GA TORE URNS OF GALCOLUS OF VARIATIONS ENN SHOWA 10 SOTMOTWS - HAASUOS | | cary S 3DOVER BOOKS ON INTERMEDIATE AND ADVANCED MATHEMATICS Handbook of Mathematical Functions with Formulas, Graphs and ‘Mathematical Tables, Milton Abramowitz and Irene A. Stegun, $4.00 Analysis of Straight Line Data, Forman S. Acton. $2.00 Elementary Concepts of Topology, Paul Alexandroff. $1.00 ‘The Mathematical Analysis of Electrical and Optical Wave-Motion, Harry Bateman. $1.75 A Collection of Modern Mathematical Classics: Analysis, Richard Bellman, $2.00 Selected Papers on Mathematical Trends in Control Theory, Vol- ume I, edited by Richard Bellman and Robert Kalaba. $2.25 Selected Papers on Mathematical Trends in Control Theory, Vole ‘ume II, edited by Richard Bellman and Robert Kalaba. 2.25 Numerical Integration of Differential Equations, Albert A. Bennett, William E. Milne and Harry Bateman. $1.35, Almost Periodic Functions, A. 8, Besicovitch. $1.75 Collected Mathematical Papers, George D. Birkhoff. Clothbound. ‘Three-volume set $20.00 (tent.) Algebraic Functions, Gilbert A. Bliss. $1.85 Introduction to Higher Algebra, Maxime Bécher. $2.25 Lectures on the Calculus of Variations, Oskar Bolza. $1.75 Non-Euclidean Geometry: A Critical and Historical Study of Its Developments, by Roberto Bonola; The Science of Absolute Space, by John Bolvai; and The Theory of Parallels, by Nicholas Lobachevski. $2.00 An Investigation of the Laws of Thought, George Boole. $2.25 Eero ti Cabie/ cf ints Differences, George Boole. 1.85 Introduction to Bessel Functions, Frank Bowman. $1.50 Introduction to Elliptic Functions with Applications, Frank Bow- man. $1.25 ‘The History of the Calculus and Its Conceptual Development, Carl B. Boyer. $2.00 Theory of Groups of Finite Order, William $. Burnside. $2.75 ‘Theory of Probability, William 8. Burnside. $1.00 ‘The Theory of Equations; with an Introduction to the Theory of Binary Algebraic Forms, William S. Burnside and Arthur W. Panton. Two-volume set $9.70 (continued on back flap)CALCULUS Or VARIATIONS BY A. RB, FORSYTH DOVER PUBLICATIONS, INC. NEW YORK * NEW YORKPublished in Canada by General Publishing Com- pany, Ltd., 80 Lesmill Road, Don Mills, Toronto, Ontario. Published in the United Kingdom by Constable and Company, Ltd., 10 Orange Street, London we 2. This Dover edition, first published in 1960, is an unabridged and unaltered republication of the work originally published in 1927. Tt is published by special arrangement with the Cambridge University Press. Library of Congress Catalog Card Number: 60-2846 Manufactured in the United States of America Dover Publications, Inc. 180 Varick Street New York, N.Y. 10014To MARION IN REMEMBRANCEPREFACE. The subject, commonly called the Calculus of Variations, has attracted a rather fickle attention at more or less isolated inter- vals in ite growth. Its progress has been neither steady nor con- seontive, From some cause, in its nature, or in its ineompleteness, or in its presentation, it has not secured an abiding interest. Not infrequently, investigators have been concerned with appli- cations of the Caleulus and, for their purpose, have been known to use fragmentary results, Thus, in the theory of the potential, Dirichlet’s Principle has been invoked. In instanees when regard has been paid to the establishment of the Principle beyond an assumption of its intuitive ‘truth, only the initial test belonging to weak conditions has been imposed ; and a general inference has been drawn, which was not justified by that test alone. Again, the Principle of Least Action has been made the support, and sometimes the occasional basis, of theoretical explanations ot the physics of the universe: though it should be added that the introduction of kinetic foci in dynamics is the equivalent of another necessary canonical test. Even so, all the recognised tests have assumed that variations in natural phenomena must be gently regular. Variations which, remaining small and continuous in their magnitude, change in », violently regular or irregular manner within a very restricted range, have usually been ignored . yet the theory of stall vibrations wields a far-flung domination. In Newton's problem of the Solid of Least Resistance, the formal solution satisfies all the customary tests which arise through variations of the gently regular type. Still, more than a century ago, Legendre proved that the solution is mathematically unsatis-vii PREFACE factory, though its neglect by engineers is not due solely to mathe- matical deficiencies. The significance of the investigations, due to Weierstrass, is not always recognised; but their importance need not be empha- sised, as though complete finality has been attained. ‘The results, usually associated with his name, relate to only the simplest class among the problems which present themselves and which require no more than the simplest form of his specially devised analysis. There is ample scope for further research by his method, in exten- sion of the range of its application. The present volume attempts a systematic exposition of the subject by what, in the main, is a uniform composite process. Though it does not purport to be a history, the gradual historical growth of the successive tests has governed the arrangement. A. fundamental (yet quite elementary) simplification, derived from the Weierstrass method, has been used from the beginning, even to obtain the results originally due to the founders of the subject. These limited results maintain their standing, because they provide tests which must be satisfied in simple forms of enquiry, and be- cause they remain significant even when they are menged in the wider results obtained by the more general method of Weierstrass. Moreover, the volume has no pretensions to an encyclopaedic range. Processes and investigations, however useful in the ex- ploration of other regions, are omitted unless they fall into the course of exposition adopted. So far as I am aware, much of its material is novel. Two sources, more than others, have been useful to me. The first of them is the Moigno-Lindelof volume Calcul des Variations, published in 1861; except for the Sarrus formalities, it seems to me an admirable exposition of the older range of investigation. The other source is to be found in such access to the work of Weierstrass as has been possible. Before the year 1895, I had read a manuscript copy of notes of lectures byPREFACE ix Weierstrass on his treatment of single integrals of the first order, including the associated isoperimetrical problems; for the loan of the volume from theiz College Library, I remain indebted to the authorities of St John’s College, Cambridge, Since that date, Pro- fessor Harris Hancock has published (1903) his volume, based on similar notes and on lectures by Schwarz, Unfortunately, a general expectation, that an authoritative edition of the Weierstrass lectures would be published, has not yet been realised. Beyond the sources just mentioned and such other sources as are quoted in the text, my work is independent. Some mathema- ticians may wish that the exposition had been differently balanced. Some will feel regret, and may award blame, for the omission of the work of writers such as Clebsch and Hilbert—an omiasion not due to lack of appreciation of their researches. Whatever its merits or its demerits, the presentation is that which has appealed to me, as leading most directly to a comprehension of the subject. An abstract of the contents of the book may be useful, as an indication of its seope. In the first chapter, the simplest form of integral is discussed. Tt involves only one dependent variable, together with the first derivative. The method adopted is, in substance, the older method for restricted variations; and the results obtained, including Jacobi’s test which limits the extent of the range of the integration, are typical of those that persist in all subsequent investigations, though they do not constitute the aggregate of tests of a general character. The second chapter deals with the same type of integral by the method of Weierstrass, which makes both the dependent variable and the independent variable in the older process to be functions of a new independent variable, usually selected so as not to be intrinsic to the problem; thus simultaneous independent variations can then be imposed from the beginning upon both the variables which occur. It is found that, for gently regular variations, nox PREFACE new tests emerge from the use of the Weierstrass method,—a con- clusion not unimportant in itself—though the formal expression of the tests is modified. In the third chapter, both methods are applied to integrals, which still involve only a single original depen- dent variable and now include derivatives of the second order as well as those of the first order. Of the analytical material in these three chapters, convenient geometrical illustration is pro- vided by plane curves. The next three chapters are devoted to the discuasion, by both methods, of single integrals which involve two dependent variables and one independent variable in their initial form, together with derivatives of the first order, and (less generally) of the second order, though the analytical development in the latter case is not carried so far as in the former. The increase in the number of variables does not lead to an inerease in the number of significant tests, though (as is almost to be expected) the expression of the several tests tends to become more complicated, For the material in these chapters, convenient geometrical illustration is provided by skew curves, The seventh chapter introduces the essential advanee made by the Weierstrass method, through the emergence of a new additional test. The advance comes through the consideration of variations which are not restricted to be of a gently regular type. The varia- tions are naturally required to be continuous and, as maxima and minima are being considered, they are required to be small in magnitude; but, within that small range, they are permitted to vary even abruptly, as violently as continuous curves representing rapid stall oscillations or even as continuous serrated curves. Many such variations can be compounded from rudimentary varia- tions of a selected type ; and the use of the latter variation leads ‘to the construction of a new test which, necessarily satisfied for the most elementary form, is cumulative in its effect for the com- posite form. This Weierstrass test is applied to single integralsPREFACE xi which, of course, involve only ordinary derivatives. In the case of the Solid of Least Resistance, it is shewn that the solution, satis- factory under the tests associated with the gently regular type of variation, does not obey the further test associated with the strong variation, and therefore does not supply a minimum. It appears also that the Principle of Least Action does not supply a mini- mum : the demands of the tests, arising out of gentle variations, are satisfied ; but the demand of the Weierstrass test, arising out of strong variations, is not satisfied. The eighth chapter is devoted to the consideration of simpler problems of relative maxima and minima—the isoperimetrical problems of even ancient interest. In particular, those problems are discussed, in which the requirement of a maximum or of a minimum is obliged to fulfil the condition of allowing a coexistent related integral to maintain an assigned value. Other types of relative problems—in which, for example, persistent relations hold among the variables—are considered, though only briefly, partly because the first stage in their treatment is to be found in treatises and memoirs easily accessible. The ninth chapter deals with double integrals which, in their initial postulation, involve one dependent variable and its two first derivatives. The concurrent geometrical illustration is, of course, provided by surfaces in ordinary space. Both the older method and the later method are used for the discussion. The treatment of the most interesting of all problems of this kind—minimal surfaces— is simplified when the Weierstrass method is used from the be- ginning. Schwarz’s theorem, which secures the determination of a minimal surface by initially assigned conditions, has been ex- tended so as to obtain an analytical expression of the Jacobi tost in limitation of the range. The tenth chapter is devoted to two issues: one, the construction of the Weierstrass test for double integrals and a proof that it is satisfied by minimal surfaces: the other, the simplest type of isoperimetrical problem. The eleventhxii PREFACE chapter is concerned with double integrals which involve the partial derivatives of the second order; but there is no attempt at a full discussion, mainly because, after the application of even the simpler tests, the analysis becomes unwieldy and the developments demand the differential geometry of the curvature of surfaces. A final chapter is devoted to triple integrals, involving the first derivatives of a single dependent variable, The convenient geometrical illustration is provided by the consideration of volumes in quadruple space. Only a slight use is made of the mathematical notions of such space; and, because the geometrical considerations are mainly concerned with volumes, a three-fold amplitude finds, for most purposes, a working representation in the ordinary space of experience. The analysis, which is requisite for the full appli- cation of the Weierstrass method to triple integrals, soon becomes laboured; it is here developed only so far as to construct the necessary tests which shew that, owing to failure under the ‘Weierstrass test, Dirichlet’s Principle is not valid. Before parting from the volume, I would thank Professor H. F. Baker, for his kindness in reading the earliest sheets of the volume. Above all, I must mention the Staff of the University Press, Cambridge. ‘Their steady and unfailing co-operation has been my mainstay during the printing of the book. Now that my task is ended, I tender my grateful thanks to all of them who have shared our A.B. FORSYTH. 81 Desembor 1926CONTENTS. INTRODUCTION. $s race oY ee acne! a 2. Early beginnings cabelas feb 1 3. Newton, John Bemoulli ; - stata 2 4, Kuler and Lagrange 3 6, Legendre. oie es 4 05. TAB ladion mckorrsi ex 3 7,8. General notion of vasia ‘ 6 9. Weak varistions : special variations strong variations 7 10, 11, Woiersires ae seh ety 12, Assumed Timifationa| on fanotionality : oa iy = eh ae 9 CHAPTER I. INTEGRALS OF THE FIRST ORDER: MAXIMA AND MINIMA FOR SPECIAL WEAK VARIATIONS: EULER TST, LEGENDRE TEST, JACOBI TES Proliminary note ao es yar ooh Po, € he A MLL 18, Types of problemas arising fom integrals with maxima or minima |) UL 4, Tntegral ofthe fist onder: its variation, wih fixed Limite . 12 16, 16, Modification of the frat variation. ria sineaey iS ‘V7. Characteristi equation, and eurre; the Buler test . a + lpi 18, Quadrotic terms in the variation : ‘second? variation...) 19-21. Subsidiary characteristic oquation; the Legendre test |.) 18 22-25, Primitive of the subsidiary equation, with aome properties. 22 26, 27. ‘The Jacobi test 5 conjugate points on a characteristic, with geomotrical Intergmestlon liber? tyr onervtat rar spiths oe saptonted > $B 28, Summary of the tests. = mex ge Bai 20, Conjugate as actual Himit ofa range. misproetiubhimuieine| 80 30, ‘Examples ; the eutanoid, and its properties connected with the teste. 30 31, Mobile tanita of an integral; terminal conditions a8 52, Complete variation of an integral with motile limite 3 30 83, 34. Conditions for a maximum or minimum, consisting of the chametoristic equation and terminal relations: examples, including brachistochrone, geodesic on paraboloid of revolution, and a failure peas. ae 35. First integral of the characteristic equation: Hilbert’s theorem | | 4844, 45. ‘The two critical equations are eqn 60, 61. Primi xiv CONTENTS: CHAPTER II. INTEGRALS OF THE FIRST ORDER: GENERAL WEAK VARIATIONS: ‘THE METHOD OF WEIERSTRASS. Proliminary note. 2 ee Ses 386, Weak variations in goneral 31, Molietion of intel, In whieh both variates become dependent on & new variable 38, Identities satisfied by now subject of integration 7, the integra form being covariantive —« ai a 39, General variation of the integral 2) 40. Tho first variation . 41-43, ‘Two equations in tho range, and terminal relations, to bo satis order that the first variation may vanish . ont to a single characteristic ‘equation, on account of identity satisfied by Fs. se 46, Invariance of characteristic equation, and of one critical quantity . 47, Two quautitios which remain contionows in passage through any free place of discontinuity in direction : 48, But tho numbor of such free places in finite range musi be limited 49, Primitive of characteristic oquation involves two easotial arbitrary constants . 0 Baants 50. ‘Tha ‘second? variation; the deviation due to small variation 51, Relations among the second derivatives of J”. 52, 53, Normal form of the second variation. '54,_ Summary of asumptions made during construction of normal form 55, 56. Discussion of this normal form: the Legendre test. ss 57-59. Subsidiary characteristic equation. « of tho subsidiary oquation; can be derived from the primitive of the characteristic equation. , 62, Proportina of any tio indapendent integrals of the subsidiary equation « 69,04, An integral 2( ¢) of the subsidiary equation, acting the socond variation: the Jacobi test. 65. Conjugate places in a range of ineprasin geometrical meaning of the ed, in ‘Tacobi test 66. ‘Tho function Z(G ¢) vanishes at conjugates and changes signin passing through any zero. 67. A range, bomided hy two conjugates, does not contain oy site bounded range. f z 68, 69, Consecutive characteristic curves: properties 70-78, Characteristic curvo can bo drawn through two asiigned points, as well 1s through one point with an assigned direction, if the second point is not within the region of eonjugatos of the frst. 73. Composite small variations: their effect... 74. Summary of results established « 75. Beamples by the Woiorstrass mothod: eatenoid: geodesics on & ones, ‘and on surfaces of revolution 70-60, Welretrmae proof that an integral range canmot extend beyond a con- Jugate a : PAGE 52 52 87 80 93. 95 7 98 110CONTENTS xv CHAPTER IL INTEGRALS INVOLVING DERIVATIVES OF THE SECOND ORDER: SPECIAL WEAK VARIATIONS, BY THE METHOD OF JACOBI; GENERAL WEAK VARIATIONS, BY THE METHOD OF WEIERSTRASS. $ PAGE Proliminary note. Mole. ei 81. Integmala involving second derivatives of one dependent variable. | 116 82, Full votiation of the integral with mobile limits... =... IT 83, 84. Vanishing of the first variation : characteristic equation =. =.) 118 85. Subsidiary characteristic equation, and its primitive... . 120 86. Terminal conditions: fourcases . - . ee se TRL 87. Results for integrals of onlorm =. | 1 1 sss TB 88-90. Second variation for special variations : its normal form ©... (136 91, 92, Properties of integrals of subsidiary characteristic equation. | | 129 98, 94. Determination of two quantities, roquired for the normal form =... 180, 9, Normal form of second varintion: the Legendre test 5. |. 138 96. ‘Tho Jacobi test ; with summary of testa. / 134 97. General wesk vatiations: the Woierstmss mothod : transformation of eee a 98, First variation of the integral Rn EmieS: ise 99. The two characteristic equations . ee 100, 101. ‘The conditions at the limits : four eases: ootparison with § 86 138 102, Continuity of four magnitudes through a fros discontinuity of direction snd curvature on characteristic curve. 12 103. The two characteristic equations in § 99 are equivalent to one only, ‘owing to the fundamental identities of § 97 : 2 M8 104. Certain invariantive forms. 146 105. Relations among the second derivatives of the integrand, incoing the single characteristic equation of $108... - 18 106, 107. Protiminary normal fore of the second variation...) AS 108. ‘Tho subsidiary charactovistic equations . Spunteee gael 109. Kquivalent single subsidiary equation, with the ‘deviation? as vatiable. 158 110. Final normal form of the second variation. + 160 111, 112, Primitive of the characteristic equations, with one essential property | 161 113. Primitive of the subsidiary equation : reek a 114,115, Properties of integrals of the subsidiary equation | ee 168 116. Consecutive characteristic curve . 167 117-119, Conjugate points; a range, limited by conjugates, cannot inolude within itsolf a range similarly liraited Ee a a ee 120. Adjacent characteristic curves: sub-conseoutive curves... . 178 191, 122. ‘Equations of sub-consooutive characteristic 176 125. Conscoutive chamnctaristic oan bo drawn through sien ‘contizguons point. ‘ eT tice Saas) 124,125, Digewssion of normal form of seoond variation =. 2.) L118 198. The Jacobitest =... " 181 127, The Legendre test... : Set 128, Summary of conditions for weak variations . 183 189, Example, in which tests are satisfied : yot thore iy nota real minimum 184xvi SS 130. 11. 132, 133, 194. 135. 136, 187. 138. 139. 140, CONTENTS CHAPTER IV. INTEGRALS INVOLVING TWO DEPENDENT VARIABLES AND THEIR: FIRST DERIVATIVES: SPECIAL WEAK VARIATIONS. Totegrals of the first order in two depondent variables. Spocial variations imposed on the integral Mobile mits. ae Candas ss ‘The first variation : characteristic equations, and terminal conditions . Hamilton form of the charactoristie oquations. =... Nature of the primitive characteristic equations : determination of arbitrary constants: various eases A ‘Special forma whon fret intagralcan to obtained: examples + « Subsidiary characteristic equations, and their primitive. - Temma auxiliary to the establishment of this primitive. - 141-145. Telation batwoon two integrals ofthe subsidiary equation. M4. 146, 146. ar. 18. 149, 10, 151. 152, 153. 164. 162, 163. 164. 165, 168. 167. 168. 169, 170. im. 172. ‘The second variation, under special variations Equations for ito teusformation, and their resolution in torms of two integral-sets of the subsidiary equation 3 Normal form of the second variation CERI ‘The Legendre test. ; Conjugate points: the Jacobi test. Tnvariantive property of fundamental group 0 of intgrabaets of the sub- liary equation — i Critical function for the Jacobi test. awe Ya eet ‘Sammary of results for special woak variations CHAPTER V. INTEGRALS INVOLVING TWO DEPENDENT VARIABLES AND THEIR FIRST DERIVATIVES: GENERAL WEAK VARIATIONS. . ‘The method of Weierstrass : a fundamental identity lations among the second derivatives of the subject of integration First variation of the intogral: three charasteristie equations : character- istic curve ; terminal conditions ‘ Continuity of certain derivatives at free discontinuities on the curve ‘The three characteristic equations equivalont to two noe Primitive of the characteristic equations : subsidiary equations ‘The second variation : proliminary modified form. v Tomark on change of method from the method of Chaptor IIT ‘Tho thre ubeitar uations edule totwoequtions wih molied variables. A One selected integral of the two subsidiney equations... Final normal form of the second variation... Primitive of the subsidiary equations, with moto nx to the Aenean variables. Proporty of fundamental sets of intagrals of subsidiary equation . Diseussion of the second variation: the Legenilre test. a PAGE 189, 190 191 192 193 194 197 ag 213 216 a7 a8 220 924 236 298, 299 21 236 238 229 2 M4 246 250 251CONTENTS xvii 8 Page 173. Conjugates on a characteristic eurye: the Juoobi test... 8B 174. Critioal equation for the dovermination of a conjugate, 8 175. Example: minimum reduced path of s say in geometuical optics.» 256 176, A conjugate-bounded range containe no similar vango . . 862 177. Contiguous eharactoristio curves. sid 178, Tntograls involving m dependent variables, with rstderivatives | | 267 CHAPTER VL INTEGRALS WITH TWO DEPENDENT VARIABLES AND DERIVATIVES ‘OF THE SECOND ORDER: MAINLY SPECIAL, WEAK VARIATIONS. 179, First variation of am integral for special variations oe er 180, 181. Characteristic equations : terminal conditions =. | |), O78 182, Form of tho primitivo of the chamaetoristic equationa 274 183. Properties of the primitive . a is 276 184,185, Subsidiary charaeteristio equations: their primitivo 278 186, Invariant property of « fundamental group of integral-sets of the sub: sidiary equations =. 281 187, Particular properties of combinations of integral-sets of these equations 988 188, Six fondamontal relations nmong four solected intogesl-sets . 384 19. First variation by the method of Weierstrass : general weak varia 386 190, 191. ‘Three characteristic equations: terminal conditions 980 192. ‘Tho threo oquations aro equivalent to two, in virtue of two identities | 201 193. ‘Terminal conditions agree with the conditions in § 181... , 292 194, Continuity of eortain derivatives through any froe isolated disco ‘on the charactoristic curve . 298 195. Seoond variation, for special weak variations batcbalisenaete © 20" 196. Retations for reduction to normal form. =... sss 888, 197. Preliminary normal form of socond variation... 307 198, Use of four integralsets of the subsidiary equations, for the deter rmination of the coafiicionts in the frst normal form. —. 208 199, Resolution of equations of redaction in § 196 by means of the six relations mn § 188. - + 301 200, Final normal form of second variation: the Legendre fost ||| 304 201. Limitation of range: tho Jacnbi test. reves 306 202, 908, Critical equation in connection with the Jacobi text || 307 204. A range, bounded by two conjugates, cannot enclose « similar range. 320 205. Summary of tests for special variations. . an 208. General weak variation: statement (without, pot) of ten results eon: cerning the second variation. ee. gat CHAPTER VIL. ORDINARY INTEGRALS UNDER STRONG VARIATIONS, AND THE WHIERSTRASS TEST: SOLID OF LEAST RESISTANCE: ACTION. 207, Goneral notion of a small variation Gaur ia 380 208. Strong variations : fondarnontal constituent type. 321 209,210, Cumulative condition, arising from condition for the type + 88888 211-218, 214, 215. 216. air. 218, 219. a2. 225, 226, 227. 228, 229, 230, 231, 252. 233 224, 235, 236, 237, 238, 230, 240, 241. 42. 243, 240, 247-249. CONTENTS Strong variation of ordinary intogral with first derivatives ‘Weierstrass Z-fanction and Weicrstmss test. Bxainplos of the £-function as atest, with extension of the fundamental ‘constituent type of variation; geodesics in general: example duo to Weierstrass. ae iat dia Expreasion for Z-function with a squared ftctor |. 5 Modified form of the test: it is distinct from the Legendre tot. ‘When the subject of integration is mtional, neither a maximum nor minimum can exist Solid of Least Rexistance: Newton's problem: alternative laws Extended type of strong variations. ae Modified £-function for extended type of strong variation Strong variations when sooond derivatives ocour . aati Form of H-function for those variations Along veiation of integrals inolrng two dependent variables and thor first derivatives. TE seoas collated ‘The #-fanetion for skew curves 5 Altemative form of the Z-funetion of § 225: it is diate’ Logendee test. Dogma of Teast Action expression fur Action sbi ‘Tests of Action, under apecial weale variations ; ‘Action is a minimum under special weal vatiations: example, from projectile. Seat tatinin Action is a minimom under goneral weal variations Strong varntions imposed on Action: expression for variation of Action ‘under strong variations Action is not a troe minimum: the Weierstrass test not being satisfied Ropresentation Gueluding the time) for configurations of moving systems; example of Action of a projectile, with strong variations . General review of all the tests. 5 from the CHAPTER VIIL RELATIVE MAXIMA AND MINIMA OF SINGLE INTEGRALS : ISOPERIMETRIGAL PROBLEMS. ‘Types of problems in relative (or limited) maxima and minima ; soporimetrical problems: simplest type, involving derivative of th first order, with a coexistent intogral remaining constant . Selection of general weak variations leaving constant integral unchanged ‘Vanishing of ‘first’ variation of variable integral Characteristic equation: terminal conditions. Likowive for integrals, aubjooted only to special weak variations. | Continuity of cortain fictions through free discontinuities on the characteristic curve, these being finite in number. oat Goma eo el ents etree 9-42) (c ‘Soond? variation: normal form. . aaa oe: Sabsidiary equation, and its primitive Discussion of the second variation: the Legendre test | PAGE B24 398, 320 335, 387 388 340 37 350 358 355 257 368, 37. 376 381 age 384 387 388 291 393 204 395 206 397 309 401 407contmnrs xix 88 PAGR 261,252. The Jacobi test: its analytical expression... 0... 07 253. A conjugnte-bounded range contains no similar range. | |. 400) 254. Consecutive characteristies . 410 265, Examples: masimum area with assigned length of contour: solid of macinom stiacioa: il of maximum volar with, anigned superficies. az 256. Strong variations for problems in relative maxima and minima, with derivatives of first order: examples... a9 257. Isoperimetrical problems, with derfuatives of omer iain than the fist 48% 258, 259, Charuoteristic oquations: terminal conditions. ae ~ 1G: 260. Second variation. 48 261, Tsoporimetrical problems, with soveral dopondont variables; exnmnplo | 428 205 28, gy ania ope in Ni ern enei cu lees, ‘threo types of frequent occurrence... es Method of proceeding . 436 Simplost typo of problem, with limiting equation not of the integral type 436 Expression of the first variation of an integral, by variations limited ‘through an imposed coexistent equation... . .. ABT. 267. Characteristic equation : terminal conditions + 49 268, Characteristio equations generally reducible in number by one unit | 440 269, Form of results for special woak variations; examples relating to goo- decics rca dt = GED 270. Statement of resulta when second derivatives oceur =.) AT 271. Reduction in the number of characteristic equations | 449 272. Forma of results for special weak variations: shortest enrve of constant circular curvature: brachistochrone in resisting medium... 451 273. Postulation of mote than one non-integral coexistent equation. . 455, CHAPTER IX, DOURLE INTEGRALS WITH DERIVATIVES OF THE FIRST ORDER: WEAK VARIATIONS : MINIMAL SURFACES, Proliminary note, mainly on minimal surfaces. to aoe 274, Double integrals of the simplest type: special weak variations 458 275. Lemma in double integration: convention a3 to order of integration | 459 276. Modification of expression for first variation, with fixed limits. 461 277. Characteristic equation: characteristic surfaoo Spang . ROS 278, First variation, with mobile limite. sae mommies 279-281. Charactoristic equation : boundary eonditions 466 282, Cauchy's primitive of the characteristic equation under = amignd ans with geometrical interprotation his 4969 288. Sooond vaviation for special vaxiations: normal form |. |) at 284,985. Subsidiary characteristic equation, and its primitive 473 286-288, Discussion of normal form of vocond variation: the Legende test; and the Jacobi test in descriptive form. . d 475 289, Differential equation of minimal surface, under special variations: tho range on a catenoid . 2 9 290, General woak variations: transformation ot ite oom nde! pendent variables 2, wwe we 481xx CONTENTS ss PAGE 291, Propertios of the transformod integrand =. swe 483 292, Relations between seoond derivatives of tho intogrand . + 486 298, First variation of the modified integral under general weak visiations . 487 94,205. Vanishing of the fit vaviation: characteristic equations: boundary condition . eas 200, Agreement of boundary conditions with result in § 280 491 997. Continuity of cortain derivatives through an isolated free discontinuity. 492 298, Tho three characteristic equations equivalent to one RT 606 299, 200. 497 301. 498 302, Boundary condi a 303, The equations of the characteristic provide an actual minimum for weak variations, subject to conjugates 7 a 304. Schwa theorem on the determination of minimal surfaces: examples 205. The second variation, under general wealc variations... 306,307. ry characteristic equations =. ss 308, 309. tion in the form of the second variation + 810 210, Relations for the transformation . ase, 513 BLL. Uso of subsidiary equations in the transformation, ney “ors 312, Remark on the process adopted, similar to § 164. - 516 13, Discussion of the normal form: the Legendre test. . eis 14, Tho Jaoobi tost ; spocial application to eatenoid . 518 315. Gonoral expression for small variation to a eonseentive minimal surface 581 316. Wquatons ofthe consecutive characteristic: general application to cate jd, and to Enneper’s minimal surface. eurmae, ies CHAPTER X. STRONG VARIATIONS AND THE WEIERSTRASS TEST, YOR DOUBLE INTEGRALS: INVOLVING FIRST DERIVATIVES; ISOPERIMETRICAL PROBLEMS, S17. Strong vari 1s of double integrals : constituent clement for surface variations. : cinier/l =. 986 218, ‘The chaincteristic component of a strong variation. =... 889 219. ‘The component from thoarbitrary surfacn 3 2530 320, Total strong vaviation: the Bicfmotion, and the Welerstnist test. 892 321, Altemative forms of the By-function 34 4332, "Typos of strong variation that can be constructed from the elernontary type. emeaanreirein cyenseoeets 2. 528, ‘Tho four tat ora dot integra ol baa 324, Nxamplos under the Weierstrass test, including minimal ssrfiess . 889 325. ‘The integral eanuot have a maximum or minimum, when the integrand is rational in the derivatives ; example, Dirichlet’s Principle, in two dimensions. ane nr ech 542 396, Tsoperimotrieal problems | ae ee bts 397. Vist variation of variable integral and constant integral. + 546 $98, 329, Variations permitted by the constant integral: charactoristic equations conditions at mobile toundary. =. se MT 200, Summary of requiremonts, from first variation =... e8908s 331, ane, 334, 370, 371. conrmNrs: Continuity of cortain derivatives through isolated free discontinuities . ‘Examples: surface enclosing maximum volume within given area. Second variation {real for isoperimetrical problem Strong variations: tho Bs-fanction : application to example in § 392 CHAPTER XI. DOUBLE INTEGRALS, WITH DERIVATIVES OF THE SECOND ORDER: WEAK VARIATIONS. Spocinl wenk variations: “frst! variation =... ‘Tho special variation and a mobile boundary. 2 |) Characteristic equation : boundary eondition zyse8 Second variation. uations of relation, to modify the expression ofthe seeond variation « Subsidiary characteristic equation. . Resolution of the equations of § 340 by means of integrals of the sub- ary equation Special relation between two intogials ofthe subsidiary equation « ‘Norial form of second vavintion : the Legendre test... Charnotovitie variation: Consecutive characteristic surface » d Conjugate of initial curve on a charnetoristic surface: the Jacobi test Gonoral weak variations = transformation of the integral. Tdentities satisfied by the modified integrand. mee ‘Firat’ variation of the intogeal: throo characteristic equations || General boundary condition. z ‘Tho threo charactoristio equations in § 954 are equivalent to one equation entifieation with equation in § 387 es CHAPTER XI ‘TRIPLE INTEGRALS WITH FIRST DERIVATIVES. Lemma in triple integration . er; Firet variation of triple integral, under special variation 3 Characteristic equation : boundary condition Primitive of the characteristic equation: Cauchy's xistonco-theorem Special forms of boundary condition. . . . Socond variation 2 2 i) eee Subsidiary characteristic equation . anaes Normal form: the Legendro test; ‘The Jacobi tost: Dirichlot’s Principle, expressed aa an. soe powesing a minimum under weak variations... 5 General weak variations — dontities sstisfied by transformed integrand in the triple integral | General weak variation of the integral: the four charactoristie equations: boundary surface condition ‘Tho four characteristic peice quivalent to the single equation mgs. nk Sees rena ee PAGE 551 552. 561 562 598 601 610 ell ars 61s es. eis. 619 21 22xxi 88 373, 373. S74. 375. 376, 37%, 378. 379-381. 82, 289, ae, 385, CONTENTS Identification of the boundary conditions in $§ 360 and 369. A ‘Summary of conditions from first variation, under weak variations ‘The Legendre test stated, for general weak variations... Romarkon tho Jecobitest . - . eee Strong variations for triple integrals ; geometrical roprosentation of a volume in quadruple space by ordinary three-dimensional space Roprosentation of volumes: (j) characteristic, (i) rudimentary strong variation. Component of the variation of the fatogral from tho characteristic con slituent of the stroug variation. Component of the vatiation of the integral, from the non-charseterstic constituent of the strong variation... ; "Total strong variation of the intagral: the Weiorstraes function. ‘The Weierstrass tost for triple intogmls Examples of the Weierstrass test: not satisfied by the Divichlet intogral of § 886, s0 that Dirichlet’s Principle docs not. express a true minimum: the four-dimensional minimal volume, which satisfies all EbeLtEMN i aaesNradiem bs, | etlaape gratin Gonctnsion . pple gatatscate sites lake 6 Imam sy of oie nn : 7 shar a PAGE on 67 or 628 629 632 634 638 439 640 oat 645CALCULUS or VARIATIONSINTRODUCTION. General range of the subject, 1. The range of Mathematical Analysis, usually known as the Caleulus of Variations, deals with one of the earliest problems of ordinary experience, The requirement was, and is, to obtain the most profitable result: from imperfectly postulated data; and the data may possibly be subjected to conditions, which likewise nre imperfectly postulated. When data and conditions are expressed in analytical form, the necessary mathematical calculations eannot be effected directly, because of some essential deficiency in the information. The gap has to be filled before the resolution of the problem is attained; and the process of supplying the lacking information is indirect, as compared with the regular methods of calculation. It consists of the construction of tests, which are the mathematical expression of general conditions ; it is composed of various gradual stages, sometimes independent of one another; and the ensuing requirements are combined into an aggregate which is adequate for the purpose. Usually, the predominating interest lies in the qualitative results that are constructed. Not infrequently, the subsequent quantitative calculations are ignored; they involve processes which belong to an ele- mentary range, that is unconcerned with the mode or modes of obtaining the information lacking in the initial stage of the original statement. Early beginnings. 2 To the ancient Grecks, with their wonderful geometry, the problem came not infrequently in a practical form : how to secure the greatest amount of land, which could be enclosed within a boundary eapable of being ploughed as a contour furrow in a given time: or, geometrically expressed, how to find the shape of the closed curve, which shall enclose the largest area within a perimeter of assigned length. ‘The Greeks—perhaps not the earliest race, in spite of their scientific achievements, and certainly not the only race, desirous of saving the most and wasting the least out of opportunities subject to assigned limitations—obtained a number of results, partly by what may be called inspired guessing or intuition ; partly by assumptions as to harmonies, and perfection, in curves; partly by experiments, ‘Though the results would not be regarded by later mathematical precisians as having been established with complete rigour, they often were sufficiently satisfactory for working purposes. Little substantial progress beyond the old attainments was made down through the middle ages. In the East, and among the Arabs in the South ‘West of Europe, the old mathematical subjects were not merely maintained,2 INTRODUCTION but even flourished, within the range of the old methods; which eame to be processes of record and description, mostly of an arithmetical and a geometrical type. Indeed, algebra was the main subject where progress was noted and notable, though often attained in geometrical guise: even the beginning of Descartes’ Geometry is occupied with the geometrical constructions devised for the solution of quadratic equations. Real progress began with the new world of science, which was created in the seventeenth century in Western Europe— not least with the new mechanics and new astronomy, under Galileo, Kepler, and Newton. It soon appeared that the old geometrical methods, even in their new amplifications, must merge into an entirely new method which, in character, was analytic and not synthetic. ‘Thenceforward, what now is called Analysis came to be the main instrument in vast fields of mathematical research. ‘The infinitesimal calculus was developed:: its historical origins are marked by the morbid hostility between the partisans of Newton and the partisans of Leibnitz, which raged for many a year to the detriment of achiove- ment in recognised knowledge. Old problems were re-stated ; and they were solved once more, not infrequently with new additions and unsuspected limita- tions. New investigations, hitherto belonging to the region of fancy, were brought within the range of practice. In particular, during the late seven- teenth century and through the eighteenth century, problems were propounded by individuals as public challenges—a habit that has survived among Aca- demies in the present day when proceeding to the award of their prizes. The actual propounding of the problem was frequently an intimation that the challenger had achieved the solution. Specially in mathematics, this was a form of initial publication. The Bernoulli family, in its successive genera- tions, was conspicuous for contributions of this type, Others, less publicly aggressive in tone, were content to achieve a result, and to be satisfied with the knowledge thus attained, without challenge or publication in any sitilar form, Of such men, perhaps Newton is the most conspicuous instance. Newton, John Bernoulli. 3. The gradual development of the infinitesimal calculus led to the formulation and the solution of new problems; and the new inquiries were not less frequent in problems, the aspect of which was mathematical rather than astronomical or mechanical or physical. Particularly within the range of problems connected with maximum attainment or minimum reduction, the new calculus proved effective; usually, the data were sufficient to allow a direct attack to be made upon the problem. But, soon, the problems some- times became of a subtler indirect nature: the very character, not merely the magnitude, of the unknown quantities was the essence of the problem; and even behind this difficulty lay processes that could not be effected. Such difficulties arose most direetly when the maximum or the minimum ‘was to be possessed by a quantity which, however veiled in expression at theINTRODUCTION 8 time, can be formulated as an integral, definite or indefinite. The quadrature —the actual evaluation—of the integral could not be made, because it would contain unknown quantities. On the one hand, the quest of these quantities was the main problem ; on the other hand, progress towards solution of the problem was barred, so far as the old methods were concerned, precisely because the quantities were unknown. Among such preliminary and isolated problems, two (among others) survive in interest to the present day. One is Newton's problem of the determination of the shape of the solid of revolution which shall meet with the least resistance to its motion through a fluid, on the assumption of a law of resistance conforming roughly to obser- vation, Newton's solution was not satisfactory so far as practice is concerned, even if the law of resistance be regarded as adequate; and Weierstrass’s analysis provided, more than half a century ago, a reason why the Newtonian solution is not satisfactory, even from a theoretical point of view. This general problem has come into more importance in recent times, owing to the develop- ments of submarines and air-craft; and the necessary association of new knowledge, derived from successive physical and mechanical experiments, has added grave complications to the mathematics of the general problem. A second problem, of historical importance and still of interest, is that of the Brachistochrone, associated (1696) with the name of John Bernoulli, In its simplest form, it requires the determination of the curve joining two given points such that the time of passage of a given mass from one point to the other along the curve (as in a smooth groove), under the influence of gravity alone and subject to no retarding force, is a minimum, that is, less than the time of passage in like circumstances along any other curve joining the two points, ‘The original solution, which requires the curve to be a eycloidal arc, is accurate as regards the quality of the characteristic curve; later investiga- tions have added limiting specifications on the range of the curve. And, naturally, analogous problems have been propounded and solved, when the motion is due to extemal forces other than gravity, and when retarding forces can come into play. Without multiplying the citation of instances unduly, it may sulice to mention the mathematical-physical conception denoted by the word Action Philosophers have been fain to deduce the mechanical movements of a dyna- mical configuration, even much of the physics of the universe, from the single property, that the Action between any two states of the configuration in continuous change is a minimum; and the property has been elevated in postulation to the Principle of Least Action, Euler and Lagrange. 