use of these relations better if we con- sider a few examples. Six. 1. The tangent galvanometer. In determining current by a tangent galvanometer we use the formula y = c tan a, where a is the angle of deflection of the magnetic needle, c is the constant of the apparatus, and y = I is the intensity of the current. Then dy _ e da cos* a and therefore Ay *%* — — Aa. The percentage error in the measurement is given by cos a lOOAy lOOcAa 200 A » f& ;^ = Aa. y c cos 2 a tan a sin 2a From this we see that the accuracy reaches the greatest possible value, Le. to a given error in the measurement of the angle there corresponds the least possible error in the determination of the current, when the angle a is equal to 7t/4 or 45°. In particular, let us suppose that it is possible to read the tangent gal- vanometer to within half a degree; then | Aa | in radians < J X 0-01745 . . . , and the percentage error will be — . If the reading on the galvano- sin2a meter is 30°, sin2a= J V3 =$ X 1-73205 ... , and the percentage error is less than 2 x , which is about 2 per cent. 1-732 Ex. 2. In a triangle ABO (of. fig. 3) we suppose that the sides b and e are measured accurately, while the angle a = x can only be measured with an error | Ax | < 8. Within what limit of error does the value y = o = Vfft* + c* — 26c cos a) vary? We have Aa fa - be sin a Aa; a «. 4 • a * lOOAa 1006<r . . T . , , ,. tne percentage error is therefore fa — — sin a Aa. If we take the VII] THE CALCULUS OF ERRORS 351 special oase where 6 = 400 metres, c = 500 metres, and a = 60°, then by the cosine formula y = a = 458-2576 metres, and 200000 w 1 .- Aa ^45¥2576 X 2 V3Aa - If Aa can be measured to within ten seconds of arc, that is, if Ace = 10" = 4848 X lO -8 radians, we find that at worst Aa pa 1-83 om.; that is, the error is at most about 0-004 per cent. Ex. 3. The following illustrates a type of application of the above methods by which we can often save ourselves considerable trouble in physical problems. It is known experimentally that if an iron rod has the length l at temperature 0, then at temperature t its length will be I = Z (l -+- <**)> where a depends only on the material of which the rod is made. If now a pendulum clock keeps correct time at temperature <,, how many seconds will it lose per day if the temperature rises to t t l For the period of oscillation we have the formula '9 & Vlg Hence if the change of length is Al the corresponding change in the period of oscillation is ._, 7t Al AT & —_=, Via where ^ = l (l + otfj and Al = aZ (< 2 — h)- This is the time lost per oscillation. The time lost per second is AT/T s» Al/71^ hence in one day the clock loses 43200 Al/l t seconds. Here the application of our methods has saved us a number of multi- plications and two extractions of the square root. In the longer direct process, moreover, we should finally have to subtraot T(lj) from the almost equal value T(l 2 ), and a very small error in calculation would cause a rela- tively large percentage error in the result.* In this case, and in most cases where the function under consideration has several factors or fractional indices, we oan reduce the calculation even more by taking the logarithms of both sides before differentiating. In the present example we have logT = log2n — - log? + - log/; by differentiating, we have ill 2/ * It is a point of this nature that makes the calculations of applied optics so extremely laborious. T(V) = 2tzJ 1 -, whence % = -?L. 352 NUMERICAL METHODS [Chap. dT AT If we replace — by — this gives dl Al AT ^ Al T 2f in agreement with the preceding result. 2. Evaluation of n. Gregory's series,* ^ = 1 — - + _— --\ ..., which we obtained in Chap. VI, § 1 (p. 319) using the series for the inverse tangent, is not suitable for the calculation of -n, on account of the slowness of its convergence. We may, however, calculate ir with comparative ease by the following artifice. From the addition theorem for the tangent, . , , o\ tana + tanfi tan(a + ft = ^ H 1 — tana tanp if we change to the inverse functions a = arc tan u, ft — arc tan v, we obtain the formula arc tanw -f- arc tanu = arc tan \1 — uv/ ' If we now choose u and v in such a way that — — — = 1, we 1 — uv obtain the value - on the right-hand side, and if u and v are 4 small numbers we can easily calculate the left-hand side by means of known series. If, for example, we put u== -, v== -, as Euler did, we obtain — = arc tan - + arc tan -. 4 2 3 If we further notice that |- + -J-r(l — — ) = -, we have \3 7/ \ 21/ 2 arc tan - = arc tan - + arc tan -, so that 2 3 7 — = 2 arc tan - + arc tan -. 4 3^7 Using this formula Vega calculated the number -n to 140 places. * Sometimes called Leibnitz's series. VII] CALCULATION OF LOGARITHMS 353 By means of the equation (- + -)-i-(l — - ) = „, we J * \5 8/ \ 40/ 3 further obtain arc tan - = arc tan - + arc tan 3 5 8 or - = 2 arc tan - + arc tan - + 2 arc tan -. 4 5 7 8 This expansion is extremely useful for the calculation of 77 by means of the series arc tana; = x — — + — ["•••» f° r o if we substitute for x the value -, -, or - we obtain with but few terms a high degree of accuracy, since the terms diminish rapidly. We can, however, perform the calculation even more conveniently if we base it on the formula 77 ,.120 . 1 , , 1 1 — = arc tan — — arc tan — = 4 arc tan arc tan — , 4 119 239 5 239 obtained by considerations similar to those above. 3. Calculation of Logarithms. For the numerical calculation of logarithms we transform the logarithmic series - log "*" = x-\- — + — — |— . . . ( | £c | -<; 1> 2 1 — x 3 5 where < x < 1 by the substitution 1 + x _ p 2 x= — 1 — x p 2 — 1 2jp 2 — 1 into the series lo 8P = s lo E(P — !) + s log(? + 1) + 2 ovx- " 2 =>' ' 2p 2 — 1 + 3(2^-l)3 + '-- ' where 2p 2 — 1 > 1, that is, p 2 > 1. If p is an integer and p + 1 can be resolved into smaller integral factors, this last series expresses the logarithm of p by the logarithms of smaller integers 354 NUMERICAL METHODS [Chap. plus a series whose terms diminish very rapidly and whose sum can therefore be calculated accurately enough by use of only a few terms. From this series we can therefore calculate successively the logarithms of any prime number, and hence of any number, provided we have already calculated the value of log 2. The accuracy of this determination of log^» can be estimated more easily by means of the geometric series than from the general formula for the remainder. For the remainder R n of the series, i.e. the sum of all the terms following the term n(2p 2 - l)n ' we ^ R n < I (l\ l \ 1 \ \ (n+2)(2f- 1)«+ 2 \ T (2p 2 - l) 2 ^ (2y 2 - 1)* ^ ' ' 7 1 1 (w+ 2) (2j» a - 1)» " (2jo 2 — l) 2 — 1' and this formula immediately gives the required estimate of the error. Let us for example calculate log , 7, using the first four terms of the series. We have p = 7, 2p» - 1 = 97, log7 = 2 log2 + _ log3 + — + 1 + . . .: e 6n 2 6T 97 3.97» -i (%( 001030928, — ~ on 0-00000037, hence 2 log2 fti 1-38629436, - log3 an 0-54930614; log.7 ** 1-94591015. Estimation of the error gives 5.97' 97 a - 1 36 x 10» We must, however, note that each of the four numbers which we have added is only given to within an error of 5 x 10 _, ( so that the last place in the value of log 7 given above might be wrong by 2. As a matter of faot, however, the last place is right also. VII] CALCULATION OF LOGARITHMS 355 Examples 1. To measure the height of a hill, a tower 100 metres high on top of the hill is observed from the plain. The angle of elevation of the base of the tower is 42° and the tower itself subtends an angle of 6°. What are the limits of error in the determination of the height if the angle 42° is subject to an error of 1°? 2. Calculate log, 2 to three decimal places by means of an expansion in series. 3. Calculate log, 6 to six decimal places, using the values of log, 2 and log, 3 given in the text. 4. Calculate n to five decimal places, using any one of the formulae in sub-section 2 (pp. 352-3). 3. Numerical Solution of Equations In conclusion we shall add some remarks about the numerical solution of the equation f(x) — 0, where f(x) need not necessarily be a polynomial.* Every such numerical method is based on the plan of starting with some known approximation x of one of the roots and then improving this approximation. Whence this first approximation for the desired root of the equation is found, and how good the approximation is, do not particularly matter. We may perhaps take a rough guess as a first approxi- mation, or better, obtain it from the graph of the function V =f( x )> whose intersection with the as-axis gives the required root (of course with an error depending on the scale and the accuracy of the drawing). 1. Newton's Method. The following procedure which comes down to us from New- ton is based on the fundamental principle of the differential calculus — the replacing of a curve by a straight line, the tangent, in the immediate neighbourhood of the point of contact. If we have an approximate value x for a root of the equation /(x) = 0, we consider the point on the graph of the function y=f(x) whose co-ordinates are x — x , y =f(x ). What we wish to find is the intersection of the curve with the a^axis; as an approxi- mation to this we find the point where the tangent at the point •Here, of oourse, we are only concerned with the determination of real roots of /(as) - 0. 35 6 NUMERICAL METHODS [Chap. x — x , y=f(x ) intersects the sc-axis. The abscissa a^ of this intersection of the tangent with the a;-axis will then represent a new and, tinder certain circumstances, a better approximation than x to the required root of the equation. In virtue of the geometrical meaning of the derivative, fig. 4 at once gives J^L=f(x ). x x^ From this we obtain the formula for the calculation of the new approximation a^: Fig. 4. — Newton's method of approximation X 1 - x n — /'W If by this procedure we have found an approximation bettei than x , then we repeat the process to find x 2 , and so on, and if the curve is of the form shown in fig. 3 these approximations will approach more and more nearly to the required solution. Fig. 5 The usefulness of this process depends essentially on the nature of the curve y =f(x). In fig. 4 we see that the successive estimates converge with greater and greater accuracy to the required root. This is due to the fact that the curve has its convex side turned towards the »-axis. But in fig. 5 we see that if we choose the original value x badly, our construction does not lead to the required root at all. From this we see that in using Newton's method we must examine each individual case to VII] NUMERICAL SOLUTION OF EQUATIONS 357 determine with what degree of accuracy we have really solved the equation. We shall return to this subject on p. 359. 2. The Role of False Position. Newton's method, in which the tangent to the curve plays a decisive part, is only the limiting case of an older method, known as the rule of false position, in which the secant appears in place of the tangent. Let us assume that we know two points (z > Vo) and (ai. yi) in the neigh- bourhood of the required intersec- tion with the a^axis. If we replace the curve by the secant joining these two points the intersection of this secant with the x-axis will certain circumstances be an in Fig. 6. — The rule of false position improved approximation to the required root of the equation. If the abscissa of this point is denoted by £, we have (fig. 6) the equation i — 3p _ £— a?i and from this we calculate £: _ Sp/fa) — »h/(So) + Sp/fop) — 3j/(Zo ) /fro) {/to)— f( x o)}K x i — * )' or !=Zp This formula, which determines the further approximation | from x and a^, is called the rule of false position. We can use it with advantage if one value of the function is positive and the other negative, say as in fig. 6, where y > and y t < 0. Eepe- tition of this process will always lead us to the required result if at each step we use a positive and a negative value of the function, between which the required root must necessarily lie. 358 NUMERICAL METHODS [Chap. The above formula of Newton results from the rule of false position as a limiting case if we let x 1 tend to x . For the denominator of the second term on the right-hand side of the statement of the rule of false position tends to f'(x ) as a^ tends to x . 3. The Method of Iteration. Another means of approximation to the roots of an equation f(x) = is the method of iteration. Here we put <f>(x) =/(*) + x and write our equation in the form x = <f>(x). We then suppose that | is the true value of a solution of our equation, and x a first approximation. We obtain a second approximation x 1 by putting x 1 — c/>(x ), a third approximation x 2 by putting x 2 = (^(xj), &c. To investigate the convergence of these approxi- mations, we apply the mean value theorem; recalling that £= 4>(£)> we have f - a* = *(fl - <£(*„) = d - x )<J>' (?) where f lies between £ and x . This shows that if for \t-x\<\t-x \ the derivative <f>'(x) is less in absolute value than k < 1, then the successive approximations converge, for | £ — X! | < k | | — x Q |, \£ — x 2 \<k*\£ — x \,..., \£-x a \<k»\£-x \, and the errors therefore tend to zero. The smaller the absolute value of the derivative </>'(x) near g, the more rapid is the con- vergence. If <f>'{x) > 1 in the neighbourhood of f the approximations no longer tend to £. We can then use the inverse function or else the following device. We choose a first approximation x , calculate A =f'(x ), and write <f>(x)=-~f(x) + x. Then the equation f(x) = can be written x = <j>{x), and here <f>'(x) = — t/'(x) + 1, which has the value at x = x and A hence will usually be less in absolute value than a constant k<l if \£- X \<\$-T. \. VII] NUMERICAL SOLUTION OF EQUATIONS 359 Keturning to Newton's method, we can now investigate its suit- ability for application at any given point. The equation/(a;) = is equivalent to x = <f>(x) = x — ~J-, provided that f'(x) =j= 0. /to) Applying the method of iteration to this last equation, from a • 1 • 1 /too) first approximation x we obtain a second, a^ = x — ~ — - ; in / too) other words, the same second approximation as Newton's method gives when applied to the equation f(x) = 0. We thus see that the smaller the value of 9(> (/'to)) 2 ' the more rapidly do the successive approximations converge. In words, Newton's formula converges rapidly for large values of f'(x ) and small values of f(x ) and the curvature, as intui- tion would lead us to suspect. We can also obtain an estimate of the accuracy of Newton's method, if we recall that, since /(f) = 0, the derivative <f>'(i) = 0. Applying Taylor's theorem, we have I - *i = #(fl - <f> too) = ^=^ 2 r (& where g lies between f and x . Thus, if the error of the original estimate is small, the method converges much more rapidly than the method of iteration applied directly to f(x) — 0. For example, if ^m _ {nx)yf"^)+f'^)f(x)f"(x)-2f(x){f"(x)Y 9 v ' {/'to)} 8 is everywhere less than 10, then a first approximation which is in error by less than -001 will yield a second approximation with an error of less than (-001) 2 X 10 -f- 2 = -000005. 4. Examples. As an example we consider the equation /(*) = a? — 2x — 5 = 0. For x a ■= 2, we have /(a; ) = — 1, while for a^ = 2-1 we have /(a^) «= 0-061. By Newton's method, * " * - m = 21 ~ 3W^2 - 21 - - 005431 = 2 - 094569 - 360 NUMERICAL METHODS [Chap. To estimate the error we find from the expression (a) above that <f"(x) is about 1 and certainly less than 2 near x = 2. Moreover, the error of our first approximation is certainly less than 1/160, for the secant joining the points x= 2, y= — 1 and x = 2-1, y= -061 cuts the i-axis at a distance less than 1/160 from x = 2-1, and the curve, lying under the secant, cuts it even nearer 2-1. So the error * of our second approximation is less than 1 • — ?— = — — < -00004. 2 (160) a 25,600 If this degree of accuracy is not sufficient, we can repeat the process, calculating /(a^) and /'(a; a ) for x t = 2-094569, and obtain a third approxi- mation x. with an error less than < -000000002. (25,600)» As a second example, let us solve the equation f(x) = x log 10 a: — 2 = 0. We have /(3) = —0-6 and /(4) = +0-4, and therefore use x = 3-5 as a first approximation. Then by using ten-figure logarithmic tables we obtain the successive approximations x l = 3-598 *, = 3-5972849 x 3 = 3-5972850235. Examples 1. Using Newton's method, find the positive root of x* + 6x — 8 = to four decimal places. 2. Find to four places the root of x = tana; between n and 2tt. Prove that the result is accurate to four places. 3. Using Newton's method, find the value of x for which 1 r Jq du = Jo l + « 2 2 4. Find the roots of the equation x = 2 sin x to two places. 5. Determine the positive roots of the equation xfi — x — 0-2 = by the method of iteration. 6. Determine the least positive root of x* — 3a; 3 + 10a; — 10 = by the method of iteration. 7. Find the roots of x 3 — 7x* + 6a; + 20 = to four decimal places. •Another way of estimating the error, without referenoe to the secant, is as follows: if we estimate that the error is less than 1/20, the error of our second approximation is less than 1/20' — -0025. Hence the root differs from 2-1 by less than (2-1 - 2-0945) + -0025 - -008. Therefore the error was not merely less than 1/20, but less than -008, so that x 2 is in error by less than (-008)* — ■000064. VII] STIRLING'S FORMULA 361 Appendix to Chapter VII Stirling's Formula In very many applications, especially in statistics and in the theory of probability, we find it necessary to have a simple approximation to n! as an elementary function of n. Such an expression is given by the following theorem, which bears the name of its discoverer Stirling: Asn-voo _ __ > 1; V2iTn n+i e-" more exactly, ■s/2^n n+i e- n <nl< V2^n n+i e-"(l + j-\ In other words, the expressions n\ and "v / 27rw n+i e"" 1 differ only by a small percentage when the value of w is large — as we say, the two expressions are asymptotically equal — and at the same time the factor 1 + l/4n gives us an estimate of the degree of accuracy of the approximation. We are led to this remarkable formula if we attempt to evaluate the area under the curve y = log a;. By integration (p. 220) we find that A n , the exact area under this curve between the ordinates x = 1 and x = n, is given by / logxdx = x logx — x n logn — » + 1. If, however, we estimate the area by the trapezoid rule, erecting ordinates at x = 1, x = 2, . . . , x = n as in fig. 7, we obtain T„, an approximate value for the area: T„ = log2 + log3 + . . . + log(n - 1) + \ logn = logn! — - logn. If we make the reasonable assumption that A„ and T„ are of the same order of magnitude, we find at once that n! and n n+i e -n are f £jj e game order of magnitude, which is essentially what is stated in Stirling's formula. 362 NUMERICAL METHODS [Chap. To make this argument precise, we first show that the difference o n = A n — T„ is bounded, from which it will immediately follow that T H = A„ 1 1 — — " J is of the same order of magnitude as A n . \ AJ The difference a k+1 — a k is the difference between the area under the curve and the area under the secant in the strip k <S x ^ k + 1. Since the curve is concave down- ward and lies above the secant, a k+1 — a k is positive, and a n = (Of. — On-l) + («n-l — «n-2) +• • • + (o 2 — a x ) + a r is monotonic in- creasing. Moreover, the difference a k +x — a k is clearly less (cf. fig. 8) than the difference between the area under the tangent at x = k 4- - and the area under the secant: hence we have 2 the inequality a*+i - a k < logU + y\ - * logfc - - log(fe + 1) = 1 Iogfl + r)- 1 log(l+ 1 ) 2 g V + W 2 g V 2(k+y < l M i+ i)-l he ( i+ w+i)- VII] STIRLING'S FORMULA 363 If we add these inequalities for k = 1, 2, . . . , n — 1, all the terms on the right except two will cancel out, and (since a^ — 0), we have «- < \ l °gl ~ hog (l + -^ < - log -. 2 6 2 2 g \ 2w/ 2 8 2 Hence a„ is bounded, and being monotonic increasing it tends to a limit a as n ->■ », Our inequality for a k+1 — a k now gives us a — a„ = S (a t+1 — a*)< - log( 1 + - ). *=« 2 \ 2n/ Since by definition A n — T„= a n , we now have logn! = 1 — a„ + (n + - j logw — n, or, writing a„ = e 1-0 ", n! = a n n n+ *e"~". The sequence a„ is monotonic decreasing and tends to the limit a = e L ~ a ; hence 1< 5? = e a -°- < e * , °«< 1 + 1 ' 2 ''> = /A H- 1^ < l + A Hence we have an n+i e -n < n ! < an n +*e- n ( 1 + — V It only remains for us to find the actual value of the limit a. Here we make use of the formula proved in Chap. IV, § 4 (p. 225): V77 = lim W2*« „_ >0 o(2n)! y/ri Replacing n! by a„n n+ *e _n and (2n)! by o 2 „2 2n +*n 2n+i e- 2B , we immediately obtain yir = Iim — 2.— = , n ->- 00 a2„ V 2 a-y/2 whence a = ^Jlit. The proof of Stirling's formula is thus com- pleted. 364 NUMERICAL METHODS [Chap. VII In addition to its theoretical interest, Stirling's formula is a very useful tool for the numerical calculation of n! when n is large. Instead of multiplying together a large number of integers, we have merely to calculate Stirling's expression by means of logarithms, which involves far fewer operations. Thus for n = 10 we obtain the value 3598696 for Stirling's expression (using seven-figure tables), while the exact value of 10! is 3628800. The percentage error is barely § per cent. Example Prove that lim V—L = _. CHAPTER VIII Infinite Series and Other Limiting Processes Preliminary Remarks The geometric series, Taylor's series, and a number of special examples which we have already met in this book suggest that we may well study those limiting processes which are called the summation of infinite series from a rather more general point of view. From its nature any limiting value S= lims„ n->oo can be written as an infinite series; for if n takes the values 1, 2, 3, . . . , we need only put a„ = s„ — s„_ t for n > 1 and Oj = s x to obtain «n = <h + a 2 + ••• + «»> and the value 8 thus appears as the limit of s n , the sum of n terms as n increases. We express this fact by saying that S is the " sum of the infinite series " «i + «2 + a a + Thus an infinite series is simply a way of representing a limit where each successive approximation is found from the preceding by adding one more term. The expression of a number as a decimal is in principle merely the representation of a number a in the form of an infinite series a = Oy + a 2 -J- a 3 + . . . , where, if ^ a <^ 1, the term a„ is put equal to a„ X 10 -n and a„ is a whole number between and 9 inclusive. Since every limiting value can be written in the form of an infinite series, it may seem that a special study of series is superfluous. But in many cases it happens that limiting values occur naturally in 885 366 INFINITE SERIES [Chap. the form of infinite series, which often exhibit particularly simple laws of formation. Of course it is not true that every series has an easily recognizable law of formation. For example, the number it can certainly be represented as a decimal, yet we know no simple law enabling us to state the value of an arbitrary digit, say the 7000th, of this decimal. If, however, we set aside the representation of it by a decimal and consider Gregory's series instead, we have an expression with a perfectly clear general law of formation. Analogous to infinite series, in which the approximations to the limit are found by repeated addition of new terms, are infinite products, in which the approximations to the limit arise from repeated multiplication by new factors. We shall not go deeply into the theory of infinite products, however; the principal subject of this chapter and of the following chapter will be infinite series. 1. The Concepts op Convergence and Diveegence 1. The Fundamental Ideas. We consider an infinite series whose " general term " we denote * by o m ; the series is then of the form 00 <h. + a 2 + • • • = ^ a y i—l The symbol on the right with the summation sign is merely aD abbreviated way of writing the expression on the left. If as n increases the n-th partial sum n «n = <h + a 2 + • • • + a o == ^ a„ r—1 approaches a limit S = lim»„ II— >00 we say that the series is convergent, otherwise we say that it is divergent. In the first case we call S the sum of the series. We have already met with many examples of convergent * For formal reasons we include the possibility that certain of the numbers On may be zero. If all the a„'s from a number JV onward (i.e. when n> N) vanish, we speak of a terminating series. VIII] CONVERGENCE AND DIVERGENCE 367 series; for instance, the geometric series 1 + q + q 2 -f- , which converges to the sum 1/(1 — q) when | q j < 1, Gregory's series, the series for log 2, the series for e, and others. In the language of infinite series, Cauchy's convergence test (cf . Chap. I, § 6, p. 40) is expressed as follows: A necessary and sufficient condition for the convergence of a series is that the number I *m — s„ I = I a„ +1 -f a n+2 + . . . -f a m | becomes arbitrarily small if m and n are chosen sufficiently large (m > n). In other words: A series converges if, and only if, the following condition is fulfilled: if a positive number e is given, no matter how small, it is possible to choose an index N = N(e), which in general increases beyond all bounds as e -> 0, in such a way that the above expression | s m — s n | is less than e, provided only that m > N and n > N. We can make the meaning of the convergence test clearer by consider- ing the geometric series where q= i. If we choose e = t 1 ^, we need only takeiV=4. For JL( 1 + L+ +_U< 1 jjn-i \2 ^ 2 a 2™-"/ 2 n - 2»-» and Jl < JL if n > 4. 2»-i 10 If we choose e equal to T £ 5 , it is sufficient to take 7 as the corresponding value of N, as may easily be verified. Obviously it is a necessary condition for the convergence of a series that lim a n = 0. n— >■» Otherwise the convergence criterion certainly cannot be fulfilled. But this necessary condition is by no means sufficient for con- vergence; on the contrary, it is easy to find infinite series whose general term a n approaches as n increases, but whose sum does not exist, as the partial sum s n increases without limit as n increases. 368 INFINITE SERIES [Chap An example of this is the series 1 +-7^+-7»+--- + J- + V2 V3 Vn the general term of which is — ,— . We immediately see that V» ,„> _L+... + J_ = » -V«. V» vn vn The n-th partial sum increases beyond all bounds as n increases, and therefore the series diverges. The same is true for the classic example of the harmonic series . + ! + ! + ! + .... =-*^+-+"--irn+— +s > s + — + s - l- *" n and m = 2n can be taken as large as we please, the series diverges, for Cauchy's test is not fulfilled; in fact, the n-th partial sum obviously tends to infinity, since all the terms are positive. On the other hand, the series formed from the same numbers with alternating signs, 1 1 + 1 _ 1 + 1 -+ -|- <- iy ~ + converges (cf. Chap. VI, p. 317), and has the sum log2. It is by no means true that in every divergent series s„ tends to +00 or — oo . Thus, in the case of the series 1-1 + 1-1 + 1 + - .... we see that the partial sum s„ has the values 1 and alternately, and on account of this oscillation backwards and forwards neither approaches a definite limit nor increases numerically beyond all bounds. With regard to the convergence and divergence of an infinite series the following fact which, though self-evident, is very im- portant, should be noted. The convergence or divergence of a series is not changed by inserting a finite number of terms or by removing a finite number of terms. So far as convergence or divergence is concerned, it does not matter in the least whether we begin the series at the term a , or o^ or a b , or any other term chosen arbitrarily. VIII] CONVERGENCE AND DIVERGENCE 369 2. Absolute Convergence and Conditional Convergence. The series 1 -\ !-- + -••• diverges; but if we change the sign of every second term the resulting series converges. On the other hand, the geometric series 1 — q + q 2 — q 3 -\ ••• converges and has the sum 1/(1 -f q), provided that ^ q < 1; and on making all the signs plus we obtain the series l + q+q 2 + q*+..., which is also convergent, having the sum 1/(1 — q). Here there appears a distinction which we must examine a little more closely. With a series whose terms are all positive there are only two possible cases; either it converges or the partial sum increases beyond all bounds as n increases. For the partial sums, being a monotonic increasing sequence, must converge if they remain bounded. Convergence occurs if the terms approach zero rapidly enough as n increases; on the other hand, diver- gence occurs if the terms do not approach zero at all or if they approach zero too slowly. In series where some terms are positive and some negative, however, the changes of sign may bring about convergence, since a too great increase in the partial sums, due to the positive terms, is compensated by the nega- tive terms, so that the final result is that a definite limit is approached. 00 In order to grasp this fact the better, with a series 2 a v having positive and negative terms we compare the series which has the same terms all with positive signs, that is, KI + |« 2 | + ... = 2K|. ••—I If this series converges, then for sufficiently large values of n and m > n, the expression I «n+l I + I °n+2 I + • • • + I «m I will certainly be as small as we please; on account of the relation I «n+l + • • • + a m I ^ I «n+l I + • • • + I «m I 13 (B798) 37° INFINITE SERIES [Chap. the expression on the left is also arbitrarily small, and so the original series S a v converges. In this case the original series is said to be absolutely convergent. Its convergence is due to the numerical smallness of its terms and does not depend on the change of the signs. If, on the other hand, the series with all the terms taken positively diverges and the original series still converges we say that the original series is conditionally convergent. Conditional convergence results from the terms of opposite signs compen- sating one another. For conditional convergence Leibnitz's convergence test is frequently useful: If the terms of a series are of alternating sign and in addition their absolute values | a n | tend monotonically to (so that 00 I Vul < | a |), the series £ a„ converges. (Example: Gregory's series (p. 352)). In the proof we assume that a 1 > 0, which does not essentially limit the generality of the argument, and write our series in the form W — &2 + \ (-•••» where all the terms b„ are now positive, b n tends to 0, and the condition b n+x < b n is satisfied. If we bracket the terms together in the two ways &l-(&2-& 3 )-(&4-& 5 )---. and (b 1 -b 2 )+(b s -b i )+(b 5 -b 6 )+... we see at once that the two following relations are satisfied by the partial sums: s l > S S > s i > • • • > s 2m+l > • • ®2 <. S4 <Z S 6 <. . . . < S 2m < . . . On the other hand, s 2 „ < s 2n+1 < s t and s 2n+1 > s 2n > s 2 . The odd partial sums s 1; s 3 , ... therefore form a monotonic decreasing sequence, which in no case falls below the value s 8 ; hence this sequence possesses a limit L (p. 61). The even partial sums s 2 , s t , . . . likewise form a monotonic increasing sequence whose VIII] CONVERGENCE AND DIVERGENCE 371 terms in no case exceed the fixed number s v and therefore this sequence must have a limiting value L '. Since the numbers s 2n and s 2n+1 differ from one another only by the number b %n+1 which approaches as n increases, the limiting values L and II are equal to one another. That is, the even and the odd partial sums approach the same limit, which we now denote by S (cf. fig. 1). This, however, implies that our series is convergent, as was asserted; its sum is S. In conclusion, we make another general remark about the fundamental difference between absolute convergence and con- 00 ditional convergence. We consider a convergent series 2<v 1—1 We denote the positive terms of the series by p x , p 2 , p 3 , . . . , and the negative terms by — q v —q 2 , —q 3 , .... If we form the n-th partial sum s n — 2 a v of the given series, a certain number, say n', of positive terms and a certain number, say n", of nega- S o o o — J — 0000 e S2 S4 S 6 S s S g S 7 Sj 5j Sf Fig. 1. — Convergence of an alternating aeries tive terms must appear, where ri + n" = n. Further, if the number of positive terms as well as the number of negative terms in the series is infinite, then the two numbers n' and n" will increase beyond all bounds as n does. We see immediately that n' the partial sum s„ is simply equal to the partial sum 2 p v of the n" positive terms of the series plus the partial sum — £ q v of the negative terms. If the given series converges absolutely, then 00 the series of positive terms S p r and the series of absolute values 00 of the negative terms 2 q v certainly both converge. For as m v = l m m increases the partial sums 2 p„ and 2 q y are monotonic non- 1/— 1 K = l CO decreasing sequences with the upper bound 2 | o, I. »— 1 The sum of an absolutely convergent series is then simply equal to the sum of the series consisting of the positive terms only plus the 373 INFINITE SERIES [Chap. sum of the series consisting of the negative terms only, or, in other words, is equal to the difference of the two series with positive terms. n ri n" For 2 a, — S p v — 2 q v ; as n increases n' and n" must also pml 11=1 V=l increase beyond all bounds, and the limit of the left-hand side must therefore be equal to the difference of the two sums on the right. If the series contains only a finite number of terms of one particular sign the facts are correspondingly simplified. If, on the other hand, the series does not converge absolutely, but oo CO does converge conditionally, then the series 2 p„ and 2 q„ must both be divergent. For if both were convergent the series would converge absolutely, contrary to our hypothesis. If only one diverged, say 2 p v , and the other converged, then separation into positive and negative parts, s„ = 2 p v — 2 q„, shows that the series could not converge; for as n increases n' and 2 p v would ■-=1 *" increase beyond all bounds, while the term 2 q v would approach a definite limit, so that the partial sum s„ would increase beyond all bounds. We see, therefore, that a conditionally convergent series cannot be thought of as the difference of two convergent series, the one con- sisting of Us positive terms and the other consisting of the absolute values of its negative terms. Closely connected with this fact there is another difference between absolutely and conditionally convergent series which we shall now briefly mention. 3. Rearrangement of Terms. It is a property of finite sums that we can change the order of the terms or, as we say, rearrange the terms at will without changing the value of the sum. The question arises, what is the exact meaning of a change of the order of terms in an infinite series, and does such a rearrangement leave the value of the sum unchanged? While in the case of finite sums there is no difficulty, for example, in adding the terms in reverse order, in the case of infinite series such a possibility does not exist; there is no last VIII] CONVERGENCE AND DIVERGENCE 373 term with which to begin. Now a change of order in an infinite series can only mean this: we say that a series Oj + a% -\- a 3 + • • • is transformed by rearrangement into a series b 1 -\-b i -\-b 3 -\- . . . , provided that every term a n of the first series occurs exactly once in the second and conversely. For example, the amount by which a„ is displaced may increase beyond all bounds as n does; the only point is that it must appear somewhere in the new series. If some of the terms are moved to later positions in the series, other terms must, of course, be moved to earlier positions. For example, the series 1 + ? + ? a + ? 4 + f+ q* + q 7 + q* + f + ? 16 + . . . is a rearrangement of the geometric series 1 + q -\- q 2 + . . . . With regard to change of order there is a fundamental dis- tinction between absolutely convergent series and conditionally convergent series. In absolutely convergent series rearrangement of the terms does not affect the convergence, and the value of the sum of the series is unchanged, exactly as in the case of finite sums. In conditionally convergent series, on the other hand, the value of the sum of the series can be changed at will by suitable rearrange- ment of the series, and the series can even be made to diverge if The first of these facts, referring to absolutely convergent series, is easily established. Let us assume to begin with that our series has positive terms only, and let us consider the w-th partial n sum s n = 2 a„. AH the terras of this partial sum occur in the m m-th partial sum t m = S 6„ of the rearranged series, provided only that m is chosen large enough. Hence t m ^ s n . On the other hand, we can determine an index n' so large that the n" partial sum s n . = S a„ of the first series contains all the terms b v 6 2 , . . . , b m . It then follows that t m 5S s„. sS A, where A is the sum of the first series. Thus for all sufficiently large values of m we have s„ iS t m g A; and since s„ can be made to differ from A by an arbitrarily small amount, it follows that the rearranged series also converges; and in fact to the same limit A as the original series. 374 INFINITE SERIES [Chap. If the absolutely convergent series has both positive and negative terms, we may regard it as the difference of two series each of which has positive terms only. Since in the rearrange- ment of the original series each of these two series merely under- goes rearrangement and therefore converges to the same value as before, the same is true of the original series when rearranged. For by the case just considered the new series is absolutely convergent and is therefor© the difference of the two rearranged series of positive terras. To the beginner the fact just proved may seem a triviality. That it really does require proof, and that in this proof the absolute convergence is essential, can be shown by an example of the opposite behaviour of con- ditionally convergent series. We take the familiar series 1,11,11,11, .„ 1 — - + - —_ + _—-+_ — --\ ... = log2. 2 3 4 5 6 7 8 under it write the result of multiplication by the factor -, + - — - H ... = - logZ, 2 4 6 8 2 s ' and add, combining the terms placed in vertical columns.* We thus obtain i, 1 1,1.1 1,1,1 1 , , 3, „ 1 + -— - + - + - H-H — - + H ...= - log2. 3 257 4911 6 2* This last series can obviously be obtained by rearranging the original series, and yet the value of the sum of the series has been multiplied by the factor 3/2. It is easy to imagine the effect that the discovery of this apparent paradox must have had on the mathematicians of the eighteenth century, who were accustomed to operate with infinite series without regard to their convergence. We shall give the proof of the theorem stated above concern- ing the change in the sum of a conditionally convergent series which arises from change of order of the terms, although we shall have no occasion to make use of the result. Let p v p z , . . . be the positive terms and — q lt —q 2 , . . . the negative terms of the series. Since the absolute value | a n | tends to as n increases, the numbers p u and q n must also tend to as n increases. As * For the addition of series see No. 4, p. 376. VIII] CONVERGENCE AND DIVERGENCE 375 we have already seen, moreover, the sum* Ep, must diverge, 00 and the same is true of 2<7„. 1 Now we can easily find a rearrangement of the original series which has an arbitrary number a as limit. Suppose, to be specific, that a is positive. We then add together the first n^ positive terms, just enough to secure that the sum Sp„ is greater than a. Since the sum Sp y increases with n^ beyond all bounds, it is always possible by using enough terms to make the partial sum greater than a. The sum will then differ from the exact value a by p ni at most. We now add just enough negative terms — 2a„ to ensure that the sum 2j>„ — Sg„ is less than o; this 1 11 » is also possible, as follows from the divergence of the series Eg,. The difference between this sum and a is now q mj at most. We "i now add just enough other positive terms S p v to make the partial sum again greater than a, as is again possible, since the series of positive terms diverges. The difference between the partial sum and a is now p„ t at most. We again add just IK, enough negative terms — 2 q r , beginning next after the last one previously used, to make the sum once more less than a, and continue in the same way. The values of the sums thus obtained will oscillate about the number a, and when the process is carried far enough the oscillation will only take place between arbitrarily narrow bounds; for since the terms p„ and q v them- selves tend to when v is sufficiently large, the length of the interval in which the oscillation takes place will also tend to 0. The theorem is thus proved. In the same way we can rearrange the series in such a way as to make it diverge; we have only to choose such large numbers of the positive terms as compared with the negative that compensation no longer takes place. * This abbreviated notation for S p y , and analogous expressions for x=l other series, will often be used in future. 376 INFINITE SERIES [Chap. 4. Operations with Infinite Series. It is clear that two convergent infinite series a t -f- a 2 -f . . . = S and b x + b 2 + . . . = T can be added term by term, that is, that the series formed from the terms c n = a n + b u converges and has the value S + T for its sum.* For Sc„= £a y +ib v -+S+T. v""l v=l l"»l It is also clear that if we multiply each term of a convergent infinite series by the same factor the series remains convergent, its sum being multiplied by the same factor. In the cases just mentioned it is immaterial whether the convergence is absolute or conditional. On the other hand, further study, which is not necessary for us here, shows that if two infinite series are multiplied together by the method used in multiplying finite sums together, the product series will not usually converge or have the product of the two sums for its sum unless at least one of the two series is absolutely convergent (cf. appendix, p. 415). Examples 1. Prove that S =-L + -!_ + _L + ...= i. ,-iv(v+l) 1.2 T 2.3 3.4 2. Prove that 5 - = I. v-iv(v+l)(v+2) 4 3. Prove that S (— 1)» 2v + 3 = I ,-o (v+l)(v + 2) 4. For what values of a does the series 1— i-f-2— 1-f converge? 2 " 3« 4° 5.