ADVANCED CALCULUS New Edition A COURSE ARRANGED WITH SPECIAL REFER- ENCE TO THE NEEDS OF STUDENTS OF APPLIED MATHEMATICS BY FREDERICK S. WOODS PROFESSOR OF MATHEMATICS IN THE MASSACHUSETTS INSTITUTE OF T?.cHNOLOGY APREFACE The course in advanced calculus contained in this book has for many years been given by the author to students in the Massa- chusetts Institute of Technology. The choice of the subject matter and the arrangement of the material are the result of the expe- rience thus gained. The students to whom the course has been given have been chiefly interested in the applications of the calculus and have felt the need of a more extensive knowledge than that gained in the elementary courses, but they have not been prima- rily concerned with theoretical questions. Hence there is no attempt to make this course one in analysis. However, some knowledge of theory is certainly necessary if correct use is to be made.of the science; therefore the author has endeavored to ii jtroduce the students to theoretical questions and possibly to incite in some a desire for more thorough study. As an example of the method used, a proof of the existence of the definite integral in one variable has been given; for the multiple integral the proof has been omitted and simply the result stated. The student who has mastered the simpler case is in a position to read the more difficult case in easily accessible texts. existence proofs have also been given for the simpler cases of implicit functions and of differential equations. In these proofs the author has preferred to make the assumption that the func- tions involved ‘may be expanded into Taylor series. This, of course, restricts the‘proof ; but the somewhat immature student gets a clearer idea of the meaning of the theorems when he sees an actual series as the solution. The more abstract concept of a function may well come later. Furthermore, the student is likely to apply his results only to functions which can be expanded into series. * Because of this constant use of the power scries that subject is taken up ‘irst, after certain introductory matter. Here again, fol- lowing the line of simplicity, the author has not discussed series in general. The gain in concreteness for the student justifies this, but the teacher who desires to discuss series of a more general type may do so with the aid of the exercises given for the student. iiiPREFACE The Fourier series are introduced later as tools for solving certain partial differential equations, but no attempt has been made to develop their theory. 4 The subjects treated in the book may be, most easily seen by examining the table of contents. Experience has shown that the book may be covered in a year’s course. FREDERICK S. WOODS Nore ror THE 1922 PrinttNG. In this impression of the book certain improvements have been made. In particular, Osgood’s theorem has been inserted in Chapter I, the discussion of uniform convergence in Chapter IT has been improved, and the treatment of the plane in Chapter V has been changed. PREFACE TO THE NEW EDITION In this edition additional exercises have been inserted at the end of most chapters. Also, in Chapter VI, certain proofs have been made more rigorous; namely, that for the existence of the” definite integral and that for the possibility of differentiating under the integral sign a definite integral with upper limit infinity. Al) the typégraphical errors that have been discovered have been corrected. FREDERICK 8S. WOODSCONTENTS CHAPTER I. PRELIMINARY SECTION PAGE 1, Functions a 1 @. Continuity rae ts teeter cae Outhe derivatives se 5 4: Composite tuncone Gcegr rc eres recoded a 7 bi Rolle's theorem gags ees i G: Theorems of the mean 2c: cee ecg eg ee esas 8 7. Taylor's series with aremainder 2... 1... 1 ee ee ee 10 S.Vhe form sree 16 9. TheformS 2... ee 16 10. Other indeterminate forms... 2.2... 22 ee eee 18 11. Infinitesimals. . . . 19 42. Fundamental theorems on infinitesimals | 5). ls. 22 13. Some geometric theorems involving infinitesimals .. . . 2... . 23 14. The first differential - 28 15. Higher differentials 29 16. Change of variable 82 CHAPTER II. POWER SERIES Ni Denton 38 18, Comparison teat for convergence ©. 2... 1. ee 40 119: Thelratio. test for convergence (act tse cnt teas 41 20. Region of convergence 42 21, Uniform convergence 45 22. Function defined by a power series AB 23, Integral and derivative of a power series... 2... 46 24. Taylor’sseries 2. 2... : 48 26. Operations with two power series . 61 26. The exponential and trigonometric functions... . 2... . 68 27. Hyperbolic functions 55 28. Dominant functions 57 29. Conditionally convergent series 58 CHAPTER III. PARTIAL DIFFERENTIATION 80, Functions of two or more variables 2.2... 2... Le 65 OL Partial detivatives 66 a2) Order iol differentiation: re + 68 88. Differentiation of composite functions»... 2... ee 69 vvi CONTENTS SECTION 84, Euler’s theorem on homogeneous functions ..........~. 35. 36. 37. 38. 39. 40. 41. 42. 43. 44, 45. 47. 49, 51. 54. 55. 56. 57. 59. 60. . Planes . Maxima and minima . The Gamma function Directional derivative The first differential Higher differentials. 2. ee Toei ee CHAPTER IV. IMPLICIT FUNCTIONS One equation, two variables. =. ee ‘One equation, more than two variables ‘Two equations, four variables ‘Three equations, six variables 2 The generalcase 2 2 1. ee ee Jacobians CHAPTER V. APPLICATIONS TO GEOMETRY Element of are SSeaghtline OUTROS eee ees Behavior of a surface near a point Cirvesee esas teeeccea tans Secret ereteaaet eae stere tart . Curvatureand tormion 2g 2 et tt tt et . Curvilinear codrdinates CHAPTER VI. THE DEFINITE INTEGRAL Denton xitonce roots Properties of definite integrals... 0... ee ee Evaluation of a definite integral Lee Gimpeonte rules Change of variables 2 ee Differentiation of a definite integral . 2... 0... 2.400. . Integration under the integral sign... 2... 1. wee finfnite lint ee ee . Differentiation and integration of an integral with an infinite limit . Infinite integrand 2 ee . Certain definite integrals 6. 0... ee Multiple integrale ee i CHAPTER VII. THE GAMMA AND BETA FUNCTIONS The Beta tancon ers Dirichlet’s integrals Special relations»... 2...CONTENTS ix CHAPTER XVI. ELLIPTIC INTEGRALS sRCTION PAGE isi Ineoducion 365 152, The functions sn u, en u, dn u a 867, 158. Application to the pendulum 369 164, Formulas of differentiation and series expansion... . . 871 155. Addition formulas. 2... 2.2... 372 166 The periods 378 167.Limiting cases 1. 815 158. Elliptic integrals in the complex plane... 2... 1. 376 159. Elliptic integrals of the second kind and of the third kind |... 379 160: The function pa) 381 let Applications 882 ANSWERS... ... a 387 INDEX 396ADVANCED CALCULUS CHAPTER I PRELIMINARY 1. Functions. A quantity y is said to be a function of a quan- tity x if the value of y is determined when the value of x is given. Elementary examples are the familiar algebraic, trigonometric, logarithmic, and exponential functions by means of which y is explicitly given in terms of x. Such explicit formulation, how- ever, is not necessary to the idea of a function. For example, y may be the number of cents of postage on a letter and z the number of ounces in its weight, or y may be defined as the largest prime number which is smaller than any number a, or y may be defined as equal to 0 if x is a rational number and equal to 1 if x is an irrational number. ~ It should be-noticed, moreover, that even when an explicit formulation in elementary functions is possible, y need not be defined by the same formula for all values of x. For example, consider a spherical shell of inner radius a and outer radius 6 composed of matter of density p. Let x be the dis- tance of a point from the center of the shell and y the gravita- tional potential due to the shell. Then y is a function of x with the following formulation : y = 2 1p(b?—a”) when x =a, 2 v= 2 xe(s?— 2) —Ste8 when azz=b, () eo —a®) when z>0b. y= 3 So we may at pleasure build up an arbitrary function of x. For example, let y = f(x), where f(z) =42? when 0<2 <1, f(z)=4% when z=1, (2) f(@~)=42+1 when x>1, 12 PRELIMINARY We shall say. that values of x which lie between a and 0 deter- mine an interval (a, b). The interval may or may not include the values a and b, according to the con- y text. In general, however, the inter- vals (a, b) will mean the values of x defined by the statement a=x=b. The student is suppesed to be fa- miliar with the representation of a function by a graph. Such a repre- sentation is usually possible for the functions we shall handle in this book, Fic. 1 although it is impossible for the func- tion mentioned in the third example of this section. The in- terval (a, b) appears in the graph as the portion of the axis of x between.z = a and x = 6, and it will be y f { } }. 1 i t f , ' t ' ' b convenient to speak of a point of the interval, meaning a value of x in the interval. Then x=a and «=b are the end-points of the interval. As men- tioned above, the interval may or may not have end-points. = The graph of the potential function in (D is the curve of Fig. 1. The graph has no breaks and the function is continuous Fie. 2 (§ 2), but the character of the curve and of the function is different in the three intervals considered. The graph of the function in (2) is the curve of Fig. 2. This graph has a break at the point for which x = 1. 2. Continuity. A function f(x) is continuous when x =a for which S(@) is defined if ~X Lim [f(a +h) —f(a)]=0, qa) or, otherwise expressed, if Lim f(a +h) = fa), @) where in either formula the limit is independent of the manner in which h approaches 0. Since h is an increment added to a, and f(a +h) — f(a) is the corresponding increment of f(a), we may express this definition as follows: A function of x is continuous for a given value of x if the increment of the function approaches zero as the increment of x approaches zero.CONTINUITY 3 A more cumbersome definition, but one which brings out the full meaning of equation (1), is as follows: f(x) is continuous for “2=a when if € is any assigned positive quantity, no matter how small, it is possible to determine another positive quantity 5 so- that the difference in absolute value between f(a+h) and f(a) shall be less than ¢ for all values of h numerically less than 6; that is, \fat+h)—f(a)| M for |h!} <6. The definition “ continuity cannot then be satisfied for z = a. For example, the functions — ig. 4) and 7 1 ig. 5) are each Fie. 3 Fig. 4 discontinuous for x = 0. as is shown by the brea i in eech of the curves representing the functions.4 PRELIMINARY The following theorems are of fundamental importance in handling continuous functions: 7 1. If f(a) is continuous at all points of an interval (a, b), it is possible to find a positive number 5 such that in all subin- tervals of (a, b) less than 6 the absolute value of the difference between any two values of f(x) is less than € when € is a positive quantity given in advance. We shall not give a formal proof. It is not difficult to see that if these theo- rems were not true, definition (3) for continuity must fail for at least one point of (a,b), Because of the property o x stated in the theorem, f(x) is said to be uniformly continuous in (a, b). 