4 The mathematical development of that section of Analysis, which has to be considered here, has been fitful, sporadic, slow; and, as 50 often is the case, not entirely free from controversy. Amid many names that now4 INTRODUCTION belong to the history of mathematical science, four stand out, because of the significance of their contributions towards the attainment of the first stage in the systematic construction of the calculus, The earliest name is that of Euler, fruitful in that branch (as in all branches) of Analysis in his day. He discovered® the characteristic differential equation which, as a necessary requisite, must be satisfied. It was obtained by considering the increment of the integral which arose through the varia- tion of a rudimentary are. In the first stage, Euler stayed his investigations at this mathematical result ; it secured a stationary quality, as a preliminary common to a maximum or a minimum property of the magnitude under discussion, Lagrange+ discussed the problem by regarding variations as active through the range of integration, not restricted to any rudimentary portion’ of that range. Indeed, he made a new foundation of the subject, entirely analytical in character as was his mathematical wont. To him is due the introduction of the symbol 5, which discharged useful duty through successive generations of mathematicians: though, as ean happen from lack of definite and precise explanation, initially unexpressed, inferences by later investigators were sometimes drawn, more comprehensive in statement than could be justified. He also extended his analysis so that it could be applied to problems involving two independent variables; and, in particular, his researches con- stitute a foundation of the theory of minimal surfaces. Subsequently, Euler resumed his consideration of the subject, on the basis of Lagrange’s work. His completed account§ is based upon the geometrical representation of the analysis, in which he regarded Sy as a change in the position of a point on an original curve, along the ordinate, to a contiguous point on the varied curve. There, for a period, advance in the progress of the subject was arrested. Detailed amplifications within the range were forthcoming, but always in the field of work already achieved by Euler and Lagrange, whose names dominate the first stage in the systematic caleulus, Legendre. 5. The results obtained by Euler and Lagrange were of the nature of necessary qualifying tests. The discrimination among merely stationary values, as between a maximum or a minimum, was left to intuitive considera tions, foreign to the analysis. + Methodus inveniendé lineas curvas mazini minimive proprietate gaudenter, sive solutio problematis tsyperimetrict latissimo aenw accept (Lausanne ot Geneva, 1744). 'f Miscellanea Taurinensia, 4. i (1760-1), pp. 178105 ; i., t. iv (1766-9), pp. 163—1 Guores, 2, pp. 885—862, 1, pp. 87—68. ‘5 For reference, see the preliminary note to Chapter IX. § Institutioner calculd integrals, til (2nd ed, 1793), pp. 881—475.INTRODUCTION 5 The next achievement in progress is due to Legendre. Thus far, all the investigations had concentrated on the stationary requirement—the Euler test, when there is one independent variable; and the Lagrange test, when there are two independent variables. All of them were concerned with the terms of the first degree in the small quantities in the increment of the integral (the so-called ‘first’ variation), consequent upon a small increment of the independent variable. The customary requirements for a maximutn or a minimum of an ordinary function, when the function is given explicitly, demand the consideration of the aggregate of terms of the second degree in those small quantities in its inerement (the so-called ‘second’ variation). Legendre was the first to discuss this aggregate. To him is due the idea of modifying its initial expression, by adding suitable arbitrary variations within the range, and subtracting their corporate effect in the balance at the boundary, always subject to unaltered conditions at the boundary. This device enabled him to express the second variation in a compact form, the sign of which Decame of essential significance for the main purpose. Legendre thus added®* a further test, discriminating as to character, which must be satisfied by integrals in one independent variable if they are to Possess a maximum or a minimum. The new requirement, like the Euler test and the Lagrange test, is a necessary qualifying test. Tacobi, 6. The next name to be mentioned is that of Jacobi. Discussing the second variation as formulated by Legendre, he proved that a quantity, of critical importance in Legendre’s expression, could always (without any inverse process) be derived from the knowledge presumed as attained in the Legendre stage. Making this advance in the purely mathematical range of the problem, he obtained extended forms for the tests already established ; he constructed an essential modification of Legendre's form; and from this modified form, he deduced+ a further test, limiting the range of the integral. A fall proof, and an extension (also with full proof), of Jacobi’s results were given in a memoir by Hesse}; and further investigations connected with the transformation of the second variation are due to Clebsch§, and to Mayer|, though their inferences for the most part are left without a state- ment of the geometrical significance of the form of the second variation, such as was made by Jacobi. The character of Jacobi’s test admits of simple illustration. ‘The shortest, are on the surface of a sphere, between two points on that surface, is the + Mémoires de VAcatémie Royate den Sciences (1186), pp. 737. t Grete Journat, xvi (1881), pp. 68-82. + Orelle’s Journal, liv (1857), pp. 227-278. § Crele's Journal, t.¥ (1858), pp. 254273, 88586. 1 Beitrige sur Theorie der Mazima und Minima einfacher Integrate (Leipaig, 1866); Crell’s Journal, t. xix (1868), pp. 288-268.6 INTRODUCTION portion of the great circle passing through them and lying between them: 50 far, the tests of Euler and Legendre are satisfied. But the portion of the great circle must be less than half the circle; if it is a half-cirele, the length (minimal in this case) is not unique ; if it is greater than a half-cirele, the length is certainly not a minimum between its extreme points, because (among other reasons) it can always be reduced by small variations. The requirement in this instance (and the cognate requirement in any corre- sponding problem) comes into Jacobi's work: the limit of the integral, as ranging between ‘conjugate’ points on the curve obtained, is bounded by a point on the circle and the diametrically opposite point. ‘The general test devised by Jacobi indicates the limit (if there be a limit) to the range along the characteristic curve within which the magnitude of the integral under consideration is a maximum or a minimus. With the achievement of these results, essential progress in the solution of the problem again came to an end for a long period. Mathematical ampli- fications, and generalisations wider in range but cognate in character, were made from time to time; but, in real effect, there were only the three types of test, to be associated initially with the names of Euler and Lagrange, Legendre, and Jacobi. General notion of variations. 7. In all the analysis hitherto applied, whatever its form, the general notion behind the mathematical expression was the same. A hypothetical maximum or a hypothetical minimum was postulated; if it were a maximum, ‘every small change (called a variation) must lead to a decrease in the hypo- thetical maximum value; if it were a minimum, every such variation must lead to an inerease in the hypothetical minimum value. Much explicit virtue lay in the assumption of the smallness of the variation. Not a little implicit difficulty was involved (though not recognised) in the lack of adequate specification of the properties constituting that smallness. Latent limitations (also not recognised) were imposed by the use of the symbol 8, which was initially employed to indicate the actual small variation of the unknown magnitude and which, by argument presumably deemed too obvious for statement, was allowed to exercise a supposed and unchallenged parsi- monious influence solely by its literal character, wherever and whenover it was made to appear. When once the range, within which the tacit assump- tions are legitimate, is actually examined, the result is to shew that the small variations are themselves gravely limited in character. Other small variations can be postulated, initially the same as regards type of magnitude, but with action gravely different from the restricted influence exereised by the other magnitudes, which are derived from them by some of the operations that were freely used. One consequence is that the old tests, when satisfied, lead to inferences which are legitimate only within the range of the smallINTRODUCTION 7 variations that remain small, not only in their own influence, but through the influence to be exercised by these derivatives. Another consequence is that the old tests can be trusted solely as connected with such range of variation: and that, precisely because they cease to be applicable when the hypotheses leading to their establishment are no longer valid, these old tests can furnish no information for small variations of the other kind indicated. They will remain, as necessary tests, because the maximum or the minimum property must be possessed for all small variations and there- fore for variations of the limited character. They are inadequate for the discussion of the magnitude, when it is subjected to small variations not of this limited character. In fact, the three tests are necessary for the solution of the problem; they are not sufiicient to secure the completeness of the solution. 8. Further, another limitation was imposed upon the small variations adopted. It was imposed explicitly; no attempt was made to shew that it is immaterial; and any significance, that the limitation might have borne, was so far ignored as not even to come within the range of examination, The limitation consisted in restricting the variations so that they should affect the dependent variable alone and should not apply to the independent variable, either independently or concurrently with the changes made in the dependent variable. If we turn to the graphical representation, whereby the relation between the two variables is illustrated by a plane curve, the small variations adopted consisted of a small displacement along the ordinate, coupled with a small displacement of the direction of the tangent; but small displacements, made sideways off the ordinate, were never entertained, Occasionally, such a displacement had to be considered for an isolated point or for isolated points; thus the extreme point might be required to lie on a given curve and be not merely a fixed point. In such instances, a special detailed argument was applied to take account of the effect caused by such a displacement of the particular point, while the possible effect of such a displacement throughout the range was ignored. When the corresponding investigation, necessary to take these small variations into consideration (subject to the old limitations as to the character of the variation of their derivatives), has been completed, it is found that no new test emerges. The sole change is of a formal character; but such a result requires definite establishment. The net conclusion, however, is that, for small variations which are subject to the limitations indicated, three necessary critical testa exist, Weak variations : strong variations. 9. Now these small variations, even when imposed upon both the depen- dent variable and the independent variable, are only the simplest type of all small variations. Thus, to return to the graphical representation, we could8 INTRODUCTION have small variations of a curve such as are given by displacements corre- sponding to very rapid small oscillations. In these, while the change of position is small, the change of direction may be finite though not small; the change of curvature may be very large; and so on. For all such variations, the analysis that has been effective in the other stages no longer applies; some new process must be devised that will eause these less simple variations to be taken into account. Small variations of the earlier type, viz. those in which the derivatives of the variation are of the same order of smallness as the variation itself, are often called weak variations. Among weak variations, those which are restricted 80 that they affect the dependent variable alone, will be called special variations (the quality of weakness being tacitly included in the name), Small variations of the later type, viz. those in which the derivatives are not limited to be small, though the actual displacement is definitely small,—are called strong variations. Manifestly strong variations will be much more versatile in character than weak variations; and progress will initially be secured by considering selected types of strong variations. Weierstrass. 10. An adequate method, new in form and (as regards later develop- ments) new in substance, was initiated by the work of Weierstrass, Although he died as long ago as 1897, no fally authoritative exposition of his researches has yet appeared. He gave lectures on the subject in 1872 and 1879, perhaps earlier, in Berlin. Notes of his lectures have circulated, though without indications as to whether their range is complete or only partial. ‘The work of Schwarz on minimal surfaces, as on branches of the theory of analytical functions, may be not unfairly presumed to have had its foundations in the teaching and the work of his master, whom he succeeded at Berlin, The results of the earlier stages of Weierstrass’s work have found a partial record in some books* published since his death. Renewed interest was stirred in the subject owing to his influence; and writers, in varying degrees of in- dependence, have been stimulated to further researches. Besides Schwarz, the names of Hilbert, Goursat, Hadamard may fitly be mentioned, as well as those of Kobb, Bolza, Kneser, Veblen, Hedrick, and ‘Tonelli. ‘The work of Clebsch also was important, though it belongs to the older range; and ‘Todhunter’s two volumes, dealing with the history of that range, may be consulted with advantage. ‘Two contributions, different in kind yet connected, and both of funda- * Hancock's Calculus of Variations may be consulted in this connection. + A considerable amount of research his appeared in scattered articles in the Bulletin of the American Mathematical Society. An artiole, by Love, in the supplement to the Zncyclopadia 18a concise account of the earlier stages of the older theory; and a fuller aocount ‘of the subject will be found in two articles, by Knosor, and by Zermelo and Han, in the Eneyelopidie d, math. Wissenschaft, voli A. 1, pp. 571625, 626—641.INTRODUCTION 9 mental importance, are made by the work of Weierstrass, In one of these, he ignores the preferential selection of the dependent variable as the subject of variation ; with him, the dependent variable and the independent variable are alike subjected to arbitrary small variations, independently of one another. The old tests, in modified amplification and with added elucidation, are obtained, in connection with weak variations. The discussion of strong variations, and the deduction of one now test—the so-called Exeess-fanetion— derived from the consideration of one simple class of strong variations, are entirely due to his initial researches in the matter. But he dealt with only the simplest class of strong variations, the particular class being sufficient for the purpose of the particular type of problem which he had under con- sideration. It may be that similar researches, involving less restricted strong variations for less particular types of problem, will only lead to generalisa- tions, however important, of his Excess-function ; it would appear as if such researches still remained for investigation. 11. As regards the weak variations, a further remark may be made to confirm the desirability of examining variations other than those which have been entitled ‘special.’ The more immediate geometrical representation, of the relation between a dependent variable and an independent variable by means of a plane curve, would imply one kind of special variation, were the analytical equivalent expressed by Cartesian coordinates; it would imply a different kind of special variation, were that equivalent represented by the usual polar coordinates; for homogeneous coordinates, and for bipolar co- ordinates, other different kinds of special variation would be used. All possible kinds of special variations are included when the most general kind of weak variation is adopted, because each special variation is only one particular case of the general type. Moreover, the general variation includes, as a particular case, that apparent variation which consists of displacement merely along the length of a curve without any alteration of its shape. Such an apparent variation is not a variation of the value of the integral because, when the integral is regarded as a sum of infinitesimal elements between its upper limit and its lower limit, the value of the integral is independent of the precise detailed mode of selection of its constituent elements. Lastly, the use of general variations makes it possible to consider (with- out the supplementary discussions needed for special variations) the cases, when the limits of the integral are liable to partial (though not quite arbitrary) mobility instead of being definitely fixed. Assumed limitations on functionality. 12, Throughout all the investigations, we shall suppose, and now explicitly assume, that the subject of integration either (i) is a regular function® of its ta derivatives) is uniform, finite and continuous. * That is, a fanetion which (wi10 INTRODUCTION arguments within the range of integration: or (ii), when not a uniform function for all possible values of its arguments within the whole field of variation, remains a uniform branch of some function, so that it behaves as a regular fanction in each part of the range of integration. If, at any place or places, an argument or arguments should suffer dis- continuity in value, so that the function or any derivative of the function potentially may cease to be regular, explicit note of the fact or of the possibility will be taken. Again, during the analysis and unless a warning to a contrary effect is given, the solutions of ancillary differential equations will be regarded as providing regular functions, when these are determined by regular data. In particular instances, it may happen that the conditions, which are imposed, cannot be satisfied by regular functions; it may even happen that, of a postulated problem, no solution exists of a type which is regular: the impossibility will be noted. But we shall proceed on an initial assumption that a regular solution may exist and shall make its discovery ‘our quest. The hypothetical solution of that type (perhaps with the assist- ance of existence-theorems, established with similar initial assumptions) will be subjected to all the conditions, and will be made to conform to all the data, that are imposed and provided. When the conditions are satisfied and the data are incorporated, the result will be regarded as the solution. In the contrary event, note will be taken of the fact, which thence will have emerged, that the imposed conditions and the assigned data forbid a solution of the foregoing type. It may be that conditions could be modified and that data could be changed; such alterations could be considered in their turn, as raising a new specific problem. Moreover, unless some contrary demand occurs requiring satisfaction, we shall expect that the fanctions concerned are of an ordinary simple character, using these words as in common parlance. Unless expressly forbidden, we shall suppose that our functions have derivatives which are unique, save at isolated singularities or in the immediate vicinity of singularities; eg» we shall not be concerned with uniform continuous functions, which nowhere have a unique differential coefficient, Equally we shall suppose that all the functions, with which we have to deal, are capable of uncon- ditional integration, eg., that they are not beset with singularities. The present purpose is the development of a calculus, without delaying over the foundations of Analysis, which will be regarded as having been duly laid and recognised in any bearing that has significance. It may happen that, when strong variations are considered in a full detail which would be essential in a complete discussion of their influence, corresponding refinements will prove necessary. For the present, its more ordinary and less critical commonplace range will be held sufficient; and, of course, the functions admitted will be subject to the corresponding limitations as regards regularity, continuity, and variation,CHAPTER I. INTEGRALS OF THE FIRST ORDER: MAXIMA AND MINIMA FOR SPECIAL WEAK VARIATIONS: EULER TEST, LEGENDRE TEST, Jacob! TEST. The chapter is devoted to the discussion of the conditions for the possession of maxima and minima by integrals which involve a single dependent variable and its first derivative. The conditions emerge from the application of weak variations, which are imposed solely upon the dependent variable and do not affect the independent variable. The problem thus is the simplest of all the problems in the caloulus; and the analysis, ‘though modified and amplified from the original form, is substantially in accord with the analysis that occurs in the researches of Euler, Lagrange, Legendre, and Jacobi. ‘The Weierstrass method, in which variations of both the dependent variable and the independent variable are considered, is reserved for the succeeding chapter. Types of problems arising from integrals maxima. and minima, 13, The calculus of variations deals with those problems which, enun- ciated in mathematical phraseology, require the determination of the form of an unknown quantity or unknown quantities as a function or functions of a variable (being a dependent variable or dependent variables), so that some integral may assume a maximum or a minimum value, The dependent variable or variables will be supposed to occur in the expression of the inte- gral: a derivative or derivatives of the unknown variable may occur in the integral; and it is supposed that the integral cannot be evaluated for un- specified forms of that variable, so as to become free from quadratures. Sometimes the demand is that the maximum or the mininum is unconditional. Sometimes there is a demand that the required maximum or minimum is to bbe possessed subject to assigned conditions, which may have a variety of forms: thus one form of condition requires that another integral of similar form may possess a fixed value. Even when there is only a single dependent variable, various cases may cccur. Thus there may be only one independent variable; advantage then accrues, if only by way of illustration, from the association of the problem with the geometry of a plane curve, There may be two independent variables; the corresponding advantage comes from an association with the geometry of a surface. ‘There may be two dependent variables and one independent variable: the corresponding advantage comes from an association with the geometry of a twisted curve. Or there may be any/number of dependent variables and a single independent variable; and then a natural association, for purposes or needs of illustration, is provided by some dynamical system,12 MAXIMUM AND MINIMUM. [oH 1 ‘The central demand is for the possession of a maximum or of a minimum. The maximum is to be a true and a full maximum in the mathematical sense of the term, as distinct from the phrase ‘largest value’ in common parlance : that is to say, the change in the value of the integral, which is caused by every small variation, however general or however particular, or by any number of small variations of every type, whether taken together or acting individually, must be an increase, uniformly definite in character as an increase, though its magnitude will be small because the variation or varia- tions effected from continuous finite quantities, are small. Similarly, a minimum is to be a true and a full mathematical minimum ; that is to say, every small variation (of the preceding range in all its extent) must lead to a decrease in the value of the integral, uniformly definite in character as a decrease. Should the integral provide a decrease for only some variations but not for all, its value cannot be declared a maximum; and, similarly, if it should happen to provide an increase for some variations but not for all, its value cannot be declared a minimum. Partial maxima and partial minima are not recognised as maxima or as minima. Thus a mountain peak is a place of maximum elevation ; the bottom of a tarn is place of minimum elevation ; the top of a mountain pass is not a place of either true maximum or true minimum in elevation. In discussing questions of maxima and minima, it is an advantage when the independent variable inereases continuously throughout the range ; if it, decreased continuously, a mere change to the opposite direction of the range would allow the desirable continuous increase. Sometimes care has to be exercised, at least initially, in the sclection of an appropriate independent variable; thus, if there were a question of determining a curve of shortest length so as to satisfy some permanent condition, it is preferable not to choose the length of the are for the purpose. Integral of the first order : its variation, 14. In order to give an indication of the method of analysis, we shall begin with the simplest general problem, and shall proceed to obtain neces- sary tests for the possession of a maximum or a minimum by an integral frewyras, where y/ denotes dy/da. The integral® is supposed taken between fixed con- stant limits; and no external conditions are imposed. ‘The unknown element, to be obtained for the purpose, is the form of y as a function of the variable @; and, in the first instance, we shall use only special variations—that is, we shall assume that y alone is subjected to weak variations. In the earlier * Sach an integral, involving only the first derivative, will be called an integral of the fret order.15] FIRST VARIATION OF SIMPLEST INTEGRAL 1B historical development of the subject, such variation was represented by the mathematical expression 5 iy, where the symbol 5 (perhaps in imitation of the d in the infinitesimal clement dy: it was introduced by Lagrange) is assumed to impose the restriction to smallness, and where the properties of the quantity Sy are not always explicitly stated, sometimes not even hinted. A varied value of y can de taken in a more definite form ytev, where « is an arbitrary constant and is so small in magnitude that any posi- tive integral power is unimportant relative to every lower positive integral power. Also, v is any regular uniform funetion of « within the range of the integral, and all its derivatives also are regular uniform functions of « within that range. Moreover, » otherwise is an arbitrary function; and we further postulate that » is independent of x. Denoting the original value and the varied value of the integral by J and J respectively, we have T-Inffey ten ¥ +e) -Se,yyhde, the integration extending between the fixed constant limits, We further assume the function f to be of such a form that f(e, y-+xe, 9’ + «0’) can be expanded in a uniformly converging series, which proceeds in powers of «. ‘Thus we have, after this expansion, Ja Lael, + hel, + By L [eZee 2) de, I, = [(o f+ 20 PL where nef agg tn) with the same range of integration as for Z; and where R, denotes the aggre- gate of terms that involve third and higher powers of x. Manifestly the quantity «J, (called the ‘first’ variation), if it does not vanish, will dominate* the value of the right-hand side; and R, is unimportant in comparison with wl, if I, does not vanish. Modification of the First Variation. 15. The expression for J, is to be modified. We have fv Bae [o%]- - fox Has, where the quantity in square brackets is to be taken at the limits, But when the limits are fixed, no variation of y is admissible for those values of 3 * See foot.note on p. 17.4 VANISHING FIRST VARIATION [eu 1 consequently, » vanishes at the upper limit and at the lower limit, Hence nefo 2 ype Now the quantity «Z,, when it does not vanish, dominates the value of JI. In that event, a change of sign for x changes the sign of w/,, that is, changes the sign of the value of J—T: in other words, one variation would lead to an inerease, and another variation to a decrease. As a maximum must be characterised by a decrease for all variations, and a minimum by an increase for all variations, the preceding possibility must be excluded: consequently the quantity «J,, and therefore the integral J,, must vanish, Let Y denote the quantity & t& ))- Then the integral Z,, which is equal to fovas, must vanish for all arbitrary values and forms assigned to the regular function v, ‘This result cannot be attained unless Y’ vanishes throughout the range, If ¥ were positive in any portion of the range, the integral would be positive, for an assigned positive value of » through the portion and a zero value elsewhere; and the integral would be negative, for fan assigued negative value of » through the portion and a zero value else- where. Similar definite values, positive or negative at will, could be secured for the intogral, if ¥ were negative in any portion of the range, Hence we must have : Hh) =o. yy acl 16. This relation Y= 0 is either an identity, or an equation not identi- cally satisfied. If ¥=0 is an identity, then the term “Fy” which would arise from the ai second term in ¥ cannot exist, for there oo involving y” to cancel it in the supposed identity. ‘Thus Bi is zero, and therefore fis of the form fayUrV, where U and V are functions of « and y only. Then 7 Y= 2 +E - 2 ji ay_aw_ you tay ae ay" fan expression which can ao identically only if av _av_y17) CHARACTERISTIC EQUATION: EULER TEST 15 that is, if some quantity M (x, y) exists such that amy _ aM vag URS In that case, we have I=[fex de -[(V Ba =U@y; and so the integral can be evaluated explicitly, contrary to the initial re- quirement (§ 13) that the initial integral is not capable of explicit evalua- tion *. ‘This possibility is excluded, as being irrelevant to our purpose; and we therefore assume that Y=0 is an equation not identically satisfied. Further, if f has the same linear form ‘=U + V, and if the equation ov av oy a 7° (which now is the form of Y= 0) is not identically satisfied, Y= 0 becomes an equation involving # and y alone: it provides y as a function of « which, when substituted, renders the integral J merely dependent upon known constants, Such a result is insignificant for our purpose; no element is left which can be used for the construction of tests concerning maximum or minimum. This second possibility is therefore excluded: that is, we reject from further consideration the case when f is a mere linear function of y’. Characteristic equation, and curve ; the Huler test. 11. In all other cases, 7, is not zero: the equation Y=0 isa differential equation, manifestly of the second order. In passing, the form of the equation is to be noted; the second derivative y” occurs only through the term E4.y" to which the othor torms are additive; and a later test, which will be imposed upon mn if a maximum or a minimum is to exist, gives to this differential equation ¥=0 a character that is of organic importance for its primitive. ‘This equation Y= 0 is often called the characteristic equation. Tt deter- mines y as a function of ©; and the curve, which is the geometrical repre- * When 1 can be thus evaluated, the problem does not fall within the range of the ealealus of ‘variations ; the value of F then depends solely upon the values of z and of y at the two limits of the range,16 CHARACTERISTIC CURVE [our sentation of the functional relation, is called the characteristic curve, The equation was first obtained by Euler, in 1744, and indeed was obtained for the more general integral which involves derivatives of y up to order 7. Sometimes it is called Euler's equation, hardly a distinctive name; it is the first of the essential tests to be satisfied if the integral is to have a maximum or 8 minimum. Note. The integral of the equation, or properties of the integral when the actual integration of the equation is not feasible, will arise later (§§ 22-25) for detailed consideration. But’ two simple cases may be noted where one stage in the integration can be attained at once. (i) Let the function f(e, y, y’) not involve y explicitly, so char £ 0 then the characteristic equation is a (of az ley, so that we have of ay 4, where A is an arbitrary constant. (ii) Let the function f(e, y, y’) not involve « explicitly. ‘Then YY oF Ley aby, a (ef), Fy (iy) +a" on using the equation Y= 0: integrating, we have _ where B is an arbitrary constant, a, ay *® In each of these simple cases, the equation remaining to be integrated is only of the first order. Ex, 1. Required the curce constituting the shortest distance between to points, or a point and a eurve, or two curves, in the same plane. ‘We shall deal only with the characteristic equation; terminal conditions, and other possible conditions or testa, either intrinsic to all the problems or accidental to any. particular problem, belong to a different (and later) range of investigation. In the present 80, to fasyo ds, 0 that /(2,9.¥) which is (1+yh in independent of 2 and of y; and the example belongs to each ofthe simple cases adduced. "The first gives yaty)~18) ‘THE SECOND VARIATION Ww ‘and the second, on reduction, gives Gtyy-ber, From either, we have 2 and therefore et) yoness ‘where m and c are constants to be determined hy other considerations Here 4?+J*=1. Generally, if the fanction fis free from explicit occurrence of # and of y the arbitrary constants 4 and 2 in () and (i) are uot induendont ‘Be. A unit muss moves in a plane fed of force ao that its velocity depends only ‘upon its distance from a fixed point, and f ude along the path, between any two points, is maximum or « minimum, Prove that the foree is directed to that fixed point; and determine the orbit. Quadratic terms in the small variation : the ‘second variation. 18. We shall assume that the characteristic curve is adopted, so that the Euler test is satisfied ‘The term «I, in the expression for JT now dis. appears; and sl, (called the second variation) becomes the governing® term, As « is constant, the sign of this term depends upon the sign of Z. If the sign is positive forall variations of the type considered, then J is positive for all such variations; and J is (so far) a minimum. If the sign ie negative, then J ~ I is negative for all the variations; and I'is (s0 far) a maximum, Ts therefore becomes novessary to make a detailed examination of [., which denotes the quantity ee oe J(eF 20 ay tH) ds, For brevity, we shall denote x. ay x dy for fou» fia» respectively. Because 7 (WA) =0N’ + 20v'd for any quantity 2, we have Tet [0A] = flo fa +) + 200 (fut) Hof] de, where [2*] denotes the term vf taken at the limite, and it is implicitly assumed that vA is continuous through the range of the integral. As the limits are fixed, v vanishes at both of them; hence, if d is assumed to remain faite at each limit, the quantity [vA] vanishes, Bearing this last assumption in mind, as one of which account must be taken, we have Tax le Fo X) 4207 (fut) 40 fp) de, ‘Thus far, no quantitative condition has been imposed upon 2; now let it be chosen so that Su Fot®)= (fat In a series atm +e2ens of finite um S, the sign of 5 is governed by that of a for talleently emall values of «. Let 8, neosserily nite, denote the largest numerical elce of the sum «-+n0e+s80)+..., for ~hep- then ¢ satisfies the equation IaSe thi £-Su-fae=0, ‘a linear equation of the second order, of which the coefficients are known functions of z, And then, with such a vaiue of 2, it follows that Iam [a(t ae = [te v-24y de, In order to take account of the most general admissible variation xv for y, it is desirable to have the most general value of ¢ possible: that is, we must have the primitive of the linear equation in z. But subject to restrictions actually imposed, 2 is at our disposal, because the main purpose of its introduction is merely a modification of the expression of Z,; therefore the general elements in z, which survive after the restrictions are obeyed, can be used for other aims that may be in view. If the value of y, satisfying the characteristic equation, is only a particular solution of that equation, then the foregoing equation for z must be com- pletely integrated. If the primitive of the characteristic equation has been obtained, the primitive of the equation in 2 can be deduced by direct pro- cesses alone, as follows. The subsidiary characteristic equation. 19. To obtain the primitive of the equation in 2, we return to the forms of I, and J,, Two expressions for J, have been obtained, viz. the original form af ye L |(-Z+"%) dz, hafo ¥-% £(¥ (Sane fora, deduced from the first in association Ge the terminal conditions; and the two expressions are equal to one another, because terms at the fixed limits vanish. ‘To both forms of J, let the same variation be applied as was applied to the original form of the integral I; let J, be the deduced value of J,, and and a form19) SUBSIDIARY CHARACTERISTIC EQUATION 19 let Y, be the deduced form of Y, From the earlier form of I, we have ~tae [ef th J-h= S020 E+ Hh) det Ky where K, represents the aggregate of terms in «* and higher powers of «; that is, neglecting X, in comparison with quantities involving only the first power of «, we have 2 From the later form of J,, we have Dae J-h= fo(¥,-¥) de, there being no variation of v, because it is assumed to be a function of « which is not subjected to small variations, Now, as we have = Yar (fav fav) — 0 Sad +fav) =" ofa) £ (fa Our immediate concern is with the sign of J,, when the equation ¥'=0 is satisfied so as to make Z; vanish ; s0, as before, we imagine the value of y, determined by that equation, to be now substituted in the coefficients. As y is a solution of ¥ =0, let y +47 be a solution of ¥,=0: that is, y +x is @ solution of what still is the characteristic equation, Hence when the value 7 is assigned to v, we have Y,=0 as well as Y= 0, and so a fo1- Eat) =0, manifestly a linear equation of the second order in 7, the coefficients in which are functions of 2, ‘The quantity y +7 has been introduced as a solution of the characteristic equation when small variations are imposed, so that 7 can be derived from the general value of y when this latter is known. But, if the value of y satisfying Y =0 be only special and particular, the quantity » must be derived as the primitive of the new equation which, owing to its source, may be called the subsidiary characteristic equation, We now have rh=S,-L, = fo Vp des20 CANONICAL FORM OF SECOND VARIATION (cn. and therefore Lem fo {cto fe0- Z Ha)} de = [Efe fonda Hard} de = [2 Eilon v0) de = [Loma] +ffa(o 9% Thus t+ [o(v-0%) 4] = [fa(v v8) de, where the terms in square brackets are to be taken at the fixed limits of the integral, and where (as before) an implicit assumption has been made as to the finiteness of the subject of integration through the range. With this assumption, and remembering that v vanishes at each of the fixed limits, we make these limit-terms zero, and finally obtain tele) ‘The equation for 7 is the same as the equation for 2: we therefore can take 7 as the value of z, and the two expressions for J, are now the same. Whether be deduced from the value of y, when the primitive of the characteristic equation is known, or be obtained by the independent inte- gration of the subsidiary characteristic equation, the magnitude m can be taken as any form of the primitive of the latter equation that may have been found. ‘The function v must be allowed all arbitrary values, subject to the requirements of regularity and of vanishing at the limits of the integral. ‘Be, Prove that the intermediate forms of J, are equivalent to one another, in virtue of the variation of the quantity the evanescence of which makes the two forma of J equivalent. The Legendre test. 20, The sign of the quantity J, is to be a steady positive, or a steady nogative, for all possible variations of y represented by wo, where v is limited to be a regular function of z, which must vanish at the extremities of the range of the integral but otherwise can be arbitrarily chosen. As regards the sign of J, we note that (vv) in always positive unless it is ero5 and, later, we shall assign the condition that it shall not be zero unless » itself is zero, in which event there is no variation, ‘Then as a first condition for the permanence of sign of I, it is necessary that fy shall keep the same sign throughout the range.21] LEGENDRE TEST a1 The requirement is established as follows. Suppose, for a part of the range, say from a lower limit a of the integral up to a value z, which is less than the upper limit of the integral, that fy, has one sign, say the positive sign, and for the rest of the range that 7, has the other (the negative) sign ; and consider the effect of the two following variations. In the first portion of the range, let v=(c—a)'(x,—2)*V, where V is a regular function of 2, s0 that v and v obviously vanish at «=a and = a,; and everywhere in the second portion of the range beginning at «=«,, let » be zero, For this composite small variation, the sign of Z, is positive. Next, everywhere in the first portion of the same range, let v be zero; and in the second portion of the range, let v= (a—a,) (c—a} U, where U is a regular function of 2, so that » and v’ obviously vanish at c=, and =. For this composite small variation, the sign of J, is negative. Hence, on the hypothesis assumed as to the change in the sign of f, within the range, the sign of Z, can be made positive by the choice of one kind of variation, and can be made negative by the choice of another kind of variation, ‘The same alternative sign can be obtained for I, by appropriate selected variations, whenever changes of sign* of f, occur otherwise in the range of integration. We therefore infer that, for the pos- session of a maximum or a minimum by the integral Z, the quantity fy cannot change its sign in the course of the range of integration. If the sign is regularly positive, the value obtained for the integral is (so far) a minimum if the sign is regularly negative, the value obtained for the integral is (80 far) a maximum. But the new condition has only been proved to be necessary; it has not proved to be sufficient, and it is not in fact sufficient, to secure a maximum or a minimum, ‘This second test was first obtained (in 1786) by Legendre+; and it is not infrequently called after him, as the first test is called after Euler. The Legendre criterion manifestly is distinct in character from the Euler eriterion. When the Legendre criterion is satisfied, the property of f, thus possessed is important, as regards both the nature of the integral of the Euler charac- teristic equation, and the nature of the primitive of the subsidiary character- istic equation. 21, The quantity v— oF remains for consideration, in relation to J,. As it occurs squared, there is no question of sign; the collapse of the signi- ficance of I will ensue if, through the appropriate selection of », this quantity can be chosen so as to vanish over a continuous portion of the range—a zero value at merely isolated places is not of material importance. Such an eventuality would occur if we could choose v as a constant multiple of , the only relation (except the irrelevant persistent zero for v) which can secure * Tho character, postulated for the fancton f,implioilly requires that, in any fnite range, ‘the number of changes of sign, for any derivative euch as fy, shall be finite, the $5.22 PRIMITIVE OF THE [our continuous zero for the magnitude in question, Now v is subject to the conditions that it shall vanish at each extremity of the range of the integral. If, therefore, the suggested relation is a possibility, the quantity » would be such that it could be chosen so as to vanish at each of these extremities. (The relation would be @ fortiori possible if » could be chosen so as to vanish at two places within the range: for we then could take v a constant multiple of within that portion of the range, and a steady zero outside that portion.) As 9 is the primitive of a linear equation of the second order, its general form is a=am tons no matter how the primitive is derived; the arbitrary constants o, and are at our disposal, so that they can be used for any desired purpose. ‘Manifestly a relation can be established which would make 7 vanish at one extremity—say, at the lower extremity of the range. If the functional character of 7 is then such that it does not again vanish within the range, or (what is organically the same restriction) if the range of the integral does not extend as far as the first place, where 7 vanishes next after vanishing at the lower limit: the contemplated relation is excluded from the possibility of occurrence. The quantity v’— ea could not be made to vanish for admis- sible variations, as these are bound to vanish at the limits; and, when the Legendre test is satisfied, J, would have a permanent sign for all such variations. The primitive of the subsidiary characteristic equation. 