* Prove that if S a„ converges, and s n = % + a t + . . . + a„, then the sequence " -1 «i + % + • • • + s s N also converges, and has S o F as its limit. 6. Is the series 2 ( ] convergent? »-i \2n +1 2ra / 7. Is the series 2 (—1)" — — convergent? *-i v + 1 * This theorem is really nothing more than another statement of the fact (cf. Chap. I, § 6, p. 41) that the limit of the sum of two terms is the sum of their limits. VIII] CONVERGENCE TESTS 377 2. Tests for Convergence and Divergence We have already met with a test of a general nature for the convergence of series, which applies to series with terms of alter- nating signs and decreasing absolute value and which asserts that such series are at least conditionally convergent. In the following pages we shall only consider criteria referring to absolute convergence. 1. The Comparison Test. All such considerations of convergence depend on the com- parison of the series in question with a second series; this second series is chosen in such a way that its convergence can readily be tested. The general comparison test may be stated as follows: If the numbers b v b 2 , . . . are all positive and the series S b, converges, and if '-1 \a n \^b n CO for all values of n, then the series 2 a n is absolutely convergent. n=l If we apply Cauchy's test the proof becomes almost trivial. For if m 2> n, we have I a n + . . . + a m I ^ I o„ I + . . . + I a m | ^ b n + . . . + b m . Since the series £ b n converges, the right-hand side is arbitrarily »=i small, provided that n and m are sufficiently large. It follows that for such values of n and m the left-hand side is also arbi- trarily small, so that by Cauchy's test the given series converges. The convergence is absolute, since our argument applies equally well to the convergence of the series of absolute values Kl- The analogous proof for the following fact can be left to the reader. If \a„\^b n >0, "■* 00 and the series 2 b n diverges, then the series 2 a n is certainly not absolutely convergent. 13 * (B798) 378 INFINITE SERIES [Chap. 2. Comparison with the Geometric Series. In applications of the test the comparison series most frequently used is the geometric series. We at once obtain the following theorem: QO The series S a n w absolutely convergent if from a certain term onward a relation of the form Kl<c? B (I) holds, where c is a positive number independent of n and q is any fixed positive number less than 1. This test is usually expressed in one of the following weaker QO forms: the series 2 a„ converges absolutely, if from a certain n=l term onward a relation of the form a n+l <? (Ila) holds, where q is again a positive number less than 1 and inde- pendent of w, or: if from a certain term onward a relation of the form v | «n | < q (IK) holds, where q is a positive number less than 1. In particular, the conditions of these tests are satisfied if a relation of the form a n+l lim k<l or lim V | a n \ = k < 1 (Ilia) (UK) is true. These statements are easily established in the following way. Let us suppose that the criterion Ila, the ratio test, is satisfied from the suffix n onward, that is, when n > n . For brevity we put a„ o+m+1 = b m and find that IM<slM> IM<<zlM<2 2 IM> \b*\«i\h\«r , \b \, and so on; hence l&«l<rlM. VIII] CONVERGENCE TESTS 379 which establishes our statement. For the criterion 116, the root test, we at once have | a„ | < q n , and our statement follows immediately. Finally, in order to prove criterion III, we consider an arbi- trary number q such that k < q < 1. Then from a certain n onward, that is, when n > n , it is certain that -2±i < q or V | a n | < q, as the case may be, since from a certain term " x ' or of V | a„ | differ from k by less onwards the values of On than (q — k). The statement is then established by a reference to the results already proved. We stress the point that the four tests derived from the original criterion | a n | < cq n are not equivalent to one another or to the original, that is, that they cannot be derived from one another in both directions. We shall soon see from examples that if a series satisfies one of the conditions, it need not by any means satisfy all the others.* For completeness it may be pointed out that a series cer- tainly diverges if from a certain term onward | a n | > c for a properly chosen positive number c, or if from a certain term onward or if lira *n+l V I a„ | > 1, = k, or lira V | a n | = k, where k is a number greater than 1. For, as we immediately recognize, in such a series the terms cannot tend to zero as n increases; the series must therefore diverge. (In these circum- stances the series cannot even be conditionally convergent.) Our tests furnish sufficient conditions for the absolute con- vergence of a series; that is, when they are satisfied we can con- clude that the series converges absolutely. They are definitely not necessary conditions, however; that is, absolutely convergent series can be found which do not satisfy the conditions. * More exactly: if Ilia is fulfilled, then Ho is fulfilled; if III6, then 116; if Ilia, then III6; if Ila, then 116; and if any of the four is satisfied, then so is I. None of these statements can be reversed. 3«o INFINITE SERIES For example, the knowledge that [Chap. lim *n+l = 1 or lim ■$/ | a„ \ = 1 does not entitle us to make any statement about the convergence of the series. Such a series may converge or diverge. For example, the series 1 S for which lim V | a n \ = 1 and lim a n+l = 1, is divergent, as we saw on p. 368. On the other hand, we shall soon see that 00 ] . the series S — , which satisfies the same relations, is con- n-lW 2 vergent. As an example of the application of our tests we first consider the series q + 2q<> + 3q* + . . . + w? + . . . . For this series J™ V'KI = I S I • lim V» = | q \, lim *«+i . lim «— >« »+ 1 1*1- That the series converges if | q | < 1 follows from the ratio test and from the root test also, even in the weaker form III. If, on the other hand, we consider the series 1 + 2 2 + g" + 2? + . . . + q™ + 2g 2 "+i + . . . , we can no longer prove convergence by the ratio test when 1 g I q I < 1; for then -=^— = 2 | q | ^ 1. But the root test immediately gives us lim -y/ 1 a„ | = q\, and shows that the series converges provided that | q | < 1, which, of course, we could also have observed directly. 3. Comparison with an Integral.* We now proceed to a discussion of convergence which is independent of the preceding. We shall cany it out for the particularly simple and important case of the series "1 11 * In this connexion see also the appendix to Chap. VII (p. 361). VIII] CONVERGENCE TESTS 38i where the general term a n is l/n% a being a positive number. In order to investigate the convergence or divergence of this series, we consider the graph of the function y = 1/a? 1 and mark off on the x-axis the integral abscissae as= 1, x—2 We first construct the rectangle of height l/n a over the interval » — 1 ^x :g ra of the a--axis (n. > 1), and compare it with the area of the region bounded by the same interval of the z-axis, the ordinates at the ends, and the curve y = l/cc° (this region is shown shaded in fig. 2). Secondly, we construct the rectangle of height l/n a lying above the interval n gign -f- 1, and similarly compare it with the area of the region lying above the same interval and below the curve (this region is cross-hatched in fig. 2). In the first case the area under the curve is ob- viously greater than the area of the rectangle; in the second case it is less than the area of the rectangle. In other words, J„ VP- »« Jn-lX* Fig. 2. — Comparison of a series with an integral as we may also prove directly from the integral itself (cf . Chap. II, § 7, p. 129). Writing down this inequality for n = 2, n = 3 n = m and summing, we obtain the following estimate * for the m-th partial sum a = 2 - : * +1 dx r m dr Jr" 1 — dx tends to a finite limit or increases 1 *" without limit according as a > 1 or a ^ 1. Consequently the monotonia sequence of numbers s m is bounded or increases beyond all bounds accord- ing as a > 1 or a S 1, and we thus have the following theorem: * From this relation for o - 1 it follows at once that the sequence of numbers 0,-1+2+- + ... + -- log nis bound ed below. Since from the inequality ~-^ < J — - l og (n + 1) - l og » we see that the sequence is monotonic decreasing, it must approach a limit lim C n - Urn (1 + 5 + 5 + . . . + i - logn) - 0. The number G, whose value is -5772 . . . , is called Eider's constant. In contrast to the other important special numbers of analysis, such as it and e no other expression with a simple law of formation has been found for Euler's constant. 38a INFINITE SERIES [Chap. The aeries .^ 1 ^~F + ^ + r« + ••• is convergent — and, of course, absolutely convergent — if, and only if, a. > 1. The divergence of the harmonic series, which we previously proved in a different way, is an immediate consequence of this. In particular, we see that the series 1 + 1 + 1 + I+ 1 +^+ 13 ^ 2 3 3 3 ' converge. 00 i The series 2 — , whose convergence we have just studied, frequently ,,= 1 v a serve as comparison series in investigations of convergence. For example, 00 c we see at once that for a > 1 the series S — converges absolutely if the absolute values | c„ | of the coefficients remain, less than, a fixed bound independent of v. Examples Find whether the series in Ex. 1-6 are convergent or not: 2. I t 1 V* CO 1 3. S l Vv(v + 1) ,te the error (_l)v + l — 1 "* 1 8. S — iv! 4. 00 * S 1 2 (l0gv)«' a fixed. 5. CO s r—2 1 (logv) Io K"' 6. CO 2 V [ the series in Ex. 7-10: 9. CO s »— 1 i v" 10. CO ^=1 V 11. Prove that 2 sin s I tc( v + -J converges. CO Q0 12. Does S e~ v ' (that is, 1 + 2 S er v ') converge? co ! 13.* Prove that 2 — converges when a > 1 and diverges when a ^ L ,-2v(logv)» VII1J CONVERGENCE TESTS 383 « 1 14.* Prove that S — - — ; converges when a > 1 and diverges when a £ 1. ,=3 v log v (log log v)- 15. Prove that if u ( 3j (*' = 1, 2, 3, . . .) and S w t converges, then 00 (=1 S Uf* also converges. t"=-i OO 00 DO 16. Show that if S a k * and S b k * both converge, then S a,Jb k also converges. *" =1 * =1 * =1 17 Prove that '+$-;+;+ 1 5 -? + ! + .. 6 7 .+ X 3n+ 1 + X - 2 +• 3» + 2 3n + 3 . . = log3 18 * Prove that if n is an arbitrary 00 „ n integer greater log », than 1 where a„" is denned as follows: — n 1 if n is not a factor of «, 3. Sequences and Series op Functions 1. General Remarks. The terms of the infinite series hitherto considered have been constants; hence these series (when convergent) always repre- sented definite numbers. But both in theory and in applications tne series of outstanding importance are those in which the terras are functions of a variable, so that the sum of the series is also a function of the variable, as in the case of Taylor series. We shall therefore consider a series 9iW + 9z( x ) + 9s(x) + • • • » in which all the functions g n (x) are functions denned in an interval a 5S x ^ b. The w-th partial sum of this series, 9ii x ) + 9z( x ) + • - - + gJ?), we denote by f n (x). Then the sum f(x) of our series, where it exists, is simply the limit lim fjx). 3»4 INFINITE SERIES [Chap. We may therefore regard the sum of an infinite series of func- tions as the limit of a sequence of functions f^x), f 2 {x), . . . , f„{x), .... Conversely, for any such sequence of functions fi( x )> fi( x )> ... we can form an equivalent series by putting ?i(*) =/i(*) an d g n (x) =f n (x) —f n ^(x) for n > 1. When it is convenient, therefore, we can pass from the consideration of series to that of sequences and conversely. 2. Limiting Processes with Functions and Curves. We shall now state exactly what we mean by saying that a function /(a;) is the limit of a sequence f x {x), f 2 (x), . .., f„(x), . . . in an interval a <^ x <^ b. The definition is as follows: the sequence f^x), f z (x), . . . converges in that interval to the limit function f(x), if at each point x of the interval the values f n (x) converge in the usual sense to the value f(x). In this case we shall write lim/„(a;) =f(x). According to Cauchy's test (cf. p. 40) n— >oo we can express the convergence of the sequence without necessarily knowing or stating the limit function f(x). For our sequence of functions will converge to a limit function if, and only if, at each point x in our interval and for every posi- tive number e the quantity \f„(x) —fjx) | is less than e, pro- vided that the numbers n and m are chosen large enough, that is, larger than a certain number N = N(e). This number N(e) usually depends on e and x and increases beyond all bounds as e tends to zero. We have frequently met with cases of limits of sequences of functions. We mention only the definition of the power x" for irrational values of a by the equation x a = lhna; r », where r v r 2 , . . . , r„, . . . is a sequence of rational numbers tending to a; or the equation = lim n 00 \ n/ where the functions f n (x) on the right are polynomials of degree n. The graphical representation of functions by means of curves suggests that we can also speak of limits of sequences of curves, saying, for example, that the graphs of the above limit functions VIII] SEQUENCES AND SERIES OF FUNCTIONS 385 C 1 x Fig- 3- — Limit curve and limit function a; a and e* are to be regarded as the limit curves of the graphs of the functions x r * and ( 1 + - 1 respectively. There is, however, a fine distinction between passages to the limit with functions and with curves. Until the middle of the nineteenth century this distinction was not sufficiently observed; only by having a clear idea of it can we avoid apparent paradoxes. We shalJ illustrate this point by an example. For this purpose we consider the functions /„(*) = *" (n=l,2,...) in the interval ^ x g 1. All these functions are continuous, and the limit function lim f n (x) = /(x) exists. n — y 00 But this limit function is not con- tinuous. On the contrary, since for all values of n the value of the funotion/„(l) = 1, the limit /(1)=1; while, on the other hand, for ^ x < I, the limit f(x) = lim f„(x) = 0, n— >oo as we saw in Chap. I, § 5 (p. 33). The function f(x) is therefore a discon- tinuous function which at x = 1 has tho value 1 and for all other values of x in the interval has the value 0. This discontinuity becomes intelligible if we consider the graphs G of the functions y = f„(x). These (cf. fig. 16, p. 33) are continuous ourves" all of which pass through the origin and the point x = 1, y = 1, and which draw in closer and closer to the x-axis as n increases. The curves possess a limit curve G which is not discontinuous at all, but consists (cf. fig. 3) of the portion of the x-axis between x = and x = 1, and the portion of the line x = 1 between y = Q and y = 1. The curves therefore converge to a continuous limit curve with a vertical portion, while the functions converge to a discontinuous limit function. We thus recognize that this discontinuity of the limit function expresses itself by the occurrence in the limit curve of a portion perpendicular to the x-axis. Such a portion must involve a discontinuity in the limit function, and, in fact, such a portion is always present when the limit function is discontinuous. This limit curve is not the graph of the limit function, nor can any curve with a vertical portion be the graph of any single- valued function y = /(x); for corresponding to the value of x at which the vertical portion occurs the curve gives an infinite number of values of y and the function only one. Hence the limit of the graphs of the functions / n (x) is not the same as the graph of the limit f(x) of these functions. Corresponding statements, of course, hold for infinite series also. 386 INFINITE SERIES [Chap. 4. Uniform and Non-uniform Convergence 1. General Remarks and Examples. The distinction between the concept of the convergence of functions and that of the convergence of curves introduces a phenomenon which it is essential that the student should clearly recognize. This is the so-called non-wniform conver- gence of sequences or infinite series of functions. Since it is well known that beginners usually find difficulties here, we shall discuss the matter in some detail. /f'~^ ^ ->y-flx)+t ' is^~ ^*Ai)-f(x) y-fnCcf J f\, t 1 1 a b x Fig. 4. — To illustrate uniform convergence That a function f(x) is the limit of a sequence fi(x), fyx), . . . in an interval a 5£ x ^ b means only, by definition, that the usual limit relationship f(x) = lim/(a;) holds at each point x of the interval. From a naive point of view one might expect that the following fact would automatically follow from this concept of convergence: if we assign an arbitrary degree of aocnracy, say e ~ Taoo oi € ~ 77wV *^ ei1 * rom a cerfcam index N onward all the functions /„(a?) will lie between/fa) + e and f(x) — e for all values of x, so that their graphs y=f n (x) will lie entirely in the strip indicated in fig. 4. That is to say, for every positive e there is a corresponding number N = N (e), which, of course, will ordinarily increase beyond all bounds as c -*- 0, such that for » > N the difference \f(x) —f„{x) | < e, no matter where x is chosen in the interval. (If this condition is fulfilled, then VIII] UNIFORM CONVERGENCE 387 \fn(x) —fmfo) I < 2e for all values of x, provided that n and m are both greater than N.) If the accuracy of the approximation can be made at least equal to a pre-assigned number e everywhere in the interval at the same time, that is, by everywhere choosing the same number N(e) independent of x, we say that the approximation is uniform. One is at first astonished to find that the naive assumption that convergence is necessarily uniform is entirely wrong; in other words, that convergence may very well be non-uniform. Ex. 1. Non-uniform convergence occurs in the case of the sequence of functions just considered, f n (x) = x"; in the interval Ogjz^l this sequence converges to the limit function f(x) = for g x < 1, /(l) = 1. Convergence occurs at every point in the interval; that is, if e is any posi- tive number, and if we select any definite fixed value x = 5, the inequality | 5" — /(?) | < e certainly holds if n is sufficiently large. Yet this approximation is not uniform. For, if we choose e = J, then no matter how large the number n is chosen, we can find a point x = y) 4= 1 at which | if — /(yj) | = rf > £; this is, in fact, true for all points x — tq where 1 > yj > ty\. It is therefore impossible to choose the number n so large that the difference between f(x) and f n {x) is less than \ throughout the whole interval. This behaviour becomes intelligible if we refer to the graphs of these functions (fig. 3, p. 385). We see that no matter how large a value of n we choose, for values of \ only a little less than 1 the value of the function /n(5) will be very near 1, and therefore cannot be a good approximation to /(£), which is 0. Similar behaviour is exhibited by the functions /„(*) = ! + »*» in the neighbourhood of the points x = 1 and x = — 1; this can easily be established. (Compare also the discussion in Chap. I, § 8 (p. 52)). Ex. 2. In the two examples above the non-uniformity of the conver- gence is connected with the fact that the limit function is discontinuous. Yet it is also easy to construct a sequence of continuous functions which do converge to a continuous limit function, but not uniformly. We restrict our attention to the interval ^ x 2j| 1 and make the follow- ing definitions for » J> 2: f„(x) = xn a for ^ x ^ -, n /«(*) = (- - *) »" for - 5 x g ?, \» / n n f n (x) = for - ^ x g 1, n J88 INFINITE SERIES [Chap. where to begin with we can choose any value for a, but must then keep this value of a. fixed for all terms of the sequence. Graphically our functions are represented by a roof-shaped figure made of two line segments lying over the interval ^ x ^ 2/»i of the x-axis, while from x — 2/n onwards the graph is the x-axis itself (cf. fig. 6). If a < 1, the altitude of the highest point of the graph, which has in general the value » a_1 , will tend to as n increases; the curves will then tend towards the x-axis, and the functions f n (x) will converge uniformly to the limit function f(x) = 0. If a = 1, the peak of the graph will have the height 1 for every value of n. If a. > 1, the height of the peak will increase beyond all bounds as n increases. But no matter ho-w a. is chosen, the sequence f^x), f 2 {x), . . . always tends to the limit function /(x) = 0. For, if x is positive, for all sufficiently n A Fig. 5. — To illustrate non-uniform convergence large values of n we have 2/ra < x, so that x is not under the roof-shaped part of the graph and / n (x) = 0; for x = all the functional values /„(«) are equal to 0, so that in either case lim /„(x) = 0. The convergence is certainly non-uniform, however, if a ^ 1; for it is plainly impossible to choose n so large that the expression |/(x) — /„(x) | = / n (x) is less than J everywhere in the interval. Ex. 3. Exactly similar behaviour is exhibited by the sequence of functions (cf. fig. 6) / n (x) = xn"er«*, where, in contrast with the preceding case, each function of the sequence is represented by a single analytical expression. Here again the equation ft" 1 fn( x ) — holds for every positive value of x, since as n increases the n — >» function e~ nx tends to to a higher order than any power of l/» (of. Chap. Ill, § 9, p. 192). For x = 0, we have always /„(x) = 0, and thus f(x) = lim /„(x) = VIII] UNIFORM CONVERGENCE 389 for every value of a: in the interval O^igo, where a is an arbitrary positive number. But here again the convergence to the limit function is not uniform. For at the point x = - (where f n (x) has its maximum) we have and we thus recognize that if a 2: 1, the convergence is non-uniform; for every curve y = f„(x), no matter how large n is chosen, will contain points (namely, the point x = , which varies with n, and neighbouring * 1 points) at which /„(*) - f{x) = f„(x) > —. Ze Fig. 6 Ex. 4. The ooncepts of uniform and non-uniform convergence may, of course, be extended to infinite series. We say that a series gi(x) + 3i(x) + • • • is uniformly convergent, or not, according to the behaviour of its partial sums f„(x). A very simple example of a non-uniformly convergent series is given by x* iB* X* /(*) = *+__ + __+___ + ... . For * = every partial sum f n (x) = x* + . f (T+l^ hMthe value 0; therefore /(0) = 0. For x 4= the series is simply a geometric series with the positive ratio < 1; we can therefore sum it by the 1 + 3? elementary rules and thus obtain for every i+0 the sum a~> 1 - 1/(1 + **) = 1 + x\ The limit function f(x) is thus given everywhere except at x = by the expression f(x) — 1 + x\ while /(0) = 0; it therefore has a somewhat artificial-looking discontinuity at the origin. 390 INFINITE SERIES [Chap. Here again we have non-uniform convergence in every interval con- taining the origin. For the difierence f(x) — f n (x) = r n (x) is always for x = 0, while for all other values of x it is given by the expression r n( x ) = Tp-r — tt=zp as the reader may verify for himself. If we require (1 ~f- ar)™" -1 this expression to be less than, say, J, then for each fixed value of x this can be attained by choosing n large enough. But we can find no value of n sufficiently large to ensure that r„(x) is everywhere less than J; for if we fix upon any value of n, no matter how large, we can make r n (x) greater than J by taking x near enough to 0. A uniform approximation to within J is therefore impossible. The matter becomes clear if we consider the approximating curves (cf. fig. 7). These curves, except near x == 0, lie nearer and nearer to the parabola y = 1 + x* as n increases; near x = 0, however, the curves send down a narrower and narrower extension to the origin, and as n increases this extension draws in closer and closer to a certain straight line, a portion of the y-axis, so that for limiting curve we have the parabola plus a linear extension reaching vertically down to the origin. As a further example of non-uniform convergence we mention the 00 series 2 g y (x), where g v (x) = x" — a;" -1 for v ^ 1, g e {x) = 1, defined in the interval OgaiSl. The partial sums of this series are the functions x" already considered in the first example (p. 387). 2. A Test of Uniform Convergence. The preceding considerations show us that the uniform con- vergence of a sequence or series is a special property not possessed VIII] UNIFORM CONVERGENCE 39 1 by all sequences and series. We shall now formulate the concept of uniform convergence again. The convergent series 9ii*) + 9a( x ) + ; • • is said to be uniformly convergent in an interval if the sum f(x) can be approximated to within e (where e is an arbitrarily small positive number) by taking a number of terms which is suffi- ciently large and which is the same throughout the interval. We suppose first that the series g^x) + g 2 (x) + . . . converges at every point of a certain interval to a limit function f(x); by f n {x) we denote the »-th partial sum of the series, f n (x) = 9i( x ) + • • • + 9dp)> and by Jt n (x) the remainder of the series after n terms, K(x)=f(x)-f n (x). The series g^x) + g 2 (x) + ... is said to be uniformly con- vergent in the interval if to every positive number e there corresponds a number N, dependent on e alone and not on x, such that for n > N the inequality | R n (x) | = | f (x) — f^x) | < e holds for all values of x in the interval. Expressed more pictorially, the partial sum f n (x) represents the sum f(x) to within an error of less than e everywhere in the interval at the same time, provided only that n is chosen large enough. By Cauchy's test we readily see that the series converges uniformly if, and only if, the difference |/ n (a:) —fjx) | can be made less than an arbitrary quantity e everywhere in the interval by choosing n and m larger than a number N independent of x. For, firstly, if the convergence is uniform we can make \f«{x) —f{x) | and \f m (x) — /(*) | both less than e/2 by choosing n and m greater than a number N independent of x, from which itfollowsthat |/„(x) -/„(») \<e; and secondly, if \f n (x)-fjx) |<e for all values of x whenever n and m are greater than N, then on choosing any fixed value of n > N and letting m increase beyond all bounds we have the relation !/.(*) -/(*) I = lim \f„(x) ~f m (x) | ^ e , for every value of x, so that the convergence is uniform. If we wish to speak of the uniform convergence of a sequence of functions we need make only trifling changes in the above 392 INFINITE SERIES [Chap. definition; the sequence f^x), f 2 (x), . . . converges uniformly to f(x) in an interval if the difference |/(a;) —f n (x) | can be made less than « everywhere in the interval by choosing n greater than a number N independent of x. As above, a necessary and sufficient condition for the uniform convergence of the sequence is that | f n {x) —fjx) j < e for all values of x when n and m are both greater than a certain number N dependent on e but not on x. We shall soon see that it is just this condition of uniform convergence that makes infinite series and other limiting pro- cesses with functions into convenient and useful tools of analysis. Fortunately, in the limiting processes usually encountered in the calculus and its applications, non-uniform convergence is a sort of exceptional phenomenon which will scarcely trouble us in our present applications of analysis. In most cases the uniformity of convergence of a series is established by means of the following criterion: 00 If the terms of the series S g„(x) satisfy the condition | g„(x) | ^ a,,, where the numbers a„ are constants which form a convergent series 00 00 S a„, then the series 2 g„(x) converges uniformly (and, we may incidentally remark, absolutely). For we then have | lg v (x) | ^S | &Gr)| ^So„ v«=n v=n v*=n m and since by Cauchy's test the sum 2 a v can be made arbitrarily small by choosing n and m > n large enough, this exactly ex- presses the necessary and sufficient condition for uniform con- vergence. A first example is ofiered by the geometries series 1 -f- x -\- x* + . . . where x is restricted to the interval \x\ g q, q being any positive number less than 1. The terms of the series are then numerically less than or equal to the terms of the convergent geometric series Eg*. A further example is given by the " trigonometric series " frsinfo— Sj) c 8 sin(a;— S a ) c g sin(a!— 8 3 ) VIII] UNIFORM CONVERGENCE 393 provided that | c n | < c, where c is a positive constant independent of n. For then we have ,„(*) = "»»*"(»-«■) , so that , (x) 1 < c Hence the uniform and absolute convergence of the trigonometric series 00 c follows from the convergence of the series 2 — • 3. Continuity of the Sum of a Uniformly Convergent Series of Continuous Functions. As we have already hinted, the significance of the uniform convergence of an infinite series lies in the fact that a uniformly- convergent series in many respects behaves exactly like the sum of a finite number of functions. Thus, for example, the sum of a finite number of continuous functions is itself continuous, and correspondingly we have the following theorem: If a series of continuous terms converges uniformly in an in- terval, its sum is also a continuous function. The proof is quite simple. We subdivide the series f(x) = g x (x) + g 2 (x) + ... into the n-th partial sum f n (x) plus the remainder R n (x). As usual, f n (x) — g^x) + . . . + g„{x). If now any positive number e is assigned, we can in virtue of the uniform convergence choose the number n so large that the remainder is less than e/4 through- out the whole interval, and hence \R n (x+h)-R n (x)\<^ for every pair of numbers x and x -\- h in the interval. The par- tial 8um/„(a;) consists of the sum of a finite number of continuous functions and is therefore continuous; for each point x in the interval, therefore, we can choose a positive 8 so small that !/.(* + *)-/.(*)!<? provided j h | < 8 and the points x and x + A lie in the interval. It then follows that \f(x + h) -fix) I = \f n {x + A) -/.(as) + R n (x +h)- R n (x) I ^ I /J* + h) -f n (x) I + I R n (x +h)- R„(x) I < e , which expresses the continuity of our function. 394 INFINITE SERIES [Chap. The significance of this theorem becomes clear when we recall that the sums of non-uniformly convergent series of continuous functions are not necessarily continuous, as our previous examples show. From the preceding theorem we may conclude that if the sum of a convergent series of continuous functions has a point of discontinuity, then in every neighbourhood of this point the convergence is non-uniform. Hence every representation of dis- continuous functions by series of continuous functions is based on the use of non-uniformly convergent limiting processes. 4. Integration of Uniformly Convergent Series. A sum of a finite number of continuous functions can be " integrated term by term "; that is, the integral of the sum can be found by integrating each term separately and adding the integrals. In the case of a convergent infinite series the same procedure is permissible, provided that the series converges uniformly in the interval of integration. 00 A series S g p (x) = f (x) which converges uniformly in an interval can be integrated term by term in that interval: or, more precisely, if a and x are two numbers in the interval of uniform convergence^ 00 /.X the series 2 / g„(t)dt converges, and, in fact, converges uniformly with respect to x for each fixed value of a, its sum being equal to f(t)dt. To prove this we write as before f(x) = Xg v (x)=f n (x) + R n (x). We have assumed that the separate terms of the series are con- tinuous; hence by the previous sub-section the sum is also con- tinuous and therefore integrable. Now if e is any positive number, we can find a number N so large that for every n> N the inequality | R„(x) | < e holds for every value of a; in the interval. By the first mean value theorem of the integral calculus we have •'a | jf (/(*)-/.(*))& el, VIII] UNIFORM CONVERGENCE 395 where I is the length of the interval of integration. Since the integration of the finite axwif n (x) can be performed term by term, this gives us /"/(«)&-£ fgJit)dt <el. But since el can be made as small as we please, this states that S f 9v (t)dt = lim S [ X gMdt= f f(t)dt, which was to be proved. If, instead of i nfinit e series, we wish to deal with sequences of funotions, our result can be expressed in the following way: If in an interval the sequence of functions f^x), f 2 (x), . . . tends uniformly to the limit function f(x), then £f(x)dx = lim f f„(x)dx for every pair of numbers a and b lying in the interval; in other words, we can then interchange the order of the operations of in- tegration and passing to the limit. This fact is far from being a triviality. It is true that from a naive point of view such as prevailed in the eighteenth century the inter- ohangeability of the two processes is hardly to be doubted; but a glance at the examples in No. 1 of this seotion (p. 387) shows us that in the case of non-uniform convergence the above equation might not hold. We need only consider Ex. 2 (p. 387), in which the integral of the limit function is 0, while the integral of the function f n (x) over the interval <: x ^ 1, that is to say, the area of the triangle in fig. 5, p. 388, has the value / Jo .1 f n (x) dx = »»-*, 'o and when a. ^ 2 this does not tend to zero. Here we immediately see from the figure that the reason for the difference between / f(x)dx /*1 Jq f n {x) dx lies in the non-uniformity of the convergence. .. , - On the other hand, by considering values of a such that 1 ^ a < 2, we see that the equation lim / f„{x)dx= f f(x)dx can hold good although the convergence is non-uniform. As a further example, the series S g n (x), where g n {x) = x n — a:"-* f or n > 1 and ?0 (x) = 1, can be inte- 396 INFINITE SERIES [Chap. grated term by term between the limits and 1, even though it does not converge uniformly. Thus while uniformity of convergence is a sufficient condition for term-by-term integrability, it is by no means a necessary condition. Neglect of this point may easily lead to misunder- standing. 5. Differentiation of Infinite Series. The behaviour of uniformly convergent series or sequences with respect to differentiation is quite different from that with respect to integration. For example, the sequence of functions f n (x) = certainly converges uniformly to the limit func- n tion f(x) = 0, but the derivative f n '(x) = n cosw 2 ^ certainly does not converge everywhere to the derivative of the limit function f'{x) = 0, as we see by considering x = 0. In spite of the uniformity of the convergence, therefore, we cannot change the order of the processes of differentiation and passing to the limit. Corresponding statements of course hold for infinite series. For example, the series . 8^2% . sin3*x , is absolutely and uniformly convergent, for its terms are numeri- cally not greater than the terms of the convergent series T5 + S5 + ^ + • • • • K > however, we differentiate the series 12 2 2 3 2 term by term, we obtain the series cosas + 2 2 cos2*a; + 3* cos3*a; + . . . , which plainly does not converge everywhere; for example, it diverges at x = 0. The only useful criterion which assures us in special cases that term-by-term differentiation is permissible is given by the following theorem: If, on differentiating a convergent infinite series S G„(x) = F(x) term by term, we obtain a uniformly convergent series of continuous 00 terms S g„(x) = f(x), then the sum of this last series is equal to the v-0 VIII] UNIFORM CONVERGENCE 397 derivative of the sum of the first series. This theorem, therefore, expressly requires that after differentiating the series term by term we must still investigate whether the result of the differen- tiation is a uniformly convergent series or not. The proof of the theorem is almost trivial. For by the theorem in No. 4 (p. 394) we can integrate term by term the series ob- tained by differentiation. Recalling that gj(t) = G r '(t), we obtain f X f(t)dt = f (2 gjtf))dt = S fg v (t)Zt = S (6 v (x) - <?») = F(x) — F(a). This being true for every value of x in the interval of uniform convergence, it follows that /(*) = F(x), which was to be proved. Examples 1. Show by comparison with a aeries of constant terms that the follow- ing series converge uniformly in the intervals stated: (a) x - x* + a? - x* + . . . (- J g x g \). (6) JVl - a* + JVl - «• + JVl - *• + . . . + 5= Vl^T^n + . . . (-lgigl). 2 , . sina: sin2a; sinna: , ( c >^r+-p- + --- + -^- + .... (d) e* + e 2 * -J- . . . + e"* + . . . (-2 g x g -1). 2. Prove that lim/„(x) = 0, where /„(») = ** -Igi^i. Prove that the convergence is non-uniform. 1 + n x* 3* (a) Find lim f n (x), where /.(*) = f ** -1 ^ x ^ 1. Prove that the convergence is non-uniform. Prove that nevertheless -1 T lim f f n (x)dx = f lim f n (x)dx. (6) Discuss the behaviour of the sequence given by f n (x) = ~ n*x* 1 -r- n-x" with regard to convergence, uniform convergence, and term -by -term integrability. 4.* Sketch the curves y = /„(*) = _, — 2 g x g 2, for n = 1, 1 + X?" 3, 10. Find lim f„(x). Prove that the convergence is non-uniform. 398 INFINITE SERIES [Chap. X 5. Show that 2 e~<*-">' converges uniformly in any fixed interval 6. Show that in the interval ^ x g ie the following sequences con- verge, but not uniformly: (a) v'sinz. (d) [/(*)]» where f(x) = !^?, /(0) = 1. (6) (sinz)» * sin a; x i \ "/ ' " h sin x (c) yzsinz. (e) y^), where /(*) = , /(0) = 1. 7. The sequence f n (x), n = 1, 2, . . . , is defined in the interval O^J^l by the equations (a) Prove that in the interval O^igl the sequence converges to a continuous limit. (&)* Prove that the convergence is uniform. 8.* Let f (x) be continuous in the interval Ogi^i The sequence of functions f„(x) is defined by Prove that in any fixed interval O^iga the sequence converges uni- formly to 0. 9. Sketch the curves x* n + y in = 1 for n = 1, 2, 4. To what limit do these curves tend as n -> oo ? 10.* Let f n (x), n = 1, 2, . . . , be a sequence of functions with con- tinuous derivatives in the interval a :£ x Ss 6. Prove that if /„(») con- verges at each point of the interval and the inequality | f n '(x) \ < M (where M is a constant) is satisfied for all values of n and x, then the convergence is uniform. 5. Power Series Among infinite series, power series occupy the chief place. By a power series we mean a series of the type 00 P(x) = c + <%x + C 2 X* + ... = £ c y x* *=.o (" power series in x "), or more generally 00 P(x) = c + c^x — a: ) + c 2 {x — x ) 2 + . . . = 2 c v (x — x ) r (" power series in (a; — x ) "), where x is a fixed number. If in the last series we introduce f = x — x as a new variable, VIII] POWER SERIES 399 it becomes a power series 2 c v £° in the new variable £, and we v = can therefore confine our attention to power series of the more 00 special form S c v x" without any loss of generality. >-=o In Chap. VI (p. 320) we considered the approximate repre- sentation of functions by polynomials and were thus led to the expansion of functions in Taylor series, which are in fact power series. In this section we shall study power series in some- what greater detail, and shall obtain the expansions of the most important functions in series in simpler and more convenient ways than before. 1. Convergence Properties of Power Series. There are power series which converge for no value of x, except of course for x = 0, e.g. the series x + 2 a a? + 3^ + . . . + n n x n + ... . For if x 4= 0, we can find an integer N such that I x I > 1/2V. Then all the terms n n x n for which n > N will be greater than 1 in absolute value, and, in fact, as n increases n n x n will increase beyond all bounds, so that the series fails to converge. On the other hand, there are series which converge for every value of x; for example, the power series for the exponential function, e»= l + x + - + -+ .... 2! 3! ' whose convergence for every value of x follows at once from the ratio test (Criterion Ilia, p. 378). The (n + l)-th term divided by the w-th term gives x/n, and, whatever number x is chosen, this ratio tends to zero as n increases. The behaviour of power series with regard to convergence is expressed in the following fundamental theorem: If a power series in x converges for a value x = f , it converges absolutely for every value x such that | x | < | £ \, and the convergence is uniform in every interval \ x | <S -q, where 7) is any positive number less than j £ |. Here r) may He as near | £ | as we please. 00 The proof is simple. If the series Sc„f converges, its terms tend to as n increases. From this follows the weaker 4oo INFINITE SERIES [Chap. statement that the terms all lie below a bound M independent of v, that is, | c v g | < M. If now q is any number such that < q < 1, and if we restrict x to the interval | x | :S q | £ |, then | cX | ^ | c„£" | q" < Mq\ In this interval, therefore, the 00 terms of our series £ c v x" are smaller in absolute value than the o terms of the convergent geometric series ZMg". Hence from the theorem on p. 392 the absolute and uniform convergence of the series in the interval —q\ g\^,xi^q\ g\ follows. If a power series does not converge everywhere, that is, if there is a value x = g for which it diverges, it must diverge for every value of x such that | x | > | g |. For if it were convergent for such a value of x, by the theorem above it would have to converge for the numerically smaller value g. From this we recognize that a power series which converges for at least one value of x other than and which diverges for at least one value of x has an interval of convergence; that is, a definite positive number p exists such that for | x | > p the series diverges and for | x \ < p the series converges. For | x | = p no general statement can be made. The limiting cases, that in which the series converges only for x = and that in which it converges everywhere, are expressed sym- bolically by writing p — and p = oo respectively.* For example, for the geometric series 1 + x + 3? + . . . we have p = 1; at the end-points of the interval of convergence the series diverges. Similarly, for the series for the inverse tangent (p. 319), arc tana; = x — x 3 /3 + xfi/S h • • • > we have p = 1, and at both the end-points x = ±1 of the interval of convergence the series converges, as we recognize at once from Leibnitz's test (p. 370). * It is possible to find this interval of convergence directly from the co- efficients Cy of the series. If the limit lira «V \c n \ exists, then n— >•« v _ 1 P limV/lcf »— >» In general p U given by the formula lim $ |c„| where Em is the symbol for the upper limit, as defined in the appendix to Chap. I (p. 62). VIII] POWER SERIES 401 From the uniform convergence we derive the important fact that within its interval of convergence (if such an interval exists) the power series represents a continuous function. 