1. If f(x) is continuous for all values of x between a and b inclu~ y sive, if f(a) = A and f(b) = B, and if N is any valug between A and B, then f(£) = N for at least one value of E between a and.b. Fie. 5 IIT, If f(x) is continuous for all values of x between a and b inclu- sive, then f(x) has a largest value M for at least one value of x betweena Y and b and a smallest value m for at least some other value of «x between a and b. These theorems seem to be inherent in the very nature of continuity and are graphically evident from Figs. 6, 7, and 8. As a matter of fact, how- ever, they are not self-evident and Fig. 8DERIVATIVE 5 are capable of rigorous proof. These proofs lie outside the range of this book and will not be given here. The difference between the maximum and minimum values of f(x) in the interval (q, 6) is called the oscillation of f(z). We can say from I, IV. If f(x) 78 continuous in (a, b), it 1s possible to find a positive number 6 so that in every interval in (a, b) less than 6 the oscillation of f(x) is less than «. 8. The derivative. A function f(x) is said to have a derivative for x =a if the expression a+ h) — f(a) A approaches a limit as h approaches zero in any manner whatever. This limit is called the derivative for «=a and is denoted by Looe pos f(a) (1) =f@. @) In order that the derivative should exist it is necessary that f(x) should be continuous when x = a, for otherwise the fraction (1) would not approach a limit. This condition is not sufficient, as may be seen by considering the function defined by the equations en Aa f(a) =x sin ] when «#40, (8) SO) a As x — 0, sin = oscillates infinitely often between + 1 and —1, but zsin = — 0. Hence the function is continuous for += 0. Using this function in the fraction (1) with a=0, we have 7 hsin a 0 aint h ah and sin 7 does not approach a limit as h-+0. Hence the func- tion has no derivative when x = 0. In 1872 Weierstrass gave the explicit statement of a siiaraiel which has for all values of z the property which sin = has for 2 =0, so that it is known now that a continuous fumetion does not necessarily possess a derivative. Hence when a new function appears in analysis it is necessary to inquire first whether it is6 PRELIMINARY continuous and, secondly, whether it has a derivative. It is only functions which possess these two properties that are of interest in this book. We have discussed the derivative of f(x) for a value a of z. If f(z) has a derivative at each point of an interval, there is thus defined a new function f'(x) by the formula $e) = Lim EDI), 4) hoo h Similarly, we define f” (w) as the derivative of f’() or the second derivative of f(x), f’’() as the derivative of f’’(x), and so on. It is assumed from this point that the student is familiar with the elemeniary process of differentiation. The proof of these ele- mentary processes involves implicitly the proof of the continuity of the function and the existence of the derivative. The student is also assumed to be familiar with the fact that if a function is represented by a graph the derivative gives the slope of the tan- gent line to the graph. ‘The graph of the function =asin™ xz is given in Fig. 9 for positive values of z. For negative values of « the curve is reflected on the line OY. Ji is of course impossible to draw the curve in the close ¥ neighborhood of the point O; but it is clear that if O be joined to any other point P of the curve, the line OP oscillates through an angle of 90°. The curve there~ fore has no tangent line at 0. Th Weierstrass function men- tioned above is represented by a curve which has no breaks, but has no tangent (that is, no definite direction) at any point. These examples illustrate the fact that a graph is at best merely a rough way to represent 2 function, and that conclusions drawn merely from the graph may be erroneous. The graphs are helpful in understanding or formulating a theorem, but an analytic proof is always neressary for rigor. Fic.9 *ROLLE’S THEOREM qT 4. Composite functions. Let y = f(x) be a function of x and let x= o(t). Then, by definition of a function y= F(t). Let tbe given an increment h and let the corresponding increment of x be k. be k= o+h) — 000. if @ is a continuous function of t, k > 0ash—0. Now FO =y=f@), FE+h) =feth, sinoe h and k are corresponding increments of t and x. Therefore FO+h) — FO =fet+h) —f@); whence Lim (Fd+h)-—FOI= Lim Che+k) — f(a). Therefore, if f(x) is a continuous function, Lim [F@+ h) — FO} =0. Hence if y is a continuous function of « and x is a continuous function of t, then y is a continuous function of ¢. Let us now form the quotients F+W = FO Seth fe) i = core — f@) eee o) , whence, by § 8, on taking the a FO =f @)- oO. a) 5. Rolle’s theorem: If f(a)=0, and f(b) =0, then there is some value & between a and b for which f’(~) = 0, provided f(x) is continuous in the interval a = x = bandhasaderiva- Y¥ tive for all values of x between a and b. By theorem III, §2, f(z) has a maximum M and a minimum m in the interval (a, b). If both M and m are zero, f(x) is always zero, its L x derivative is zero (by (2), §3), and ol é the theorem is proved. Suppose that M is not zero, as in Fig. 10, and let f() = Then E+ -JO© fE+h —fE) is negative. Hence ae a Fig. 108 PRELIMINARY is positive when h is negative, and is negative when h is positive. But, by hypothesis, fE +h) —1© h approaches a limit f’(£), which is independent of the sign of h. fa SO =0. Again if M=0 but m < 0, as in Fig. 11, the same argument applies. y The student should notice that the condition that f(z) should have a deriva- tive rules out such graphs as shown in Figs. 12 and 18, for in neither case is there a derivative in the strict sense of the definition when x=c. It is true that in Fig. 12 we may speak of a left-hand derivative and a right-hand derivative, but in so doing we modify the defini- Fie. 11 tion by first restricting h to negative ‘ values and afterwards réstricting h to positive values. In Fig. 13 we may write f"(c)=00, but again the limit of L¢+#) —SO does not exist in the sense of having a definite value. Y Yy oy @ tba ofa ¢ be Fie. 12 Fig. 13 6. Theorem of the mean. J. If f(x) is continuous in the interval a = x = band has a derivative between x =a and x = b, then : $0) — f(a) = (b — a)f'(€), wherea 0 as h—>0, so does | F(z +h) — F(z)|—>0, and - b Fe) =r) O10. Now F(a) =0 inn F(b) = as is seen by direct substitu- tion. Hence, by Rolle’s theorem, F’(é) = 0 for some & between a and b. Thatis, ro -2 From this it follows at once that f(b) =f) + (b—afid), @<& <>) ay which is the theorem to be proved. In (1) we may write £ = a+ 6(b— a), where @ is an unknown proper fraction, and have f(b) =f@) + (b—a)flat+ 6b—a)} (a, ¢«+a—>2a; therefore w: —a? f this fraction is y, by definition, that the value of when «=a is 2a. To obtain a general method for finding this limit we begin by applying Rolle’s theorem to the function (a) JR LO. (0) — o(0)] Ue) ~ Fh (8) — which obviously vashes when x = @ and when « LO) —fa) _ fe) 60) 6@) #®" In (1) let f(a) = 0, eee =0,and b=. We have Hence @<& a, £—» a, and therefore im Jee im £® ot aay eae)16 PRELIMINARY Now unless f’(a) = 0 and ¢/(a) == 0 we have the result f(z) _ f@, oe $e) ~ oa) If, however, f(a) =0 and ¢’(a)=0, we must apply (3) again with the result Se) _ £® _ 7"@ Lim Lim —— = zea @(n) eae GE) o"(@) unless f(a) = 0 and $’(a)= 90. In the latter case formula (8) must be applied again. We may sum up in a rule known as L’Hospital’s rule. (4) To find the value of a fraction which takes the Jorm 6 7 when x = a, replace the numerator and the denominator each by its as and substitute «=a. If the new fraction is also repeat the process. For example, e—2costt+e* .. e%+2sinr—e7* Lim ——.—— = Lim 40 © sin x z+0 sinz+2cosx @+2cosa+e™ |= Zeosz—asinz Jeno 9. The form 2. Consider the fraction S@) $2)’ and let f(a) = % and ¢(a) = % by hypothesis. The value of the fraction for x = a is then defined as the limit approached by the value of the fraction as x increases without limit. We shall prove that L’Hospital’s rule holds also for a fraction which takes the form =. We shali first take the case in which a = 0. From (1), § 8, we may write Sa) -fo _ f£@ $(z) — d(e) o'(E)” where c is a large but finite value of x. From (1) we derive, by simple algebra, (6) fa) _£@ _ o@), o@) $@ 1 _ FO f(z) ( <& <2) qa) (2)INDETERMINATE FORMS 17 We shall now assume that oe has a te Aas E> o. He may consequently take ¢ so large that TO and therefore differs from A by less than any assigned positive quantity «. This fixes c. Then f(c) and $(c) are finite, and x may be taken so large that A ©) a fo) differs from unity by less than any assigned positive quantity €2. We then Mave, from (2), Sx) =(A+m)(1+ m2), where Imf <4). [ml Then, by L’Hospital’s rule, as ne"-2— n(n —2)0"-4 a a Proceeding in this way it appears that eventually x will dis- appear from the numerator of the fraction, no matter what n is, and therefore Lim 2"e-2? = 0. z0 Again, the difference sec x — tan x becomes 0 — 0 when z= 1—sinz cos © , nla But sec —tan x= which becomes 2 when z=. Therefore 0 2 Lim (sec x — tan z) = Lim —. i ue ie aE cos x 2 An expression of the type Lar? may, when x = a, give rise to forms 0°, «0° 1°,INDETERMINATE FORMS 19 and the like. Such an expression may be reduced to a type 2 0 or 2 = by the use of logarithms. Thus, we write u =a. Then log u = $(a) - log f(x). if Lim (a) - log f(z) can be found by the previous methods, the limit approached by w can be found Considerasanexample (1— 2)? when «= 0. A plausible procedure would be to place « = 0 and, obtaining 1°, to say that this is 1 since any power of lis 1. But. this would ignore the fact that we are interested in the limit of 1 (1 — 2)? as x approaches 0, which we have defined as the value of -the function when x= 0. We therefore write 1 u=(1—2)*, eek log (1 — x) 1 == log (1-2) =—-——.. log u 2 log (. a) iB By L’Hospital’s rule, , log (1 Lim mC) ae 40 Therefore ~ Lim _ u=-1, 240 and Lim u =e! 240 11. Infinitesimals. An infinitesimal is defined as a variable which approaches zero as a limit, When two or more infinitesimals ap- proach zero at the same time, they may be compared by consider- ing their ratios. We say an infinitesimal @ is of the same order as an infinitesimal aif B Lim — a k, a) where k is a finite quantity different from zero. An infinitesimal 6 is.of higher order than an infinitesimal o if Lim 80, 2) 720 PRELIMINARY As an example, consider an infinitesimal angle a (represented in Fig. 15) and describe an are of a circle PQ of radius unity. Let B=PN=sina, P y=NQ=1-cosa. By a well-known theorem which is used in deriving the formula for the derivative of sin 2, Lim == a Fig. 15 Therefore PN is of the same order as a. Also Lim 2 = Lim oe Hence ¥ is of higher order than a. A measure of an order of an infinitesimal may be given as fol- lows. If @ is taken as an infinitesimal of the first order, then a’, a3, at,--- are called infinitesimals of the second, third, and fourth orders, respectively, and £ is an infinitesimal of mth order with respect to a, where 7 is positive, if Aus ae 8) where k is a finite quantity not zero. For example, consider y = N@ of Fig. 15. We