22, ‘The function 7 thus becomes of critical importance; we therefore proceed to its discussion, and we shall deal with its construction by the most general process, viz. by derivation from a supposed knowledge of the primitive of the characteristic equation. ‘This characteristic equation is x_ 4/7 vod de) eee by dno ~ ayay! aye" =0, being of the second order and (usually) not linear, ‘The only term, which is # y's and, when the Legendre test is completely satisfied, involves the coefficient x, does not vanish for any value of a in the range of # that is considered, ‘Then the general existence-theorem* for ordinary differential equations gives us the result:—if values 2 =a, y=, y=0,, be chosen ‘+ Seo my Theory of Dierentiat Equations, vol. ii, § 08.22) SUBSIDIARY CHARACTERISTIC EQUATION 23 arbitrarily, subject to the conditions that a lies within the range of in- tegration and that these values constitute a set of ordinary (that is, non- singular) values for the derivatives of f which occur; a uniform integral of the equation exists, uniquely determined by the requirement that y and y’ acquire the respective values Q, and Q, when 2=a. This integral is of the form y=0,+0,(7-a)+(e-a¥ R(w—a, G,, ), where R is a regular function; we may denote it more briefly by y= (2, G, G), where $ is a regular function, and where C, and C, may be regarded as arbitrary constants, the presence of which constitutes a primitive. ‘The first form is not the unique expression of primitive for all initial requirements. But the quoted theorem establishes the existence of a primitive, containing ‘8 couple of arbitrary constants; the later form, taking no special account of the initial value of 2, will be adopted as the analytical basis for farther developments. Every primitive y= (a, C,, C,) satisfies the characteristic equation Y¥=0, for all positive arbitrary values assigned to the constants C, and C,, It therefore will provide a primitive, when variations are applied to C, and C,. such as a choice of O,+xo, and C,+ xc, in place of Q, and C, respectively, where ¢, and ¢, themselves are arbitrary, and where x is the current small constant. When the new value of y is denoted by y+ «7 on the customary assumption as to the smallness of « in magnitude, then ag, ab ama, + eae 5 or, if 08 ag "= 50 ™ 50, we have 2= 0m + en. Now the new value of y satisfies the equation Y,=0; hence 7 satisfies the equation Y,—Y=0, that is, it satisfies the subsidiary characteristic equation pene ofa) - Fe fast)=0. As 7 involves the two arbitrary independent constants, the value of 7 provides the primitive of the equation, which may be written in the form Ya fat + hid —(fa-fu)1=0.ory PROPERTIES OF THE PRIMITIVE [ou 1 Properties of the primitive of the subsidiary equation ¥ = 0. 23. One or two useful properties of this integral may conveniently be established at this stage. The quantities m and 9, are clearly not in a constant ratio, because (when they are taken in the initial form) mals(e-ayh, noe-at(e— aye, and B is a regular function of #— pendent. Now : that is, 9, and 9, are linearly inde- Son" + fan! — (Safa!) = 0 Sane" + fain —( foo fa’) = 05 and therefore Farm” — mn”) + fal Cen’ — tan’) = 0. Thus Fan! —nm’)= A, where A is a constant, Asm and n, are linearly independent, the quantity ‘ants’ — mm’ does not vanish. The Legendre test is supposed to be satisfied throughout the range, and therefore fi, does not vanish; consequently A does not vanish, But fi contains no arbitrary element in its first form, merely as derived from the subject of integration in the original integral ; it usually will contain C,, or O,, or both C, and O,, after substitution is made for y. Also n, and 7, contain no arbitrary element other than C, and Q,. Hence A may involve both O, and C,, because it is equal to the value of the left-hand side in the last equation ; but A, different from zero, is not a new independent arbitrary constant. From the last equation, it follows that dyn) 4 del) 5h On the right-hand side, the signs of A and of »,* do not change; and, by the Legendre test, the sign of fi: does not change throughout the range; hence the sign of the right-hand side remains unchanged through the range. If the sign is always positive, then 7 increases throughout the range. If the sign is always negative, then ” decreases throughout the range. This result, as to the persistence of sign, is (or may be regarded as) a consequence of the uniformity in the sign of fy through the range; therefore it is a consequence of the Legendre test.25) OF THE SUBSIDIARY EQUATION 25 24, There is a temptation to infer, as an immediate and obvious corollary, that the quantity **, which is always increasing or always decreasing through- out the range, can never acquire again, in the course of the range, the value which it possesses at the lower limit. The inference would be justified if the quotient were always a continuous function; it is not necessarily correct, if the range should include a zero of »,, which would cause a discontinuity through an infinite value, always provided such a zero is not simultaneously cose and 74=sin #, so that £2) is always positive and never zero; we know that the function tan e increases as » increases, But tan can resume an initial value, say the value zero for «=0, if the range is extended as far as w=; the explanation is that, in spite of the steady increase in its value as « increases, it suffers a sudden set- back through its discontinuity as # passes through the value #= 4a. It is not therefore justifiable to assume, as a consequence of the persistence of a zero of m. For example, let 9, sign of 7 that this quantity cannot resume the value which it had at the beginning of the range. Note. From the relation Fa (tin! — mm’) = A, where the constant A is not zero, it follows that there can be no value of # within the range, for which 7, and », vanish simultaneously, (This result has been anticipated in the proceding argument.) It equally follows that there can be no value of 2 within the range, for which m’ and m,' vanish simultaneously, It is unnecessary to consider a possibility of a value of z for which m, and a (or, alternatively, m, and n/) vanish simultaneously ; it would follow, from the linear differential equation ¥=0 of which », and m are integrals, that m (or m2) would be zero over a continuous range. 25, The preceding construction of » is based upon a supposed knowledge of the primitive y of the characteristic equation. It shews that, if such primitive is known, then formal integration of the subsidiary characteristic equation is superfluous: the primitive of that equation can be obtained other- wise, by purely direct operations. But if the primitive y is not known—if, for example as an unfavourable case, only a very special value of y can be obtained satisfying the character- istic equation—then, to complete the set of present operations, the formal integration of the subsidiary equation must be achieved. By some process, its primitive must be obtained.26 JACOBI TEST [ou 1 Note. It might happen in this event, that the couple of fundamental integrals composing the primitive are different from », and my. If so, let them be denoted by & and &. Then, by the properties of an ordinary linear equation of the second order, we know that there are relations maGht at, m=bE + bibs, where a, dy, b,, by are constants such that a,b,— a,b, does not vanish; and then mn! — mm’ = (dibs — abs) (Es Be — Fob’). The corresponding equation fa (Gib - B68) = B, where B is a constant, agrees with the earlier equation funn’ — nmi) =A, provided the relation A =(a,b,~a,b,) B is established between the constants ‘A and B. All the argument and the conclusion are the same as before. The Jacobi test : conjugate points on a characteristic. 26. It has been seen that, if maximum or a minimum is to exist for the integral, we must be precluded from the choice of a non-vanishing function » which (i), could be a constant multiple of the magnitude throughout the range and (ii), could also be consistent with the condition that » is to vanish at both the lower and the upper limit of the integral. If the choice were possible, then » would have to vanish at the lower limit, Because 1 = Om + OTs this requirement can always be satisfied by choosing one relation between the two constants which are at our disposal, so that —o,/o, is equal to the value (say m) of ™ at that lower limit. Now, owing to the other conditions that have been imposed, 7 is always increasing through the range of the integral or is always decreasing through the range. If, then, whatever be the range, the quantity ; never returns to the value which it acquires at the lower limit, then cannot again vanish. A non-vanishing function » cannot be chosen (i), to be a constant multiple of 7 through the range and (ii), to satisfy the terminal conditions. No new test emerges as an additional requirement to be satisfied in connection with the integral, If, however, the quantity can return to the value which it acquires at the lower limit, and if the range is sufficiently extended so that it includes the value of « where this return is made (and made for the first time, should27) CONSUGATE POINTS a7 the initial value recur more than once), then 9 acquires the value zero at this place. We then can take v= jn, where . is any constant (its magnitude is significant only as regards the scale of the small variation x to which y is subjected) ; and, for that variation, J, would vanish. In other words, some variation is possible under which a claim for a maximum or a minimum could not be regarded as established. Such a possibility must therefore be pre- vented; hence the range must not include the first place at which the mono tonic function again assumes the value assumed at the lower limit. It therefore follows that the range of the integral must not extend so far as that first place, if such a place exist; while, if no such place exist, there is no corre- sponding restriction on the range of the integral when there is to be a maximum or a minimum. ‘The place, where the initial value of a is next assumed, is customarily called the conjugate of the initial place. We thus have a new test—the third. in, the range of the integral must not extend as far along the characteristic curve as the conjugate (if any) of the lower limit, This test, imposing a limit upon the range of the integral, is due to ‘Jacobi*, and is frequently called the Jacobi test. The earlier discussion as to the variation of * shows that, asa test, it is independent of, and additional to, the Legendre test, 27. The Jacobi test can be expressed in a simple geometrical form. Let a characteristic curve AP... be drawn, passing through an initial point A; the position of the point A, and the tangont at A to the characteristic curve, are the geometrical representation of the arbitrarily assigned initial values determining the solution y of the characteristic equation. The special small variation consists of a small displacement of a point P of the curve along the ordinate, as to a point Q; when this small variation is effected along the curve, then (for weak variations) the inclination of the tangent Q to the displaced curve differs from the inclination of the tangent at P by a correspondingly small amount. The new ordinate y +7 is the ordinate of a new characteristic curve AQ..., passing through the initial point A, and having its tangential inclina- tion at A differing by the appropriate small quantity from the tangential * It was first stated by him in 1887 (L.c., §6); for the establishment, reference may be made to a memoir by Hesse, Grell, t, liv (1857), pp. 227—273.28 SUMMARY OF TESTS [our inclination at A on the original characteristic. If this consecutive character- istic cuts the original characteristic in some point after A, we consider the first of such points of intersection ; its limiting position, say B, as the initial deviation between the two curves at A is incessantly diminished, is the con- jugate of . ‘The Jacobi test requires that the range of the integral, taken along the characteristic curve, and beginning at A, shall not extend so far as the point B, thus defined as the conjugate of the initial point A. Ez, 1. We know that « straight line is the shortest distance hetwoon two points in a plane. Thus the characteriatc curve (Ex. 1, § 17) a straightline. Another characteristic ‘curve (also a straight line) through the initial point, slightly inclined to the first, does not ‘again moct the initial straight line, ‘The initial point has no conjugate on the characteristic ‘curve; there is no finite limit to the range of the integral, Ex,2, We know that the smaller are of the groat circle through two points on & sphere is the shortest distance on the surface of the sphere between the two points. Tater (§ 76), a* an illustration of the analysis, the great circle will be obtained as the ‘characteristic ourve for the spherical geodesic. ‘A consecutive characteristic, through @ point A on the great circle, is a consecutive great circle; the two circles intersect at the point B, which is diametrically opposite to 4. ‘The particular great circle between 4 and B cannot be declared a unique minimum between 4 and B: in order to have an actual minimum, the great-cirole range, which begins at A, must not extend beyond 2. Be. 3. Show that, if B is the forward conjugate of 4 on the characteristic curve, A is the backward conjugate of B on that curve. Summary of the tests. 28, It may be convenient to summarise the results thus far obtained in connection with special variations, as follows: L. The Luler test: the characteristic equation v-2-2(@) 0 must be satisfied; and (for subsequent purposes, but not as a part of the Euler test) the primitive of the subsidiary characteristic equation Y= (fo- fa’) — fal’ — fan" =0 must be obtained, in a form =c.m + ons where ¢, and ¢, are arbitrary con- stants, IL, The Legendre test: the quantity oy By (when, in it, the value of y derived from the characteristic curve has beon substituted) must keep its sign unchanged through the whole range of the integral.29) CONJUGATE AS ACTUAL LIMIT OF A RANGE 29 IIL The Jacobi test: the range of the integral must not extend s0 far, as to allow the monotonic magnitude n/n, to resume the value it assumes at the lower limit; or, in geometric phrase, the range of the characteristic curve must not extend so far as the conjugate of its initial point. Conjugate as actual limit of a range. 29. Tt may however happen that an assigned range extends so far as to include the conjugate of the initial point, In that event, a small variation, given by v= pn where pis a constant, will make the quantity I, vanish: that ia, will make the term in J'~J, depending on , to vanish; while, for all other variations, Z, does not vanish. Except for this particular variation, the terms in J—T involving powers of « higher than the second are negligible. But for this particular variation, those terms require consideration; and, among them, the most important initially is the term gx*Z,, where gO, a ae OF ys _ Lao Sit te + aon OL iA) ae In order that JI may keep an unvarying sign for variations—so that a maximum or a minimum may be sccured—it is necessary that Z, should vanish for the variation v= jan which makes Z, vanish ; otherwise a change of sign for the arbitrary constant « would alter the sign of «'Zy. The requisite condition is that a ae a I(r3 0 Ls ay apy tm" “aha aei)ae eee the original integral, or taken between the range bounded by the conjugate points, shall vanish. If this condition is not satisfied, then the integral does not provide a maximum or a minimum. If this condition is satisfied, then the sign of I,, eile , 2 OF m ve Lm | 5+ son aplag * Ot gE + agit aE) ae, must be uniformly positive for a minimum and uniformly negative for a maxitauin, A fuller discussion is given at a later stage (§ 76—80). Note. The same kind of consequence would follow, if an assigned range of integration extended beyond a conjugate. In that case, it is possible that I, might be made to vanish; then the term, involving «¢ in the expression for J=1, would have to vanish under the conditions making I, and I, vanish, and the term involving + would have to keep one sign under those conditions. But it will be proved (§ 76—80) that no maximum or minimum can be possessed by an integral over a range, which extends beyond the conjugate of its initial place.30 EXAMPLES [om 1 EXAMPLES. 30. We pass to some examples: in the first of these, the catenoid is discussed. Bz. 1. Find the plane curve joining two points such that, when the curve is rotated ‘round a line in a plane which passes also through both the points the eurface of the generated solid is « minimum. (A) Take the line for the axis of 2, The clement of are is (1+y')4 dz, where the positive sign ie taken for the radical. The ares of the generated surface is ae yds, =90 fy (ey? ds, the integral being taken in the positive direction of the axis of = between the values of = for the two points, Here the subject of integration is f; where fay (lty9, which does not involve explicitly 5 and therefore (§ 17, Note) a astyte: +6 garey ey ‘where ¢ is an arbitrary constant, Thus mc(ityt; and therefore : Le yeecosh 2, where b is another arbitrary constant independent of c This equation, containing two arbitrary constants b and o, is the primitive of the characteristic equation ; it represents a catenary, the directrix of which is the axis of revolution (the curve is the form of & ‘uniform heavy string, when it hangs freely). (B) Deferring temporarily the discussion of the possibility of drawing such a catenary through the two points, we proceed to the other tests For the Legendre test, we have oe ay? a+ynt ‘The sign of the radical is positive. The catenary does not meet the axis, 20 that y is a alway positive and does not vanish. Hence £4 is always positive ; and therefore, 6o far as the Legendre test is concerned, the catenary provides a minimum superficial area, (©) To apply the Jacobi test, the quantities 7, andy, taust be constructed. As the arbitrary constants in the primitive are b and ¢, we take30] CATENOID AS MINIMAL SURFACE 31 ‘To trace the value of m/m, we use the geometrical representation. At any point P of the curve, draw the tangent cutting the y directrix in 77; the vertex is A; and AB, 4 PN are ordinates: and OB=b, AB=«. Oo 8 to (e-bacath Sa y-nraar; consequently, for successive points on the curve, the postion of the point 1? on the Ainectrix gives tho measure of the value of 7. Also = (2) is positive, so that 7? : ; n ‘increases with 2, corresponding to the motion of 7'to the right as P moves along the curve im the direction of increase of Now lot a moving point @ pass from the initial point Z of the curve corresponding to te lower limit of the intagral, 17, being the initial tangent. As Q moves along the curve towards 4, the intersection of the tangent at Q and Or moves to the right; ax @ moves through 4, that intersection sufers a violent discontinuity in position from the extreme right of Ox to the extreme left of Ox; and then, as @ still moves onward, the intersection still moves in the positive direction towards the right. When the intersection returas to its initial position, the postion of @ in the point of contact (say J) of the other tangent from 7, to tho catenary. This point Fis the conjugate ofthe initial point Z Accordingly, the Jacobi test requires that a range of the intogral, which begins at Z, shall terminate before J, the conjugate of the point Z. We shall assume that the Jacobi test is satisfied: the rango of the integral is to be restricted within the indiated limita, (D) The three critical testa are satiafed by the curve, as drawn, But there persiste the fundamental question : ean a catenary be drawn, passing through two arbitrarily given points and having the azis of 2 for ite directrix ! There are subsidiary questions: if a catenary can be ¢o drawn, is it unique? if it is not unique, which (if any) of the possible catenaries can bo taken a8 satisfying the requirements? In the frst place, no such eatenary ean be drawn when the two points lie on opposite sides of the axis Oz; the analysis becomes illusory. If, in the first integral ytyy~ of tho chaructriti equation, zro vale is assigned to tho arbitrary constant e we could have y=0, or y indfnitely large, of both; but such variations do net accor with tho orignal asruzption that y(1+/), th sujoot of integration, ia regular function for auch arguments. Manifstly, some diferent analysis must be devised, in order to admit the Ihypotesis tt the points lie on opposite sis of Os—a. hypothesis which ail leaves a reasonable problem. Next when tho two points are on the same side of the axis, we shall stume (in onde thin stago, meray to simplify the immediate algebra) that the twopointa are atthe same distance # from the axis; and wo shall denote thir distance apart by 2a. When the origin in taken atthe point 2, we have 6-0; the equation of the catenary is sroon!,32 CATENOID AS MINIMAL SURFACE [on 1 and the quantity © is to satisfy the equation kemecosh ®. ‘To determine whether this transcendental equation has any real positive root , let ook ber, then any real point common to the curves veld, Yeacoh will give areal value of Let € denote the abscissa of a point X on the 7 now catenary where the tangent to the curve passes through the origin 0; as the equation of that tangent is : : Foacnhfa(x-Esioh§, we have oth £m £, ana ° % Simple geometrical considerations shew that this equation hes only a single positive root, From tables, we find that franco 33° 327 ‘approximately. ‘Thus, ifthe angle XOK be denoted by a, then a= 56°28, also approximataly, (@) Hf £T7- Note. When dealing with the eycloid, advantage sometimes acorues in the discussion of the reality or the uniqueness of the curve by choosing the current variable to be the ‘angle 6 through which the generating circle of the eycloid has, in the usual generation44 ‘EXAMPLES (on. 1 rolled instead of the angle ¥ through which its tangent has turned. In particular, the aulty at the cusp arising from the latter choice—the variable suddenly increases by at such a point—is avoided by the former choice. We have vate 405 the coordinates, referred to the cusp an origin and the directrix as the axis of 2, are X=a(d—sin®), Y=a(l-cos6). In the preceding example, the initial value of 6 is zero. The final value is @=27~28; for the final value the point (X, ¥) lies on the Kine y=(e-o)tanB, ‘and, as before, we have 2a(n—8)=6. Ex, 4, When the particle starts downwards along the curve in Ex. 3 with an initial velocity (29h), and the curve is referred to the initial -point as origin and the horizontal line as axis of, obtain the coordinates of a point on the curve in the form ythma(l-cosd), #+k=a(0—sin 6), where a and £ are the two constants of integration. Shew that, if the time of quickest descent to a given point below the axis of « is required, one (and only one) cycloid exists satisfying the requirement, Ez. 5, Prove that an ellipse is a brachistochrone between two points on its range, under a central forve to one focus varying inversely as the square of the distance from the other focus. Find the limit to the range between the two points, in order that the property may be secured. Ex, 6. A plano curve isa free orbit under one central foree, and is a brachistochrone under a different central force; and v, «, aro the velocities of the respective moving particlos at any point, Provo thet ev is constant, Ee.7. Prove that, in an ellipse under a central forve to the focus varying inversely as fhe aquare of the distang, the quantity fods between two points on the urve (v being the velocity at any point) is a minimum, provided the length of the are is loss than a. semi- circumference by a vectorial angle greater than -1_Ssing 200" F econ ate where ais the angle betwoon the radii vectores to the nearer apse and the initial point of the arc, and e is the excontricity of the ellipse Be.8 Prove that a minimum of the integral [orstssey +04 ride 1 provided by y=0, if the upper Limit and the lower limit of the integral have the same sign; ‘and that, for a minimum when the lower limit of the integral ia —a and the upper limit is 0, where a and c are positive, e must be loss than the root of the equation sinh 'etsinbtan 4am) 420409), A geometrical interpretation is immediate, The curve is a meridian on the paraboloid of revolution 2=21+y!; the meridian is a geodesic between any two points on its length, if the vertex does not lie between them ; there is a limit to the range of the meridian as the geodesic between the points, when the vertex does lie between them. Also the example is adduced to indicate the analysis necessary (§ 25) when only a special integral—here, it is y=0—of the characteristic equation is known,34] EXAMPLES 45 ‘The characteristic equation is (etsy _ uta ty 0 at de at where 4 denotes 142%4 22yy/+(+y%)y/* A4etty)y"=(ey¥-9) (+9). A first integral of the equation is fan equation which reduces to ay-y~at, where a is an arbitrary constant; the primitive is expressible by circular and logarithmic functions, and will be obtained later. Manifestly, y=0 satisfies the characteristic equation ; it is a very special integral, and it cannot be used in the same way as the primitive for the construction of the critical function in Jacobi’s test. For the Legendre test, we have in general, and oe <0+9-4 for tho particular solution. Thus $f is always postive. The stated intogral, so far as concern this test is 4 minimum. For the subsidiary characteristic equation ¥'=0, we have ay te a> Byay when the value y=0 is inserted in the general values; hence the equation for» is 2 (L+a%)~4, dtasey-tnpin Zea se heo, that is, (L429) 1" = a7'+ ‘The primitive of this equation is Bet A (esinh 12 =(14258. Thus we can take =a maesinh“te-(1ts9h, and therefore inh-he Ea tet n me Hence a 2) 0208 dz\n aie that 2 is always increasing so that 2 is away 1 1f, in passing from the lower limit to the upper limit, does not acquire the value 2er0, then ™ remains finite throughout; as this function increases through such a range, it i cannot again acquire its initial value in the range.46 RANGE OF A GEODESIC ON A [on Now let # acquire the value zero in the range, supposed to begin with the value —a For that value of #, the value of = is at ~sinh1a 4 CE Anza nga to psn rough he ln ‘passes from a large positive ‘value through infinity to large negative value, and then again increases, Tf when == where 1 is positive, it again acquires the initial value, we have that is, w is given by sinbtutsinh-ta=2a tum + Lagan. ‘The possibility of aoquiring the initial value must be excluded if« true minimum or a true maximum is to be acquired; hence the range of the integral boginning with the value —a for 2 must not extend as far as the positive root w of this equation. ote, ‘The primitive of the charactaristic equation for the general problem of minimising the given intogral, viz, . (tatty?) 7 =(0y/-9) 497) can be obtained as follows, We have and so Thus =v mnstant = Tay} Tr ata omant ie where a is an arbitrary constant, Let the variables be changed according to the relations sarcosé, yersind, ‘The equation becomes ae(eeyt arr (+3) and the primitive is @-angatog (41) ~ tana), where tl ttyl at Weng mppcae | Omtante ‘This equation gives the characteristic curve; interpreted geometrically in connection with the paraboloid of revolution, itis the projection of a aurface geodesic upon a plane perpen- dicular to the axis, For the Jacobi test, we have so that the critical quantity is34] PARABOLOID OF REVOLUTION 47 or, if we write coth vaun(tattt)t this critical quantity is tanbe ~ianh¥ ora" Now Valways increases ; hence the only way in which V can resume its initial value is by Passing on ® positive increasing range for v through the value infinity, then becoming nogative and gradually increasing again. If then », and —v, are the two limiting values, we have Veo. tanh oy ~ 8 aoa ta? which limits the range. In the form propounded*, the problem leads to a meridian goodesio; the projection is ‘ straight line through the origin, and a is zero, ‘Then, for a limited range in a minimum, the projected range encloses the origin ; if dy and dy are the distances of the limits of the range from the origin 1 1 qtantcoth a, gotant -coth v2, so that Sheed. hoon Ee.9, Pisoun the intgrad [gh ytd betwen fed init, with a view to a masi- ‘mum or minima value, As (1+) y/~, the subject of integration, does not involve the independent variable, a first integral of tho characteristic equation (§ 17) is (+x) ¥-*-¥ {-2 (+9) ¥~}=constant, that is, Ltytacys, where ¢ is an arbitrary constant. ‘The primitive is yosinh (ce-+2), whore a and ¢ are the arbitrary constants. ‘We have Ff gly ey which does not change its positive sign for any range; hence the Legendre test will permit a minimum, For the Jacobi test, we have ey, mm Ymoosh(ceta, m= Yaromh (cea); om r, which does not resume an initial value through any steadily increasing range, there is no conjugate of an initial point on the character It thus appears that the three tests, which have been established in connection with special weak variations, would point to the existence of a minimum for the integral when ‘y is an ordinate of the characteristic curve y=sinh (cr +a). * The question was set in the Mathomatical Tripos, Part I, 1804.48 INSTANCE WHERE WEAK VARIATIONS ARE INSUFFICIENT [CHL I But we must notice that this minimum, while provided by special weak variations, ia not a true minimum. Its value can bo decreased by other variations as follows, Let the curve be required to pass through two points (41, 94) and (24, 72) such that H>H>0, >>; ‘then ade 1m 7 ‘so that ¢ is positive. Take a range for the integral such that, whatever be the value of 4, the increasing argument ex-+a is positive through the range. Then y is positive through the range; while y and y/, along the curve, increase continuously with increase of, Now take successive points P, Q, 2, S,... along the curve, in the direction of increasing, these points being very close together. Draw tangents at Q, 2 8, ... succes sively, meeting the ordinates through P, Q, & ... respectively in Py, Qty Ryy n+ 80 that the points Py, Q1, 2, .. lie between the curve and the axis of «. ‘Thus we have a broken line PP,QQ,2R,..., which is a small variation of the curve PQR... but is not a weak variation; it isa strong variation, Consider the integral along this broken line and com- pare the portion along a section PP,@ with the portion along the corresponding section PQ of the characteristic curve, Along PP,, the quantity 9’ is infinite; the contribution is ero. Along P,Q, the value of y’ is greater than its value along PQ, while the value of y less than its value along PQ; on both grounds, (1+y%) /~* is less for P,Q than for PQ. Thus the element of the integral for P,Q is leso than the clement of the corresponding integral for the curve PQ. Similarly for every pair of associated elements, "Thus the ‘whole intogral along ...PPQQ,AR;... is less than the integral along ...PQZ... Consequently the minimum provided by the weak variations ia not a true minimum; for there are other small variations which lessen that minimum (see § 215, Ex. 5). Note. ‘The systematic consideration of strong variations is deferred until Chapter VIL First integral of the characteristic equation : Hilbert’s theorem. 85. Tho solution of any problem of the present type requires the inte- gration of the characteristic equation v2 (¥ that is, of ae of Of of by Gedy’ ~ dyay Y ~ ‘The equation is of the second order. For a form of the function f, which is quite general and unspecialised, the explicit expression of a primitive or of a first integral (except as an infinite series, usually to be associated with an existence-theorem) is not possible, ‘When we postulate a first integral which, when resolved with regard to y/ may be expressed as ‘we have35) ‘HILBERT'S THEOREM 49 If, then, we regard p as a new variable, to be determined as a function of and y, we can take it as satisfying the partial differential equation OF (p32 42) 28 OF _, oF apt \P ay * be) ~ ay ~ deep ~? dyap” where F denotes f(a, y, p), a known function of its arguments. As this partial equation of the first order, when p is the dependent variable and a and y are the independent variables, is of the Lagrange linear type—linear, that is to say, in the derivatives of p—we may be able to obtain an integral. Let some such integral", necessarily involving p in its expres- sion and also involving an arbitrary constant, be obtained ; it will arise as an integral of the subsidiary equations dp T (oF _ or oF = 4) a (55 ~ aap ay?) a and will be satisfactory, provided it contains the variable p. We denote this integral, resolved so as to express p in terms of # and y, by P=P(@y). Now take ¥=P=P(ey) We have Dp, Pay. dat Pay HV the function F becomes f(e, y, y’); and the partial differential equation is transformed back into that is, into the characteristic equation. Hence the relation y=P@y) provides a first integrated equivalent of the characteristic equation, when p satisfies the partial differential equation of the first order, The transformation is used by Hilbert+, in an investigation with the different aim of substituting, for the original integral Z, an ultimately equiva- Tent integral, chosen s0 as to become independent of the path between the lower limit and the upper limit. Hilbert’s investigation is based upon the requirement that an integral, allied to f [/(@, y y’)de, and taken in the form Jisev 2+ -p) A) de, * me primitive ofthe partial equation i specified type, will sufie, + Gate. Nackr, (1900), p. 291. 10 required for the purpose ; any integral, of the50 HILBERY'S THEOREM CONCERNING fou. 1 where A is some function of a, y, p, shall yield a maximum or minimum as regards the variations of p, and then shall be independent of the path, p being some function of « and y. For the first property, we must have Lew-nZ-4=0, which can be satisfied by taking as this relation leaves the result provided wi is not zero. With this value of A, the integral becomes [v 5+ erg) If this is to be independent of the path, and therefore is to depend only upon the variables at the limits, the subject of integration must be the exact differential coefficient of some function UJ. Then wy Wy, #, oy op’ oe z (r- if) - 2a. that is, as p involves @ and y, ‘Of , Of ep) _ {ap ar OF , Bop) _ af (+o) Bain? nla ap enh -aalp * aes ay ~ beap~ ? Spey ~ op \ae Y_ oF o_o, w (i+) which is the characteristic equation. For the use of the analytical property, reference may be made to Hilbert’s memoir. and therefore ep ba" or 0, Bx. 1, When the integral is foo -+y¥ds, the intogral for the catonoid (§30, Ex. 1), wo have Fayatyh For this function, Fate, Rae eateyt, ayaa pt;35) ‘THE CHARACTERISTIC EQUATION 51 OF the subsidiary equations ma de_dy_ ydp T itp? an integral ia _ yaa(ltg, Hence a first integral ofthe characteristic equation is yraceynh, Ex. 2, When the integral is Jorssron +0 +) y Wa, the integral for a geodesic ‘on & paraboloid (§34, Ex. 8), we have Pa(iestseyp+ty) py After substitution and reduction, the partial equation for p is C+P)(p2—y) eo amere, From the subsidiary equations de_dy__ lestiy? Top epg er we find ‘0 that and an integral is spaynak. “Hence a first integral of the characteristic equation is ay -yaa{ltattaeyy + +yy yhCHAPTER II. INTEGRALS OF THE FIRST ORDER; GENERAL WEAK VARIATIONS; THE METHOD OF WEIERSTRASS. ‘The chapter is devoted to the discussion of maxima and minima of single integrals, involving one dependent variable and its first derivative, ‘The method adopted is due to ‘Weierstrass. It enables the application of variations, still limited to the weak type, to be made simultaneously to the independent variable and the dependent variable, instead of limiting the discussion to the effect of variations upon the dependent variable alone; ‘and it thus allows tho consideration of general weak variations, ‘The most direct account (as yot: published) of the method seems to be that given in Harris Hancock's Caleulus of Variations (the Weierstrassian. Theory). Weak variations in general. 36. We now proceed to the consideration of weak variations, which are not restricted to the preceding type and have the general property of admitting simultaneous weak variations of both the dependent variable and the independent variable, each without regard to the other. The integrals, in the discussion that ensues immediately, will still be restricted to the first order, that is, they will involve derivatives of the first order and no deriva- tives of order higher than the first. ‘The results to be obtained must include all the preceding results connected with the special weak variations, More- over, they will have the added advantage of proving useful for the considera- tion of strong variations. ‘Accordingly, the variables © and y will now be regarded as subject to small variations, whereby the variable # is changed to #+«u and the variable y to y+«v. As before, « is a small arbitrary constant quantity, so small that any positive integral power is negligible in com- parison with every preceding positive integral power. Also, u and v are regular functions throughout the range; they are quite independent of one another; and they are completely at our arbitrary choice, subject (for the immediate purpose) solely to the condition of being regular. But in order to secure this general variation of # and y, we make both of them functions of ‘anew independent variable which can be chosen at will. It might, initially, seem natural to choose, as this new independent variable, a magnitude intrinsically connected with the characteristic curve to be obtained, such as the length of the are of the curve measured from any assumable point, Yet any such intrinsic magnitude may itself be the subject of discussion; for instance, the problem may be limited by an are of given length, or we may be required to find a geodesic arc on some surface, ‘Thus there is an87] METHOD OF WEIERSTRASS 53 advantage in assuming an initial independent variable, unspecified by any particular intrinsie relation; when once the analytical results have been obtained, they may be subsequently modified or simplified by any speci- fication that can be appropriate and convenient. We therefore take « and y to be functions of an independent variable denoted by ¢; and then u and v will be arbitrary independent regular functions of &. But it will be assumed (60 far, the assumption is a partial limitation on arbitrary functions intro- duced) that the variable ¢ increases continually throughout any range of integration, Modification of the form of the integral. 87. Derivatives will be denoted by a single subscript number, so long as, the derived quantity is a function of only a single argument; thus de, @e_, dy Gn Ga Be and s0 on; modifications will be required for partial derivatives, when the derived quantity involves more than one argument. ‘Thus Yer and so on. Making this change of variable in the integral, we have [revsrde= [ilo 98) adem [PG ye y) de ‘Thus our integral becomes [Fe ne.wae, where the range of the real variable ¢ is from 4 with continuous increase to t; and where ¢, and f may be variable quantities, if occasion demands. Moreover, owing to its source, the function F is such that Foxe w=af(s92) aasf@ yy) We at once have and therefore54 MODIFIED SUBJECT OF INTEGRATION (cu. a permanent identity satisfied by every function F which arises in the manner indicated. ‘What is desired is the determination of # and y as functions of ¢ so that the integral, extended over an assigned range, shall acquire a maximum or a minimum for the weak variations indicated, of every type possible within the limitation, 38. Manifestly, the change to a variable t is not unique; thus, if » variable 7’ were chosen, we should have da ds F (2. Go Gh) a= fe wae =F(an%, ¥ at. In particular, let t= 7, where sis taken to be a constant for our immediate purpose; then F(a, y, we, wy) = wF (a, y @, %)- ‘This relation holds for all values of 4. ‘Two important inferences can at once be made. Firstly, let w= +e, where ¢ is a small arbitrary quantity ; then Fey area, n+ yar Fe, % ay) Expanding the left-hand side in powers of «, and equating the coefficients of the first powers of ¢ on the two sides, we have ace in ne oF =F, the permanent identity satisfied io F; and from the other powers, relations such as or oF || ar anges t dh gay, tM ys immediate consequences of the sdntiy being the customary property of an infinitesimal transformation in a continuous group. Secondly, let «=—1; then Fy a -y)=-F OY aH) It is an immediate corollary that, if E-Pen ew SAV yews then Dy, —%, —H) =P, Y, BH), Y@ ym —W=V@ % mH)40) CHARACTER OF THE WEAK VARIATIONS 55 General variation of the integral. 39, We now consider general weak variations of the integral Z, where r=f'F dt ip (@, yt» ys) at, the function F being subject to the permanent identity In the first place, we shall assume (though, later, the assumption will be modified) that all the arguments and all the functions of the arguments, which occur, are continuous. We take a variation by which @ and y are changed into ©+«u and y+ ev respectively, x, u, » being limited in the modes already stated. Such variation changes a curve into an immediately neighbouring curve in such a way that, to each point of the original curve, there corresponds one (and only one) point of the modified curve: and that, as x is a small constant quantity at our arbitrary choice, the distance between two such corre- sponding points can be made less than any previously assigned small quantity. Initially, this is secured by the finiteness of w and o for all values of @ and y within the range. But the wider restriction has been imposed that w and v are to be not merely finite, but aleo are to be regular functions. This restriction carries the property that, in these small variations, the inclination between the tangents to the original curve and the modified curve at the corresponding points also is small; for the trigonometrical tangent of this inclination is «(Gt — Yith) at yt te Gy tym)” a small quantity of the same order as x. ‘We shall further assume that u and v are independent of x, so that the magnitude of any derivative of u or of v remains unaffected by the magnitude of «. Thus, for example, we shall not consider, at this stage, the possibility of a variation such as u=asin (51), t Le sn ace-rens (5), which, when n differs from zero and is less than unity, is small but is not of the same order of small quantities as xu; it is Gnite if n=1; and it is large when n is greater than unity. 40. Within these limitations, we take the most general small variation of the integral. Let the upper limit be varied to t,+ «7, and the lower to for then56 FULL VARIATION OF THE INTEGRAL [ow a t,+4T,, where 7, and 7, are finite quantities independent of one another, though u and » may possibly be subject to limitations at %, or at f, or at both t and f,, for such variations, Denoting the modified integral by I, we have I- ie Pet eu, y tHe, a+ ath, nom) dt [" FG, Hayat. tens 4 ‘Then, for a maximum or for a minimum, the magnitude T—T is to have a persistent sign, whether « be positive or negative, and whatever functional values be assigned to the functions w and v, independently of one another. The right-hand side is te, fete -{fr +f- fe Pe ten yee abe, ten) dt Arar - [Fe war wde Now, as in § 32, tet, i Fo + au, y +40, 2, + 12, Yo x0) db= eT PO + Ro, where F° denotes the value of F(#, y, #, ys) when t=t, and Ro! denotes an aggregate of terms of the second and higher orders in x, Similarly [Or Pe+e, yt e, a +h, Ye) d= KT PO + RY, 5 where Fe denotes the value of F(a, y, a, y,) when t=%, and R® denotes another aggregate of terms of the second and higher orders in «. Also a i {P(e Wn, y+ 00, y+ 1th, e+ 4%) — F(x, yr, yy} dt rh OF ok oF oF ne fi(u Bat ay tba, ty, where R denotes an aggregate of terms of the second and higher orders in «, arising out of the expansion of F (w+ eu, y-+40, +h, +40) in powers of. We thus have 7) d+ B, =e {U4 2,F9 — TF} 4K, * (xP ee a rf 3a tay tae, dy, K=R0-R+R, Here U, F®, F%, 7, 7, are independent of «; and the quantity K is the complete aggregate of terms of the second and higher orders in «. where at, and When regard is paid to the smallness of the arbitrary constant x, 0 that K becomes negligible in comparison with non-vanishing terms of the first42] FIRST VARIATION aT order in x, the value of I~ I is governed * by the quantity {0+ 2,F° — 7,P0} when this quantity does not vanish. In the latter event, the value of 1—I can be made to change its sign by a change in the sign of x: that is, IZ does not have a persistent sign, the prime condition for either a maximum or a minimum, It follows that, in order to secure our aim, the quantity U+2,Fo—7,F% must vanish. It must vanish for all admissible variations, separately for cach however arbitrarily chosen, and collectively in all combinations. For the present condition, the latter requirement is fulfilled if the former require- ment is fulfilled, because of the linear character of the expression in the quantities u, », Z%, 7%; and therefore it is sufficient to require that the quantity shall vanish for every admissible variation separately. Two critical equations. 41. The expression, denoted by U, can be modified through integration by parts, such integration being possible because the subject of integration is continuous through the range. We have hh OF oF} 4 od fom “ge, = [» zl -fiu wales a, rr oFr)* fie Poe atm fo SE] — fo SO) as and therefore re (ar od & (ard /@l OF aFyt w=fi[+fae~ a Ge)} +* opt Gh] + [ose eae ‘The whole expression for T now consists (i), of an integral, which involves the arbitrary variations u and v but not their derivatives: (ii), of terms at the upper limit: (iii) of terms at the lower limit. 