2. Integration and Differentiation of Power Series. On account of the uniformity of convergence it is always permissible to integrate a power series 00 f{x) = Sc/ ■-=0 term by term over any closed interval lying entirely within the interval of convergence. We thus obtain the function F(x) = c + Z, 4r^ for which F\x)=f{x). Further, since J, I c„ I for all values of v, the series v+1 obtained by integration converges more rapidly than the original series. We can also differentiate a power series term by term within its interval of convergence, thus obtaining the equation f'(x) = 2 vc v x-~ l - In order to prove this statement we need only show that the series on the right converges uniformly if a; is restricted to an interval lying entirely within the interval of convergence. Sup- pose then that f is a number, lying as close to p as we please, for which 2 c r £" converges; then, as we have seen before, the numbers \c y ^\ all lie below a bound M independent of v, so that M I c v ^~ l I < j — i = N. Now let q be any number such that < q < 1; if we restrict x to the interval | x | 5S q | £ | , the terms of the series under discussion are not greater than those of the series £ | vc v q"~ 1 £*~ 1 \, and therefore less than those of the •-=.1 00 14 ( It 7981 402 INFINITE SERIES [Chap. series 2 NvqT 1 . But in this last series the ratio of the (n + l)-th term to the n-th term is — — — q, which tends to q as w increases. n Since < q < 1, it follows (criterion Ilia, p. 378) that this series converges. Hence the series obtained by differentiation con- verges uniformly, and by the theorem at the end of last section (p. 396) represents the derivative f'(x) of the function f(x), which proves our statement. If we apply this result again to the power series 00 f'(x) = 2 vc v af~\ we find on differentiating term by term that /"(x) = Sv(v- l)c„x"- 2 , *=2 and, continuing the process, we arrive at the theorem: Every function represented by a power series can be differentiated as often as we please within the interval of convergence, and the differen- tiation can be performed term by term.* 3. Operations with Power Series. The preceding theorems on the behaviour of power series are our justification for operating in the same way with power series as with polynomials. It is obvious that two power series can be added or subtracted by adding or subtracting the corresponding coefficients (see p. 376). It is also clear that a power series, like any other convergent series, can be multi- plied by a constant factor by multiplying each term by that factor. On the other hand, the multiplication and division of two power series requires somewhat more detailed study, for * As an explicit expression for the fc-th derivative we obtain f<*>(x) -£■>(»- 1) ... (v - i + l)^"-*, y-t or in a slightly different form, 4! These two formulas are frequently useful. £©—-£(* ;>"«-■ VIII] POWER SERIES 4°3 which we refer the reader to the appendix (p. 416). Here we merely mention without proof that two power series 00 f{x) = Io/ »- 00 and g(x) = Tib y 3f K-0 can be multiplied together like polynomials. To be specific, we have the following theorem: throughout the common part of the intervals of convergence of these two series their product is 00 given by the convergent power series 2 c y x", where the coefficients »— o c„ are given by the formulae c = ct b , <H. = a o°i + <hK> C a = a b 2 + <hW + a 2 b , c« = «o*» + °i&«-i + • • • + «n*0» (For the proof see the appendix, § 1, p. 416.) 4. Theorem of Uniqueness for Power Series. In the theory of power series the following fact is of impor- 00 00 tance: if two power series So/ and Si,/ both converge in an v=0 >-=o interval which contains the point x = in its interior, and if in that interval the two series represent the same function f{x), then they are identical, that is, the equation o„ = 6„ is true for every value of n. In other words: A function f(x) can be represented by a power series in x in only one way, if at all. Briefly, the representation of a function by a power series is " unique ". To prove this we need only notice that the difference of the 00 two power series, that is, the power series <j>(x) = ~Lc v x" with coefficients c,= a„ — 6„, represents the function <j>(x)=f(x)-f(x) = 404 INFINITE SERIES [Chap. in the interval; that is, this last power series converges to the limit everywhere in the interval. For x = 0, in particular, the sum of the series must be 0; that is, c = 0, so that a = 6 . We now differentiate the series in the interior of the interval, 00 obtaining (f>'(x) = S vc v x v ~ x . But <f>'(x) is also throughout i—i the interval; hence for x = 0, in particular, we have c x = 0, or %= b v Continuing this process of differentiating and then putting x = 0, we find successively that all the coefficients c„ are equal to zero, which proves the theorem. We see, in addition, that we can draw the following con- clusion from the above discussion: if we take the v-th derivative of a series /(a;) = Ea.X and then put x = 0, we at once obtain a, = l/e)(0), that is: Every power series which converges for points other than x = is the Taylor series of the function which it represents. The uniqueness of the expansion is here expressed by the fact that the coefficients are uniquely determined by the function itself. 6. Expansion op Given Functions in Power Series. Method op Undetermined Coefficients. Examples Within its interval of convergence every power series repre- sents a continuous function with continuous derivatives of all orders. We shall now discuss the converse problem of the expan- sion of a given function in a power series. In theory we can always do this by means of Taylor's theorem; in practice we often meet with difficulties in the actual calculation of the ra-th derivative and in the estimation of the remainder. But we can often reach our goal more simply by making use of the following device. CO We first write down the statement f(x) — 2 c y af, where the coefficients c„ are unknown to begin with. Then by some known property of the function f(x) we determine the coefficients, and then prove the convergence of the series. The series represents a function, and it only remains to prove that this function is VIII] EXPANSION IN POWER SERIES 405 identical with f(x). Because of the uniqueness of the expansion in power series we know that no other series than the one just found can be the required expansion. We shall now consider some examples of this method. Actually, we have already obtained the series for arc tana; and log(l + a;) by a method which forms part of the range of ideas of the present chapter. For we simply integrated term by term the series for the derivatives of these functions, which we knew to be geometric series. 1. The Exponential Function. Our problem is to find a function /(*) for which f'(x) — f(x) and /(0) =1. If we write down the series with undetermined coefficients f(x) = c + c 1 x + c&?+ ... , and differentiate it, we obtain J'(x) = d + 2CaX + 3c,,s> + . . . ; Since by hypothesis these two power series must be identical, we have the equation "Cfl = c «-i> true for all values of n ^ 1. If we observe that because of the relation f(0) = 1 the coefficient c must have the value 1, we can calculate all the coefficients successively, and obtain the power series m=1 + Y< + i + i + '— As we easily see by the ratio test, this series converges for all values of x and therefore represents a function for which the relations f'(x) = f(x), /(0) = 1 are actually fulfilled. (Here we intentionally avoid making any use of what we have previously learned about the expansion of the ex- ponential function). Now the function e" certainly possesses these properties; we readily deduce that the function f(x) is identical with e*. For if we form the quotient <p{x) = f{x)jei" and differentiate we have ^■ ^-/^ ■t The function <p(x) is therefore a constant, and since it has the value 1 for x = 0, it must be identically equal to 1, thus proving that our power series and the exponential function are identical (cf. the analogous dis- cussion on p. 178). 4°* INFINITE SERIES [Cha> 2. The Binomial Series. We can now return to the binomial series (Chap. VI, § 3, p. 329), this time making use of the method of undetermined coefficients. We wish to expand the function/(a;) = (1 + a:)* in a power series, and therefore write f(x) = (1 + *)«• = Co + c x x + c^ 2 + . . . , the coefficients c„ being undetermined. We now notice that our function obviously satisfies the relation (1 + x)f'(x) = af(x) = S a.c„x'. On the other hand, if we differentiate the series for f(x) term by term and multiply by (1 + x), we obtain (1 + x)f'(x) = c, -f (2c 2 + cjx + (3c s + 2c 2 )*2 + . . . ; and since these two power series for (1 + x)f'{x) must be identical, o^o — 6 i» «<h = 2cj + Cn ac a = 3c« + 2c 2 Now it is certain that c = 1, since our series must have the value 1 for x = 0, and so we obtain in succession the expressions r - „ , - (a- 1 )* ,. _ (a-2)(a-l)q cj — a, c 2 , c 4 — , ... i for the coefficients, and in general, as is easily established, we have c» (a- v + !)(«- v+2)...(a-l)a = /a\ v(v— 1)...2.1 W" 00 M Substituting these values for the coefficients, we have the series S I Ja;'j y—0 vv/ we have yet to investigate the convergence of this series and to show that it actually represents (1 + x) a . By the ratio test we find that when a is not a positive integer, the series converges if | x \ < 1 and diverges if | x | > 1; for then the ratio a — Ti+1 of the (n -f- l)-th term to the m-th term is x, and the absolute n value of this expression tends to | x | as n increases beyond all bounds.* Hence, if | x | < 1 our series represents a function f(x) which satisfies the condition (1 + x)f'(x) = ocf(x), as follows from the method of forming the * Here we state, without proof, the exact conditions under which this series converges. If the index a is an integer ^0, the series terminates and is there- fore valid for all values of x (becoming the ordinary binomial theorem). For all other values of a the series is absolutely convergent for | x \ < 1 and divergent for | a: | > 1. For x — + 1 the series converges absolutely if a > 0, converges conditionally if - 1 < a < 0, and diverges 3 a ^ — 1. Finally, at X — - 1 the series is absolutely convergent if a > 0, divergent if a < 0. VIII] EXPANSION IN POWER SERIES 4°7 coefficients. Moreover, /(0) = 1. But these two conditions ensure that the function J(x) is identical with (1 + x) a . For on putting 9(*)« /<*)/(! + ■)■ we find that 9 '(x) = (1 + xYi ' (x) ~ g(1 + x r~ l W = 0; (1 + a;) 2 " <f(x) is therefore a constant, and, in fact, is always equal to 1, since <p(0) = 1. We have therefore proved that when | x \ < 1 (! + *)« = 2 (")*", which is the binomial series. Here we quote the following special cases of the binomial series: the geometric series = (1 + *)-i=l — x + x* — x* + x*— + ... 1+x the series 1 (1 + *) 2 . 2 (-iy x -, v=0 = (1 + a:)- 2 = 1 — 2x + 3x* — 4ic" + — . , S (-l)"(v+l)x", v=0 which may also be obtained from the geometric series by differentiation; and the series 2.4.6.8 + ■••• 1 .»(l + a .j-*»i_U+LL- 8 -«i--i^*ai V(l + x) 22.4 2.4.6 , 1-3-5.7 + 2T4T6^* + "-' the first two or three terms of which form useful approximations. 3. The Series for arc sin;*;. This series can be obtained very easily by expanding the expression l/v'(l — t a ) according to the binomial series, 408 INFINITE SERIES [Chap. This series converges if | t | :£ 1, and so converges uniformly if | t | ^ q < 1. On integrating term by term between and x, we obtain , lx* , 1.31 s . arc smx = x + A + . . • ; 2 3 2.45 by the ratio test we find that this converges if | x \ < 1, and diverges if |*| >1. The deduction of this series from Taylor's theorem would be decidedly less convenient, owing to the difficulty of estimating the remainder. 4. The Series for ar sinn* = log {x -j- -y^l + x 2 )}. We obtain this expansion by a similar method. Using the binomial theorem we write down the series for the derivative of ar sinhx, 1 1 2 1.3 , 1.3.5 , V(l + * 2 ) 2 2.4 2.4.6 and then integrate term by term. We thus obtain the expansion 1 x 3 , 1 . 3 x 5 ar suini = x — + 1- . . . , 2 3 2.45 whose interval of convergence is — 1 ^ x ^ 1. 5. Example of Multiplication of Series. The expansion of the function log(l + * ) 1 + x is a simple example of the application of the rule for the multiplication of power series. We have only to multiply the logarithmic series , „ . , 3? x 3 x* l0g(l + X)=X— ~ + j—j+ — ... by the geometric series 1 = 1 — x + x* — 3? + x*— + ... ; 1 + X as the reader may verify for himself, we obtain the remarkable expansion ^— (>+5)-+0+j+J)* -(>+H + jW-- for j x | < 1. VIII] EXPANSION IN POWER SERIES 409 6. Example of Term-by-Term Integration (Elliptic Integral). In previous applications we have met with the elliptio integral J V(l - k* sin 2 <p) (the period of oscillation of a pendulum (p. 302) ). In order to evaluate the integral we can first expand the integrand by the binomial theorem, thus obtaining 1 1 + -& 2 sin 2 ? + — jfc 4 sin 4 ? V(l-fc 2 sin 2 9) 2 r 2.4 . 1.3.5... . „ + -— — -i 8 sm 6 <p+... i 2.4.6 Since k 2 sin 2 9 is never greater than k % this series converges uniformly for all values of <p, and we may integrate term by term: ' ?1 =/ d<p + -k* sin*? dtp V(l- ifc a sin 2 <p) J r 2 J * , 1.3,, W 2 . , . , + e 4 / sin 4 <p <J<p + . . . . 2.4 J The integrals occurring here have already been calculated (cf. Chap. IV, § 4, p. 223). If we substitute their values we have For further examples on the theory of series we refer the reader to the appendix (p. 415). Examples Determine the intervals of convergence of the series S a n x n , where a„ is given by the formulas in Ex. 1-20: n_1 J 15. (-01 - l) n . 16. W (2n)l ._ »+ Vn 1. 1 n 2. n. 3. 1 Vn 4. Vn. 5. 1 »T 2 ' 6. n 7 1 a + n I4» 8. an -f- 6 9 1 log(n + 1) 10 1 log log lOn 11. 1 $n' 12. a n . 13. a Vn . 14. a l °e". n 2 — n 18. 1 1 + a n 19. 20. Vn 1 _l)n n (■798) 4io INFINITE SERIES [Chap. Expand the functions in Ex. 21-26 in power series: 21. a x . 24. cos 2 x. •"• -^ • 25. sin's. 23. sin 2 *. 26. arc sins". 27. Using the binomial series, calculate V2 to four decimal places. 28. Obtain approximations in series for the following integrals by expanding the integrand in a power series and integrating: (a) f lsi ? x dx. (c) /• 1 l°g(l + *) fa . Jo x J X (b) J ft dx , r lQ dx I Vd - *•)• w J s V(i + x*) 29. By multiplication of power series obtain the expansions of the following up to the terms in a 4 : i„\ .js „,•_ ~ / v arc sin x (a) er smx. (c) Vl-a; (6) Dogd + x)?. . (a) sin 2 a;. 30.* By multiplication of power series prove that (a) e*e* = e»"-». (6) sin 2x = 2 sin x cos a. 31. If the interval of convergence of the power series Sa B i" is | x | < p, and that of 'Zb n x n is | x \ < p', where p < p', what is the interval of con- vergence of S(a„ + &„):*:"? 32. Using the method of undetermined coefficients, find the function f(x) which satisfies the following conditions: (o)/(0)=3; (&) /'(*) = /(*) + *. 7. Power Series with Complex Terms 1. Introduction of Complex Terms into Power Series. The similarity between certain power series representing functions which are apparently unrelated led Euler to set up a purely formal connexion between them, found by giving complex values, in particular, pure imaginary values, to the variable x. We shall first do this formally, unhindered by questions of rigour, and shall investigate the results of the process. The first striking relation of this sort is obtained if we replace VIII] COMPLEX POWER SERIES 4" the quantity x in the series for e* by a pure imaginary ij>, where (f> is a real number. If we recall the fundamental equation for the imaginary unit i, that is, i 2 = — 1, from which it follows that i s = — i, i* = 1, i 5 = i, . . . , then on separating the real and the imaginary terms of the series, we obtain or, in another form, e»* = cos^ -)- i sin^. This is the well-known and important " Euler's formula "; as yet it is purely formal. It is consistent with De Moivre's theorem (p. 74), which is expressed by the equation (cos<£ -\- 1 sin<£)(cos^ + * sin^r) = cos(<£ + t[i) + i sin.(<f> -\- ifi). In virtue of Euler's formula this equation merely states that the relation C" . e v = e*+ v continues to hold for pure imaginary values x = i(f>, y^itfi. If we replace the variable x in the power series for cos a; by the pure imaginary ix we at once obtain the series for cosh a;; this relation can be expressed by the equation cosh x = cosmj. In the same way we obtain sinha;= - sinia;. Since Euler's formula also gives e^'*= cos^ — % sin<£, we arrive at the exponential expressions for the trigonometric functions, e ix — e~ ix e ix 4- e~ ix sins= — , cosa;=- — ! . 2t 2 These are exactly analogous to the exponential expressions for the hyperbolic functions and are, in fact, transformed into them by the relations coshx = cosia;, sinhz = T sinix. i 412 INFINITE SERIES [Chap. Corresponding formal relations can, of course, be obtained for the functions tanas, tanhx, cot a;, cotha;, which are connected by the equations tanha; = - tana;, cotha; = icotix. Finally, similar relations can also be found for the inverse trigonometric and hyperbolic functions. For example, from « = tana; = = u »(«*» + «-*■) i(e**+l) we immediately find that 1 — iy If we take the logarithms of both sides of this equation and then write x instead of y and arc tana; instead of x, we obtain the equation 1 , 1 + ix arc tana; = — log -, %, 6 1 — vx which expresses a remarkable connexion between the inverse tangent and the logarithm. If in the known power series for _ log ' - (p. 318) we replace x by ix, we actually obtain 2 1 — x the power series for arc tana;, 1 .. . (ix) 3 . (ix) 6 . . arc tanx = - (ix + ^-~~ -f ^— '— + . . . ) i 3 5 x 3 . X s , The above relations are as yet of a purely formal character, and naturally call for a more exact statement as to the meaning they are intended to convey. In the next sub-section we shall indicate how this can be given with the help of function theory. For later use, however, we shall only need Euler's formula e*'* = cos <j> + i sin <f> and, this being so, we can avoid a thorough analysis. We need only regard the symbol e** as a formal abbreviation for the right-hand side cos<£ + *sin<£, in which case De Moivre's formula e** . e 1 '* = e , ' M,+ ' w appears merely as a consequence of the elementary addition theorems of trigonometry. From this formal point of view, in order to make VIII] COMPLEX POWER SERIES 413 the relation c* . e v = e* +v remain valid for any complex arguments, we set up the further definition e* = e* (cos 17 + i sirnj), where x = f + i-q (f, 17 being real). 2. A Glance at the General Theory of Functions of a Complex Variable. Although the purely formal point of view indicated above is in itself free from objection, it is still desirable to recognize in the above formulae something more than a mere formal connexion. To follow out this aim leads us into the general theory of functions, as (for the sake of brevity) we call the general theory of the so-called analytic functions of a complex variable. In this we may use as our starting-point a general discussion of the theory of power series with complex variables and complex coefficients. The construction of such a theory of power series offers no difficulty once we define the concept of limit in the domain of complex numbers; in fact, it follows the theory of real power series almost exactly. But as we shall not make any use of these matters in what follows we shall content ourselves here by stating certain facts, omitting the proofs. It is found that the following generalization of the theorem of § 5, No. 1 (p. 400), holds for complex power series: If a power series converges for any complex value x — £ what- ever, then it converges absolutely for every value x for which I x I < I 1 1; if it diverges for a value x = £, then it diverges for every value xfor which | x | > | £ |. A power series which does not converge everywhere, but does converge for some other point in addi- tion to x = 0, possesses a circle of convergence, that is, there exists a number p > such that the series converges absolutely for | x | < p and diverges for | x | > p. Having once established the concept of functions of a com- plex variable represented by power series, and having de- veloped the rules for operating with such functions, we can think of the functions e*, sina;, cosa;, arc tana;, &c., of the complex variable x as simply defined by the power series which represent them for real values of x. Then all the above formal relationships reduce to trivialities. We shall merely indicate by two examples how this intro- duction of complex variables helps us to understand the elemen- 4H INFINITE SERIES [Chap. tary functions. The geometric series for 1/(1 -\- x 2 ) ceases to converge when x leaves the interval — 1 ^ x ^ 1, and so does the series for arc tana;, although there are no peculiarities in the behaviour of these functions at the ends of the interval of convergence; in fact, they and all their derivatives are con- tinuous for all real values of x. On the other hand, we can readily understand that the series for 1/(1 — as 2 ) and log(l — x) cease to converge as x passes through the value 1, since they become infinite there. But this divergence of the series for the inverse oo tangent and the series 2 (— l)"a^" for | x | > 1 immediately i/=0 becomes clear if we consider complex values of x also. For we find that when x =i the sum-functions become infinite and so cannot be represented by a convergent series. Hence by our theorem about the circle of convergence the series must diverge for all values of x such that | x | > | i |; in particular, for real values of x the series diverge outside the interval — 1 ^ x jS 1. Another example is given us by the function f(x) = e~ 1/xS for x 4= 0,/(0) = (see pp. 196, 336), which, in spite of its apparently regular behaviour, cannot be expanded in a Taylor series. As a matter of fact, this function ceases to be continuous if we take pure imaginary values of x — i£ into account. The function then takes the form e ll( ' and increases beyond all bounds as $ -> 0. It is therefore clear that no power series in x can represent this function for all complex values of a: in a neighbourhood of the origin, no matter how small a neighbourhood we choose. These remarks on the theory of functions and power series of a complex variable must suffice for us here. VIII] MULTIPLICATION OF SERIES 4*5 Appendix to Chapter VIII 1. Multiplication and Division of Sebies 1. Multiplication of Absolutely Convergent Secies. Let i = 2o„ B=S6 F be two absolutely convergent series. Together with these we consider the corresponding series of absolute values I = S|a„| and B = Jl\b r \. We further put A n = Za„ B„ = S&„, I„ = SK|, B n = i\b y \ V—0 V=0 V=0 F— and c„ = a b n + a 1 b n _ 1 -f . . . + a n b . 00 We assert that the series S c„ is absolutely convergent, and that its sum is equal to AB. " =0 To prove this, we write down the series «(A) + <*A + °J>i + «A + °2*0 + °A -f" «A + «A + a oK + • • • + « A + <* A + . . . + aj>n + ... + <hf>n + %?>„ + • • • » the w 2 -th partial sum of which is A„B n , and we assert that it con- verges absolutely. For the partial sums of the corresponding series with absolute values increase monotonically; the n 2 -th partial sum is equal to A n S n , which is less than AS (and which tends to AB). The series with absolute values therefore con- verges, and the series written down above converges absolutely. The sum of the series is obviously AB, since its n?-th partial sum is A„B n , which tends to AB as n -*■ oo. We now interchange the order of the terms, which is permissible for absolutely con- vergent series, and bracket successive terms together. In a 4i6 INFINITE SERIES [Chap. convergent series we may bracket successive terms together in as many places as we desire without disturbing the convergence or altering the sum of the series, for if we bracket together, say, all the terms (a n+1 + a B+2 -f- . . . + a m ), then when we form the partial sums we shall omit those partial sums that originally fell between s„ and s m , which does not affect the convergence or change the value of the limit. Also, if the series was absolutely convergent before the brackets were inserted, it remains absol- utely convergent. Since the series 00 2 c v = (a b ) + (<*„&! + ajbo) + (a b 2 + a x \ + a 2 b ) + ... i/=0 is formed in this way from the series written down above, the required proof is complete. 2. Multiplication and Division of Power Series. The principal use of our theorem is found in the theory of power series. The following assertion is an immediate con- sequence of it: the product of the two power series 00 00 2 a v x v and 2 b„x v v=0 i/=0 is represented in the interval of convergence common to the two CO power series by a third power series 2 c„x", whose coefficients are given by "~° c ^= a oK + <hK-i + • • • + aj> . As for the division of power series, we can likewise represent the quotient of the two power series above by a power 00 series 2 q v x", provided b , the constant term in the denominator, »— o does not vanish. (In the latter case such a representation is in general impossible; for it could not converge at x = on account of the vanishing of the denominator, while on the other hand every power series must converge at x = 0.) The coefficients of the power series 00 »— o VIII] MULTIPLICATION OF SERIES 417 can be calculated by remembering that 2 q v x* . 2 b r x* = 2 a„x r , bo that the following equations must be true: a o = ?(&)> °i = Vo b i + ?i & <» a 2 = Voh + qfo + g- a & , °v = %K + ?A-i + • • - + ?A- From the first of these equations (ft, is readily found, from the second we find the value q x , from the third (by using the values of q and j x ) we find the value q 2 , &c. In order to give strict justi- fication for the expression of the quotient of two power series by the third power series we have still to investigate the con- 00 vergence of the formally-calculated power series 2 q v x". We »-o * shall pass over this general investigation, of whose result we shall make no further use, and shall content ourselves with the statement that the series for the quotient does actually converge, provided x remains within a sufficiently small interval, in which the denominator does not vanish and both numerator and denominator are convergent series. 2. Infinite Series and Improper Integrals The infinite series and the concepts developed in connexion with them have simple applications and analogies in the theory of improper integrals (cf. Chap. IV, § 8, p. 249). Here we confine ourselves to the case of a convergent integral with an infinite interval of integration, say an integral of the form f f(x) Ax. If we divide up the interval of integration by a sequence of numbers x = 0, a^, . . . tending monotonically to + °o, we can write the improper integral in the form f o f{x)dx=a 1 + a 2 + .... where each term of our infinite series is an integral; f{x) dx, <%=) f(x)dx,..., 4i8 INFINITE SERIES [Chap. and bo on. This is true no matter how we choose the points 05„. We can therefore reduce the idea of a convergent improper integral to that of an infinite series in many ways. It is especially convenient to choose the points as, in such a way that the integrand does not change sign within any indi- 00 vidual sub-interval. The series S | a„ | will then correspond to K=l the integral of the absolute value of our function, jf|/(aO|cfa. We are thus naturally led to the following concept: an improper X 00 f (x)dx is said to be absolutely convergent if the integral |f(x) | dx exists. Otherwise, if our integral exists at all, we say that it is conditionally convergent. Some of the integrals considered earlier (pp. 250-251), such as /•" 1 r 00 a r m / 7- — .dx, I e-^dx, V(x)= e-'t®- 1 *, Jo l + X* Jo Jo are absolutely convergent. On the other hand, the integral Jp°° sins , ,. r A sin* , ! dx = hm / dx, o x ^->oo Jo £ studied on p. 251, is a simple example of a conditionally convergent integral. In order to give a proof of the convergence of this integral which is inde- pendent of the former proof, we subdivide the interval from to A at the points x v — vtc(v = 0, 1, 2, . . . , jj.^) where \x A is the largest possible in- teger for whioh \l a tz g A. We therefore divide the integral into terms of /"" r sin* the form a v = / dx{y = 1, 2, . . .), and a remainder R A of the form f -fH^tfce (0 g A — y.jit < 7t). It is clear that the quantities a v have alternating signs, since sins is alternately positive and negative in consecutive intervals. Moreover, | a^+j | < | a v |; for on applying the transformation x = 5 — tz, we have [sins | ^ = /•("+!> |sin(g-7r)| = X^+D- | sm g[ Jim 5 — W Jvk 5 — n J(v—Vfr ffr+l>| single VIII] IMPROPER INTEGRALS 4*9 Hence by Leibnitz's test we see that £a v converges. Moreover, the remainder B A has the absolute value ^ 1 /-^ + i> 2 :g / sins \ ax = , few.,, H* TC and this tends to as 4 increases. Thus if we let A tend to « in the equation C^sinie r „ I da; = % + <% + a, + . . . + o„ + ic^, Jo x A the right-hand side tends to Ea„ as a limit, and our integral is convergent. But the convergence is not absolute; for , , f I sinal , 2 ,„,,,. \a v > / dx = — , so that 21 a, diverges. 3. Infinite Pboduots In the introduction to this chapter (p. 366), we called atten- tion to the fact that infinite series are only one way, although a particularly important way, of representing numbers or func- tions by infinite processes. As an example of another such process, we shall introduce the infinite product. No proofs will be given. On p. 223 we met with Wallis's product, w = 2 2 4 4 6 6 2 1'3'3'5'5*7"* in which the number w/2 is expressed as an " infinite product ". By the value of the infinite product 00 II a„ = Oj . Oa . a 3 . a 4 . . . we mean the limit of the sequence of partial products «!, % . Og, Oj . Oj . Og, «! . a 2 . a 8 . a 4 , ...» provided it exists. 430 INFINITE SERIES [Chap. The factors a^, a 2 , a 3 , . . . , of course, may also be functions of a variable x. An especially interesting example is the " infinite product " for the function sinsc, sinwa;= 77a; (*-©(>-»(-» which we shall obtain in § 4 of the next chapter (p. 445). The infinite product for the " zeta function " plays a very important part in the theory of numbers. In order to retain the notation usual in the theory of numbers we here denote the independent variable by «, and we define the zeta function f or a > 1 by the expression We know (cf. § 2, p. 380 et seq.) that the series on the right converges if 8 > 1. If p is any number greater than 1, we obtain the equation ^~1 =sl+ p> + p Ts + p T.+ "- P* by expanding the geometric series. If we imagine this series written down for all the prime numbers p t , p v p& . . . in increasing order of magnitude, and all the equations -thus formed multiplied together, we obtain on the left a product of the form 1-Pi- 1-Pf It, without stopping to justify the process in any way, we multiply to- gether the series on the right-hand sides of our equations, and in addition remember that by an elementary theorem each integer n > 1 can be ex- pressed in one and only one way as a product of powers of different prime numbers, we find that the product on the right is again the function £(«), and so we obtain the remarkable " product form " r^ 1 * 1 This " product form ", the derivation of which we have only briefly sketched here, is actually an expression of the zeta function as an infinite product, since the number of prime numbers is infinite. In the general theory of infinite products we usually exclude the case where the product a^ • • • a « nas * ne 1™"* zer0 * Hence it is specially important that none of the factors a„ should vanish. In order that the product may converge, the factors a n must VIII] INFINITE PRODUCTS 421 accordingly tend to 1 as n increases. Since we can if necessary omit a finite number of factors (this has no bearing on the ques- tion of convergence), we may take it that a n > 0. The following theorem applies to this case: a necessary and sufficient con- 00 dition for the convergence of the product II o„, where o„ > 0, 00 V=l is that the series S logo„ should converge. For it is clear that n the partial sums Sloga y = log {c^a^ . . . a») of this series will v=l tend to a definite limit if, and only if, the partial products a^a 2 . . . a n possess a positive limit. In studying convergence we usually apply the following criterion (a sttfficient condition), where we put a, = 1 -J- a„. The 00 product II (1 + a„) converges, if the series =1 S|a„ converges and no factor (1 -f- a„) is zero. In the proof we may assume, after omission of a finite number of factors if necessary, that each | a y | < -. Then we have 1 — | a„ | > -. By the 1 mean value theorem log(l + h) = log(l + h) — logl = h 1 -+- 6h for < < 1. Therefore I log(l + «,)!■ l+0a„ 00 and so the convergence of the series 2 log (1 -J- «„) follows from 00 K«»l the convergence of 2 | a„ |. „=i Prom our criterion it follows that the infinite product given above for sin nx converges for all values of x except for x = 0, ±1, ±2 where factors of the product are zero. Moreover, for p ^ 2 and * > 1 we readily find that 1 , , 1 1 2 1 — p-» p* — 1 p' — 1 p' 422 INFINITE SERIES [Cha*. Now if we let p assume all prime values, the series S— must converge, P° 00 1 since its terms form only a part of the convergent series S — . The oon- vergence of the product II for a > 1 is thus proved. I — p-» 4. Seeies involving Beenoulli's Numbees So far we have given no expansions in power series for certain elemen- tary functions, e.g. tana;. The reason is that the numerical coefficients which occur are not of any very simple form. We can express these coefficients, and those in the series for a number of other functions, in terms of the so-called Bernoulli's numbers. These numbers are certain rational numbers, with a not very simple law of formation, which occur in many parts of analysis. We arrive at them most simply by expanding the function x 1 1 + 2! + 3! + -'- in a power series of the form s ^v. e»-l „-ov! If we write this equation in the form 00 n *—(e»- 1) 2 —x" and substitute on the right the power series for fs" — 1, we obtain, as on p. 417, a recurrence relation which enables all the numbers B„ to be calculated. These numbers are called Bernoulli's numbers.* They are rational, since in their formation only rational operations are concerned; as we easily recognize, they vanish for all odd indices other than v = 1. The first few are JL - 5 30' B * = — nn > 5 io = HH> • • • * In some works a slightly different notation is used, the basio formula being written — - i - i- + v /_ iy+1 -Ql x *» e»- 1 2 X „_i ( l; (2v)l !E • VIII] BERNOULLI'S NUMBERS 4*3 We must content ourselves with a brief hint as to how these numbers are involved in the power series in question. First, by m akin g use of the transformation B 2 x ,"_x e?+ 1 _ x el* + e~** 1 + 2[ * + •••=■ jTTi + 2 ~ 2 " e^~l "" 2 " e** - e~** we obtain ?coth?= 2 ^fx". 2 2 „= (2v).' If we replace x by 2x, we have the series 00 OScR iBCOth*= S ~^**% v- o (2v)! for |*| < 7T, from which, by replacing x by — &r, we obtain * cot* = 2 (-1)* ^%**, I x I <k. By means of the equation 2 cot 2* = cot a; — tana; we now obtain the series co 2 a "f2 2 " — l\ tans = 2 (-1)- 1 \ 9 ., ^ a^- 1 . -=i (2v)! which holds for I x I < -. 1 ' 2 For further information we must refer the reader to more detailed treatises,* owing to the lengths of the proofs involved. Examples 1. Prove that the power series for V(l — x) still converges when *= 1. 2. Prove that for every positive e there is a polynomial in x which represents V(l — x) in the interval ^ x ^ 1 with an error less than e. 3. Prove that for every positive e there is a polynomial in ( which represents 1 1 1 in the interval — 1 g t ^ 1 with an error less than s. 4. Weierstrasa' Approximation Theorem. Prove that if f(x) is con- tinuous in a ^ x ^ 6, then for every positive e there exists a polynomial P(x) such that \f(x) — P(x)\ < e for all values of x in the interval o ^ x ^ &. 5. Prove that the following infinite products converge: n (i + <$)*»); n !Lz_j n (i - -), a | «i< l. »-i n-a n s + 1 n-l * »/ * See e.g. K. Knopp, Theory and Application of Infinite Series, p. 183 (Blackie ft Son, Ltd., 1928). 424 INFINITE SERIES [Chap. VIII 6. Prove by the methods of the text that II [ 1 -f- - ) diverges. 7. Using the identity "1 " / 1 \ 2 - = II ( ) (where p { is the t-th prime) »_i n* t-i \1 — p t ~*J Pi prove that the number of primes is infinite. 8. Prove the identity 1 n (1 + a; 2 ") = v=l 1—35 for |*| < 1. CHAPTER IX Fourier Series In addition to the power series there is another class of infinite series which plays a particularly important part both in pure mathematics and in applications. These are the Fourier series, in which the individual terms are trigonometric functions and the sum is a periodic function. 1. Periodic Functions 1. General Remarks. Periodic functions of the time, that is, functions which repeat their course after a definite interval of time, are met with in many applications. In most machines a periodic process takes place in rhythm with the rotation of a flywheel, e.g. the alter- nating current developed by a dynamo. Periodic functions are also associated with all vibration phenomena. A periodic function with period 21 is represented by the equation f(x+2l)=f(x), true for all values of x. We specially call attention to the fact that 21 is called the period* It is worth notice that in addition to the period 21, the function f{x) necessarily has the period U * In representing periodic functions it is often convenient to denote the independent variable * by a point on the circumference of a circle instead of the usual point on a straight line. If a function f(x) has the period 2ir, say, that is, if the equation f{x + 2*)-f(x) is true for all values of x, and if we denote by x the angle at the centre of a circle of unit radius which is included between an arbitrary initial radius and the radius to a variable point on the circumference, then the periodicity of the function /(a;) is expressed simply by the fact that to each point on the circum- ference there corresponds just one value of the function. In the case of a machine, for example, the periodicity may be expressed in terms of the position of a point on the flywheel. 425 426 FOURIER SERIES [Chap. also, since f(x+4l)=f(x+2l)=f(x); f(x) likewise has periods 6Z, 81, ... ; and it is also possible (though not necessarily true) that f(x) may have shorter periods, such as I or J/5. Graphically, in any two consecutive intervals of length 21 the graph of the function has exactly the same form. In order to have available a second interpretation which some readers may prefer, we may think of the variable x as the time (and accordingly we some- times write ( instead of x), the function f(x) then representing a periodic process or, as we shall also say, a vibration (or oscil- lation). The period 21 = T is then called the period of vibration (or oscillation). If any arbitrary function i (x) is given in a definite interval, sa y> — I ^ x ^ I, it can always be extended as a periodic function; we have only to define f(x) outside the interval by the equation f(x + 2nl) =f(x), where n is an arbitrary positive or negative integer. Here we must point out that if f{x) is continuous in the interval — I <L x ^ I, but /(— 1) =# f(+l), our extended periodic function will be discontinuous at the points +1, +21, . . . , (cf. figs. 7 and 8 (pp. 441, 442), in which I = -n). Further, in this case the extension fails to give us a single- valued function f(%) at the points x = ±1, +3J, . . . since e.g. we have defined f{2>l) as f(l -f 21), which gives /(3?) = f(l), and we have also defined it as /(— I + U), which gives f(3l) ==/(— I). We avoid this difficulty by extending, not the function as defined for —l^x^l, but the function as defined either for — I < x 5S I or — I j£ x < I; that is, we discard either the original value f(—l) or the original value f(+l). Here we would point out a general fact relating to periodic functions, which is expressed by the equation /, f(x)dx = f f(x)dx, or, in words: the integral of a periodic function over an interval whose length is one period T = 21 always has the same value, no matter where the interval lies. To prove this we need only notice that in virtue of the equation f(£ — 21) =/(£) the sub- stitution x = £ — 21 gives us f(x)dx=J f(£)d£=J f(x)dx. IX] PERIODIC FUNCTIONS 427 In particular, for a = — I — a and /? = — I it follows that J , /(a) dx = f f(x) dx, J —l— a J I— a and hence r l-a f f(x)dx=f~ f(x)dx + f~"f(x)dx pi pi — a pi =J l _J( x ) dx +j f( x ) dx =J_ l f( x ) dx > which proves our statement. If we recall the geometrical meaning of the integral, the statement is made obvious by fig. 1. Fig. i. — To illustrate the integral over a whole period The simplest periodic functions, from which we shall later build up the most general periodic functions, are the functions a sinoxr and a cos cock, or more generally a sin w (x — £) and a cosco(x — f ), where a(2:0), <u(>0), and £ are constants. The processes represented by these functions * we call sinusoidal vibrations or simple harmonic vibrations (or oscillations). The period of vibration is T = 2irlw. The number a> is called the circular frequency of the vibration; -f since 1/T is the number of vibrations in unit time, or frequency, w is the number of vibra- tions in time 2n. The number a is called the amplitude of the vibration; it represents the maximum value of the function a sinaj(a; — £) or a cosco(a; — £), since both sine and cosine have the maximum value 1. The number co(x — |) is called the phase and the number wg the epoch or phase displacement. * Either of these formulae taken by itself (for all values of a and £) represents the class of all sinusoidal vibrations; and the two formulae are equivalent to one another, since asinco(x — f) — acos<x>{x — (( + tt/2o>)}. t The reader should take care to distinguish between the frequency and the circular frequency (Ger., Kreisfrequem). 