have i—ecosa_ 1 ren a Therefore NQ is of the second order with respect to a. Equations (1) and (8) lead to the forms B=ka+ae, (4) B=ka"®+a"e, 6) respectively, where € is another infinitesimal. In each case the first term on the right of the equation is called the principal part of the infinitesimal. An infinitesimal and its principal part are obviously of the same order and differ by an infinitesimal of higher order than either of them. If we denote by 3; the principal part of 8, formulas (4) and (5) become € B=h + Big = Bi + Bie. (6)INFINITESIMALS 21 Theré are two convenient ways to determine the order of an infinitesimal 6 with respect to a: One is to evaluate Lim ie «40 a” by L’Hospital’s rule, so choosing that the limit is finite and not zero. A more expeditious way, when £ can be placed equal to a function of a, is to expand 8 = f(a) by Maclaurin’s series ; then, if B= ka" + katt! + R= ko" + ka"! + o"+2p, we have ‘ & =k+ha+a?P a Lim a k, «+0 a” which shows that the degree of the first term in the expansion of 8 determines the order of 8. We shall illustrate these methods by inquiring by what order of infini- tesimals an infinitesimal arc of a circle exceeds its chord. In a circle of radius a and center O (Fig. 16) let AB be an infinitesimal arc and AB its infinitesimal chord. Draw the radii OA and OB, and draw ODC perpendicular to AB. Let the angle BOC=9@. Then AB = 208, Fic. 16 AB = 2asin 6, and : AB — AB =200 —2asin 0. Take AB as the a of our general discussion, and let B= AB — AB. We have Lim 8 = Lim 2089 =2e8in 8 «9, 20% 040 200 Hence 8 is of higher order than a. Similarly, Lim £ =0, Lim 8 1 , ono Dba? and therefore 6 is of the thitd order with respect to a.22, PRELIMINARY The second method is to place B=200—2asin0 We have accordingly shown in two ways that the difference between an infinitesima! are of a circle and its chord is an infini- tesimal of the third order with respect to the arc. 12, Fundamental theorems on infinitesimals. There are two im- portant theorems involving infinitesimals; namely, I. If the quotient of two infinitestmals has a limit, that limit is unaltered by replacing either infinitesimal by its principal part. To prove this let us place B=8itBie, a= a+ue, in accordance with (6), §11. Then By eickeBieneBi lic a. ataue alt+e whence Lim > 8 tim a a II. If the sum of n positive infinitesimal has a limit as n increases indefinitely and each infinitesimal approaches zero, that limit is unaltered by replacing each infinitesimal by its principal part. Let #1, 82,---, Bn be a set of n positive infinitesimals, and let i, O2,- ++, a, be their principal parts, also positive. Then OG + OE, and LB = Yor + Dare. (2) Let y be a positive quantity which is equal to the largest absolute value of ¢;., Then, for any 7, -Y=eG2%, and — Yo; S 4: = VO, the multiplication being allowable since a; is positive. Then — Yor = Yeas vdeo (3)INFINITESIMALS 23 By hypothesis, as n increases indefinitely, Za; approaches a finite limit and y approaches zero. Therefore, from (3), Lim Ya, 0, and therefore, from (2), Lim $}6;=Lim Sia. The theorem stated is known as Duhamel's theorem. The proof assumes that all the infinitesimals are positive. The theorem is obviously also true if all infinitesimals are negative but is not necessarily true if the infinitesimals are not all of the same sign. The proof also assumes that Lim ¢,=0 no matter how 7 depends upon n. For the important applications to definite integrals the theorem may be replaced by Osgood’s theorem, as follows : Let a1 +a2+++++an be a sum of n infinitesimals, and let ox differ uniformly by infinttesimals of higher order than Ax; from the > elements }(x:)Ax; of the dejinite integral [ S(x)dx where f(x) ts con- Jo tinuous in the interval a =x =b. Then the sum a, +a2+-+++ +o approaches the value of the definite integral as a limit as n becomes infinite. To prove this, let ai =f(xi)Ax:+ §; Az, where [f.|< €, by hy- pothesis. ‘Then | Sa Spaaz|< (YAn=e(b—a). But (§ 54) we can make | Svteoae - L "ade Therefore | Sa~ [ *Ka)dx whence Lim Sai = f S(z) dx. <é ++ On) =8. (2) a0 Now 8; is the projection of a; on AB. Hence, by the law of projections, B; = 0; c08 8: where @; is the angle between the chord of length a; and the chord AB. Hence (2) is Lim (1 sec 0; + B2 sec 02+ +++ + Bn sec On) = 8. (8) ne Under our hypotheses sec 6; is always positive, although 6; may be negative. Our hypotheses allow us to apply the theorem of the mean to any portion of the curve between A and B. Therefore Fic. 17INFINITESIMALS 25 there is some tangent which makes an angle 6; with AB. Hence, if @ is the largest angle in absolute value which any tangent makes with AB, we have 1 = sec 0; = sec ¢. Therefore Bit B2ot+-+++ Bn = Br sec 6; + Be sec 2-+---+ Bn sec On sec (8: + B2+---+ Bn), or, from (1) and from (8), lss=slsec ¢, (4) WA WA or Ls} s sec em (6) where the equality signs could hold only if AB coincided with AB. ‘This result is true for any finite arc for which the hypotheses that have been made hold. It remains true as B approaches A. But then sec @ approaches unity. Hence we have Lim? =1, (6) ino or s=l+le, which was to be proved. I. Under the hypothesis made in I the perpendicular distance from one end of an infinitesimal arc to the tangent at the other end is an infinitesimal of higher order than the are, and the length of the tangent q from the foot of this perpendicular to the B point of tangency is an infinitesimal of the same order. Consider again the are AB (Fig. 18) with the properties as before. Draw the tangent AT at A, and drop the perpendic- LP ular BT. Let AT'=¢ and BT=h and angle BAT = Then t=lcosa, Fig. 18 and ‘ = E COs a, istese Let s > 0 and apply (6). We have Lage See 0: : fo) s-08 520 826 PRELIMINARY The relation between ¢ and h may also be found by means of coérdinate axes and Maclaurin’s series. Take as origin a point on the curve (Fig. 19) and as OX the tangent at O. Let y y =f). be the equation of the curve. Then f(0) =0 and f’(0) = 0, since the curve P passes through O and the slope of the tangent at O is zero. Then, by Mac- laurin’s series, Mu x 2 x O y=I@) =I") + I'"0) 5 + B a But OM=xz and MP=y. Hence it appears that MP is of the second order with respect to OM unless f’’(0) = 0. III. Except for infinitesimals of higher order than the lengths of the arcs, an infinitesimal right-angled curvilinear triangle obeys the same trigonometric laws as a siraight-lined right-angled triangle when the hypotheses of I are satisfied. Consider a triangle 4 BC (Fig. 20) whose sides are ares of curves which satisfy the hypotheses of I and which may be made to approach zero together. Let the ares intersect at a right angle at C and let the angle at A be ¢. Draw the chords AB, BC, and CA. The angle between the chords AB and AC is @ + & and that between BC and CA is 90° + €2, where € and e2 are infinitesimals approaching zero as the sides of the curvilinear tri- angle approach zero. Then BC _ sin(¢+a) AB sin (90° + €)” which we may write as BC 4B BC _ sin(@+ 4) Fig. 20 AB AB BC sin (90° + €2) where BC means the are BC, and so on.INFINITESIMALS 2F Taking the limit as A, B, and C approach coincidence, and using theorem I, we have — Lim = = sin , AB ¢ a oe a or BC = ABsin $+ AB. In a similar manner, AC = AB cos b+ & 4B, BC = AC tang + €5/ AP? = AC? + BC? 4 As an example of the use of the foregoing theorem consider an ellipse with foci F. and #” (Fig. 21). Let P and Q be two points in- finitesimally near on the ellipse, and draw PF, PF’, QF, and QF’. By the definition of the ellipse, PF + PF’ =QF + QF. With Fo as 2 center and a radius FQ construct an arc of a circle cutting FP in S. With £ as a center and a radius F’Q construct an are of a circle cutting F’P in R. Then SP~RP = QF— PF—(PF’—QF’) =0. Then in the infinitesimal triangles SQP and RQP SP =QP cos SPQ, RP = QP cos RPQ, except for infinitesimals of higher order. Therefore cos SPQ = cos RPQ except possibly for infinitesimals of higher order. But the angles SPQ and RPQ are independent of the position of Q. Hence SPQ = RPQ; and, since SPQ = FPA, we have the result that in an ellipse the tangent at any poin* makes equal angles with the focal radii to that point. FIG. 2128 PRELIMINARY 14. The first differential. Consider now a function f(x) which has a derivative f’(x). If Az is an infinitesimal increment of x, then the increment Ay = f(x + Az) — f(x) is an infinitesimal, since f(x) is continuous. Now Lim 48 arvodg 7 5 whence Ay =f’ (z)Aa + € Ax. (1) Aside from the values of x for which f'{x) is zero, or infinite, this is of the form (4), § 11, and f’(x)Az is the principal part of Ay. This we shall call the differential of y and denote it by dy. The case of the independent variable x, however, is different. For in that case f(x) = 2; therefore y=, and formula (1) is simply Ax = de. (2) There is no possibility, therefore, of separating Az into two parts; in other words, the principal part of Az is the whole of Az, and we may take this as dz. Summing up, we say: The differential of an independent variable x is equal to the incre- ‘ment of the variable; that is, dr = Ax. (3) The differential of a function y = f(x) is the principal part of the increment of y and is given by the formula dy = f'(a)de. i (4) Suppose, now, we have y = f(x) and z= $(t); then y= F(i). Now, by the definition above, we have dt = At, dx = $'(t)dt, dy = F'(tydt. Substituting for F’(?) the value derived given in (1), § 4, we have dy = S'(e) 6 de; whence dy = f'(x)dzx. (5) Note that this is the same form as (4); but in (4) dx is the entire increment of x, whereas in (5) dz is the principal part of that increment. The result is, The differential of a function y is given by the formula dy = f'(x)dz, whether x is the independent variable or not.DIFFERENTIALS 29 We are now ready to write the derivative as a quotient ; namely, dy (2) =e. 6 l@=7 (6) In differential form (1), § 4, becomes dy _ dy de (D dt dx dt 15. Higher differentials. If w= f(x) and v = ¢(x) are two func- tions of x which possess derivatives, we have, by well-known formulas for differentiation, £urn=smr+oe, J(u) = F@)60) + 109"), 2 (2) fede) fave, dz\o] [o@)P i‘ whence, by the definition of the differential, d(u+v) = du+ do, wD d(uv) =v du+u do, (2) ‘u\ _vdu—udv a3). 2 Let us apply formula (2) to df = f'(x)dx. We have d(df) = d[f’ (x) dx + f'(x)d(dz). (4) Now we have, by the definition of § 14, aff (a) =f" dx. It is natural to express d(dy) by d?y and it is customary to express (dc)? by dx”. This must not be confused with d(z”), which, by § 14, is equal to 2 x dx. We have, then, d?f =f" (x)dx? + f'(x)d?z. (5) This is called the second differential of f. Formula (5) contains a factor d’x which has not been defined. However, if x is a function of another variable t, so that z= Fb, we have, by another application of (5), the result d?x = F’'(t)dt? + F'(t)d*t,30 PRELIMINARY but here d’¢ is still to be defined. It is evident then that (5) is not sufficient to define the second differential of the independent variable. We are accordingly free to frame that definition as we will, and we say The second differential of the independent variable is by definition zero. With this and (5) it follows that when zx is an independent variable, af =f" (a)de?; (6) but when z is uot an independent variable, af = f"(w)dx® + f(@)dPx. The fact that the second differential has different forms aecord- ing as x is independent or not is in striking contrast to the fact that the form of the tirst differential is always the same. Second differentials must therefore be used with more care than the first. We notice that. when < is the independent variable we have, from (6), a, de oS 3) whereas when x is not an we have, from (5), siogy CFL @ ee _ Pf de of de I) = ea o which agrees with (7) only when dx = 0. 