42, We write Ha a x-ad | ar_d rl” vy a | In the first place, take variations through the range of the curve such as to vanish at each extremity; thus we might have u=(h-H(t-) dO, r=(h-DE-t) YO, * See foot-note on p. 17.58 TWO CRITICAL EQUATIONS (cn. where ¢ and y are regular through the range; and, simultaneously with variations through the range, keep the upper limit fixed and the lower limit fixed. In this case, we have zero, both when #=¢, and when t=t; also 7-0 and 7,=0, Hence the requirement is that fs [olx+er) amo. “ First, take 0=0 throughout; then we must have [" uXdt=0 for all values of u If X, which is independent of w and of v, does not vanish everywhere along the range, then [Puxae can be made positive, by taking u positive when X is positive, and « negative when X is negative; and it can be made negative, by taking « negative when X is positive, and u positive when X is negative, choices that are admissible. ‘These possibilities must be ex- cluded ; and the exclusion is possible, only if X=0. A similar argument leads to the necessary result that the equation Y=0 must be satisfied. Hence, from the foregoing requirement, we have X=0, ¥=0, as necessary equations; and when these are satisfied, the integral in the modified expression for U' vanishes, whatever admissible variations u and v are adopted. It will appear that, if the limits of the integral are not given as definitely fixed by the data of the problem initially, they are definitely determined by other assigned conditions. When the limits are definitely fixed, we have T,=0 and T,=0. In the alternative when they have to be definitely deter- mined, then, as soon as this determination is achieved, we have 7, =0 and 1,=0. Accordingly, in any event, we can take 7,=0 and 7,=0; and so no condition accrues from the terms involving T, and 7,. 43, It therefore remains to secure that the magnitude shall vanish at the upper limit and at the lower limit, independently of one another. ‘Various inferences are drawn, according to the various terminal data that are given, If the range of the integral terminates, not merely in a fixed value f,44) ‘TERMINAL CONDITIONS 59 but in a fixed point giving values of « and y that are not subject to variation at that point, then v=0 and o=0 at that extremity; and the terms, thus vanishing, provide no condition that is new. If the range of the integral terminates in a point that is required to lie upon a given curve represented by some equation such as 9% y)=0, then the varied extremity #+.u, y+ev also lies on that curve; thus at the extremity, we have from the geometry 29 oy aP dg _aF ag utes and therefore we have =0, thus providing a condition that is new. But it is only a terminal condition ; and therefore it can be expected to prove of use for the determination of arbitrary constants which have arisen or may arise. Similarly, the same type of equation is satisfied at the other extremity. Note. This relation to be satisfied at a mobile extremity (not determined by initially assigned data which require u=0 and v=0), is in actual agreement, though superficially not in formal agreo- ment, with the relation obtained (§ 34) for the special variation. To harmonise the two forms, we note that ar, oF Fong thi Faaflayy) s0 that ‘ = i : _ u and therefore ae : ar, a agnale-ne, that is, m x aa TY ay" ‘The form of the relation, for the present type of weak vs -y L) 82%. (-1)§-9 8 ‘hich isthe form forthe other type. ‘The two equations are equivalent to one characteristic equation. 44 It follows that, in order to make the terms in I~ J which involve the first power of « (the first variation of I) vanish, we have two kinds of conditions to be satisfied, The one set of conditions belongs to the limits we must have tion, thus changes to FoF aa * ay,60 THE TWO CHARACTERISTIC EQUATIONS [om at each limit, ‘The other set of conditions persists throughout the range of the integral; the two equations X=0, ¥=0, must be satisfied everywhere along the range. But when the identity satisfied by F is taken into account, these two ‘equations are equivalent to a single equation, in virtue of which both are satisfied. This claim can be established as follows, We have ar, oF a ea, +h oy Differentiating this identity with respect to «, and to y, separately, we find oF) OF ar oF a a =0, h dat * Maedy, 9 esd, +H dy ler_ ol #F 1 eF Ye dae mY, Ody, a Oy," when we denote the common value by P, a critical quantity that recurs ‘through the whole investigation. ‘We have Hence a at \Sn, oF oF Or | oF ) = Gnea, * ayou,Y + Be3 * aay, oF, OF = geaa,” + ayant rye wey) P. ‘The same identity, when differentiated partially with respect to «, gives the relation or_ OF, OF We dade, * * andy, Hence ar i (is) or @F anf ge Bet onan Ph =- 98, where ar @F OF Fea, CWP Similarly _oF_d@ “ay de @)-s6 Is therefore follows that the two equations X=Oand Y=0 are satisfied in virtue of the single equation @=0; and this is the characteristic equation.46] EQUIVALENT TO A SINGLE EQUATION él 45. Sometimes it is convenient to take the equation @ = 0, sometimes the equation X=0 or the equation Y=0; and sometimes it is convenient to retain both equations. Moreover, the two relations X=-y@, Y=a@, have been deduced by means of the permanent identity affecting the form of the function, and without reference to the subsequent requirement that the first variation of J shall vanish. ‘The integral in the former expression for U is [raxsorya a = [Ror yay at If then we write Wa fev yu) Gat, and if we assume that the terms at the limits vanish—as, from the foregoing argument, they are bound to vanish—we have 1-1=4W+K,, where K, is the aggregate of terms of the second and higher orders in the small quantity «. Covariantive character of P and @. 46. We at once have the important: properties that, whatever independent variable be chosen, P is a relative covariant, and @ is an absolute covariant, Let t and 7’ be two distinct independent variables, so that (as in § 88) de dy . F (2 yap gp) a2 =F, yo ya; and write : dy Feng G=% Fayawe=F de dy_y @e_y ay nk, Bar, Fa-k, Hey, as ar-* so that is a variable quantity in general. Then Xenn, Yreny, Fao. oF oF oF or aX, am,’ BY, dy." CF ek ax,0Y," ~ Seay? Also62 CHARACTERISTIC EQUATION AN ABSOLUTE COVARIANT — [H. IT and therefore er ( nee RAY, axeY, wi ay deity, shewing that P is a relative covariant. Again, we have Xi= w(t nm), Yo= a (myet myn and therefore XAY,~ X= 8 (Gys— ay). Also or _@F OF _ oF @ad¥,” Gandy,’ dydX, dyoe,’ 80 that or OF _XY.-XY, OF _¢ dyoX,~ ax0Y, XY, 8X0%, shewing that @ is an absolute covariant. In what follows, it will appear that the permanence of the sign of P throughout a range is of fundamental importance as a test. The independent variable is assumed to increase continually throughout the range; hence, at any place, dé and a have the same sign, and so pis positive. As P=—P, the signs of P and P are the same; a test, as regards permanence of sign in P, is unaffected by the change. Further, as the characteristic equation is thus essentially unaltered by change of the independent variable, we may change that variable after the equation has been constructed, should advantage accrue from any particular choice. Later, it will be found convenient (for some purposes) to choose the length of the are measured from some fixed point as the variable, especially if intrinsic properties of the characteristic curve, or of variations from that curve, are under consideration; the change can be made after the characteristic equation has been obtained, ‘Two quantities continuous in passage through a discontinuity. 7. One important theorem can be established for the subject of integra- tion in the case of a maximum or a minimum, when some of the assumptions already made as to continuity of arguments are not satisfied, viz. Should the characteristic curve suddenly change its direction at a free place, that is, a place unkampered by conditions as to fizity, the quantities ¢ and. 7 suffer no discontinuity in value,47] ‘TWO DERIVATIVES CONTINUOUS THROUGH A DISCONTINUITY 63 As there is a sudden change of direction, y or a or both y, and = must suffer a sudden discontinuity in value there. Let such a place be 7’; and sup- pose that the curve possesses no singu- larities in the immediate vicinity. Take 4 4 point t, near 7 between 7’ and , and another point f near 1’ but between 7 and ,; the direction is taken continuous from f, to 7, taken continuous also from 7° /y to f, and it undergoes a sudden change in passing through 7. te Let the limiting values* of = and ¥ as f, approaches indefinitely near 7 in the range t,77 be denoted by Gf. Gre and let their limiting values as f approaches indefinitely near 7' in the range Tt, be denoted by ) Gs, Take a variation of the characteristic curve such that u=0 and »=0 along ty, and also along t,t,; as the are of the curve itself is continuous between t, and f, we can take u=G-Dt-%)$O, v=b-)E-H¥O, within that range, where $(t) and y(t) are arbitrary regular functions of ¢ which do not vanish at 7. ‘The first variation of the integral I is to vanish; taking account of the fact that u and » vanish (for the present variation) between t, and t,, and between f, and f,, we have the analytical condition (as in § 41, 45) as [fof}or-nnene [Abo onn Now @=0 everywhere along the characteristic curve; thus the integral from to T, and the integral from T' to &, vanish. Also [ose ++ae limo Ga)_ +e)» [xe5* 9h le=—e Gee). -°*G5). ‘The condition therefore becomes ary . G),} =0. {@).-@). +16) and64 DERIVATIVES CONTINUOUS AT FREE DISCONTINUITY [CHL ‘The quantities up and vz are arbitrary ; their values can be chosen, each inde- pendently of the other, and distinet from zero, Hence the condition can be satisfied for all such quantities, only if QR) _ (oF @) a @)-@), @-¢ ‘Thus the proposition is established. But the property that £— oF and z are continuous in value, even if 2, and 4h are not continuous in vale, in passing through any place, can be inferred solely if the place be free. When the place is fixed by external requirements, both wp and vp are zero: the condition is then satisfied without leaving a residuary relation. Ex, 1. Lot it be required to find the shortest curve passing through three fixed points A, 6, B, not in a straight line. ¢ ‘The characteristic eurve is always a straight line; so the shortest curve will consist of two sides of the triangle ABC. If the two sides are AC and CB, the limiting conditions arise from wel, But 1c=0 and vo=0 in the present instance; the conditions at C are satisfied without limitation, eel elena Ez. 2, In the case of a curve, rotated about the axis of and providing a minimum area for the surface of revolution (§ 30), we avo Pany.y(ltyniay (ettyat ‘Therefore, at a free point, vey aa(ateath va : (etsy (ateyoh suffer no discontinuity in value, in passing through the point. In particular, the curve is yity) 4-0; that is, if there is a sudden discontinuity in direction, © is unaltered when the point is passed. ‘An exceptional case was noted, when the two limiting points were on opposite sides of the axis; the catenary could not be drawn. But the possibility of the catenary has assumed a finite non-zero value for @, We must then consider C=0: that is, ee ary tent Before reaching the axis of », we must have z,=0 or =a; the curve is the straight line perpendicular to the axis of «, When it reaches the axis, the equation is satisfied by ‘y=0 and 2, not zero; the curve now lies along the axis. At the point of chango, which is free from external conditions, there is a discontinuity of direction while the (zero) value of the constant is unchanged. ‘Similarly, in passing to the other extremity taken, in the present case, to lie on the is, the remainder of the curve is the ordinate of that extremity.49] DISCONTINUITIES LIMITED IN NUMBER 65 The number of discontinuities of direction in a finite range is limited. 48. One important inference can be made, ‘The result shews that a place may exist within the range of integration at which there can be discontinuity of direction and that, at such a place, and # are continuous, if the place is not absolutely fixed. In order to establish the result, it is necessary that the place should be isolated as a place of discontinuity in direction for it has been assumed that, on either side of the place as a position on a continuous curve, fall small variations xu and sv are admissible, whether the place be absolutely fixed or not. But there is nothing to restrict the number of such places in any range to a single unit; the general necessity for any such place is isolation. Accord- ingly, there may exist a finite number of places within a finite range of inte- gration at which discontinuity of direction may occur; each such place is isolated. The number of such places within a finite range cannot be infinite ; if it were unlimited, there would be concentration of a countless number of discontinuities at one or more of such places, and then a place of concentra tion would not admit solely continuous small variations xu and ev at each side along the curve. The primitive of the characteristic equation. 49. The characteristic equation @=0 is, effectively, the former (§ 17) characteristic equation Y= 0; the reduction can be made as follows. We have Fenn w=af(ayt)=as@ yy); hence Oro oF _ of dy,” By’ Aady, = Dady” OF pn F_ of ae, "Fm by ~F—¥ ay» er _o thus Also ar or pi, oy?” so that os o Ler en66 PRIMITIVE OF CHARACTERISTIC EQUATION [om 1 Again, so-4Q)-25: hence, when « is made the independent variable, dy 1d), _ ayes, pak Sy hes and therefore » Wy (ay —ay) Pau Zh. Consequently oF oF Buay, 7 ve mY) P= eady, -§-a(@): establishing the exact equivalence of the old characteristic equation and of the new characteristic equation, ‘The old characteristic equation was a differential equation of the second order in y, with « as the independent variable, Its primitive contained two arbitrary constants, which were to be determined by two assigned data such as an initial value of y and an initial value of y/ for an initial value of Initially there are two equations, in the new investigation, for the deter- mination of « and y, viz. X=0, Y=0, in each of which # and y, ocour raised only to the first power. When &, is eliminated, we have some relation such as P= P(%, Gs Ys) ‘Two derivatives of this, and subsequent elimination of # and 2, lead to an equation of the fourth order in y alone; its primitive involves four arbitrary constants, When this value of y is substituted in yp=® and in y= #, and 2 is eliminated, we have expressed as a function of ¢, the expression involving those four arbitrary constants. But the four constants are not essential for the equation of the character- istic curve in terms of « and y alone, Thus, as ¢ never occurs except in a differential coefficient, the quantity ¢ must occur in the primitive in a form +—A, where A is an arbitrary constant; and A will disappear with ¢, in the elimination of ¢ when the equation between « and y is obtained. Again, as initial conditions, we may have initial values of «, «;, y, ys given for an initial and so these two constants for y, and @ merge into a value of f, But y/ single constant for y'; and the two initial constants for 2 and y become, for the equation in # and y, an initial value of y for an initial value of «. ‘Thus when we pass to the final relation between « and y, there will survive only two arbitrary independent constants, although in the initial form of the50] SECOND VARIATION 6r primitive of X=0 and ¥=0, giving the most general values of and y as functions of t, there are four arbitrary constants, Moreover, because X=—y,@, ¥=2,@, the primitive either of X=0 or =0 could lead to the primitive of @=0; and a first integral of either would lead to a first integral of @=0. It thus appears that, in practice, the integration of either X= 0 or ¥=0 would determine the curve: but, for the derivation of farther results, it is convenient to retain both the equations. Finally, unless there is express warning to the contrary, it is assumed (as stated in § 12) that the functions 2(¢) and y(t) are analytic functions of t, so that they are continuous in variable and in parameter, are differentiable, and are one-valued along a finite range or in each one of a finite number of portions of that range, The * second’ variation of the original integral, 50. Now that the terms of the first order in x, which occur in the ex- pression for I—J, have been made to vanish, we must consider terms of the second order in «, 60 as to determine the necessary and sufficient conditions which shall secure a permanent sign for the aggregate of those terms for all admissible variations, We may write 1-I=n0+4e [ods k, where, as before, : Ue *( ar ar, ar of) (Get Gy Maa, +5) as here K, denotes the aggregate of terms in I~ I, which are of the third and higher orders in «; and OF aug OF 4 ys OF On egeat Saag, +E By or oF oF oF + 2a 5a, 2a aay, + Bo Sat iyi, or OF or tu gg tue te As the first variation has been made to vanish, when the characteristic curve is adopted and the boundary conditions are ‘used, the terms of the second order now govern the sign of I~; they are often called the ‘second’ varia- tion, We proceed to modify the expression for { “@dt, on the assumption ‘ that the first variation vanishes; so we substitute the values of # and y in the cooficients 2, .., sf, which accordingly may now be considered to be functions of t alone,68 RELATIONS AMONG [on Farther, instead of using u and v separately, we find it convenient to introduce a quantity w, defined as the combination WH HV — Yih ‘As the point (2, y) is displaced to (e+ xu, y+) by the small variation, the quantity « («+ ys!) # (av —yru) is the distance of the displaced point from the tangent to the characteristic curve at the undisplaced point. Because w thus measures that distance, we shall call ew the deviation due to the varia- tion, The vanishing of w means that the point is merely displaced along the curve; when this deviation vanishes everywhere, the ends being fixed, then there is no new curve obtained from the variation: in such an event, we should expect the second variation (and every other variation) to vanish. Relations among derivatives. 61, We have already obtained the relations 1eF_ 1 OF _10F das Hy Omoy, a ‘Thus the first line in the preceding expression for @ can be written Plan-yny From the characteristic equation & oF aay. * 7 aan ‘and we introduce three quantities Z, M, N, by the definitions GF b+ yyP ane, ane Bean, we aoe +a yP= +ymeP; =M-ayP seox- imP| Or ayoys ‘Thus the second line in the expression for @ can be written 2 {Lan tly + M (th 02 + 0th) + Novy} + 2P (ev, — gate) (mv — yard). Next, we write temporarily =N+a,2,P oF. ap ab Gan At wP+S, ar am dady 7B teP + a? oF52] DERIVATIVES OF INTEGRAND 69 But, by partial differentiation of the identity with regard to #, we have L+yM; whence, differentiating completely with regard to ¢, at aE eee ate yeh ey tne gM, or, on substituting for the second derivatives of F, aA +yB=0. Differentiating the same identity partially with regard to y, and proceeding in the same manner, we find aB+y0=0. A Tay ae introducing a new symbol @ to denote the common value* of the three fractions ; and so we now take Hence a : sp, Gen w+ yer er am Feag 7 ANG = ay P 4ST, oF an aye G+ asP +7 Thus all the second derivatives of F, ten in number, are expressed in terms of five quantities P, L, M, W, @. Normal form of the second variation, 52, When these expressions for the derivatives of F are substituted in @, and the terms are suitably collected, we find B= Pein 1m) + (ee — yu) + G (a0 yuu + S(t + 2Muo + No) = Pui + Gut +S he + 2Muv + Nv’). Denoting by H, a quantity at our disposal and not involving w, we have @= Pus — 2H, + (G— Hy) ut + 4 (Lat + 2Mun + No+ Hw’), Now let H be so chosen, that Pu! —2Hww, +(G—H,) w* is a perfect square qua fanetion of w and w,; thus H*=P(G-H,). * As x, and y, do not simaltanconsly vaniah, this common value i70 NORMAL FORM OF (ca. 0 In this relation, let H=P, 2 where 2 is a new variable in place of H. After reduction, the equation for 2 becomes P2,+ Py2,— G2=0, which, as P and G do not involve z, is a linear equation in z of the second order. To this equation we shall return later; meanwhile, for such a value of z, Pw? —2Hww, +(G— H,)w' = P (= -# wy, and therefore =P? (u~ Zu) +5 (Lae + 20w + e+ PS a). Thus [ Odt= “ : ‘ P(m—Zu) dts [Doe + 2ahuns Nor PE wl”, 4 the last term being taken at the limits of the integral, Note. Tt might appear as if, when taking solely the combination waav-yu in the construction of the modified expression for @, a limited part of the small variation were taken into account; and that a combination T=muty also ought to be considered. The quantity w measures the normal deviation caused by the small variations eu and xv; the quantity 7’ measures the tan- gential sliding due to the same source. Though 7 will cause no change in the shape of the curve and therefore no apparent variation, it is not at first sight analytically clear that 7 should disappear from the expressions. ‘That T does so disappear is seen from the fact that @ has been expressed in the form P (ein — yin + 20 yu) +E (a0 + F (Lat + 2Btun + Ne), where P, G, L, M, N do not involve u and v; and this expression becomes Puss Gus ¥ (Le + 2Mtuo + Ne, so that the integral [““@dt becomes Te “ \ [ieors owyars [e+ 2atue ave]! : “ 5 Now ulerty)=—ywta?, varty)=nwtyT,53] THE SECOND VARIATION n so that the quantity 7’ could only occur at the limits, if at all. At each of the limits, w and v (by given data, or after limits have been fixed from given con- ditions) vanish, so that wand 7' vanish at each of the limits. ‘Thus P disappears altogether from the expression of the second variation when the combination, denoted by w, is used, 53. The preceding expression for ®, which is of fundamental importance for the second variation of the integral J, can be established also as follows by somewhat different analysis, Let, if possible, quantities a, 8, y be determined so that the magnitude PO-P Feat + 2Buv +0), manifestly homogeneous and quadratic in w, », 1, %, is a perfect square, which (after the values oe xa 5) must be {P (a.m; — yity) + wo — dup. The necessary conditions are or Mn=LtyyP—a MeaP aa —Pa ha, = M—2,y,P 8 aor , —r™u=P2™ _ pg). -ny=M-ymP 8 en Panay ~ FA: p= N +0,0,P —y, > 2k _p,, From the second and third equations in the first set, we have dees — HY = (Ye — ays) P 5 and therefore we may take A=mP typ, womP+ap, so that P(e, — yu) + pv du = Puy, + pu. ‘The first set of equations now reduces to a=L-yip, B=M+ amp, y=N-a2p, 0 that and + 28uv + oe? = Lut + 2Muv + Not + pur, The second set of equations now gives oan) =82 — sep—28, egos ce au on (5 1) Baby TPT (Go) ee = ep a **(G-n) ae — #P—a-72 ASSUMPTIONS DURING CONSTRUCTION OF [on Hence there exists a quantity @, such that a result in exact accordance with the expressions for aE, Ee a. deduced from the identity satisfied by the form of F when account is taken of the characteristic equation. As regards this equation for p, let A, p=-Phi then Pe, +P,s,—G=0, the same equation as before. Gathering the results together, we obtain the expression for ®, given by 9-4 (Lu + 2htu + Ne + PB w') =P (mF 54, Before discussing the significance of the form obtained for the second variation, it is worth while to note assumptions which have been made, overtly or tacitly. ‘The variables « and y, as functions of t determined by the characteristic equation, and the quantities u and », vary continuously. The variables «, and 1, vary continuously ; but places may occur on the characteristic curve, where ‘the ratio y/i suffers a sudden discontinuity in value. Between such a place and each limit of the integral, and between every two consecutive such places (if there be more than one of them), these assumptions hold and the analysis is valid. ‘The number of such places (if any) within a finite range of the integral must remain a finite integer. The original integral involves no derivative higher than the first, and therefore is formally unaffected by possible discontinuities in «and y,. These quantities occur in the characteristic equation @=0, as, of course, also in X =0 and Y=0; and their continuity is assumed, as also is a tacit assump- tion that 1, is continuous. But the quantities u, and v, do not oceur in the analysis, even by implication, although 1 and », are assumed continuous ; yet such quantities arise, if the curvature of the varied curve has to be considered, ‘Thus there might be a continuous varied curve, which is continuous also in direction, but which might have violent discontinuities of curvature at any number of successive points; for instance, any portion of the characteristic curve might be divided into very small ares, and semicircles on these ares in succession might be constructed on alternately different sides of the curve. Such possibilities must not be ignored; they belong, however, solely to the variations imposed upon the characteristic curve and not to the curve itself In order that the varied curve may be continuous, u and v must be con- tinuous. But for such a curve, % and 9, are not necessarily continuous, merely53) NORMAL FORM 7B because w and v are continuous; thus the varied curve might be sharply and rapidly serrated, For such variations, the preceding analysis is not valid; yet such variations must not be ignored: again, they belong not to the character- istic curve but to the imposed variations. For variations which thus (or in other respects) deviate from continuity, a different analytical calculation must be effected. ‘The more obvious occasions, where such deviations occur, arise at fixed points, where u and v are bound to vanish: terms Tat+ 2Muv+ Not + P2 wt vanish at such places. Discussion of the normal form of the second variation. 55, We now return to the expression for the second variation, given by foa= |p (w.- Su) are [soe +2atio + Ne PS wr), 2 z taken between limits. If none of the preceding types of discontinuities occur in the whole range of integration, the integral extends from t to t. In this event, we may assume (after the discussion of the first variation of Z) that the limits are fixed, and that therefore u, v, w vanish at each limit. The terms at the Vinmits then vanish; and a JP (@-24) at is the quantity measuring—except as to the positive constant factor }e— the second variation, If some of the preceding types of discontinuities do occur and are subject to the prescribed limitations, we suppose the range to be made up of a finite number of sub-ranges, each sub-range extending from one such place to the immediately consecutive place. At each place, the limit terms vanish, and the whole second variation is the sum of the second variations [P(w- Su) ae for the successive sub-ranges. In the former case, the tests will apply to the whole range; in the latter case, they will apply to each sub-range separately and thus, cumulatively, to the whole range. We thus have to deal with the magnitude eas a=[P(w,—2w) de,™ LEGENDRE TEST [om over the whole range, or over the aggregate of the sub-ranges taken separately. With the assumptions made, the second variation of Fis $A; and therefore its sign is determined by the sign of A, while it can vanish only if A oan vanish, The sign of A, which is to be permanent for all variations w that are admissible, depends upon the quantity P and upon the quantity (w~ fw) In the latter, « is an integral of a definite linear differential equation, while w is (within a broad range) at our arbitrary disposal ; the quantity therefore is always positive, unless it can be made zero by a choice of w. If then ¢ is such that, consistently with the conditions, we ean choose w= me, where m is a constant, and the choice is valid over the whole range or over part of the range, then the quantity is zero for the whole range or for that part. In the former case, we should have A=0; in the latter case, by choosing w to be zero throughout all the range except that part, we again should have A=0. The second variation of J then vanishes for that small variation or combination of small variations; we should no longer be in a position to assert a permanence of sign for 1—I as derived from the coefficient of x. It is therefore desirable, where possible, to exclude this chance: and such exclusion will depend upon the character of the quantity z, that is, upon the primitive of the equation Pat Pyx,- Gz=0. The Legendre test, 56. In the first place, we consider the effect upon A of the sign of P. If P be sometimes negative and sometimes positive, then (@ by choosing w zero where P is positive, and different from zero (or from mz) where P is negative, we can make A negative: and (ii) by choosing w zero where P is negative, and different from zero (or from mz) where P is positive, we can make A positive: that is, variations can be chosen at will which make A negative or A positive. This possibility must be excluded; as it depends upon having different signs for P in different stretches of the range, this previous possibility must be excluded, Hence P must not change its sign in the range: consequently P must not pass through the value zero and change its sign in doing 0, nor pass through an infinite value and change its sign in doing eo. A zero value of P at some place, without a change of sign, would not be excluded by this con- sideration ; but owing to the differential equation satisfied by z, itis desirable to exclude even that possibility. Similarly as regards any infinite value of P without change of sign in passing through such a value, quite independently of the assumptions concerning the continuity of the quantities that occur.58) SUBSIDIARY CHARACTERISTIC EQUATION 15 ‘We therefore can summarise the results by the following requirement: In order that the integral I may acquire a mazimum value or «minimum value, the quantity denoted by P must never vanish within the range. Manifestly this is the Legendre test, ‘The subsidiary characteristic equation. 57. We now must deal with the equation satisfied by the quantity z, viz. Pat P.4-G2=0, especially with reference to the foregoing requirements. We shall suppose that the Legendre test is satisfied, so that the quantity P remains finite, continuous, and distinct. from zero, for all values of ¢ within the range; thus P can be regarded as a regular function of ¢ for all such values. Further, suppose that @ does not become infinite within the range, As Bo Gis the common value of the three equal fractions 4. and as @, and y, do not simultaneously vanish—otherwise 4 would vanish—we then have the requirement that A,B, C do not become infinite, a require- ment usually satisfied in consequence of all the preceding assumptions as regards F, «, and y. Thus the ordinary linear differential equation of the second order for z has all its coefficients regular and finite for all the values of ¢ within the range, these cooflicients being known functions after the (presumably) known values of « and y as functions of t have been inserted. For such an equation, the customary existence-theorem* yields the following result: ‘An integral of the equation exists, which is a regular function of ¢, and which is uniquely determinate by the conditions that, for any assigned value of f, the variable + acquires any arbitrarily assigned constant value and its first derivative , acquires any other arbitrarily assigned con- stant value. Moreover, the two arbitrary constants occur linearly, so that 2 is of the form z= cht ey (t), where cand ¢’ are arbitrary constants, while ¢(t) and y(t) are regular functions of t 58. If this method of determining « is adopted, we have to integrate a linear equation of the second order. We already have an equation, not linear, of the second order, @ =0 (or the equivalent equations X=0 and Y=0). We now proceed to prove that, when the primitive of the characteristic equation is known, the primitive of the s-equation can be deduced by direct operations involving no quadrature. + eo my Theory of Diferential Equations, vol. ili, § 200.76 SUBSIDIARY CHARACTERISTIC EQUATION [ou u ‘As the second variation is to be a criterion for the possession of a ‘maximum or a minimum by the characteristic curve, the values of z and y ocourring in the coefficients in that variation are the values satisfying the characteristic equation, that is, satisfying the equations X=0 and Y=0. When « and y are submitted to small variations eu and ev in general, let X become X, Y become ¥, and @ become EH. Then, as X=-y,8, ¥=2, we have —X=(yteyE, Y=(2,+«u)E, and therefore =(K-X)=y,(E-@)+mB, ¥-Y=0,E-@)+ cub, so that (@.ti— yn) (B - @)=u (K— X) +4 (¥— ¥). oF d (aF Xo Eo ai Gm)’ and therefore OF oF. cn oF oF xan + aay * “\guan * asap) d (aF oF oF oF @F WS eu th ee SE en rua the unexpressed terms being of the second and higher orders in x. When these higher terms are neglected, because of their relative unimportance owing to the smallness of x, we have oF oF er or X-X=e(u et seat Gabe t “aay or oF or ~ 65 (“aa eal ayoe, * ast "anan) Now Then oF, OF Mana dedy, so that AP (wy, ayv+ you), OF yy OF 4 OF Fo uLh+ eM -yPu. uo BE ane Also oo 4 Bi tng mal toMy? (use 90); and therefore oer dL oF dM ad xa Kae uw) 40 (SF - Gi tan ) +n un} a ney {- Gus Gah, by the results obtained in § 51.59] DEDUCED FROM ORIGINAL CHARACTERISTIC 7 Proceeding in a similar manner, we find ¥-Yonn, {ou—$ (Pu : Hence (1-H) BB) = e(an—nu) {60 -F Pop}, and therefore B-@=« {Gw— p Pa) Now the quantitios © and y satisfy the equation @=0, that is, also the equations X=0 and ¥=0; and they involve arbitrary constants which are completely independent of the characteristic equation. Let small arbitrary Parametric variations be imposed upon these arbitrary constants, and let. the consequently varied values of # and y be +E and y+ un respectively ; also, let: . Saain— nk, s0 that { is the consequent value of w. Just as a and y, when substituted, make @=0, s0 24+x£ and y + en, when substituted, make E=0. We there. fore have a4 (Ps) =0, where xf and x7 arise as the small variations of « and y, when small para- metric variations are imposed upon the arbitrary constants in the values of and y that constitute the primitive of the characteristic equation, The equation for { and the equation for z are identical; and therefore we can take The quantity ¢ is known, by direct operations upon # and y through these small parametric variations of the most general kind imposed upon the primitive of the characteristic equation ; thus z, that is, the integral of the. equation Pa,+ Py —Ge=0, can be regarded as known when the primitive of the characteristic equation is known, On this account, the linear equation for s (and £) is called the subsidiary characteristic equation. 59. The expression for the second variation now becomes, on omitting the factor de, foat=[P (1-8 a, with the usual limits for the integral.78 ALTERNATIVE CONSTRUCTION OF NORMAL FORM [ou ‘This expression can also be derived as follows. The initial form of U is oF, OF oF T= [ue oe a Let U be the form of u a @ and y are subjected to the small variations w+ xu and y+«v; then U-U=« fe at, in the original form of the function ®, because the varied value of er is oF oF oF er oF ae * (« dat t Gedy * “dade, + am)’ and so for the other terms in U. The second form of U, when account is taken of the conditions at the limits, is (§ 41) w= SL fae~ ah fat = [uxs oxy ar; and therefore” u- T= fuce—X)+0e- Yat we fora —$ (Pap at Hence [ood = fur —$. Pus} at = [Efe Gem—G Pah 24 Pak, fu) ae When we integrate by parts, and in the portion outside the integral impose the condition that u and v, and therefore also w, vanish at each extremity, we have Joaem— [P wt, —t09 Zn — wb pat = [P(w- bu) as, the former expression, _ An integral of the subsidiary characteristic equation® derivable from the primitive of the characteristic equation, is necessary for this expression of the second variation. We proceed to consider this integral in closer detail. seen by an example (Bx. 8, §34), the resulte for « particular ease may be ined more simply by solving the subsidiary equation than by first obtaining the primitive of the characteristic equation. ‘The analysis, which now ia to be given, relates to the general case when the primitive is known and appliss, with unchanged inferences, to any particular ease.60) PRIMITIVE OF SUBSIDIARY EQUATION 79 The primitive of the subsidiary characteristic equation, 60. The primitive of the equations X=0 and ¥=0 leads to the (2, y) equation of the characteristic curve, in which the variable ¢ does not appear. As ¢ does not appear explicitly in X =0 or ¥=0, one of the arbitrary con- stants in the primitive is merely additive to t; it therefore is not an essential constant for the curve, Again, the initial values of a and y, combine into a single constant (say m), as the initial value of y/, that is, of y,/2: that is, these two arbitrary constants coalesce into a single constant for the curve, ‘Thus, in effect, there are two essential arbitrary independent constants, which survive in the primitive whatever form be adopted. Let these be A; and Ay; and suppose that the expression of the primitive is 2= (4,4), yall A Ay At an initial point of the curve, let «=a, y=, t=¢, where ¢ may vary from one curve to another; then, if and "be denoted by 6’ (t) and y(t) respectively, we have a=$(6, 4,4) b= H(t, Ar 4s), ve), #@% The quantity a, without loss of generality, can be regarded as non-parametric, because the origin can be chosen at will; hence, when ¢’ is eliminated, we have beh(A,, A), m=k(Ay Ad), where the functions & and k must be independent of one another. Also, conversely, the independent constants A, and A, are functions of b and m. ‘Moreover, we have a= (64,4), a= $(t, Ar As), and therefore a—a=(t-t)¢'(t, Ar, A,) + powers of (t—¢); and similarly y—b=(t—-t)¥ (t, Ai, A.) + powers of (t— 1). Hence y—b=m(e—a) + powers of 2a, say ¥=9 (a,b, m), the expansions having been taken in the vicinity of the place (a, 8). Thus the primitive of the characteristic equation can be taken in the general form a= (64,4), y=¥(t Av 4s);80 LINEARLY INDEPENDENT INTEGRALS OF (on. 1 and also in the more particular form yab+m(e-a)t.. =g(@, b,m), where b and m are functions of 4, and 4,, as A; and A, are funetions of } and m, the two independent arbitrary constants in either pair being the essential arbitrary constants of the primitive, ‘These quantities « and y are such that the equation @=0 is satisfied, whatever be the arbitrary constants A, and A, Let A, and A, suffer small arbitrary variations so that they become A, +o and A, +40, respectively, where ¢, and ¢ themselves are arbitrary constants. The characteristic equa- tion, now E=0, is satisfied. But the quantity «€ is the term of the first order in x in the expansion of $(t, A, +xc,, A,+ 0), and the quantity en is the term of the first order in « in the expansion of y(t, Ay + 01, Ay +0); hence ; 1b brane sagt on ‘eA, =o +¢, aay We write xO= 08h "Ou 20-6 OF -¥Ox and then Saan— nt = ax (t) too (2). . But this quantity £ as was proved in §58, satisfies the equation da GP) - Gb= and it contains linearly two independent arbitrary constants ¢, and c. If then x(t) and o(t) are linearly independent functions of 1, this quantity £ provides the primitive of the subsidiary characteristic equation. 61. Ifx(t) and «(¢) are not linearly independent, a relation ay (t) + Bo(t)=0 would subsist, « and 8 being constants. Now =$(t4 4), y= (t Av Ad), a a= $l ya=¥ 1) yab+m(e-a) +(e- a R(e~ a,b, m)=9 (5, m),61) ‘THE SUBSIDIARY EQUATION 81 where R (ea, b, m) is a regular function of «a, and where A, and A, are two independent functions—any we pleaso—of the two independent arbitrary constants 6 and m. Now Ov _ ay 9 Oe 8g | ay Om 9A, 04, 020A, "0b 9A, * dm 94, = 49 28 09 0, ag Im ~ $0 04,” 6b 04,” Om GA,’ and therefore xO=8 OF -vO matey (20 2 4 Bg, Bem 80 (Senthia): Similarly — 1g ig Om 0-8 O85 sa, ama)” Ir, then, the foregoing postulated relation could exist, we should have 29 /, % , 9 2), ay (am pam\_ 3B (*3a,* aa) tim ("aa +Poa,) —° for $'(f) is not a permanent zero: that is, there would be a linear relation between the variable quantities 2 and | sin which the coefficients are con- stant. But, a3 ag aR gpa ite—ay Ss, Mmw—ar(e— anh, it is clear that no such linear relation betwoon 2% and 2? ean exist; hence the foregoing coefficients must vanish, and therefore a 4 am, 3m ‘Now 6 and m, as functions of A, and A,, are independent of one another, so that is not zero; we must therefore have a=0 and 8=0. Consequently, there is no linear relation between x(t) and w(t); and 60 S=ax®) +e0(t), where ¢, and ¢, are independent arbitrary constants, is the primitive of the subsidiary characteristic equation,82 ONE SPEOIAL INTEGRAL OF THE SUBSIDIARY EQUATION [CH II 62. Thus x(t) and w(t) are individual integrals of this equation, and therefore diy , dP dy apt “a dt -Gz=0, Peo dP dw PRtaa hence, by the well-known property of linear equations of the second order, : a - ® Now P does not vanish with the range of t, the Legendre test being supposed satisfied ; and there is no relation of the form ay (t)-+ Bw (t)=0, where a and A are constants, so that the quantity multiplied by P does not vanish, ‘Thus the new constant is not zero and, because neither P nor x(t) nor w(t) can involve the arbitrary constants in the primitive of the subsidiary equation, the new constant is determinate and not arbitrary. Denoting it by 0, we have do ax). P(E ae B= We infer, from the fact that the constant @ is not zero: that (j) the linearly independent integrals () and w(t) cannot vanish simultaneously for any value of ¢ in its range (Gi) the first derivatives of x(t) and a (¢) cannot vanish simultaneously for any value of ¢ in its range ; and Gii) no root of x (#), oF of w (t) oF of any integral of the subsidiary equa- tion, ean be multiple. P —Gw=0; constant. An integral Z(t, t’) of the subsidiary equation, affecting the second variation. 