428 FOURIER SERIES [Chap. We obtain these functions graphically by stretching the sine curve in the ratios 1 : o along the a;-axis and a : 1 along the j/-axia, and then translating the curve a distance ij ™ the positive direction along the s-axis (cf. fig. 2). By the addition formulae for the trigonometric functions we can also express sinusoidal vibrations in the form a coswx -f- /} sinaxt and j3 coswas — a sinwx Fig. 2. — Sinusoidal vibrations respectively, where a = — a sincuf and jS = o cos<o£. Con- versely, every function of the form a cosgjx + fi sinwa; represents a sinusoidal vibration a sinco(a; — f) with the ampli- tude a= \/(a?-\- /J 2 ) and the phase displacement w£ given by the equations a = — a sinw|, /$ = a cosa>|. By using the expression a cosajcc + ft sin <mx we see that the sum of two or more such functions with the same circular frequency o> always represents another sinusoidal vibration with the circular frequency a>. 2. Superposition of Sinusoidal Vibrations. Harmonics. Although many vibrations are found to be sinusoidal (cf . Chap. V, § 4, p. 296), it is nevertheless true that most periodic motions have a more complicated character, being obtained by the superposition of sinusoidal vibrations. Mathematically this IX] PERIODIC FUNCTIONS 429 simply means that the motion, e.g. the distance of a point from its initial position as a function of the time, is given by a function which is the sum of a number of pure periodic functions of the above type. The sine waves of the function are then piled up on top of one another (that is, their ordinates are added), or, as we say, they are superposed. In this superposition we assume that the circular frequencies (and, of course, the periods) of the superposed vibrations are all different; for the superposition of two sinusoidal vibrations with the same circular frequency gives us another sinusoidal vibration with the same circular frequency (but with a different amplitude and phase displacement), as shown above. If we consider the simplest instance, the superposition of two sinusoidal vibrations with the circular frequencies ci> x and u> 2 , we find that there are two fundamentally different cases, depend- ing on whether the two circular frequencies have a rational ratio or not, or, as we say, whether they are commensurable or in- commensurable. We begin with the first case, and by way of an example take the second circular frequency to be twice the first; to 2 = 2^. The period of the second vibration will then be half the period of the first, 2^/20)! = T % = TJ2, and so it will neces- sarily have not only the period T 2 but also the doubled period T x , since the function repeats itself after this double period; and the function formed by superposing them will also have the period T v The second vibration, with twice the circular frequency and half the period of the first, is called a first harmonic of the first vibration (the fundamental). Corresponding statements hold if we introduce a further vibration with the circular frequency w 3 = 30^. Here again the vibration function sin 3^3; will necessarily repeat itself with the period 2^/^ = T x . Such a vibration is called a second harmonic of the given vibration. Likewise we can consider third, fourth, . . . , (n — l)-th harmonics with the circular frequencies a> 4 = 4*u 1; o) 5 = 5ft*!, . . . , <o„ = nco v and, moreover, with any phase dis- placements we please. Every such harmonic will necessarily repeat itself after the period T r = 2-n\ot x , and consequently every function obtained by superposing a number of vibrations, each of which is a harmonic of a given fundamental circular frequency <xt v will itself be a periodic function with the period 2tt\u> x = T v By superposing vibrations with circular frequencies ranging from 43° FOURIER SERIES that of the fundamental to that of the (n ■ obtain a periodic function of the form [Chap. l)-th harmonic we S (x) = a + 2 (a„ cos vcox + 6„ sin vwx). Fig. 3*. — Combination of vibrations (The constant a, which we have introduced here in order to make the formula slightly more general, does not affect the periodicity, since it is periodic for any period.) Since this function contains sin 2* sm*-5% 2 *+ 8 J n A* i 3 j_ sin 2» , sin Sx _ sin 4» 2 3 4 Fig. 4. — Combination of vibrations 2n -f- 1 constants which we can choose arbitrarily, we are thus able to generate very complicated curves which are not at all like the original sine curves. Figs. 3-5 illustrate this graphically. * The proportions of the figure correspond to the assumption at - 1. IX] PERIODIC FUNCTIONS 43i The term " harmonic " originates in acoustics,* where we find that if a fundamental vibration with circular frequency a> corre- sponds to a note of a certain pitch, then the first, second, third, &c, harmonics correspond to the sequence of harmonics of the funda- mental, that is, to the octave, octave -f- fifth, double octave, &c. In general, in the case of the superposition of vibra- tions in which the circular frequencies have rational ratios these circular fre- quencies can all be repre- sented as integral multiples of a common fundamental circular frequency. The V>\ superposition of two vibra- ^ Fig. St. — Combination of vibrations tions with incommensurable circular frequencies <a 1 and <w 2 , however, represents an intrinsically different type of pheno- menon. Here the process resulting from the superposition of sinusoidal vibrations will no longer be periodic. We cannot go into the mathematical discussions that arise from this, but merely remark that such functions always have an approxi- mately periodic character, or, as we say, are almost periodic. Such functions have just recently been studied in great detail. A final remark on the superposition of sinusoidal vibrations is concerned with the phenomenon of so-called beats. If we superpose two vibrations both with unit amplitude but with different circular frequencies a> 1 and w z , and if for the sake of simplicity we take the same value of | (see p. 427) for both (the generalization to arbitrary phase can be left to the reader), then we are merely concerned with the behaviour of the function y = saiw-jX + amourfc (o> x > w 2 > 0). * In acoustics the terms overtone, (tipper) partial are also used, t The curves drawn in the figure correspond to the trigonometrical poly- nomials obtained by taking 3, 5, 6 and 7 terms respectively of the series sins , 2 sin 2a; . gin 3s sin 5a ; „ sinoa; . sin 7a; , sinto 1 2 3 5~ 6~~ ~T~ ~T" 432 FOURIER SERIES [Chap. By a well-known trigonometrical formula we have y = 2 cos^toj — o> 2 )x sin^ajj + <o 2 )x. This equation represents a phenomenon which we may think of as follows: we have a vibration with the circular frequency ^(o> x + w 2 ) and the period in/^ + o> 2 ). This vibration, how- ever, has not a constant amplitude; on the contrary, the " am- plitude " is given by the expression 2C0SKWJ — cd 2 )x, which varies with a longer period 477/(0^ — o) 2 ). This point of view is particularly useful and easy to interpret when the two circular frequencies ca 1 and co 2 are relatively large while their difference Fig. 6. — Beats (a^ — w 2 ) is small compared to them. Then the amplitude 2 cos^t^i — oj 2 )x of the vibration with period ^/(toj + <o 2 ) will vary only slowly compared with the period of vibration, and this change of amplitude will repeat itself periodically with the long period ^/(a^ — o> 2 ). These rhythmic changes of amplitude are called beats. Everyone is acquainted with this phenomenon in acoustics and perhaps also in wireless telegraphy. In wireless telegraphy the circular frequencies a^ and w 2 are as a rule far above those which the ear can detect, while the difference w i — '"a f au,s m * ne ran g e 0I audible notes. The beats then cause an audible note, while the original vibrations remain imper- ceptible to the ear. An example of beats is illustrated graphically in fig. 6. IX] COMPLEX NOTATION 433 2. Use of Complex Notation 1. General Remarks. The investigation of vibration phenomena and periodic func- tions gains in formal simplicity if we make use of complex num- bers, combining each pair of trigonometric functions cos cox and sincua; to form an expression of the type cosoxz; -f- i sincox = e"** (cf. Chap. VIII, § 7, p. 411). Here we must bear in mind that one equation between complex quantities is equivalent to two equa- tions between real quantities, and that our results must always be interpreted and made intelligible in the real domain. If we everywhere replace the trigonometric functions by exponential functions in accordance with the formulas 2 cos 6 = (3* + e-' e , 2t sin 6= <?* — er ie , we express sinusoidal vibrations in terms of the complex quan- tities e'"*, er™*, or ae t„c*-8 ae -t»(*-{) respectively, where a, to, and to£ are the real quantities ampli- tude, circular frequency, and phase displacement. The real vibrations are obtained from this complex expression simply by taking real and imaginary parts. The convenience of this mode of representation for many purposes is due to the fact that the derivatives of the real vibra- tions with respect to the time x are obtained by differentiating the complex exponential function just as if i were a real con- stant, as is expressed by the formula — a{ cosw (a; — £) + i sinw (x — £) } ax = aw{ — sina>(a; — £) + t cosoj(a; — £)} = iato{ coato(x — £) + i sinco (x — £)}, or — a^"- & = wto^*-®. dx 2. Application to the Study of Alternating Currents. We shall now illustrate these matters by means of an important example. Here we shall denote the independent variable, the time, by t instead of x. Jf (K798) 434 FOURIER SERIES [Chap. We consider an electric circuit with resistance B and inductance L, on which an external electromotive force (voltage) E is impressed. In the case of direct current E is constant, and the current / is given by Ohm's law, E = BI, If, however, we are dealing with alternating current, E is a function of the time t, and consequently so is J, and Ohm's law takes the form (cf. p. 182) E-L^ = BI. dt In the simplest case, to which we restrict ourselves here, the external electromotive force E is sinusoidal with circular frequency co. Now instead of taking this oscillation in the form a oosat or a sin<oi, we combine both possibilities formally in the complex form E = se'"' = e coscaf + *s sintat, where e( >0) represents the amplitude. We shall operate with this " com- plex voltage " as if i were a real parameter, and we thus obtain a complex current /. Then the significance of the relation thus found between the complex quantities E and / is that the current corresponding to an electromotive force e cos<ot is the real part of J, while the current correspond- ing to an electromotive force e sincoi is the imaginary part of /. The com- plex current can be calculated immediately if for I we write down an expression of the form / = ae"""' = a(cos&>« + i sin<o<); that is, if we make the assumption that / is also sinusoidal with circular frequency <o. The derivative of / is then given formally by the expression dl . . , dt = aa>(— sino>< + icoscot). By substituting these quantities in the generalized form of Ohm's law and dividing out the factor e""t we obtain the equation e — aiito = Ba, or B + uaL so that E = (B + iaL)I = WI. We may regard this last equation as Ohm's law for alternating currents in complex form, if we call the quantity W = B + iaL the complex resistance of the circuit. Ohm's law is then the same as for direct current: the current is equal to the voltage divided by the resistance. IX] COMPLEX NOTATION 435 If we write the complex resistance in the form W = we is = w cosS + iv> sin 8, where we obtain V(i2 2 + LW), tan 8 = ^, v> According to this formula the current has the same period (and circular frequency) as the voltage; the amplitude a of the current is connected with the amplitude e of the electromotive force by the equation e w and, in addition, there is a difference of phase between the current and the voltage. The current reaches its maximum, not at the same time as the voltage, but at a time 8/to later, and the same is of course true for the minimum. In electrical engineering the quantity w = Vf-R 2 + £ 2 <o 2 ) is frequently called the impedance or alternating current resistance of the circuit for the circular frequency oi; the phase displacement, usually stated in degrees, is called the lag. 3. Complex Representation of the Superposition of Sinusoidal Vibrations. So far the complex notation has been used to denote the com- bination of two sinusoidal vibrations. But a single vibration or a compound vibration of the type n S(x) = a + S (a„ cosv^c + b y sin vx) (for simplicity we have taken co = 1) can also be reduced to com- plex form by substituting cosra = - {£"* + e- ,v *) and sin vx = — . (e ,v * — e~ ivX ). The above expression then becomes an expression of the form S(x) = 2 a v e iv *, 43<S FOURIER SERIES [Chap. where the complex numbers a„ are connected with the real numbers a, a„ and 6„ by the equations a v = a v + a--v> a- = <* , b v = i(a„ — a_ v ). In order that the equation a v = a„ + a_„ shall formally include the case v = 0, we often put a = Oq = a /2. Conversely, we may regard any arbitrary expression of the form S a„e ,v * as a function representing the superposition of vibrations, written in complex form. In order that the result of this superposition may be real it is only necessary that a„ -f- a_„ should be real and a, — o_„ a pure imaginary; that is, a v and a_„ must be con- jugate complex numbers. 4. Deduction of a Trigonometric Formula. By using complex notation we obtain a very simple proof of a formula which we shall need later. This is the trigonometric summation formula <*»(«) = i + °os<* + cos2a + . . . + coswa = ! m ( w + i)« 2 sin J a which is true for all values of a except the values 0, ±2n, ±4tt, .... To prove this we replace the cosine function by its exponential ex- pression and thus bring the sum a„(a) into the form a„(a) = J 2 ««■». On the right we have a geometric progression with the common ratio q = e 1 '" =t= 1. Using the ordinary formula for the sum, we have „ (a) = I e - im 1 ~ g 2n+1 = 1 «^»-e6.+i» " 2 l- 3 2 1 - e« * On multiplying numerator and denominator by e~ ,a / 2 we obtain c a (a) = !H^ + i^ 2sin£<x as was stated. IX] COMPLEX NOTATION 437 Examples 1. Sketch the curves w = 2 for N = 3, 5, 6. n-i n n cos/ 2. Sketch the curves y = 2 — - for N = 3, 6, 8. 3. Evaluate the sum sin a + sin 2a + — + sinraa. 4. If s m (a) = C ° (a) + CT i( a > + • • • + a ™(* ) where c„(a) has the value c n ( a ) = J + cos a + cos 2a + . . . + cos»a, prove that (m+ l)a"i a fl » (a)== ^Tl sin- . a sm- 2 (The expression s m is called the " Fejer kernel ", and is of great importance in the more advanced study of Fourier series.) 5. Show that * f\ m(a )da=l, where s m {tx) is the Fejer kernel of Ex. 4. 3. Fourier Series The function n S(x) = a + S (a„ cos va + 6„ sinva;) resulting from the superposition of sinusoidal vibrations contains 2n + 1 arbitrary constants a, a„, 6„. The question now arises whether these constants can be so chosen that in the interval —it ^x^tt the sum S(x) shall approximate to a given function f(x), and if so, how they are to be found. More precisely, we inquire whether the given function /(a;) can be expanded in an infinite series 00 f(x) = a + 2 (a„ cos vx + b v sin vx). If we assume for the moment that this expansion of the function f(x) is actually possible and that the series converges uniformly in the interval — it ^ x 5S it, we readily obtain a simple relation between the function fix) and the coefficients a = £a , a„ and b r . (We shall soon see that the notation a = £a 43» FOURIER SERIES [Chap. is justified by its convenience.) We multiply the hypothetical expansion above by cos vx and integrate term by term, as is per- missible on account of the uniform convergence. In virtue of the orthogonality relations r + ". . 7 (0, if m =fe n, J sm mx smnxdx = \ J -n \tt, ifm=n4=0. /' sin mx cosnxdx = 0, - + *_ j_ (0, ifm4=n, if m= n, / cos mx cosnxdx = \ J -„ \1T, proved in Chap. IV, § 3 (p. 217), we at once obtain the formula 1 r +7r o„ = - / f(x) cos vxdx for the coefficients. Similarly, by multiplying the series by sin vx and integrating, we have 1 r + " 6„ = _ / f(x) sin vxdx. These formulae assign a definite sequence of coefficients a v and b v , usually called the Fourier coefficients, to every function f(x) which is defined and continuous in the interval — tt ^ x 5S it, or has only a finite number of jump discontinuities there. If the function f(x) is given, we can use these quantities a„, b v to form the Fourier partial sums n S„ (x) = %a -f- 2 {a v cos vx -f 6„ sin vx), and we may also formally write down the corresponding infinite " Fourier series ". Our problem is to distinguish simple classes of functions /(x) for which this Fourier series does actually con- verge and does represent the function. In order to formulate the result which we wish to prove, we introduce the following definition. A function f(x) is said to be sectionally smooth * in an interval if it is itself sectionally continuous f (that is, continuous in the interval except for a * Ger. stllckweise glatt. f Ger. gtiichweise stetig. IX] FOURIER SERIES 439 finite number of jump discontinuities) and if in addition its first derivative f'(x) is sectionally continuous. We shall imagine the function /(a;) originally defined in the interval — -n ^ x 5£ 77 to be periodically extended. At each point at which the function /(x) has a jump discon- tinuity we shall alter the function, if necessary, and shall assign to it the value which is the arithmetic mean of the left-hand limit and the right-hand limit oif(x); that is, we write /(*) = *(/(*-0)+/(a! + 0)), where f(x — 0) and f(x -f 0) are simply the limits of f(x) as x approaches x from the left and from the right respectively. This equation is obviously true for every point x at which f(x) is con- tinuous. Our goal is the following theorem: If the function f (x) is sectionally smooth and satisfies the above equation, then its Fourier series converges at every point x and represents the function.* Further, we shall prove the following theorem: In every closed interval in which the function f(x) (imagined periodically extended) is continuous as well as sectionally smooth, the Fourier series converges uniformly. Finally: If the function f (x) is sectionally smooth and has no dis- continuities, the Fourier series converges absolutely. The proofs of these theorems will be postponed until § 5 (p. 447). Here we merely wish to emphasize that the functions which can be expanded according to these theorems have a very high degree of arbitrariness; it is by no means necessary that the functions should be given by a single analytical expression. In the next section we shall display the extraordinary fertility of the Fourier expansion by discussing a number of examples. * It may be remarked incidentally that this theorem can be proved for more general classes of functions. The result formulated here, however, suffices for all applications. 44° FOURIER SERIES [Chap. 4. Examples of Fourier Series I. Preliminary Remarks. We shall assume that our functions f(x) have the period 2tt and are denned in the interval — v < x < v. Beyond this interval to the left and the right, they are to be extended periodically, as on p. 426. Before going into details we remark that if f{x) is an even function (cf. p. 20), then clearly f(x) suike is odd and f(x) cos vx is even, so that b„ = - / f(x) sin vxdx = 0; a v = / f{x) cos vxdx. TrJ—w TT"'0 We thus obtain a " cosine series ". If, on the other hand, the function f(x) is an odd function, then 1 r +1r 2 r" a„ = - / f(x) cos vxdx = 0; 6„ = - / f(x) sin vxdx. We therefore obtain a " sine series ".* 2. Expansion of the Functions \\>(x) — x and <p(x) = X 2 , id lunction x we nave o„ = tion by parts For the odd function x we have &„ = - /a; sinvsda;, and on integra- nt J It, — xoosvs -b„ = + - feosvxdx = (-1)" +1 -. v Jo v Hence for the periodic function <\i(x) which in the interval — 7t < x < it is equal to x (cf. fig. 7) we obtain the expansion i / > « fsinx sin2a; , sin3« , \ + (*)-2(— _ + _- + ...]. If we put x = 7t/2 we obtain Gregory's series, "=1-1 + 1- + ..., 4 3 5 with which we are already familiar (p. 319). The function i/(%) represented by this series is not a continuous function; on the contrary, it jumps by * Consequently, if the function f(x) is initially given only in the interval < x < 7t, then we can extend it in the interval — w < x < either as an odd function or as an even function, and correspondingly expand the function in the interval < x < jt either in a sine series or in a cosine series. IX] EXAMPLES 441 an amount 2n at the points x = kn, k = ±1, ±3. ±5 At these points of discontinuity, that is, at the points x = kit, k= ±1, ±3, ±5, . . . , each term of the series is zero, and hence the function itself is zero. Hence at the points of discontinuity the series represents the arithmetic mean of the left-hand and right-hand limits. If 5 is any fixed number between — 7c and it, and if we replace x in the above series by (x — £), we obtain the series «h* - a = 2 ( sin(a: - * > - sin2( * ~ g) + sin3(a: ~ %) - + . . .) \ 1 if o / 2 2 2 = — - sinE cosa: + - cosE sina; + - sin25 cos2a: 1 1 2 2 2 2 — cos2E sin2a: — - sin3E cos3a: + - cos3E sin3z + . . . . 2 3 3 Fig. 7 This may also be written in the form of a Fourier series with coefficients 0, a„=2 (-1)" • sinwJ;, b n = 2 = 9.(-l)" COS 71 5, which tend to zero as n increases; this series represents a function having the discontinuities described above at the points a;=5±7r, i=E ± 3n For the even function <p(x) = a? we find, on integrating by parts twice, that a„=- C x* cosva:cte = (— I)" - (v > 0), TZJ V 2 27t 2 so that we obtain the expansion cp (x) = — y\ i 3 cosa; cos 2x cos 3a; ~2?~ ~W (cosa; - + > By differentiating this series term by term and dividing by 2 we formally recover the series for f ]i{x) = x. 15« ;i798) 442 FOURIER SERIES [Chap. 3. Expansion of the Function .r cos at. For this odd function we have 2 /•" a„ = 0, b r = - / a; cosa; sinvaiix. ■a J By using the formula •'o aia i ixdx= (— l)e+i - (n = 1, 2, . . .) found in the previous sub-section, we evaluate 6„: b y = - I xcosxainvxdx= - I a;(sin(v + l) 31 + sin(v — l)x)dx tzJq t:J (— 1)"+ 2 (—1)- 2v 6i=-; Fig. 8 We therefore obtain the series 1 x cosa; = — - sina + 2 X 2, (-l) P v . 2 ,_, v* - 1 and if -we add the series found on p. 440 this becomes smvas, ... . 3 . , / sin2a; xll -f cosa;) = - sina; +21 — v ' 2 \l. sin3a; sin4a; , 2.3 2.3.4 3.4.5 )• When the function which is equal to x cos a: in the interval — -k < x < n is extended periodically beyond this interval, the same discontinuities (of. fig. 8) occur as are exhibited by the function <\>(x) considered in No. 2. On the other hand, if the function x(l + cos a:) is periodically extended it remains continuous at the end-points of the intervals, and in fact its deri- vative also remains continuous, since the discontinuities are eliminated by the factor 1 -f- cos a;, which together with its derivative vanishes at the end-points. IX] EXAMPLES ♦43 i. The Function f{ x) = \x\. This function is even; consequently 6, = and 2 o„ = - / x cos \>x dx, 77 Jn and on integration by parts we readily obtain f Jo x cos va; dx = - x sin va; v * l r v -4 sinvxix f 0, if v is even and 4= 0, = \ 2 :, if v is odd. Consequently ... 77 4 / , cos3cc , cos5a; , \ f{x) = 2 ~ * l 008 * + T + ~sr + • • •> If we put x = 0, we obtain the remarkable formula <L 3 =1 + I + A+.... 8 T 3 ! 5' T 5. Example. The function defined by the equations (—1, for — 7t < x < 0, 0, for x = 0, + 1, forO < x < it, m ^_^^^ y . .i -x\ -f Jl\ 2jJI 5 Fig. 9 as indicated in fig. 9, is an odd function. Hence a„ = and / if v is even, b = - I sin vxdx= \ 4 "Jo — if v is odd, so that the Fourier series for the function is ,, , 4 (smx . sin3a; . \ /{x)= __[^_ + _+...j. For a; = ", in particular, this again yields Gregory's series. This series can be formally derived from that for | x | by term-by-term differentiation. H4 • FOURIER SERIES [Chap. 6. The Function f(x) = | sin x \. The even function f(x) = | siux | can be expanded in a cosine series, the coefficients a y being given by the following calculation: 7t /•" I /* - o„ = / srnxcosvxaJa; = - / {sin(v + \)x — sin(v — \)x}dx 2 Jo ^ Jq ' if v is odd, —2 (.V s — 1 We thus obtain if v is even. i - i 2 4 ? cos2ixa! k n^x 4(x a — 1 7. Expansion of the Function cos y.x. Resolution of the Cotan- gent into Partial Fractions. The Infinite Product for the Sine. Let /(x) = cos [jue for — iz < x < tc, where y. is not an integer. Since f(x) is even we again obtain 6„ = 0, while 7T z" 17 1 /" lr - a„ = / cosjjue cosvxcfa: = - / {cos(n + v)x + cos((ji — v)x}dx 2 Jo "Jo _ 1 j sin(ti, + v)tc , 3in((i,— v)rc \ 2 \ {x + v [i — v / |X(-1)" . (X 2 — V 2 We thus have 2u.sintATC / 1 cossc , cos2a: . \ «,.,«=__ ^_____ + ____ + ...j. This function remains continuous at the points as = ±tc. If we put x = jc, divide both sides of the equation by sin [i.7t, and then write x instead of \x, we obtain the equation 2x( 1 , 1 , 1 , \ This is the so-called resolution of the cotangent into partial fractions, a very important formula frequently discussed in analysis. We now write this series in the form cot not ■KX KT nil'-** 2»-s» / If x lies in an interval iS x :S <? < 1, the n-th term on the right is less in 2 1 absolute value than Hence the series converges uniformly in it n a — « 8 IXJ EXAMPLES 445 this interval, and can be integrated term by term. We thus obtain Jr* i 1\ sinra: ,. , sinrai sinra; 1 (cotTrt J <« = log v^ lo 8 — — = lo S — — on the left, and ^0-5) +k8 ( i -*) + ---5 n .ii ,Dg ( l -5) on the right, multiplying both sides by n. If we pass from the logarithm to the exponential function we have RimTX lim 2 k>g(l -«•/•'•> S log (1 -**/■.») warm = ^^ „_, = hm e — i TZX = lim Hence ■h"w="w(i-y( i -J)( 1 -^)--- • We have thus obtained the famous expression for the sine as an^irifinite product.* Prom this, by putting x = -, we obtain Wallis's product 7t_ » 2v 2v ^2 2 4 4 2 ~ .i^v - 1 "2v+l — l'3'3"6'"" as on p. 224. 8. Further Examples. By brief calculations similar to the preceding we obtain the following further examples of expansions in series. The function f(x) which is defined by the equation f(x) = sin(ix for — n < x < 7c can be expanded in the series 2sin(i7t/ sins 2sin2a 3sin3z \ /(*) = sinp; = — {jpZTp ~ j^Y* + il^T 2 +•••)■ If we put x=- and use the relation sin(irc = 2 sin— cos—, this gives us the resolution of the secant into partial fractions, that is, of the function ; this expansion is COStl - 2 * • (-l)"(2v-l) jr sec nx = = 4 i = co8tux %_i4a?-(2v- 1)»' where for (J./2 we have written r. * This formula is particularly interesting because it shows directly that the function sin irx vanishes at the points x = 0, ±1, ±2 In this respect it corresponds to the factorization of a polynomial when its zeros are known. are 446 FOURIER SERIES [Chap. The series for the hyperbolic functions cosh \xx and sinh iix(—n<x<n) cosh ( x :B =^sinh ! x TC /'J___^^, + ^^__£?^ + _ ^ t \2(i a [i 2 + 1* ^ + 2 2 ^+32 ' " 7' sinh^=2 sinh(Jt7r ('_iiE^_i sin l^ + i sin ^_ + \ EXAMPLES 1. Find the Fourier expansions for the functions which are periodic with period 2n and which in— iz <x^tz axe defined by the formulae (a) e ax . (6) (a* — ti 2 ) 2 . (c) sinox;(l + cos*). (d) f(x)=l(a^x^b), f(x) = 0(—n:<x<a), f(x) = 0(b <x^ jt). 2. The function /(«) is periodic with period 1, and inOgi<l it is given by /(*) = <. Prove that /(«)--- - S !E^f. 2 -re n =i » 3. The polynomials S n (t) (Bernoulli polynomials) are defined by the relations: (a) £,(t) = t - J; (6) £„'(*) = raB,,^); (c) /"^-(i) * = 0. Find £,(<), B s (t), B t (t). J (Note.— The numbers B n (0) are rational, and are, in fact, the same as Bernoulli's numbers B n ; cf. pp. 422, 423.) 4. Verify the Fourier expansions for the Bernoulli polynomials: B l{ t) = _i { £ ****** B s (t) = A ( S «n2^n tc U-i raj 2tc 3 l„_i n* f BS) = \ { £ S?5^}. B 4 (|, = - 8 { £ °a^\ 5. Prove that S i=-, S I = H-. »-i re 2 6 „-! ra 4 90 6. Prove that I_I+I-I + _...=V I s 3 3 5 3 7 3 32 7. Prove that (a) 1 + 1 + — +l+...= 7 ^. 32 ^ 5 2 72 8 /7.X 1 ^ p 1 1 1 I* ( 6 ) 1 -22 + 32-ii+--'=i2- ( 6 ) 1-1+1-1+.. . = Z^. 2* 3 4 4 4 720 8. Obtain the infinite product for the cosine from the relation sin2rca; costm;= — : . 2 sin7ce IX] CONVERGENCE OF FOURIER SERIES 447 5. The Convergence of Fourier Series We now proceed to establish rigorously the theorems which were stated in § 3 (p. 439) and illustrated in § 4 (p. 440). 1. The Convergence of the Fourier Series of a Sectionally Smooth Function. We first recall that if /(«) is any function which is denned and sectionally continuous (that is, continuous except for a finite number of jump discontinuities at most) in the interval — it ^ x 5S 77, we can form the Fourier coefficients of f(x) accord- ing to the formulas 1 r + * 1 r + " a v = - / f(t) cos vtdt, b v = - / f(t) sinvtdt, and we can formally write down the series CO \a -f 2 (a y cos vx + b„ sin vx). This series is called the Fourier series corresponding to f(x), irrespective of whether it converges or not. We are now about to find the conditions which must be imposed on f(x) in order to ensure that the Fourier series corresponding to f(x) does con- verge and does represent f(x). We assume that f(x) is extended periodically beyond the interval — it <.x^tt. The theorem which we shall now prove is as follows: If the function f (x) is sectionally smooth * and at each point of discontinuity (x) satisfies the equation f (x) = £ { f (x — 0) -f f (x + 0) }, then the Fourier series corresponding to f (x) converges at every point and represents the function. To prove this theorem we consider the partial sums n S n {x) = |a + 2 (a„ cos vx + 6„sin vx). If for the coefficients we substitute the integral expressions found above and then interchange the order of integration and summation we obtain S n (x) = — I fit) ; + 2 (cos vt cos vx -f sin vt %m.vx) [ dt, * That is, f(x) is sectionally continuous and its derivative /'(a;) is sectionally continuous also. 448 FOURIER SERIES [Chap. or, using the addition theorem for the cosine, SJx) = - f + f(t) \l + Xcoav(t- x)\ dt. If we now apply the summation formula obtained on p. 436, this becomes 2tt j -„ smf(,J — x) Finally, making the transformation t = (t — x) and noting the periodicity of the integrand, we obtain sjx) =z- f( x + T ) — • T dr - XttJ—v sinfi Starting with this form of the partial sum S n (x) we can prove, by means of the following lemma, that S n (x) tends to f(x). Lemma. If the function s (x) is sectionally continuous in the interval a :£ x ^ b, then the integral <J> 1= f s(t)smXtdt tends to as A increases. In the proof we may take s(x) to be continuous in the whole interval, since otherwise we need only carry out the argument for each sub-interval in which s(x) is continuous. As in the similar argument on p. 418 et sea., we notice that if A is positive the function sin M is alternately positive and negative in successive intervals of length 77/ A. For large values of A the contributions to the integral from adjacent intervals almost cancel one another, since on account of continuity the values of s(x) in two such adjacent intervals differ from one another only slightly. We make use of this circumstance by transforming the integral I by the substitution t= r -\- h, where h = 7r/A; then sinAf= — sin At, and we obtain J~b-h s(r -f- h) sin At dr. ' s\ J a-h If we again write the letter t instead of t and then add the two IX] CONVERGENCE OF FOURIER SERIES 449 expressions for I, we have 2/ = — fs (t + h) sin tidt + f ~{s(t) — s(t + h)} sin Xldt J a-h J a r b + s{t)smXtdt. J b-h If M is an upper bound for the absolute value of s(x), that is, if for all values of a; in the interval under consideration | s(x) | ^ M, then from this expression for I the inequality 2 I 1 1 ^ 2Mh + f ]s(t) — s(t + h) I dt follows at once. Now let e be any positive number; if we choose A so large that in the whole interval a ^ t 5S b — h the expres- sion I s(t) — s(t + h) I remains less than e/(6 — a) and also Mh = < -, then 1 1 1 < e; and consequently,* since e can A 2 be chosen as small as we please, lim 1=0. Besides this lemma we need the integration formula Jo 2sin£t 2 which is true for every positive integer n. This we readily estab- lish by using our summation formula for the cosine, since r sin(m + A) t ,. f w .. . " ,. ,, w / — i — z/ ■ dt = / (A + 2 cos vt) dt = -. •A) 2sin^ Jo V2 ^ x ' 2 Proo/ of the Main Theorem. — By means of the lemma it is easy to prove our main theorem, i.e. to prove the formula lim S n (x) = lim 1 f + 'f( x + t) ain(n . + * ) * dt = /(*). M— >-oo Z7T 2 sin^f * If we assume that s(x), besides being continuous, has a sectionally con- tinuous derivative «'(»), the proof of this lemma follows simply on integration by parts. For js(t) sinAi* - i js(o) cosAo - «(6) cosA6+ J «'(<) oosAtdf j. Here we see at once that as A increases the right-hand side tends to zero. 450 FOURIER SERIES [Chap. We begin by subdividing the interval of integration at the origin. For fixed values of x the function * 2 sin^ is sectionally continuous in the interval j£ t ;g it. For this is obvious when < t ^ 77, while the continuity when t = follows from the assumed existence of the right-hand derivative lim /(*+*)-/(»+<>) _ lim /(»+*)-/(»+ 0) # 2aJnj< i->o, «>o < i-*-+o 2sin£< ' £ -lim /(* + <)-/(*+<» «-»-+o 2 sin^f Hence, as A = w -f \ increases, the integral 1 r" - / s(t) sin M it 77 ■'0 1 f ti , ^sinAi^ 1 r*,, , „ 4 sinAi , = ^- / /(*+*) -^— r- dt — — I fix 4- — — eft tends to zero. Since, however, the factor f(x 4- 0) can be taken out of the second integral on the right, and since for A = n 4- £ the integral f" sinAi , . . J 77 . ,.,,., _/ . , «f is equal to -, we immediately obtam the equation f lun 9 - f'f(x +t)^dt= I f(x + 0). x-^„277- / o sin# 2 In the same way we obtain lim * /° /(* + <) ^S& =£/(*- 0), x->oo2tt-'-» smff 2 for the interval —rr^Lt^O, and by addition x-*-» 277 J -" sinji * For this notation see p. 439. fBy putting x — 0,f(t) = (sin i«)/' in this equation, and then replacing i by k/A, we obtain the important relation (of. pp. 251-253) x-+«J u 2 IX] CONVERGENCE OF FOURIER SERIES 45 * 2. Further Investigation of the Convergence. In the neighbourhood of those points where the function f(x) is discontinuous the Fourier series does not converge uniformly; for according to Chap. VIII, § 4 (p. 393), a uniformly convergent series of continuous functions possesses a continuous sum. Never- theless, we have the following important theorem: If a sectionally smooth periodic function has no discontinuities, its Fourier series converges absolutely and uniformly. The conver- gence of the Fourier series for any sectionally smooth function what- ever is uniform in every closed interval which contains no point of discontinuity of the function. In order to prove this theorem we start from a fundamental inequality satisfied by the Fourier coefficients of any function f(x) which is sectionally continuous (note that f(x) is not assumed to be sectionally smooth). This so-called Bessel's inequality states that for all values of n I V 4- £ (a* + b*) 2£ - J + \f{x)fdx. The proof follows from the fact that the expression \ f( x ) ~ h a o — ^ (a v cos vx + b sin vx) \ dx — <- „=.i J is always positive or zero. If we evaluate the integral by expanding the bracket under the integral sign and recalling the orthogonality relations and the definitions of the Fourier coefficients, we at once obtain Bessel's inequality in the form f^{f{x)Ydx- *\L* + i(a* + b y *)} ^ 0. In addition to Bessel's inequality we make use of Schwarz's inequality (p. 13): if u^, w 2 , . . . , u n and v±, u 2 , . . . , v n are arbitrary real numbers, it is always true that (Jes- 9 -)' l 2 w„ 2 . 2 v*, v=l v=l the sign of equality occurring only when the sequence u is pro- portional to the sequence v. We now assume that the periodic function f(x) is sectionally smooth, and also continuous. The derivative g(x)=f'(x) is 452 FOURIER SERIES [Chap. sectionally continuous, and we easily show that c„ and d r , the Fourier coefficients of g(x), satisfy the relations c = 0, for on integration by parts we have 1 r + " c„ = — / g(x) cos vx dx = - f( x ) cos vX \ + - / /(*) sin vx <£b = v&„, 77 I — it 77 ./— it a similar proof holding for the other statements. Bessel's inequality applied to the function g(x) therefore gives us S *(<** + K 2 ) = S (c, 2 + <i 2 ) ^ - f +T {g(x)Ydx. v=i v=i ir J -* If for brevity we denote the right-hand side of this inequality by M 2 , and apply Schwarz's inequality, we find that when m> n m m S | a„ cos vx -|- fc„ sin vx | ig £ \/(a r 2 + fe„ 2 ) since -\/( a * 2 + K 2 ) * a tne amplitude of the periodic function a„ cosvx + 6„ sinvx. 00 J But owing to the convergence of S — the right-hand side, v=.l v a which is independent of x, can be made as small as we please by choosing n and m large enough, which proves the absolute and uniform convergence of the series.* In order to prove the above theorem for sectionally smooth functions which are discontinuous, we first consider a special function ip{x) of this type. * The same considerations show incidentally that for periodic functions with continuous derivatives of the (ft — l)-th order and derivatives of the n A-th order which are at least sectionally continuous, the sum Sv^a,,* + 6„*) remains below a fixed bound. This gives us a definite statement about the order to which the Fourier coefficients vanish. For such a function the Fourier series of the derivatives up to the order (ft - 1) converge absolutely and uniformly. IX] CONVERGENCE OF FOURIER SERIES 453 In the interval — it < x < 77 we define i/i(x) as equal to x; out- side this interval t/i(x) is extended periodically. According to p. 440 its Fourier series is „ /sina; sin2a; sin 3a; \ 2 V^ 2~ + "l — + 'T This series cannot be uniformly convergent, for its sum is the discontinuous function ifi(x). We shall show, however, that the convergence is uniform in every interval — I ^x^l for which < I < -n. The proof is based on a special artifice.