2 In spite of this the symbol is used to represent the second derivative f(x) even when x is not the independent variable. af This may be explained by interpreting 2 nm as a symbol for aug) ~ tl) dz a(t) de de, = aff'(z)) (xy) _ f(a) - Then de rege: (9) 2, It follows that at is used in two’ senses: first as a symbol for the second derivative and secondly as the quotient of d?f as given in (5) by dx*. These two senses agree only when z is theDIFFERENTIALS 31 independent variable. The context usually makes clear which @y sense is meant. Asa matter of fact, the use ‘ot 2 pe as a derivative is more common than the other use. i. If in (8) we place y = f(x), and interpret a as a symbol for the second derivative f’(x) and rot as the quotient of differen- tials, we have (0) This formula may also be obtained by direct differentiations, thus: ay dy _ di dx da” & (oe 4(@ adhd) du de dt dt der} a= “de a i (a (a as before. ‘This result may also be obtained by. dividing the numerator and the denominator of the fraction in (8) by (d#)*. @ third, fourth, and higher differentials are d(d2x) = d*x, Tee diz, ete. Since, if « is the independent variable, d?x = 0, it follows at once that d"x =0. That is, The nth differential of the independent variable is zero if n is greater than 1. The higher differentials of a function of x are found by oper- ating with the laws (1), (2), and (3). Thus we have df = f(x) dx? + f'(w)d?x. (il) "Then Bf = fi'(ajdx? +3 f" (ade dx + fl (a)de. (12) It is to be noticed that (12) gives f'"(@) = dy as as)32 PRELIMINARY only when x is the independent variable. Otherwise, if ¢ is the independent variable we have, from oe aS _ ge “ Bas@(Z+sr@ZSrroZ a As in the case of the second derivative, the expression ay dx’ is used for the third derivative even when z is not the independent variable. In this case es. af e(a) and Fa is not the quotient of d?f by da’. From (12) we have : Bf —3 f" (x)dx da — f'(x)dx 10) = useage By means of (8) of this section and (6), §14, this reduces readily to the form dx(d°f da — df dx) ~ 8 [df de — df d?x]d?x I"@= mee If ¢ is the independent variable, we may divide all the terms of the numerator of (15) by dt® and obtain a result in derivatives which may otherwise be obtained by direct differentiation. Similar results are readily obtained for the fourth and higher differentials. 16. Change of variable. The methods and formulas just obtained may be used to solve certain problems connected with the change of the variables in a given expression. 1. Let there be given an expression involving y, we , ana = where the last symbol means a second derivative, and let it be required to replace y by z where y = f(z). We have at first, by § 4, (45) apy OE =f'@ i a and then, by Geet aiaiananDy Sa pa(Z)+roS @)EXERCISES + 33 @& 2. Let it be required in an expressién involving y, a, and oy to replace x by ¢, where x = @(t). dy 1 dy _ at _ 1 dy We have de de FO a (8) dt and, again by direct differentiation or by (10), § 15, dydx dyde dy. dy, vy a ave @eO-4rO a ee i dt 3. Let it be lake to interchange x and y in an expression involving 2, ay, and at This is a special case of (4) in which 2 = y, and therefore wy a: oy =0. We have dy 1 t-z 6) ay Pe dy ay? de? Ta ® Ga) EXERCISES 1, Prove that if f(z) and $(z)are continuous at + = a, then f(z) + (x), {[@) , f(x) - &(z) are continuous at x= a, and that ——— is also continuous at c= a unless (a) = 0. @) 2. If y=f(z) is continuous when += a, and z= ¢(t) is continuous when = b, where a = $(b), show that y = f[¢()] is continuous when t=b. 8. Show that the theorem of the mean as given in § 6 may be trans- lated into the theorem of the mean for definite integrals; namely, ['P@de=b-aF®. @ 0, n+1 lant] and that |Ri< ——*_ 1 __. when x < 0. @eynatart 7. From the result of Ex. 6 estimate the error made in +: : log 1.2 from three terms of the series. How many terms of the + are sufficient to compute log 1.2 accurately to six decimal places? 8. From the result of Ex. 6 how many terms of the expansion oi log (1 + <} are sufficient to compute log .9 to five decimal places? 8. Show that in the expansion of log : ae Qeret a iRi< @ebadea when x > iene wepars when x < 0, where x is the exponent. of x in the last te. retained in the expansion. 10. From the result of Ex. 9 how ad terns of the expansion of L log — es 11. Show that in the expansion of (1 + x)* n)an + and that IRI< are required io compute log 25 to four deeima! ulaces? when x > 0, k(k-VD-e k=”) | (w+ DL +2)" wei! and that [Ri < a"*| when z <0, ifu-k+1> 12. From the result of Ex. 11 find how many terms of the binomial series are sufficient to compute W162 to four decimal places. 18. By integration find an expansion for sin@' x.EXERCISES 35 Sartet . sin? x 14. By division find an expansion for - . . =, Pe 16. Find an expansion for f/“e~*dz. * 16. Find an expansion for "cos «*de. Find the limit approached by each of the following functions as the variable approaches the given value: sec 3 x m “sec 5a 2 ag, VEZ, y+ 0, x oe 24, —_, 2 2. tan x 2 os, 28 x ae 25. ==, x —» © (n positive). 28. ’ 2x-—> (n positive). 27. Find the limit zpproached by ayn” $ ayat—) box™ + bia™—1 + as z—> ©, where rand m are positive integers, under each of the three hypotheses n < m,n =m," > m. 28, Show that for all positive and negative values of n Lim x"e-* = 0. 29. Show that: for all positive values of n and m og 2)" Lim aan em 30. Show that for all positive values of 7 and m Lim x*(log x)" = 0. 81, Evaluate Lim 2 b . 84. Evaluate Lim (1 + az)*. 40 $2. Evaluate Lim x*. 35. Evaluate Lim (sin x)#"*, 2-0 7 1 aoe oa i ee 86. In Fig. 16 find the order of CD.36 PRELIMINARY 37. Let T be the intersection of the tangents to the circle (Fig. 16) at A and B. Find the order of TC — CD. 88. Two sides AB and AC of a triangle differ by an infinitesimal «, and the angle 6 at A is an infinitesimal of the same order as a. What order of infinitesimals is neglected if the area of the triangle is taken as 4 AB? 6? as} AB- AC- 62 39. If y = f(x) is a curve in Cartesian codrdinates, show that the area bounded by the curve, the axis of x, and two ordinates separated by an infinitesimal distance Az differs from y Az by an infinitesimal of higher order. 40. If r= f(@) is a curve in polar codrdinates, show that the area bounded by the curve and two radii making an infinitesimal angle A@ with each other differs from 4 r?A@ by an infinitesimal of higher order. 41. What order of infinitesimals is neglected in taking as 4 rr?h the volume of a spherical shell of finite inner radius r and infinitesimal thickness h? 42. A parallelogram has an angle which differs from 90° by an infini- tesimal of first order. What order of infinitesimals is neglected by taking the area of the parallelogram as the product pf two adjacent sides? 43. Show that in a hyperbola the tangent makes equal angles with the focal radii drawn to the point of contact. 44. Show that in a parabola the tangent makes equal angles with the focal radius to the point of contact and the line through the point of contact parallel to the axis. 46. A circle of radius a rolls on a straight line. A point P on its rim describes a eyeloid. If P moves to P’ by an infinitesimal rotation d¢, show by the method of infinitesimals that PP! =2asin }$d¢. Note that the linear dispiacement of the circle is a dg, and that the motion of P takes place in a direction normal to the line from P to the point of contact of the circle with the straight line. 6. Find d?y when y =sin x? under the two assumptions (1) that x is the independent variable; (2) that x =e‘. In the latter case first use formula (5), § 15, and check by substituting for x in the given value of y. 2 47. In Ex. 46 find © as a quotient of differentials and also by direct differentiation. a 48. Given y =e, find dy (1) when 2 is the independent variable; (2) when x= log z and z is the independent variable. Verify by substi- tution and direct differentiation. 49, In the equation dy, /dy\? 3 — 4944 (We aay a-ha 2Y+ a-at=o place y = sin 2.EXERCISES 387 50. Inthe equation 2224 4 4 4 (92 — ny =0 place y = 2%. dx?" de 81. In the result of Ex. 50 place x = 2-Vi. 62, In the equation a2) 22 4 amt py =o place x = cos 6. a. ae 58. Show that if the equation of a curve y = f(z) is transformed to polar coérdinates by the substitutions x =r cos 6, y=r sin @, the derivative a) becomes ee £9d6? +2 d0 dr? — r dr d0 (cos 6dr—rsin Odd)? where @ is the independent variable. 54, Show that the formula for the radius of curvature of a plane curve y = f(z), namely, ‘dy\? \t [1 +(# | ax! y dx? [de® + dy? dx d?y — dy Px for the curve x = fi(t), y = fo(t) where ¢ is the independent variable. 55. Show that the formula for p given in Ex. 54 for the curve y = becomes i) becomes : oF dr\? 24 (Sy) [r @) | aiceecae +2(%) eee in polar codrdinates. 46} 6? 56, Show that the formula for p in Ex. 54 becomes dx\?}4 aa = dx . dy? if the curve is taken as x = $(y). 57. if z=1 cos 0, y=r sin 0, show that Note that this is the expression for the tangent of the angle which a curve makes with a line from the origin.CHAPTER II POWER SERIES 17. Definitions. The expression do + a1% + aoe? + azz? +--+ tage ™+ees a is a power series. lf the number of terms is finite, the pcwer series reduces to a polynomial; if the number of terms is unlim- ited, the power series is an infinite series. It is with infinite series that we are concerned in this chapter. The series (1) is said to converge for a given value x = 2; if the sum of the first x terms approaches a limit as n is indefinitely increased. The limit is called the value of the series or, somewhat inacex.wately, the sum of the series for x = x. A simple and i- portant exampie of a power series is that of the geometric series do + aor + dont? + aye? +++» gH +++ 2) The surn of the first terms of the series is, by elementary algebra, (3) Now if x is numerically less,than unity, the last fraction in (3) approaches zero as » increases indefinitely, and the sum of the first » terms approaches the limit oo + l—z Hence the geometrig s ries (2) converges for any value of x in the interval -l<2<1. The series (2) defines, then, the function + ka = in the interval (— 1, 1), but does not define the function outside of that interval nor at its ends. : A series which is not convergent is called divergent. As an example consider the harmonic series for values of x 1,1,1 1 Lt gtgt gt tates 33ABSOLUTE CONVERGENCE 39 Now ati gt+et+ht+s>h and in this way the sum of the first n terms may be seen to be greater than any multiple of 4 for sufficiently large n. Hence the sum of the first » terms does not approach a limit, and the series diverges. Return now to the general series (1) and set « = 2, obtaining Qo + arty + eas? + agar? +--+ aqm"+---. 