63. We know that every integral of the equation can be expressed linearly in terms of x (¢) und « (1), with coefficients that are independent of ¢; and any such linear combination is an integral. ‘Thus, if ¢' denote any particular value of tin the range, Ze th=ot)xO-xC)o® is such an integral. It will appear that this function Z(t, t') is of critical importance, especially because of its zeros, We therefore note that, effectively, it is independent of any particular choice of fundamental integrals y (t) and (0) of the subsidiary equation : for, if any two others be chosen, necessarily of the forms kyy (t) +k, (¢) and Ly (i) +he(t), where kyl,—kyh is not zero, the new form of the function Z(t, ¢') is kx® +o, bx @) + he () Ax) + he), bx (C) + he (t)64] JACOBI TEST 83 which is equal to (k,l,—hyd,) Z(t, t). The function Z (t,t) is, in fact, covarian- tive amid all possible choices of fundamental integrals; and it cannot become evanescent, because x(¥) and w(t’) cannot vanish simultaneously for any selected value ¢’ in the range. The Jacobi tet 64. We retum to the second variation of our integral J, and specially to the consideration of the permanence of sign of fou, where fos Jo (mbe) a where now the limits of the integral are regarded as fixed values. The quantity Pis to preserve a uniform sign along the range, in accordance with Legendre’s test; the quantity (w -Suy is always positive when it does not vanish. ‘The acquisition of a zero value, save for isolated values, must be avoided if possible; and such acquisition can be attained if, over some continuous stretch in the range, we could have w=f, where v is a constant. Let the stretch be the whole range—the argument is a fortiori valid if it be only a portion of the range, because we could take w=yf over that portion, w=0 before that portion, and w=0 after that portion. We have w=0 at the lower limit t,, and therefore at the lower limit { would have to vanish: that is, we should take f=2(t 4). We know (§62) that £ cannot anywhere have a repeated zero; thus, in the immediate vicinity of f, we must have £=2(¢—t)+ higher powers of ¢—f, where 2 is a non-vanishing constant. Again, our quantity w is to be a uniform function of the variable, which must vanish at t, because w vanishes at the lower limit; hence any admissible value of w is of the form w= (t—t)" + higher powers of tf, where* n>1, In the immediate vicinity of f, we have w— = (n—1)(¢— 6)" + higher powers of tt, a quantity which vanishes at t, whether n=1 or n be greater than 1. Thus there is no infinity at the lower limit, For the rest of the range, the possibility of w= yf makes * In the text, w is uniform and so n is an integer; for the immediate argument, w might be ‘a radical. But w; is not to suffer discontinuity for values of ¢ that aro considered; in that case, the positive quantity n would be greater than unity,8 LIMITATION OF RANGE OF INTEGRATION [on. a If therefore at the other extremity of the range, or earlier than that other extremity, the selected quantity can again vanish, the terminal condition as regards w is satisfied by the assumption w= f. All the other conditions are satisfied ; and so the second variation would vanish for the particular small variation xw. We should not then be in a position to assert the existence of a maximum or a minimum for the integral, because of the lack of a uniform sign, either negative or positive, for all small variations of the charactezistic curve. To settle the doubt, we should (in the absence of other knowledge*) then have to consider the behaviour of the third and higher variations for the particular small variation w =f under which the second variation vanishes. But if the quantity {=Z(, 4), vanishing at the beginning of the range, does not again vanish in its course—or, in other words, if the range beginning at f, doos not extend so far as the first zoro of Z(t, &) for a value of t greater than t—then we cannot have a variation ew, given by w= (t, t), because w would not vanish at the further extremity for such a variation. Over a range thus limited, (m- $m) is always positive for all non-zero small variations ew; and then fou remains permanent in sign for all small weak variations, that sign being the same as the sign of the quantity P which docs not vanish throughout the range. We thus find the extended form of what has been called (§ 26) the Jacobi test, viz. ‘The range of the integral, beginning at a value ty, must not extend so far as the first zero of the function Z(t, t,) which is greater than ty. Conjugate places in a range of integration, 65. Once again, attention must be called to an implicit elusive assump- tion, ‘The range, beginning at f,, must not extend so far as 7, where 7, is the first value of ¢ after t for which Z(¢, &) vanishes. Let a value ¢’ be taken within that range, that is, such that 4<¢ <1; and consider a function (t,t), which of course vanishes at ¢. We could have a small weak variation sw, where w is made zero from t, to ¢, and is made equal to Z(t, ¢) from ¢ onwards; and such a variation could continue onwards, even if Z(t, ¢’) should again vanish within the range or at its limit, that is, if the range contains a value 1” oft where 1” is the first zero of Z(t, ¢) next after ¢. Manifestly this possibility has been implicitly excluded in the preceding statement of the range; the implicit exclusion must be justified +. For this purpose, and also specially in order to indicate a geometrical significance of the results, the characteristic curve may be considered with advantage. ‘The expressions for 2 and y lead, when the variable ¢ is eliminated, to the equation of that curve, The quantities xf and ey, leading to the forma- + See § 29: and 3g 76—80, post. + The anslytical justification will be found in § 67, post.65) CONJUGATES IN RANGE 85 tion of fie. of the function Z(t, ¢), are small variations of « and y which still satisfy the characteristic equation and therefore give a consecutive character- istic curve. The latter, as represented by the variation Z(t, 4), meets the original characteristic curve in the point determined by the value t of ¢; and the range of the integral, which is to possess a maximum or a minimum, is not to extend so far as the first value of ¢ greater than %, which corresponds to the next point of intersection of the two curves, Accordingly we take a characteristic curve ...APB... and, on it, choose any ordinary point as the initial point for a range. In connection with the characteristic equation, the curve is made precise and unique, by the assign- ment of the valuo of y and the value of $Y for the initial value of 2—or, as already explained, by the assignment of initial values of 2, a, y, gr for an initial value of the variable t: that is, the curve is made precise by the assignment of an initial point and an initial direotion, Now consider a cha- acteristic curve AP’... consecutive to APB, passing through the same initial point’ 4; as the variation is small, the inclination of the tangents at A is small. Let P on the curve ...A PB... be given by a= $l dn Ad, y(t 4s, 4), so that, if & be the value of ¢ at the initial point A, the coordinates of A are $ (ls, 4s, 42), (lo, Ay, 43). For the consecutive curve, there is a small varia- tion; letit be represented by t+ «T,, A, + xa, Ay +a, instead of t, Ay, Ay respectively, and by t-+«7' instead of t, where 7’ is not necessarily the same as T,, As the initial points are the same, we have O-« {1.6 (+04 2) ay 2S terms of at least the second order in «, (i , dwelt era tee Da ht Me. and therefore, on the elimination of 7, 0 =« {a,x (t) + dae (t,)} + terms of the second order in «: that is, in the limit as « diminishes indefinitely, 4x (4) + age0(b) = 0. Let the point P’ be the displaced position of P due to the small variation, s0 that the coordinates of P’ are «+ xf, y +7, where f=T¢ +a eye Oy 028, a + terms of at least the first order in x, od nal¥ +a nyt +a .86 GEOMETRICAL FORM OF JACOBI TEST (cH. 0 that is, S=an—yk x (t) + a, (t) + terms of at least the first order in x. Lot the curves, if they meet after A, mect again in B for the first time after A; and let @ be the value of t at B for the characteristic curve. Then £ and ‘n, and therefore &, vanish when t= 6; that is, 0 = ayy (0) + ao (9) + terms of at least the first order in x. Hence, in the limit as « diminishes indefinitely, O= any (8) + ax (0). ‘The arbitrary quantities a, and a, do not vanish; hence Z (8, te) = 0 (te) x (8) — x (t) @ (8) = 0, and @ is the first value of ¢ greater than t,, for which this equation gives another intersection of the consecutive characteristic with the original characteristic. ‘Then Bis called the conjugate of A on the characteristic curve. On the one hand, it is the limiting position of the first point of intersection of the consecutive characteristic with the original after the initial point of inter- section. On the other hand, it is given by the first value @ of ¢ greater than t, where Z(t, 4) vanishes. We may therefore re-enunciate the Jacobi test by requiring that the range of integration must not extend. along the characteristic curve as far as the conjugate of the initial point, 66. We may take x(t) and w(t), more briefly denoted by x and @ when there is no doubt as to the argument, as the fundamental integrals for the discussion of the covariantive function Z (t,t); and the relation dw qd P(x Ga? B)-0 has been established, where C is a specific constant. Also Z(t) = 0 (0) xO- xt) ox), so that. GEC 1], elt x dx (to) c ae where P,, not zero, is the value of P at t; thus T(t te) verifying the former inference that t, is a simple zero of Z(t, t)- Then as 2(@, t,)=0, we can take XO) = 0x (6). w(0)=cw(t), ¥ (¢— 6) + higher powers of tt,67] PROPERTY OF CONJUGATE-BOUNDED RANGE 87 where o is not zero, for otherwise x() and »(@) would vanish together; and 80 oS (t, ty) = w (8) x(t) — x (8) w(t) =Z(,6). As before o " where P, is the non-zero value of P at t= 6; and again, as is to be expected, 9 is a simple zero of Z(t, 0), that is, it is a simple zero of Z(t,t). Thus Z(t, t) changes its sign as the variable ¢ passes through the zero 6: and so for any zero. Wote, We have the corollary that, if we pass backwards along the curve, the limiting position of the first point of intersection of the original curve by a consecutive characteristic through B, is the point A. For Z(t, 0)= 02 (t&), and the first value of ¢, less than @ in the range between ¢, and 6, for which Z(t, &) vanishes is t,; thus A is the backward conjugate of B. Z(t, 0)=—F (t— 0) + higher powers of t— 6, Evelusiveness of a range bounded by conjugates, 67. We can now establish the proposition that when any point A’ is taken in an are of the characteristic curve, limited by two reciprocally con- jugate points A and B, the conjugate of A’ lies outside the range AB. (It will be sufficient if we establish the result for the forward conjugate of A’, on the assumption that B is the forward conjugate of A.) Let A be the point determined by the value f of £; and let @ be the value of t at its conjugate B, so that 4(@, t) =0, where 6 is the first value of ¢ greater than ¢, for which this equation is satisfied. Let f be the value of ¢ for A’, so that 4<& <0. As Z(6,t)=0, we have x@O)=ox(t), o(8)=o0(t), where o is not zero; and therefore 40) (h) x (8) — x (4) @ (8) {a (t) x (4) x (4) @4)} =— 02 (t,t). As t, lies between f and @, we cannot have Z (4, %) equal to zero: hence Z(6, t,) does not vanish. Consequently, the conjugate of ’ does not coincide with B; and A’ is any point within AB.88 NO CONJUGATE-BOUNDED RANGE [ona Now take a point A, within A.A’ and near to A; and let the value of ¢ at that point be f,+ 7,, where 7, is small and positive. As A, is near to A, the conjugate B, of A, must be near to B. Let the value of t for B, be +0; then @ definitely is not zero when T, is not zero, because the con- jugate of A, (a point in AB) cannot coincide with B. Also, when 7, is zero, then © is zero; so that @ and 7, vanish together, and @ must be small when 7; is small. ‘The sign of ®, as well as its magnitude, has to be deter- mined, Because @+@ gives the point conjugate to the point given by t+ Ty, we have Z2(04+0,t+T); and therefore as Z(6, t,) = 0, and @ and 7, are small, eZ 2). 2, 2G. a, at the unexpressed terms being of the second and higher orders in @ and 7, combined. But, as Z(t &) = 06) x(O-x(h) @ (0), we have FED w(e)x (O- x(t) oO; and therefore AE) - ater x(t) a! (0). Now xO) oO), XG)" ot)” where o is not zero; and so FE) 1 tw x OKO} o — Also (4) x (8) — x (be) 0 (8) =o {a (te) x (th) —X (t) «(t)} oo Py ‘Thus the equation connecting @ and 7, becomes o c -0G, + Te pt or, for sufficiently small values of 7, Pop, +higher powers of 7,68] ENCLOSES A SIMILAR RANGE 89 Now P, and Py have the same sign because the Legendre test is satisfied, and @ is a non-zero constant; thus ® is positive, because ‘7, is positive. Consequently 0 + @ is greater than @: and so the conjugate of A,, which is near A in the range AA’, lies near B but beyond B. Let it be denoted by B,; then A,B, is another range bounded by two conjugate points, BB, ‘being of the same order of magnitude as A.4,. Now take a place Ay, in the range A,B, and near Aj; its conjugate, say B,, lies beyond B,. And so on, with points A in succession from A towards A’; the successive conjugates lie beyond B in succession, each beyond its predecessor: until, finally the conjugate of A’, say B’, is reached, and it thus lies in the stretch beyond B, Lemma on consecutive characteristic curves. 68, The function Z(¢, %) is of significance in a cognate investigation. Hitherto the small variations upon the characteristic curve have started from the curve itself; and the deduced characteristic curve has not been required toconform to any external condition. At a later stage, it will be found desir- able to draw a characteristic curve, closely contiguous to a given characteristic curve in the sense that it is to pass through two assigned points very near the given curve: thus they may be taken near the initial point and the final point of some stretch in the range. ‘The conditions, be it. noted, differ from those hitherto imposed, viz. that the curve shall pass through an assigned point and shall have an assigned direction at the point; the new conditions require the curve to pass through two assigned points which lie very near a given characteristic curve. A question arises as to whether there are limi- tations—and, if so, what are the limitations—upon the data, in order that the new characteristic curve may be drawn. We therefore postulate a characteristic curve, given by @=$ (tA, 4) y=vlt Ai, A.) ‘We consider o stretch of the curve, beginning at a point (2,, y,) with an initial value %, and ranging with increasing values of ¢ through a point (@,, yj) where t has the value f; thus = Olle Ay Ad, =H Ay Ad, = bb Av AD, = Hey Ay ADD Let (25+ Xo, Yo+ ¥,) be a point contiguous to (2s, y,), 80 that X, and Y, are small: and (2+ X,, y, + ¥,) be a point contiguous to (z,, y,), 80 that X, and Y, are small. If a consecutive characteristic curve can be drawn through these contiguous points, its arbitrary constants ean be denoted by A,+«d,, A, + na;; the value of t for its initial point (e+ Xs, yo-+ Ys) by + «To; and the value of ¢ for its final point by t, + «7, where Z, and T, are finite. Should this be possible for small values of «, then, except for relatively90 ADJACENT (on negligible quantities of the second and higher orders in «, we must have Ayal (0+ 94 BO } fry I, 84 BO. ny oh. 0 FO + 0, Fa ny for the potential determination of Z,, 7), a, dz, all of them finite quantities, while X,, Y,, X,, Y;, « are small quantities. ‘The determinant of the coefficients of 7, a, «a;, eT, on the right-hand side of these four equations is equal to Flt be); and therefore each of the quantities KDE (hsb), KZ (tsb) eZ, %) «TZ (t,t) is equal to a linear and homogeneous function of X,, Y,, X,, Yi, the coefficients being integral combinations of the third order in the derivatives Of (ly Ar, Ads $ (bs Av, A), (ts Av, Ay) and Y(t, Aa, Ay). If then Z(t, t) is not zero, or is not itself a very small quantity of the same order of magnitude as any one of the four small quantities X,, Yj, X,, Y,, the resolved equations give values for «7,, xa, x4, «7, of the same order of magnitude as X,, Y,, X,, Y,: that is, a consecutive characteristic can be drawn as required. ‘Thus the possibility depends upon the magnitude of Eh, b). We have seen that the conjugate of a point given by t= t, is the point given by the first root of Z(@,t)=0 which is greater than %. If t, is appreciably less than 6, that is, if the point on the characteristic curve given by & falls short of the conjugate of the initial point by an are-distance of a magnitude greater than the order of the small quantities X,, ¥s, X,, Yi, then the consecutive curve can be drawn: and it is a unique curve. If however t, = 6, the equations certainly do not give small values for «T., xa, d,, «T;; and if t,, while less than 6, differs from it by a small quantity of the same order as X,, Y,, X,, ¥,, no safe inference as to the magnitude of xT), xa, ea, eT, can be made, without investigating the values of the coefiicients of X,, Y,, X,, Y, in the expressions for those quantities. It is unnecessary to consider values of t greater than @, because the range of the characteristic curve is restricted (for reasons already adduced) not to extend beyond the conjugate of its initial point.69) (CHARACTERISTIC CURVES 91 69. Lastly, part of the preceding analysis (go far as it applies solely to the curves at their. extremities corresponding to the value f of f) can be used to establish an assumption slightly wider than one which has already been made: a characteristic curve, consecutive to a given characteristic, can be drawn through a point very near the latter so that the inclination of its tangent to the tangent of the characteristic is very small, We know that a characteristic curve can be drawn through any point with any assigned initial direction. For the present purpose we have to prove that a new characteristic, drawn as required, is consecutive to the given curve. ‘Taking the given curve in the usual form, lot 4y+ a and Ay + xa, be the constants for the new curve, where it must be proved that xa, and xa, cour as small quantities under the given data. Let the initial value ¢ for the given curve be %, as usual, and for the assumed curve be + x7), 50 that the value of ¢ anywhere is ¢ + «7, where 7= 7, when t=, (the value of Twill disappear from the analysis); we must prove that «7, is small. Let the initial points of the two curves be (a, y.) and (+ X., y+ Y,), where X,and Y, are small; thus we have a Xia Ded (t+ em BOD + ky BO, we Top () + 0, LAO) 0a EO 4 Y= where the unexpressed terms are of the second and higher orders in the small quantity *, and are negligible in comparison with the retained terms of the first order. Let (a, y) be any point in the vicinity of (z,, ys) aloug the range; and let (7+ X, y+ Y) be the displaced point on the new character- istic, We denote by t the current variable at (#, y), and by t+«T' the current variable at (e+ X, y+ ¥) along the new curve; so that 7’ may be a fametion of ¢, and is equal to 7, when t=f. Then Ha 28 (0+ 00, BO 4 ga, BO 5 PaaTy (+ 10 FO + 10, CO + and therefore X! mal" $ (0) + wT (0) + a, OA 2 $0,220 5 yo +0, KO 4 Va ely Otel yO +c, FO92 ADJACENT CHARACTERISTICS [ou where the unexpressed terms are negligible compared with the retained terms of the first order in «. Hence aY'-yX'=¥'$'()-X'¥ (t) =AD IS OY O-¥ OG) a ‘ a +f OO yO} + cafe Oyo Salt The inclination of the tangents to the curves, at en y+Y) and (@, y) respectively, is : ee w4x" a At the initial places this inclination is small, and consequently a’¥’ — y/X’ is small there, that is, when t=, and therefore 7=7,; denoting its value at t=, by a small quantity ¢, we have = Ay ($' (he) W" (h) — (be) $” (H)} +0, {e yO 2 + HO) EI eo {er oe io -¥ ef he. ‘Thos there are three equations connecting «2, xa, xa, with X,, Ya three given stall quantities; in each of the expressions, there are linear terms in Ty, #0, ey, The determinant of the coefficients in the three sets of linear terms is FOVO-¥OPO, 8 -HOEO, GEO yw EO OA, 0A, OA, . Od (te) Op (to) ow : ate) : ie : Or (th) Bye (4) ¥@ : “td, : Wa, HOY OHS (9) PEP OBO FO} - {9 oo (te) -y 0) °O) FO oto +{6 wr Wy SIO to. The first line of i last right. -hand side = {POO ¥ OBO ORO yr Q4M = {8 OHO 4 WHO WFO yw WRO} ~ {6 @%O 0} ye) {6° BO ow (te) |70] CHARACTERISTIC THROUGH TWO ASSIGNED POINTS 93 and therefore the whole of the right-hand side =u § {6 EO ¥ wy BO} = 06) {6 to Oy BO x(t) 9 wo c ao by §62. Now Cis a finite non-zero constant; and P,, the value of P at the beginning of the range, is finite and is of the same sign whatever value 4, is initially chosen. The foregoing equations therefore give AD, Kip, eS as expressions that are integral functions of X,, Y..¢, beginning with linear terms in these three small quantities; and © is finite, being different from zero. : Thus ea, kay, #2, are small, and therefore the new characteristic is con- secutive. We infer that a characteristic consecutive curve can be drawn as required. Characteristic curves through any two assigned points, 70. The limitation of the range of integration, so that it shall not extend as far as the conjugate of its initial point, has an important bearing on another issue, already established. It has been seen that a characteristic curve is uniquely determined by an initial point and an initial direction, as data which yield the values of the arbitrary constants in the primitive of the characteristic equation; and no question arose concerning the range of the curve so determined. There is an implicit limitation in the theorem quoted as leading to the unique establish ment of the curve; the theorem is an existence-theorem, and the validity depends upon the convergence of the power-series employed—that is, partly upon the so-called radius of convergence of that series. Consider a range along the characteristic curve, extending from a point, to its conjugate. By the definition of the conjugate, a consecutive character- istic curve can be drawn through the two points; if therefore we substitute (as often is done) a requirement, that the curve shall pass throngh two given points, for the requirement, that it have an initial position and direction, the characteristic curve would not be unique if the two points could be conjugate on any curve through them,4 TWO MODES OF DETERMINING CHARACTERISTICS [ox ‘The insistence on the Jacobi test, that limite the range, excludes the admission of the possibility indieated ; when the test is satisfied, the charac- teristic curve determined by the initial data remains unique for the sub- stituted data, To the discussion of this equivalent requirement we now proceed. 71. The property of conjugates upon a characteristic curve—which, in its analytical form, secures that the function Z(t, 4) of § 63 does not vanish within a range bounded by f and its conjugate on the curve—enables us to compare the two modes of settling the curve. In one of these modes, which is the expression of the conditions used to secure a precise and unique primi- tive of the characteristic equation, the essential arbitrary constants in the general primitive are made specific, by the assignment of any initial arbitrary point on the curve and of any initial arbitrary direction through that point as determining the tangent there. In the other of the modes, the essential arbitrary constants are made specific, by requiring the curve to pass through two assigned arbitrary points. In the former mode, the constants are actually determined by the data, The latter mode is more descriptive in character; in order to complete the precise expression of the primitive, equations (not always even algebraical in form, because the general primitive is not always or often algebraical) have to be resolved. As is ween in the example of the catenary (§ 30, Ex. 1: §75, Ex.1), the resolution of the equations may provide results, which may be satisfactory, but among which discrimination is not made without further examination if there is more than one result; or the resolution of thé equations may prove initially impossible in terms of real quantities. The case of failure, as to the catenary, is treated by reviewing an initial assumption that the curve necessarily is a catenary and that, in the assumption, a zero value for the arbitrary constant is necessarily excluded. ‘The real significance of the difference is that the two modes of determination are not always equivalent in all circumstances; and the property of the function Z(t, i.) shews why the second mode of determination does not neces- sarily lead to the same result as the first. 72. To compare the modes, let A be the initial point common to both; and let B be the final point in the second mode, For the initial arbitrary direction through A, let the straight line AB be taken; then, in ite earlier course at least, the characteristic curve K lies on one side of the line AB. With B as contre and gradually increasing radius, let a circle be drawn until it touches K, say at a point ; and along CB take a point ’ near C, so that CC" is finite though small. Along K, let the value of ¢ at A be f and its value at C be 7, Then (§ 68) we can draw a consecutive characteristic through A and ’, provided the funetion Z(7, t)) is not zero nor a small quantity of the same order of magnitude as the small length C’B. Thus, if C is not near A’, the conjugate of A, and if it lies within AA’, a consecutive characteristic73) COMPOSITE SMALL WEAK VARIATIONS 95 curve AO’ can be drawn; and the distance of B from AO’ is less than the distance of B from AC, Now take AC’ as the first: characteristic to be drawn, As before, with B as centre and gradually increasing radius, let a cirele be drawn until it touches the new curve K’, say at a point D. In DB, take a point D' near D, so that DD’ is finite though small: and let the circle contre B and radius BD cut OB in 0”, We now can draw a characteristic curve through A and 0”, which is consecutive to AC’; and it is always possible to draw the new con- secutive curve, provided 0” is not near A”, the conjugate of A along the curve AC’, and provided 0’ lies within the range A.A”, bounded by the initial point A and its conjugate. And the new curve, still a characteristic curve, passes nearer B than does AC’, the distance from the preceding curve being finite though small, We thus may pass from curve to curve in succession, each curve being drawn nearer to B than its predecessor. The one persistent requirement, exacted as the condition of drawing a consecutive curve at each stage, is that the function Z(t, 4) must never be zero nor (at any stage) be of the same order as the small (though finite) distance taken nearer to B as a point for that consecutive characteristic. When therefore the field is such that, at no stage in drawing the succession of characteristics, a conjugate of A is approached within a small distance, we can pass from the initial arbitrarily drawn characteristic to a characteristic, still having A as ite initial point, and passing through the second point B, It follows that, when B lies within a region of the field bounded by the locus of the conjugate of A for different characteristic curves, a characteristic can certainly be drawn to pass through the two points A and B which are given initially. It thus appears that, within certain limits as to the range of B depending on the nature of the characteristic equation, the two modes of rendering the general primitive a precise curve are equivalent to one another, while, outside that range, their equivalence cannot be asserted®, Composite small variations. 78. Finally, one inference from the general result is to be noted, It adds nothing to the conditions, and it does not limit them. In character, its interest is not immediate but prospective, as regards the subsequent analysis in the discussion of relative maxima and minima (Chapter VIII). In all the variations which have been imposed, we have dealt with only a single general arbitrary small variation of « and of y represented by xu and xv. All our results have been deduced from the consideration of that single variation, quite general and quite arbitrary within the limits of continuity. * An illustration of the possible range of the conjagete of an initial point, for varying initial direotions, i given in §75, Ex. 8, for the eatenary.96 CUMULATIVE EFFECT OF SMALT, VARIATIONS [own But a small variation may be compounded of a number of small variations, equally general, and equally arbitrary ; thus we could have Kit F Keylig bee tigi, KU MaDe Hoe + nny as a combination of arbitrary small variations, «,,..., ¢q being small arbitrary constants, and* 1, and 1, 1%, and %,...,q and t, being regular functions of f. It is not inconeeivable that a combination of two such variations, distinet from one another, might have some combined discriminating effect upon the second variation, which is quadratic in the quantities u and v and their derivatives, though not upon first variation, which is linear in those elements. To estimate the significance of the possibility, we note that the quantities @ and y, constituting the primitive of the characteristic equations, do not depend upon «, u,v and are not affected by them, ‘The same remark applies to & and », constituting the primitive of the subsidiary characteristic equa- tions; and it therefore applies also to ¢ which is defined as any. If then we write Wy mt TU — Yadiry (for r=1, ..., m), denote the foregoing composite variations by xu and xv (where, if we please, we may take « equal to any multiple of «+... +n) and write waay— yr we have KW = Ky Wh +... + RW Now xw, so defined, is an arbitrary small continuous variation; and all the results, that have been obtained, have been established for ew. ‘They hold for each variation taken by itself, such as xu, and ev,, because it is merely a special instance of xu and xv. ‘That they hold for two or more such varia~ tions taken simultaneously, and that the requirements of such simultaneous variations lead to no additional conditions, ean be seen at once from the full expression for the change of the integral. This change, due to the small variation eu and «v, has been expressed in the form «feuds ye [P (w, 7 ) de + Ba, c where R, is the aggregate of terms of the third and higher degrees in x. Now xw= 5 x,y, and £ (with its derivative) is independent of the arbitrary constants ; hence the change of the integral is equal to & (er f@mrat) +4 [P {5,5 (eo) — 23 (ea h de Ky Because @=0, the first variation vanishes. The expression of the second variation takes account of the simultaneous variations xu, and xv,. It remains * For the immediate purpose, the subscript indices merely denote distinct variations and not derivatives of s single variation.74] SUMMARY OF RESULTS 97 steadily of one sign when the Legendre test and the Jacobi test are satisfied. No new test, to meet the composite variation, is required in order that the second variation should remain of one sign. Hence the analytical results, holding for one arbitrary variation, remain ‘unamplified if any number of simultaneous variations are imposed. Summary of results. 74, We may now summarise the conditions which have been obtained as, necessary for the existence of a maximum or a minimum in an integral [Fewawae, when the variables « and y are subject to weak variations of the general type. ‘These necessary conditions correspond exactly to the conditions obtained for the special weak variations, Both of the variables are now subject to the changes, so that the form of the conditions differs from the form of conditions for the simpler type; and other properties have arisen incidentally which lie outside the considerations governing the former case. ‘The conditions still consist of three tests : I. The Euler test: the characteristic equation @ = 0, or the two equations X=0 and Y=0, must be satisfied by « and y as functions of t; and their values determine a characteristic curve (§ 42, 44, 49). IL. The Legendre test: a quantity P, being the common value of Bea i oF 1eF way?’ — my, Or0y,” ys? Oat? must keep an invariable sign throughout the range of the integral and may not vanish, the sign being positive for a minimum and negative for a maximum (§ 44, 56), IIL. The Jacobi test: the range of the integral must not extend over the whole stretch of variation represented by the are of the character- istic curve between a point and its conjugate (&§ 64, 65). Incidentally, it was shewn that the function F satisfies the identity ar, | oF. Sag th Gy oF and that the second derivatives of F are expressible in terms of a diminished number of magnitudes (§ 37, 38, 51): that, if the limits of the integral are not given as fixed, they are determined by boundary conditions (§ 43): that, if the characteristic curve (supposed to be otherwise everywhere definite in direction) suffers a discontinuity of direction at a single free place, the quantities 2° and z are continuous ab the place (§ 47); and properties of conseeutive characteristic curves, useful in connection with the Jacobi test, have been established (§§ 68, 69).98 EXAMPLES OF ow. 11 Moreover, the threo necessary tests, if satisfied, are sufficient to secure a maximum or a minimum under weak variations; but strong variations remain for consideration. ‘The first (Buler) test, together with the boundary conditions, secures that the first variation of the integral. vanishes. The second (Legendre) test and the third (Jacobi) test, taken together, secure that the second variation cannot vanish; according as the sign of this second variation is positive or is negative, a minimum or a maximum respectively is possessed by the integral. Note. If it should happen that the Jacobi testis just not satisfied because the range extends to the conjugate, the second variation may vanish specially. ‘The discrimination then can be made (as in the example on p. 38) by the examination of the third variation of the integral for that special variation of the characteristic curve. 75. Some examples will now be considered Ex. 1. 'To illstrate tho preceding analysis, wo consider again the problem of the catenoid already (§ 30, Bx 1) discusod, to determine the plane eurve which, by revolution round the ais of inthe plane, provides the eurface of minimism area. (Ay), As the integral is ae fyarynbas, =2nfyiarennban, wo have FoF (2,9 ay y=y at+ynh, which in at ones seen to verify the permanent identity of § 87. As F does not involve # explicitly, we choose the equation Y=0 to obtain the characteristic curve, as being the easier one fo integrate, Tt is and therefore that is, or, if we take # the intrinsic equation of the characteristic curve, a catenary. ‘The variable 4 even in implicit oocurrence, has disappeared. (Ay fi). If we proceed from the equation @=0, we here have —Setein—nad) — 249 2+y:2t etty ity (atin?) jetty YY (A192 Yita) Hi (HP HY") =O. pat always, (anni) (Pye) 4 (2) = Se be ai \x,, a : so that 1h C)=Lerss m2 b(t75) ‘THE WEIERSTRASS METHOD 99 for the equation: oF if yy=, tan yy this is oy, tan yy ad ‘and therefore yoos =o. If we proceed from the equation ¥=0, which here is ay om catenot-5{ Mo} Ho Katty) this becomes : wn - Russ Be =o, ‘that is, = yoo yyy sin ye yo08 aman sin ye Et and therefore sarhiany, with the same result, (A, ii. "The only limitation, imposed upon the independent variable ¢ is that it shall be capable of continuous increase through the range. When account is taken of the form of the curve, and the positive direction of the tangent lies in the direction of increasing 2, the variable y isalways increasing—it may begin by being negative. ‘The variable ¢ can be changed arbitrarily, subject to its one limitation ; 90, a8 see is always positive for the range and is nevor numerically less than unity, we take see y=eosh 7. nh 7; hence 2, =07", and so we have snate?, y= coosh 7, where the function "isto increase continually with ¢, but otherwise is arbitrary. Manifestly the simplest form is given by ‘Then tan yesinh 7, and ys=e?" eeate srccab ‘where an arbitrary constant implicitly additive to ¢ (as in § 49) is absorbed in ¢, and where the two essential arbitrary constants are a and ¢. (B). As regards the Legendre test, we have OF got Beary? cosh that is, Pay(asey)”', the postive sign being taken for the radical (z32+9;2)8 throughout the discussion, ‘Thus P in alwayn positive. The Legendre teat is satisted; if other teste are satisfied, it allows minimum, : (©), Nest, for the Jacobi test. With the notation of the text, we have abetmzm$lla,0, ceoshtaymy(s a, 0) where a does not ocour in y. Hence xO OE-¥ © Sa -esinns, 29-9 0%-v(0% =eeote-teiohs100 CONDITIONS FOR [om m and therefore Z(t Wael) x O-x (al) se fsinh f(osh ¢~ ¢sinh &)—sinh ¢ (cosh ff sink f)) which manifestly vanishes when f=%. For the conjugate of fy, we require the first value of ¢ which, being greater than f, makes Z(¢, ) vanish. Now az, Fm ~ ct cosh ¢ (¢sinh ty+cosh fy~f sink the initial value of which is the negative quantity—c*cosh® ty. Hence Z(t, ) zero at foy Decomes negative at once as ¢ increases from fy; as it remains finite for finite values oft it continues negative until the value 4, the conjugate of t. If is positive, then sinh f sinh ¢ (with the range ¢ > ‘) cannot vanish therefore, as Z(t) =< sinh ty sinh tooth - ¢—(coth %—6)}s Z can only vanish for a value of ¢ greater than fy, such that coth ¢—t=eoth fo fy. Within this range, coth tt initially is equal to coth &—%, and is always decreasing because its derivative is negative, and it remains finite for all finite values of ¢; hence it ‘cannot again resume its initial value. Consequently if ty be positive, there is no conjugate ‘with a value of ¢ > fg: that is, no forward conjugate. When % is negative, Z(t, f,) becomes negative at once as ¢ increases from f; when ¢ is equal to 0, then Z( (9) is equal to esinh , ‘that is, it still is negative. Denoting the nogative value of f by —f, the place 4, where Z frst vanishes after the beginning of the range, ia given by coth 4 f4=¢ —eoth f. ‘Whether ¢—cosh ¢ is negative oF positive, this equation gives one (and only one) positive root f;: and it gives no negative root. (A \ “L negative value of ¢—ooth? gives B in the figure: a positive value gives in the figure.) The tangent to the curve at ¢ is Y¥=coosh =(X~a—et)sinhe, 0 that the abscissa of the intersection of the tangent with the axis of ais ate (¢—coth’, the same as the abscissa of the intersection of the tangent. at ¢, with that axis; the con- jugate of the initial point (or D) is obtained by drawing the other tangent through U (or V) to the curve. The construction of the conjugate is thus the same as in the former investigation (§ 30, Bx. 1) ‘When the initial point and its conjugate are symmetrically situated with regard to the axis, we have 4=1'; and the equation is coth ¢=e, ‘the former limiting case, (D). For an effective solution by the preceding analysis (with o real, and different from zero), there must be a real catenary. Hence the initial data must be examined, to indicate whether they allow a real catenary ; manifestly they admit no such curve if, for example, ft is to join two points on opposite sides of the axis. The only alternative left is that e should be zero, when the foregoing analysis does not apply ; and the alternative must be considered independently.75) REALITY OF CATENOID 101 ‘We therefore take the case when the curve is to join two points on the same side of the and we'shall assume now (as distinct from the assumption in § 30, p. 81) that the Points are not equidistant from the directrix. Taking the axis of y midway between the points, we denote these by (£, a) and (4, ~a); 0 that, as zeate, —ymocosht, we have 0 that and therefore, on eliminating a, E@ate (86-6 ‘n equation for the determination of ¢. For the existence of a real catenary or real catenaries, this equation must have a real root or real roots, (E). Consider the march of (0), as the positive quantity ¢ ranges between zero and infinity. Let cun ta, that the positive quantity u ranges between infinity and 0; then Blo)= he? (cosh 2u 1) +(L— A)? 28h (cosh u—1). “Henoe, when o is very large and therefore w is very amall, Z(c) differs only slightly from dat + (kA), ‘and thus is positive, For any finite non-zero value of e and therefore of 1, H(¢) is finite, As c becomes small and therefore w becomes large, H(c) grows large and is positive; and 4 zero value of ¢ makes H(c) infinite. Henco as £ (0) is positive and E (co) is positive, the ‘number of real roots of (0), for 0 $a?for the possibility of any real catenary. Though @(), ‘and therefore £'(c), then has one real root, it may not lead to a real root of Z(@). The existence of a real root, or of real roots, will depend upon the magnitude of the quantity Ak— $a, necessarily positive, (F). Asan illustration, consider a case Meme, ‘The instance, when Eek, has already (§ 30, Ex. 1, p. 81) been considered ; so we take Eand unequal, eg. Kea, A=Va We now have Gynt (a4 and itn eaay to verify that (@) when u is small, G6) ia positive, when wa12}, (0)—-014, (ii) when v2=13, 0 (¢)= - 018. (iv) a8 u®inoreases further, @ (c) increases numerically and remains negative. Hence Gc) has one and only one real root, lying between (299)* a and (394)? a; thus ‘E(@) has one minimum, and the minimum ocoura for that root, It is easy to verify that, ‘when w= 124, Z(e) ia negative ; hence £(c) has one, and only one, root, such that o 0-< 2m, be the smallest value of 6, for the initial position, and if ,,>0-<2m, be the stnallest value of @, forthe final position, then (i) if 9,>9,, while s, and sy are the initial and the final values of 2, we have an un- limited number of values of eot 8, given by cot m2 (8,~9,+2pm), where p is any positive or negative whole number; and if 9,<3,, the values of oot 8 are given by cot B= —*— (9 +93 Bai Ort. 5143p, with the like significance for #1, 2, 2. Each value of p gives a value to 8, the inclination of the curve to the cross (cireular) section of the eylinder. ‘The Legendre test for a minimum is satisfied for all the forms of f, and so is satisfied for the oylinder. For the Jacobi test, we have 24, tegiing aw plice yen by 4 beer rare ffi value along any belix? that is along any selected heli, there is no lit. ach such helix provides a minimum; small varintions of each helix provide an increase in the length of ‘the curve, ‘There are minima, each corresponding to each value of p; these minima are the numerical values of the critical quantity 24 sing” and the smallest of all the minima is provided by the largest numerical value of sing, among all the values of 8, that is, by cot = Sod ana? with 0 t,14 SACOBIAN RANGE 18 4 {ou 1 As Z(t, t:) vanishes at f, consider two places 4— and t, +7, where 7 is positive and t,+7< 7; and to arrange for even a small extension of the range beyond &,, we take 7 small, We know (§ 66) that the characteristic, which is consecutive to the central curve and passes through f, and f,, crosses that central curve at each of these points. As t, is the first point of crossing after 4, the quantities Z(t, 7, t) and Z(t,+ «Ty, t), where « and 7, are positive, have the same sign; that is, Geom, [$26 0] have opposite signs. Let the former be denoted by Q; the latter (§ 67) is ene, -E where P, is the value of P at t,, the sign of P being steady through the range. Thus Q and C/P have the same sign. Hence D(y—7)=—7Q+ M42) +(e, a4] Dht= 1Q+NG+7 +(e 7} Now choose 7 arbitrarily, keeping it small and making it positive, We have chosen ¢ (our arbitrary constant, also small) of the same sign as P throughout the range. Finally, choose ¢ so small that, through a range of t from t;—7 to +7, a [21 <7/Ql. If Q(t) is positive, then when Q is positive, e (t)—7Q is negative in that range, <2 (t) +7Q is positive in that range: and when Q is negative, 9 (t) +7Q is negative in the range, <2 (t)—7Q is positive in the range. If 9.) is negative, the same results hold. Hence, in every case, =1Q+Q(4—7), 1Q+eNG+7), have opposite signs; that is, there is some place ¢", which lies between &—7 and f, +7, where 7(t) vanishes, This place t” may lie in the range between t—7 and f,, or it may lie outside that range and between t, and t, + 7. We therefore take the range of our integral from f up to the place £”; and to make certain of including ¢”, we assume that the original rango of the in- tegral extends beyond ¢, to 7. A function 7 has now been chosen which (i) vanishes at f,, the lower limit of the integral, di. ii) is not zero everywhere, (iv) satisfies the equation (@-+)0— everywhere along the curve.80) STRICT LIMIT FOR THE INTEGRAL 1s ‘We thus have an admissible variation, conforming to all the earlier conditi and satisfying* additional conditions; and, for this variation, nae) Hence this value of J, is of a sign, opposite to that of ¢; and the sign of e was chosen the same as that of P; so that this value of J, is of a sign opposite to that of P. But for all the other small variations that have been considered, the sign of the corresponding second variation was of a sign the same as that of P. Thus, in the circumstance that the range of integration of the original integral extends beyond the conjugate of the lower limit, there are variations, some of which give one sign to J, and others of which give the reverse sign to I;, while for all of them the first variation vanishes, Hence the integral does not possess a true maximum or minimum: that is, if the range of the integral extends beyond a complete range, bounded by two conjugates on the characteristic curve. Consequently the Jacobian range is a strict limit for the integral, sola 80. The theorem is due to Weierstrass. It proves that, if the range of the integral certainly extends beyond the conjugate of the lower limit, the integral does not possess a maximum or a minimum; but it provides no decision on the issue, if the range merely extends as far as the conjugate. A discussion of the question, when the integral ranges exactly between con- jugates, will be found in a memoir by Osgood. * The non-zero value of @ along the curve ie due to the fact that the quantity Zlo)-HZ (etd {is manifestly not zero everywhere along the curve. + It is published in accounts of the subject, based upon notes taken at his Berlin lectures; e-ge, soe Harris Hancock's Calculus of Variations (the Weierstrassian Theory), 1904, Chapter X. $ Trans, Amer, Math, Soe., vol. ii (1901), pp. 160—182,CHAPTER III. INTEGRALS INVOLVING DERIVATIVES OF THE SECOND ORDER: SPECIAL WEAK VARIATIONS, BY THE METHOD OF JACOBI; GENERAL WEAK VARIATIONS, BY THE METHOD or WEIERSTRASS, ‘The chapter* is concerned with integrals which, in their simplest initial form, involve ‘one dependent variable together with its first and second derivatives, Some propositions are stated for integrals, when derivatives up to the general order n occur. Many extensions almost suggest themselves, so far as regards mere form ; and frequently the establishment of these self-suggested extensions can be made by formal enlargement of the analysi actually given. Thus the analysis, similar to that in §§ 8896 and applicable to intograls containing derivatives up to order n, can be constructed without difficulty. ‘The first portion, comprising §§ $1—96, contains the discussion for the special weak variations of the dependent variable alone, ‘The method, used in Chapter I, is again used. The second portion, comprising §§ 97—128, contains the discussion for the general weak variations of the dependent variable and the independent variable simultaneously. ‘The method used in Chapter II originating with Weierstrass has been used, in a necessarily amplified form. Integrals involving second derivatives. 81, We now proceed to the determination of criteria for the possession of maxima and minima of integrals, which involve the second derivative as well as the first derivative of the single dependent variable. At this stage, only those variations will be considered which are of the ‘weak’ type. Also, initially, the weak variations admitted will be those styled ‘ special" being variations that affect the dependent variable only, the independent variable being free from the imposed arbitrary variations, We thus, in the first place, deal with an integral of the second order, defined as =[Peny.yee, where y/ and y/" respectively denote the first derivative and the second derivative of the dependent variable y. ‘The subject of integration F is assumed to remain continuous, so long as its arguments 2, y, y’, y/’ are continuous. It need not be everywhere a uniform function of its arguments, as it may (eg.) be a radical. But, if it is not uniform, we shall take only such a range as will admit no branch-points: that is, the selected radical will be uniform within the range, so that it may be considered as expansible in regular power-series of appropriate variables. + me substance of the chapter has alresdy appeared in a memoir by the anthor, Proc. Roy. ‘See. Edin, vol. xvi, part ii (1025), pp. 149-195. a connection with the fret portion of tho chapter, the memoirs by Hesse and by Clebech, already quoted (p. 8), may be consulted: thete remoirs dealing with integrals containing the derivative of general order.82] VARIATION OF INTEGRAL OF SECOND ORDER ay Variation of the integral. 