* We observe that in the interval — I < x <L I the function cos - is never less than the ~ ~l 2 positive quantity cos - = k. If we multiply the absolute value of the difference between the »»-th and ra-th partial sums of the above series (m > n), that is, the expression \S m (x)-S n (x)\ = 2 sin(w+l)a; sin(w+2)a; , sin mx\ — -j . . . ~r n -f 1 w-(- 2 m \ by the function cos -, then in accordance with the well-known trigonometric formula 2 sin to cosv = sin(w -\- v) + sin(w — v), we obtain the absolute value of the expression x isin(w + l)a; sin(w -\- 2)x sinma;) £i COS — { — - r~ — • • • *t~ — } 2 \ n+ 1 n+ 2 ~ mx l __ sin(w-f |)a; sin(w+|)a; , sin(m-f|)a; m+1 w+2 °" — m sin (n -f- 1) a; _ sin (w -f- 1) a; sin(w+f)a; , n+ 1 m+2 w+3 ~~ +•••• * We are led to this artifice naturally by observing that the function 2y cosy, when extended periodically beyond the interval —~^y^~, remains con- 2 2 tinuous, so that according to the first part of the theorem its Fourier series must converge uniformly and must represent the function. This series, however, is obtained if we multiply the Fourier series for 2y by cosy. If we now put y — xj2, this multiplication leads to the steps in the text. 454 FOURIER SERIES [Chap. If we combine the terms on the right which have the same numerators, we obtain the expression sin(n + |)a; sin(m + \)x n + 1 — m , sin(w+|)a; sin(w+f)a; sin (to— i)a; (n+l)(n+2) (n + 2) (n + 3) + '"~ t ~ (to-I)to ' and since cos - ^> k and | sinw | ^ 1, we obtain the estimate \S m {x)-S n (x)\ /cLw+1 to (w -f 1) (ra + 2) (to — 1)toJ But the expression on the right does not depend on x, and in 00 1 virtue of the convergence of the series S it can be made v =iv{v-\- 1) as small as we please by choosing n and to large enough. This implies the uniform convergence of the Fourier series, which we asserted. Now that we have obtained the expansion for a particular discontinuous function, we can (cf. p. 441) transfer the discon- tinuity to any arbitrary point in the interval by translation of the curve or of the co-ordinate system. In fact, the function ^(r -g) — o/ 8 " 1 ^— ^) _ sm2(x — g) | sin3(a;-g) | \ is continuous except at the points (2k + l)n + i, where k is an integer. On passing these points, however, the function jumps by an amount — 2-rr, from the value 77 to the value — tt, while at these points themselves the value of the function is zero. If now f(x) is any sectionally smooth function which in the interval — 77 ^ x ;S 77 is discontinuous only at the points £i> $2' • • • » £™> an( i ^ on passing these points from left to right the function jumps by the amounts S 1; 8 2 , . . . , S m respectively, then the function /(*)+ h 0(* + w - &) + h ,/,(*+ tt- fo + . . . In Ztt IX] CONVERGENCE OF FOURIER SERIES 455 will be continuous and sectionally smooth, and hence by the previous proof can be expanded in a uniformly convergent Fourier series. We now obtain the Fourier series of the function /(x) by adding the finite number of Fourier series corresponding to the functions —_!,/, (a; +7r—^), .... — ^t/,{x+ir— g m ) term by term. Our theorem is thus proved. This result is quite adequate for most mathematical investi- gations and for applications. We would, however, point out that the investigation of Fourier series has been pushed much further. The conditions for expansion in Fourier series which we have found here are sufficient, but by no means necessary. Func- tions with far fewer continuity properties than those discussed here can be represented by Fourier series. There is an extensive literature devoted to these questions and to the general problem of the expansibility of a function in a Fourier series. As a remarkable result of such investigations we mention the fact that there are continuous functions whose Fourier series do not converge in any interval, no matter how small. Such a result does not in any way impugn the usefulness of Fourier series; on the contrary, it must be regarded as evidence that the con- cept of a continuous function involves fairly complicated possi- bilities, as has already been shown by the example of continuous functions which nowhere have a derivative. Appendix to Chapter IX Integration of Fourier Series One of the remarkable properties of Fourier series is their term-by-term integrability. In general, a series can be integrated term by term if it is uniformly convergent; otherwise term-by- term integration may lead to false results. In contrast with this, for Fourier series, we have the theorem: If f(x) is sectionally continuous in — ti 5S x ^ -n, and if the series £a + 2(a„ cosx-(- b„ sinx) is the Fourier series corresponding to f(x), then this series can be integrated term by term between any 456 FOURIER SERIES [Chap. IX two limits £ and x lying in the interval — it :£ x £S tt; in symbols, J f(x)dx=J \a dx-\-~Ll\ a v cosvxdx-\- 1 b„sm.vxdx 1. Moreover, for every fixed value of £ the series on the right converges uniformly in x. The remarkable feature of this theorem is that not only do we not require that the Fourier series for f(x) shall be uniformly convergent, but we need not even assume that it converges at all. To prove this, let the function F(x) be denned by the equation {f(x) — $a }dx. This function is sectionally smooth, and by the definition of a we have F(tt) = F(—tt) = 0, so that F(x) can be extended periodically and continuously. The Fourier 00 series \A + S {A v cosvx + £„ sinvx) of the function F(x) there- fore converges uniformly to F(x). We now investigate the co- efficients A v and B v . By integration by parts as on p. 449, we find that, for v > 0, A p = — bjv and B„ = ajv. Hence for any values £ and x in the interval — tt ^ x ^ tt, we have 00 F(x) — F(£) = S {A v (cos vx — cos v£) + B v (sin vx — sin v£) } v = l = Z j — " (sin vx — sin v£) —-- (cos vx — cos v|) }, converging uniformly in x. If we replace F(x) by its definition, this becomes / f{x)dx — %a fdx = S (a r I cos vxdx-\- 6„ / sin vxdx), which was to be proved. It is easy to see that if f(x) is periodic and sectionally con- tinuous the term-by-term integration can be performed over any interval whatever. CHAPTER X A Sketch of the Theory of Functions of Several Variables Up to this point we have concerned ourselves exclusively with functions of a single independent variable. We must now go on to consider functions of several independent variables. Even the applications of the calculus force us to take this step. In almost all the relationships which occur in nature, in fact, the functions in question do not depend on a single independent variable; on the contrary, the dependent variable is usually determined by two, three, or more independent variables. Thus, for example, the volume of an ideal gas is a function of a single variable, the pressure, if we keep the temperature constant, but not otherwise. As a rule the temperature also varies, and the volume depends upon a pair of values, namely, the value of the pressure and that of the temperature; it is therefore a function of two independent variables. From the point of view of pure mathematics also, the need for detailed study of functions of several independent variables is urgent. Here we shall be able to take advantage of what we have previously learned, so that in many cases we have only to make simple extensions of our arguments. It is usually sufficient to consider the case of only two inde- pendent variables x and y, so long as no essentially new con- siderations are required for an extension to functions of three or more variables. In order to keep our statements and notation simple, therefore, we shall consider only two independent variables as a rule. A systematic presentation of the differential and integral calculus for functions of several variables is impossible within the compass of this volume, but will be given in Vol. II 458 FUNCTIONS OF SEVERAL VARIABLES [Chap. of this treatise. All that can be done here is to give the reader a preliminary view of some of the most important new concepts and operations. We shall frequently rely on intuitive plausi- bility, the full proofs being developed subsequently in Vol. II. 1. The Concept op Function in the Case op Several Variables 1. Functions and their Ranges of Definition. Equations of the form u = x 2 + y 2 , u= x — y, u= xy, or u = -\/(l — x 2 — y z ) assign a functional value u to each pair of values (x, y). In the first three of our examples this correspondence holds for every system of values (x, y), while in the last case the correspondence has a meaning only for those pairs of values (a;, y) for which the inequality as 2 + y 2 :gj 1 is true. In these cases we say that u is a function of the independent variables x and y. This expression we use in general whenever some law assigns a value of u as dependent variable, corre- sponding to each pair of values (x, y) belonging to a certain specified set. The relation between x, y and u may be stated in terms of a "functional equation", as above, or by means of a ver- bal description such as "u is the area of the rectangle with sides x and y ", or it may follow from physical observations, as, for in- stance, in the case of the magnetic declination for different lati- tudes and longitudes. The essential thing is that a correspondence exists. Similarly, u is said to be a function of the three inde- pendent variables x, y, z if for each triad of values (x, y, z) of a certain set there exists a corresponding value of u given by some definite law; and similarly for the general case of functions of n independent variables %, x 2 , . . . , x n . The set of values which the pair (x, y) can assume is called the range of definition of the function u =f(x, y). For the pur- poses of this chapter we shall restrict our attention to the sim- plest types of range of definition. We shall consider that (a;, y) is limited either to a so-called rectangular region (domain) a iS*x<Lb, c fS y <>d, X] FUNCTIONS OF SEVERAL VARIABLES 459 or else to a circle determined by an inequality of the form (a; — af + (y — b) 2 S «*• In the case of functions of three variables x, y, z we shall again consider only rectangular regions a 5£ a; <1 6, c^y^d, e 5S z S/ and spherical regions (a; — a) 2 + (y — bf + (z — cf ^ r 2 . In dealing with more than three independent variables geometri- cal intuition fails us, but it is often convenient to extend geo- metrical phraseology to this case also. Thus for functions of n variables x 1> . . . , x„ we shall consider regions <h ^ *\ ^ b v a 2 <; x % <; 6 2 , . . . , a„ ^ x„ <: b n and also regions (x 1 -a 1 ) 2 +(x 2 -a 2 ) 2 + ... + (x n -a n ) 2 ^r\ which we call rectangular regions and spherical regions respec- tively. 2. The Simplest Types of Functions. Just as in the case of functions of one variable, the simplest functions are the rational integral Junctions or polynomials. The most general polynomial of the first degree (linear function) is of the form u = ax -f- by + c, where a, b, and c are constants. The general polynomial of the second degree has the form u s= ax 2 + bxy -\- cy 2 -\- dx -\- ey +/. The general polynomial is a sum of terms of the form a mn x m y n , where the quantities a mn are arbitrary constants. Rational fractional functions are quotients of polynomials; to this class belongs e.g. the linear fractional function u= ga; + ty+o a'x + b'y + c'* 4<5o FUNCTIONS OF SEVERAL VARIABLES [Chap. By extraction of roots we pass from the rational functions to certain algebraic functions,* e.g. In the construction of more complicated functions of several variables we almost always fall back on the well-known functions of one variable; f e.g. w = sin (xy) or u = log (y 2 + cos %x). 3. Geometrical Representation of Functions. Just as we represent functions of one variable by means of curves, we seek to represent functions of two variables geometri- cally by means of surfaces; henceforward we shall consider only those functions which can actually be represented in this way. We achieve this representation very simply by considering a rectangular co-ordinate system in space, with co-ordinates x, y, and u, and marking off above each point (x, y) of the range (R) of definition of the function the point P with the third co-ordinate u — f(x, y). As the point (x, y) ranges over the region R the point P describes a surface in space. This surface we take as the geometrical representation of the function. Conversely, in analytical geometry surfaces in space are represented by functions of two variables, so that between such surfaces and functions of two variables there is a reciprocal relation. For example, to the function u = V(l - x 2 — y») there corresponds the hemisphere lying above the x, y plane, with unit radius and centre at the origin. To the function u = x 2 -f- y 3 there corre- sponds a so-called paraboloid of revolution, obtained by rotating the para- bola u = x 3 about the M-axis (fig. 1). To the functions u = x 3 — y 3 and « = xy there correspond hyperbolic paraboloids (fig. 2). The linear function u = ax + by + c has for its graph a plane in space, f * For an accurate definition of the term " algebraic function " see p. 485. f Cf. also the section on compound functions (p. 472). J If in the function « — f(x, y) one of the independent variables, say y, does not occur, so that u depends on x only, say « = 9(x), the function is represented in xyu-sp&ce by a cylindrical surface obtained by erecting the perpendiculars to the uz-plane at the points of the curve u — g(x). X] FUNCTIONS OF SEVERAL VARIABLES 461 This representation by means of rectangular co-ordinates has, however, two disadvantages. Firstly, intuition fails us whenever we have to deal with three or more independent vari- ables. Secondly, even in the case of two independent variables it is often more convenient to confine the discussion to the xy- plane alone, since in the plane we can sketch and make geometrical constructions without difficulty. From this point of view another geometrical representation of the function, by means of contow lines, is to be preferred. In the ay-plane we take all the points for which u=f(x, y) has a constant value, say u—k. These Fig. 1. — u = *' + y* Fig. a. — u= x* — y* points will usually lie on a curve or curves, the so-called contour line for the given constant value of the function. We can also obtain these curves by cutting the surface u=J(x, y) by the plane u = k parallel to the ay-plane and projecting the curves of intersection perpendicularly on to the tcy-plane. The system of these contour lines, marked with the corresponding values \, k 2 , . . . of the height k, gives us a representation of the function. As a rule k is assigned values in arithmetic progression, say k = vh, where v = 1, 2, . . . . The distance between the contour lines then gives us a measure of the steepness of the surface m = /(•£, y); for between every two neighbouring lines the value of the function changes by the same amount. Where the contour lines are close together the function rises or falls steeply; where the lines are far apart the surface is flattish. This is the principle 462 FUNCTIONS OF SEVERAL VARIABLES [Chap. on which contour maps such as those of the Ordnance Survey and the U.S. Geological Survey are constructed. In this method the linear function u = ax-\-by + cJB represented by a system of parallel straight lines ax + by + c = k. The function u = a; 2 + y 2 is represented by a system of concentric circles (cf. fig. 3). The function u = x 2 — y 2 , whose surface has a saddle point at the origin (fig. 2) is re- presented by the system of hyperbolas shown in fig. 4. The method of representing the function u = f(x, y) by con- tour lines has the advantage of being capable of extension to functions of three independent variables also. Instead of the Fig. 3. — Contour lines of u *= x* + y* Fig. 4. — Contour lines of u «■ x* — y* contour lines we then have the level surfaces f(x, y, z) = k, where k is a constant to which we can assign any suitable sequence of values. For example, the level surfaces for the function u = x 2 + y % -f- z 2 are concentric spheres about the origin of the co-ordinate system. Example 1. For each of the following functions sketch the contour lines corre- sponding to z = — 2, — 1, 0, 1, 2, 3: (a) z = x*y. (d) z = y*. (6) t = x* + y 2 - 1. / ^ x (c) z = z 2 - y\ (e) * V \ & + y 2 / X] CONTINUITY 463 2. Continuity 1. Definition. As in the case of functions of one variable, the basic require- ment that functions should be capable of being represented geometrically leads to the analytic condition of continuity. Here again the concept of continuity is given by the following de- finition: a function u = f(x, y), defined in a region R, is said to be continuous at a point ($, 77) of R if for all points (x, y) near (£, 7}) the value of the function f(x, y) differs but little from i(£, ■??), the difference being arbitrarily small if only (x, y) is near enough to (€, v). More precisely: the function i(x, y), defined in the region R, is continuous at the point (£, ij) o/R, provided that for every posi- tive number e it is possible to find a positive distance 8 = 8(e) (in general depending on e and tending to with e) such that for all points of the region whose distance from (£, tj) is less than 8 (that is, for which the inequality (^-a 2 +(y-^) 2 ^s 2 holds) the relation \f(*,y)-Mr,)\^ e is satisfied. Or, in other words, the relation \f(i+h,r,+ k)-f(i,r,)\^ e is to hold for all pairs of values (h, k) such that F+Pg § 2 and (i+ h,7] + k) belongs to the region R. If a function is continuous at every point of a region R, we say that it is continuous in R. In the definition of continuity we can replace the distance condition A 2 + Jfc* ^ 8 2 by the following equivalent condition: To every e > there shall correspond two positive numbers 8 t and S 2 such that l/(f +*»!»+*)-/«, ij)|^e whenever | h | ^ S 2 and | k | ^ 8 2 . The two conditions are equivalent. For if the original con- dition is fulfilled, so is the second if we take 8 X = 8 2 = 8j-y/2; 464 FUNCTIONS OF SEVERAL VARIABLES [Chap. and conversely, if the second condition is fulfilled so is the first, if for 8 we take the smaller of the two numbers S x and 8 2 . The following facts are almost obvious: The sum, difference, and product of continuous functions are also continuous. The quotient of continuous functions is continuous except where the denominator vanishes. Continuous functions of continuous functions are themselves continuous (cf . the note on pp. 473-474). In particular, all polynomials are continuous, and all rational fractional functions are also continuous except where the denominator vanishes* 2. Examples of Discontinuities. In the case of functions of one variable we met with three kinds of discontinuities: infinite discontinuities, jump discon- tinuities, and discontinuities at which no limit is approached from one side or both. With functions of two or more variables no such simple classification is possible. In particular, the situa- tion is made more complicated by the fact that discontinuities may occur not merely at isolated points but also along whole curves. Thus for the function u = the line x = y is a line of infinite x — y discontinuity. As we approach the line from one side or the other the values of u increase numerically beyond all bounds through positive or through negative values. The function u = has the same line of (x — yf discontinuity but tends to +00 as we approach the line from either side. The function u = has the single point of discontinuity x = 0, <* a + y 2 1 v = 0. The function u = sin — , tends to no limit as we approach " V(a; 2 + y 2 ) the origin; the surface which it represents is obtained by rotating the graph of the function u = sin - about the u-axis. Another instructive example of a discontinuous function is given by the rational function u = — ^— . In the first instance the function is & + y a * Another obvious fact, which, however, is worth stating, is as follows: •/ a function f(x, y) is continuous in a region R and is different from zero at an interior point P of the region, it is possible to mark off about P a neighbourhood, say a circle, belonging entirely to ~R, in which f(x, y) is nowhere equal to zero. For if the value of the function at P is a, we can mark oft about P a circle so small that the value of the function within the circle differs from a by less than o/2, and therefore is certainly not zero. X] CONTINUITY 465 undefined at x = 0, y = 0, and we supplement the definition by assuming that «(0, 0) = 0. This function has a peculiar type of discontinuity at the origin. If we put x = 0, that is, if we move along the y-axis, the function becomes «(0, y) = 0, which has the constant value for all values of y. Along the a-axis we likewise have u{x, 0) = 0. Thus at the origin the function u{x, y) is continuous in a; if we keep y at the constant value and is continuous in y if we keep x at the constant value 0. Nevertheless, the function is discontinuous when considered as a function of the two vari- ables x and y. For at every point of the line y = x we find that w = 1, so that arbitrarily near the origin we can find points at which u assumes the value 1. The function is therefore discontinuous at the origin,* and cannot be defined at the origin in such a way as to make it continuous. The above example shows that a function can be continuous in x for every fixed value of y and continuous in y for every fixed value of x and yet discontinuous when considered as a function of the two variables. The essential point in the definition of con- tinuity is that the value of the function at a point P must be arbitrarily close to the value of the function at a point Q, provided only that Q is near enough to P; it is not permissible to restrict the position of Q relative to P in any other way. Examples x 2 4- V 1. Examine the continuity of the function 2 = - ^___„i_^. Sketch the Vx* + ]p level lines z = k (&= — 4, — 2, 0, 2, 4). Exhibit (on one graph) the behaviour of z as a function of x alone for y = — 2, — 1, 0, 1, 2. Similarly, exhibit the behaviour of z as a function of y alone for x = 0, ±1, ±2. Finally, exhibit the behaviour of z as a function of r alone when is con- stant (r, 6 being polar co-ordinates). 2. Show that the following functions are continuous: (a) sin (a? + y). . x 3 + y 3 ... ataxy a? + y* W + / (d) 3? log(** + s,»). * More generally, on the straight line y — x tan a inclined at the angle o to the z-axis we have u — 2 tano/(l + tan 2 a) — 2 sin a cos a -= sin 2a. The sur- face corresponding to the function u — 2xyftx* + y*) is therefore formed by rotating a straight line at right angles to the w-axis about that axis until it coincides with the x-axis, and simultaneously raising or lowering it so that the height sin 2a is associated with the angle a. As a increases up to 45° the straight line rises to the height 1, and subsequently falls to the level of the y-axis and below it to the depth 1, thereafter rising again to the level of the a;-axis. The surface enveloped by the moving straight line is known as the cylindroid; it is of importance in mechanics. 16 (K798) 466 FUNCTIONS OF SEVERAL VARIABLES [Chap. 3. Find whether or not the following functions are continuous, and if not, where they are discontinuous: (a) sin?. x (b) ** + y* x*+y*' (C x* + y* (d) 3? + y* x* + y' 3. The Derivatives of a Function of Several Variables 1. Definition. Geometrical Representation. If in a function of several variables we assign definite numerical values to all but one of the variables and allow only that one variable, say x, to vary, the function becomes a function of one Fig s Fig. 6 Sections of « = /(*, y) variable. We consider e.g. a function u = f(x, y) of the two variables x and y and give y a definite fixed value y = y = c. The function u = f(x, y ) of the single variable x which is thus formed may be simply represented geometrically by letting the surface u = f(x, y) be cut by the plane y = y (cf. figs. 5 and 6). The curve of intersection thus formed in the plane is represented by the equation u=f(x, y ). If we differentiate this function in the usual way at the point x = x (we assume that the derivative exists), we obtain the partial derivative of f(x, y) with respect to x at the point (x , y ). According to the usual definition of the derivative this is the limit * iim /( x o + ^ y )— /(% y ) _ * If (x„, y ) is a point on the houndary of the region of definition we make the restriction that in the passage to the limit the point (x + h, y„) must always remain in the region. X] PARTIAL DERIVATIVES 4^7 Geometrically this partial derivative denotes the tangent of the angle between a parallel to the x-axis and the tangent line to the curve u = f(x, y ). It is therefore the slope of the surface u = f(x, y) in the direction of the x-axis. To represent these partial derivatives several different nota- tions are used, of which we mention the following: iim f( x o + h >y<>)-f(*o,y ) =fx(Xot yo) = UxiXoj yo) . If we wish to emphasize that the partial derivative is the limit of a difference quotient we denote it by iT or a - /* ox ox Here we use a special round letter 3, instead of the ordinary d used in the differentiation of functions of one variable, in order to show that we are dealing with a function of several variables and differentiating with respect to one of them. It is sometimes convenient to use Cauchy's symbol D, men- tioned on p. 90, and write ox but we shall seldom use this symbol. In exactly the same way we define the partial derivative of f(x, y) with respect to y at the point (x , y ) by the relation Km /(*o,yo + *)-/(*o,yo) =M y ) = d l = D v f(x , y ). *-»o k oy This represents the slope of the curve of intersection of the sur- face u=f(x, y) with the plane x=x perpendicular to the jo-axis. Let us now think of the point (x , y ), hitherto considered fixed, as variable, and accordingly omit the suffixes 0. In other words, we think of the differentiation as carried out at any point (x, y) of the region of definition of f(x, y). Then the two deri- vatives are themselves functions of x and y: ««(*> V) =/.(*, V) «= d -^P^ and u v (x, y) =f y (x, y) = d J^dl m ox oy 468 FUNCTIONS OF SEVERAL VARIABLES [Chap. For example, the function u = x* -f y a has the partial derivatives u x = 2x (in differentiating with respect to x the term y 2 is regarded as a constant and so has the derivative 0) and «„ = 2y. The partial derivatives of u — x?y are u x = Zx*y and u v = x 3 . We similarly make the following definition for any number (n) of independent variables df(x 1 , x 2 , . . . , x„) dx 1 _ hm f(xi +h,x 2 , . . . , x„) —J(x x , a; 2 , . . . , x n ) *->0 h = fxS X l' X » • • • ' *») = D xJ(^ X 2> ■ • • > *»)> it being assumed that the limit exists. Of course we can also form higher partial derivatives of f(x, y) by again differentiating the partial derivatives of the "first order ",f x (x, y) and/„(a;, y), with respect to one of the variables, and repeating this process. We indicate the order of differen- tiation by the order of the suffixes or by the order of the symbols dx and dy in the " denominator ", from right to left,* and use the following symbols for the second partial derivatives: 3 (df\_dj = =Z)2 . 3 /9/\_ *f -f =Z) 2 f dx\dy/ ox ay d (df\_ dj , =Di f dy\ox/ oyax 8y \3y/ By 2 We likewise denote the third partial derivatives by di\d*)~ W~ Jxx " o hy\ _ a 3 / = f 9 / dj \ = dx \dxdyj 3 3 / _, ' — J X. dx z dy • In Continental usage, on the other hand, g- f J J is written ^-J- . X] PARTIAL DERIVATIVES 469 and in general the n-th derivatives by 9 /9"-y\ 3*/ 3a; Xdx"- 1 ) dx n Jx ' 3 /3»-y\ 3«/\3a; n_1 / dydx n ~ Finally, we shall study a few examples of the actual calcu- lation of partial derivatives. According to the definition all the independent variables are to be kept constant except the one with respect to which we are differentiating. We therefore have merely to regard the other variables as constants and carry out the differentiation according to the rules by which we differen- tiate functions of a single independent variable. Thus for example we have: 1. Function f(x, y) — xy; first derivatives, f x = y, f v = x; second derivatives, f^ =0, f XB = f vx = 1, / w = 0. 2. Function f(x, y) = V(x 2 + y a ); first derivatives, f x = * — -, /„ = ■- y — -. V(x 2 +y 2 ) V(x 2 + y % ) (Thus for the radius vector r = V(x i + y 2 ) from the origin to the point (a;, y) the partial derivatives with respect to x and to y are given by the cosine, cos<p = x/r, and the sine, simp = y/r, of the angle <p, which the radius vector makes with the positive direction of the x-axis.) Second derivatives, V(x* + y*) , _ V(x 2 + y 2 ) J xx ~~ J xv Jyx x 2 +y 2 xy V(x* + y 2 f V(** + J/ 2 ) - . f f _ V{x 2 + y 2 ) y* sin 2 9 V(x 2 + y 2 f r ' sin 9 cos 9 r x> cos 2 9 V(x 2 + y 2 f r 47° FUNCTIONS OF SEVERAL VARIABLES [Chap. 3. Reciprocal of the radius vector in three dimensions: 1 1 f(x, y, z) == V(a^ + y* + z 2 ) r ' first derivatives, V(a; a + y* + z 2 ) 3 r 3 ' f = y = _y " V(a^ + y a + z 2 ) 8 r 3 ' 8 V(a; a + y* + z 2 ) 3 r 3 ' second derivatives, f =-1 + ™ f - - 1 + W f - 1 , 3z» / = f _ 3x V f _ * _ 3yz , _ , _ 3zx Jxv Jvx -ft Jvz — Jzv ~S~> Jzx — Jxz £"• From this we see that for the function / = - the equation V (* + »» + *) Jxx ~ Jmi \ Jzz — ~i T ; — V r 3 r 5 holds for all values of x, y, z except 0, 0, 0; as we say, the equation Jxx i Jyy + Jzz = is satisfied identically in x, y, z by the function f(x, y, z) = 1/r. 4. Function f(x, y) = -i- e - ( JC_a ) , ' 4 J'; VJ first derivatives, second derivatives, f = C 3 i_ _ 3 im# + <* - «) 4> \ e -(,- a) .,*v X] PARTIAL DERIVATIVES 471 The equation fxx - Sv = is therefore satisfied identically in x and y. Just as in the case of one independent variable, the possession of derivatives is a special property * of a function. All the same, this property is possessed by all functions of practical impor- tance, except perhaps at isolated exceptional points. In contrast with functions of one variable, the possession of derivatives does not imply the continuity of the function. This is clearly shown by the example u = i— already considered x*+y* on pp. 464-465; for the partial derivatives exist everywhere, and yet the function is discontinuous at the origin. But, as is stated by the following theorem, the possession of bounded derivatives does imply continuity: If a function f(x, y) has partial derivatives f, and fy everywhere in a region R, and these derivatives everywhere satisfy the in- equalities \Mx,y)\<M, \f v {x,y)\<M, where M is independent of x and y, then f (x, y) is continuous every- where in R. In particular, if f x and/ s are continuous they are necessarily bounded, so that f(x, y) is also continuous. The proof of this theorem we shall leave for Vol. II. The reader will have noticed that in all our examples the equation f xv =f vx is satisfied. In other words, it made no dif- ference whether we differentiated first with respect to x and then with respect to y or vice versa. This is no accidental occurrence. In fact, we have the following theorem: If the " mixed " partial derivatives t^ and t^ of a function f (x, y) are continuous in a region R, then the equation Jvx == J XV holds everywhere in the interior of this region; that is, the order of differentiation with respect to x and y is immaterial. * The expression " differentiable " implies more than that the partial deri- vatives with respect to x and to y exist. Cf . Vol. II. 472 FUNCTIONS OF SEVERAL VARIABLES [Chap. By applying this theorem to f x and f„, then to f xx , f xy , f yy , and so on, we find that Jxxv Jxvx Jyxxy Jxyy == Jvxv == Jvyxt Jxxvv ^ Jxyxy === Jxwx ^ = Jvxxy ^ Jyxyx == Jyyxxi &C., and in general we have the following result: In repeated differentiation of a function of two variables the order of differentiation can be changed arbitrarily, provided only that the derivatives in question are continuous functions. For the proof of this theorem we refer the reader to Vol. II. Examples 1. Find the first partial derivatives of the following: (a) vte + t,*. (d) * . V(l + x + y* + z 2 ) (b) sin(a^ — y). (e)ysin(xz). (c) e*~y. (/ ) log V(l + *» + «/ a ). 2. Find all the first and second partial derivatives of the following: (a) xy. (d) x". (6) log as/. (e) e(**). (c) tan (arc tana; + arc tany). 3.* Find a function f(x, y) which is a function of (x* + y 2 ) and is also a product of the form ty(x)ty{y); that is, solve the equations f(x, y) = ^(3? + y 2 ) = 4t{x)4/{y\ for the unknown functions. 4. The Chain Rule and the Differentiation of Inverse Functions 1. Functions of Functions (Compound Functions). It often happens that a function u of the independent variables x, y is stated in the form u=f(£,-n, ...), where the arguments £, -n, . . . of the function / are themselves functions of x and y: t = 4>(x, y), ij = jj(x, y), ... . X] THE CHAIN RULE 473 We then say that w =/(£ v> ■ • ■) =f(<f>( x > y)> 4>( x > y)> • • •) = F i x > y) is given as a compound function of x and y. For example, the function u = e*'y(x + yf may be written as a compound function by means of the relations u = <£ 7) 3 = /(5, 7]); % = x*y, 7) = x + y. Similarly, the function u = log(a; + 1) . arc cos V(4 — x 2 — y*) can be expressed in the form u = 7) arc cos 5 = /(£, yj); £ = V(4 — a? — j/ 2 ), 7) = log(* + 1). In order to make this concept more precise, we assume to begin with that the functions $ = <f>{x, y), tj = ^(x, y), . . . are defined in a certain region R of the independent variables a;, y. Then to every point (x, y) of R there corresponds a point (£, ij, . . .) in the space with co-ordinates g, 17, . . . . As the point (x, y) ranges over R, the point (£, 77, . . .) will range over a certain set of values. We assume that the point (f, 77, . . .) always lies within a region S in which /(£, ■>?,...) is denned. The function u = f(<f>{ x > y)> #"i V)> • • ■) = F(x, y) is then defined in the region R. Referring to our examples, in the first we find that 5 and 7) are denned for every x, y and/(£, 7)) is defined for every 5, vj, so that our region B can be taken to be the whole xy-pkme. In the second example, however, the region S is restricted by the inequality | ? | £? 1, since for | 5 I > I the function arc cos £ is undefined. Secondly, the region B is restricted by the inequalities x + 1 > and x 2 + y 2 fg 4, since for other values \ and tj are not both defined. Thirdly, the region R must be further limited by the inequality 3 g x 2 + V* in order that the point with co-ordinates 5, t) shall fall in 8; that is, the restriction [ Z, [ ^ 1 implies that x 2 + y 2 ;> 3. Hence B consists of the part of the ring 3 5S x 2 -\- y 2 ^ 4 lying to the right of the line x = — 1. The following theorem on compound functions is an im- mediate consequence of the definitions: If the function u = 1(£ , tj, . . .) is continuous in S and the functions £ = <f>(x, y), tj = i/i(x, y), . . . are continuous in E, then 16* IE 798) 474 FUNCTIONS OF SEVERAL VARIABLES [Chap. the compound function u = F(x, y) is continuous in R. The reader will be able to prove this for himself. 2. The Chain Rule. We now turn our attention to compound functions of the type u=f(g, 7], . . .), where £, •>?, ... depend on the single variable x: € = tfr), V = <M»), For such functions we have the important theorem known as the chain rule: If the function u = f(£, tj, . . .) has continuous partial deri- vatives of the first order in S, and the functions £ = <f>(x), i) = </r(x), . . . have continuous first derivatives in the interval R, a ^ x 5S b, then u = f {<j>{x), ^r(x), . . .} = F(x) has a continuous derivative in R, and F'(x)=f l i'(x)+f vV '(x)+.... The right-hand side of this equation is an abbreviation for /*{#*), #*),...}*'(«)+.... To simplify the notation we shall assume that /is a function of the three arguments £, 77, £. We shall denote by x Q an arbi- trary fixed point of the interval a ^ x ^ b, by £ , tj , £ the corresponding values & = <f>(x ), -q = ifi(x ), £ = x (x ), and by £, ij, £ the values <f>(x), i/i(x), %(x) corresponding to a variable point x = x -f A. We first write down the identity *(*) - *■(**) = /(£» i?» £) — /(&> *?o> £0) In each bracket on the right we observe that only one of the independent variables changes its value. Hence to each bracket we can apply the mean value theorem for functions of one variable, and obtain F{x) - F(x ) = (£- lo) Ml % 0+ (v- Vo)L (£0, v> Q+(t- W/< (U v , 0> X] THE CHAIN RULE 475 where | lies between £ and $, -rj between rj and -q, and £ between 6 and £. Further, by the mean value theorem, we have £ — £ = <f,(x) - <£(x ) = (a: — a; ) f fo), *l — Vo= <A(») — 'I'M = (a — a ) "A'^), C - Co = Xfc) — x(«%) = (* — afe) X^s). where a^, x 2 , and x 3 all lie between x and a;. Substituting these values in the last equation and dividing by x — x , we have F(x) - F(x ) X Xq We now let x tend to x . Owing to the continuity of (f>(x), if>(x), X(x) the quantities |, 17, £ tend to £ , ij , £ respectively, and a fortiori f, ij, £ do likewise. Also x 1 , x 2 , and x 3 tend to % Since all the functions on the right are continuous, we have ]hQ F M -JW = r{Xo) =/f(£o> %. WfW+/,(^ Vo, ?o)fW+/;(?o. %. Co) x'K), thus establishing the formula for F'(x). The continuity of F'(x) follows immediately from the formula, since <£', if/', and x' » re continuous by hypothesis and / f , /,, and f i are continuous functions of continuous functions. This theorem may be extended to compound functions of two or more variables, as follows: If the function u = f(£, ij, . . .) has continuous partial deri- vatives of the first order in the region S, and the functions (;= <f>(x., y), rj = tfi{x, y), have continuous partial derivatives of the first order in R, then u = F(x, y) = i{<f>(x, y), ifi(x, y), . . .}has continuous partial derivatives of the first order in It, and these derivatives are given by the formulas, f* = i ' t *.+J ;<a* + — Fy=f(<f>y+fr l 4>v + These formulae are often written in the abbreviated form u x = u ( £ x + M,1J a + . . . , m v = m 4 |, + «,i7«+ • • • • 476 FUNCTIONS OF SEVERAL VARIABLES [Chap. To establish them we temporarily introduce the notation g(x) = <f>(x, y ), h(x) = ifi(x, y ), . . . , where y is a fixed value of y. By the definition of the partial derivatives it follows that g'(x) = <f> x (x, y ), h'(x) = ifi x (x, y ), .... Similarly, if we write H(x) = F(x, y ), we have H'(x) = F x (x, y ). We now apply the theorem just proved to the function u = H(x) =/(!, r), . . .) = /{g(x), h(x), . . .}, and obtain H'(x ) =ft 9 '(x ) +f,h'(x ) +... . Returning to the original symbols, we have -Fsfoo Vo) =/^*( a; o> Vo) +f v >Px(xo> y ) + • • • • The other formula is proved in a similar way. If we wish to calculate the derivatives of higher order we need only differentiate the right-hand side of our formulae again with respect to x and y, regarding f it /,,... as compound functions. Thus for u = /(& 7]) =f{<f>(x, y), <j,(x, y)}, we have u xx =f(S <f>x 2 + 2/f, <f> x lfi x +/„„ </> x 2 +/( 4>xx +/, <A*S, u xv =/« <f>x <f>y +/|, (<f>x>Pv+ 4>v >Px) +fm 'I'x $v +/f fxv +/, <l>xv, ««» =/« ( t>v 2 + Vir, 4>n <t>v +f m <A» 2 +ft <t>vv +/, <l>vv 3. Examples.* 1. u = e xt!M y+y coax . Here we put £ = x tanj/, in = y cosa;, bo that SL = tany, L. = . oos 2 y r) x = — y sins, y]„ = cosx. Since u = e£ +T >, uj = m, = «f +i, ttiid Ujt _ e x tany+y cosx ( tan ^ _ y sm X ) t „ _ ^cVmy+ycasx(_^__ + CO sa;). \cos 2 2/ / 2. An example of a compound function of a single variable is u ={<?(*) }**>= 5" = /(5,v]), where we put 5 = g(x), i\ = h(x). We immediately obtain ^ = ft V + /,tf = ^~ 1 5' + 5" ^g I ■ V = fo (*) }**> f *(*) ^ + *'(*) logg(x) } . I g(x) ) We have already dealt with a special case of this by rather artificial methods (p. 203). * We would emphasize that the following differentiations can also bo carried out directly, without using the chain rule. X] THE CHAIN RULE 477 4. Change of the Independent Variables. A particularly important type of compound function occurs in the process of changing the independent variables. For ex- ample, let u = /(£, 13) be a function of f and rj, which we interpret as rectangular co-ordinates in the l^-plane. If we rotate the axes in the |rj-plane through an angle 6 we obtain a new system of co-ordinates x, y, related to the co-ordinates £, ij by the equations: £ = SCCOS0 — ysin0, ij = zsinfl-f y cosfl, or x = |cos0 + rjam.0, y= — £sin0+ ij cos0. The function u = /(£, yj) can then be expressed as a function of the new variables x, y: u=f(Z,- n ) = F{x,y). Then the chain rule immediately gives u x = u ( cos0 + m, 8in.6, w„ = — w f sin0 + u v cosO. Thus the partial derivatives are transformed by the same formulae as the independent variables. This is true for rotation of the axes in space also. Another important type of change of co-ordinates is the change from rectangular co-ordinates x, y to polar co-ordinates r, 8. This is done by means of the equations x=rcoad, y = rsind, r = a/(x* + y% 6 = arc tan y ~. x We then find that for an arbitrary function u=f(x, y) with continuous partial derivatives of the first order we have « =/(z, y) =f(r costf, r sinfl) = F(r, 6), a x V a sm ^ u x — u r r x + ufl x = u r - — u e 2- = u r cost) — u e , r r* r n V , £ • a 1 cos ^ u v = u r r y + u e y — u r z + u e — = u r sine/ -f- u e . r r r 478 FUNCTIONS OF SEVERAL VARIABLES [Chap. From this we obtain the equation «« 2 +W» 2 =Mr 2 + i«»«. 1 r 2 which is often useful. In general, let us consider a pair of functions £ — <f>(x, y), ■q = i/j{x, y) which are continuous and have continuous deri- vatives in a region R of the ay-plane. To each point (po, y) in R these equations assign a point £ = <f>(x, y), rj = ip(x, y) in the ^ij-plane. As (x, y) ranges over R, the corresponding point (|, rj) will range over some set of values S in the fq-plane. It is of course possible that several distinct points (x, y) will give the same values for |, 17, so that to several points (x, y) there corre- sponds only one point (£, rj). We shall assume that this is not the case, but instead that to one point Q{£, rj) in S there corresponds exactly one point P{x, y) in R. We may therefore look at the correspondence from either point of view — saying that Q cor- responds to P or that P corresponds to Q. The latter point of view can be expressed thus: to each point (£, rj) in S there cor- responds one x and one y, namely, the co-ordinates of P, or, in equations, there are two functions x = g{£, rj), y=h(£, rj), denned in S, which represent the correspondence inverse to $=$(x, y), rj= iP(x, y). It often happens that the functions g (£, rj), h(£,rj) are by no means easy to calculate, even when they do exist. Hence we shall now find how to obtain the partial derivatives g t , g v , h ( , A, directly from the partial derivatives <f> x , <f> v , ift x , ip y , without cal- culating g and h themselves at all. For this purpose we observe that if we choose any point Q($, rj), find the corresponding point P{g(S, rj), h(i, r])} in R, and then find the point in S correspond- ing to P, which is <f>{g(£, rj), h(£, rj)}, ${g(£, rj), h(£, v )}, we have simply returned to the point Q. That is, the equations f= <f>{9(£> V)> h(£> V)}> V = H9(£> V)> *(£ V)} are identities in £ and ij. We now differentiate * both sides of both equations with respect to £ and to rj. We have 1 = 4>xfft + <i>v h, = c/> x g v + <f> v \, = fafft + iMf. 1 = <Px9 v + 4>vK * If an equation expresses an identical relationship, differentiation with respect to any independent variable in it yields an identity, as follows im- mediately from the definition. v XJ THE CHAIN RULE Solving these equations, we find that „ _ fa „ _ _ fa j. fa i _ fa 9t ~~D' 3 * D' f— ~D' D' or D' ~ v I' D' Xe =^, x.= -^. y t =-H, y,= ^, 5? 01 = ^ x D' where by D we mean the determinant D= ixVv— iyVx = d£ d£ dx dy dr) dx dr) dy 479 which we assume is not zero. This determinant D, called the functional determinant or Jacobian of (f, r?) with respect to (x, y), occurs so frequently that a special symbol is often used for it: Z> = _3(£l) d(x, y)' Examples 1. Calculate the partial derivatives of the first order for: 1 («)/- (C) /= x 2 + y log (1 + x* + J/ a + z 2 ). V(x 2 + ^ 2 + 2xy cosz) (6) /= arc sin — 5—. (<Z) / = arc tan V (x + yz). z + y 2 2. Calculate the derivatives of (a) / = x^\ (6) / = (( \ J . 3. Prove that if J(x, y) satisfies " Laplace's equation " 8a? dy* does <p(x, «)=/( -, — - — _). 4. Prove that the functions («) /(«. ») = log vV + y ! ). (6) g(x, y, z) = y (g , ^ + ^ . (c) A(x, i/, z, w>) = , X 2 + «/ 2 -+- Z a + U! 2 480 FUNCTIONS OF SEVERAL VARIABLES [Chap. satisfy the respective Laplace's equations: («) fxx + fw = °- ( 6 ) 9xx + SW + S,i = 0. (c) h xx + h yv + h zll + h ww = 0. 5. Given z= r 2 cos 6, where r and 6 are polar co-ordinates, find z x and z„ at the point 6 = - , r = 2. Express z r and z e in terms of z x and z„. 6. By the transformation %= a -{- ax -\- py, i] = 6 — fix + «y> in which a, 6, a, p are constants and a 2 + p 2 = 1, the function m(k, y) is transformed into a function {/(£, tj) of ? and tj. Prove that 7. Find the Jacobians of the following transformations: (a) %=ax-{-by, i) = ex + dy; (6) r = V(a; 2 -f y 2 ), 6 = arc tan"; (c) ? = 3?, 1) = y 2 . 8. If x = a;(w, »), y = y(«, «) and «=«( 5, ■»)),»= f(5, v))> prove that 8(s, y) _ 6 (a:, y) 6 (it, u) 8(5,7)) 8(«,»)'S(5,yi)" 0. As a corollary to Ex. 8, prove that 8(x, y) _ 1 d(u, v) d(u, v) 8(x, y) 10. Using Ex. 9, find the Jacobians of the transformations which are the inverses of those in Ex. 7. 