4) Let.us replace each quantity by its absolute value, obtaining the new series {ao|-+ jax |]21|-+[a2[]217|-+]as{far3|-+--- +lan||ai"|+- (5) We shall prove that if (5) converges, then (4) also converges. Let s, be the sum of the first terms of (4), and Jet s,’ be the sum of the first n terms of (5). Now s, contains a certain num- ber of positive terms whose sum we call p, and a certain number of negative terms whose sum we call — g,, so that Sn = Pn — Ine 6 The positive terms of (4) appear in (5) unchanged, whereas the negative terms of (4) appear in (5) with signs changed. Hence Sn! = Pat dn @ Now as 7 increases, $n’, Dn, and g, each increases, since each is positive. Suppose s,’ approaches a limit A. Then p,,, always increasing but always less than A, must approach a limit B,* and, similarly, g, must approach a limit C. Hence, from (6), s, approaches a limit B—C. We have ac- cordingly proved the proposition that if (5) converges (4) also converges. It is te be noticed that we cannot prove conversely that if (4) converges (5) does; for s, in (6) may approach a limit even if p, and q,, separately do not. Summing up, we say that o series which converges when each term is replaced by us absoluie value converges as it stands. It is said to converge ebsolutely. ‘The determination of the absolute convergence of a series reduces, then, to the determination of the ne as ovident that a yuantity which "ways increases either vesomes fines, Ueeth,40 POWER SERIES convergence of a series of positive quantities, and for that we shall find two tests useful: the comparison test (§ 18) and the ratio test (§ 19). 18. Comparison test for convergence. If no term of a series of positive numbers is greater than the corresponding;term of a known convergent series, the first series converges. If no term of a given series is less than the corresponding term of a known divergent series of positive numbers, the first series diverges. Let Cr tcetes-tes++entes: qd) be a series of positive numbers, and let Mi+ Mo+Ms+---+Mnat+--- @) be a known convergent series such that Cn = My (3) for all values of n. Let s, be the sum of the first x terms of (1), S, the sum of the first n terms of (2), and M the limit of S,. Then, from (8), all terms M,, are positive and therefore S, < M, and also from (3) s, = S,, so that we have & 1, we may compare (5) with the series 1,1 1,1 1,1 1 ltotatptptptptget (6) where there are four terms equal to *» eight terms equal to > and in general 2* terms equal to a Now no term of (5) is greater than the corresponding term of (6), and (6) is the same as ge a ee a PT BP he : This is a geometric series with ratio equal to 2 ‘y\p-1 #7) <1 Hence the series (5) converges when p > 1 and diverges when p2l 19. The ratio test for convergence. If in a series of positive nur. bers the ratio of the (n + 1)st term to the nth term approaches a limii i as n increases without limit, then, if L <1, the series converges; ij L > 1, the series diverges; if L = 1, the series may either converge or diverge. Let Cr+ cat est ++ +ent enti te: @) Cott _ be a series of positive numbers, and let Lim L. ‘We have three cases to consider: *"* ° Case I. L <1. Taker, anumber such that L 1. Since the ratio approaches L as a limit, this ratio eventually becomes and remains greater than unity for n sufficiently large, se: 2 > m. Then Cm+1 > Cmy Cm+2 > Cm+i > Cm, ete. Each term of (2) is then greater than the corresponding term of the divergent series Cm + Cm Om bry and therefore (1) diverges. Case II]. L=1. Neither argument given abc e is valid, and experience shows that the series may either converge or diverge. As a first example consider 2,38 4 n n+l ltgtgstgit tga tga te The ratio of the (x + 1)st term to the nth term is a - which n approaches the limit 3} as increases without limit. Hence (4) converges. : Again, consider 22 33 nr (n+ yt Mtotate tat (~+1)! The ratio of the (n + 1)st term to the nth term is ee (42) n” n/ which approaches the limit e as n increases. Hence (5) diverges. As a Jast example consider 1 OP (5) 1 1 1 +t tot @ppet (6) ? The ratio of the (w+ List to the nth term is (4) » which approaches 1 as n is increased. Hence the ratio test fails, but it has been shown in § 18 that (6) converges if p > 1 and diverges ifp=1. 20. Region of convergence. We now proceed to determine the values of x for which a power series converges. We begin with 1+REGION OF CONVERGENCE 43 the theorem tiat if a power series converges for x = 2), it con- verges absolutely for any value x = xy such that |z2|<|eaj. We assume that the series Qo + art, + age? +--+ fan" ++-+ a converges in any way and wish to prove that Qo + Ayatz + O22? +--+ + Opto” +++ + (2) converges absolutely if |22|<{2|. Since (1) converges, all its terms are bounded ; that is, there 1s a. positive number 3 such thai for all values of n Jao" | } Hence the region of ee is determined as ee L L As an example take first the series L4+2e48a?4---4 ner? + (n+ Da + The ratio of the (n + 1)st term to the nth term is as x. The limit of this ratio as » increases indefinitely is x. The region of convergence of the series is -1 |a|. Let a b the suin-of the first » terms of (1) be = 0 Rr denoted by s,(x) and the sum of the Fic. 28 remaining terms by r,(z), where Tn(L) = And” + An 41Z"*! + dng oart? foes, Place Ry(x)=|Gnx"| [dng it™t! | + anyon" t2|4---. (2) ‘Then Ir.(x)| = Ra(z). (3) If (1) converges absolutely, it is possible to take n so great that for any assigned positive quantity € R,(z) < & (4) but the value of nv in (4) depends in general on x. However, if n is so determined that R,(b) <6, then with the same value of n and any value of z in the interval (a, 6) Ry(2) < Ra(b) <€ since |xz|<|b|. Hence, from (3), Irn(a) |<. (5) A series is said to be uniformly convergent in an interval (a, b) if, when e has been chosen, a value of » can be found, independent of x, so that equation (5) is true for all values of x in the interval. We have proved that the power series is uniformly convergent in the region (a, 6). The boundaries of the region may extend as closely as we please to the boundaries of the region of convergence. 22, Function defined by a power series. We shall now prove that @ power series defines a continuous function of x for values of x within the region of convergence of the series,46 POWER SERIES Any value of x in the region of convergence determines another definite value; namely, the limit of the sum of the first n terms of the series. This limit we define as the value of the function and write f(x) =a) + ait + ae? + asx +--+ page" +--~. (1) It remains to prove the continuity of f(x). For that purpose let us write (1) in the form F(©) = $n(e) + Trl), (2) where s,(x) is an algebraic polynomial and r,,(r) is an infinite series whose value is the difference between s,(x) and f(z). Let x be a value of x in the region eh . Pa er ceca aaa Lara of convergence and consider an in- Ou 8 2,298 R terval (v1 ~ 6, +8) (Fig. 24). By a the property of uniform convergence, if € is any assigned positive quantity we can take n so great that Irae) < 5 for all values of x in the interval (2: — 6, 21+). This fixes n in (2). Now take x +4 in this interval and form Sle +h) = 8n(@ +h) + ale +h), 8) where Irae + h)| < From (2) and (8) we have L@ +N) S(t) = 8n(a +h) — 8n(t) + rae +h) — a(x). (4) Now s,,(x) is an algebraic polynomial and therefore continuous. Hence h can he taken so small that. [sn +h) ~ SO1< 5 Then in (4) M@ +h) — f(x)|s|s.(@+h) — s,(x)|+] raz + h)|+1rn(@) [5 that is, [f@+h) -f@l Therefore n can be taken so large that 2\r-! n(2) anX"; and since the series with terms a,X” diverges, so does the series (4). We have now in two ways proved that the series (4) converges in the interval —-R<“ ae" H+ (5) defines f(x) for x within a region of convergence (— R, R) (Fig. 26). The series enables us to find the value of f(x) and all its derivatives for z= % when 2 is in the region of on oO RK convergence. Placing x’ = x — xy we RT one t may obtain a new power series Fic. 26 do! + ay"! + ate'r!? + ag’e’? + ++, (6) which will converge in a region — R’ (2) be a second function defined by another power series in which the coefficients A, are all positive and such that An > |anl- (8)58 POWER SERIES Then the function (x) is said to dominate the function /(x) in the region of convergence of the two series. For a given f(x) there are an infinite number of dominant func- tions. A simple one may be found as follows: Let r be any positive number in the region of convergence of (1). Since (1) converges absolutely for z=, there is some positive number M which no term can equal or exceed in absolute value; that is, lan|r?< M. (4) Consider the function M M M OG) eee ee (5) y—2 7 This series converges when |x| < r and, by (4), M @ > lanl. Hence ¢(z) is a dominant function in the region (— 1, r). 29. Conditionally convergent series. We have been concerned in the previous sections entirely with absolutely convergent scries. It has been noted, however, that a series may converge even when it does aot converge absolutely. Such a series is called condition- ally convergent. For example, consider tie series 1-$+4-d4d-$4e. @ The sum of the absolute values of the term of (1) is the harmonic series (§17), which is known to diverge. Hence (1) does not. converge absolutely. However, it does converge as it stands. Tc show this let us plot on a number scale successive values of lee ook gg (-) in The appearance of the graph (Fig. 27) may be compared to the swing of a pendulum, every s with an odd subscript corresponding to a swing to the right, every s with a 5, an even subscript corresponding to oa ae a swing to the left. Every swing in one direction is less than the previous swing in the other direction, since the numbers added or sub- tracted are growing smaller. Hence se, increases with n but remains less than 1, and therefore sz, approaches a limit L.* +See footnote on page 39. S,=1CONDITIONAL CONVERGENCE 59 Similarly, s2n+1 is growing smaller, but is always greater than 4. Hence s2n41.approaches a limit L’, But 1 2n+1 so that Lim|s,+1—8,|= 0 and therefore L = L’. The series (1) is a special case of a general type of series of importance for which we have the theorem [S2n+1 — S2n|= If in a series of alternately positive and negative terms each term is less in absolute value than the preceding term, and the absolute value of the terms approaches zero as a limit as the number of terms increases without limit, the series converges. We have do — a: + A2—Agt-+-+ (1) an + (— W"dngr tes, (2) in which the a’s are positive numbers with Qn41 <@, and Lima, = 0. ne The proof that this series converges may be given on the lines on which the convergence of (1) was proved. It should be noticed that the limit approached by the sum of » terms of a conditionally convergent series may depend upon the order in which the terms are arranged. Suppose A is any arbitrarily assumed number, which for convenience we take as positive, and let us arrange the terms as follows. Take at first just enough of the positive terms in the order in which they appear in (2) so that their sum exceeds A; then just enough of the negative terms so that the sum of all the terms taken shall be less thar. A; then just enough positive terms so that the sum of all the terms taken shall again be greater than A; and so on. In this wey s, is alternately greater and less than A, and it is easy to see, since the terms are decreasing in absolute value, that S, approaches A as a limit. The limit of the sum of m terms of an absolutely convergent series, however, does not depend upon the order in which the terms are taken. To prove this let Mo + 1 +2 +A3+--++an++-> (8) be a series of positive numbers and let bo + bit bet bst-+-+ bates 4)60 POWER SERIES be the same series with the terms in different order, but.so that each term of (8) is found somewhere in (4) and each term of (4) is found somewhere in (8). Let s,,’ be the sum of the first m terms in (4). Then » may be taken so large in (3) that all the terms in s,,’ are found in the n terms of (8). Hence Bn’ < 82 t Ta 2POWER SERIES 20. Show th Ayo tar taat- ++ ayers converges if Lim Wa, <1 and diverges if Lim Wa, > 1. 