82. As before, the special weak variation adopted consists in the sub- stitution of y+ev for y—with the corresponding substitutions for the derivatives—where x, as usual, is so small that any power, which occurs, is negligible compared with lower powers; and where v is a function of a, which can be arbitrarily assigned subject to the requirement of being regular (§9) within a given range of the variable 2, supposed to increase con- tinuously through that range. The limits of the integral may be fixed definitely by assigned data, or they may be required merely to conform to assigned conditions: in the latter event, their determination is part of the problem. To take the wider possibility into account from the beginning, we shall assume that, when the whole variation is imposed, the upper limit, of the integral becomes 2+, instead of a, and its lower limit becomes % + €; instead of @,. ‘Thus the new value of the integral is ts fis F(a, y+ nv, of + 0, y" + 10") deo = yee ae +f" Fay sway tev, y" +00") de, Now (as in § 82) the first integral ["""" Fae differs from ef,F,, where F, ale Japhet, denotes the value of F(z, ys yy") at a, by quantities of the second degree and higher degrees in «; we therefore take it as bP + Ky, with the customary significance of K,’ as the aggregate of terms of the second and higher degrees in x. Similarly, the second integral i + Pas differs from «£,F,, whore F, denotes the value of F(c, y, y', y") at 2, by quantities of the second degree and higher degrees in ej we thereforo take it as «EF, + Ky’, with the customary significance of K,”. Hence the increment of the original integral is «(GF — FF) + Ki — Ky" + [Pe yteny ter, +0") ~ F(a, y, y', y")} de. But the last integral (08 | OF OF a wef (oF eee ee WF) da Ke", where K,” denotes an aggregate of terms involving the second and higher powers of «, Writing K,=K,— K,’+K,", so that K, is the completeus FIRST VARIATION [ex. ut aggregate of terms of the second and higher powers of « that occur in the whole increment of the integral, we have that increment equal to ofan an+["(° Faw vB) as} +e = eGR B41) + Kn "(oe wires irae ‘The whole term in the increment, a involves the first power of x, is called the first variation of the integral. where Vanishing of the first variation: characteristic equation. 83. If the first variation does not vanish, the quantity K, is relatively unimportant in magnitude. In that event, we can make the increment of the integral positive or negative at will, solely by changing the sign of «. If the integral is to have a maximum or a minimum value, that increment must have a persistent sign (negative for a maximum, positive for a minimum); hence the hypothesis admitting the former possibility must be excluded, that is, the first variation of the integral must vanish. We therefore proceed to make the first variation vanish, for all variations which are possible: that is, for arbitrary small variations ev of y throughout the range, and for terminal small variations as arbitrary as the conditions permit. Now, with the assumptions as to the function F, we have ECS) Yde Gy) -8 ap dal de) a and therefore (ofS 2G) Bone R-£ G0} +B). where the integral extends over the range, and the terms outside the sign of integration are to be taken at the limits. Hence the first variation of the original integral is «fre fe & Ey) +d Gh valve 2G} oo and it is to vanish for all small variations, admissible within the range and at the limits. First, consider variations which vanish at the limits but otherwise are arbitrary through the range. ‘The terms at the limits then vanish, because £, 0’ vanish at each limit; and the integral must then vanish also for all84] CHARACTERISTIC EQUATION 9 such variations which remain arbitrary within the range. This result can be attained only if the equation a. e 2) (ip) +a Fy) =° is satisfied everywhere along the range. Otherwise, if @& were sometimes positive and sometimes negative, a choice of » to have the same sign as @ would make the integral positive, and a different, choice of v s0 as to have its sign different from that of @ would make the integral negative: that is, the integral could be made positive or negative at will, a possibility that is precluded. Next, consider variations which do not vanish at the limits, Now that @ vanishes all along the range so that the integral in the first variation vanishes, and because the first variation as a whole vanishes, the terms at the limits must vanish ; hence we must have [er+o Hee ]-o. This requirement may be satisfied automatically by assigned duta. If not thus satisfied, it must be satisfied by inferences conforming to assigned con- ditions, affecting the limits only and determining the limits. 84. This critical equation (called the characteristic equation) @ =0 usually is of the fourth order, and not linear in y and the derivatives of y. ‘The only term involving the highest derivative of y is y”" oF 4 fact of importance in connection with the character of the primitive because of the property to be imposed later (§ 95) that in is never to vanish within the range of the integral. ‘The primitive is therefore of the form y=y(a, Ay Ay, As, Ad, where ;, 4s, As, A, are four arbitrary independent constants, When the form of the second variation (§ 88), as a function of , v, v”, is considered in detail, this value of y (with the consequent values of y/'and y”) must be substituted in the six coefficients ay, dy Qo, x, Giz» Om, Which then can be regarded as functions of z alone, Ex. 1, Prove that, if POI IYVAILEWV4I HYD the equation § =0 is only of the second onder in y. Ex, 2, Prove that, if the equati fan actual identity and not a differential equation, then F(x, y, 9") must be equal exactly to a quantity 100, 2G 2G yt gta where @ can be any function of 2 ¥,120 SUBSIDIARY EQUATION AND ITS PRIMITIVE [on. mm Subsidiary characteristic equation : its primitive. 85, The characteristic equation is satisfied by the foregoing value of y for all values of the arbitrary constants A,, Ay, As, A,. It therefore still will be satisfied, when we effect any small variation upon these constants by substituting Ayt nt, Ay t Wy Ay + ey, Ay + 0 for A,, Ap, Aj, Ay respectively, the new constants ay, a,, ds, a, themselves being arbitrary and independent. ‘The constant « is so small that second and higher powers are negligible in value compared with the first power; henee, if Oe eye Or ey OSE aa ad Pd! the consequent value of y is y+ xn + ..., where the unexpressed part involves second and higher powers of «. Thus x7 can be regarded as a small variation of y, the value of » being immediately derivable when the general value of y is known, ‘As the fundamental change has originated in the arbitrary constants, for any set of which the characteristic equation is satisfied, that equation must remain invariable and characteristic, though individual parts are altered, : ar oF oF The new values 55 Fp gy? are oF, aa oF ao ar, 20 By ag toe By tag to aye tage where the unexpressed parts involve the second and higher powers of «, and where 20 = ayer? + 2ana7! + Laon" + an4* + Layo!" + don”, Consequently the new value of @ is an da an any) e+ f- z e+e Gt te with the same significance as to unexpressed terms. The characteristic equation still is satisfied ; and therefore (22_ a am "lon ~ de Ge +e a wht Hence, as « becomes indefinitely small, this last equation is Sr de () * de Gt) ‘This new equation in m, the coefficients in which are dy, dus Ge, dus y, my (all of them are functions of 2), is linear and of the fourth order. Its 0 in the limit,86] ‘TERMINAL CONDITIONS 121 Primitive therofore involves four constants, arbitrary so far as the equation is concerned. But we have seen that ay ay ay ay. DAG ey th ag tee tha this expression for n involves four arbitrary independent constants, a, de, a, 4, which do not occur in the equation, If therefore the four quantities # ae, fare such, that no linear relation subsists between them involving only constant coefficients, the foregoing expression for » provides the primitive of the differential equation for 7. This property, of linear independence among the four derivatives of y, will be established later (§ 112) and will be assumed in the meanwhile. Thus we have a new equation derived from the characteristic equation ; and its primitive can be derived frota the primitive of the characteristic equation. Accordingly, the equation in 1 is called the subsidiary character- istic equation, sometimes (more briefly) the subsidiary equation When the primitive of the characteristic equation is known, the primitive of the subsidiary equation can be derived without farther integration. If however the primitive of the characteristic equation is not known, but only some integral (perhaps a very special integral) is known which may lead to @ maximum or minimum, this special integral is to be substituted in the six coefficients de, da, Ge, xr, Me, Gu; and the subsidiary characteristic equation remains, Its primitive, which cannot now be obtained by the former process, must be obtained by the customary processes of integration, In fact, to complete the analysis up to this stage, it is always necessary to integrate one of the two equations of the fourth order—the non-linear characteristic equation or the linear subsidiary characteristic equation ; for, if the former has not been integrated in general and only a special integral is known, the latter cannot be used in order to construct the primitive of the subsidiary equation Terminal conditions : four cases. 86. Next, it is necessary to consider the inferences to be derived from the condition at the limits, as represented by the relation [oon -£ Gh eeR-0 to be satisfied in connection with all variations. We can select one variation, which leaves one extremity unchanged: that is, the position of the extremity and the direction of the extremity (given122 VARIETIES OF (cx. by the values of y and y’) are fixed ; 80 &, v, v’ are zero there, Hence we must have @F_ oF) | ar tof 7 -E(B)p eo ro at the other extremity, Similarly, we can select another variation not rigidly settled at the Bist extremity. For the second extremity, the condition just deduced is already required. Hence the same condition, now still remaining at the first extremity, is to be satisfied there. Hence, as an initial (but not completely detailed) form of requirement, the relation (@F 4 (oF OF reo a Getta must be satisfied at each extremity. Various detailed forms emerge, according to the variety of character in the assigned terminal data, As the relation is to be satisfied at each extremity by itself, without reference to the other extremity, the data (and the con- sequent condition or conditions) are not necessarily the same for the two extremities. The result, appropriate to the data at one extremity, must be associated with the result appropriate to the other. (@) Let fixed values be assigned to one extremity (the case really has already been considered). We then have E=0, v=0, v’=0 for that extremity; the relation is satisfied, without leaving any residual analytical condition. (ii) Let the terminal values be required to satisfy a given relation x@ ye ¥)=0, this relation being the only restriction at the extremity. Now, at that extremity, we have supposed « varied into ©+£; let the corresponding values of y and y’, connected with this relation, be changed into y+ and y +7. Then the sole condition imposed upon £,n,1, because of the assigned requirement, is Kx 4X 4 Bet hat 0. by 7" oy ‘We have to obtain the relation between v and 7, and the relation between v and 9, each such relation possibly involving £. Now the relation between v and 7 can be obtained exactly as for the simpler ease discussed in § 33. We take the same figure as before. The curve EPN represents a current relation between a and y; the dotted curve ...PQ... represents a terminal relation between them. In that figure, QN=«n, SP=QR=nr, PN =x;86] ‘TERMINAL CONDITIONS 123 and we have eps coty’ xb, that is, ven—yé ‘The same diagram, with a different significance for ordinates, leads to the relation between v and 9’, We now take Y=, and regard the curve LPN as a current relation between # and Y; the dotted curve ...PQ... now is a terminal relation between « and Y, that is, between and y’. For this representation, QN =variation of ¥ along PQ= en’, SP = QR = variation of the curve EPW along the ordinate = xv’, PN=cé, RN = PN tan RPN =&. Y= Ey"; and thus 1 =QN =QR+ RN =e + by’, that is, van —y'é With these values for v and 2’ in terms of 9, 7’, €, the terminal condition or da yok werol aay) te ay” for the integral, becomes {oF d (aF OF ‘OF , OF ty ay lay) But the only condition, representing assigned requirements, is & x ax e+ Mn Hence we have, as ie ra fs requirement at the extremity, ~9 ipa) -8 ap B-£@)-08) OF _ = together with the terminal condition XG YH ¥)=0, that is, after the elimination of 2, there are three equations associated with the extremity of the range in question, (ii) Let an extremity of the range be required to lie upon a given curve h@ y)=0, and let the direction at the extremity be fixed.124 ‘TERMINAL CONDITIONS [ex m1 Because of the latter requirement, we have y' = given constant, so that v’=0 at the extremity: and the terminal condition becomes oF d oF wre alr} s[P-9 mae Gel] + by ae Gp) The condition, representing assigned requirements, is now that is, } and therefore, as requirements at the extremity, we have (OF d (aF))_ ah P-¥ lisa) lay’ da \ay")} Moe ar _d 2) | Sy ae (ap) 5p where » is merely a multiplier; together with the terminal conditions h(a, y)=0, 9 =given constant, (iv) Let an extremity of the range be required to lie upon a given curve R(@, y)=0, and let the direction at the extremity be unconditioned. ‘Then v’ is arbitrary and variable. The conditions at the extremity are roy Eo (Z)} mo lay aay ae aF_ a (aF\_ ak -E)--8). oF ay” together with the single terminal condition k@ y=0. 87. In every special kind of terminal relation, but with forms that vary from one kind to another, we have three conditions surviving at an extremity, and therefore six surviving conditins in all. As will be scen immediately, the primitive of the equation @ =0 involves four arbitrary constants. When any freedom is left to the position of an extremity of the range by the character of the imposed conditions, both the initial limit 2, and the final limit 2, have to be determined, as being quantities initially unknown, ‘Thus, when the greatest freedom is allowed by the assigned conditions, there are87] CONDITIONS FOR INTEGRAL OF GENERAL ORDER 195 six constants to be determined. But we have six relations from the terminal conditions ; and thus all the quantities are potentially determinate. When the assigned conditions are more precise and the limits 2, and ay are actually known from the beginning, the four constants in the primitive are otherwise obtained (§ 111). And so for the various forms of condition that occur: either they impose limits, which initially are definitely fixed ; or they lead to relations, arising out of imposed demands and definitely determining the limits as known, Thus, in order to secure that the first variation of the original integral shall vanish for all admissible variations, we obtain a critical equation @ = 0 and we have conditions which determine limits. ‘These limits, henceforth to be regarded as known and fixed, admit no variation. ‘The way is thus cleared for simplification in considering the second. variation of the original integral The corresponding results for an integral, which involves a derivative of order higher than the second, are given in the following example. Ex1, Prove that ifan integral Pde of order m, where Vm FH IY oooy YO) and ©) denotes 2%, bas a maximum or minimum, then y must satisfy the equation a far ea eae Bp ae lig + de gah — + (-NSE fa ee ee Pay (2595.9 YMA 96H, a M. Writing a f Pena ds (ggornh t= (rr oo) for r=1, .. m, prove that, if an extremity of the integral lies on a curve F(x, y)=0 and bbe displaced to a position +4, y-+e¥ upon that curve, then the condition {Py Py Pym... fo) Ph X FPYEPIU4.. + PAPO must be satisied at that extremity, where OF yak xe er Feo, Fy PPL, (PF PP ay yw ik Ry PPV ay (OF yy PPL yy pe (Beton Sa} + fate ae} OH and so on. Prove also that, if V is the exact differential coetfoient of a function of 2, 95 yf,.., yo", then #=0 is the condition of integrability: and that, if V is the exact second differential coefficient of a function of «, y, ¥® ..., y*~%, the additional condition av yd fav) 4 @ fav ey z {ja aa wot ‘must be satisfied.126 NORMAL FORM OF SECOND VARIATION (eu. ur Ee. 2. Prove (a) that, if the function V in the preceding example does not involve y explicitly, « first integral of the characteristic equation is P=4, where A is an arbitrary constant; (b) that, if it does not involve » explicitly, a frst integral is Ve B+ yO Py $y Prt nt 9 Pas where B is another arbitrary constant: and (2) that these two integrals are independent of, fone another, so that they coexist, when V does not involve = or y explicitly and does con- ‘tain derivatives of y higher than the first, Is the latter statement correct, when V contains only the derivative of the first order ? ‘The second variation, under special weak variations. 88, We proceed to the consideration of the second variation of the original integral; but now it is necessary to consider it only in the cireum- stances established by the vanishing of the first variation. In particular, the limits of the integral are fixed (§87). ‘Thus the variation of the integral is [re yr eng rey" +e FG yh VI de =} fi (Fae+ K, where SH Out? + age + Laon" + dy? + Behav" + de, in which the six coefficients dy, dy, Gn Qh, ray Gea Tespectively denote OF OF OF OF OF OF. ayh? ayoy” Bay” ay”? Byay” Oy and where K, denotes the aggregate of terms arising ont of the third power and higher powers of « in the expansion, The term involving the first power of « does not appear; it is the first variation of the integral and has been made to vanish. The governing magnitude in this variation is the ‘second’ variation 4+ [fFde, unless it vanishes, We must, therefore, consider, in detail, the integral f [$ae: and it will be noted that Jf is the same function of », v, v’ as 20, connected (G85) with the subsidiary equation, is of 9, 9’, n”. 89. For this purpose, we take a quantity U, homogeneous and of the second degree in v and v, choosing it (if possible) so that on( H+ ®) (ag0" + lo! + mop, where J and m are magnitudes also to be determined.89] AND SUBSIDIARY EQUATION 127 In the earliest discussion ($19, 21) of the second variation of integrals originally involving derivatives of only the first order, the squared quantity in the modified subject of integration in the eritical integral was such that, when v was made a constant multiple of the associated integral of the subsidiary equation, the quantity vanished. By analogy, we are led to con- sider the possibility that the two quantities ! and m should satisfy initially the sole condition dgx” + In’ +m =0, where 7 is any integral of the subsidiary equation @& an + ae Gp) = Other conditions, to render J and m precise, will be imposed; they need not be stated at this stage of the investigation. On this possibility, we have an oe pyr ee + aa + at! == Or — $n, where @ and ¢ are two quantities dependent upon l and m by the relations T=a,+0, m=an+¢, that is, @ and ¢ are not new unknown quantities, additional to 7 and m. Next, ao _ d aa by da a) dan + onn + aaa’ + 2 (0x + $n) = (dnt 8) 0" + (an +8 + $) 0+ (Gut $')0 (a +0) + (A+ +9 E) a (aato =- orn Pada (an+8'+2$), Imman (ant $ +4): the former of these equations is a second relation (apparently) affecting Uand m, while the latter defines a new quantity yr. And now a0) , @ (22 6) = 460 + an +d" +L (bn! +0) = (n+) 0" + (an +8 + Ya! + (ant) a if wim, n = ma a + (dot vy T+ Got¥)0 = gg (tn +l’ + mn) + (ov1238 NORMAL FORM OF SECOND VARIATION ox. m As the left-hand side vanishes and as 7 is not zero, we have mM? = Ae (deo + Y")- It therefore follows that, if » denote any integral of the subsidiary charac- teristic equation, and if J, m, 8, @, are such that aan" + ly’ + mn =0, L=ay+8, m=dutd, Pada (tn +O +24), Im=an(ant 6 +4), m= An (d+ ¥), then (a0! + lo’ + mv) = gb"? + Lota (dy, + 8) 0"! + Qty (dn +h) 00 $a (ty + 0+ §)8+ te (y+ BP) UH On (Oy) a =an (# — @). where U = Ov + Que! + yor. Determination of two quantities, for the expression of the second. variation. 90. Thus far, we have considered only a single integral » of the subsidiary characteristic equation in connection with the two quantities and m; and no limitation has been imposed on the integral. We proceed to prove that all the determining relations affecting 1 and m (and the quantities 6, 4, y+) are satisfied, when two distinct integrals of that equation are taken, subject to a single condition connected with the subsidiary equation itself Let two such integrals be denoted by a and 8, so that we have dna’ + lal +ma=0, an8" +18 +mB=0. In order that J and m may be determinate, af’—a/8 must not be zero, a requirement satisfied because @ and 8 are two distinct integrals such that 8 is not a mere constant multiple of a. In order that Z may not be zero always, we must have @8”—a"8 not always zero, s0 that af’ — a8 must not be a constant. And in order that m may not be zero always, we must have a8” — a6’ not always zero, so that we do not have 8 = ka+ h, where h and k are constants: in other words, the subsidiary equation must not admit any single integral that is a mere constant, In connection with the equations for Land m, it is convenient to write Gy Ld — yy =, yy — On’ + en” and then the subsidiary characteristic equation is aai9"” + Qala" + (da — b) 9” — Bn! + 09 = 0, 0 that the exclusion of the possibility @=ka-+h, where a and 8 are two integrals of this equation, merely requires that ¢ must not vanish, clearly not a limitation of a general kind.92] AND TWO INTEGRALS OF SUBSIDIARY EQUATION 129 91. First, as regards the indicated condition connected with the equation itself, The equation, being linear in the dependent variable » and of the fourth order, possesses four linearly independent integrals: and the primitive is a linear combination of these four integrals with arbitrary constant coefficients. Let a and 8 be any two of the four; and introduce new quantities zand p by the definitions z= af'—aB, p=a'p"~a'B. 2 = af" —0"B +p, f= aB""—0"B 4%, wad B”—a"®, wade" aa + a'8" —a"p. One property, to be used immediately (p. 131), may be noted in passing. If ¢ denote "8" ~ a8", wo have ‘Then one's" 28", awe" ~2"8, and therefore 1B oes 2 way B monte (er so that beet =a" Bm pd pt — pa", As a and are integrals of the subsidiary characteristic equation, nf" + 2ag'8" + (da," ~b) B"- UB’ + 08 =0, age” + 2ag'a” + (dn” —b)a” — Ba’ + ca =0. ‘Multiplying the first by a, the socond by 8, and subtracting, we have yy (2 ~ 2p!) + Ltn’ (2”— pw) + (des — b) 2’ —B'2 = 0. Thus : ag Ite (2"— 2a) + age — be} = and therefore Gz, (2" ~ 2) + y's —be =O, where C is a constant, which manifestly is specific and not arbitrary, because all the quantities on the left-hand side are specific. 92. We have taken a and 8 to be any two linearly independent inte- grals, out of a complete set. We now proceed to prove that, if C be not zero for any two initially selected integrals, two can be chosen so as to make the specific constant zero. Let a and y be a pair with 2, pn, C, as the values of 4, 4, O; and let a and 8 be another pair (where 8 is distinct from y and is not a linear combination of a and with constant coefficients), this second130 DETERMINATION OF COEFFICIENTS (ow. ur pair having 2, #2, O, as the values of z, u, 0. (It is assumed that neither C, nor O, is zero; for, if either were zero, the desired combination would be to hand.) Then yn (24” — 2ptz) + Oh’ 2y' — 2, = Cry a 44" — Bye) + 6 ~ Be, = One B=Cry— 6,8; Let then 220.2" Qz,", B= Orth — Cine» and therefore a, (2"— 2p) + a9/2'— be =0. But 8, thus chosen as a linear combination of the two integrals y and 5 with constant coefficients, is itself an integral of the characteristic equation ; and it is distinct from a, because 8 is not a linear combination of a and ‘Thus two integrals @ and 8 can always be chosen, so that the specific constant. (is zero—which is the property in question. Moreover, the choice manifestly is not unique. 93. We now proceed to shew that all the equations for J, m, 8, , yr are satisfied by the equations aud’ +lal+ma=0, dn8" +18’ +mB=0, defining J and m in terms of the two integrals a and 8 with the property just established. From these two equations, it follows that Z # tae, madat. ‘The other three equations are Pa tin(dn + 0 +29) = dn (b+ +2), Im = dn (dy + 8+ Y)= toa lh Ol + +), m= (OY). When the values of 2 and m are substituted in the first of these equations, it becomes ban 8a. Fe ay which is satisfied because n(2"— 2p) + dy's— be =0. When the values of J and m are substituted in the second of these equa- tions, it determines a quantity ¥ in the form ww fomu8), ey ’ Hp ty al +} (aala tdag)s ‘that is, with this value for y, the second equation is satisfied.93] TN NORMAL FORM 131 When the value of m and this value of y are substituted in the third ‘equation, it becomes Og — On (Ct + apt’), ol — te! +S (a + Rat H+ ap”) ~ that is, oteee “ ee / om ta + (Oa wt ial + Oy") — F(a + det) Now, taking the two equations satisfied singly by 8 and a because they are integrals of the subsidiary characteristic equation, multiplying the first by a’, the second by f’, and subtracting, we have ule” —0"')+ 205/808") + (0-8) ('8"-0'8)4 0(0/8 a8?) =0, and therefore n(n" (a'8” ty (0B ~ a"B") = dy (u's'+ pu" — us”), always, whatever integrals a and 8 be chosen; and for the present choice, a"8")| + 2ay’y’ + (dn — 6) w— cz = 0. Now ogg2” = Degg fe — 2+ be 5 thus at (0"B"—a"B") = apt + apt — 1 (2dr — ant + be) tia (We— 1) + Onl na’ bye. ‘Therefore ape" (tao! 2 — wt) + ani — bys) + Dain + (an! D) ce =0, that is, on dividing by z, Pla ue e in Fat 5 Ont" + 2p + de") 5 (Op + dyin) — 0= 0. ‘Thus the third of the three equations is satisfied; and the five quantities 1, m, 8, $, are expressed in terms of a and 8 by the equations a Oe + Oa rl, n pat —a6' Haale + Ont’), where 2=af'—a8, p=ap’—a'p’,132 NORMAL FORM OF SECOND VARIATION [en mt and where the condition On (2”— 2) + axle'— be =0 is satisfied. 94, The preceding analysis can be modified and restated as follows. We have #= cae + bo + on S {abag0!? + Lag 0" + (Ay — dra’) v*}, where, as before, beady — ay — Lay, C= yy — dy + On”, Also Sos 2mov’ + ne) = Dei’ + Lave" + (U'+ 2m) v2 + (2m! + Qn) ov’ + n'dh; and therefore a 7 : o[$-% Haig 0) 08 + 2 (yg — m) v0'+ (dg) — On’ — oe} = 0,30" 2aglo'v" + 2aqmv0” + ag (b + U + 2m) v8 + Lata (m' +n) vv’ + dn (on) oF = (aav" +lv'+ mo, provided Pag (b+l'+2m), Im= dg (m' +n), = dn (0+1'). This is the earlier form (§89) of the result, when we take Aq— m=, aq yn = ‘The remainder of the analysis, for the establishment of the equations aga’ +la'+-ma=0, an8"+1s'+mB=0, with the relation betweon a and 8, is unaltered. ay —1=6, Normal form of second variation : the Legendre test. 95. It follows that Cs (# +2) = (dav + bo’ +e); and therefore, on inserting the values of and m, aU cn wv ol $e e-aBP| a’, a acoeeae =O,95) LEGENDRE TEST 133 where (a8 — oY =| v", vo, 0 |e oe 8,8, BI The quantity determining the second variation of the original integral frenyyyavia f spat, But [Pea 00+ 2g’ + pot], taken at the limits of the integral. As these limits have become fixed by the inferences from the vanishing of the first variation, we have v=0 and o’ =O at each limit; and thus [U]=0. Hence the second variation ate f. Fae <4 foaOde; and the whole variation of the integral is 40 fon de-+ K, where X is the aggregate of terms involving the third and higher powers of «. ‘Thus the integral foatide is critical for the second variation. If it is positive for all admissible variations », all such variations make the increment of the original integral an inerease: subject to any other conditions that may cecur, the original integral is a minimum. If it is negative for all admissible variations, all such variations make the increment of the original integral a decrease: subject to any other conditions that may occur, the original inte- gral is a minimum, Hence faaCdz is to be regularly positive or regularly negative for all arbitrary admissible variations, Then aq must have a persistent sign throughout the range. For, if a» were sometimes positive, sometimes negative, then by choosing »=0 when ay, is negative, we make fonOae Positive; and by choosing v= 0 when ay is positive, we make f aude negative; that is, the integral in question can be made positive or negative at wil, if au can change its sign through the range. Henee we infer, as a necessary requirement, that the quantity am must be134 JACOBI TEST: SUMMARY OF TESTS [en. ur monotonic: that is, zs must have a constant sign throughout the range of the integral along the characteristic curve. This test, an extension of the Legendre test, for the former ease, may be called the Legendre test. The Jacobi test. 96. But the quantity () must still be considered. Apparently, it is a positive quantity. The magnitudes a and 8, which occur in its expression, are linearly independent, so that no difficulty arises from the quantity @@’— @’8, or from the (excluded) possibility that the second and third lines of the deter- minant might effectively be the same. Tt vanishes of course when v= 0: in that event, there is no variation. Tt may however vanish if some variation » is possible such that through the range v= Aa+BB, where A and B are constants. The conditions now require that v shall vanish at the limits of the range. If, therefore, the range is such as to allow constants A and B to be chosen, so that Aa+ BB shall in general be distinct from zero, shall vanish at the lower limit, and shall vanish also at the upper limit (or earlier), then the non-zero variation v= Aa+B8 makes [) vanish. For that variation, the second variation of the original integral [F (e, yy, y’)de vanishes: we cannot assert the existence of maximum or a minimum, without examining the quantity K. If, however, the range of the integral along the characteristic curve, beginning at a lower limit where a quantity a+ B8 vanishes, does not extend as far as the nearest place where that quantity again vanishes, then the indicated variation cannot occur. There then is no non-zero variation » which annihilates []; and the former argument now remains valid. ‘The further significance of this result will be indicated later (§ 124-126), when we come to consider the conjugate of the lower limit of the integral as fa point on the characteristic curve. Meanwhile, we have the initial form of the Jacobi test limiting the range of the integral. Summary of tests. ‘Summarising, we have three tests, thus far: (i) the characteristic equation (the Euler test), leading to the character- istic curve; Gi) the non-evanescence of es along the range (the Legendre test); (iii) the limitation of the range (the Jacobi test).97] (GENERAL WEAK VARIATIONS 185 General weak variations : the Weierstrass method. 97. We now pass to the consideration of the more extensive type of variation, still required to be ‘weak, wherein both the dependent variable and the independent variable are subject, separately or simultancously, to small unrelated changes. To facilitate the representation and expression of such variations, it is convenient, as before (§ 36), to make @ and y funetions of another variable t which is parametric along the characteristic curve. Initially at least, this new variable will not be restricted to represent any element of the curve 0 esten- tially intrinsic as the length of the arc from a fixed point. The introduction of the variable ¢ changes the expression of the integral itself and substantially modifies the analysis; a reversion to the previous form can always be made, 80 far as some results are concerned, by adopting as the independent variable t. We write for integer values of m and of n, while a, is # and y, is y. To express partial differentiations of any function @ involving 2, 2), 2.5% Yis Yor ons We write ee ace Bnd“ Deg, for m, n=0, 1, 2; so that there are six coefficients © (because Cun = Cyun)s and six coeflicients + (because Yan=an). Also we write ee Pra Foun for m=0, 1, 2. and n=0, 1, 2; so that (as ky and kym are not the same, when ‘mand n are different) there are nine coefficients k. To transform the integral [F (w, y, y’, y”)de, we have I y= i (erye— me); and we write Ga, Wy, oy Ys Yas Y= OF (a Y's Y") =aF (« ” so that Gis a function of the same character in its arguments as the original fanction F. The integral becomes [o Cas a4, 205 y, yo wat136 IDENTITIES SATISFIED BY INTEGRAND (ex. m We assume that ¢ increases steadily throughout the range from the lower limit to the upper limit of the integral ; thus does not become infinite. The function F is a function of only four arguments, while G (as expressed) is a function of six arguments. Thus G cannot be a quite general funetion of those six, and therefore the limitations on generality must be indicated explicitly. We have a@ Ha 2s _ Sym) OF $n p03 4 etme) a6 af _ a oF oy oy ay” oa aa, eas ys By climinating @*, 9 and F, using the relation @ =. F, we have two (and iy 8 Oy By ie only two) identical relations satisfied by @; and these, which will be called the fundamental identities, can be taken in the form* a, 2G ag Gans ty, +2528 2H Fy, ag, | aG : O= ae, + Hoy, ‘These two identities are the expression of the limitations upon G as a fanetion of its six arguments, Variation of the integral : the frst” variation. 98, To obtain the conditions for a maximum or a minimum of the integral [Gat the variables 2 and y are subjected simultaneously to small variations so that they become #+«u and y+xv, where « is the usual small arbitrary constant, while u and v are arbitrary functions of f, regular within the range * For another mode of derivation, based upon the fact that the fntegral { Gat, which (as in § 88) is an invariant for changes of ¢, is to be unaltered by infinitesinial changes in t, see § 105 pot. tis to be noted that if, instond of rogarding the two identitios as relations satisfied by every given function G arising as in the toxt, the two rolations are regarded as simultaneous partial differential equations satisfied by an unknown variable G, the most general primitive of those ‘ovo equations is where, so far as concerns the two differential equations, F denotes the most general arbitrary fametion of its four arguments.98] FIRST VARIATION OF INTEGRAL 187 admitted, and independent of one another along the range. The values of t, which are the limits of the integral, are taken to be fixed, as in § 42, Tho new value of the integral is [Coren 2. ten, 2,44, ybHD yb, yb ende Hence the increment of the integral is cl +40 [Gd K, where ae, t Maa, *” By tay, * where K, denotes the aggregate of terms involving the third power and higher powers of «; and where @ is the combined aggregate of all the terms of the second degree in w, 1, t, 8, %, 0, via. I f(u2 Boy 26, 26, 06, 20, 8) as G= cat? + 2am + Pee + yt? + Lontity + cnt? pet + Bate + Dy + 7168+ 2a e+ pst + LUkyuy + Bky ur, + By UYy HBr 0 + By tv + Baty ty + Degg + Ut + Byte Proceeding as before in § 83, we have ag og a@ Gd (0G) ag [(ege regi ge) a= foxaes [u fe- au(ge)| +3) oe ag ae IG dG og iG By tat sie) a= foras[o Na 2} +032], where aG_ da (a@ @& (6@ (Se) +a X= Hla, a@_ dG), a 0G a oy dl Sn) te Gy) and the terms outside the sign of integration are taken at the limits of the range of the variable t ‘The customary argument concerning the aximum or minimum of the integral requires the total increment of the integral to have one uniform sign, persistent through all the possible small variations which ean be imposed ‘upon the variables # and y. Consequently the first, increment J,, containing only the first power of « which can be changed at will, must vanish for all such variations,138 ‘TWO CHARACTERISTIC EQUATIONS (cu. ur The two characteristic equations. 99. Now J, consists of two parts: an integral, and a set of terms at the limits. Consider, first, the possible variations which, arbitrary along the range, are chosen to be such that 2, w, 0, # are zero at cach limit; thus, at each limit, the position and the directions of # and y are kept fixed. For these variations, the limit-terms vanish ; and J, reduces to the integral fox oY) at. Consequently, to achieve the purpose, this integral must vanish ; and, in it, the quantities u and v are arbitrary regular fanctions of z, independent of one another. The only way, in which this requirement can be met, is by having the two equations X=0, Y=0, satisfied everywhere along the range. Otherwise, by choosing w positive or negative as X is positive or negative, and choosing v similarly positive or negative as Y is positive or negative, we could make the integral positive: and by choosing u negative or positive as X is positive or negative, and choosing » similarly negative or positive as ¥ is positive or negative, we could make the integral negative. These possibilities are to be excluded; hence the two equations X=0, ¥=0, must be satisfied. They are the characteristic equations. It will be noted that, in their expression, the variable does not occur explicitly. Conditions at the limits. 100. Assuming now that these necessary equations are satisfied, the integral I, for all other variations reduces to the aggregates of terms at the limits. But variations of « and y are possible, vanishing at either limit though not at both, When these are taken for the limits in turn, we infer that the magnitude “fea Geld esata aah eae ‘must vanish at each of the limits of the integral. It is convenient to conclude, at once, the consideration of these conditions, ‘They require that the equation 2-4 CB} rule -4 CQ} ndl-o shall be satisfied at each limit. As before (§ 86), various analytical conditions emerge according to the character of the assigned data,100] ‘TERMINAL CONDITIONS 139 (i) Let the extremity be rigidly settled, so that 2, y, y’ are not subject to variation. Then u=0,0=0; the variation in y/ is wt _ Eat ete oe ©, (@, + €u)" 50 that, as this variation is to be zero at the limit, we there have myth = 0. In the foregoing equation, we take u= 0, 0=0; the other terms are a@, , aG “Ge, t ay, =hen-nm, ‘on account of the identity a ug =o. ‘The terminal equation is satisfied without leaving any residual condition. (ii) Let the quantities 2, y, y’ be subject to a condition H@eyy)=0 ata limit, Then the elements of the small variation at that limit are subject. to the single equation oH oH at? yt dens win keeping only terms of the first pa in x, as we are dealing with the first variation. But the terminal condition for the integral is “fea Dh fina Gh Sermo, and there is only the foregoing linear relation connecting the otherwise arbitrary quantities u, », 2% —yith- Hence there is some quantity d, such that aG_d (ae oH) Se ai Ge) = aG_ d /a@ 0H | By. 7 at (oy,) = dy ' az. oH Gy. together with the equation w H(ay, 2 0, in effect, three equations when 2 is eliminated,140 ‘TERMINAL, (ow. mm Gil) Let the extremity be required to lie upon a curve A@y=0 and be rigidly fixed in direction. ‘Then, at the limit, we have we eotno, an—nm=0; and so there exists a quantity 4 such that, ane) ~*35 2G _a ®) = pat ay, at \oy,) ~ Yay together with the equations h(a, y)=0, 4! =given constant, (iv) Lot the extremity be required to lie upon a curve k@y=0 and be perfectly free in direction, ‘Then the arbitrary quantities u and v are subject to the single condition ak, al “Set Gy while the arbitrary quantity 2,0, —y,2 remains unconditioned. ‘Then a _a (2) Bh ae, di (ae) =? ae} a _ a 0G) ay ae Ge : 0 together with the equation k(a, y)=0. In every type of boundary condition, except the first (where all the variations are zero at the limits), we have three conditions at an extremity, amounting to six in all. ‘Their significance is the same as before: they serve to determine the constants in the primitive and (by means of the values of ) the actual positions of the extremities, When these are thus settled, no further variations at either extremity are possible; we then may regard them as definitely settled by the first variation. This settlement completed, we can consider the second variation of the integral for all variations which are subject to fixed extremities, that is, which are such that u, ¥, 20%- y.th are zero. It will be convenient to con- sider a quantity w= qua yu,101) CONDITIONS: wu alike for the current variation and for the terminal variation; as eee it follows that, at any fixed extremity (and we now consider each extremity as fixed), the quantities w and i vanish. 101. These results, and the results obtained (§! 86) for the special variation, can be harmonised as follows. We have Ganka yyy") a (ye ne), so that nek age at ey”” dy, am and therefore 0G _aF ; ay” Again, ay, ay ~ aay” 5-08) 14 (aP)\_%aF 2a ay) Sy 4 (28) OF, wa lap) say and therefore 204 (0)_an_ a (ary iy, 7a (3y,) ay? ae Gy : Next, 6 e ag at a Gans ange tins as og ag = ys omy an, * May, Also, @=2,F; hence ¥ ag , OF ae P-¥ 5p ay Again, and therefore OG a yhence 142 CONTINUITY OF FOUR MAGNITUDES THROUGH (ce. ur (Gd (a » OF fey (yd) “Yap -aGe) -1-1 BEB) 78 When these equivalent values for combinations of derivatives of @ and com- Dinations of derivatives of F are used, the two sets of terminal relations are ‘seen to agree for the corresponding types of terminal condition. Bz.1. The integral [ro de ¥¥'.¥") de is transformed to an integral [PGs mH ge by changing the current variable to so that both 2 and y are functions of , Prove that the function F satisfies the three identities OF fon FF yy, OF. Fe one tng Hing Mea Rey Syn oma, Fane th aF On Ga yg Bx.2, Prove that if the preceding integral f(s 21220 9,2 ds 28) dt sto be isfy the two characteristic a) +e) an (e)=9 z -E£+8 & ae -3@) ere ae ee rae of sng oqstin which he alo ya & fF (FY 5a) tae as G)=% Find the terminal condition (or conditions) to be satisfied ata free limit; and identify it (or them) with the result stated in Ex. 1, § 87. ‘maximum or minimum, # and y as functions of ¢ must si equations Continuity of four magnitudes through a free discontinuity on the characteristic ourve. 102. One further inference (similar to the proposition in § 47) may be derived from the consideration of the terms at the limits of the original integral, as follows: At a free place on the characteristic curve where, while the are of the curve is continuous, there is discontinuity of curvature or of direction or of both curvature and direction, the quantities ag ag a@_a aa, ~ dé \ax,)* Bay" dys db are continuous in value. a@ ” Ot102] A FREE DISCONTINUITY ON CHARACTERISTIC CURVE 148 Let P be such a place*, given by a value T'of t. Let Q” (for the value ¢”) be a place immediately beyond P, and Q (for the value 1) be a place immediately behind P. Consider a variation of the characteristic curve AQPQ'B such that wand 1 (as also v and 1) are zero along AQ’; are zero along Q”B; and along Q'PQ’, usE-tF-9 UW, va(t-tPe" -HPVO, where U(t) and V(t) are arbitrary regular functions of t in the vieinity of 7, These functions U() and V(t) are not zero, because the place P is supposed to be free (that is, not rigidly settled by assigned data); also the two functions are independent of one another. For variations of this type, continuous along the whole range, the first variation of the original integral along AQ is zero, and is also zero along Q'B; because both w and v are steadily zero. ‘The first variation along the whole path is to be zero; and therefore the contribution by the portion QPQ' must be zero. Thus we must have the relation s i (uX +0¥) at ¢ (ag d /a@ ag : [fo a e)} mg es + f (uX +0¥)dt 7 2, @G_ ad AG) 20)" et es (eh +nze]° 0. Along Q’P, we have X =0 and '=0,and likewise along PQ”. Hence the integral in the first line vanishes, and likewise the integral in the third line, For the particular variations selected, u=0, u4=0, »=0, .=0 when tt, and also when t=¢". Hence the terms at the lower limit in the second line vanish, and the terms at the upper limit in the fourth line vanish. We denote by 6 the value of a quantity @ in the range Q'P in the limit when t approaches 7’ from t’ along QP; and we denote by 0} the value of the quantity @ in the range PQ” as ¢ approaches 7’ from ¢” along QP. The foregoing requirement is “fern de Gadh + Ge). Ged ** [Hein a Gh. + Ge), +s Gd}, ts + in sn hair pnemy sania os et ad other intrinsic properties, because the functions (1) and y(t) aatistying the characteristic equation are assumed to be analytic.144 FOUR SPECIAL CONTINUOUS MAGNITUDES [ca. ur and this requirement must be met for all forms of the functions u and » specified, the quantities u, 1, 0, », acquiring their values at the place 7. First, let v be a persistent zero through the range Q'PQ”, so that V(¢) is zero for all values of ¢ in the range; thus »= 0 and »,=0 for this selection. At the same time, let the arbitrary function U(t) be chosen, so that, U(=A+ Bt-T)+(t- Tb, where $ (¢) is a regular function of ¢, and A, B are arbitrary constants at our Gisposal ; thus, for this selection, U(L)=A, ([U' Oher=B. ‘Then, at 7, u=(T-tp(t'- TA, nara one Te fa HEED a Bhs =H e=7)* and the condition gives ~ [fia Ge} faa Ct ] +» (@)_-G).] -° while A and B are arbitrary constants. When, as one choice, we take 2(t"+¢ 27) B+ Ame T=” the condition can only be satistied if eG d @ (eG _d (ag fea G)}- fe 2). ‘Thereafter, for any other choice of B, the condition can only be satisfied if Gz). = Ge), Secondly, as the terms involving w and x, now disappear from the con- dition whatever be the choice of u, the condition now involves v and », alone. By a corresponding choice for the function V(t) in the assumed form of v, we have: aG da Gy) _jaG_ a ae ton dea = lage Gh ag ). Note. (a) If however the place 7, where the hypothetical discontinuity may occur, is a definitely and rigidly fixed place, that is, fixed in position and with assigned direction at the position, the inferences cannot be drawn, For then we have v= 0, a%—yn=0, at 1; and the requirement is satisfied without any residual inference.103] SINGLE EQUIVALENT CHARACTERISTIC EQUATION 145 (8) Again, if the place 7, where the hypothetical discontinuity may cecur, is definitely fixed in position without any condition imposed on the direction, at such a place while «x, —y,u, is arbitrary. ‘The requirement is »()_+% Gi)» GS), +96, ag ae Bigg, + jy, 9 But we have along Q'P up to P, and so #(G5).