5. Implicit Functions In the study of functions of several variables we have as yet had no analogue to the inverse function. We can regard the inverse function of y = f(x) as the function obtained if we solve the equation y — f{x)=0 for x. In this section we shall seek more generally to solve equations F(x, y) = for x or for y, and to discuss functions of several variables in a corresponding way. Even in elementary analytical geometry curves are frequently represented, not by equations y = f(x) or x = <j> (y), but by an equation involving x and y in the form F(x, y) = 0. For example, XJ IMPLICIT FUNCTIONS 481 «J*2 a.2 we have the circle x 2 + w 2 — 1 = 0, the ellipse — + y — — 1 = 0, a 2 b 2 and the lemniscate (cc 2 -f- y 2 ) 2 — 2a 2 (x 2 — y 2 ) = 0. In order to obtain y as a function of x, or a; as a function of y, we must solve the equation for y or for x. We then say that the function V — f( x ) or a; = <f>(y) found in this way is denned implicitly by the equation F(x, y) = 0, and that the solution of this equation gives us the function explicitly. In the examples cited and in many others the solution can be carried out and the solutions stated explicitly in terms of the elementary functions. In other cases the solution can be obtained in terms of an infinite series or other limiting process; that is, we can approximate to the solution y =f(%) or x = <j>(y) as closely as we desire. For many purposes, however, it is more convenient to base our discussion on the implicit definition F(x, y) = 0, instead of resorting to an exact or approximate solution of the equation. The idea that every function F (x, y) yields a function y=f(x) or x= <f>(y) given implicitly by means of the equation F(x, y) = is erroneous. On the contrary, it is easy to give examples of functions F(x, y) which, when equated to zero, permit of no solution in terms of functions of one variable. Thus for example the equation x 2 -\- y 2 = is satisfied by the single pair of values x = 0, y = only, while the equation x 2 -\- y 2 + 1 = is satisfied by no (real) values at all. It is therefore necessary to investigate the matter more closely in order to find out whether an equation F{x, y) = can actually be solved, and what properties the solution has. Such an investigation we cannot undertake in detail here, but content ourselves with a geometrical interpre- tation which suggests the required results, the rigorous proofs being left for Volume II. 1. Geometrical Interpretation of Implicit Functions. To discuss this problem geometrically we represent the function u = F(x, y) by a surface in three-dimensional space. Finding values (a;, y) which satisfy the equation ^(a;, y) = is the same thing as finding values (x, y) which satisfy two equations F(x, y) = u, u = 0; in other words, we wish to find the inter- section of the surface u = F(x, y) and the plane u = 0, which is the a;y-plane. We then suppose that we have a definite point { x o> Vo) which satisfies the equation F(x , y ) = 0; that is, at 482 FUNCTIONS OF SEVERAL VARIABLES [Chap. ( x o> Vo) * ne surface u = F(x, y) has a point in common with the plane u = 0. (If no such point exists, there is no intersection, and the equation F(x, y) = cannot be solved.) If the tangent plane to the surface u = F(x, y) at the point (x , y ) is not hori- zontal, it cuts the plane u = in a single straight line. Intuition then tells us that the surface u = F(x, y), lying near the tangent plane, likewise cuts the plane u = in a single well-defined curve. How far this curve extends does not at present concern us. The tangent plane will be horizontal if the two curves m = F(x , y) and u = F(x, y ) both have horizontal tangent lines at (x , y ); that is, if F x (x , y ) = and F v (x , y ) = 0. Thus if either F x (x , y ) 4= or F v (x , y ) #= the tangent plane is not horizontal, and, as we have just seen, we may expect that a solution in the form y—f(x) or x = <£(*/) will exist. If, on the other hand, both F x {x , y ) and F„(x , y ) have the value 0, we readily see that there is no guarantee that the solu- tion is possible. For example, f or F = 1 — V(l — a 2 — j/ 2 ) the corresponding spherical surface u = 1 — V(l — a? — y 2 ) has the point (0, 0) in common with the ay-plane. The partial derivatives F x (0, 0) and ^,,(0, 0) are both zero; and we find that no point other than (0, 0) satisfies the equation F — 0. For the function F(x, y) = xy we find that F(0, 0) = 0, while ^(0, 0) = F v (0, 0) = 0. Here all the points on the jc-axis and all the points on the y-axia satisfy the equation F(x, y) = 0; and in the neighbourhood of the origin we have no unique solution x = <p(y) or y = f(x). Thus we see that when F x (x , y ) — F y (x Q , y ) = we cannot be sure that a solution exists. If we accordingly return to the case in which one of the partial derivatives — say F v (x , y ), to be specific — is not zero, the graphi- cal suggestion that a smooth surface should be cut by a non- tangent plane in a smooth curve leads us to suspect that the following theorem is true: If the function F(x, y) has continuous derivatives F x and F and if at the point (x , y ) the equation F(x , y ) = is satisfied, while F y (x , y ) is not zero, then we can mark off about the point (xo. y<>) a rectangle x x 5S x ^ x 2 , y x ^ y 5S y 2 such that for every x in the interval x x 5S x :gj x 2 the equation F(x, y) = determines just one value y = f(x) lying in the interval y x <S y <S y 2 . This function y = f(x) satisfies the equation y = f (xj, and the equation F{x,f(x)}=0 XI IMPLICIT FUNCTIONS 4§3 is satisfied for every x in the interval. Moreover, the function y = f(x) is continuous and has a continuous derivative. This can actually be rigorously proved, and will be proved in Vol. II. Assuming it to be true, we can add the following: The derivative of the function y = f(x) is given by the equation y>=f {x )=- F * This follows immediately by using the chain rule. For i F{x,f(x)}^F x ^+F v ^=F x + FJ'. But since F{x,f(x)} is identically zero, its derivative is also zero; hence F x -\-F v f'= 0, and the formula is established. If we regard the right-hand side of the formula as a compound function of x and differentiate according to the chain rule, replacing y' by —FJF V , we have v >. = _ F V (F XX + F m y') - F X (F XV + F m y') V F v * _ _ FxxFy 2 — ^FxyF x F y -\- F yy F x 2 F 8 Continuing the process, we may calculate y'", y iv , &c. By using this formula we can usually find the derivative of a function given in implicit form much more easily than by solving first and then differentiating. For example, for the circle F (x, y) = a* + y* - 1 = we have y" = x = . f* y This is easily verified. For on solving the equation of the circle for y we obtain two solutions, namely, y=V{l — x 2 ) and y = — V(l — x i ), giving the upper and lower semicircles respectively. For the upper we have " V(l - 3?) for the lower «' = -, — — . * V(l - **/ so that in either case y' = . 484 FUNCTIONS OF SEVERAL VARIABLES [Chap. As another example, we have F(x, y) = e x+ y + y — x = 0. We find that F x ( -, — -J = 0, while F y ( -, — - J = 2. Thus the equation has a solu- tion y = f(x); but the actual explicit calculation of the function f(x) would not be simple. Nevertheless, we have ' = _ z? = _ e * +y - 1 y F y = e*+y + 1 In order that the function f(x) may have a maximum or a minimum we must have y' = 0, that is, e x +y —1 = 0, whence y = —x. Substitution of y = — x in the equation F(x, y) = gives I — 2x = 0, whence x = -, ll 2 3/ = If we calculate /"(x) for x = -, we find it to be negative, so that — -is the maximum value of y. An extension of this theorem for implicit functions to func- tions of a greater number of independent variables readily sug- gests itself. The extension is as follows: Let F(x, y, . . . , z, u) be a continuous function of the indepen- dent variables x, y, . . . , z, u with continuous partial derivatives F x , F y , . . . , F z , F u . For the system of values (x , y , . . . , z , u„) let F(xo, y , . . . , z , u„) = and F u (Xo, y , . . . , z , «o) # 0. Then we can mark off an interval u x ^ u :S u 2 about Uq and a region B. containing (x„, y , . . . , z ) such that for every (x, y, . . . , z) in R the equation F(x, y, . . . , z, u) = is satisfied by just one value of u in the interval u x :|S u ^ Ug. TAis wihte 0/ u, which we denote by u = f (x, y, . . . , z), is a continuous function of x, y, . . . , z and possesses continuous partial derivatives f,, f y , . . . , f z , and w o ~J\ x o> Vo' • • • > z o)- TAe derivatives of i are given by the equations F x + F u f x =0, F y + F u f v = 0, F.+ Fuf. = 0- For the proof of the existence and continuity of u we again refer the reader to Vol. II. The formulae for f xv , &c, follow im- mediately from the chain rule. Incidentally, the concept of an implicit function enables us to give a general definition of the term " algebraic function ", X] IMPLICIT FUNCTIONS 485 We say that u =f(x, y, . . . ,z) is an algebraic function of the independent variables x, y, . . . , z if u can be denned implicitly by an equation F(x, y, . . . , z, u) = 0, where F is a polynomial in x, y, . . . , z, u; that is, if u satisfies an " algebraic equation". Functions which do not satisfy any algebraic equation are called transcendental (p. 24). As an example of our differentiation formula we consider the ellipsoid a 2 o 2 c 2 For the partial derivatives we have _2x c 2 c' x a 2 2u a 2 u = _ 1y c 2 c* y m Uy ft 2 ' 2u ~ b 2 ' u and by differentiating again c 2 1 c 2 x _ c 2 a 2 u 2 + c*x* a 2 u a 2 u 2 a*u 3 c 2 x c 4 xy M = — _?, a 2 u 2 a 2 b 2 u 3 c 2 1 c 2 y _ c 2 6 2 m 2 + c*y* Uyv ~ b 2 'u + V'u* Uy -~ bW Examples 1. Prove that the following equations have unique solutions for y near the points indicated: (a) x 2 + xy + y 2 = 7 (2, 1). (6) x cosxy = ('•?• (c) xy + logxy = 1 (1, 1). (rf) z 6 + y 6 + xy = 3 (1, 1). 2. Find the first derivatives of the solutions in Ex. 1. 3. Find the second derivatives of the solutions in Ex. 1. 4. Find the maximum and minimum values of the function y = f(r) defined by the equation x 2 + xy + y 2 = 27. 5. Show that the equation x + y + z = sinas/u can be solved for z near (0, 0, 0). Find the partial derivatives of the solution. *86 FUNCTIONS OF SEVERAL VARIABLES [Chap. 6. Multiple and Repeated Integrals 1. Multiple Integrals. We consider a function u = f(x, y) which is denned and continuous in a rectangle R(a f^x^Lb, c ^ y 5S d), and which takes positive values only. We wish to assign a volume to the portion of three-dimensional space bounded by the rectangle R, the surface w=/(a;, y), and the four planes x= a, x=b, y = c, y = d perpendicular to the xy-plane. Moreover, the volume should be denned so as to satisfy certain elementary conditions: (1) if the three-dimensional region is a prism — i.e. if the function u is a constant k — the volume should be the product of the base by the altitude, V = (b — a) (d — c)k; /SST~~^y \ [^Tr^/j i A \A V A / V v*,v / V Fig. 7 (2) if we divide the rectangle R into smaller rectangles R^ and R 2 by drawing straight lines, then the volume over R should be equal to the volume over R^ plus the volume over i? 2 ; (3) if the three-dimensional region Rj completely includes R 2 , the volume of R 1 should be at least as great as that of R 2 . These considerations lead us to a method of denning V which is an immediate extension of the method of denning area in Chap. II (p. 77 et seq.). By constructing lines parallel to the sides we subdivide the rectangle R into smaller rectangles R i , R 2 , . . . , R n , whose areas we denote by AR lt AR 2 , . . . , AR n . In each rectangle Rj the function has a least value m f and a greatest value Mj. Therefore a prism whose base is R< and whose height is Mj completely includes the portion of our region over R it while this portion of the region contains the prism with base R, and height m f (cf. fig. 7). We see, therefore, that the volume of X] MULTIPLE AND REPEATED INTEGRALS 487 the portion in question lies between nijR, and MjR t . Thus the total volume V should be such that S mjARf ^ V ^ S M,AR t . -1 y-i Suppose now that the number n of rectangles increases beyond all bounds in such a way that the length of the longest diagonal tends to zero. Intuition leads us to expect that the two sums 'Zm j AR i and HMjAR, will both converge and will tend to the same limit. This limit we therefore call the volume V. The reader will have observed that we have carried out an immediate generalization of the discussion in Chap. II (p. 78). As in Chap. II, we call the common limit of the sums Swi^AE, and TiMjARj the integral of the function u = f(x, y) over the rect- angle E, and we denote it by the symbol fff(x,y)dr. It is at once clear that if in each rectangle R } we choose a point (£j, 7]}) and find the corresponding value of the function /(£,, ij,) then the limiting relation lim S/(& Vj )AR,= fff(x, y)dr must hold; for the sum 2/(&, t] } )AR } lies between l.m i AR j and HiMjARj, both of which approach the integral as a limit. As a particular method of subdividing R into smaller rect- angles, we may divide the side a^Lx^b into n intervals of length Ax = (b — a)/n and the side c 5S y ^ d into m intervals of length Ay— (d — c)/m, and then draw parallels to the axes through the points of division thus marked. The area of each rectangle R t is then AR, = Ax Ay. Choosing a point (g h rj f ) arbitrarily in each rectangle R t , we form the sum Z*/(6, „)AK, = S l /(f / , rjJAxAy. As n and m both increase without limit, this sum approaches the integral as a limit. This type of subdivision suggests a second notation for the integral, which has been in common use since the time of Leibnitz, namely, fffl x > V)dxdy. 488 FUNCTIONS OF SEVERAL VARIABLES fCHAP. The proof that such a limit exists if u = f(x, y) is continuous can be carried out as in the appendix to Chap. II (p. 131 et seq.). We shall, however, assume without proof an even stronger state- ment, namely, the following: If the function f(x, y) is continuous except along a finite number of smooth * curves y = f (x) or x = <f>(y) along which f(x, y) has jump discontinuities, then the double integral fff(x,y)dr exists. The proof of this we leave for Vol. II. It depends essentially on the fact that as the number of rectangles increases the total area of the rectangles having points in common with the curves of discontinuity tends to zero. Thus even though Mj and m f may differ considerably for such rectangles, they give rise to little difference between the sums SMjARj and Em^A-fi^. With this assumption we can find the area under surfaces u = f(x, y) for which (x, y) ranges over quite complicated regions R. For suppose that the region R is bounded by a finite number of curves x = <f>(y) or y = i/j(x) with continuous derivatives, and that f(x, y) is continuous in R. We enclose R in a rectangle R', and at the points of R' which do not belong to R we assign to f(x, y) the value 0. Then we take the integral / I f(x, y)dr, taken over the region R', as the volume under the surface u = f(x, y), where (x, y) is in R. This integral is usually denoted by I ff(x, y)dr. Certain simple but important theorems relating to these double integrals follow directly from the definition. Here we simply state the theorems; the reader will be able to prove them without any trouble. If f (x, y) and g(x, y) are integrable over a rectangle, then so are f ± g> and °f » where c is a constant: If tf( x ' y} ±s( x > y)} dr = fff( x ' y) dr ± // 9( x > y) dr > ff cf(v, V)dr = cfff(x, y)dr. * By smooth curves we mean, as before, curves with continuous derivatives. XI MULTIPLE AND REPEATED INTEGRALS 489 If f (x, y) ^ g(x, y) in R, then fff( x > y) dr ^ff gfr, y) dr - Ij R is the sum of two regions R x and R 2 , then f ff(x, y)dr = [f f(x, y)dr + ff f(x, y)dr. 2. Reduction of Double Integrals to Repeated Simple Integrals. We now have a definition of the double integral, with its interpretation as a volume and with the many possibilities of usefulness which our experience with the single integral suggests; but as yet we do not possess a method for evaluating such integrals. In this section we shall see how the calculation of a double integral can be reduced to that of two single integrals. We suppose that u=f(x, y) is a function which is denned and continuous in a rectangle R, a 5S x 5S b, c ^ y ^ d. If we fix upon any value x in the interval a :£ x ^ b, the function f(x , y) is a continuous function of the remaining variable y. Hence the integral r d J f{x ,y)dy exists, and can be evaluated by the methods of earlier chapters. This integral has a definite value for each value of x that we may choose; in other words, the integral is a function <£(a; ) of the quantity x ; f c f(x,y)dy= <f>{x). For example, suppose that u =f(x, y) = x 2 y 3 , 5S x ^ 1, £S t/ ^ 3. For each fixed x in the interval fS x ^ 1 the x^y 3 dy can be evaluated, and is, in fact, — x 2 ; that is, it is a function of x. Or if f(x, y) = e xv , 1 ^ x ^ 2, r* 1 1 <= w ^ 4, we have / e xv dy ■= (e ix — e x ). Having thus found the function <f> (x), we can prove that it is continuous; this is a simple consequence of the uniform con- 490 FUNCTIONS OF SEVERAL VARIABLES [Chap. tinuity of f(x, y). It is therefore possible to integrate <f> (x) betweeD the limits a and b, thus obtaining the " repeated integral " J 4(x)dx =f (j f(x, y) dyj dx. By reversing the order of the process, first calculating the func- r b tion of y denned by / f(x, y) dx and then integrating from c to J a d, we obtain the other repeated integral I e (f a f( x >y) dx ) d y- These integrals, as we have seen, are obtained by a double application of the ordinary simple integration which we have studied in previous chapters. Their importance lies in the following fact: For continuous functions f(x, y), and for functions f(x, y) having at most jump discontinuities on a finite number of smooth curves, the repeated integrals are equal to the double integral: j Ij^' y } dr = L (Ie f( x ' ^ dy ) dx = L [f a ^ x '^ dx ) dy - We shall content ourselves with an intuitive discussion of the case where /(cb, y) is continuous. In our original discussion of the double integral regarded as the volume lying above the rect- angle a^,x<Lb, c^y 5S <Z and below the surface u =f(x, y), we obtained this volume by subdividing the solid into vertical columns and then letting the diagonals of the bases of these columns approach zero. Instead of this we can divide the solid into slices of breadth h = {d — c)/n by drawing the lines y = c + vk {v = 0, 1, . . . , n) parallel to the x-axis and then constructing a plane perpendicular to the a^-plane through each Kne (cf. fig. 8). These planes cut the solid into n slices which grow thinner as n increases, and whose total volume is equal to the double integral. We now see that the volume of each slice is approximately (but of course not as a rule exactly) equal to X] MULTIPLE AND REPEATED INTEGRALS 49 1 the product of the thickness k by the area of the left-hand face, that is, equal to r* k / f(x, c + vk) dx. J a Therefore, if we write , r b <t>{y) = l f{x,y)dx the desired volume is represented approximately by n-l S k<f>(c-\- vk). ■— Fig. 8 As n -> oo these sums tend to £<i>(y)dy- It is therefore reasonable to expect that the volume or double integral is exactly equal to f e <f> (y) dy =£ (f a f(x, y) dx) dy, which is the statement made above. A similar discussion makes it equally plausible that the statement ! a (X -f( x > y) dy ) dx = SL ^ x ' y) dr is also true. 492 FUNCTIONS OF SEVERAL VARIABLES [Chap. 3. Examples and Remarks. A few examples will serve to illustrate how this theorem may be used to evaluate double integrals. For the function u = f{x, y) = x*y, 0^icgl,0^yg2,we have -/ •'O r-1 1 I 1 1 2a?dx = „«* = . 2 lo 2 The above example belongs to a general class of functions whose integration is often simplified by the following theorem: If the function u = f(x, y), a 5£ x ^ b, c ^ y ^ d, can be represented as the product of a function of x alone, and a function of y alone, f(x,y)=4>{x)tj,{y), then the double integral of f is the product of two simple integrals: ff t f{x, y) dr = (J J (x) dx) (jf V (y) dy) . For on integration with respect to y the function <f>(x) can be treated as a constant and placed in front of the integral sign, while, on integration with respect to a, / >p(y)dy is a constant; hence f a \f e H*)'l> (y) dy) dx =f a \4> (x) £ tfi (y) dy) dx = (fj(y)dy)(fj(,x)dx). For the function u = sin (x + y), Oga;^ tt/2, Ss y Sa Jt/2, we have /• r /-t/2 / /.ir/2 \ / / sin(a; + y)dr = ( / sin(a; + y)dy)dx — / ( — cosf x + -) + cosa;J dx = I (siax + cosx)dx IW2 = ( — cosa; + sins) =1 + 1 = 2. lo Again, let us calculate the volume V of the vertical prism whose base in the zy-plane is bounded by the co-ordinate axes and the line x + y = 1, and which lie? below the plane u = 2x + Sy. We first extend the func- X] MULTIPLE AND REPEATED INTEGRALS 493 tion u = f(x, y) to the square ^i^l.Ogygl by equating it to outside the triangle (the base of the prism). Then for each x in the inter- val the function f(x, y) is different from for ^ y ^ 1 — x only; hence f /(*. V)dy = f l(x, y)dy = f (L + 3y)dy Jo Jo Jo = 2s(l - x) + I (1 - x? ^ _ * + ?, ^ F = !Ij (X > V)dT =/ 1 (- i" " « + D* " f The device just used is capable of extension to any function u = f{x, y) which is defined in a region R bounded above and below by curves y= ip(x) and y = <j>(x). For suppose R is de- fined by the inequalities a^x<^b, </>(x) <i y 5S *p(x). We mark off a rectangle R', a^x-^b, c<ly ^d, completely containing R, and outside R we put/= 0. Then / /(*. y) fy = 1 , /(*» y) d y for every x in the interval a ^ x ^ 6, so that ff B /(*, */) * =/£ /(a?, «/) dr = £* (fj(x, y) dy) dx = J. Uiwf&'yWy)**- Thus to find the volume V of the ellipsoid — +^+"—1 = a* ft 2 c 2 ' we notice that -F is the volume under u — f(x, y) = c . / ( 1 — * — — Y 2 v \ a 2 6V this function f(x, y) being defined only inside an ellipse 3+S" - -»V('-3 s 'W0-S)— *■** Calculating the repeated integral we first have / nx,y)dy=l c v( 1- 5-y* l ~(t, bx *\ y , cj/ //, a; 2 „2\|+»^(l-»»/a>) = — -c( 6 — — larc cos—, (- _2 . / [ 1 — *_ i 2\ « 2 / V(6'-W/o') T 2Vl a 2 6«/L 6 va-«'/««) --|('-5)<»-")+»=?'('-S> 494 FUNCTIONS OF SEVERAL VARIABLES [Chap. Proceeding with the integration, we have J F -J!:(j!>'»*)*-£i*( 1 -3*-?(-»)L 2tt = - cb [ - a ) 2 \S J ' abc, 3 so that V = - nabc. 3 4. Polar Co-ordinates. In our definition of the double integral, the subdivision into rectangles was of course chosen simply because such a subdi- vision is most convenient in connexion with rectangular co- ordinates. As we already know, however, there are many applica- tions in which polar co-ordinates are much more suitable than rectangular. If we are considering a function f(p, fa where p and </> are polar co-ordinates, the most convenient subdivision is not into rectangles, but into regions bounded by arcs of circles p = constant and radii j> — constant. Suppose, then, that our function f(p, fa is defined in such a region R, specified by the inequalities a^.p<Lb, a ^ <f> <* ft. (If f( p , fa is originally defined in a region R' not of this type, we enclose R' in a larger region R of the desired form and put f(p, fa = outside R'.) Then, just as on p. 486, we can insert points of subdivision p = a, Pi> P2> • • • j Pn = b, <f> = a, fa, fa, . . . , <j> m = ft and construct the corresponding radii and arcs of circles, thus dividing R into regions R {i , of area AR H . In each R ti we choose a point (p if , <f> ti ) and form the sum 2/(p ,<^) AR ti , and then let m and w increase without limit. Then the sum will again tend to the volume under the surface u =f(p, fa, and we may denote this by the integral flf(p,<p)dr. So far we have encountered nothing essentially new. The point of importance is to learn how to evaluate these integrals by reducing them either to repeated integrals or to integrals in terms of rectangular co-ordinates. For this purpose we mark off a pair of rectangular axes in a new plane, the p^-plane, and call them the />-axis and the ^-axis respectively. Correspond- ing to the point in R with polar co-ordinates p, <f> we plot the X] MULTIPLE AND REPEATED INTEGRALS 495 point in the p^-plane with rectangular co-ordinates p, <f>. Thus the region R, a 5S pf^b, a^<f><L ft is represented in the /j^-plane by a rectangle R', a<L p^b, a <S <£ <S £, and each small region R i} , p t _ x <; p <S p<, <p t _^ <S <£<:<£, is represented by a small rectangle R i} '. But the area AR it ' of the rectangle R t / is not the same as the area AR i} of R t) . The relation between them is easily found. The area A.R*/ is simply (4>j ~~ <Pi-i) (.Pi — Pi-i)> while the area AR ti is given by the formula =Upi+ pi-i) {& — <t> f -i) (Pi - Pi-!) =Upt+ Pi-d&Ru'- In each region R ti let us now choose the point p ( = %(p t + p^, <f>i — i(<f>i + <f>i-i)- Then by definition fjnp, fldr = lim S/fo, ^kRi,. But 2/fo, £,)AB, = Z/fe, ^)p t Ai?,./, and the latter expression is just the sum which we form in de- fining the double integral of the function f(p, </>)p over the rect- angle R' in the p^-plane. Hence as the fineness of the subdivision increases the sum approaches this integral and ffj(p, <p)dr=ffj{ P> <f>)pdr'=ffj-(p, <p)pdpd<p = L (fj {p ' *)/"*#) ^=jf(j[/(/>> <f>)pd P y<f>. As an example, let us calculate the volume V of the sphere of radius a. The upper hemisphere is given by the equation u = V(o a — p 2 ), Ogp^o, g 9 S 2tt. Thus i r <(j:^'-*»<>>H:(-i<*-<-nh 1 r 2 " S j 2jta 3 "3"' so that V = - 7ta 3 . 3 496 FUNCTIONS OF SEVERAL VARIABLES [Chap- /oo e~ x *dx. -00 The formulae of the preceding sub-section enable us to calculate the area under the curve y = e~ **, — oo < x < oo, which frequently occurs in the theory of probability. This integration is especially interesting, in that we can evaluate the definite integral from — oo to oo of a function for which we cannot find a primitive function or an indefinite integral at all. Let us first consider the integral /„ of the function e — <*'+3 , " ) = e~ p " over the circle 2a p f^ a. This is given by 7 « = r (f ■"■' ^ = rs - 1 •-*)* - mi - e ~ ai) - The square — o ^ x 5j a, — a £= j/ 5j a contains the circle ^ p ti a and is contained in the circle ^ p ^ 2a, and the integrand e~ x '~ y is everywhere positive; hence tc(1 - «-«*) = /„ g ("" (T e-*'-y'dy)dx g J 2o = jt(l - «-*"'). The integral can be written in the form f" e ~ x '( f e ~ y ' d y) dx = ( f e-*'<te) , hence rt(l — e-°") ^( f e-*'dx\ g tc(1 - e~ ia '). If we now let a increase without limit, this gives the equation e- x 'dx— Vjt, /: and our integral is evaluated. 6. Moments and Centre of Mass; Moments of Inertia. In Chap. V, § 2 (p. 283) we saw that the moment of a system of points P ls P 2 , . . . , P„ with co-ordinates (x v yj, (x 2 , y 2 ), . . . , (x n , y n ) and masses m^, m 2 , . . . , m n about the z-axis is given by n £ m t yi, and that the ordinate of its centre of mass is given »— 1 by the equation -in n r) = ■=--- S m„y p , where M= S m„ M. v =\ »=i with analogous expressions for the moment about the y-axis and the abscissa of the centre of mass. We now extend these X] MULTIPLE AND REPEATED INTEGRALS 497 ideas to masses distributed uniformly over a region R. We suppose that a mass is distributed with density 1 over the region R; that is, that each portion of R with area A.R has also mass AjR. Then the total mass M of R is the same as the area of R, M= f [dr. -If.' Let us now divide R into portions R l , . . . , R„ with areas Ai^, . . . , AR„, and in each portion R v choose a point (£„, ij„). If we imagine that the total mass A2J K of the portion R v is con- centrated at the point (f„, tj v ), the moment of the resulting system of points with respect to the as-axis will be Si^AB,,, and the ordinate of the centre of mass will be Xr, v AR v = S^AR„ SAfl, M If we now let n -> 00 and let the diameter of the greatest R„ tend to 0, these sums tend to the integrals .= //**. V = respectively. These expressions we take as the definitions of the moment T x of R about the a;-axis and of the ordinate ij of its centre of mass. Similarly, the moment about the y-axis and the abscissa g of the centre of mass are respectively given by T^flxdr, eJ-lj^. where M-ff*. For example, the moment of the semicircle R, —p^x^p, £s y ^ V (p a — 3?), about the z-axis is and since M = I I dr= area M — - up*, ,, (E798) 498 FUNCTIONS OF SEVERAL VARIABLES [Chap. By a similar argument, starting from the definition of the moment of inertia I x of a system of particles, I x =~Lm v y*, we arrive at the expression for the moment of inertia of the region R about the o>axis, and similarly we obtain the moment of inertia with respect to the y-axis, II x 2 dr. Analogous formulae hold for three-dimensional regions R; the co-ordinates £, 17, £ of the centre of mass are given by fffxdr fffydr f f 7 ' zdr S- m ' v== M ' 4= M * where M = J f J \dr = volume of R. To find the moments of inertia l x , l y , I z of R about the x, y, z-axes respectively, we must remember that the distance of the point (x, y, z) from the ataxia is V(2/ 2 + z2 )> hence for a system of particles the moment of inertia about the aj-axis is Sffl»V(y, 2 + 2, 2 ) 8 = Srojy, 2 -)- a, 2 ), and on dividing R into sub-regions and passing to the limit as before we obtain the formula I x = fffy + z*)dr. Similarly Z„ = fjj (x* + z 2 ) dr, I z = fff B (x* + y*)dr. Thus the moment of inertia of the cube — h^x^h, — h<.y ^h, —h^z^h about the z-axis is I z=f_ {f (f^i^ + y^dz^dy^dx = f if 2ft(a~>-fy a )<fyjefe= f 2A (2*% + - hA dx -h yj-h ) J-h X] MULTIPLE AND REPEATED INTEGRALS 499 The significance of the moment of inertia, as we have already remarked in Chap. V (p. 286), lies in the fact that in rotatory motion it plays the part taken by the mass in translatory motion. For example, if the region R rotates about the s-axis with angular velocity w, its kinetic energy is ^I x <» 2 - This, however, is not the only application of the concept of moment of inertia; for example, it is also important in structural engineering, where it is found that the stiffness of a beam of a given material is proportional to the moment of inertia of the cross-section taken about a line through its centre of mass. The reader will find further information about this in any textbook on strength of materials. 7. Further Applications. The student should not assume that the applications already discussed exhaust the possibilities of the double integral. For instance, we have not proved the important theorem that the area A of the surface z=f(x, y), where {x, y) is in R, is given by the integral A /a i+ G9 ,+ G0> provided — and -^ are continuous; and we have left many other dx ay interesting fields untouched. These further developments, how- ever, do not come within the scope of the present book and must be left for Vol. II. Examples 1. Perform the following integrations (a) f f xy(x*—y*)dydx. Jo Jo (6) / / ooa(x + y)dydx. Jo Jo c r 2 i (c) / / — dydx. J\ Ji xy (d) f f xe*"dydx. Jo Jo 500 FUNCTIONS OF SEVERAL VARIABLES [Chap. X JO •'0 Vl-x' y 2 dy dx. ' / ydydx. •'o 2. Find the volume between the aa/-plane and the paraboloid z = 2 — s 2 — y 2 . 3. Find the volume common to the two cylinders je 2 + 2 2 = 1 and S/ 2 + z 2 = 1. 4. By integration, find the volume of the smaller of the two portions into which a sphere of radius r is cut by a plane whose perpendicular dis- tance from the centre is h ( < r). 6. For the following figures find the area, the centre of gravity, the moments about the x- and t/-axes, and the moments of inertia about the x- and y-axes: (o) the semicircle ^ y ^ V(r 2 — x 2 ); (b) the rectangle gi^o, ^y f^b; (c) the rectangle — a ^i^n, — 6 ^ y :£ 6; (d) the ellipse | y | ^ 6 ^^1 - 5|); (e) the triangle with vertices (0, 0), (a, 0), (0, 6): 6. For the following figures find the volume, the centre of gravity, and the moments of inertia about the x-, y-, and z-axes: (a) the parallelepiped O^x^a, 0<^y?£b, Of^zSic; (6) the hemisphere Ogzg V(a 2 — a? — t/ 2 ); (c) the triangular prism with vertices (0, 0, 0), (a, 0, 0), (0, 6, 0), (0, 0, c). CHAPTER XI The Differential Equations for the Simplest Types of Vibration On several occasions we have already met with differential equations, that is, equations from which an unknown function is to be determined and which involve not only this function itself but also its derivatives. The simplest problem of this type is that of finding the inde- finite integral of a given function/(a;). This problem requires us to find a function y = F(x) which satisfies the differential equation y> —f(x) = 0. Further, we solved a problem of the same type in Chap. Ill, § 7 (p. 178), where we showed that an equation of the form y' = ay is satisfied by an exponential function y = ce°*. As we saw in Chap. V (p. 294), differential equations arise in connexion with the problems of mechanics, and indeed many branches of pure mathematics and most of applied mathematics depend on differential equations. In this chapter, without going into the general theory, we shall consider the differential equations of the simplest types of vibration. These are not only of theoretical value, but are also extremely important in applied mathematics. It will be convenient to bear the following general ideas and definitions in mind. By a solution of a differential equation we mean a function which, when substituted in the differential equation, satisfies the equation for all values of the inde- pendent variable that are being considered. Instead of solution the term integral is often used: in the first place because the problem is more or less a generalization of the ordinary problem of integration; and in the second place because it frequently happens that the solution is actually found by integration. 601 502 DIFFERENTIAL EQUATIONS [Chap. 1. Vibration Problems of Mechanics and Physics 1. The Simplest Mechanical Vibrations. The simplest type of mechanical vibration has already been considered in Chap. V, § 4 (p. 295). We there considered a particle of mass m which is free to move on the ataxia and which is brought back to its initial position x — by a restoring force. The magnitude of this restoring force we took to be proportional to the displacement x; in fact, we equated it to — he, where &is a positive constant and the negative sign expresses the fact that the force is always directed towards the origin. We shall now assume that there is a frictional force present also and that this frictional force is proportional to the velocity dx/dt = x of the particle and opposed to it. This force is then given by an expres- sion of the form — rx, with a positive frictional constant r. Finally, we shall assume that the particle is also acted on by an external force which is a function f(t) of the time t. Then by Newton's fundamental law the product of the mass m and the acceleration x must be equal to the total force, that is, the elastic force plus the frictional force plus the external force. This is expressed by the equation mx + rx + he = f(t). This equation determines the motion of the particle. If we recall the previous examples of differential equations, such dx as the integration problem x=-—=f(t) with its solution at x = J f(t) dt-{- c, or the solution of the particular differential equation mx + he — on p. 296, we observe that these problems have an infinite number of different solutions. Here too we shall find that there are an infinite number of solutions, which are expressed in the following way. It is possible to find a general solution or complete integral x(t) of the differential equation, depending not only on the independent variable t, but also on two parameters Cj and c 2 , called the constants of integration. If we assign special values to these constants, we obtain a particular solution, and every solution can be found by assigning special values to these constants. The complete integral is then the totality of all particular solutions. XI] THEORY OF VIBRATIONS 5°3 This fact is quite understandable (of. also Chap. V, § 4, p. 298). We cannot expect that the differential equation alone will determine the motion completely. On the contrary, it is plausible that at a given instant, say at the time t = 0, we should be able to choose the initial position s(0) = x„ and the initial velocity £(0) =* * ( m short, the initial state) arbitrarily; in other words, at time t = we should be able to start the particle from any initial position with any velocity. This being done, we may expect the rest of the motion to be definitely determined. The two arbitrary constants (^ and c a in the general solution are just enough to enable us to select the particular solution which fits these initial con- ditions. In the next section (p. 508) we shall see that this can be done in one way only. If no external force is present, that is, if f(t) — 0, the motion is called a free motion. The differential equation is then said to be homogeneous. If f(t) is not equal to zero for all values of t, we say that the motion is forced and that the differential equa- tion is non-homogeneous. The term f(t) is also occasionally referred to as the perturbation term. 2. Electrical Oscillations. A mechanical system of the simple type described can actually be realized only approximately. An approximation is offered by the pendulum, provided its oscillations are small. The oscil- lations of a magnetic needle, the oscillations of the centre of a telephone or microphone diaphragm, and other mechanical vibrations can be represented to within a certain degree ,uP of accuracy by systems such as we have ? described. But there is another type of *• £) ' phenomenon which corresponds far more <}>(*) exactly to our differential equation. This is 'fiectrica^clrcuft 7 the oscillatory electrical circuit. We consider the circuit sketched in fig. 1, having inductance (i, resistance p and capacity C = 1/k. We also suppose that the circuit is acted upon by an external electromotive force <f>(t) which is known as a function of the time t, such as the voltage supplied by a dynamo or the voltage due to electric waves. In order to describe the process taking place in the circuit we denote the voltage across the condenser by E and the charge in the condenser by Q. These quantities are then connected by the equation CE = E\k = Q. The current I, which like the voltage n 5o 4 DIFFERENTIAL EQUATIONS [Chap E is a function of the time, is defined as the rate of change of the charge per unit time, that is, as the rate at which the charge on the condenser diminishes: / = — Q ■= — dQ/dt = — EJk. Ohm's law states that the product of the current and the resis- tance is equal to the electromotive force (voltage); that is, it is equal to the condenser voltage E minus the counter electro- motive force due to self-induction plus the external electromotive force <f>(t). We thus arrive at the equation Ip = E — /*./ + <f>(t) ot—?-E=E + ?E+ 4>{t), that is, fiE + p E + kE= —Kd>(t), K K which is satisfied by the voltage in the circuit. We see, therefore, that we have obtained a differential equation of exactly the type considered in No. 1 (p. 502). Instead of the mass we have the inductance, instead of the frictional force the resistance, and instead of the elastic constant the reciprocal of the capacity, while the external electromotive force (apart from a constant factor) corresponds to the external force. If the electromotive force is zero, the differential equation is homogeneous. If we multiply both sides of the differential equation by — 1/k and differentiate with respect to the time, we obtain for the current / the corresponding equation id + /J + kI = 4(t), which differs from the equation for the voltage on the right- hand side only, and for free oscillations (<f> = 0) has identically the same form. 2. Solution of the Homogeneous Equation. Free Oscillations 1. The Formal Solution. We can easily obtain a solution of the homogeneous equation mx -+- rib + he = on p. 502 in the form of an exponential ex- pression, by seeking to determine a constant A in such a way that the expression e xt = x is a solution. If we substitute this and its derivatives x = Ae* 1 , x = A V in the differential equation XI] THE HOMOGENEOUS EQUATION 505 and remove the common factor e*', we obtain the quadratic equation mA 2 4- rX + k = for A. The roots of this equation are Each of the two expressions x = e* 1 ' and x = e^'* is, at least formally, a particular solution of the differential equation, as we see by carrying out the calculations in the reverse direction. Three different cases can now occur. 1. r 2 — Amk > 0. The two roots X t and Aj are then real, negative, and unequal, and we have two solutions of the differen- tial equation, x = v^ = e**' and x = u% == e x »*. With the help of these two solutions we can at once construct a solution in which two arbitrary constants are present. For on differentiation we see that x = c^ + c^ is also a solution of the differential equation. On p. 508 we shall show that this expression is in fact the most general solution of the equation; that is, that we can obtain every solution of the equation by substituting suitable numerical values for <\ and Cj. 2. r 2 — imk = 0. The quadratic equation has a double root. Thus to begin with we have, apart from a constant factor, only the one solution x = w l = e~" t2m . But we easily verify that in this case the function x— w 2 — te~ nt2m is also a solution of the differential equation.* For we find that x=(l- — t) e~ rti2m , x = (— t - -\ e-"' 2 ", \ 2m I \4m 2 m) and by substitution we see that the differential equation r 2 mx+ rx-\- — x=mx+rx-\-kx—0 im * We are led to this solution naturally by the following limiting process: if X 1 4= A 2 , then the expression (e*i* — «*•«)/( A, — Aj) also represents a solution. If we now let \ t tend to A, and write A instead of A x , A„ our expression becomes * e \t „ U M aA IT (1798) «;o6 DIFFERENTIAL EQUATIONS [Chap. is satisfied. Then the expression x = c 1 e- rtlim + c 2 fe- r " 2m again gives us a solution of the differential equation with two arbitrary constants of integration c± and c 2 . 3. r 2 — 4mk < 0. We put r 2 — 4mk = — 4m 2 v 2 and obtain two solutions of the differential equation in complex form, given by the expressions x— u^= e ~ rtl2m + irt and x= w 2 = e -r " 2 ' B-, ''". Euler's formula e ±,v ' = cos vt ii sin vt gives us for the real and imaginary parts of the complex solution «!, on the one hand the expressions v x = e ~ rtl!!m cos vt, v 2 = e ~ rt l 2m sin vt, and on the other hand the representation _ "i+m 2 _ _ %— M 2 Vl 2~" ' " 2 2T" From the second form of representation we see that v t and v 2 are (real) solutions of the differential equation. To verify this directly by differentiation and substitution forms a simple but valuable exercise. From our two particular solutions we can again form a general solution x = CjV x -J- <h% = ("l cos vt + c 2 sin vt) e ~ rtl2m with two arbitrary constants q and c 2 . This may also be written in the form x = ae~ rtl2m cos v(f — 8), where we have put c^ — a cosvS, c 2 = a siixvS, and a, S are two new constants. We recall that we have already come across this solution for the special case r = (Chap. V, § 4, p. 296). 2. Physical Interpretation of the Solution. In the two cases r > 2-\/mk and r = 2\/mk the solution is given by the exponential curve, or by the graph of the function te~ nlZm , which for large values of t resembles the exponential XI] THE HOMOGENEOUS EQUATION 507 curve, or by the superposition of such curves. In these cases the process is aperiodic; that is, as the time increases the " dis- tance " x approaches the value asymptotically, without oscil- lating about the value x = 0. The motion, therefore, is not oscillatory. The effect of friction or damping is so great that it prevents the elastic force from setting up oscillatory motions. It is quite different in the case r < 2^/mk, where the damping is so small that complex roots \, \ occur. The expression x= a cos v{t — 8)e~ rtl2m here gives us damped harmonic oscillations. These are oscillations which follow the sine law and have the circular frequency v ~ \ { 7~i)' ^ u * wnose amplitude, instead of being constant, is given by the expression ae~ rtlim . That is, the amplitude ciiminishes exponentially; the greater the expression r/2m is, the faster is the rate of decrease. In physical literature this damping factor is frequently called the logarithmic decrement of the damped oscilla- tion, the term indicating that the logarithm of the amplitude de- '[..