31. Show that stole) + a) + tele) bee banat) eee, where vo(x), wz), --+ u(x} are functions of (x), is uniformly con- vergent in a region (a, 5) if convergent: series of positive numbers Mo+ Mi + Machos++ Mates ean be found such that jue)! < My for all values of x in the interval. $2. If Uta tagt--- ta, te-: is an absolutely convergent series, prove that a sine +agsin2e+--+-+a,sin ne+es- is uniformly convergent in the interval (0, 2 3). 33. Prove that 2 ae ES a + 4 sy tripe aie Gary is not uniformly convergent in any interval which includes x = 0. 84. Prove that any series Ug (E) + Mo(Z) b+ UZ) to { continuous functions u;(r) which is uniformly convergent in an terval (a, b) defines a continuous function in the interval. 8. Show that any series of continuous functions which converges uniformly in the inteeval (a, ) may be integrated term by term be- ‘tween Linnil ‘h lie in that interval. 36. Show that any series of continuous functions may he differentiated er the resulting series is one of continuous functions which converges uniformly. apa Show by consideration of R that the following series really repre gent the function and determine the region of convergence: Salt ete betes at 88. cos & 89. log (i -+ x) 40. (L+ a)" =14n2¢ 202)EXERCISES be Obtain expansions of the following functions by integration of a series and determine the region of convergence: ce 2 de 42. 3in7? x =f" 48. tanta = : So Mi- “OL + x? J Vien Obtain series expansions for the follgwing integrals and determine the region of convergence: 45. ["e-" dx. Jf 44. log (e+ V1 + Find by combinations of elementary series already obtained expan- sions of the following functions: sing! x 48. 52. 1 - Bt 53. sine, sec r= — aor 54, eM" =. goes", 5B. tae, 50. cot 2 = 8%. . a sin < Dewees 51. e* sec x. 57. log (1 + sin z). Yipee. vnev 58. Show that sin (c+ iy) =“ +" sin 2 +i cos x. 59. Show that oe te# ¥ : cos (x + iy) = —F*— cos » sin x. 2 60. From formula (9), § 26, assuming that the real and imaginary parts of two equal complex quantities are equal, derive the formulas cos nf = cos"9 — =) at nin — 1)(n — 2)(n — 8) 4 2 d a n(n — 1)(n — 2) = 3i 61. Apply the method of Ex. 60 to find cos 36 and sin 36. 62. Apply the method of Ex. 60 to find cos 56 and sin 56. 63. Prove formulas (6), § 27. 84, Prove formulas (9) and (10), § 27. 85. Prove formulas (12) and (13), § 27. 86. Construct the graphs of sinh x, cosh x, tanh 2. 87. Find formulas for sinh 2 x and cosh 2 2. cos"~?@ sin? cos"~ 49 sint@—---, sin nO = ncos"~'@ sin 6 cos*~*4 sin’ +--+.64 POWER SERIES 68. Find formulas for sinh ; and cosh = 69. Prove that sinh (x + iy) = sinh x cos y + i cosh x sin y, cosh (x + iy) = cosh x eos y + i sinh x sin y. 70. Show that if ao = 0 in (1), § 28, the function is a dominant function. 71. Show that in (1), § 28, (y [ra(e)| < M———- x ae 7 72. Show that the series 1 1 1 Vet en et see v2 Va V4 converges conditionally but not absolutely. 78. Show that the series Uy bug tuster buy tees nm sinc where uy =. Sf een ae, converges. 74. Cauchy's integral test. If the terms of a series @; + a2 + a3+ ++ +ay+--- are all positive and decreasing, and if a constantly decreasing function f(z) can be found such that f(n) = as, show that the series converges if . the integral f ‘{(z)dz approaches a definite limit as Roo and diverges if it does not. 75. Apply Cauchy’s integral test to the series 1 1 1 Bog 2) * Boga)? +" * iaiogayeF*™ 46. Apply Cauchy’s integral test to the series cot“! 14 cot? 2+ cot 3 +--+CHAPTER III PARTIAL DIFFERENTIATION 30. Functions of two or more variables. A quantity f(z, y) isa function of two variables if the value of f is determined by the values of a and y. The values of x and y may be represented in the usual way, as codrdinates of a point on a plane. Then to every point of the plane is associated a value of f, and we may speak of the value of f at a‘ point (a, y) as a convenient way of saying the value of f for the number-pair (x, y). This manner of speaking may have a physical meaning; as, for example, if f is the intensity of illumination or electric potential at each point of the plane. The method is useful, however, when there is no physical meaning of the function and it is simply an abstract function. The function f(x, y) is continuous at a point (a, 6) for which it is defined if Lim f(x, y) = f(a, b) () eee independently of the manner in which x approaches a and y approaches b. As an example of a discontinuity, consider ep +e The function is not defined by (2) for «= 0, y= 0. It is in order, therefore, to complete the definition of f(x, y) by assigning to it a definite value A for z= 0, y= 0. It is not possible, however, so to choose A that f(r, y) is continuous at (0, 0). For let (@, y) approach (0, 0) along the straight line y= mz; then along this line I@ w= (2) and by changing m the funétion on the right may be given any value whatever. Hence f(z, y) will not approach A for all ways in which z~-»+ 0 and y-+0. Hence f(z, y) is not continuous at (0, 0). 6566 PARTIAL DIFFERENTIATION The definition of continuity may be more explicitly given as follows: Let ¢ be any assigned positive number no matter how small; then if f(z, y) is continuous at (a, b), it is possible to find a number 6 so that [fa+h, b+k) —fa, b)|0, k-+0. Since by hypothesis f,, and fy, are continuous at (a, b), we have, from (11), Say(a, b) = Sye(a, b), (12) which was to be proved. From this result the statement that the order of differentiation is immaterial for any number of differentiations or variables read- ily follows, of course with the proper assumptions as to continuity of the functions involved. For example, since alex) ~ Gy) we have, by replacing f by 4, aa (2) 22(%); dy Ox\dx]~ dx Oy \ Ox, and again, interchanging the a of two differentiations, é oy ale oe Gx (i) @f by ox? so that $8. Differentiation of composite functions. We shall consider in this section a function of any number of variables z, y, z,--- when 2, y, z, etc. are functions of other independent variables 7, 8, t,- The number of variables in each set is immaterial. For defi- niteness, however, we shall consider S@, v) and fix our attention on some one of the independent variables, say t. If tis given an inerement At, then 2, y, and f receive inere- ments Az, Ay, Af, and Af =f t+ Az, y+ Ay) — f(z, v) = Say + 6 Aa + €2 Ay, qd) i Ae +> whieh is obtained trom (5), 7 81, by writing h = Az, k = Ay and noting that the coefficients of h and k differ by infinitesimalswo PARTIAL DIFFERENTIATION from f, and f,, respectively, under the hypothesis that these functions are continuous. Divide, (1) by At and take the limit as At > 0. if t is the only independent variabie, oo Pees ge vere ee Tota, & of and we have af _ fda | of dy, @) dt ex dé : oy dt ff, however, there are other independent variables besides t, m8, igg AE 28 Tim r= pe him a= 9’ ete» jae af by and we have dy ae (8) Tf we differentiate (2) with respect to ¢, we ie ay _d (He af dx #(2) dy i. ai? dt\éx/ dt * dx dt? * dt by) dt os ae 4) Now xi is a function of x and y, and therefore (2) may be applied, with f replaced by of Then we have ar a /afy @fde, OF dy at \3z/ da Oy at Similarly, d d (of af de PF dy di $(2)— da oy dt * dy? dt” Substituting in (4), we have df Of (de fF dady , of (dy) of dx of a i ma) ein att oy ak eaet wa Similarly, starting with (3) we get, when « and y depend upon sand t, oF _ os (2): 2 Of bx dy, OF 5(2). af Ox | Of Oy oe ax? Oa Oy Gt Ot" dy? bx at? * dy at? ef — Of dx dx OF (24 oy , x zu). Of Oy dy Os Gt Ot Os ay? ds at of ey it by Os at D (6) as ot Ou? 0s Ht" Ox 4) a ne Obviously (7) reduces to (6) if s = 1.COMPOSITE FUNCTIONS UW Ina similar manner expressions for the third and higher derive tives may be found. We shall consider special cases of importance with changes in notation. CasE I. f(u), where u = (x). Here formulas (2) and (5) take the following forms: Lapa, x Cy a As an example, let it be required te show that f(x" of isfies the equation y x —x—=0. Weplaceu=2?-+y*® Thea oy f(a? + y?) = Se a and, from the equations given above, =2aflu), = ce 2 af’), by whence the required result follows. Case III. f(u, 0), where u = di(x), v= fol. Formulas (2) and (4) may be written with u change of © to t: y to v, and t tox. Case IV. f(u, v), where w= d(x, y), v= Wr, Y)- ie Formula (3) with a slight. change of notation gives us On and a uf (uy; formula (6) gives = 7 92 5 and oF and formula (7) gives of . 7 a oy? dx 3: x oY12 PARTIAL DIFFERENTIATION As an example, consider f(x — y, y— x) and place u=x—y, v=y—zx. Hence, from (2), 4 _ of af on ~ Gu 60 af _ _ af by eu" oe * from which it appears that f(« —y, y— x), no matter what the form of f, satisfies the equation a 42 Again, consider z = fi(x + at) + f(x — at) and place u = x + at, v=2x—at. Applying (6) we have bx? du? * ao? Bz ah | 2 Vhs, aa ete ae! whence it appears that z satisfies the equation Ce Ca =a ot? ex? Case V. An important application of Case IV occurs when we have a function f(x, y) in which we wish to replace x and y by polar codrdinates r and 6, where xz=recos@, y=rsin 6. (8) Then f becomes a function of r and @ and, by (3), of Of Ox, Of Oy _ af FF Be Oalon gy Ore on oy . Of fbn, Of Oy a of Des obey 0 ae oye By solving these equations we obtain HW ay B08, en Or 00 (10) 8 = sing 4 HF e088 dy ér "00 +COMPOSITE FUNCTIONS 73 These last equations may also be obtained directly from (8), written in the form aaa ee ey éx Or dx 00 Ox Or Vx? $y? oe ap af _ af or, aaa af _y ez : oy ‘Or Oy 00 dy rN + ¥P On +y? Or @r 06 086 where to get me ou Qe On we write r= Ve? +y’, 6= tan. Again, by use of (6), we may get #2 ef af 3 pa2° 0876 + 2 oy cos Osin 6 + 5 sin’, OF 2 OF ene 2 _OF i 2 FF sage 36? aa? sin 6-27 Diy ee eae By? 089 =r ZL eosg—rZ sino, : ax °° oy , Ori, Sal aa Oe oe OdenaUn ae? Or? a? +y? Gr 86 (x? +9)? 06? (x? + 2)? ev et ay ar @? ty?) 80 (2? +9)? of _ OF wg OF oy fF ay? artat ty? Or 08 +a)? 28? + PP fg Fw + ; rare Ore? +y?)? G0 (aw? +”)? In this way we verify the important relation St of of 1497 1a, (12) by? Or? 7? 