*» Ge =). * Gi. ‘Thus the requirement becomes sen wo{@) -@Z) Joo while 2,0, — yu at 7’ is arbitrary. Hence G)_~ Ge, 2) -@,. (7) But when the place 7' is completely free, all the four quantities specified in the proposition are continuous, and similarly and thence The two characteristic equations of §99 are equivalent to a single equation. 103. Before the consideration of the second variation of the original integral, we return to the two characteristic equations X=0, Y= It is assumed that they determine « and y as analytic functions of the variable ¢, When ¢ is eliminated between the equations giving the values, we have a relation between z and y which is the equation of the characteristic curve, Now t is a parametric variable, not essential so far as the curve is concerned ; and therefore it may be expected that the elimination of t would leave a single characteristic equation. It is possible to prove that the two equations are, in fact, equivalent to a single equation, though, for subsequent analysis it is convenient to retain both of them.146 SINGLE CHARACTERISTIC EQUATION fou. mm Because G is a function of 2, a, 25 Ys Yi» Yay We have aG_ 0G, GG, ae, GG Ge ag 5a, tgs, tH 5g + Vey, t Sy, . dG) Pay, ag, aa wa {x+a()- deli} aet ee dg) @aG)) ag, ag +of¥ +a) ae Gol Hay ta, = — 2g OF, 4 =aX+n¥ aol" 5n* 5p a/, 0G, ag, a, a +i (® Ge dy Es eG, and therefore, owing to the two identities 2G 2G, 9 2G, 4 ae Gam, 4 yO 4 20, SS 4 24, ia, tag, + rag, * dy, ag, ae : Ommae thay, it follows that aX +yF=0. ‘Thus we can have YeoE, X=-y5; and there is a single characteristic equation H'=0, the explicit form of which is obtained later (§ 105, p. 151). But, as already stated, the simultaneous equations X =0, ¥=0, may prove more useful than the single equation E =0. The identities (597) satisfied by G: invariantive forms. 104. At the beginning of the investigation, a non-intrinsie new inde- pendent variable ¢ was chosen, in terms of which and y and their derivatives were to be expressible; and there is convenience in having some non- intrinsic variable, so long as the characteristic curve is unknown. But when ‘once the curve can be regarded as known, and its properties are to be derived from its known character, advantage may accrue from the choice of the arc- length (that is, the length of the are measured along the curve from some fixed point) as the independent variable. ‘The forms, that result from the choice of s, may be regarded as canonical, Meanwhile, let another variable @ be taken; we write dy my, pean n= tags my oh104) IDENTITIES SATISFIED BY THE INTEGRAND 147 where de de ‘The forms of F(e, v.42, $4) de, after the respective changes, are FY Ts Day Be, FAO, OCH, Ys yy Yrs te, yp) at, which are necessarily equal to one another ; hence G ©, Ys Bis Has Fas Hr) = ME (GY Hs Ys Wey Ys) 5 and therefore a a G (2, y, me, my, ma, + may, mye +m BN) AG (0.015 94 2 Ye Here ¢ and @ are arbitrary variables; by appropriate sclections some useful inferences ean be derived. (a) Let m=—1; then EY — 21, — Yrs a, Y= E(B YB, Ys Bas Ys a result that will be usefal when strong variations are considered. @ Let m=1+e, where ¢ ia a small arbitrary variable quantity such that 4 is a emall quantity of the same order of magnitude as ¢. ‘Then =(L +6) EO], ya, Yrs te, He), where the unexpressed terms on the left-hand side are of the second order of small quantities, Expanding in powers of the arbitrary small quantity e and de de O(a yee ite at dente Ee yet ey ne + of a. and equating the coefficients of ¢ and of : ‘on the two sides, we have aG G4 AG, ae de, tH ay Peat Ha ag, ae Onn at hay, Ge. the fundamental identities satisfied by G. Thus the two fundamental identities are @ partial expression of the invariance of the differential element GG Ys ry Ys ay yo) at for all changes of the independent variable t. (©) When the characteristic curve is such that the independent variable may be taken to be the length s of the are measured from a fixed point, and we write dz_ dy de an ash as we have Ge yay cy yV=tC (a, y, a, Hy te He);148 RELATIONS AMONG (ow, and the fundamental identities are 26 4 28 99" 2 5 947 2G Gna ty wee Bry's 28 : ona +936 together with wtytel, ve tyy’= Ex. Let 0-6 (96 Fi Tu Fy THOS 4 By Yo Os so that : oan? we that — de da er _ d 2 @ (or ea. + aa ie)" fea) +a Ge 2- % e+ 2@)- ne a, +a Go} ‘Thus X and Y are invariantive under chango of the independent variable. Relations among the second. derivatives of G. 108. We now proceed to consider the integral foea in the second varie- tion of the original integral. ‘The variational quantities w and », which occur in G, are arbitrary regular functions of t. As the limits of the integral are completely specified by the enforced vanishing of the first variation, w and» (as well as their derivatives) vanish at the limits. Let (J denote any homo- geneous quadratic function of 1, 2, %, t, so that [) must vanish at each limit. ‘The integral in the second variation, which can be expressed in the form [e-®aven is equal to the integral 1G D)a. ‘The quantity [] will be chosen so as to simplify the expression under the integral sign, though it does not affect the complete value of the second variation. In securing this modification to a normal simplified form, special account, must be taken of the coefficients of the combinations of u, v, u’,v, u", 0” which oceur in Gi. ‘These coefficients (as defined in § 97) are not independent of one another; certain relations subsist among them, owing to the two identities satisfied by the function G. As in §51, a knowledge of these relations is a convenient preliminary to the reduction of the sepond variation. (i) From the identity @ y ae atten 8 o,105] SECOND DERIVATIVES 149 partially differentiated with respect to « and y, successively, we have Mat Yka=0, tha + Yrla=0. We therefore may take a quantity Q, such that a= 27Q, ke=- 472 Y= It is easy to verify that or aiQ= ay ‘Thus, as 2, does not vanish or become infinite in the range, because « in the original form and t in the present form increase continuously, Q reverts to the quantity a, (of §88) when # is made the independent variable. (ii) When the same identity is differentiated partially with respect to a and y, suecessively, we have az, t Ma + Yika=0, Set akatucan0. When the other identity : 26 5 95,06, 94,98 Gm Ban, tH ay, + Bes * Gy, is differentiated partially with respect to 2, and y, nccessively, we have iG 0= Fo pat gk + 2a + 2k, a + ikea t the + Qaigkeg + Wrenn Combining the two first equations of these two sets, we have bu ~ kin) + 2 (tas Q — @y2Q) = 0, Keg + 20 92Q = key + 2eay.Q. The same result follows from combining the two second equations of the two sets. Accordingly, new quantities Z, M,N are taken, such that m= L + 2y.yQ that is, Jg=M—2eyQ Fa = M -2y,2,Q Oy = + 22,0, With these definitions of Z, A, N, we retain the two relations ag My,=-28, =- La, + My, 3m Mx, + Nj Oy, and then fall account is taken of the four derived equations. Gil) When the identity 26, 2G 5 2G, ag Gage tH 3g, + eege +2150 RELATIONS AMONG DERIVATIVES OF INTEGRAND (cx. ur is differentiated partially with respect to 2, and y, successively, we have O=mynt tikn + 22+ Woloa =, (ym — 4y2Q) + ys (kn + AtayeQ) + La, + 2Mys, Oy (kn + Aaya) + 1h (Cu — 40°Q) + 2M + 2N ye. We introduce quantities R, 8, 7, defined by the relations m= B+ 4y2Q — dey | 5 + 402Q these new derived equations become Ra, + Sy, + 2Le, + 2My,=0, Se, + Ty, + Mey + Vy, (iv) When the identity is differentiated partially with respect to « and y successively, we have Yat thka=0, taka + Yr0u= 0. We introduce quantities C and D, defined by the equations he=—mD |, oo ale but sometimes it will prove preferable to revert to the actual coefficients ‘Yas Faas ns Com (v) Differentiating, completely with regard to ¢, the relation aa Be, La, + My, obtained in (ii) above, we have ~L+ ye sab+yl= 22) =a, Yat yokat ets + Yalen + y+ Yoke = ayy O— ye + aL + Yo + ys (ree — ays) Q5 that is, on the substitution of — (Ra, + Sy:) for 2a, + 2y,M from (iii), a{R-L— (0-0) + (SM + .(D=24Q)] =0. Proceeding similarly from the relation Ma, + Ny, also obtained in (ii) above, we find a {S— M’ +a, (C— ysQ)} +9 (2 — NW’ — a, (D - #,Q)} = 0.105) CHARACTERISTIC EQUATION AND CURVE 1oL ‘The two relations, thus deduced, shew that a quantity J can be taken such that ytd +2y,(C—y.Q) ayT— %(C—yQ)— n(D-mQ) >. aid 4+ 2n,(D—a,9)! (vi) Differentiating, partially with regard to y, the identity Ga a 28 + 9 28 4 20,97 4 04,28, we have : Ba thiet seu + akin + Wyle Also 4 ac Fe (Gpe) he ta alin + tnt ln @ Gy a ae G@ = Gy Cikea + Yaa t dake + Yorn t Saka + oC). Hence, writing % Jey! and similarly for the other quantities, we have aG_d (2G), & (26 Yoo Gy) * ae Gy = 2 (hig — hn + He) + (Dean ~ ha + Hen + ey!) + 205 (— a + Ha + He!) + i tna + Yo (Zen — Cn + Cen + Or!) + Yom + Yaron When substitution is made for the various coefficients cim and kim in terms of the quantities that have been introduced, we find Y=, where B= heyy — eg + Ke’ — hee! +n — Gays + Ye) Q+ (iys— mys) QO + (Hiya tays) J. Proceeding in the same way to a similar expression for X, denoting tei Gn) * de) we find Xa-yE. We thus verify the result, already (§ 103) established, that the two equa- tions X=0 and Y= 0 are satisfied in virtue of a single equation, in virtue of the identities satisfied by @. This single equation is E=0; it is the characteristic equation (constituting the Euler test). Its primitive gives, in effect, the characteristic curve. ‘The characteristic equation can be expressed in the form fea, — Kea’ — Q (rye 2s) — Vays — Tory = Kia ken! — Q(yreta~ yates) — Qt — Tyree.152 RELATIONS AMONG DERIVATIVES fox. un We therefore may take a quantity T’, such that bey — ke! = T+ Q (iy mays) + Q'aays + Jerys Kay — he = T+ Q (Yatra Yate) + Q's + Spits the equivalence of the two values of I’ representing the characteristic equation. (vii) Lastly, proceeding from the relation ie = ah + Yr0nt Weaken + 2YoCn derived at the beginning of (vi), and differentiating both sides completely with regard to f, we have Bikey + Ys00 + tall + YOu + alin + Yolen sgh + akg + Ys0u + Yon + Wl’ + Lerslian + Bae’ + BysCone Hence 2 (keg — ley’ + ea!”) + 9h Coe — On! + Cea") = Okey! + Doak! + Laken t YCua" + Balu + Ysera a = Gp ikea + rea) = 0. Proceeding similarly from the relation 3 = 2170 t Yio + 2a + Bakar obtained by differentiating the identity 2G, 2G, 2G,» a Gamma thy, + mae + sy, partially with respect to #, we find (Y= Ya! + et") + 9h (hao — Ke’ + eg”) = 0. We thus can take two quantities U and V, such that ome! tye! = 20, lg ke + hy! = 40, Fog fad es” == YeV C90! hon” = 0,7. Equating the two values of ki, we have iV — 050 = ba — bg! + he” == Coy — ay)’ —Coays— ay) T = QCeye— ng) ~ 20 (wp. ay) Q" (eye), by means of the characteristic equation F=0; thus {PF ol)- Fa} =a {0-F wl) FO} 3, such that Hence a quantity A exis Vaart Sad) Gla Taud+$ Yd) + $n106] FIRST NORMAL FORM OF SECOND VARIATION 153 and therefore, when these values of U and V are used, we have on & C= Ca Oy" = afan+ Ged + F(a ag Ie Beg” A+ 2 (aJ)4 Zed oe nat Fed) + Be aha bg! == a {yh + F40d) +5 (Hh an’ tan” nfd +E cved)+ (Oh ‘These various equations constitute the aggregate of relations among the cocfiicients yma, knns Cun, to be used in the transformation of the second variation. Yo First normal form of the ‘second variation. 106. In order to modify the expression for @, which occurs in that second variation, we introduce the former linear combination w of u and v, defined by the relation w AY ls s0 that w is a measure of the deviation* along the normal to the curve caused by the small variation eu and . Then ¢, is the smallest value of ¢ greater than t, such that £(2) vanishes there: or, if wo write £(f, #) in place of £(2), it is the smallest value of f,, >, such that S(t H)=5 | Ch), A(h), O(h), Olt) |= (to) » Oa(te) , Ost) » Oat) OY (te), (te) (te), 4 (te) 4"), 4," Ct), 04" (te), 84" (C0)170 CONSUGATE-BOUNDED RANGE fon. mm Here is a mere non-zero unessential constant, independent of ¢, and t. ‘The point on the characteristic curve, determined by this value of t,, is called the conjugate of the point determined by the value of t,. As the determinant in the expression for {(f, 4) is invariantive under the linear transformations to which a fundamental set of integrals of the equation E=0 is subject, manifestly any four linearly independent integrals of E suffice for the construction of the equation £(%, &)=0. Thus the settle- ment of the conjugate of the initial point ¢, is not affected by the particular choice of fundamental integrals 6, 6,, 6, 8 of the subsidiary characteristic. Also, the position of f, is independent of 8, which affects only the scale of the deviation of the consecutive from the original characteristic. Finally, £(t,, t) is a regular function of ¢. After vanishing when t=), it does not again vanish for increasing values of t, until =f; that is, for values of t such that t, f, 60 that we have 0 = 918, (t) + g29e(t) + gu (ts) +.9.84(b)5 the adjacent characteristic is represented by the equation SO | 8" (to), 85" (bo), 0," (t), 84" (ts) |= | Cs Olt), BCE), BCL) x(t) alte) » At) Oto) s(t), Bala), Os(t)» Bxbo) (te), (te), Ox (te), 84 (ts) Y(t), O4'(t), Os (te), 1 (ts) A(t)» Pate), Ost) , Ox(te) (t+), Oo(h), 91(t), (4) So far, the place f, can be chosen at will; because one arbitrary element in £( has been left, a survival after the assignment of the three initial ‘conditions, as regards position and tangeney (both unchanged) and curvature (changed). Among these adjacent curves, let that one be selected which touches the original curve at the next point of meeting after &. Denoting this value of ¢ for this point still by %, we have 420 when ¢=4; that is, the equation Alt, 6)=| (6), Ol), A(t), A(t) |= 81 (to), Out), 80 (ta)s 4 (te) (te), Ort), Os(t), OC) 8 (G), Ox (te), Os (te) Od (ty) is satisfied. It may be that A (f, &)=0, as an equation to determine f,, has more than a single root >t; iffso, we shall assume t to be the smallest of such roots,176 SUB-CONSECUTIVE [ex mm ‘A characteristic curve, touching an original characteristic curve in two distinct points ¢ and t, instead of osculating it at t, and merely meeting it again in a distinct point t, will be called a sub-consecutive characteristic. Equation of sub-consecutive characteristic. 121. The analytical expression of a sub-consecutive characteristic, when explicit account is taken initially of the fact that it touches the characteristic at th, is SO=E) AO, BO, AO, AO J3 ,(t), Balt), lle)» O4(t) Oi (to), 8 (ts A(t), 80 Ch) A(t), Olt)» A(t), 0.(4) and the relation, securing tangency to the characteristic at t,, is Ata 4) =0. ‘When explicit account is taken initially of the fact, that the sub-consecutive characteristic touches the characteristic at ¢,, the analytical expression of the sub-consecutive curve is LO=H14M, &, &, aH |, | A(t), Ct)» s(t), A(t) (ta), A(t)» Pat), Alte) | Or (ts), Oc (ty), Os (he), 81 (ta) together with the same relation A(t) =0, which now secures the tangency to the characteristic t. |As the two expressions for £(t) represent the sub-consecutive curve, each taken in conjunction with the equation 4 (t, t)=0, it is to be expected that the two expressions will agree, ‘To secure full agreement of the two expressions along the whole curve, it is necessary that the coefficients of 6,(¢) in the two expressions must be the same: and likewise for the two coefficients of 6, (2), of @,(t), and of @,(t), respectively. Consequently, there must be four relations of the form | Oulte), alle), Alte) [=H] Ct), (le), Bulb) 5 | ag (td), A(t), Ot) | | alba) Ble), Ob) A(t), Ct), (6) | | OG) (HD, 6 (4) and they must be consistent with one another, in virtue of the relation A (fy &)=0 connecting the variables f, and t,.122] CHARACTERISTIC CURVE qt Let AONE AC HECAC HEE) ae sy Guy Gay Coy Cas | (4), Ole), Als), O4(6) | | A(t), OF, BL (t), 82 Ct) [ea ‘Pins Yee Pas Yar Ye 18) Of), OCH, 084) ‘Then, if CO, () + H8, () =K,, for r= 1, 2, 8, 4, the foregoing relation becomes Kyou + Kyea+ Kita =0. There are three other such relations, One of them is Kyeut Keta+ Kita =0. Eliminating K, between these two modified expressions of relation, and using the identity Cata + tata + Cate=0, wo find Kyoa + Ket + Kita 0. In a similar way, we deduce from those two relations Kc + Keea + Kien = 0. Again, eliminating the ratio @ : H between the two relations Kyout Kyat Kita =0, Kyeu+ Kieu + Kies=0, we have £8) cau + 8x! (to) Can + Be (ta) Can} {01 (ta) Cae + 8 (2) Ca + OK (ta) Crs} = {81 (to) Cop + O,' (bo) Ca +O: (to) Cus} {8 (Ca) One + 8, (te) Cen + 0, (ts) Cs}, that is, Ca (aoe + YesCre + YarCae + Yorn + Yale + Yun) = 0, = CA (toy fe) = which is satisfied. Thus the four relations, which express the equivalence of the two forms for £(¢), are consistent with one another. All of them, when account is taken of the equation A(f, %.)=0, are satisfied, if @ and Hf are connected by a single one of thom, such as Kyou t Kicat Kut =0. or 122, Take a point on the sub-consecutive curve between f, and f, and very near to 4, Then from the G-expression for £ we have, effectively, EO= FE ty GE | Oh), Os" (te), 5" (le), 4" (be) |» A(t) + Bolte), Ps(b)» a(t) ON (te), A(t), (te), (te) A(t), A(t), P(t), (4)178 PROPERTY OF CONSECUTIVE CHARACTERISTIC {om ur and from the H-expression for £ we have, effectively, EO MHE WP H| O76), 7H, OG, BU) |, A(t), Galt), ste), Alte) | Ct), Olt), Ble), Ole) | BNC), Ot), B(), 4. (6) | the coefficient of t—f, in the H-expression for { being the vanishing quantity A(t, t). The two expressions for {(t) in the near vicinity of t, agree. Similarly, for a place in the range in the near vicinity of f,, we have LO=HG-9E| 4°), OH), 8: ), OC) |, F.(6) + Balt), (4), B.(l) 45 (t), Bi), 85 (t), 8: (b) 4,(t) » Blt), Ot), (4) | =H | O"(t), O(H), O"(), 8" (t) |, (te), Oa(te) + Pa(te) » Pa(to) x(t), Palte), A(t), A4(4) Oi (b), OH), AH), OE) from the two expressions respectively, the coefficient of ¢,— tin the G-expression for {being the vanishing quantity — A(t, t,). These two expressions for £(¢) in the near vicinity of t, agree. ‘Manifestly a sub-consecutive characteristic does not cross the characteristic curve with which it is associated. and Property conserning @ consecutive characteristic. 123. One other proposition, connected with characteristic curves, will be established here; it is required, later, in the discussion of small variations ” ‘that are not weak. 3 7 Let APR... be a characteristic Q ® curve on which A’ is the conjugate of A; and let P be any point on the curve within the range AA’. Take any point Q, off the curve, and con- tiguous to P; then one, and only one, i consecutive characteristic curve can be drawn from the same initial point A go as to pass through Q. If possible, let the consecutive characteristic curve be represented by 8, (t) + KO, (t) + Fy, (t) + kO, ();124] NORMAL FORM OF SECOND VARIATION 179 because it is consecutive to APR..., we have =k, (bo) + haOs (te) + hss (ho) + he (4), O= Ky (te) + Kay (bo) + a8,’ (te) + Hi. (te), = 8," (bo) + fa" (te) + Fa." (ba) + sO." (bs where f, is the value of t at A. At P, let ’ be the value of ¢ on the original curve, and ¢’+«7" its value at Q on the hypothetical consecutive character- istic. If # and y be the coordinates of P, those of Q can be represented by a+ 0X’ and y+e¥’; thus X= Gi (L) T+ bags C) + babe (D+ kode (l) + kage), Yee y OC) T+ btn (t) + kefe) + bas) + heal), and therefore $C) Py CX! = HG, (0) + aba (t) + hy Resolving the four equations for ki, ka, hy, ky, we have Slt, tke = {8 (C)Y'-W' (OX Yen, for r=1, 2, 3, 4, where ¢,, ¢, ¢, ¢ are the coefficients in the expression of the consecutive characteristic curve in § 118, and {(f, ¢) is the former expression for the deviation in that consecutive curve. As 4’, given by t=, is the conjugate of A, so that t, is the first value of 4, (>t) satisfying the equation £(%, f,) = 0, and as t’ is intermediate in value between t, and 4, the quantity {(f, ’) does not vanish; hence k,, ky, ks, ky are finite and definite. The consecutive characteristic can be drawn as required. Bs, Let B be a point on the characteristic in the range 4P, and let C be @ point contiguous to B. Prove that it is possible to draw a characteristic curve through C and @, when the difference between the directions of the tangents at B and C, and the difference between the curvatures at B and Q, are of the same order of magnitude as the small variation displacing B to C and P to @. (C) + k(t). Discussion of the final normal form of the ‘second’ variation, 124. We now come to the normal reduced form for the second variation of the original integral, which has been proved (§ 110) to be Q ww, ww |tat. lame 7 o 8’, B, BI Here, a and 8 are two linearly independent integrals of the subsidiary characteristic equation E=0 such that the quantity Q(aB” —a"B— ap" + a'B) + Y (w8” —a"B) + J (ap'-e'8), known to be a constant for any two integrals, is made equal to zero by the choice of a and 8. For the discussion, we shall need the properties of adjacent180 DISCUSSION OF SECOND VARIATION {en. ur (but not sub-consecutive) characteristics and, ultimately, the properties of the consecutive characteristic. ‘The first variation of the original integral has been made to vanish. In vanishing, it secured fixed limits for the integral from assigned conditions, if these were not already fixed by the preliminary assigned data, Thus all the variations, which now may be considered, must be such as to vanish, at the lower limit of the integral and also at the upper limit of the integral. Consider, first, the lower limit of the integral, being the value t, of ¢ for the initial point of the range on the characteristic. We take a couple of independent adjacent curves, represented by small variations xa and «8. If ‘the coefficients for a be m;, ms, ms, mg, and those for 8 be nm, a, ns, me, then, ‘1 9, (te) + 19s (te) +150 (ta) + 2494 (bo) = 0, 1,0, (te) + 14s (te) + 10s (te) +,8, (t,) =0, because the characteristic represented by the deviation ea is adjacent to the central curve; and 1,8; (te) +20 (te) +105, (te) +748, (be) = 0, 1,8, (t.) +204 (t) +145 (ts) +401 (6) =0, because the characteristic represented by the deviation «@ is adjacent to the central curve, The two former relations give a=0 and a’ =0 at f, and the two latter relations give 8=0 and 8’=0 at t,; hence the quantity Q(a8" — a8 — a/8” + 0"B’) + Q (a8 —a’8) + J (ah’ — a8) is zero at t=, and so (because it is a constant along the range) it is equal to zero for all values of ¢, Thus the two selected integrals « and § of the equation E=0 satisfy the prescribed demand; and the second variation is expressible in the foregoing form. 125. For the adjacent curve represented by a, let, p, and g; be the values of” and @” at the initial point; and for the adjacent curve represented by B, let pr and o; be the values of 8” and 8” at the initial point: then Pr m6," (to) + 1220," (ty) +150,” (te) + m8." ol 1 = 1B" (to) + 12982" (te) + 585" (to) + 148.” (ty) P2= M8," (bo) + 00," (to) + 12283" (to) + m8,” a = m8," (to) + M83" (bo) + 390" (be) + 48." (to); The determinant of the coefficients in the four linear equations, in which ‘my, Mg, M,, m, ovcur, is —W (t), @ non-vanishing quantity; and likewise for the four linear equations in which n,, m, 1g, n, occur. The coefficients m, and the coefficients , are thus uniquely determinate ; they determine two adjacent curves given by m8, (E) + 1120, (t) + 11, 84(E) + 248,(0), B= MBO + OO) + Os) + m4 (1).126] SACORI TEST 181 ‘Now consider a characteristic curve, represented by the combination put PB. ‘The coefficients for this characteristic are p,m, — pity (for 7=1, 2, 8, 4). These four coefficients satisfy the relations 3 Goame— in) x(t 3 (om, — p.m) 8b 3 foam, = pine) 8" (by). ‘We must not have : & (pate — pate) 8," (ta) a vanishing quantity. Otherwise, as the determinant of the coefficients of the four quantities pym,— py, is W(t) which is not a vanishing quantity, we should have Patt — pity = 0, for r=1, 2, 8, 4; and then pye—p,8 would be zero for all values of t, so that aand 8 would not be linearly independent. Hence pra Pita must not be zero. And now, the combination p,a—p,8, instead of being a zero combination, gives a consecutive characteristic curve. ‘The Jacobi test. 126. Now the first value of ¢, greater than %, for which a consecutive characteristic meets the original characteristic, is the value denoted by f. If therefore the range of the integral, beginning at &, should extend as far as f,, we could have @ non-zero variation pa—p,8 of the characteristic. ‘That variation vanishes at %,, because @ and f represent adjacent curves; it vanishes at t,, because p,a— p,8 represents a consecutive characteristic; and it is not zero in the coursé of the range between t, and f,. In these circum- stances, we could take w= p.a—p.B; such an assumption would meet the requirements of vanishing at each ex- tremity of the range, and of being distinct from zero throughout the whole of the range save at the extremities But p, and p, are constants; and therefore, for such a value of w, the magnitude w, wy w a’, a, BBB182 LEGENDRE TEST (ox. ur vanishes, while af’—a' is not zero because « and @ are linearly independent, Should this oceur—that is, by selecting a particular non-zero variation of the characteristic curve—the second variation vanishes. But if a maximum or a minimum is to be possessed by an integral, it must arise for a stationary conformation settled by the characteristic curve. ‘The maximum or minimum is certainly possessed, if no admissible small variation can occur which makes the second variation vanish, (It might be that admissible small variations could occur, of the type just indicated, for which the second variation vanishes. In that eventuality, the third variation would have to vanish for those variations if the maximum or minimum is to ‘be possessed; and the fourth variation must not vanish for them. The case is not sufficiently general to justify discussion here; when it arises in any particular instanee—as in Ex. 1, § 30, p. 33—it can be considered by a special discus- sion. It will therefore be omitted from further consideration in the present general argument.) We therefore exclude the possibility which renders small variations of the foregoing type admissible: that is, pya—p,8 which vanishes at the beginning of the range must not again vanish within the range and, as has been proved, it cannot vanish earlier than at the conjugate of the initial point. Consequently the range of the integral must not extend as far as the conjugate of the ‘initial point on the characteristic curve, We thus obtain the range-test (the Jacobi test), limiting the range. The Legendre test. 127. There still remains the quantity Q for consideration, as it occurs in the subject of integration in the integral expressing the second variation. If Q conld be sometimes positive and sometimes negative, then, by taking ‘w zero over a range where @ is negative and distinct from zero over a range where Q is positive, the second variation could be made positive; and, by taking w zero over a range where @ is positive and distinct from zero over a range where Q is negative, the second variation could be made negative. But, the second variation must have a persistent sign, if the integral is to possess a maximum or a minimum. The possibility, just indicated, invalidates the existence of cither, and must therefore be precluded from occurrence in con- nection with a favourable issue for the problem. ‘Thus Q must be either always positive through the range (and then a minimum can exist, subject to the other conditions) or it must be always negative through the range (and then a maximum can exist, subject to the other conditions), ‘We thus obtain the test (the Legendre test) : The quintity Q(t) must have a uniform sign throughout the whole range of the integral,128] AGGREGATE OF CONDITIONS FOR WEAK VARIATIONS 183 Summary of conditions for weak variations. 128. We thus have established the following three tests for the existence ofa maximum or of a minimum of an integral when its variable quantities are subject to weak variations : (A) The curve, along which the integration takes place, must satisfy the characteristic equations, (This test may be called the Euler test, as Euler was the first to obtain the critical differential equation in virtue of which the characteristic equations are satisfied. Sometimes it is called the Lagrange test, because of his more extended investigations at a date later than Euler’s first investigation.) Moreover, either the terminal data explicitly, or inferences from terminal descriptive conditions onstruetively, determine the limits of the integral as known quantities which, once known, are immune from variation, (B) The range of the integral along a characteristic curve must not extend as far as the conjugate of the initial point of the range. (This is the Jacobi test. In any exceptional instance when the range is extended up to the conjugate, the second variation of the integral can be made to vanish for a particular type of emall variation of the variables; the higher variations of the integral, for that type of variation of the variables, must then be investi- gated before declaration can be made in favour of the existence of a maximum or of a minimum.) (C) The quantity, which has been denoted by Q or Q(t), must not vanish nor become infinite throughout the range of integration: it must keep a uniform sign throughout. (This is the Legendre test. It is conceivable that, at isolated points, Q may vanish; but the zero then must be of even order, and that order must not exceed four, in the present case: also, in that event, there would be limitations on the quantities J and A. It is simpler to set these special possibilities aside, examining them only when they arise in particular instances.) ‘The three conditions are necessary conditions, arising out of the discus- sion of the integral under weak variations. When satisfied, the conditions are sufficient to secure a maximum or a minimum for weak variations. For the first condition (the Euler test, when satisfied), together with the accompanying inferences as regards the terminals, secures that the first variation of the integral vanishes for weak variations. ‘The third condition (the Legendre test) and the second condition (the Jacobi184 EXAMPLE [ox. um test), when they are taken together and are satisfied, secure that the second variation of the integral is either definitely positive in all circumstances, or definitely negative in all circumstances, for weak variations. When the definite sign is positive, the integral possesses a minimum: when the definite sign is negative, the integral possesses a maximum. 129. An example follows, in which the tests for weak variations are satisfied, yet for which a real minimum does not exist. Ex. Find the curve (if any) sueh that the area, comprised between (i) any are, (ii) the radii of curvature at the extremities of the arc, and (iii) its evotute, is a minimum. ‘The area in quetion is dfoa ni [fae oo that we may take Leann g Oth, (1) As fdoos not involve y explicitly, a first integral of the characteristic equation is £-i(G)-come that is : 12080942 (Yan But ER ey ryt yn Leny_aay aby. eo ny - ee and therefore oe Pe at By: that is, changing the arbitrary oonstants 4 and B to 2a and 28, Cy ae yx tye ty Greeny tain, where, a5 usual, we hav ae : yoteny. ‘Tous [ BpT eos y+Dsin pcos yy F(a con y+ and therefore 4 (e-W)=2ay-+asin 24 —beonah} 4 (y—#)=2by —a 008 2y—b sin By)” where h, F are arbitrary constants, We thus have a primitive in which the four essential arbitrary constants are a,b, f,129] EXAMPLE 185 (II) If eX and xY denote the displacements at the limit, the equation y a_i Y 4 (9), y [Lv h-a@) yale o> rust be satisfied there, We shall assume that the required curve is to join two fixed pints, so that each extremity is fixed in position ; we thus have Y=0, at the extremity, and therefore the condition becomes at that extremity. (i) If the direction at the extremity be fixed, as well as the position, Y’=0: no further condition emerges from the limiting term, and the value of y is then known as a given. datum, i) If the direction at the extremity is free, so that there is a condition to be satisfied for all directions there, then of a ¥ that is, (+y2)y'-? is zero there, and the radius of curvature is zero, that is, the extremity is a cusp, or, reverting to the current equation p=a.cosy-+bsin y, ‘we should have doo +b sin y=0 at an extremity, when itis fixed in position and free in direction, (Gi) Let one extremity be the point (29, yo), and let yy be the value of y there: the other, the point (=, yn), and y, the value of y there: where Yo and yj, may be regarded as known, if initial directions are given: or, otherwise, are subject to the foregoing external condition. In the former case, the four constants a, 5, f, kare determined by the conditions 4 (24h) =2afy-+a.sin 2y—B cos 2} 4(Y—B)=2b fp — a.008 2Yy~ Bsin BY 4 (a,-B) 2a, 4.4 sin 2), — b c08 24, 4(1— B= By —a.008 Bain By ‘and, in these foar equations, yp and yy are known. In the latter case, we have the same four formal equations of condition, where now yp and , are not initially given. But now we also have the two terminal conditions acosyo+bsinyy=0, acos yy +bsin arising out of the evanescence of the first variation of the integral, ‘The results a=0, b=0, deducible from these last two conditions, are out of the question ; and so we have sin (Jr Yo)=0, wwe can take (as the simplest case) Yi~Ya=m, oF (as a more general case) mm, where m is an integer. We at once have which, with the condition Y4—Y=m, shows that the tangents to the curve at the fixed points aro perpendicular to the line joining the fixed extremities. When we write sino, b= e000 Yo,186 MINIMUM UNDER WEAK VARIATIONS (cx. ur the equations of the curve are 4 (=A) =c.08 (2) — Yo)+2op 5in yy 4 (y—B=enin (2) Yo) 204 000 and the terminal conditions give 4 (A) = 008 port 2eposin yy 4 (gE) =esin Yo—Ze}r008 Yo 4 (41) =c 008 Yo +2 (+ Ya) 80 Yo 4(4—H)=esin yo 2e (w+) 008 4 ‘The curve is @ cycloid having the line joining the fixed points for base ; and the fixed points are cusps, If mis greater than unity, the curve is made up of m equal eyelids, all along that base and having all the cusps along the base. (IID) As regards the Legendre tost, we have yt ay and, naturally, the area is to be taken postive, #0 we may take y” positive*. Now ay _aty)? event 4 positive quantity ; hence, if other testa are satisfied, the Legendre test will allow the integral to possess minimum, (IV) To apply the Jacobi tost, we must construct the equation which is to settle the conjugate (say yi) of an initial value of Y (say Yo). Merely changing the arbitrary constants a and B, we have the primitive of the characteristic equation in the form eahta(2y-tsin 2y)—b cos 4, yaksb (2) sin 29)—a.c08 2. We have Tein y, = Teo y-+2psi0 Y), . -He == Ticin ¥-2y 0089), T=4(acosy+bsin ‘The critical equation is a 4 (Yo 8 (eh 8 (Yar (Ye) |=. 8 (Yo, 8 (War (a & (Ho) "(os 6" Was 44" (bob 8" (Ho) 4 Was & Was hh &% Gn) + Were y” negative, the radius of curvature would be taken equal to —(1+y8 for the purpose of making {pds to be positive: and the Legendre test would then be satisfied as in the teat,129] DECREASED BY STRONG VARIATION 187 Substituting, and reducing, we have the left-hand side equal to 513 (woos Yo-+b sn Ya} (2.608 Yr +b sin Ya) sim (fa — Yo) {(Ja— Yu) 0b Ga Ya)—D- We have to consider the roots of this equation. The factors, which give rise to 2008 Yo+5sinYy=0, acos rt+bsiny,—=0, ‘as in the case where the end directions are free, and as connected with the equations A(acosy-+deinyyeosy, Yond (a cosy-+bsin y)siny, indicate cusps of the cycloid. Thus the equation precludes a range which actually includes a cusp. ‘The equation (Ho) 00 (bY) — for & rango +2 > Yo, has only one real root, vie. point of one complete branch of the cycloid ; the factor Wh-Wo) oot i-W=1 provides no conjugate in the range, ‘The equation 0, Yo, merely giving the initial in (Yi - Ya)=0 gives Yum +o, a9 the first value of y, greater than y and satisfying the equation ; the factor therefore, in the extreme case, merely indicates a cusp as the conjugate of a pre- ceding cusp. Tt thus appears, when account is taken of all the equations into which tho critioal equation can be resolved, that the range of the oycloid must not include a cusp. ‘The inference therefore would be that, if a curve is to be drawn joining two fixed points with fixed directions at each point, we take & cycloid through those points so that no cusp of the curve shall occur within the range : while, if the directions at the extreme points are not fixed, we obtain a minimum area though the Jacobi condition is just not satis, Note. In the latter case, it would follow that, for one special type of variation, the second variation of the integral vanishes ; as an immediate preliminary, it would bo necessary to consider the third variation of the integral. (V) Other considerations shew that, for even the most generally satisfactory case where the Buler test, the Legendre test, and the Jacobi test are satisfied, there is no real ‘The fact is that only weak variations have been taken into account. Tho characteristic ‘curve, thence derived through the Kuler test, isa eycloidal arc AC! ‘to simplify the discussion, we may regard the directions at 4 and B as fixed at the cusp. Take sny point @ near A in the line 4B; and draw a small eycloid on 4@ as base. Also draw a eycloid on GB as base. The ‘two cycloids on AG, GB lie within ACB ; they constitute a small variation of ACB. The area, associated with the double cycloidal arc from A to @ and from G to B, is less than the area associated with the are ACB. But the variation from ACB to AG, while small in the magnitude —4¢ of the displacement of corresponding points, is not a weak variation;188 ILLUSTRATION OF STRONG VARIATION fou. um ‘the changes in the direction of the tangent and in the curvature are violent. ‘Thus a strong variation has led to a reduction in the minimum. We can see otherwise that the area produced by the eycloidal aro can be reduced. Divide 4B into n equal parts, one of which is AG; on exch such part as diameter, let a semi-circle be drawn. The total area, under the requirements of the problem, and provided by tho m somi-ciroles, is ABV? oe ante (42)) te Ean, which, with increase of n, can be reduced indefinitely.CHAPTER IV, INTEGRALS INVOLVING TWO DEPENDENT VARIABLES AND THEIR FIRST DERIVATIVES: SPECIAL WEAK VARIATIONS. Integrals of the first order in two dependent variables. 180. We now proceed to deal with integrals which involve two dependent variables and one independent variable, denoted by y, z, and « respectively. We begin with the simplest problem; only the first derivatives of the de, It will be convenient to associate a geometrical interpretation with the analysis for all such problems; manifestly, skew curves offer the natural medium. In the geometry of such curves there is less tendency, than in Plane curves, to a preferential selection of one variable (such as 2), over the other two, to rank as the independent variable; and so there will be added reason for ultimately considering 2, y, z as functions of a new independent variable, Such an ultimate selection will, as before, allow the small variations to be made in all the variables, instead of merely special variations to be made upon the initially selected dependent variable. ‘Then (as in the pre- coding chapters) possible discontinuities, as of direction, may be discussed, Moreover, the selection simplifies the expression of properties of tangoncy and curvature, dependent variables (viz. y, =; and 2, -%) will be supposed to occur. As, however, certain results, fundamental in character, are derivable from the analytically simpler supposition, which makes y and 2 functions of «, and subjects y and z to small variations, without regard to small variations of , it is advantageous to derive the corresponding results at once. ‘They will indicate the type of test required for the general weak variations. We accordingly consider an integral T= [fenyra eds, the limits of which may be definitely fixed or are to be deducible from assigned data. ‘The problem is to determine the conditions to be satisfied in order that J may possess a maximum or a minimum.190 VARIATION OF INTEGRAL INVOLVING [curv Variation of the integral under special variations. 181. We begin with special weak variations. We therefore assume y + xv as a varied value of y, and z-+ew as a varied value of z. Here « is an arbitrary constant quantity, of sign that can be taken at will, and of magni- tude sufficiently small to render any power negligible in comparison with every lower power. The quantities » and w are arbitrary regular functions of 23 and we assume that xo’ and xw are small quantities. If the limits are fixed, they remain unaffected for variations made upon I. If the limits are to be deduced from assigned data—thus, if the extremities of the curve, representing y and z as functions of «, are to lie upon a given curve or curves, or upon a given surface or surfaces—the varied upper limit may be taken with +X, as the varied value of a, and the varied lower limit may be taken with «,+«X, as the varied value of ay. In this event, X, and the values of v and of w at «, must conform to the data at the upper limit, while X, and the values of v and of w at 2, must conform to the data at the lower limit. Further, the function f(a, y+ ev, tev, 2+eu, 2+ ew’) is assumed to be expansible in a series proceeding in powers of x. Then, if J denote the inuegral arising from the complete variation of Z, we have ve 1-[ Fa, y+ wv, of +a, 24 ew, 24 eu! yd - [Pe uyss2)d0 a [ey ten ote ee, 2 +0) Se, 9, 6,2} de = eyte Ky + [payee yf to 2, 2 4) de aebee ] f “F(a, y+ ee, y tev, 2+ ew, 2 ewe, ‘The integral in the first line of the right-hand side is = f"(° Lav vwLewd L) ao K, where K, is the aggregate of terms involving second and higher powers of Now iz feed ad ar A Irv n — 2) a” da \ay’ 7) de, Peters [wd |t-[lod 3) as; and therefore ofa Gl} ae Chere be] lt Ky132] TWO DEPENDENT VARIABLES 191 Let f, and f, denote the respective values of f(e, y, y/,2, 2) at a, and ,; then (as in § 32) [Fe yt, yf + ot, eb 0, ft ele = afi Ke, a [Fe yt y+ el, 2+ 00 ol) dom eXefet Ki, where K,' and K,” denote aggregates of quantities involving the second and higher powers of «. 132. We have to bring «X, xv, ew into relations with one another in connection with the curve or the surface upon which a displaced limit stands: and the necessary relations must be estab- lished for each limit. In the figure, P is the point of a curve at the limit which is subjected to a displacement taking it to Q. Through P, we draw PA, PB, PC, parallel to the axes of «, y, z. Through Q we draw planes QDA and QDB, parallel to the planes Oyz and zz, The curve which is being varied is ...PR... cutting the plane QDA in R; so that, if the are PR=o and if A denote (1 +/+ zt, we have ao iv 7 PA=xe, AN=4o, NR= Se The curve along which the limiting point P is compelled to move is PQ, the dotted curve: or, if the limiting point P is compelled to move on a given surface, then ...PQ... is some curve (and, indeed, may be any curve) on that surface. As Q is the displaced position of P on this hypothesis, let the coordinates of Q relative to P be kX, kY, kZ, so that relations (or a relation) will exist between &X, LY, AZ, determined by the equations of the curve ---PQ... (or the equation of the surface on which ...PQ... lies). The small variation xv and xu, to which P is subjected by variation of the curve that is to provide the maximum or minimum, takes P to S, where S is in the plane BPO. Up to small quantities of the first order, QR is parallel and ‘equal to PS; and therefore the coordinates of Q relative to R are equal to the coordinates of § relative to P, Thus «v= AD-AN, nw=QD-RN. But PA =kX, PB=kY = AD, P0=kZ=QD; also AN=Yo=KyX, NR=5 =KrX; and therefore wvsnY~nyX, ew=0l—02'X, that is, at an extremity of the integral, v= ¥-yX, w=Z-7X.192 ‘TWO CHARACTERISTIC EQUATIONS [ou.1v Note. These results can also be established as follows. When P is taken temporarily as origin, the coordinates of Q are «X, «YZ; and the equations of the tangent at Q are ya bea where £, », {are current coordinates along the tangent, the equations being accurate up to the first power of « inclusive. Now § is a point on this tangent, to the same order of accuracy; and its coordinates are 0, xv, «w. Hence Y-yX, w=Z-7X. 183. Gathering together the results, we now have =f" [oe -2 Qh + ¥-2 OF staf - EG + Ra] +e [x (-vB-2P)vvhsegl en, where R,=K,+K; + K,", being the aggregate of all the parts of J—I which involve terms of the second and higher powers of « arising from the expansion. In this expression, ev and xw in the integral are the arbitrary small variations of the points on the curve, while «X, «Y, x2 represent the displacement of a limit upon the terminal curve or the terminal surface and variations of every type are admissible and must be taken into account, arbitrary, independent of one another, subject to the conditions (i), that and w are regular functions of 2; (ii), that X, ¥, Z are consistent with the maintenance of a terminal position ; and (iii), continuity between v and w, and X, Y, Zat the terminal positions. The two characteristic equations, and the terminal conditions. 