^..((.^-^ creases at the rate r/2m. A Fie . 2 .-Dam P ed harmonic dilutions damped oscillation of this kind is illustrated in fig. 2. As before, we call the quantity T = 2n/v the period of the oscillation and the quantity v8 the phase displacement. For the special case r = we again obtain simple harmonic oscillations with the frequency v = Vk/m, the natural frequency of the undamped oscillatory system. 3. Fulfilment of Given Initial Conditions. Uniqueness of the Solution. We have still to show that the solution with the two constants Cj and c 2 can be made to fit any pre-assigned initial state, and also that it repre- sents all the possible solutions of the equation. Suppose that we have to find a solution which at time t = satisfies the initial conditions x(Q) = x w sc(0) = *o> where the numbers x Q and aS can have any values. Then in case 1 on p. 605 we must put ^1 "T c 2 — x o> 508 DIFFERENTIAL EQUATIONS [Chap. Foi the constants Cj and c 2 we accordingly have two linear equations, and these have the unique solutions Aj Aj A2 Aj In case 2 (p. 505) the same process gives the two linear equations *, + *-*, (x— £) from which Cj and c 2 can again be uniquely determined. Finally, in case 3 (p. 506) the equations determining the constants take the form a cos vS = x , with the solutions a( v sinvS cosvS ) = a! , \ 2m I \ ar ° oos ?- ° - \ V( vV + (*° + £ *»)'}• Thus we have shown that the general solution can be made to fit any arbitrary initial conditions. We have still to show that there is no other solution. For this we need only show that for a given initial state there can never be two different solutions. If two such solutions u(t) and »(t) existed, for which u(0) = x , ii(0) = x and t>(0) = x , v(0) = x w then their difference w = u — v would also be a solution of the differential equation, and we should have w(fi) = 0, j£(0) = 0. This solution would therefore correspond to an initial state of rest, that is, to a state in which at time t = the particle is in its position of rest and has zero velocity. We must show that it can never set itself in motion. To do this we multiply both sides of the differential equation mw 4- rw + lew = by 2u> and recall that 2wib = - to 2 and 2ww = — to*. dt dt We thus obtain t (mut) + i. (kw*) + 2nb* = 0. dt dt If we integrate between the instants t = and t = t and use the initial conditions w(0 ) = 0, w(0) = 0, we have r T /dw\" miv'iy) + kw*(t) + 2r {— J dt = 0. This equation, however, would yield a contradiction if at any time t > the function to were different from 0. For then the left-hand side of the equation would be positive, since we have taken m, h and r to be positive, while the right-hand side is zero. Hence to = u — v is always equal to 0, which proves that the solution is unique. XI] THE HOMOGENEOUS EQUATION 509 Examples For the equations in Ex. 1-5 find the general solution, and also the solution for which «(0) = 0, x(0) = 1: 1. * — Zx + 2x = 0. 2. x + 3x + 2x = 0. 3. 2x + x — x = 0. 4. £ + 4x + 4x = 0. 6. 4x + 4* + x = 0. 6. Find the general solution, and also the solution for which x(0) = 0, sr(0) = 1, of the equation x + x + x = 0. Determine the frequency (v), the period (T), the amplitude (a), and the phase (S) of the solution. 7. Find the solution of 2x + 2x + x = for which a;(0) = 1, ac(0) = — 1. Calculate the amplitude (a), the phase (8), and the frequency (v) of the solution. 3. The Non-homogeneous Equation. Forced Oscillations 1. General Remarks. Before proceeding to the solution of the problem when an external force f(t) is present, that is, to the solution of the non-homogeneous equation, we make the following remark. If w and v are two solutions of the non-homogeneous equation, the difference u = w — v satisfies the homogeneous equation; this we see at once by substitution. Conversely, if u is a solution of the homogeneous equation and v a solution of the non- homogeneous equation, then w = u + v is also a solution of the non-homogeneous equation. Therefore from one solution * of the non-homogeneous equation we obtain all its solutions by adding the complete integral of the homogeneous equation, f We therefore need only find a single solution of the non-homo- geneous equation. Physically this means that if we have a forced oscillation due to an external force, and on it superpose an arbi- trary free oscillation, represented by a solution of the homo- • Often called the particular integral. t Often called the complementary function. 510 DIFFERENTIAL EQUATIONS [Chap. geneous equation, we obtain a phenomenon which satisfies the same non-homogeneous equation as the original forced oscilla- tion. If a frictional force is present, the free motion in the case of oscillatory motion will fade out as time goes on, because of the damping factor e~ rtl2m . Hence for a given forced vibration with friction it is immaterial what free vibration we superpose; the motion will always tend to the same final state as time goes on. Secondly, we notice that the effect of a force /(«) can be split up in the same way as the force itself. By this we mean the following: if /$), f 2 (t), and f(t) are three functions such that and if x 1 = x^t) is a solution of the differential equation mx + rx + lex =f 1 (t) and x 2 = cc 2 (£) is a solution of the equation mx+rx+ kx=f 2 {t) then x(t) = x^t) + x 2 (t) is a solution of the differential equation mx + rx + kx = f(t). A correspond- ing statement of course holds if f(t) consists of any number of terms. This simple but important fact is called the " principle of superposition ". The proof follows from a glance at the equation itself. By subdividing the function f(t) into two or more terms we can thus split the differential equation into several equations, which in certain circumstances may be easier to manipulate. The most important case is that of a periodic external force f(t). Such a periodic external force can be resolved into purely periodic components by expansion in a Fourier series, and can therefore * be approximated to as closely as we please by a sum of a finite number of purely periodic functions. It is therefore sufficient to find the solution of the differential equation subject to the assumption that the right-hand side has the form o cos cot or b sin cot, where a, b, and at are arbitrary constants. Instead of working with these trigonometric functions, we can obtain the solution more simply and neatly if we use complex notation. We put f(t) = ce'"', and the principle of superposition shows that we need only consider the differential equation mx + rx -+- hx = ce*"*, * Provided that it is continuous and seotionally smooth (p. 439), which is the only case of importance in physios. XI] THE NON-HOMOGENEOUS EQUATION 511 where by c we mean an arbitrary real or complex constant. Such a differential equation actually represents two real differential equations. For if we split the right-hand side into two terms, if e.g. we take = 1 and write e"" = cosarf + t ainwt, then %i and x 2 , the solutions of the two real differential equations mx + rx + fee = co&cot and mx + rx + kx = sinarf, combine to form the solution x = Xy + ix 2 of the complex differential equation. Conversely, if we first solve the differential equation in complex form, the real part of the solution gives ub the function a^ and the imaginary part the function x 2 . 2. Solution of the Non-homogeneous Equation. We solve the equation mx + rx + Tex = ce' w by a device naturally suggested by intuition. We assume that c is real and (for the time being) that r 4= 0. We now make the guess that a motion will exist which has the same rhythm as the periodic external force, and we accordingly attempt to find a solution of the differential equation in the form x = ae ia \ where we have only to determine the factor a, which is indepen- dent of the time. If we substitute this expression and its deriva- tives x — taxxe*"", x = — to W*" in the differential equation and remove the common factor e"°' we obtain the equation — mafia -f- irwa + kcr = or — ma? -f irai -f- k Conversely, we see that for this value of a the expression ae* mi is actually a solution of the differential equation. To express the meaning of this result clearly, however, we must perform a few transformations. We begin by writing the complex factor a in the form k — ma? — ira> ... ° *" c n »«■ 22 = cae ' (« — mut^Y + r 2 a> 2 where the positive " distortion factor " a and the " phase dis- 512 DIFFERENTIAL EQUATIONS [Chap. placement " to8 are expressed in terms of the given quantities m, r f h, by the equations a 2 = -., sinajS = rcoa, coswS = (k — mo) 2 )a. (k — mw 2 ) 2 + r 2 ca 2 With this notation our solution takes the form and the meaning of the result is as follows: to the force c coswi there corresponds the " effect " ca cosa){t — 8), and to the force c smcot corresponds the effect ca sina>(£ — 8). Hence we see that the effect is a function of the same type as the force, that is, an undamped oscillation. This oscillation differs from the oscillation representing the force in that the amplitude is increased in the ratio a : 1 and the phase is altered by the angle cuS. Of course it is easy to obtain the same result without using the complex notation, but at the cost of somewhat longer calculations. According to the remark at the beginning of this section (p. 509), by finding this one solution we have completely solved the problem; for by superposing any free oscillation we can obtain the most general forced oscillation. Collecting the results, we have the following: The complete integral of the differential equation mx + rx + kx = ce*"' (where x =)= 0) is x = cae'™ (t_!) + u, where u is the complete integral of the homogeneous equation mx + r± + kx = and the quantities a and S are defined by the equations 1 (k — ma) 2 ) 2 + r 2 co 2 ' sincuS = rioa, coscoS = (k — mio 2 )a. The constants in this general solution leave us the pos- sibUity of making the solution suit an arbitrary initial state, that is, for arbitrarily assigned values of x and x the constants can be chosen in such a way that a;(0) = x and x(0) = x . 3. The Resonance Curve. In order to acquire a grasp of the solution which we have obtained and of its significance in applications, we shall study XI] THE NON-HOMOGENEOUS EQUATION 513 the distortion factor a as a function of the " exciting frequency " oj, that is, the function ^ W ' = V(k — mw 2 ) 2 + rW* The reason for this detailed investigation is that for given con- stants k, m, r, or as we say for a given " oscillatory system ", we can think of the system as being acted on by periodic ex- citing forces of very different circular frequencies, and it is important to consider the solution of the differential equation for these widely different exciting forces. In order to describe the function conveniently we introduce the quantity a> = -y/k/m. This number a> is the circular frequency which the system would have for free oscillations if the friction r were zero; or, briefly, the natural frequency of the undamped system (cf. p. 507). The actual frequency of the free system, owing to the friction r, is not equal to co , but is instead V \m 4m 2 / where we assume that ikm — r 2 > 0. (If this is not the case the free system has no frequency; it is aperiodic.) The function <f>(a)) tends asymptotically to the value as the exciting frequency tends to infinity, and, in fact, it vanishes to the order 1/w 2 . Further, <f>(0) = 1/k; in other words, an exciting force of frequency zero and magnitude 1, that is, a constant force of magnitude 1, gives rise to a displacement of the oscil- latory system amounting to 1/k. In the region of positive values of a) the derivative <f>'(u>) cannot vanish except where the deri- vative of the expression (k — mco 2 ) 2 -f- rW vanishes, that is, for a value a> = oy l > for which the equation — 4mo)(k — mio 2 ) + 2r 2 cu = holds. In order that such a value may exist we must obviously have 2km — r 2 > 0; in this case y\m 2m 2 / V\ ° 2 2m 2 / * Since the function <f>(to) is positive everywhere, increases mono- tonically for small values of <a, and vanishes at infinity, this SH DIFFERENTIAL EQUATIONS [Chap. value must give a maximum. We call the circular frequency a»! the " resonance frequency " of the system. By substituting this expression for w x we find that the value of the maximum is V v» 4m 2 / As r -»• 0, this value increases beyond all bounds. For r — 0, that is, for an undamped oscillatory system, the function <f>(a>) has an infinite discontinuity at the value a> = w v This is a limit- ing case to which we shall give special consideration later. The graph of the function <f>(ao) is called the resonance curve of the system. The fact that for to = o> x (and consequently for small values of r in the neighbourhood of the natural frequency) the distortion of amplitude a = <f>(w) is particularly large is the mathematical expression of the " phenomenon of resonance ", which for fixed values of m and k is more and more evident as r becomes smaller and smaller. In fig. 3 we have sketched a family of resonance curves, all correspond- ing to the values m = 1 and k = 1, and consequently to to„ = 1, but with different values of D = $r. We see that for small values of D well-marked resonance occurs near co = 1; in the limiting case D = there would be an infinite discontinuity of <p(o>) at <o = 1, instead of a maximum. As D increases the maxima move towards the left, and for the value D = 1/V2 we have ^ — 0. In this last case the point where the tangent is horizontal has moved to the origin, and the maximum has disappeared. If D > 1/V2 there is no zero of <p'(to); the resonance ourve no longer has a maximum, and resonance no longer occurs. In general, the resonance phenomenon ceases as soon as the condition 21cm — r 2 ^ becomes true. In the case of the equality sign, the resonance curve reaches its greatest height <£(0) = 1/k at to-^— 0; its tangent is horizontal there, and after an initial course which is almost horizontal it diminishes towards zero. 1. Further Discussion of tbe Oscillation. We cannot, however, rest content with the above discussion. In order that we may really understand the phenomenon of forced XI] THE NON-HOMOGENEOUS EQUATION 5*5 motion an additional point requires to be emphasized. The particular integral cae"' ( ' -4) is to be regarded as a limiting state which the complete integral x(t) = cae , '"<'-*> +c 1 u 1 + c 2 u 2 approaches more and more closely as time goes on, since the free oscillation c^ + c^u^ superposed on the particular integral fades away with the passage of time. This fading away will take place slowly if r is small, rapidly if r is large. ]«_ 2Z it fZ-0 1-9 is 17 1-6 1-5 H t '' 3 X. ,., | 1-0 C) 0-9 0-8 on 0-6 0-5 04 0-3 0-2 0-1 OS 1-0 IS 20 Exciting Frequency — *■ (»> Fig. 3.— Resonance curves Let us suppose, for example, that at the beginning of the motion, i.e. at time t = 0, the system is at rest, so that x(0) = and x(0) = 0. From this we can determine the constants q and c^ and we see at once that they are not both zero. Even when the exciting frequency is approximately or exactly equal to w l5 so that resonance occurs, the relatively large amplitude a = ^(a^) will not at first appear. On the contrary, it will be masked by the function c^ + c^u^, and will first make its appearance when this function fades away; that is, it will appear more slowly the smaller r is. lr 1 D-Oj /D-lf \ \P-° \\ / \\ J '/ \\ - ^^^ ^D-i \\ ^;& \\ \ X \i>-' \ - \ \ sP-2 - \ ^^ ^■^ =8 ' 516 DIFFERENTIAL EQUATIONS [Chap. For the undamped system, that is, for r = 0, our solution fails when the exciting frequency is equal to the natural circular frequency a> = Vk/m, for then ^(a> ) is infinite. We therefore cannot obtain a solution of the equation mx -f- kx = e*"' in the form ere""'. We can, however, at once obtain a solution of the equation in the form crfe""'. If we substitute this expression in the differential equation, remembering that x = ae , '"'(l -f- iwt), x = ae'"'(2iw — tat 2 ), we have oiflimco — mcoH -f- kt) = 1, and, since mu? = k, 2imoi Thus when resonance occurs in an undamped system we have the solution t ,-..., t x = e* " = — -== e*" 2ima) 2iVkm Using real notation, when/(<) = coswt we have x= - — sinwf, * Vkm and when/(<) = Binwt we have 1 * x = — - cosarf. 2 Vffwi We thus see that we have found a function which may be referred to as an oscillation, but whose amplitude increases pro- portionally with the time. The superposed free oscillation does not fade away, since it is undamped; but it retains its original ampli- tude and becomes unimportant in comparison with the increasing amplitude of the special forced oscillation. The fact that in this case the solution oscillates backwards and forwards between positive and negative bounds which continually increase as time goes on represents the real meaning of the i n fin it e discon- tinuity of the resonance function in the case of an undamped system. XI] THE NON-HOMOGENEOUS EQUATION 517 5. Remarks on the Construction of Recording Instruments. In a great variety of applications in physios and engineering the dis- cussion in the previous sub-section is of the utmost importance. With many instruments, such as galvanometers, seismographs, oscillatory electrical circuits in radio receivers, and microphone diaphragms, the problem is to record an oscillatory displacement x due to an external periodic force. In such cases the quantity x satisfies our differential equation, at least to a first approximation. If T is the period of oscillation of the external periodio force, we can expand the force in a Fourier series of the form l_-«o or, better still, we can think of it as represented with sufficient accuracy by x a trigonometrio sum S y l efW"l' r ) t consisting of a finite number of terms 1- -j only. By the principle of superposition (p. 510), the solution x(t) of the differential equation, apart from the superposed free oscillation, will be represented by an infinite series * of the form x(t)= £ a,^ 2 *' 2 ")', I— as or approximately by a finite expression of the form *(*)= S <,, e tf(2W2X I s In virtue of our previous results <7,= - r ,a,e-''¥2*'/D and 1 27tZj. 2-Klr We can then describe the action of an arbitrary periodic external force in the following way: if we analyse the exciting force into purely periodio components, the individual terms of the Fourier series, then each com- ponent is subject to its own distortion of amplitude and phase displace- ment, and the separate effects are then superposed additively. If we are interested only in the distortion of amplitude (the phase displacement is only of secondary importance t in applications and, moreover, can be dis- cussed in the same way as the distortion of amplitude), a study of the resonance curve gives us complete information about the way in which * Questions of convergence will not be discussed here. \ Since e.g. it is imperceptible to the human ear. 518 DIFFERENTIAL EQUATIONS [Chap. the motions of the recording apparatus reproduce the external exciting force. For very large values ofZorto(=-=n the effect of the exciting frequenoy on the displacement x will be hardly peroeptible. On the other hand, all exciting frequencies in the neighbourhood of <o„ the (circular) resonance frequenoy, will markedly affect the quantity x. In the construction of physical measuring and recording apparatus the constants m, r, and k are at our disposal, at least within wide limits. These should be chosen so that the shape of the resonance curve is as well adapted as possible to the special requirements of the measurement in question. Here two considerations predominate. In the first place, it is desirable that the apparatus should be as sensitive as possible; that is, for all frequencies to in question the value of a. should be as large as possible. For small values of <•>, as we have seen, a is approximately proportional to \jk, so that the number 1/k is a measure of the sensitiveness of the instru- ment for small exciting frequencies. The sensitiveness can therefore be increased by increasing l/k, that is, by weakening the restoring force. The other important point is the necessity for relative freedom from dis- tortion. Let us assume that the representation f{t) = £ f^^wirtt jg ^ i- adequate approximation to the exciting force. We then say that the apparatus records the exciting force f(t) with relative freedom from dis- tortion if for all circular frequencies <■> :& N — the distortion factor has approximately the same value. This condition is indispensable if we wish to derive conclusions about the exciting process directly from the behaviour of the apparatus; if, for example, a gramophone or wireless set is to repro- duce both high and low musical notes with an approximately correct ratio of intensity. The requirement that the reproduction should be relatively " distortionless " can never be satisfied exactly, since no portion of the resonance curve is exactly horizontal. We can, however, attempt to choose the constants m, k, r, of the apparatus in such a way that no marked resonance occurs, and also in such a way that the curve has a horizontal tangent at the beginning, so that tp(to) = a remains approximately con- stant for small values of a. As we have learned above, we can do this by putting 2km — r 2 = 0. Given a constant m and a constant k, we can satisfy this requirement by adjusting the friction r properly, e.g. by inserting a properly chosen re- sistance in an electrical circuit. The resonance curve then shows us that from the frequency to circular frequencies near the natural circular frequency a> a of the undamped system the instrument is nearly distortionless, and that above this frequenoy the damping is considerable. We therefore obtain relative freedom from distortion in a given interval of frequencies by first choosing m so small and k so large that the natural circular frequency co of the undamped system is greater than XI] THE NON-HOMOGENEOUS EQUATION 519 any of the exciting circular frequencies under consideration, and then choosing a damping factor r in accordance with the equation 2km — r 2 = 0. Examples For the equations in Ex. 1-5 find the solution satisfying the initial conditions a;(0) = 0, sc(0) = 0. For equations 1-4 state also the amplitude, the phase, and the value of <o for which the amplitude is a maximum: 1 . x + 3a6 + 2x = coaat. 2. x + x + x = cos cot. 3. x + x + * = sinfcrf. 4. 2x + 2x + * = coscoi. 5. x + 4x + ix = costoi. 4. Additional Remarks on Differential Equations A more systematic study of differential equations is made in Volume II, Chapter VI. Here only a few additions to the preceding special theory will be given. 1. Homogeneous Linear Differential Equations of Order n with Constant Coefficients. More complicated vibration problems lead to a linear differen- tial equation for the unknown function x(t) of the independent variable, of the form where c^, . . . ,a„ are constants and n is a positive integer. We can solve this by a method similar to that for the case n = 2 (p. 504). Let x = c x '. If we substitute this function and its derivatives in the differential equation and remove the common factor e*', we obtain the following equation of the w-th degree for A: /(A) = A" + OiA"- 1 + • • • + o« = 0. If A is a root of this equation, e*' satisfies the differential equation. We shall now examine the various possibilities. Let A l7 Ag, . . . , A„ be the roots of the equation /(A) = 0, so that /(A) = (A-A 1 )(A-A 2 )...(A-AJ. 520 DIFFERENTIAL EQUATIONS [Chap. First assume that all the roots are different. If all the A„'s are real, then we obtain n linearly independent solutions e*"', exactly as before. The general solution is any linear combination Cje* 1 ' -+- Cjje*'* + . . . + c n &* of these. The constants c„ can be so determined that x and its first n — 1 derivatives take arbitrary pre-assigned values at time t = 0. To do this we must solve the following system of n linear equations *: ^ + 02+ ... + C„=x(0), A^ + A 2 e 2 + . . . + A„c„ = x'(0), K"- 1 * + A a "- 1 c 2 + . . . + A„«-ic„= ^-i»(0). If two of the roots are equal, say A x = Ag, then not only e x,f but also te Kt is a solution. This can be verified as follows: since /(A) = has a double root A = X 1 = )^, by a well- known theorem in algebra it follows that /'(A) = wA- 1 + (n - 1M- 2 + . . . + «„~i = 0. Now, by Leibnitz's rule for the derivative of a product (p. 202), * (to") = t-e" + k % ^ e w = <AV + hX*' 1 e". dt* dt k dtdt"- 1 Substituting in the differential equation, we have te"(X»+a 1 X"- 1 + ...+ a„)+e x '(wA»- 1 +(w-l)a 1 A«- 2 + ...-\-a„_ 1 ) = te»f(X) + e Xt f'(X)=0, since /(A) = and by the above remark on double roots /'(A) = 0. In the same way if A^ Aj, . . . , A^ are equal, we obtain the following linearly independent solutions: e x '«, to*"', . . . r-V 1 ', which may be combined to give a general solution depending on Cj, Cg, . . . , c„. These parameters again enable us to adapt the solution to n pre-assigned conditions, so that for t = we can fix the value of x(0) and its first n — 1 derivatives. * This set of equations always has a solution if the roots are unequal, for the determinant of the coefficients is not zero. XI] HOMOGENEOUS EQUATIONS 521 If the equation has complex roots, then by a theorem in algebra the roots occur in pairs, each one with its conjugate. Just as in the case n = 2, we obtain solutions of the form cos fit . e°* and sin/fc . e"', where \ = a + *yS, Ag = a — i/3. A few examples will serve to illustrate the above. Example 1. — + 2 — - - - 2x = 0, dfi dt* dt /(X) = X s + 2X a - X — 2 = 0. The general solution is x = c x e~* + c 2 e* + c 3 e~ ! '. A particular solution for which x = 2, *' = at t = is given by x = e' + e - '. d os tit or dx Example 2. — — — 4- x = 0. dt 3 <ft a <fc The general solution is x = c^e* + c^te* + c 3 e - '. ^a»m^e 3. ??-2- + 4=0, /(X) = X» - 2X + 4 = (X + 2) (X - 1 + ♦) (X - 1 - »). The general solution is z — c^ - *' + c 2 e* cost + c 8 e* sint. 2. Bernoulli's Equation. An equation of the type ^+A(t)x=B(t), where A and B are functions of t alone, is called a linear equation. In the case B = 0, if x = a(t), x = ;8(i) are solutions, any linear combination of a and ft is also a solution. We shall now consider the slightly more general type ^+A(t)x=B(t)x», where n is a positive integer. This is known as Bernoulli's equation. First consider the simpler case where B is zero, i.e. where ^+A(t)x=Q. dt w 523 DIFFERENTIAL EQUATIONS [Chap. dx Rewriting the equation as — = — A(t)dt, we see that it can be x integrated immediately as follows: logs = — jA(t)dt + c, if we write e° = v. Let us now try to satisfy Bernoulli's equation by a function of the form x = ve~i Adt , where we assume that v is a variable, so that dt dt Substituting, we have fo v — = Be—S A *et A *. dt which can be integrated at once, giving x i-« = (1 - w )e<»- 1 >P<" ffB&-*>S A "dt]. The above method is very important and may be applied in many cases. It is called the method of variation of para- meters. (For further details, see Volume II, p. 445.) Note that our solution is expressed in terms of integrals which can- not in general be expressed in terms of the elementary functions. Example. — Consider the equation — - tx = fix', dt Let x=*veS , *=ve i, ' ; then — — tx = — e" + vte v — tve u = — e ' , dt dt dt and the equation becomes dv u* dt v By integration, _ \ = (t* - 2)e*'' + c, or - = 2 - fi + ee'* 1 ". v * This result could have been obtained by direct substitution in the formula given above, but actually to carry the method through is far more instructive. <*V"=<W, or *L=fie v 'dt. .i2 XI] SEPARATION OF THE VARIABLES 523 3. Other Differential Equations of First Order Solvable by Simple Integration. There are a few other types of differential equations of the first order which can be solved by integration (although in most cases the integration cannot be performed explicitly in terms of elementary functions). The first method we shall consider is that of separation of the variables. If the differential equation can be brought into the form * A(x) dx + B(y)dy=0, the variables are said to be separable. The solution obviously is [A(x)dx+fB(y)dy + c = 0. Example. — Consider the equation yy" + xy* = x. Here ydy + x[y* — \)dx = 0, or y y + xdx = 0; hence ilog(y*-l) + ix i =e. or (j/ a - l)e*' = h. Another type of equation which can be solved is of the form M(x, y)dx+ N(x, y)dy = 0, where M and N are homogeneous functions of x and y of the same degree. In this case the fraction M/N is a function of y/x only, and we may write dy dx "©• If we put y =xv, this becomes X dx f The variables x, v are now separable as follows: dx dv x f(v) — v • That is, y'B(y) + A(x) = 0. 524 DIFFERENTIAL EQUATIONS TChap. Integrating, we have logs = fjr^r — + c ' Example. — Consider the equation (2 Vxy — x)dy + ydx = 0. Substituting y = vx, we have (2c 1 " - l)x(v + x Y) + «» = °» v{2v xlt -l) + v+ x(2v*i* - 1) ~ = 0, ax *! = _ 2vl * ~ * <fo = — * + — ~x 2« s ' 2 " » 2» s/2 ' Integrating, we have logo; = — log» — v~ * + e or logy 4- Vxjy = c. 4. Differential Equations of the Second Order. There are a few types of non-linear differential equations whose solutions can also be found by integration. One type has already been discussed implicitly in Chap. V (p. 297) when we studied the motion of a particle on a given curve. This type is as follows: dP JK ' Let v = — , so that dt cPx __dv dv dx _ dv dfi ~~dl ~dxdl ~ dx' and our equation becomes dv ,. . v— =f{x). dx XI] SECOND ORDER EQUATIONS 525 This may be regarded as an equation of the first order with v as dependent and x as independent variable. Separating the variables and integrating, we have vdv =f(x)dx « 2 = 2 J f(x) dx -\- or v = -y 2 jf{x)dx -f- e. Then dx ^2jf(x)dx = dt, + e which can be solved by integration (although in general it is impossible to carry out the integration explicitly). This device aids us to solve equations of the following types: ,/riPx dx\ . .(d?x dx \ JdH dx \ which reduce respectively, when we write v = — , to *{'&'■*)=*■ These are equations of the first order which may be solvable by the preceding methods. This solution, after v has been re- placed by -, will again be a differential equation of the first order, which must be solved for x. A few examples will make the process clear. 526 DIFFERENTIAL EQUATIONS [Chap. Example I. dy d*y . dxdx* Let -? = p. The equation becomes ox ox Integrating by separating the variables, we have op 2 = 3+0,, or Va^= Vx~+1^ ax By integration, Squaring, we have Example 2. Letf'= P . dx Va(y +c z )=* f (x + cj"*. a(y + e a ) s = f (x + c,) 8 ax 8 ax xdx ix p 1 + #* Integrating, we have logp = — £ log(l + x 2 ) + e, P = Cl (i + x«r i/a , or dy °i dx Vl + x 2 ' whence y = c 2 + Cj ar sinhx. Example 3. " dx* \dxJ Let^ = dx = P. then<^ = dx 8 - p ~. We have dp dy , „ pdp dy = 1 — p a , or J ^ = -s. 1-p* Jf XI] SECOND ORDER EQUATIONS 527 By integration, -4 log(I -p>) = logy + e, that is, or and y = Cl (l - p*r lft , 2/ 2 (l - p*) = c,", f~ p - **-°*\ or **L- = dx. dx y Vyi — ^ Integrating, we have ■VV - c, 2 - x + Cj, that is, y a = a? + C& + c 4 . Examples Solve the differential equations in Ex. 1-22. 1. (1 + y*)dx - (j, - VT+^) (i + x f dy = . 2. (x 3 + y*)dy = Zx*ydx. 3. y(loga; — logy)dy — xdx=0. 4. »f/' + y = j/> loga;. 5. (1 + y*)dx = (arc tany — *)%. 6. y^ + hy* = sina;. 7. (aV + x*y* + xy + l)y + (afy 3 _ »V - **/ + 1)^ = . 8. 3y V + y 3 =x—\. D. sina;eosy(te+oosa!sinj/iy=0. 10. (1 + e xlv )dx + e* lv (l -*)dy = 0. «-S-»S+«S--i ».§+©■+«-* 12.^-6^ + 9^=0. 18 ^ = ^ "■S-S+!=°- »<i--g+ »(|)=o. 15. ^ — 2 rf,y ' - - " ■" <*** « ^^ 2 + 2,= 0. 21. s^?=2 (^Y <ft> w d* 8 <te*^ y * * «ft> *V<a M.«^=* 22.(1-^-^=2. da; 2 (ia; d« s dt 23. Find the motion of a particle moving in a straight line under the attraction of a force varying as the inverse square of the distance from the origin. SUMMARY OF IMPORTANT THEOREMS AND FORMULAE 1. Hyperbolic Functions. 2. Convergence of Sequences and Series. 3. Differentiation. 4. Integration. 6. Uniform Convergence and Interchange of Infinite Operations. 6. Special Limits. 7. Special Definite Integrals. 8. Mean Value Theorems. 9. Expansions in Series: Taylor Series, Fourier Series. 10. Maxima and Minima. 11. Curves. 12. Length of Arc, Area, Volume. 1. Hyperbolic Functions (pp. 183-189) ^ , sinha; e* — e~°° sinha; = ««- - «-). tanhs = _ = ^-~. cosh* = *(e" + e"«). cotha; = ^ = j±£ ooah>« - sm&x = 1. cosh** = _ i-^. cosh (a; + y) = cosha; coshy + sinha: amhy. emh(x ±y) = sinha; coshy ± cosha; sinhy. eosh 2 a; = £(cosh2a; + 1). sinh 2 x = |(cosh2a; — 1). ar sinha; = log{x + s/(x* + 1)}. ar cosha; = logla; + y/{pP — 1)} (x ;> 1). 18 629 (E7981 53© SUMMARY OF FORMULA 1 4- x ar tanha; = \ log —I— (| x I < 1). 1 — x X -+- 1 ar cothz = £ log (| x I > 1). a; — 1 2. Convergence of Sequences and Series 1. Infinite Sequences (p. 38). Gauchy's Convergence Test (p. 40). A sequence of numbers a n is convergent if, and only if, for every positive quantity c there exists a number N such that I a» — ««. | < e when n > N, m> N. Operations with limits (pp. 41-42). If lima n and lim&„ exist then «->» ii-*- » lim(a„ + b n ) = lima„ + lim&„; « — >-ao b— >-oo n— >-oo lim(a„.6„)= lima„ . lim&„; #i— >-ao n— >ao 71— >-oo limo„ lim ^" = "~ >0 ° , provided lim&„ 4= 0. n->ooO n Iimo„ i»-*-« «->-00 2. Infinite Series (p. 365 et seq.). Gauchy's Convergence Test (p. 367). The series Sa n converges if, and only if, for every positive quantity e there exists a number N such that I On + « +l + • • • + Offl I < « when m > n > N. Note. — All the following criteria are sufficient but not necessary. Principle of comparison of series (p. 377). Ea„ converges if numbers b n exist such that b n 2g | o„ | for all values of n and Eft„ converges. CONVERGENCE 53* Ratio test and root test (p. 378). So„ converges if there is a number N, and also a number q < 1, such that a n+\ <q or y I a n | < q for all values of n > N; in particular, if there is a number k < 1 such that lim *n+l = k or lim -y/ 1 o„ | = A. Sa„ diverges if there is a number k > 1 such that lim °n+l = k or lim -(/ 1 a n \ = k. Leibnitz's Test (p. 370). Sa n converges if the terms have alternating signs and | a n | tends monotonically to zero. 3. Differentiation 1. General Rules (Fundamental Ideas, p. 88 et seq.), {/(*) ± £(*)}'=/'(*) ± •(*)• {/(*)*(*)}' = f\x)g{x) + f(x)g'(x). W)/ (<K*)} 2 > 9¥i 4= (pp. 136-139). {f(x)g(x)}M = /W {x) g( x) + Qf-^\x)g'(x) + (l)f n - 2) (*)9"(*) + • • • +f(x)g^(x). (Leibnitz's Rule, p. 202.) Chain rule. If f(x) = g{<f>(x)}, df dg d(f> dx d<f> dx dx* d*g /MY , dg d 2 <b , = JTl ( -T I + lh. X5' and so on (PP- 532 SUMMARY OF FORMULAE If u = /(£, t), i, . . .), where £ = £{x, y), i? = r)(x, y), . . . , «.=/«£. +f v V x +f(Zm + • • • » «_ =/«£•■ +/™ 1 ?* 2 +/?<r£* 2 + . . . + 2/ h *.ij. + 2/« £,£„+... + + f(£x* + fr,Vxx +fitxx + • • • , with corresponding formulae for u xy and w„„ (p. 476). Implicit functions. If F(x, y) = 0, dx F v ®y_ = __ F xx F y *-2F xv F x F v +F vv F x * dx 2 F* ^' Functions expressed in terras of a parameter. If x = x{t), V = »(«). dy = dyldx (p 262) dx dt/ dt Inverse functions. ^=l/^ (p. 145). dx I dy If £ = #&, y), 7) = 0(a;, y), dx_t/j v dx___^y dy = __ l h <ty_—$? H~D'd : n D' d£ D' d n D' where <f>x <t>v (functional determinant or Jacobian) (p. 479). d{x,y) ff> x >py — <f>y^x 2. Special Formulas (pp. 94-96, 139-141, 149-150, 167 et seq., 186-187). (a;")' = nx n ~K (sin a;)' = cos a;. (arc sin a;)' = (cosx)' = — sina;. (arc cosa;)' = 1 V(l - a?) DIFFERENTIATION 533 (tana;)' = — — = sec 2 2;. (arc tana;)' = cos 2 x 1 + x 2 (cota;)' = — =-s- = — cosec 2 a;. (arc cotx)' = sin 2 a; l + i 2 1 (sinha;)' = cosh a;. (arsinha;)' = (cosh a;)' = sinhx. (ar cosh a;)' = + 1 (x > 1). (tanha;)'= _— — = sech 2 aj. ( ar tanha;)' = 1 . ( I x I < 1). cosh z a; ' \ — ar (cotha;)'= — _^=-cosech 2 a;. (ax cotha;)' = — — ( I x I > 1). sinh 2 a; x ' 1 — x 2 (log a x)' = - log„e; ( a *)' = a * l g, a; in particular, in particular, ogx)' = -. x («")' = U V (VU'/U + »'logM), (logx)' = -. (e')' = e?. X 4. Integration 1. General Rules (Fundamental Ideas, p. 79 et seq.). ff{x) dx + fj(x) dx =f f(x) dx. f a f(x) dx = — f b f(x) dx. f a {/(*) + 9(x)}dx =f a f(x) dx +f a g(x) dx. J cf{x) dx— c I f(x) dx (pp. 81 et seq., 141). Estimation of integrals. If /(#) 2; g{x), b^a, r b r b I f(x) dx ^ / g(x) dx (p. 126). J a J a 534 SUMMARY OF FORMULA Integration by Parts (pp. 218-219). J a f(x)g'(x) dx = f(x)g(x) ||-jf * f'(x)g(x) dx. Method of substitution (pp. 207-212). fj(x) dx =f" f{<t>(u)}<f>'(u) du, where a = <j>(a), b = c£(£). Connexion between differentiation and integration (p. Ill et seq.). Improper Integrals (pp. 197-254). If f(x) is continuous except at the point x = b where it becomes infinite, / f(x) dx is (absolutely) convergent, if in the neighbourhood of x = b UK n ~ (b-x)"' where v < 1 (p. 248). J f(x)dx converges (absolutely) if i/<*>is£ where v > 1, for values of x 2> A (p. 250). 2. Special Formulas (pp. 82-87, 128-130, 142 et seq., 151, 168 et seq., 206, 208-209, 210, 213-217, 220 et seq.). J x n dx= . Jhgxdx= xlogx— x. - = log I x |. J - logs & = J (log x) a . sMoga;<Za;=— — (logx — ); o=)=-l. a + 1 \ a + 1/ INTEGRATION 535 f sinx dx = — coax. /sinhx dx = coshx. J coax dx = sinx. /coshx dx = sinhx. / tana; dx = — log | cosx |. /tanhx dx = log coshx. fcotxdx = log I sinx |. fcothx dx = log| sinhx |. Tare shixdx = x arc sinx + -\/(l — x 2 ). / arc cosx<Zx = x arc cosx — -\/(l — x 2 ). / arc tanxefee = x arc tanx — \ log(l + x 2 ). Tare cotxefce = x arc cotx -f- \ log (1 + x 2 ). Tar sinhxrfx = x ar sinhx — -\/(l + £*)• Tar coshxrfx = x ar coshx — \/(x 2 — 1). /ar tanhxefc = x ar tanhx + £ log(l — x 2 ). Tar cothxtfx = x ar cothx -+- \ logfx 2 — 1). /dx 1 L x I r dx , I , , x I = log tan- . / -— — = log tanh - . sinx I 2 1 J sinhx | 2 1 / = log tan 1 - + - 1 / — r— = 2 arctan I tanh J J cosx 6 | \2 4/1 J coshx \ 2/ = 2 ar tanh ( tan - I. /- = log I tanx I. /-— — = log I tanhx I. J sinx cosx J sinhx coshx /dx r dx -— - = —COtX. / . , = —COthX. sin a x J sinh^x /dx c dx , — — = tanx. / — — - = tanhx. cos z x J cosh 2 x 536 SUMMARY OF FORMULAE Jsm 2 xdx = \(x — sina; cosz). I cos 2 xdx = $(x -+■ sina; cosa;). dx — arc tan ( - tana; J ao \b / / . . , ri r = ~ -j-artanhf^tana;) J a'Bm^x— b 2 cos 2 x ao \b /) J a 2 sin 2 a; -f- b 2 cos 2 a: dx a, b 4= 0. /•da; I ^ x / 3-r— , = arc tan -. J x 2 + a 2 a a ar tanh = — log ?, if I x I < a. /dx \ a a 2a a-\-x *"-»* | 1 x 1 x—a - ar coth - = --- log , if I x I > a, a > 0. a a 2a x+a /dx Via 2 - x 2 ) ' 4- arc sin' dx -arc cos . a a f dx x J xVix 2 — a 2 ) 1 . a — arc sin - . a x , 1 a + arc cos-. a x f xdx = V(« 2 + x 2 ). Via 2 + x 2 ) /xdx . dx V> 2 "+ * 2 ) dx V& 1 - a 2 ) dx f V (a 2 +x 2 } = ar Shih a = l0g ^ ±a: + V{X * + fl2) >- --3-— - ar cosh . = log{x ± V(* a ~ a 2 )}. xVix 2 -\-a 2 ) dx 1 a f „f, „, = - I ars inh a =- 1 ln g ±^±V(g 2 + ^) log 1 , a 1 . a + Via 2 — x 2 ) // 2 2^ = ~' ai cosh = — lo g -=-?- — . Via 2 — x 2 ) a x a x r ax J xVia 2 — : INTEGRATION 537 / \/{a 2 — x 2 )dx = — a 2 arc cos - + - x^/(a 2 — x 2 ). f-\/(x 2 — a 2 ) dx = — - a 2 ar cosh + - x y/(x 2 — a 2 ). IVi® 2 + a 2 )dx= - a 2 ar sinh - -f- - x ^/(x 2 + a 2 ). J 2 a 2 f: dx x 2 +2bx+c \/{b 2 — c) 1 2^{b 2 -c) aitanh log x+b V(b*-c ) V(b 2 -c) — x—b V(6 2 - c) + x+b if c < b % , i.e. x 2 -f- 2bx -\- c = has real roots. / dx arc tan sc+6 x 2 + 26x + c V(« - &2 ) " " V(« - &2 )' if c> 6 2 , i.e. (E 2 + 26x -f c = has imaginary roots. e ax sinbxdx = e ox (a sin bx — b cos&r). a 2 + b 2 v /I e oa! cos fee dec = e ax (a cos foe + b sin fee). a 2 + o 2 /sin"a;cosa;aa; = J n+1 Recurrence Formulce (p. 221 e£ seg.). cos n xdx = - cos n_1 a; sina; -\ / cos n-2 a;<fcc. n n •> fsm n xdx — — - sin n_1 a; cos a; + / sin n_2 a;tfa;, J n n J fx n cosxdx = x n since — n Jx"' 1 sinxdx. fx n smxdx = — x n cosx + n \x n ~ x cosxrife. /• , sin m+1 a;cos n-1 a; , n — 1 /si sin™ x cos" x dx = m + n ■ + m + -7' sin m xcos n-2 a;ia;. 18* ( IS 798) 538 SUMMARY OF FORMULA j(logx) n dx = a;(logx) n — n uloga;)" -1 ^. fx n e x dx = x"e x — n fx"~ 1 e x dx. ^(log^fc^ ^'y - -4- [x°(hgx)»-idx (a =f* -1). f dx _ x In — S r dx J (1 + z 2 )" "" 2(n-l)(l + ai»)"- 1 2(n— lj^l + aj 8 )"- 1 ' 3. Integration of Special Types of Functions. (a) Rational Functions. These are reduced to the following three fundamental types by resolution into partial fractions (pp. 226-234): r dx 1 1 J (x — a) n ~~ n — 1 (a; — a)" -1 ' r dx 1 r du J (x* 4- 26a; 4- c) n ~ (c — 6 2 ^-* J '( (a? 4- 2bx 4- c) n (c — 6 2 )"-* •/ (1 4- ««)»' where c — 6 2 > 0, w = (a: + b)/\/(c — 6 2 ), the integral on the right being evaluated by the last recurrence formula given above; / xdx (x 2 4- 2bx 4- c) n 1 1 , r dx x 4- c)" -1 J ( 2 (n — 1) (x* 4- 26a; + c)"" 1 ./ (cc 2 + 26a;4- c)"' where the integral on the right is of the type immediately pre- ceding. In what follows R denotes a rational function. (6) jR(sinx, cosx)dx (p. 237). Substitution: t =•= tan , so that sin a; = , cos a; = dx_ 2 2 l + i2 1 + * 2 ' A 1 + t*' INTEGRATION 539 If, however, R is an even function or involves tana; alone, the following substitution is more convenient: i -a « 2 e 1 dx 1 u = tana;, sin 2 x = — — , cos 2 x = — — , — = — — -. 1 + u 2 1 + u 2 du 1 + u 2 x 2t 1 + < 2 Substitution: «= tanh -, so that sinh x= cosha;= -J— > (c) JR(coshx, sinha;)<& (p. 237) Substitution: «= tanh-, so that s„~~ dx_ 2 2 1 ~ t > 1_ <ft 1 - P (d) fR(^">)dx. Substitution: t = e""", — = — . at mt (e) fR{x, V(l - a: 2 )) da: (pp. 237-238). Substitution: (/) yi2(«, V(x* -l))dx (p. 238). Substitution: 1 V W+i/' ~ r^ 2 ' V( "r^T 2 ' s ~ a - < 2 ) 2- (g) y*i?(x,V(l + ^ 2 ))<fe(p.238). Substitution: i // » i i\ <2 ~ 1 // i i i\ 1 + fi dx t2 -fl t = x + \/(z* + 1), X = — - — , \/(x 2 + 1) = — ■ — , — = — l —. ^ v v ~r ;> 2* Y v ' 2t (ft 2f 2 (1+* 2 ) 2 (A) /fifo Vfa** + 26a; + c))(& (p. 239). . j, aaM The substitution f = — n Vl «w- one of the three preceding types, _ CLX I ■ 6 The substitution £ = — n — , reduces this integral to Vl ac— 6 2 | S4o SUMMARY OF FORMULA (t) [R(x, y/(ax + b), V(«b + d))dx (p. 239). Substitution: £ = ^/{cx + d) or x = - (£* — d), — = — (i) /<^(^))^ (p - 240) - Substitution: f V \« + «*/' c£» - a' d£ (c£» - a) 2 5. Uniform Convergence and Interchange of Infinite Operations For the definition of uniform convergence see p. 391. A series which is uniformly convergent in a closed interval and whose terms are continuous functions represents a con- tinuous function in that interval (p. 393). If \fjx) | ^ a„ and Sa„ converges, S/„(a;) converges uniformly (and absolutely) (p. 392). Interchange of Summation and Differentiation pp. 396-397). Any convergent series of continuous functions may be differentiated term by term, provided the resulting series converges uniformly. Interchange of Summation and Integration (p. 394). Any uniformly convergent series of continuous functions may be integrated term by term. The resulting series also converges uniformly. 