60 or 84, Euler’s theorem on homogeneous functions. A function f(z, y) is said to be homogeneous of the nth degree in x and y if the mul- tiplication of x and y each by a factor \ multiplies the function by 5 if, symbolically, #(d2, dy) = XY/(2, 9). a Thus x? —3 zy+4y? is homogeneous of the a degree, Va? + y® is homogeneous of the first degree, and = ey each homogeneous of the zero-th degree. = »e are4 PARTIAL DIFFURENTIATION Euler’s theorem in its simplest form is as follows: If f(x, y) isa homogeneous function of the nth degree, then aes 2a +y Wg 7 (2) To prove this differentiate (1) with respect to A. We have, placing u = Az, v = Ay, and applying (2), § 33, of of Saya nrr-y, ata, mr"—'f. (3) This is true for all values of \. It is therefore true when \ = 1. Substituting \ = 1 in (8) gives (2), which was to be proved. Differentiate (3) again with respect to 4. We have Z 2 (eZ tuZ)s=nm—mer-rs, and, placing \ = 1, (eZ petY 3 Z)i= n(n — If. (4) So by successive differentiations we have @ a\F (cZ+ud)rana- ke De &) These results may obviously be obtained for any number of variables, so that we have as the general form of Euler’s theorem a é a i - (eZtugeeZe---)pamn-Y---@—k+ De. 35. Directional derivative. Let P (Fig. 29) be any point of the plane at which f(z, y) is defined and y Ze has partial derivatives g and z. 2 . Draw any curve PS through P, take Al Ay Q any point on PS near to P, and Y= let PQ = As, where s is the length of the curve. Let Ar and Ay be the inerements of x and y corresponding to As, and let Af be the change in f as (x, y) moves from P to Q. Then Lim a is the rate of change of f along the curve PS. Also ae Az dx Ay : ds. Lim Re ge 8 Lime gy ain Fig. 29DIRECTIONAL DERIVATIVE 15 where a is the angle which PT, the tangent to PS at P,-makes with a line parallel to OX. But from (2), § 33, af _ofdx , ofdy_ of ot. Weccoe ds py dagen @) It appears that if the point P is fixed and therefore ag and x are fixed, the value of of depends on the angle a@ and not on the ds equation of the curve PS. Hence (1) is called the directional derivative of f in the direction a. As a changes, P being fixed, the directional derivative changes. ‘Therefore we may write 2 0S a ++ a sin a = F(a). ox oy of When a = 0 we have F(O) = Be’ TT wT and when w= gwe have r() = i, of of . That is, be and ay are the rates of change of f parallel to OX and OY respectively. The directional derivative is zero when of fee: Pla) == cosa + FF eine =03 ex cy a that is, when tana =— 2. (2) of oy This determines two values of a, a and oj +7, differing by 180°. Between these two values there must be, by Rolle’s theorem, at least one vaiue of a, say a2, for which F(a) has its maximum value and whence16 PARTIAL DIFFERENTIATION An easy substitution from (8) in (1) gives \. (ay, (ay F(a) = \) + (Fy. 4) From (2) and (8) it follows that tan a; tan a2 =—1, and therefore the directions a, and az are at right angles to each other. We may interpret these results geometrically still further. Let us imagine that we have marked on the plane of (z, y) those points for which f(z, y) has a constant value c. By connecting such Points we have a plot of the equation f@, y)=e which will be in general a continuous curve. Let the same thing be done for the equations F(z, y) = 02, > f(a, y) = es The plane is then covered with a set of curves called contour lines, along any one of which f(x, y) has a constant value. Thus, for f(z, y) = x? + y? the contour lines are concentric circles x? + y? =e, and for f(x, y) = zy the contour lines are the equilateral hyperbolas zy =c. The expression ‘contour lines” is borrowed from a topo- graphical map, where such lines give the projections on the plane of the map of points at which the height above sea level is constant. Along a con- tour line the rate of change of f is zero. Hence the angle a, determined by (2) gives the direction of the contour line. Suppose now we take two contour lines, and let the difference between the values of f on the two lines be Af. Draw a curve PN (Fig. 30) perpendicular to both lines and any curve PS mak- ing an angle ¢ = SPN with PN. Let PN = Anand let PS = As. Then, by § 18, An= Ascos ¢+€, where ¢ is an infinitesimal of higher order than As. Now Af is the same, whether taken along PN or PS. Hence Fre Fic. 30 A x 7 x cos } + 1, and in the limit Ge cos . (6) ds” dnGRADIENT ° qT This shows that & < a + and therefore a is the maximum rate ds ~dn dn of change of f at the point P. Therefore the direction of PN is the angle ae given in (3). As special cases of (5) = PP (6) by -£ sin a, (7) ef zy, A and therefore (2) +(4Y ~(£ (8) d; which agrees with (4). The quantity o is called the gradient of j. To sum up: The gradient of a function of x and y is the maximum rate of change of the function at a point. I é takes Bee. tna direction normat The extension to a function of three variables f(x, y, z) is obvi- ous. We construct in space the contour surfaces F(x, y, 2) =e, along each of which fis constant. We may use Fig. 30, interpreting PN and PS as drawn between two such surfaces, PN being normal to both. Then, as before, yf a pole £9) eda cos , g : df : showing that -s is the maximum rate of change. Special cases ‘of (9) are to a contour line and is equat to »\( 2 (# ay +(2 a (10) where «, 8, y are the angles made by PN witi the axes of x, y, and z respectively. Then -(# a [ ( an) an (+BY +) since cos? a + cos? B + cos? y = 1.* * The etudent to whom this relation is not familiar will find a demonstration in 6 45,78 PARTIAL DIFFERENTIATION The gradient of a function of three variables is the maximum rate of change of the function at a point. It takes place in a direction perpendicular to the contour surfaces and is equal to ( 36. The first differential. Considering a function of two vari- ables f(z, y), let z take an increment Ax =h and y an increment Ay =k. Then f takes an increment Af, where of of Af= 5, bet 5 Aut Ant eAy, () as has been shown in § 33. The third term here is in general an infinitesimal of higher order than the first term, and the fourth term is in general of higher order than the second, but there is no means of comparing Az and Ay as infinitesimals. However, we shall take the first two terms of (1) and call them the differential of f, writing a= 2 ye F * ay. @) We now complete the definition of . by saying that if x and y are independent variables, dx = Ax, dy = Ay. (3) Then (2) takes the form 8 df => dx+ by dy. (4) Expression (4) is called the total differential of f(x, y). The methods and results of this section may obviously be ex- tended to functions of any number of variables. For example, for f(, y, 2) af as df= Oe dx + cay y+e 6) This definition has been based upon the ae that x and y are independent variables. We need to examine the cases in which this is not true. We consider first the case in which z=hlh, y=d2(0, so.that S(@, y) = Ft).DIFFERENTIAL 19 Here t is the independent variable. Hence, by § 14, dt= At, dx=dyi'()dt, dy= oz’ (dt, dF=F(bdt. (6) af af t 18) cx dat Fb ger But, by (2), § 88, F’() on? @+ by 2'(t). Multiplying through by dt and applying (6), we have again Z dF =df= 52 + By which is the same form as before. Again, let us suppose that x and y are functions of three inde- pendent variables u, 7, w. Then f is a function of these same variables. From, (5), since'w, v, and w are independent, Ox Ox ox =, +5, eta, (7) a 2 ay 4 dy = Oa du+ as dv + on dw, (8) ee df = bu du + do dv + bu dw, (9) But, from (3), 938, 2F _ 2F 2 | Of dy ” Ou dz du dy du af _ of de, Of oy 0” Bx Ov" By bv’ af _ of ae, a ov dw dz dw dy'dw Substituting in (9), we have ._ of (ox ox ox of (oy a= E(u +S do + = a) + oe (gta whence, by (7) and (8), df = 2 dz+ # dy, 2Y ay 4 aw); n+ He do + 2H a); again the same result as before. It is clear that the results obtained are independent of the number of variables involved, and we have the following theorem: 1. The differential of a function f(z, y, z, +++) is always of of of . an + 5g Wt a eT 2 whether x, y, Z,-++ are independent variables or not.80 PARTIAL DIFFERENTIATION We shall rlext prove the following theorem: IL If fy, %- The relation f(x, y, 2, -- -) = ¢ cannot exist when 2, y, 2, --- are independent variables unless it is an identity. We will therefore ¢, where c is a constunt, then df == 0. suppose that x, y, 2,:-- are functions of independent variables u,t, w,---. By substitution f becomes a function of w, v, w,---, which is necessarily an identity, and, by I, f df= F ayy Lint Zs... ou ov But since wu, tv, w,+-- are independent variables, « may be changed without changing any of the others or the value of f. ‘Therefore flu, 0, w, -«) =e, Fut Au, v, w, ++) 1 Ws) — SU, v, W.--)= tek At 2 wy) ~ Sly % we) Au In a similar manner of 0, of =0, ++, av ow and hence df = 0. Therefore, by I, =F aye bay df= +i ts et: It is important that the student should understand just what is meant. Consider, for example, S@yvn=2+y¥4+7=0’, which defines a sphere in space. Here z, y, and z cannot be inde- pendent variables, and evidently of af Bp 2 HO, By Be 22 #0. But a. y, z can be expressed in terms of two independent vari- ables u and v, where z=acosusinvy, y=asinusiney, z=acosv. =2y+#0, a Then S(x,y, 2) = (a cos u sin 0)? + (a sin usin v)? + (a cos v)? = is an identity, and z =0, of =0. a0DIFFERENTIAL 81 of Hence & ut Liv 20 ou ov for all values of du and dv, and of af be dx e+ 3y S ay ips = = for all values of dx, dy, dz which are consistent with the given equation. I. If by any means we hare found that df = Xda, + Xedt2+---+X,dtj, then X;= ——, where the partial derivative is to be taken under the dn; assumption that all variables except x; are const int. The proof is simple. If all variables except x; are constant, all differentials dx, =0 except for k=7, and the above equation becomes df = X;dx;; See whence It does not follow, however, that any expression such as Xi dx, + Xodtz+-+-++ Xndaq which may be written down is equal to the differential of some function. In this connection we will prove the following theorem : IV. The necessary and sufficient condition that an expression Mdzr+Ndy, where M and N are functions of x and y, should be the exact differ- ential of some function f(x, y) is that M and N should satisfy the equation OM ON o By = or" (10) Suppose, first, that M dx + N dy is an exact differential, so that we have df= Mdz+Ndy. Ok _ of Then = 5 N= ay’ omM_ Ff _ ON dy dx dy oz" so that condition (10) is necessary. and82 PARTIAL DIFFERENTIATION Again, let us assume that (10) is satisfied and we wish to show that a function f(x, y) may be found so that ox To do this we may first integrate M dx, regard y as constant, and have fs dz = $(x, y). We now write f(x, y) = $(a, y) + ¥Y), (dy whére yy) is a function of y only, and we shall show that it is pos- sible to determine y so that f is the function required. Frorn (11), Of of 2), ae By ay t YO which must equal N. yy Hence VYy)=N- By (12) This equation will be absurd unless the right side is free of x, since the left side is free of x by hypothesis. The condition for this is that the partial derivative with respect to x should be zero; that is, that . . an _ a = E/E f wae] = 2 ox iy dylex ~ ey’ ox which is exactly (10). Then (12) will determine ¥(y) and (11) will determine f. Hence the condition (10) is sufficient. Extending this to three variables, we have the theorem: V. The necessary and sufficient condition that Pde+Qdy+Rdz should be an