6. Special Limits Stirling's Formula (p. 361). n! lim ,V2^n n+i e-» Wallis's Product (pp. 223-225, 363, 445). 1= n ( 2w 2n \ 2 — „_! \2n - 1 In + 1/ ,. (n!) 2 2 2n Vir = lim -5 — - — — . v „-+ao{2n)\^n (For infinite products, see pp. 419-422). SPECIAL LIMITS 541 «" = Km (l + -Y (p. 175). i (s) = s - = n _L_ « > 1 ( P . 420). „=ltt* p 1 — p * 00 / x 2 \ sinwc = 77a; II ( 1 1 (p. 445). Definition of the Gamma Function (pp. 250-251). T(x) = fe-H'^dx (x ^ 1): ■'0 r(* + i; = xr(x); if x is a positive integer n, T(n) = (n - 1)! Order of magnitude of functions (pp. 190-195). lim — = 00, if c > (p. 192). ft— >oo X" lim !££? = 0, if a > (p. 192). *-»ao X* lim x* logx = 0, if a > (p. 195). «->o 7. Special Definite Integrals Orthogonality relations of the trigonometric functions (p. 217). f + " . . , (0, if m 4= n. I sinmx sinnxax — { ., , „ J-* Itt, 11 m = n, n =)= 0. / sinmx cosnx da> = 0. r + * , (0, ifm=f=n. / cosmx cosnx ax ={ ., , „ ^-■r In-, 11 m = w, n 4= 0. jTV-'dx^V (P- 496). £ !™£ (fo = I „ (pp. 251-253, 418, 450). 542 SUMMARY OF FORMULA 8. Mean Value Theorems Mean Value Theorem of the Differential Calculus (p. 103). f{x + h) - f{x) =f(x+eh), o<0<i. A If f(x)—f(x+ h) = 0, this gives Rolle's Theorem (p. 105): between two zeros of the function there is always a zero of the derivative. Generalized Mean Value Theorem (pp. 135, 203). f(b)-m _nt) 9(b) -g(a) g'(i)' where f is a value between a and b. Taylor's Theorem (pp. 320-323). f(* + A) =/(*) + £ /'(*) + ~f"(x) + . . . + £/*•»(*) + Rm with the remainder (pp. 323-324). K = ^ / * (h - T)»/ ( " +1) (* + T)dT h"* 1 (n + 1)! /««+«(» +0A) «."(!_ 0)» /(«+« (35 + A ) (0 < < 1). Afean Value Theorem of the Integral Calculus (p. 127). f*f(x)dx = (6 — a) f(£), where a ^ f ^ 6. ff(x)p(x) dx =/(£) ^(z) <fcc, if p(x) ^ 0. EXPANSIONS IN SERIES 543 9. Expansions in Series: Taylor Series, Fourier Series 1. Power Series (for definition, see p. 398). (a) Power Series in General. Any power series R-0 in one variable has a radius of convergence p (which may be zero or infinite); the series converges when | x\ < p, and in fact it con- verges uniformly and absolutely in every interval | x | ^ 17, where 7) < p; when | x | > p, the series diverges (p. 400). If the remainder in Taylor's theorem tends to zero as n increases, we have the infinite power series (p. 325) fix + h) =/(*) + */' (x) + *f"(x) +... + ^f\x) +.... (b) Special Taylor Series (pp. 316-319, 326-330, 405-409, 422-423). iog(i + *) = *-£ + J- 5 + -... + (-ir l - + .-- for — 1< x ^ 1. ) x in+l sinx=x-^ + |;- + . ..+ (-!)« (2n+i)! 00.. -.1--, + -- + . .. + (-!)» — + . + for all ' values Of 3. X 8 3* , X 2b+1 sinhx = 2; + — + — +...+ S! 51 (2n + 1)! + cosha; = 1 + + - + (2n)! + „_i (2v)! for--<»<-, 2/COtZ 1 2 2 "B„ S (_i)-H_^?'iB 8 ' for— 7T<x<it, v-o (2v)! where the quantities B 2t , are Bernoulli's numbers (p. 423). 544 SUMMARY OF FORMULA . la* , 1.3a* , 1.3.5a; 7 , arc sinx = x + — + -A — + . . . 2 3 2.4 5 2.4.6 7 . , la? , 1.3a? 1.3.5 x 1 , for ar sinha; = x — \- ... V z. - , 2 3 2.4 5 2.4.6 7 -l^^l- x 3 x 5 arc tanas =x— — + — 1-... artanhx = as + — -f — + ... for|x|<l. Binomial Series. (1 + xf , , , o(a — 1) , . in particular, 1 a(a-l)(g-2)...(a-n+l) d -* + -'- for — 1 < x < 1, if a > — 1 for x = 1 also, if a S: for a; = — 1 also; = 1 — x-\- a? — a? -\ ... = 1 — 2x + 3a; 2 — 4a? + — 1 + as 1 (1 + xf 1 , 1 , 1.3 . 1.3.5 , , 1.3.5.7 . , =1 — -xA x a a? A or r ••• • V(l + x) 2 ^2.4 2.4.6 ^2.4.6.8 X Elliptic integral: V(l- ^sinV) {^©"^GrO'^Gifr^-}' EXPANSIONS IN SERIES 545 2. Fourier Series. If the function f(x) is sectionally smooth in the interval —17 ^ x ^ 77, i.e. if its first derivative is sectionally continuous, the Fourier series 1 °° f(x) = - a + S (a, cos vx + o„ sin vx), 2 1—1 where 0,= - f /(<)cosvi<&, 6,= - f /(*)sinv<<ft TlJ—rr TI J —Tt is absolutely convergent throughout the whole interval. If f{x) has a finite number of jump discontinuities, while elsewhere f'(x) is sectionally continuous, the series converges uniformly in every closed sub-interval which contains no discontinuities of f(x). At every point at which f(x) is continuous, the series represents the value of the function f(x), while at every point of discontinuity of f(x) it represents the arithmetic mean of the right-hand and left-hand limits of f(x) (pp. 447-450). 10. Maxima and Minima The following rule holds only for maxima and minima in the interim of the region under consideration. In order that £ may be an extreme value of the function y = f(x), f'(g) must vanish. When this condition is satisfied there is a maximum or minimum if the first non- vanishing derivative of f(x) is of even order; if it is of odd order, there is neither a maximum nor a rninimum. In the former case there is a maximum 01 a minimum according as the sign of the first non-zero deri- vative is negative or positive (p. 158 et seq.). 11. Curves In what follows £, -q are current co-ordinates. Equation of the curve: («) y =/(*), (&) -Ffo y) = °> («) * = * W, y = * W- 546 SUMMARY OF FORMULA Equation of the tangent at the point (x, y) (p. 263): (o) , - y = (£ - x)J'{x), (b) ($- x) F„+{ri- y) F v = 0, («){*-* (') }f (0 - {v - (0 } M) = 0. Equation of the normal at the point (x, y) (p. 263): (o) f - * + (ij - ?)/'(*) = o. (&) (*- *)*.- to - y)*. = 0, (c) {£- <£(*)}f M + {*? - *(0}f (0 = o- Curvature (p. 281): . . 7 y" /M t F xx F v i —2F xv F x F y + F^FJ {a) h ~ ("!+?¥' (6) . ... W+W ' Radius of curvature (p. 282): 1 Evolute (locus of centre of curvature) (pp. 283, 307-311): 1 4- w' 2 .1 + V' 2 {a)i=X -y>L±JL, v =y+^; P 2_|_ p 2 (6) £=*+*. ^^ a _ 2F xv fJ v + F„F„» ' Involute (p. 309): | = a; + (a— s)*, tj = y + (a — *)& where a is an arbitrary constant and s the length of arc measured from a given point. Point of inflection (pp. 159, 266). Necessary condition for a point of inflection is (a) y"= 0, (6) F„F* - 2F XV F X F V + F„F* = 0, (c) xy—xy = 0. CURVES Angle between two curves (p. 264): 547 (b) cosa> = VW+^VOvW+W (c)cosa, **i + m V(* 2 + y 2 )V(V + 2/! 2 ) In particular, the curves are orthogonal if (6) F X G X + F V G V = 0, (c) x*! + yy t = 0; the curves touch if (6) F X G V - F V G X = 0, (c) s& - ^y = 0. Two curves y=f(x), y = g{x) have contact of order n at a point x, if /(*) = *(*), /'{*) = ?{*), .... /^^"V), / (n+1) (a')*5r ( ' ,+1) ^) (pp. 331-333). 12. Length op Abc, Akea, Volume Length of Arc (pp. 276-280). Let a plane curve be given by the equations (a) y =f(x), (b) F(x, y) = 0, (c) z = <f> («), y = 0(0. (<Z) (polar co-ordinates) r = r(0). The length of arc is (a) • = /V(l + V' 2 )dx, (c) « = /"V(* 2 + 2/ 2 )*> ® * = PI vW + F*)ix, (d) s = f 1 V(r 2 + r' 2 ) «W. Area of Plane Surface. The area bounded by the curve r=r(0) and two radii vectores 6 , 6 V where r, are polar co-ordinates, is given by lpr*d6 (p. 275). 548 SUMMARY OF FORMULA The area enclosed by the curve y=f(x), the two ordinates x = x , x= x 1 , and the x-axis, is Cy&x (p. 80). Volume. The volume lying over the region R and bounded above by the surface with the equation 2 =f(x,y) is given by V=fff(x,y)dxdy (p. 487). MISCELLANEOUS EXAMPLES CHAPTER I 1. Prove that if p and q are integers the expansion of p/q as a decimal either terminates or recurs from a certain point onward. Prove also that every terminating or recurring decimal represents a rational number. 2. Express 39 in the ternary scale (scale of 3). 3. How would the number one hundred and fifty-six be written if (a) the binary scale (scale of 2), (6) the scale of 4, were in common use? 4. Express the following numbers in the scale of 12: (a) 1076, (6) 10,000, (c) 20,736, (d) 1/6, (e) 1/64, (/) 1/5. 5. We can find V 2 to one decimal place thus: I s = 1 < 2, 2 2 = 4 > 2, therefore 1< V2 < 2. Next, l-3 a = 1-69 < 2, l-4 a = 1-96 < 2, l-5 a = 2-25 > 2, therefore 1-4 < V2 < 1-5. (a) Continue this process one step further. (6) Calculate V7 to two decimal places by the same method. 6. For what values of x do the following inequalities hold? (a) x* + 3x + 1 ^ 0. (c) ' l S6. (b) X* - x + 1 ^ 0. (d) Sx - 2 ^ a* x + X 7. Prove that the arithmetic mean of two positive quantities a, 6 is not less than the geometric mean Vab, i.e. that -±r > Vab. State when the equality sign holds. 8. The quantity | defined by - = ( +-)is called the harmonic 5 2 v» 6/ mean of the two positive quantities a, b. Prove that the geometric mean is not less than the harmonic mean, i.e. that Vofi ^ %. When does the equality sign hold? 9.* Show that the following inequalities hold, if a, b, c are positive: (a) a 1 + 6» + c» g ah + be + ca. (6) (a + b)(b + c)(e + a) ^ 8a6e. (c) o 2 6* + 6«c» + cV ^ abc (a + b + e). 549 550 MISCELLANEOUS EXAMPLES 10. The numbers «,, x* *, and a ik (t, k = 1, 2, 3) are all positive. In addition, o J]k g 3/ and a^' + xf + * s a ^ 1. Prove that a^xf + 2ai 2 a: 1 a; 8 + . . . + a^V g 3Jf . 11.* Prove that if the numbers %, a 2 a n and &„&,,..., &„ satisfy the inequalities Oj ^ a a ^ . . . ^ o n , & t ^ b 2 2: . . . S 6„, then 12. Prove the following properties of the binomial coefficients: <<)+«GM;) + --- + -C)-»*"- (c) 1 . 2g) + 2 . 3 (J) + . . . + (n - 1)»(*) = »(»- l)2»-«. .«'+iG)+iGD+-+.-TiO-^r- 13. By summing v(v + l)(v + 2) ... (v + ft + 1) - (v - l)v(v + 1) . . . (v + k) from v = 1 to v = n, show that Sv(v + l)(v + 2)...(v +fc ) = "<" +1 )---( re + fe+1 ) . v- j e+2 14. Evaluate I s + 2 3 + . . . + n 3 by using the relation v> = v(v + l)(v + 2) - 3v(v + 1) + v. 15. Evaluate , . 1 . 1 , . 1 ( a > , „ „ + i— jr— ; -. + ••• + 1.2.32.3.4 n{n + l)(n + 2) /m l _L. * -I. X J- -L. l 1.3 2.43.5 n(n + 2) 1 1 ' 1.2.4 2.3.5 n{n+ l)(n + 3) 16. Find a formula for the n-th term of the following arithmetio pro- gressions: (a) 1, 2, 4, 7, 11, 16, . . . . (6) -7, —10, —9, 1, 25, 68, . . . . MISCELLANEOUS EXAMPLES 551 17.* Show that the sum of the first n terms of an arithmetic pro- gression of order k is aS k + 6S t _, + ...+ pS, + qn, where S„ represents the sum of the first n v-th powers, and a, b p, g are independent of n. Evaluate the sums for the arithmetic progressions of Ex. 16. 18.* Prove the binomial theorem (o + 6)» = a B + ("V" -1 * + ("V"""* 6 ' +••• + &" by mathematical induction. (See also Chap. Ill, p. 201.) 19. Find (a) lim (— + .— + . . . + )■ »-^.\1.2 2.3 n(n+l)J (6) lim ( — h — h ...+ V '»->.\1.2.3 2.3.4 n(n + l)(n+2)/ / 1 1 1 \ (c) lim ( -n- + ,.— — + . . • + -rif J. 20. If Ss f = 0, prove that lim S a^ Vn + » = 0. i-o »— >■» «-o 21. Prove that lim " = 0. »->«2 n 22. Prove that lim (n + 1)5 = 0. 23. Prove that lim tfn* = 0. (2» + 1),, 24. Prove that lim -y/(n* + n) = 0. * — >■« 25. Use Cauchy's convergence test to show that the following sequences converge: 1 (a) a n = . n ... n + 1 (6) c = — L- . n (c)*«„=. + 1 1 j + 2 1 ! +... + n V 26.* Show that the limits of the sequences (c), (d) of the previous example are reciprocals of one another (so that the limit of the sequence (<*)isl/e»). 552 MISCELLANEOUS EXAMPLES 27.* Prove that the limit of the sequence V2, V2 + V2, - v / 2 + V2 + V2, ... (a) exists, (6) is equal to 2. 28.* Prove that the limit of the sequence exists. Show that the limit is less than 1 but not less than £. 29. Prove that the limit of the sequence - X +-..+ 1 n+1 2n exists, is equal to the limit of the previous example, and is greater than i but not greater than 1. 30. Obtain the following bounds for the limit L of the two previous i 37 ^, , 57 examples: — < L < — 60 60 31.* Let a v 6, be any two positive numbers, and let a, < 6,. Let a ' = a 1 + b l ' 6 ' =Vo » 6 i' and in general a n-i T- ° B -i Prove that the sequences a„ a 2 , . . . and b lt b 2 , . . . converge and have the same limit. 32.* If a n > 0, and lim V+> = L then lim jy = L 33. Use Ex. 32 to evaluate the limits of the following sequences: (a) yn, (b) v'fn* + » 4 >, (c) ^/(~)- 34. Use Ex. 33(c) to show that re! = ra B e _n o n , where a n is a number whose ra-th root tends to 1. (See Chap. VII. Appendix, p. 363.) x -4- 2 35. Prove that lim — — - = 2. Find a 8 such that for I x I < 8 the dif- z-^.0 x + 1 a: + 2 ferenoe between 2 and is. in absolute value, (a) less than A> (6) less a: + 1 than 10 1 00> (c) less than e, e > 0. MISCELLANEOUS EXAMPLES 553 36. (a) Prove that lim a = -. Find a 8 such that for | 1 — * | < 8 «_>i x + 1 2 the difference between - and is, in absolute value, less than e, e > 0. 2 x + 1 Do the same for (6) lim V(l -f x 3 ); (c) lim z— >2 *— »-o a; 37. Prove that (a) lim - (1 + a:) ~ 1 = i. *->o x 2 (6) lim V* + i( Vx + 1 — Vx) = J. 2 — ^-w 38. Prove that lim (cos to) 2 "* exists for each value of x and is equal to m — >•» I or according as x is an integer or not. 39.* Prove that lim [lim (cos im\ a;) 2 ™] exists for each value of x and is equal to 1 or according as x is rational or irrational. 40. Determine which of the following functions are continuous. For those which are discontinuous, find the points of discontinuity. {a) f(x)=t±^±^f, /(0) = 0. sinx (6)/(x) = a;5 + 5 ^ + 3a: , /(0) = 0. smx (c) f(x) = lim (cosnx) 2 "*. m — >■» (d) f(x) = lim [lim (cosmilx) 2 "]. 41. Let /(x) be continuous for ^ x ^ 1. Suppose further that f(x) assumes rational values only, and that f(x) = £ when x = £. Prove that /(a) = J everywhere. 42. Has the function fix) = 2 sin 3* + 10 cos5x any real zeros f 43.* If /(x) satisfies the functional equation /(*+»)-/(») + /(*) for all values of x and y, find the values of /(x) at the rational points and prove that, if /(x) is continuous, f(x) = ex, where c is a constant. 44.* Prove a converse of the theorem of uniform continuity; namely, that if /(x) is uniformly continuous in the half-open interval a < x SJ 6, then f(x) tends to a unique limit asi->j (which may be taken as the value of /(a)). 554 MISCELLANEOUS EXAMPLES 45. Plot the following graphs, and express the equations in Cartesian co-ordinates: (a) r=*a+b cos 6 (Limacon). (6) •—=—^— s (Ellipse). 2 — cos8 . . 2a sin a 6 ,— . ,. (c) r = (Cissoid). cos 6 (d)r= ******«»* (Folium of Descartes), sin* 9 + cos'o 46.* Show that the equation of an ellipse with one focus at the origin is = k l — ecos(e — e )' 47. Let c be the complex number x + iy represented by a point in a Cartesian co-ordinate system. Plot the ourves c— t («) (&)* o + t e — a 2. = k, a, |3 complex constants, c- p (c) |c»-l| = *. 48. Let Cj, c s be two complex numbers. Prove that (o) \c 1 ±c 2 \^\c 1 \ + \c i \. (b) I^ical^lcjl-Kl- 49. Prove the equality I "a + c a | * + | <H - e a | « = 2 | c, I s + 2 1 c, |* and state its geometrical interpretation. 50. Prove that (oos6 + » sine)" = cos»6 + » sinnG by mathematical induction. CHAPTER II 51.* Prove directly that the derivative of the functioD f(x) = x" sin -. x =*= 0; /(0) = x exists at every point and is equal to — cos - + 2x sin -, x 4= 0; at x = 0. X X Show that although f'(x) is not continuous at x = 0, nevertheless the mean value theorem still applies and the property of Ex. 67 below holds good. (See pp. 199, 200 of the text.) MISCELLANEOUS EXAMPLES 555 62. Draw the graph of the Amotion /(x) = x sin -, x dp 0; /(0) = x and find its derivative for x 4= 0. Show that its derivative does not exist at x = 0, but that the difference quotient ^-^ ^-/ as x -> has the x upper and lower limits 1 and — 1 respectively. (See p. 199.) 53. Investigate the behaviour of the function f(x) = x sin - + a? sin -, x #= 0; /(0) = x x with regard to differentiability. 64. Prove that the derivative of the function f(x) = - sin x, x 4= 0; /(0) = 1 x exists at every point and is equal to fix) = — — sin x + - cos x, x #= 0; /' (0) = 0. X 2 X Show that f'(x) is continuous, and find f"(x). 55. If f{x) is continuous and differentiable for a g x £J 6, show that if /'(a;) ^ for »:£*<!; and /'(x) ^ for $ < x ^ 6, the function is never less than/(§). 56.* If the continuous function f(x) has a derivative /'(x) at each point x in the neighbourhood of x = £, and if /'(x) approaches a limit L as x — *■ 5, then /'( 5) exists and is equal to L. 57.* If /(x) possesses a derivative f'(x) (not necessarily continuous at each point x of a ^ x ^ 6, and if /' (x) assumes the values to and M) it also assumes every value ft between m and M . 58. If /"(x) ^ for all values of x in a ^ x ^ 6, the graph of y = /(x) lies above the tangent line at any point x = §, y = /($) of the graph. (The curve is convex upwards.) 59. If f"(x) ^ for all values of x in a ^ x ^ 6, the graph of y = y /(x) in the interval 1,^1^1, lies below the line segment joining the two points of the graph for which x = x x , x = x s . 60. If/"(x) => 0, than /(**+*) g /<*»> + /(**>. 61. Given /(x) = $x* — x* + 1, find a number <5 such that for every h less in absolute value than S and every x in the interval — J Sa x 5> i the following inequality holds: A JL loo 556 MISCELLANEOUS EXAMPLES 62. Differentiate directly and write down the corresponding integra- tion formulae: (a) x 11 *; (b) tana:. 63. Evaluate (o) lim J- (l +*+■•• + 4-Y (6) lim I ( 1 + sec* - + sec a - + . . . + seo a — Y »— >.» » \ 4n 4ra 4»/ 64. Prove that 2"+ 1 (n!)« (a) T 1 (x> - l)»(fa = J?; (6) (-D" f 1 (*" - l)»«fe = (2m + 1)! 65. Show that 1 /-"+ 1 & 1 +1 J. * v+ 1 J x v 1 v and > + » + .. . + i<r*<i + i + ... + . » 2 3 n ./, a; 2 n— 1 1 1 f v dx Prove that the sequence 1 + -+... H / — , v=l, 2, .... is a 2 v Ji x decreasing sequence and is bounded below. 66.* Let f(x) be a function such that f"(x) S for all values of z, and let u = ti{t) be an arbitrary continuous function. Then 1 [ a f(u(t))dt>f( l f a u(t)dt). ajQ \ctJo ' 67.* If a particle traverses distance 1 in time 1, beginning and ending at rest, then at some point in the interval it must have been subjected to an acceleration =£ 4. CHAPTER m 68. Differentiate the following functions: / a \ gtan'x + logsinz^ (6) (x + 2) 4 (1 - a^'V + l)" 7 . x 3 sina; — a; 5 cosa; («) a^ tana; 69. What conditions most the coefficients a, (3, a, b, c satisfy in ordei that ax+ g V(aa? + Zbx + e) shall everywhere have a finite derivative which is never zeroT MISCELLANEOUS EXAMPLES 557 70. Sketch the graph of the function y = (**)*> y(0) = i. Show that the function is continuous at x = 0. Has the function maxima, minima, or points of inflection? 71. Among all triangles with given base and given perimeter, the isosceles triangle has the maximum area. 72. Among all triangles with given base and given vertical angle, the isosceles triangle has the maximum area. 73. Among all triangles with given base and given area, the isosceles triangle has the maximum vertical angle. 74.* Among all triangles with given area, the equilateral triangle has the least perimeter. 75.* Among all triangles with given perimeter, the equilateral triangle has the maximum area. 76.* Among all triangles inscribed in a circle, the equilateral triangle has the maximum area. 77. Prove the following inequalities: w^rb- x>0 - (6) e» > 1 + log(l + x), x> 0. (e) e* > 1 + (1 + x) log(l + x), at> 0. 78.* Let a, 6 be two positive numbers, p and q any non-zero numbers, p < q. Prove that [6a'+(l-e)6g] 1/>> 5;l [6a« + (1 - Q)b"f lq for all values of in the interval < < 1. (This is Jensen's inequality, which states that the p-th power mean [00*+ (1 — Q)V*f v of two positive quantities a, 6 is an increasing function of p.) 79. Show that the equality sign in the above inequality holds if, and only if, a = 6. 80. Prove that lim [8o» + (1 - e)^] 1 * = a"* 1 -*. 81. Defining the zero-th power mean of a, 6 as o*6 1 -*, show that Jensen's inequality applies to this case, and becomes (a 4= 6) a 96i-« > [0 O « + (1 — e^ 9 ] 1 '" according as q ^ 0. For 8=1, a?&-« ^ 6a + (1- 6)6. 82. Prove the inequality «j,i-« ^ 8a + (1 — 0)6, o, 6 > 0, < 6 < 1, without reference to Jensen's inequality, and show 558 MISCELLANEOUS EXAMPLES that equality holds only if a — b. (This inequality states that the 0, 1 — 6 geometric mean is less than the corresponding arithmetic mean.) 83. If <p(x) -*■ oo as x -*- oo , show that log<p(x) is of a lower order and e* te> of a higher order of magnitude than <p(x). 84, If the order of magnitude of the positive function f(x) as x •* oo is higher, the same, or lower than that of x m , prove that / f(^)d^ has the corresponding order of magnitude relative to a; m+1 . "'" /(5)<Z5 relative to j v ~y x^x mo xuxiuwxug imiuuuxlH J W ; (a) -^ (c) *e»*. (&) e». (d) log*. 86. Prove that if f(x) is continuous and /(*) = [ X f(t)dt, then f(x) is identically zero. 87. Prove that 2 w;* -1 = v - 1 — . i-i (x - If 88. Show that ' = u n (x]e x l*, where u n (x) is a polynomial of degree n. Establish the recurrence relation M n+i - *»n + «»'• 89.* By applying Leibnitz's rule to - (e»'/2) = xt**!*, ax obtain the recurrence relation "n+i = *"» + »«„-i- 90.* By combining the recurrence relations of Ex 88, 89, obtain the differential equation u," + mt n ' — nu n = satisfied by «„(*). 01. Find the polynomial solution u„(x) = x n + ai*" -1 + . . . + o, of the differential equation «„" + »u B ' — nu n = 0. MISCELLANEOUS EXAMPLES 559 1 d» 92.* If P n (x) = _L *!. (a;2 _ i)», prove the relations (a) P '— x *~ l P » 4. ( n + 2 )* p , » + 2 (a) ^» +i ~ 2^+"D p » + ^+r F - + ~r f - (&)P„ +I , = *P„'+(»+l)P B . (c) £ ((*» - 1)P„') - n(n + 1)P„ = 0. 93. Find the polynomial solution of the differential equation £ «* - i)p n o - «(» + i)p„ - 0. 94. Determine the polynomial P„(x) = (a? — l) n by using the binomial theorem. ^" re! ^ 95 that * Let W*) = (n)^ 1 ~ x ^~ n ' » = 0» 1. 2 p. Show 1=SW*). , g) * = ^ U X„Jx). n-ip »-* (f) CHAPTER IV Perform the integrations in Ex. 96-101. 96. f] + <{ X dz. 99. f Xt ~ l dx. \>x VV" + 1) "*" *""' J xV(x*" - 1)' m <• dx -ry-^ dX. 100. / 98 /* xdx 101 f *** ' J ^(l + x)- V(l + x)' ' J x(x + 1) . . . (x + n Evaluate the integrals in Ex. 102-107. coa n xdx. 103. / cos' 39 sin* 68 i8. 56o MISCELLANEOUS EXAMPLES 104 l 1 x* n dx 105. f •4) 1 a* n -*dx V(l - x*)' V(l — X 2 )' 106. f x*V(\ - 3?)dx. 107. f ^(1 _ x*f*dx. Jo J o Obtain recurrence formulae for the integrals of Ex. 108-112. 108. fx<>(logx) m dx. 111. fe ax aiahbxdx. 109. fx n e ax einbxdx. 112. je ax coshbxdx. 110. /V'e * ooabxdx. results /dx — ; in three different ways and compare the V(a a — x 2 ) 114.* Let P„(a;) = — — (a? - l) n . Show that " 2»»! dx n f P n (x)P m {x)d* = 0, if m * n. T 1 2 115. Prove that / iy(a:)<fo:= — — 5 . J-l 71+1 116. Prove that f x m P n {x)dx = 0, if m < n. 117. Evaluate f x"P n (x)dx. Test whether the improper integrals in Ex. 118-131 oonverge or diverge. 118 r" dx i i '" Jo V(ox — X 2 )' /•" dx 119 - h *V(*-1)* 120. jf^)"** 121. jf X^D" -CO 122. / «-*B»»(]oga;)»<te; •A) 123. / log sinztfz. , f -log dx. 124 'sinxdx. / x log sinztix. Jo / e-"dx. J—<X> r x^-ig-x'dx. t/2 a;m(fa! 'o (sins)" r m dx J 1 + x* sin 2 *" X 00 itfa; 1 + X s sin'a ^ r m x*dx J 1 ■+- aP sin* MISCELLANEOUS EXAMPLES 561 132.* If f — dx converges for any positive value of a, and if f(x) tends to a limit Lasx^O, show that / J ->— - — ^— dx converges and p Jo x has the value L log -. a 133. By reference to the previous example, show that (a) / ix = log -. J x a /•°°cosaa; — cosSa; p (6) / — <&: = log -. 134.* If / -Q dx converges for any positive values of o and b, and if J a x f(x) tends to a limit M as x -*■ 00 and a limit i as x -* 0, show that f-/(-0^«W *.-(£_ Jf, kg S. Jo x a 135. Obtain the following expressions for the gamma function: T(n) = 2 f* x u *- 1 e-*'dx, Jo Xi / IN" -1 flog- 1 dx. CHAPTER V 136. Plot the following curves and find their equations in non-para- metrio form: . . 5aP oat 8 (a) x = — — ,, V = 1 + *»' " 1 + « 5 (6) x= at + b sint, y = a — 6 cost. 137.* Show that the two families of ellipses and hyperbolas, °* J y*— = 1, for X < 6, a 2 _ x T 6 2 - X -| ? — = 1, for a < t < 6, a 2 _ T &2 — T are confocal and intersect at right angles. 138. Find the pedal curves (see p. 267, Ex. 11) of the following: (a) the ellipse x = a cos 6, y = 6 sin 6 with respect to the origin; (6) the hyperbola x = cosh 6, y=b sinhO with respect to the origin; (c) the parabola y 2 = 4px with respect to the origin; (d) the parabola y 2 = 4px with respect to the focus. 19 CB798) 562 MISCELLANEOUS EXAMPLES 139. Show that the tangent to an ellipse is equally inclined to the focal radii drawn to the point of contact. 140. Show that the tangent to a hyperbola is equally inclined to the focal radii drawn to the point of contact. 141. A constant length I is measured off along the normal to a para- bola. Find the curve described by the extremity of this segment. 142. Find the area bounded by the loop of the curve x 5 + y b — Sarty 2 = 0. 143. Find the area enclosed by the curve a?(x* + y 2 ) 2 (6 2 a^ + aV) = (a a — & 2 ) 2 6 2 a;\ 144. Find the length of arc of the epicycloid x = (a + 6) cost — 6 cos — — — t b y = (o + 6) sin* — 6 sin -~ — t reckoned from the initial point t = 0. 145. Prove that the radius of curvature at a point of the polar curve r = /(9)i3 r-+(S)T d6 2 W 146.* If the curvature of a curve in the a^-plane is a monotonic function of the length of arc, prove that the curve is not closed and that it has no double points. 147. Find the moment of inertia of a rod of length L (a) with respect to its centre; (&) with respect to one end; (c) with respect to a point on the line of the rod at a distance d from the centre; (d) with respect to any point at a distance d from the centre. 148. Find the equation of the curves which everywhere intersect the straight lines through the origin at the same angle a. 149. Find the equation of the curves whose normal is of constant length k. (The " length " of the normal is the length of the portion of the normal intercepted between the curve and the a;-axis.) 150. Show that the only curves whose curvature is a fixed constant k are circles of radius \jk. 151. Find the equation of the curves whose centre of curvature lies on the x-axis and whose radius of curvature is therefore equal to the length of the normal. MISCELLANEOUS EXAMPLES 563 152. Find the equation of the curves whose radius of curvature is equal to the length of the normal but whose centre of curvature does not lie on the a-axis. 153.* Obtain the formula for the length of a curve in polar co-ordinates. CHAPTER VI 154. Deduce the integral formula for the remainder B n by applying integration by parts to /(* + A) - /(*) = /"/'(* + t)*t. •'O 155. Integrate the formula R n =- f\h - T)»/ <n+1, (* + t)*t, to! Jo and so obtain A", K = /(* + *) ~ /(*) - V(«) - • • • - „y/ (*>• 156.* Suppose that in some way a series for the function f(x) has been obtained, namely f(x) = a t) +a 1 x+atf?+ ... + a„x n + B„{x), where a , a v . . . , a n are constants, B n (x) is to times continuously differen- tiable, and Bn ^ -+ as x -* 0. Show that a k = £-±-± (fc = to), x n fc! i.e. that the series is a Taylor series. 157.* Find the first three non-vanishing terms of the Taylor series for sin 2 a; in the neighbourhood of * = by multiplying the Taylor series for sina; by itself. Justify this procedure. 158.* Find the first three non-vanishing terms of the Taylor series sina; for tana; in the neighbourhood of x = 0, by using the relation tana; == — -, and justify the procedure. 159.* Find the first three non-vanishing terms of the Taylor series for V cos a; in the neighbourhood of x = 0, by applying the binomial theorem to the Taylor series for cos a;, and justify the procedure. 160. Find the first four non-vanishing terms of the Taylor series for the following functions in the neighbourhood of x = 0: (a) a;cota:. (c) seoa;. (e) ef- (6) Vsjnx . (d) e*". (/) log sina; — loga:. Vx 564 MISCELLANEOUS EXAMPLES 161. Find the Taylor series for arc sinx in the neighbourhood of x = by using r* dt are sinx = / ,- Jo Vl — fi (Cf. p. 203, Ex. 5.) 162.* Find the Taylor series for (arc sins) 2 . (Cf. p. 203, Ex. 5.) 163. Find the Taylor series for the following functions in the neigh- bourhood of x = 0: (a) sinh- 1 ^ (6) f er*'dt. (c) f ^- dt. Jo Jo * 164.* Estimate the error involved in using the first n terms in the series in Ex. 163. 165.* Two oppositely charged particles +e, — e situated at a small distance d apart form an electric dipole with moment M = ed. Show that the potential energy (a) at a point situated on the axis of the dipole M at a distance r from the centre of the dipole is — (1 + e), where e is approximately equal to — 4r 2 ' (6) at a point situated on the perpendicular bisector of the dipole is 0; (c) at a point with polar co-ordinates r, relative to the centre and axis of the dipole is — — (1 + s), where s is approximately equal to — (5 cos 2 6 -3). 8r a (The potential energy of a single charge q at a point at a distance r from the charge is q/r; the potential energy of several charges is the sum of the potential energies of the separate charges.) 166.* Find the first three terms of the Taylor series for f 1 + - ) in powers of -. x 167. Evaluate the following limits: (a) lim z|"(l+ -)*-«"]• (c)* lim x [(l + *)*- e log (l + ?)"] . (d) lim (™y. W lim ( s ^y". 168.* Show that the osculating circle at a point where the radius of curvature is a maximum or minimum does not cross the curve. 169. Find the maxima and minima of the following functions: (a) \x\, (b) xsin(l/x). MISCELLANEOUS EXAMPLES 5^5 CHAPTER VTI 170. Show that the length of the ellipse x = a cost, y = 6 sin< is /■v/2 o 2 — 6 2 s = 4a I VI — e 2 cos 2 tdt, where e 2 = ; JO a Calculate the length of the ellipse for which e = \ to four significant figures, by using Simpson's rule with six divisions. 171. Expand the integral of Ex. 170 as a series, and estimate the number of terms necessary for accuracy to four significant figures. 172. Evaluate / ° g ^ ' dx, using Simpson's rule with h = 0-1. Jo x 173. The hypotenuse of a right-angled triangle is measured accurately as 40, and one angle is measured as 30° with a possible error of i°. Find the possible error in the lengths of each of the sides and in the area of the triangle. 174.* By considering / log (a + x)dx, a > 0, show that •'l/a a(oe + 1) . . . (a + re) = a n n\ re", where a n is bounded below by a positive number. Show that a n is mono- tonically decreasing for sufficiently large values of n. (The limit of a n as re -* » is l/r(a).) , , re,! re,! . . . nA , 175. Find an approximate expression for log j , wnere n 1 + n l + ... +»j = n. 176. Show that the coefficient of x n in the binomial expansion of 1 1 = is asymptotically given by — ==. Vl-x v™ CHAPTER VIH * °° tx 177. Prove that if S a„ 2 converges, so does S — . „_1 v-l v 178. If a is a monotonic increasing sequence with positive terms, when does the series - + + - - + . . . converge? Oj a x a^ a^a % a 3 179.* If the series S a v with decreasing positive terms converges, then lim na n — 0. v ' 1 n — y «> 00 TT 180. Show that the series S sin- diverges. u 566 MISCELLANEOUS EXAMPLES 181.* Prove that if Sa„ converges and if 6,, 6 a , 6 3 , . . . is a bounded monotonio sequence of numbers, then ~Zafi v converges. 182.* Prove that if 2a„ oscillates between finite bounds and if 6,, is a monotonio sequence tending towards zero, then Sa„6„ converges. 183. Discuss the convergence or divergence of the following series: (a) S tJl. (6) S (-l>*°° s (8/v>. (c) S ^ 9 . V v V «£) S !^- 6 . (e) S (- 1 )" C0Sv9 . (/) S ( - 1),/Sinve . V V V 184. Find the sums of the following derangements of the series l-i+$-l+l-i+... ioi log2: (a)i-i-i+i-*-i + i-A-A+ •••• (6) 1 + * + *— * — i— i+ + + — 1851 For what values of a do the following series converge? (a) 1 +- i — H .... w 2" 3 4" 5 6* (6) 1 + 1 — 1 + 1 + 1—1+ + — w 3" 2" 6" 7 a 4" 186. Find whether the following series converge or diverge: (a) l + i-i+i'+ *-* + *+!-*+ + -.... (B)l+i— f+i+i— f+t+i— f+ + — 187. Show that (") S 7^ converges. 00 log(v + 1) — logv (6) £ M , ',, converges. „ =2 (logv) 3 » 1 2 3 ... v (c) S — r converges if a > 1 and diverges W „_ 1 (a+l)(a + 2)...(«+v) 8 if a^l. CO J 188.* By comparison with the series 2 — , prove the following test: c = l ^ If lo gwl a »D > 1 + s for every sufficiently large n, the series 2a„ converges absolutely; if ^fo ° < * ~ e f or ever y 8uffioien % ^S 6 », the series 2a„ does not converge absolutely. 2 / 1 V 189. Show that the series 2 1 1 — —i-\ converges. MISCELLANEOUS EXAMPLES 5&7 190. By comparison with the series S — , prove the following test: The series S | a v | converges or diverges according as log(l/»| a nl) log logn is greater than 1 + e or less than 1 — s for every sufficiently large ». 191. Derive the »-th root test from the test of Ex. 188. 192.* Prove the following comparison test: if the series 26 v of posi- tive terms converges, and | g n+l I d« from a certain term onwards, the series So,, is absolutely convergent? if £&,. diverges and a n b n from a certain term onwards, the series Za„ is not absolutely convergent. 193. Obtain the ratio test by comparison with the geometric series. 00 1 . 194.* By comparison with S — , prove Raabe s test: The series S | a v \ converges or diverges according as "(r^i- 1 ) I °n+l I is greater than 1 + e or less than 1 — e for every sufficiently large n. 195. Bv comparison with 2 — -, prove the following test: J v(logv) a The series 2 | a v | converges or diverges according as nlogJlM^ -I- 1 -) is greater than 1 + e or less than 1 — e for every sufficiently large ». 196. Prove Gauss's test: where | iJ„ | is bounded, then S | a„ | converges if (i > 1, diverges if |i g 1. 197. Test the following series for convergence or divergence: / ^ a ,«(«+!) , a(q+l)(a+2) W ^ + p(P+l)^P(P+D(P + 2) ... . , a.B , a(a+l).&(P+l) , a(«+ l)(a + 8). P(p+ 1XP + 8) . W + I7y 1.2. Y (Y+1) 1.2.3. Y (T+l)(Y+2) (6) (2v)! ' (e) v 1 " V*' (d) v(-l)' 568 MISCELLANEOUS EXAMPLES 00 1 198. (o) Show that the series 2 — converges uniformly for x ^ 1 -+- e. (6) Show that the derived series — 2 — — converges uniformly for * ^ 1 + e. v°> 199.* Show that the series £ , a > 0, converges uniformly for c S x :£ 2tt - c. v ° 200. The series <e + 1 3 Va; + 1/ 5 Ve + 1/ " converges uniformly for s ^x ^ N. 201. Find the regions in which the following series are convergent: (a) Ex-'. (e) X%> a < 1. </>s£ «>i. 1*" (A) 2 * v» 1 — x" 202.* Prove that if the series £ converges for x = x , it converges for any x > x a ; if it diverges for x = x , it diverges for any x < x . Thus, there is an " abscissa of convergence " such that for any greater value of x the series converges, and for any smaller value of x the series diverges. 203. If £ -" converges for x = x , the derived series — £ — — — con- verges for any x > x . 204. If a„ > and £a„ converges, then lim 'Za v x v = £<v x—yi-a 205. If a„ > and £a„ diverges, lim 'Za^x" = 00. 206.* Prove Abel's theorem: If £a„X" converges, then £«,,£" converges uniformly for ^ x ^ X. 207.* If Sa^X" converges, then lim £<ve" = Sa^". X—&-X-0 208. Find the rational functions represented by the following Taylor series: (a) x + x 2 — x 3 - x* + x 6 + x e h + • • • • (6) 1 + 2* — 4a? — 5a* + 7a: 8 + 8a:' 1- + • . . • MISCELLANEOUS EXAMPLES 569 209. Show that w i + 5 + i + "- 1 (6)i + J^_ + 1 - 3 - 5 - 7 + ...= * V8. v/ 22. 4. 62. 4. 6. 8. 10 2 1 » 210. Let z = re l9 = »-(cos0+ i sin 6). Prom the expansion = S z v , show that 1 ~ z "=° 1 — J-CO80 S „ „ = S r" cosv0 and 1 — 2rcos0 + r 2 ,_ r sin0 S r"sinv0. 1 — 2rcos0 + r a „_„ CHAPTER IX 211.* Using the expression for the cotangent in partial fractions, ex- pand tzx oot izx as a power series in x. By comparing this with the series given on p. 423, show that s l - < iy»-* (27c)am B 212. Show that S 1 _ (— l)""- 1 ^ 2 " 1 — l)n 2m B v=1 (2v-l) 2m_ 2(2m)! *"' 213. Show that » (_!)■»_ (_ i)«»(2 2 "' — 2)7t 2m R v-l"^™" - 2. (2m)! 214. Prove that a 6 + Z 12 215. Using the infinite products for the sine and cosine, show that . x , /sinz\ " (-l)-i22-i^ (a) log I ) = — 2 ^— * . \ x / „=i (2v)!v (6) log cos* = -^ ^ *>• 19* (1798) 570 MISCELLANEOUS EXAMPLES 216. Using the infinite products for the sine and cosine, evaluate (a) f . f • 5 . f • 9 • tl • if • • • • m o 1 4 1 U U U (0) ^.§.f.$. 9 • 15 • 15- 217. Express the hyperbolic cotangent in terms of partial fractions. CHAPTER XI 218. Find the curves whoso tangent ia of constant length a. (The "length" of the tangent is the length of the portion of the tangent intercepted between the curve and the a;-axis.) 219. Find the curves which are orthogonal to the family y = ce kx . 220. If « denotes the length of arc of a chain measured from a point at which the tangent is horizontal, the form of the chain is determined by the differential equation d dx ,™-l<P& x Show that the equation of the chain is y = c cosh -fa. 221. Integrate the equation for the electric circuit |jj + pi = E, where E = E a sin<o<, and u., p, E w u> are constants. 222. A particle falls towards a point which attracts inversely as the cube of the distance and directly as the mass. Find the motion and time of descent if v = and x = a at t = 0. 223.* Integrate y = — xp + x*p\ where p = X 224. Integrate y = p + logj). 225.* Solve the difference equation "»+» + 2aM n+i + 6 "» = °» where a, 6 are constants, by putting u„ = X n . Show that the solution can be expressed in the form u„ = ar-f + P»»"» where r„ r a are the roots (sup- posed distinct) of the equation X 2 + 2aX +6=0. Show that the form of the solution when 6 = a 2 is u n = cc(— o) n + p»(— a) n . ANSWERS AND HINTS CHAPTER I § 1, p. 13. 1. (d), (e). Show that x satisfies an equation of the type X s + a,?* + . . . 4- a e = 0, where a v . . . , a 6 are integers; prove that * is then either irrational or an integer. 2. Use the irrationality of sin 60° = V3/2. 4. Write ax 1 + 26a; + c as al x + - ) H • \ a) a 7. If a >0 and 6 a — oc 5^0, it is possible to make ax i +2bx-\-c=0 for some value of x if, and only if, 6 a — ac = 0; then use Example 6. 8. The cosine of the angle between two straight lines is ^ 1 in absolute value. 9. Use Schwarz's inequality. 10. Square both sides and then use Schwarz's inequality. The sum of the lengths of two sides of a triangle is not less than the third side. §§ 2, 3, p. 26. 2. (a), (d), (e), (g) odd; (6) even. 3. (6), (c), (A) monotonic; (a), (d), (e), (I), (m) even; (d) and (e) identical. § 4, p. 28. 2. (n + 1) (2» + 1) (2» + 3)/3. 3. (c) Expand (1 + 1)" by the binomial theorem. 4. (a) n(n + l){n + 2)19. (b) Sum from v = 1 to v = n. n/(n + 1). v+ 1 v (c) Sum from v = 1 to v = n. n(» + 2)/(n + !)*■ (v + l) 2 v a 5. 3; 193. 7. J(2ra* + 3» ! — 11» + 30). 671 572 ANSWERS AND HINTS § 5, p. 36. 1. (a) 1; (6) 333; (c) 333,333. 2. (a) 0; (6) oo; (c) 6; (d) o /6 ; <e) 1/3. 4. 19. 5. (a) 6; (6) 10; (c) 14. 6. (a) 25; (6) 2500; (6) 250,000. 9. (a) 0; (6) no; (c) yes; (e) 30. 15. The greatest of %, . . . , a k . 16. 2. 17. Use the fact that n/2 n -> 0. § 6, p. 45. 1. (a) For every number M, no matter how large, there exists aD n such that | «„ | > M . (b) There exists a positive number e such that for every number M there exist numbers n, m greater than M for which | o„ — a m | ^ s. 5. Error is less than ; e = 2-71828 .... n(n\) § 7, p. 49. 1. (a) 6; (6) 15; (c) J; (d) 3. 3. Limits (a) and (b) do not exist; limit (c) exists and is equal to 1. § 8, p. 55. 3. (a) 1/60; 1/600; 1/6000. (6) 1/10(1 + 2| 5-1 1), &o. (c) 1/120(1+ | 5|) 3 , &c. (d) 1/100; 1/10000; 1/1000000. (e) 1/10; 1/100; 1/1000. 4. (a) 1/600; e/6. (6) 1/400; e/4. (c) 1/77600; e/776. (rf) 1/10000; e 2 . (e) 1/100; e. 5. (a), (6), (c), (i), (g) continuous; (e) discontinuous at x = 2, 4; (/) „ x = 3; (A),(fc),(m) „ „ x = (ra + J)tc; W, (j) „ a; = wtc; (0 „ a; = nn, n 4= 0. Appendix I, p. 70. 1. (a) Upper bound _ £|* ( * lower = 0, upp (6) i * — l * — 2» " — x ' (c) 9 * 2 * — 10' " — 3' (d) 19 * 1 * — 10' » ~~ 3> (*) = 2* „ = 0, The quantities marked * belong to the sequence > 0, lower = 0. 0,* „ = 0.* _ A — 2> tf _ 1 — 2 3 — 2' 99 1 — 2 = 1, „ = 0. ANSWERS AND HINTS 573 2. Divide the interval into a finite number of sub-intervals by points o = x , x x , x t x„ = b so close that | f(x) — f(x) \ < e if x and x lie in the same sub-interval. Join adjacent points x = x it y = /(a; f ) by straight lines. 3. The expression —-\x—x t \ + -\x— x t ^ | has the slope zero 2 ^ outside the interval (x t - ly x t ). Add suitable terms of this kind. t + *-f|*-2|+|*-S|-l|*-S|. 4. (a) e/6; (6) e/na"- 1 , » > 0; (c) e 8 /2. Appendix II, p. 75. 2. (a) r= a; (6) r = 2a eos(<p — <p ); (c) r = a/cos (<p — <p ). 3. oos26 = cos 2 8 — sin 2 0, sin20 = 2 sin© cosO; cos36 = 4 cos 3 9 — 3 cos0, sin30 = 3 sin — 4 sin 3 0; cos50 = 16 cos 5 — 20 cos 8 + 5 cosG, sin50 = 16 sin 6 — 20 sin 8 + 6 sinO. 4. (a) — 6i; = n, r = 3; 6 = tt/2, r = 2; = 37t/2, r = 6. (J) 1 + V3 + t(l - V3); = 7t/4, r = 4V2; = ti/3, r = £; 6 = 77t/12, r=2V2. (c) 2; = tc/4, r = V2; = 7tt/4, r = V2; = 2ti, r = 2. (d) 2 - 2» V3; = 5tc/6, r = 2; = 6jt/3, r = 4. (e) ±1; = 0, r = 1; = 0, r = ± 1. (/) ± (^ + ^} 6 = W2, r = 1; = rc/4, r = ± 1. ( ? ) {VV2 + 1 + t-v/^2 - 1}/V2; = tc/4, r = V2; = tt/8, r = «(/2. (h) — v/18 (V3 + »)/2; 8 = 7n/4, r = 3V2; 6 = 77t/6, r = (3V2) a/3 = ^/18. (*) 1, (-1 ± »V3)/2; = 0, r = 1; = 0, ^, ~, r = 1. (Z) y'2{\/V2 + 1 + *VV2— 1}; 8 = tt/2, r = 16; 6 = it/8, r = ± 2. 5. Note that e" satisfies the equation x n — 1 = 0; then factorize x n — 1. CHAPTER H § 2, p. 87. 1 . Use formulae of § 2 and the basic rules: 70/3. 2. Required area may be regarded as the difference between the area under the line and the area under the parabola, taken between the points of intersection of curve and line: 10V5/3. 3. V5/6. 574 ANSWERS AND HINTS 4. J(o» + 4bf*. 5. (a) {(1 + b) 1 *"- — (1 + a) 1+a }/(l + a); (6) — (cosaft — cosaa)/a; (c) (sin a& — sin aa)/a. 7. (6 4 - a 4 )/4. 8. l/(n + 1). § 3, p. 109. 1. For every number a there exists an e such that for every positive number 8 there exists an x for which I x - 5 I g S and a - fW-f&) ^ e _ x-l 2. (a) —!/(*+ I) 2 ; (6) -2x/(x* + 2) 2 ; (c) -4a;/(2a; 2 + l) 2 ; (d) — cos a: /sin 2 a:; (e) 3cos3x; (/) — asinaa;; (g) 2siua;cosi; (h) —2 cosaisina:. 3. (a) 5 has any value; (6) 5=^ + ^/2; (c) S=^?!±3|l±fl). § 4, p. 119. 2. (a) J; (6) J. § 5, p. 121. 1. ~= 0-785. 4 §§ 6, 7, p. 130. a + b ' f/a n + a"-^ + . . . + b n \ r , - 1. (6) 5 = -f - ; (c) 5=V( „ + I ) ; (<£) ^^ /a6 ' 3. (a) /„ = aW»/(l + 1/n), lim /„ = a; n — >■» (6) 7 n = a"+V(« + J )> lim /„ = for — 1 g a ^ 1, » for a > I. 4. | JP(a) — /(a;) | S -I f | /(s + <) — /(*) I *• Use tne uniform con- 28 J _ j tinuity of /(#) in a :£ x ij 6. Also, we may write F(x) = 1 { J" /(«)* +f X+ *f(t)dt}, where c is a fixed number. 5. Express the integrals as limits of sums, using equal subdivisions of « g » ^ 6, and applying Schwarz's inequality (p. 12) to these sums. Another method is to integrate {fix) + tg(x)} 3 ^ and use Ex. 6, p. 13. Appendix, p. 135. 1. Let <f(x) = f\x) where f'(x) S: 0, <p(a:) = e