exact differential is that P, Q, and R satisfy the equations 0Q ROR _ OP. (13) dz by’ Ox be In the first place, if Pdr+Qdy+Rdz=df, ae then P= ae’ = By’ whence equations (18) immediately follow.DIFFERENTIAL 83 Again, let the conditions (18) be satisfied. Then, by theorem IV, Pdx+Q dy =d¢, where ¢ is a function of 2, y, z and z is to be regarded as constant. We forin the function L,Y, 2) = b@, yD) +H) (14) so that V@=R-S. (15) (cod ox! oR oR a (<6) dy dy dz by ge (ay a But these are just the conditions (13). Hence (15) determines y¥(z) and (14) gives the required f. It is to be noticed that it is possible to have in applied mathe- matics expressions of the type dw = M du +-N dv, expressing relations between infinitesimal increments dw,du,and dv, where M and N do not satisfy the condition (10). Then w is not a function of wu and », the partial derivatives of which are M and N. For example, if H is the heat, U the energy, p the pressure, and » the volume of a gas, a small amount of heat dH added to the gas causes an increase dU in the energy and does an amount of work p dv in expanding the gas, and we have dH =dU+ pd. The right-hand side of this equaticn does not satisfy (10), since pis not constant. Therefore H is not a fanction of v and U in the sense that it is determined when v and U are given. In fact, the amount of heat in a gas depends not merely on its present state but on the manner in which that state has been reached.84 PARTIAL DIFFERENTIATION Again, suppose a force with components X and Y applied to a particle. Denote by dW the infinitesimal amount of work done by displacing the particle from (z, y) to (c+ dz, y+dy). It is easily shown that aW=Xdr+¥dy, but it does not follow that W is a function of x and y. In fact, that this may be so it is necessary and sufficient that the force should be such that ax _ ay, oy x ‘These equatiohs are satisfied by many of the common forces. Such forces are called conservative. But other forces, of which friction is an example, do not satisfy the condition. These statements do not contradict theorem IIT, since it is there assumed that f is a function known to exist. 37. Higher differentials. We have df= Fae +5 a y ei q) whence d(df) = Pe +2 “ Pn “(Lv tay of a F op 2, y From (1), replacing f by On or oy we have OH FS ayy OF az) =a + aay ‘Of _ _@f af aa)= Bar by Ot ay rnd these values in (2), we have ap ena one af o oa 2 af= Sas +257 ay dy + oe ay +o F ity (3) Equation (3) does not define d?x and d*y. If x and y are func- tions of other variables, say r and s, then an equation similar to (3) would give d’x and dy but leave d?r and d?s undetermined. It is therefore necessary to define the differentials of the inde- pendent variables, which we shall do as in § 15; namely, All differentials of the independent variables higher than the first ‘are taken as zero.TAYLOR’S SERIES 85 Hence in (8), if x and y are independent variables, the last two terms disappear; but if x and y are functions of other variables, these terms must be retained. Expressions for d"f (n>2) are readily formed in the same way. 38. Taylor’s series. We wish to extend the results of §7 to f(e, y) under the assumption that at (a, 6) the function f(z, y) is continuous and has continugus partial derivatives. Let us place z=ath, y=b+ kt, where a, b, h, and k are constant and ¢ is a variable. Then f(a, v) = fat ht, b+ kt) = FO. @ Then, by Maclauri#’s theorem, i: _ t t S(z, y) = FO) + (Ot + FO) atoct FO (0) ra R (2) By (2) and (5) of § 38 we have FO= Fete Oatia) vy an Bt ay 24 _ (2 PD = at Bik tS 5=(iZ tee Zyr where the meaning of the symbols on the right of these equations is obvious. Similarly, wn ay a, OF 2_ OF af Fr’ = =WS ga tke +8 hk an ay t ® He (s+) a o\" is Oy the ee So, in general, F' o=(s cree) @) This may be proved by showing by direct differentiation that if (3) is true for any value n =k, it is true forn=k-+1. Then, since (8) is certainly true for n = 3, it is always true. a ay m9) = (= ae Therefore F® (0) = (i a +k as where the subscript indicates that the values ¢ = 0, or r =a, y=, are to be substituted after differentiating. Substituting in (2), we may write rom FO thE + EL) I++ Anne end ayite.86 PARTIAL DIFFERENTIATION This is true for all values of t. Place’ =1. Then, from (1) and the last result, we have Kew) = flat h, b+ B= fla, 1) +(h Z+ed)s (n+ ays. +5 (ne rer )irR © This result may be written in another form by placing h = x — a, ka=y-b. It is easy to show from § 7 that 1D a ye =o (ha the 5 wait ast ap w” . where the subscripts indicate that we are to substitute x = &, y = wherea | ain]. If (9) is known to converge for x = 7, y = p, where r and p are positive, a dominant function is easily found. There is then a number M so that lacn|rip® M 3) By (10), lace] Fp whence, by (10), Aix>|aiel and (13) dominates (9). EXERCISES Find the first and second partial derivatives of these functions: 7 1 log (2? + 9%). 2. tant 2. 8. 4 ev sin-H(r ~ 9). way 2 Verity x ce < c for the following functions: 6 224. 6 logVz?+y?. t.sin-t4. 8. esiny. at+y x 1 af. ar a Wh aTR sin (x? + y?), os : 1 7 10. Iff = y? + 2 ye, prove that 2? Liye a4 =2y, i 7 1. If f= V2? — 7? sin” 4, prove that 2 Ly Fay, = 2 yy? 1 oF, oF 12. If f = log (x? + y?) + tan 4%, prove that 5 Be haa 0. . Vv 1av, Lav 18. If V = e** cos (a log r), prove that => + = — —> =a? ——. (ner. et ccat tov toa de 28, Given x =r cosh 0, y=r a 9, prove that av jay “7008 + Ore 29. If x=f(u, v) and y= $(u, v) are two functions which satisfy the equations af _ ae of _ ae tor Op sou and V is any function of < and y, ian that atv, av _ (av af any) But Bee Ge =) + @ 5” find the contour lines and the direction and magni- tude of the gradient. 381. If an electric potential V is given by V = log Vz? + y’, find the direction and magnitude of the maximum rate of change of V. 32, Find the direction and magnitude of the gradient of a potential [eae at the point (0, a).90 PARTIAL DIFFERENTIATION 83. Find the direction and magnitude of the gradient of wae sin 2 + pe sin 3x at the point @ 0). 34. Show that in polar codrdinates the rate of change of a function f of g, 1 along a radius vector is Zz + and normal to a veetor it is >< 35. Show that in polar codrdinates the directional derivative is oa te 7 a ae with ie radius from the origin, 86. Show that in polar codrdinates the gradient is we 9g y, where y: is the angle made by the direction con- \00! 87. Show that the sum of the squares of the directional derivative in any two mutually perpendicular directions is constant and equal to the square of the gradient. 38, If y= ¢(z) is any curve, show that the directional derivative along this curve is @ c Show that the following differentials are exact, and find « function of which each is the differential : 43. (22y? — 22)x dx + (222 + 2%)y dy + (x?y? — 22 + y? — Dede. 44. (yte—b—e)de + (e+4—-—0¢—a)dyt+ (x +y—a—b)dz. 45. (ytaQutytadet+ @t+x2yt+e+ax)dy ta@ty@etat yds. 46. G- S)tr+ G-4 2 )ay + G- By as 47. Show that a force directed toward the origin and inversely pro- portional to the square of the distance from the origin is conservative 48. Show that a force directed toward a center and equal to any fune- tion of the distance from the center is conservative.CHAPTER IV IMPLICIT FUNCTIONS 39. One equation, two variables. We are accustomed to say, somewhat roughly, that an equation S@, y) =0 @ defines 7 as an implicit function of x and is equivalent to the equation y=f(2). We shall proceed to inquire just what this means, to examine the hypothesis underlying the statement, and to put the state- ment in a more scientific form. In the most elementary cases equation (1) can be solved to obtain y. For example, . a +y—a=0 gives y=VaF— 2; 3 : but a little reflection shows that as soon as equation (1) becomes complicated such an elementary solution is impossible. We need to show, then, that y in (1) satisfies the definition of a function in the sense that a value of x assumed in (1) determines a value of y. This statement, however, is by no means self-evident. We sha’ proceed to prove it in the case in which the solution may be expressed as a power series. We shall assume that equation (1) may be satisfied by x = 20, y = yo, and that Ge v= Without loss of generality we may take ro = 0, yo = 0, as this amounts merely to a change of coordinates. We also assume that f(x,y) may be expanded into a series, so that we may put Sle, y) = biox + bovy + b20x? + duty + boay? +--+, (2) where the term boy is omitted since, by hypothesis, f(0, 0) = 0. Equation (1) becomes byox + Bory + boox? + busy + boxy? + +++ =0. (8) 91 #0.92 IMPLICIT FUNCTIONS \ Now bo = (2) ; and since, by hypothesis, this is not zero, we Oeue may divide through by it and, after transposition, have, from (3), y = aot + azox? + arity + doay® +: --. (4) This means: that we have transformed equation (1) into the form y= F(z, y), where F(z, y) is defined by the series in (4). In (4) let us substitute the series with undetermined coefficients y= cn + con? + esx? +--+ (5) By equating coefficients of like powers of x in the result the coefficients c; are easily determined. For the first three we have 1 = ayo, 2 = Gao + dit + doze”, €3 = G30 + dase + Arata” + doscr® + aisc2 + 2 doztice. The series (5) then formally satisfies the equation. It remains to prove that (5) converges for sufficiently small values of x. The series (4) is, by hypothesis, convergent. Then, with the notation of § 38, the function ou — Mu M OTe) a dominates F(x, y). That is, 6(x, y) may be expanded into a series with positive coefficients $(t, y) = Arot + Avot? + Auzy + Aozy? +++, (6) where Ax > ||. (7) If we solve the equation y= (x, y) (8) by a series expansion y= Cie + Cor? +--+ )) in the same manner in which we have solved (1), the coefficients C; are obtained from (6) by the same formulas by which the coeffi- cients of c; in (5) are obtained from (4), and it is evident, therefore, aed C:> leil. Hence the series (5) converges if (9) does. But the series (9) may be obtained by solving (8) as a quadratic equation in y and expanding the result by the binomial theorem. The resulting series is known to converge for sufficiently small values of x and can be no other than (9). Hence the series (5) converges.DERIVATIVES 98 The existence of the function y of x having thus been proved, its derivatives may be found by use of the theorems of § 36. For we have by that section =F at Fy = 0; = ode + 5 du = 05 (10) of whence a S ay ey Higher derivatives may be found by differentiating (11), or if we divide (10) by dx and denote we y', we have Cia ies Bet py =o on ay Then differentiating successively with respect to x and placing im FY ds »+++, ete., we have the equations Tn this way we may find the derivatives y’, y”, y’”, ete., pro- vided the partial derivatives of f(x, y) exist, since in all cases the coefficient of each of the derivatives y’, y”, y’” in the equation in which it first appears is x which has been assumed not to van- ish. In,this way we may write down series (5) without following the method of the text. 40. One equation, more than two variables. The equation FY, 2) =0 ql) defines any one of the variables, sey z, as a function of the other two, x and y; namely, 2= (2, 9). (2) af provided = +#0. az