184. By the usual argument relating to the possession of a maximum or minimum, the ‘first’ variation in J must vanish ; and it must vanish for all small variations of the curve. Consider, first, a small variation xv and xw for y and z, with real * variations at the extremities, ‘The terms included in the expression [ | * vanish for such variations, because X, Y, Z are zero at each limit. Hence ao zh af EQ} . ie [8 a) vu fl 28 =0, for all arbitrary regular functions v and w independent of one another. The conditions, necessary and sufficient to meet this requirement, are that the135) ‘TERMINAL CONDITIONS 193 equations oid (a 2m iy ae f x _ 4 (af B= Fase) = shall be satisfied everywhere along the curve. Were it not so, by choice of » where 3 is not zero, and of w where % is not zero, the integral could be made positive or negative at will, contrary to the necessity that it shall vanish. ‘The two equations are the characteristic equations, Theit primitive, being the complete simultaneous integral of the two equations, determines y and z as functions of 2; it thus determines a curve in space, which is called the characteristic curve. We shall now suppose that we are dealing with such a curve, so that the characteristic equations are satisfied. Consider, next, a small variation, which is zero at the lower limit of the integral and not zero at the upper limit. Because the curvo is now restricted to be a characteristic curve, the portion of JI depending upon the integral vanishes. ‘The requirement, that the first variation shall completely vanish, can now be met only if the relation of ¥ x Z ( t-9 eA) +V4a%—0 is satisfied at the upper limit. Similarly, by considering a small variation not zero at the lower limit, we infer that the same relation must be satisfied at that limit, Consequently, in order that the first variation in J~J shall vanish, it is necessary and sufficient (i), that the characteristic equations shall be satisfied, thus determining the characteristic curve: and (ii), that the condition ~y Fie Pf of 7 of x(s ty yt Yer eem =0 shall be satisfied at cach limit separately, in accordance with some terminal property or properties for the characteristic curve, Hamilton form of the characteristic equations 185. The characteristic equations for the present problem, when f does not explicitly involve the independent variable, as also for many such problems involving more than two dependent variables, can be expressed in the Hamilton form which is used in theoretical dynamics, We write of af ay” &194 HAMILTON FORM OF CHARACTERISTIC EQUATIONS [on 1v taking » and as now variables, being of course dependent variables; then the characteristic equations are -2(¥ eid (7 iy ay) Se te (ae) 7 Thus y, yf, 4,’ are expressible in terms of 7,7’, £, ¢’. We now introduce a new function H, by the definition eg any he eZ -f syntet—f. ‘As f does not now explicitly involve «, we have dH =(y/dn-+ndy) + (edt + Ede’) (Lay +L ays tar +%, (Gay tHatee the) sy dnt edh—n'dy— de. Let the function H, initially defined with reference to f and the arguments in f, be transformed by the equations oo oe ay ae & ‘90 as to become a function of y, 2,7, { In this form, BH ay 5 Hag 4 Hay Ha, aH = an 4 Fe ae The two forms, obtained for dH, are oe for all variations, Hence Hy Hy Hy, ra ye cent and therefore ay dy which is the Hamilton form. Nature of the primitive of the two characteristic equations. 186, If we are to obtain the most general solution of the problem, we must obtain the primitive of the characteristic equations. (It might happen that particular solution of the problem could be provided by a special integral of those equations.) Now the equations are of . a" + sabe +916 ws 6 2)=0, Hythe the yy, % 2) =0,136] PRIMITIVE OF CHARACTERISTIC EQUATIONS 195 leading to equations Oy’ = Oley. 9,4), Oe M09, 59,2) provided the quantity @, where ondh 2h (OE ay ae (ayer) * does not vanish. Then the elimination of z and its derivatives leads to an ordinary differential equation of the fourth order in y; and the new equation has a primitive of the form Y=(@, Ay, Aa, As, Ad, where A,, 4,, 4s, A, are arbitrary constants, When this value of y, once known, is substituted in the two equations 798, By and 2’ is then eliminated oe these two equations, we have e= (a, A, Ay, As, Ad). The primitive of the characteristic equations manifestly is y= (@, Ay, Aa, As, Ad z=4(e, Ay, dy Ay, ap which, accordingly, are the equations of the characteristic curve. Note. Later, it will be proved that, for the possession of a maximum or a minimum, the quantity ar ar ee afy ay? as* ay must usually be positive and must never be negative; but, in exceptional circumstances and subject to further conditions attaching to the function f it may vanish. If it should happen that @=0, the two initial characteristic equations are equivalent to Shu’ + phe +0 nya 2)=0, ka yf, 2)=0. Differentiating the second equation and combining the result with the first, we obtain equations YtP@mys52)=0, "+Q(mmy, 4 2)=0, simultancously with &(2, y, y,2,2')=0. A differentiation of y”+ P=0, leads to a relation y+ Rw yy, 2, 2)=0; and we then have, on the elimination of z and 2’, an ordinary differential equation of the third order in y. ‘The primitive of the characteristic equations becomes, Y=X (1, Ce, 03), 2=0 (2, C, Cr, O), which then constitute the equations of the characteristic curve.196 FORM OF PRIMITIVE (ox. 1v ‘As however the occurrence is exceptional when a maximum or a minimum could be possessed, the detailed examination may be left for any problem when it actually arises. We shall therefore assume that @ is always positive. Ex, Prove that each of the characteristic equations is only of the first onder when F (2,9 ¥, 4 #) is of the form F(x, yy) +2 (ay +H (2, 952); a IPH, y, 2) 486 (ey 2) +H (Hy, 2) Find the form of (a, x, xs #), when tho characteristic equations aro satisfied identically. 137. As regards the determination of the four arbitrary constants that occur in the primitive, various cases may arise, (i) Two fixed extremities may be given: that is, we have y=y, and 2=2, when #=a), and y=y, and z=% when #=«,. Thus Yom $ (ae, Ay, sy An, Ay) 4-4 (te, Ar, Any As, AO) =o Gn Avy As An ADL” A= 4m An Ay As, Ad) equations potentially sufficient to determine A,, Ay, As, Ay. Gi) An extremity may be fixed in position and direction: that is, the values of y, 2, y/, 2’ are given when =a,. Again, there are four equations to determine the constants. (iii) One extremity may be fixed in position, and the other be required to lie on a given curve y=F@, 2=G(@). At that extremity, say #2, where 2, is not known beforehand, we have Y=F(a,)X, 2=G'(a,)X, so that, from the terminal condition at a mobile extremity, we have s-vE-rLrrotrew@Z=o when 2 =. With F(@)= $ (@y Av As, Av Ad, G (@) = 4 Ar, An Ar, Ad, we have three equations. Also Yo= $ (a, Ay, An, As, Ad, y= (@, Ai, As, As, A,), making five equations in all, potentially sufficient to determine ty, Ay, Ay Ass Ae138] SPECIAL FIRST INTEGRALS 197 (iv) One extremity may be fixed in position, and the other be required to lie on a given surface S(ey,2)=0. Let the extremity be ay, ys, 2, 80 that Slax, $(@), ¥ (a) = 28 y 8 y 28 Also, at 2, while the terminal condition at a mobile extremity is PL! a y x(t-¥3 # of) + rg, +2 for all variations X, Y, ee the surface S=0. Hence, at a, fy vy. ef Ff a ae = oe ee %. a which, with S=0, are three — re the fixed extremity, I= $ (Cy Av Ay, As, Ad, YG Avs Aa, As, Ade ‘Thus there are five equations in all, potentially sufficient to determine By Au Aes Ay Aw (v) Similarly for other cases, when both extremities are mobile separately, ‘upon a curve or upon a surface. % Forms when a general first integral can be obtained, 138. There are some forms, not unimportant, for which a first stage toward the integration of the characteristic equations can be attained, (a) Let f not explicitly involve y, or not explicitly involve z, or explicitly involve neither of them. For the first of these, we have z- 2 as a first integral ; for the second, we oe ¥ z-c, ‘as a first integral ; for the third, we have as two first integrals,198 SPECIAL FORMS OF FIRST INTEGRALS (on. 1v (®) Let f not explicitly involve 2. Then FF Hy Fe Dray tag tit tee b2@yev 204 from the characteristic equations ; and therefore we have f-vp-eBa4, as a first integral. (0) Butif f does not explicitly involve « or y or 2, we do not acquire three first integrals by combining the two cases of (a) with the ease (6); the three constants A, B, C, each arbitrary, are not independent, the relation between them being obtained by the elimination of y/ and 2’, the only quantities which now occur, between the three equations. Thus, if fadsy+ 2, we have A*+ B?4C?=1. (@ The simplest (and perhaps the most important) ease of all occurs, when f is of the form fare yaddtyreeyt =n tyes}, where p is a function of 2, y, s. Then the equations are % ey a oa y sitet s + ae een tyt+2t-— 1 — a tye When the length of the are is made the variable, so that ds a Er atyt sent, these equations are138) EXAMPLES: 199 a . 5 dz te . dy Multiplying this equation by 52, the two characteristic equations by 4Y and & reapectively, and adding, we have reed i)-20-$ ‘Thus de ue=0, and therefore U=0. ‘We thus have three equations, symmetric in «, y, £: they will be discussed later*. Ez. 1, Shew that the shortest curve from one fixed curve (or surface) to another fixed curve (or surface) in space is a straight line which, at each extremity, is normal to the curve (or surface). Prove that the shortest distance between the circle 2*+y!=a%, s=0, and the hyperbola Aa yteaot, 20, in (4400), Ez. 2, Find the integral equations of the curve, joining two points not lying in & plane with the axis of x, such that its moment about the axis of 2 is a minimum, (1) We are to minimise the integral f trntatyt+enh de. ‘The characteristic oquations are yasy set { yyrsank Jao, Gre ae lapse wwe d fe grenh Jao teat de ayeeenl Because f, the subject of integration, does not involve » explicitly, there is a ft integral x, %, Svs bene where ¢ is an arbitrary constant: that is, Yembec(tyt+enh, ‘When this intogral in used, the charscterstic equations become 1 4a Fino, tno Fen, * They will appear (§175) as the characteristic equations when z, y, # are simultancoualy atfocted by small vatiations,200 EXAMPLE (om. 1v Heneo y= co 240 sinh 2 sma’ cosh =+0' sinh = where a, 6, a 8 are arbitrary constants, Substitution in the intagral Gttmbacityt+eyt PB Hed, so that there are four independent arbitrary constants, the five arbitrary constants being connected by this single relation. ‘When new constants a, 8, yare introduced®, such that leads to the relation ancomecen!, dmenineconZ, deccosasinn®, sin a sinh”, ‘which satisfy the relation and now leave ¢, a, 8, y as the four independent arbitrary con- stants the equations of the curve are atB a+8 a acces acosh #8, zmesin cosh Manifesty the ourve is not a plane curve unless S=y: when B=, the curve lies in the particular plane ysin a—2 cos a=0 through the axis of 2. ‘The projection on any plane through the axis of z is, in that plane, an orthogonal pro- jection of a catenary. In the foregoing case 8=y, the curve, in the plane y sina—z cos. fn the catenary yoonartsxinasceosh £2, ‘the projection ona plane, perpendiular oo the axis of, is the hyporbola = ayesine waa ob =? pointe estan ince sinh 227, hich, in the foregoing particular case, bocomes the repeated line y'sina—z00s a=0. (ID) We thus have the characteristic curve. To complete the discussion, so far as weak varttione are conoorel, we anticipate the etabliahinent ofthe further ona, For the Legendre tot (§ 148), wo have ay ie He og otf =o ye Ot art Ho ysy en) 1 _ a OO ae 8 cyte ent— = aya he aye none SY > 0,20 20, > (BL a thre 0 a at Lege tn concerned) the solution admits a minimum. * On the assumption that no one of the constants a, b, a’, Bis zero, such as may happen in. ‘ particular ease.138} EXAMPLE 201 ‘To apply the Jacobi test (§ 149), the conjugate (say 2) of the initial point (say =) is required, It can be obtained as the limiting position of the intersection of the character- istic curve with a consecutive characteristic. Using, first, the form ynaconh £4 buinh?, fcosh! 48 ih, vith the relation BB 4a—Vad, wwe apply a small variation to the constants by taking ated, bieB, ated’, Ute, e+eCs and we denote the consequently varied values of y and ¢ by y-rxY and 2x2. ‘The Yerd cooh 4 Bsinh2— 3 0 («sinh = 4bcosh 2) = 4 cosh 24 Bsinh == cy, 7 2 we. oooh 40 aah? But for the initial value 24 of 2, we have ¥=0 and and for the conjugate value 2, ‘the interseotion of the consecutive characteristic with the original curve, we again have Y=Oand Z=0. Thus, because F=-0 at the two place 22 oy/=0, Aceh 24 B sinh A cosh 28+ B sinh? and therefore 4 B ray! sinh Similarly, because Z=0 at the two places, yank 2! yyemeh ayy) comb 4 gz Sivainb maja = naj coh aya oh ‘When these ar ubsitted in the rlaion ad-0B a4 arising out of the relation among the five arbitrary constants by varying these constants, we have zaysey! —niyleyd + tyxboe! — 2, 2{00) +08 sink © that is, (eo— a) (yop! thx) +0 sinh 210, an equation for determining x, in terms of zy. When the later expressions for the constants are used, we have nit, ¢osinasioh 227, ¥=eos asin} ‘and so the equation determining the conjugate is 225 font asin 22a 142 patent 242 ah 2127 sinh Ano202 EXAMPLES [ca wv ee {ae a Gaon} +1=0, =m oth FB teinta {1 -. (ere ‘he curve becomes the plane curve 48 In the particular ease when 2. n= yconats sin a=ceosh the plane of the curve passing through the axis of 2; the necessary boundary con- dition is % % and the relation giving the conjugate is oth 4B _ cot ate, in effect, the old relation (§ 30, Ex. 1, p. 81) which determines the complete range upon the characteristic catenary. Es, 3. From a given point, in one of two parallel planes, a curve of given length is drawn to some point in the other plane. From every point on the curve a perpendicular is drawn to the first plane, those perpondioulars forming a eylindrical surface, ‘The area of the intercepted portion of this surface is to be a maximum, Prove that the curve must be a circle; and find the relation between the circular aro and the planes, at each of the extremities. Ex. 4. Find the equations of a curve joining two points, such that the moment of inertia of the curve about the axis of = is minimum, when the points do not lie in one plane with the axis of ‘The integral to be minimined is fore Ctyteenbde; the characteristic equations are A42) eyteent 2 {7 e 0, fe ale crea spent 2 eee As J, the subject of integration, does not involve « explicitly, there is a first integral eee soy b-2¥ mconanta, that ig yee aera ‘When this integral in used, the characteristic equations Income 2y (yt422)—at Tao, sey —at Fe138] ‘EXAMPLES 203 OF these equations, there is an integral dy_ods 2 yh, ‘whore cis an arbitrary constant. For further integration, take a new variable w given by sestaaty, 80 that the actual value of u is (1+y?+2%)4, that is, advantage in the specific use of the variable. Then yy tee gat, as ut, at this stage, there is no But GA) +e) my tlh + ey ys and therefore au (ut 1) fatu+ oh Write ae am 0 that-g may be taken as a new constant instead of o: then @ (fase tu-g, the differential equation of Weierstrass elliptic functions with 4 and g for invariants, a ae yersing, r=1008 6, ya where rte; then frm the equation fy St ac, wo have that is, Let + be a value of ¢ such that 9 (¢)=0, so that ~gh=aid 5 then and therefore 2 (0— B)= 240 (7) ~log o (¢+7) 40g @ (¢—2), that is, eo) e(t=r) oo ot) eet204 SUBSIDIARY CHARACTERISTIC EQUATIONS [oH.1v ‘Thus the primitive of the characteristio equations is reaytat yorsing, s=rcos9| Haatp =a PO- 9} a(r—Da(r+o) na HED +0) o¥(7) 03 (0) a(t~r) (o-pymetgin 2) Seco) ‘The four arbitrary constants are 4, a, 8, with either ¢ or + (which are equivalent to one another). ‘As in the last example but one, the equations of the characteristio curve have been obtained, ‘The teste, as to whether it provides an actual minimum, must be deferred ; their formation constitutes an oxercise in elliptic functions, Subsidiary characteristic equations, and their primitive. 189. The characteristic equations are £3 @)mo, xy 4 ay” de ay) ae ae and their primitive is of the form Y=, Ay Ay As Ad, 2 = (@, Ai, Ae, As, Ad, where A,, a, 4s, A, are arbitrary constants. When variations of integrals involving only one dependent variable with one independent variable were discussed, it was found convenient to consider a ‘consecutive’ characteristic curve: and advantage arose for two issues, one determining the limit (if any) to the range of the integral for the conserva- tion of a maximum or minimum property, the other leading to the construc- tion of a critical normal form for the second variation of the integral. The same advantages can acerue in this investigation, Accordingly, we consider a small variation of the foregoing primitive, so that it shall still satisfy the characteristic equations, We submit the arbitrary constants A, 4s, 4s, A, to small arbitrary variations by changing them into Ay+ nay, Ag+ eiy, A,+ dy, Acted, whore a, ay, a5, a are themselves independent arbitrary constants, « being the usual small arbitrary constant. When the changed values of y and z are denoted by y+4n and 2+ xf, we have i) ee a 90 FF a BE a FP ta, OE as A es feast 54, +%5a,+% 54, while the new values y + «7 and s+ xf still satisfy the original characteristic equations.139] AND THEIR PRIMITIVE 205 The function f involves five arguments 2, y, y’, 2, 2. Derivatives with regard to these will be denoted by subscripts 0, 2,3, 4, 5 respectively, attached to f, the notation being used here only for second derivatives as given by the definitions a om fy nfm pan he Si=tu ay and so on, We need the modified form of the equation, under the change of yond ¢ into y+.n and z+ ef, subject to the restriction that the equation of OF OF OF remains the same, ‘The modified forms of 3°, 50, 5f, 2%, are Le ant fant + fab fob) on Be Fan fal tab fab) x te (funt fart + fub + fob) + oy D se fan bhanl +fab+ fab) +o respectively : or, if 20 = fan? + und + ful? +2(fann’ + fant! + fu $+ fall) + Sal + Yolo + fsb, the four modified forms are x, aa Oe x. oF, 20 ay tam ay* a tof Bt era Bgl TH Berens where, in every instance, the unexpressed terms constitute aggregates of terms in the second and higher powers of x. But the characteristic equations are to remain unaltered. Hence of oF ayt* Be. 2 -& H+ 38 oe that is, in the limit when « tends to zero, =0, an d an ‘nae (5y) =P and similarly 0 _ $2) <0. og da ag) ~~ In attaining this result, we have supposed that the original characteristic equations are satisfied, and that therefore the quantities y and z (with their derivatives) are given by the primitive which expresses the characteristic206 PRIMITIVE OF [caw curve; hence all the coefficients fn are to be regarded as functions of «. Thus the two equations just obtained, viz, and fan irae or, in full, San" + fe" =(fa-fa) 0 fen Praia Fatt" + fa" = (Sufi) 1+ fu-Sa—fa' + (furSa) ofa : are two linear equations in and with variable coefficients. On the assumption (§136, Note) of a non-vanishing quantity fs fu—fu’, their form shews that the primitive involves four arbitrary constants. But we have [ee amare tase tse +a, oy a oy ay Seas tag, tb 54, tsa where 4, dy, as, a are independent arbitrary constants. Hence these two equations, when combined, can constitute the primitive of 4 (eM), 4 Ga). te (by) = Se ade @)-°. ‘The quantities y +«n and z+ xf determine a consecutive characteristic curve. Accordingly, as before, these new equations, satisfied by ©, are called the subsidiary characteristic equations. Note. The primitive of the subsidiary equations has been derived from the primitive of the origiual equations. If however that original primitive is not known, but only a special integral is known and has to be tested, the special integral is substituted in the form of the subsidiary equations as they stand; and these equations must then be integrated, without assistance from the special integral except in so far as it may contain some arbitrary elements *, Lemma necessary for the establishment of the primitive of the subsidiary equations, 140. In declaring that the values of » and £ can constitute the primitive of the subsidiary equations, it is tacitly assumed that the four quantities 28 (91, 2 & & ase linearly independent, and that the four quautitics * In the most restricted instance, the epocial integral will contain no arbitrary element ; no ‘assistance towards the construction of the primitive of the subsidiary equations ean thence be Gerived. In other instances, a special integral may contain some arbitrary elements, though not enough of them to makeit the primitive; in that event, some assistance towards the construction of the primitive could then be derived, of w rauge limited by the amount of generality of the special integral.140] SUBSIDIARY CHARACTERISTIC EQUATIONS 207 Beet 2, 8, 4) also are linearly independent, ‘The assumption must be justified, We have seen that the elimination of ¢ between the two fundamental equations leads to an equation of the fourth order, deduced through by" =4(e yy,2 2) 02" = Hayy, 2,2), and arising from the elimination of s and 2’ between Oy” =G(a yy a2), Oy” =K (a, yy, 22), oy” (a yy, 22), where K and Z are similar in form to @ and H, and where @ denotes Safa-fu’. Thus the equation will be of the form OY =Me ASS % where 6 is always positive. Then the customary existence-theorem* provides unique integral of this equation defined by the initial data that, for an assumed value «=a, the quantities y, yy”, y” respectively assume arbitrarily assigned values By, By, By, B,, independent of one another, provided that the function M is finite for those values and is regular in their immediate vicinity. This unique integral is consequently of the form y=By+ B,(e—a)+4By (wa) + 4B (w—a)+(a—a)' R(@—a, By, By, By, By), where R is a regular function of «—a, Thus, as y=$(@, By, By, By, BY, we have #- 4a +(e-ay ot, Sn be—ays (ea Clearly no linear and homogeneous relation of the type ae 2 a ba bat +5 thse =O exists among the four quantities, the coefficients being constants; and con- sequently the expression for is satisfactory as containing four (and not fewer than four) independent arbitrary constants. * Soe my Theory of Diferential Equations, vo. ii, oh. xi, oh. xiii208 PRIMITIVE OF SUBSIDIARY (on. 1v Similarly, when we eliminate y between the two fundamental equations, we obtain an equation of the fourth order in 2. Of this equation there exists a unique integral, defined in the same manner as the foregoing integral y, by means of arbitrarily assigned initial values for 2, 2, 2”, 2”; the integral is regular in the vicinity of #; and it has the form 220.4 0,(@—a) +40, (0a) +401 (2—a)?+(e—a)'S (ea, Ox, Cx, C2), where S is a regular function of «—a. As before, there exists no linear and homogencous relation of the type he et a0, * *a0,* “ac, +80, where the coefficients ¢ are constants. In order that the value of y and the value of z thus obtained may be simultaneous integrals of the equations Oy" = G(x, yy, 4,2), Oe" the four independent constants B,, B,, By, B, and the four independent constants C;, C,, C, C, must be brought into relations with one another, 80 as to leave only four independent constants in all. ‘To indicate these relations, we can proceed as follows. ‘The equations can be written in the forms =H@yya7) y=P % Of =A Bs OY =H P.4D5 set of four equations of the first order, Here, @ never vanishes within the range allowed to the variable «; and the functions @ and H are regular, for values of «, y, p, 2, q, in the vicinity of a, By, By, O,, Q, respectively. By the existence-theorem applying to such equations, there exists a unique set of regular integrals y=B,+B,(#—a)+4B,(e—a)+tB, (2—a)+(2—ay R (x ~ Just, then fy and fag must have the same sign. The statement is made as in the text, in order to leave the actual sign of fay and fas in more direct evidence.149] JACOBI TEST: CONJUGATES 219 where p=iv, 0, wi, v, wl, jm om m &| lam be hom fl and where the quantity 7,£2— maf, does not vanish in the range. So long as p and q do not vanish together—and, of course, simultaneous zero values of » and w through the range are excluded for the present aim— the quantity A remains of uniform sign. But A would vanish if p and q could vanish simultaneously—a contingeney to be excluded: and this could happen when v=pmton, w=plit ot, where p and # may be functions of 2, if np’ +mo'=0, fip' + ba’ =0, that is if p’ and o’ both vanish so that p and o are constants, Now v and w are to vanish at the lower limit, while xy and xf represent the variation to a consecutive characteristic curve. For the present purpose, we take any such consecutive curve to begin at the same point as the original curve now under examination; hence, at that lower limit, we can take pn,+on,=0 and pit of, =0. If, after passing that lower limit, pn, +on: and pf, + 0%, can again vanish simultaneously within the range or (what is another mode of stating the same possibility) if the range of integration should extend as far as (or further than) the first succeeding value of where pm+on, and pt,-+ot; vanish, then we have a possible variation v=pn,-+ om and w=pf, +6, such that (i) v and w vanish at the lower limit and at the upper limit, (ii) v and w are variations that are not zero, and (iii) p and q vanish along the whole range. ‘Then we should have A=0 throughout the range for a certain non-zero variation; and we could not assert the existence of a maximum or a minimum for the integral. Accordingly, the contingency, which admits this result, must be prevented: that is, it must not be possible to have a variation (pm+on,) and «(pf,+o6,), vanishing at the beginning of the range and also at the end of the range (or earlier than the end of the range). Such a variation represents change to a consecutive curve: and the vanishing of such a variation can take place only at an intersection of the original curve and the consecutive curve; hence the range, beginning at any point of a characteristic curve, must not extend so far as the first point of intersection with a consecutive characteristic curve drawn s0 as to meet tho original curve. Such a point of intersection (if any such point exists for one con- secutive curve among all the consecutive curves that can be drawn) is called the conjugate of the original point which marks the lower limit of the integral; and consequently we have the third (Jacobi) test for the existence220 PROPERTY OF [caw of @ maximum or a minimum:—the range of integration, beginning at any point on a chiracteristic curve, must not extend as far as the conjugate of that initial point. We proceed to construct the analytical expression of this limitation upon the range of integration. Property of a fundamental group of integral-sets: covariantive function. 150. We write Bo HB, B60, 6, Ho, #00) a oF nO nO nO. ne so that $,(2) and ¥, (2) (for r=1, 2, 3, 4) constitute an integral-set of the two equations Son + fal =(fo-fan—- Sal + Su Fu 8+ Sos fu Fos EY Sat" + fab" = (Safa) + (Fu-Sa—fis) 0 + Su fa) b= fey Let W=| file), fle), bz), @ |i hE), WO, We), we) He, &#@, H@), Wo [WO WO Ww, We) +/ PO +b)» d(@) , d(@) then AW _|h@), h@), He), d@) & AO, Ke), h@, W) WO» We(2), We), We(@) Hi" (@), Go" (ew), a" (a), Gu") || a’ (@) s Ge (a) , Ps (@) , di (@), IW’ @) Wi), Ww), WA] LW’ @)s We" @, WO), W'(@ ‘The two characteristic equations, on being resolved for $,”(a) and " (2) a8 values of 7" and £”, give Od:" () = (—fa' fst fos fos — ffs SuSs) by (2) + linear terms in $, (2), Yr («), Wr (2), OMe" (2) = (— fafa! + fafa! — Safa t fubo) Vr (@) +linear terms in $,(«), ¥+ (2), $r' (2), where 0=fsfs—Ju’. When these expressions for ¢,”(a) and y;"(c), for 7 =1, 2, 3, 4, are substituted in the two determinants in the foregoing value and aw : for Ti 1 we have 0S (fob fut af Saft fafe) W +(-fafe + fafa —fafat Suto) Ws151] FUNDAMENTAL INTEGRAL-SETS 221 and therefore aw 07 W, that is, ow=A, where A is a constant, The Legendre test is to be satisfied; thus @ is positive through the range of the independent variable. Again, the quantities ¢,(a), for r= 1, 2, 8, 4, are linearly independent, so that there is no identical linear relation along them; and likewise there is no identical linear relation among the quantities ¥,(z), for r=1, 2, 3,4. Now constants & can always bbe chosen go that, if 1 = hyd (0) + keds (0) + keds (0) +h dy (2), £ = Kay (@) + kate (@) + kas (2) + hae (2), a =I, by (w) + ka fy (a) + ka ds (w) + bes $s (@), ‘n, 697 vanish for any assigned value of a, But Kal (@) + harps (2) + haya (2) + Bas (a), being ¢, cannot vanish for that assigned value; because then the differential equations would require =0 and {” =0 and, by successive differentiations, would make every derivative of and every derivative of £ vanish for the assigned value. The quantities » and £, given as expansions in the vicinity of the assigned value, would then vanish everywhere in that vicinity— contrary to the property that the set of quantities ,(2) and the set of quantities ,(e) are, each of them, linearly independent, among themselves. Hence W cannot vanish within the range; and therefore the constant A is not zero. 161. Further, this property does not depend upon any particular choice of integrals $, and 4, (for r= 1, 2, 8, 4) of the subsidiary characteristic equations. If any other complete set », and ¢, (for r=, 2, 8, 4) be chosen so that See b= enyps (2) + emPs(@) + Cahs(2) + Onvra(@))” where the determinant D (oy, ¢e, és Cu) must be different from zero, we have W (n, 8) = Wb, W) D(Cns Cur Cas Cu) OW (9, 6) =A’, where the constant A’ is not zero. ‘The determinant, W is, in fact, a covariant for a fundamental system of integrals of the subsidiary equations. Note. We have the corollaries : (i) The fanetions $, (2), ¢4(2), $s (2), (2) cannot simultaneously vanish, nor can the functions (2), ¥u(2), ¥s(e), ¥i(a) simultaneously vanish ; and and therefore222 ORITICAL FUNCTION FOR [on. wv (ii) The derivatives $,'(z), $y’ (a), $y'(@), $'(#) cannot simultaneously vanish, nor can the derivatives yy (@), py (©), Ws (@), Wi (@) simultaneously vanish. Critical function for the Jacobi test. 152. We take one set of integrals of the characteristic equations =, bs (2) + Ba ha (2) + ba bs (0) + bi ha (a) = She de (a) Si = dit (@) + Daye (2) + bys (2) + baal Debra)" and determine the constants b,, by, b,, 6, by the conditions that m=0, &=0, m=, f'=0, at ©=4,, where 2 is an arbitrary constant, and W,,= W(x), is the value at @, of the function W (of § 150) which nowhere vanishes. Then Lbebe (a) = 0, Eber (o)=0, Ede Ge! (a= AWo, Eber’ (a) = 05 consequently baat Kaa (80) + Bests! (20) + Kees (2), bad (hea (eo) + habs (a) + kai)» fear’ (2) + haps (a) + kus @)), Uap (0) + heaps (0) + aay (ee) h where | (ea). a ta, ba (tas Ge () |] Hin Fins Bir ay Ba Be Lalo, Yale), Ya Fela) We take another set of integrals of the characteristic equations nem Cups (©) + ads (2) + ds (©) + 04 ey Leah) tava) tay) + ela)!’ and determine the constants ¢, ¢,, ¢s, ¢, by the conditions that =0, t’=-aM, at 2= 2, where wis an arbitrary constant. The set of integrals m and £ is manifestly independent of the set of integrals », and {. Then Lerge (a) =0, LeWr (ee) = 0, Lerbe (m=O, Serpe! (a) =— eos consequently m=0, &=0, a { Kaudh! (@) + Keats! (te) + hehe (te), Hah’ (ao) + kads’ ©) + kisi (m)}, (Rsudh' (a) + Kas (aa) thai (a) (His u’ (a) + ken ha' (ae) + Kesh (aa) I152] JAcoBr TEST 293 We write A(x, 2)=| G1 (ao), haa), Ga (a), a (ae) Hi), Fal), Fale), Yelm) Hi(@), bale), dale), (x) WO), Ww), we), we) is (2a) Hes (2) + Hes (20) ne (22) + Ke (a) a (#0) Hie) Kas (©) + Ks (te) Koa (2) + Bie (te) as (@)5 also we write Wo= W (aa) = kaalas + alae + healer + kale + hauls + haulray Hi (@), Gal (od, Gs! Ce), Bal (Ha) |] = as bay es bas bow tue Wi (me), Ya (a), Ye (0), We (eo) |) ‘Then we have mabe ite= (bebe (2)] [Zeer (2)} — [Ebrhr (@)] (Berdr (a)}- On the right-hand side, the terms, involving by, 1, bs, cx, gr (2), Fi (a), $2 (2), ‘Ya(2) alone, are Now where (iea— Baer) {fr (©) Yo (@) — Ya (2) a (a). $:(@) Ya) Hr (@) b1(@) = hal); and ~ babe) Uhh (ta) + haps (4) + abs (2)} hut (22) + Rad (0) + ind (0) = tha (2) + has) + hah (ea) (0) + hap (Os) + hai ()} wag heal ala Balt bal + Bal) + (bab habe at Ika + Rhea Kaye = 05 and therefore 7 = 5 hha) alt) We ‘The other terms can be arranged in similar groups, five in number; and the final form of $,— mf, is pe Wo {ie (a) Kaa (Wty) + «+ + Kea (0) Kae (0)} rw, A (sy, 2). ‘The function A(z, @) vanishes when «=a; let a, be the first value of « greater than 2, for which it vanishes, so that A(t, a) =0. ‘Then the quantity mf—nsfi, which vanishes at the initial value a, of « and is not identically zero, afterwards first vanishes for a value of z. mbe— mb224 JACOBI LIMITATION OF RANGE {cu. 1v 153. Now consider a variation v and w, such that mn +on, w= ph tot, where p and o are non-zero constants. At o=.,, we have , wad, v=pr wi=—op, where pd and oy are non-zero constants; thus » and w constitute a non-zero variation of the characteristic curve. If it is possible to have a consecutive characteristic through the initial point «, meeting the original again in the subsequent point 2, (or, if there be other points of intersection, in the first subsequent point which will be denoted by 2,), then at that point we have oa v= v=0, w=0. Because p and ¢ are not zero, we have, for ©= a, mb—mb=0, and therefore A (es, 2)=0. Bat the range of the characteristic curve, when it begins at 24, must not extend so far as the conjugate of that initial point, that is, not so far as the limiting position of its intersection with a consecutive characteristic through 2. Henco the range of integration must not extend s0 far as the value a, where a is the first root of the equation A (e,, 2,)=0 which is greater than zy. We thus have the analytical expression of the geometrical limit, which restricts the range along the characteristic curve to lie within the eonjugate of the initial point, Note. Ono assumption has been made tacitly—that, if A, be the con- jugate of an initial point A,, no point can be chosen in A,A, so that its conjugate shall also lie in A, 4,. The justification of the assumption will be deferred until the discussion of the variations which allow variation of 2 as well as of y and = Summary of teste, 154. ‘The results thus far obtained for special variations may be sum- marised as follows, when an integral i ie y x 2, #)da is to possess a ‘maximum or a minimum: L (a) The variables y and 2, as functions of 2, must satisfy the two characteristic equations fay #_4(%)_o. zZ ay) = £-(f)=95 and quantities 7 and £ must be determined as the primitive of the subsidiary equations eon 20) 0, 20, i) 0, 3m da\aq') =" GE ae \3e') = where 02 is a homogeneous quadratic function of 9, 1, ££’.154) SUMMARY OF TESTS 225 (®) The primitive of the characteristic equations is of the form Y=O(@, Ar, Ae, Ay, A), 2= (2, Ay, Ay, Ay, Ad; and the primitive of the subsidiary equations is 9= $4 (2) + 04$4(2) + $s 2) + 0460(0), Saar (2) + daa (a) + aayps (ec) + aya (e), where y(2)— 26 and Yr (2)= for rm, 2, 3, 4. II, The quantity ef ef (ey ayaa (ayer, rust be positive through the range of integration, the common sign of x and a (persistent through the range) being positive for a minimum and negative for a maximum, LIL The range of integration, beginning at # value a, must not extend along the characteristic curve as far ns the conjugate of xy, this conjugate being determined by the least value of z,, greater than 2, which is a root of the equation AG, %)=| Gr(m), br(%), b(%), d.(@) | FC), We), Vole), falas) H(i), dul), Gale), dul) WG) Hila), Yala) vq)CHAPTER V. INTEGRALS INVOLVING TWO DEPENDENT VARIABLES AND THEIR FIRST DERIVATIVES: GENERAL WEAK VARIATIONS. ‘Tue MeTHOD oF WEIERSTRASS. 185. We now proceed to consider the effect of general weak variations, under which a, y, 2 are subjected simultaneously to small arbitrary variations, so that they become #-+«u, y + xv, 2+xw respectively. ‘The quantities a, y, 2 are now made functions of a new independent variable ¢, which is supposed to increase throughout the range of integration; u, v, w are arbitrary regular functions of t, 80 that they (as well as their derivatives) remain finite and con- tinuous over the range. Denoting derivatives of #, y, ¢ DY 2,35 -+us Yr Yor yy May vee, We have Bat a a , #) dz, when transformed, become y Le the integral [7(e, vy T= foley sens 2) a where 90,9 8.2 BLE I Is 2 186, Now this function g satisfies a single permanent identity, whatever be the form of the function f (and therefore of g), because g involves six arguments, while f involves only five. We have Ae @, 02” a 32,” a2” and therefore : oe oe ae, tH 9g, TH ag, — LHD the identity in question, It can be obtained also as follows, the method leading to other relations. ‘As there is nothing to specify any requirement concerning an independent variable, we may substitute any other variable 1’ for ¢; and then, if dt = pa? where « may be variable or constant, so that de dy _ dz_ gpa hte gpa ele gpm ee156] GENERAL WEAK VARIATIONS 227 we have de dy di 9 (%1.4 Se Bh, Sp) AP =9 (0, 82, 8) ab that is, Let I Yr 2 pir, MY» HE) = 1 (8 Yr 2 Hy Yay He a Bi, 7 Hh 4 Bin Thy) = (Ys es Yor Bs a By 7 YH ir Hr £1) = IaH, 8 Bs Yo Bae a Ba, FW Bis Yor 2) = Ie (@ Hs 2 is Yr Hi then I Yr % Mis MY» bes) = Is (BY 4 is Yrs Be), Gus Yo 2 Her, Myr» ME) = I (2 YB Wy Yrs &)y Ia Ys % BY, Ha» ME) = Ys (@, YB ay Yor 4) Various corollaries may be derived. () The first derivatives of g, viz. gr, gs, go, are unaltered when 2, ys, 2 are multiplied by any quantity 4, variable or constant (but not zero). In. particular, let 4% =1; then “Hence the first derivatives of g are unaltered when we make the are-length # to be the independent variable, so that ROH HI =A Yo 0, Y', 2); and similarly for g, and g,. (ii) In the equations for g, 9, gx, gs, let 4=—1; then if the simultaneous change of sign of «,, y1, # does not affect any radicals, we have IAG B— Yo A=—G (GH AH, HH), IAG By — I~ H) = GG HH By Hs HH), 3 (ili) Let w=1+6, where ¢ may be constant or variable; and take the expansions, in powers of ¢, in the relation GG Ye 2 MEiy WY HE) = MY (®, Ys 4 Bs Yay 2). When the coefficients of e are equated, we have a ge, tag t 5g I the identity satisfied by the function g.228 RELATIONS AMONG DERIVATIVES OF INTEGRAND [ow. v Relations among second derivatives of the subject of integration. 157. This identity (as in the earlier instances in § 51 and § 105) gives rise to certain relations among the second derivatives of g with regard to its arguments, one sot of them being necessary for use almost immediately. ‘When we differentiate the identity partially with respect to a, y,, # in turn, we have Agat YGut A Gu=0, 2 GIat Gat AJa=0, ©, 90 NGut %Ga=0, where differentiations with respect to @, #1, y, a, 4 #1, are denoted by sub- scripts 0, 1, 2, 8, 4, 5, 80 that eg orien 8003," and so on. ‘These three equations shew that all the six derivatives can be expressed in terms of three of their number, or in terms of three independent quantities. ‘Thus, if we write gu=Ayte? , gu= Beene , gu= Cnty: } gu=Patyc, gn= Gaye go= Hoye? we have A+H+G=0, H+B+F=0, G4+F4+0=0. Selecting appropriate trios, we have A=B+0+2F) B= a O=A+B42H -H=B +F}, -F=0 en 0+ PF) -H- a Fo B+ Also we have —2F: +0-A =-2G=C+A— | —29H=A+B-C together with relations BC-F'=04-@=AB- } (BC +204 + 24B At B02), ‘One case (§ 156) arises by taking eat, m=1, nay, ane. ‘The critical conditions in the Legendre test (§ 148) were that gs and gs, should have the same sign while gugm—Jut should be positive: that is, the quantity BO-F158] FIRST VARIATION : CHARACTERISTIC EQUATIONS 229 must be positive. In the circumstances, itis clear that A, B,C must have the same sign; but this condition is not sufficient to secure all the foregoing tests, and there is the additional condition 2BC +204 +248 > A+ B+ 0 ‘We shall return, later (§ 162), to the relations among the second derivatives, First variation of the integral : characteristic equations : terminal conditions 168. Let J denote the value of the integral, when a small variation is effected upon 2, y, z, 0 that Tm fg leben y te 2+ eM, +, Yb, 6+ et) dt =Lteh+}el+K, where eZ, and 44°F, are the respective aggregates of terms containing the first power alone of « and the second power alone of «, and where K; is the aggregate of all the terms involving the third and higher powers of x. The ‘first’ variation in J — I, viz, «l,, governs the value of JI, unless Z, actually vanishes. Tf I, does not vanish, J—Z can be made to change its sign by changing the sign of «; and the integral I could not then possess a maximum or a minimum value. It is therefore necessary that, if I is to have either kind of value, the quantity Z, shall vanish, and that it shall vanish for all arbitrary small variations, We have [2 Fy gv Oy, Ba [ero Be wl mB 4 2 + 0,20) a = [nav Z ro] +[[+8-a@)}+-a G+ 8-4 @h] a where the terms in the first line are to be taken at the limits, Consider, first, variations which vanish at the upper limit and the lower limit alike: for them, I, consists solely of the integral in the second line in the last expression. Then, the functions u, », w are arbitrary, and they are independent of one another ; and for all such functions, the integral (being the value of Z, for these variations) must vanish. Hence we must have B= _ 4 (2) =0 ag _ d (ag ray ay) ™9230 TERMINAL CONDITIONS [on v For, if they did not vanish, we could choose u to have the same sign as E or the opposite sign, » to have the same sign as Ey or the opposite sign, w to have the same sign as H, or the opposite sign; and the freedom of choice in each selection is unaffected by the choice in the other selections. Such choices could make I, positive or negative at will, a result that is excluded; hence the choices must not be open, though w, v, w remain arbitrary: that is, E,, Ey, E, must vanish separately. Next, for J, consider a small variation, which is zero at the lower limit but not necessarily zero at the upper limit. ‘The integral in I, now vanishes, owing to the equations H,=0, Hy=0, H,=0. Hence, in order that J, may vanish, we must have yy we uo +a =o at the upper limit, If the upper limit be fixed by assigned data, the require- ment is automatically satisfied because u, , w then vanish, If the upper limit be not thus fixed, then the relation a uf oD ew Po provides a condition or conditions at that mobile upper limit. Similarly, by considering another variation which does not vanish at the lower limit, we must have at the lower limit. The requirement is automatically satisfied if the lower limit be fixed by assigned data, because u, v, w then vanish. If the lower limit be not thus fixed, then the relation a Bros wi ao provides a condition or conditions at that mobile lower limit. u ‘We thus have two kinds of requirements, necessary and sufficient to secure that J, shall vanish for all weak variations. Firstly, the equations By=0, By=0, Ep=0 must be satisfied everywhere in the range of integration, Secondly, the relation a ur o sw a0 must be satisfied at the upper limit and at the lower limit, separately, for all admissible variations. The second set of requirements will, directly or in- directly, determine the limits of the range of integration when they are not actually fixed by the assigned data.159] DERIVATIVES CONTINUOUS THROUGH FREE DISCONTINUITY 231 Accordingly, we shall now consider the limits as known; they will no tonger be subject to variation, when further investigations (in particular, concerning the transformations of Z,) are being pursued. The threo current equations are called the characteristic equations. They determine «, y, £ as functions of the independent parametric variable ¢; and we shall assume that these functions are analytic, ‘The elimination of ¢, if desired, gives two relations between 2, y, z, free from the unessential variable t. By either form of primitive, a curve, usually a skew curve, is determined : it is called the characteristic curve. Continuity of certain derivatives through free discontinuities on the curve. 169. One property of the function g, the extension of the property § 47, 102) possessed in the case when there is only a single dependent variable originally, can be established. It is as follows: In passing through a free® point, where a discontinuity in the direction of the characteristic curve occurs, the quantities i ae, By? de, are continuous in value. With the same notation and the same figure as before (/.c) to consider the integral near such a discontinuity at a value 7’ in the range between % and f, we take a variation of the characteristic curve, such that u=0, v=0,w=0 along ff and also along t,t. ‘The are of the curve (though not the direction of the are) is continuous in the range 47h, so that we may assume u=(-D(t-4)®O, v=(4-)C-4) VN, G-O(¢t-4) XO, where (0), V(t), X(@) are regular functions of t, which do not vanish at 7 and otherwise aro arbitrary. In the range 7%, and indefinitely near 7, we denote the values of the derivatives of g by G2). QD.» @),» in the limit as ¢ approaches 7; and, similarly, in the range 47° and in- definitely near 7’, we denote their values.by G2). Gi @ in the limit as ¢, approaches 7, The integral J, is to vanish for all small variations, and therefore for the foregoing variation. Between t, and f, the contribution to Z, is zero, because * That is, a point which is not fixed by assigned dats.232 ‘THREE CHARACTERISTIC EQUATIONS fon. v u,v, ware kept zero in that part of the range. Between ¢, and 4, the con- tribution likewise is zero, for the same reason, ‘Thus *pfule-@ ao @ 2 (Broo (¥ Si+fe [¥f2 a} +" ay 5 (2) +0 t3~ ai \ae)S | a re [ 29 4 y 9 4 9)" + [Meow Pa [nile ew)” Now JZ, must vanish. ‘Tho integrals vanish, because the characteristic equations are satisfied everywhere along the curve. ‘The part, outside the integral and taken at the limits t and 1’ of the portion t,7' of the range, vanishes at f; but not at 1°; and the part, outside the integral and taken at the limits 7’ and &, of the portion 7, of the range, vanishes at f, but not at 7. ‘Hence we have va (@E)_+ee(QB)_er() — fer GB), ree). vee G8 Now wr, v7, we are arbitrary fanetions; they are independent of one another; ‘and, because the place 1 is free, up, vx, wr are not required (by assigned data) to vanish. Hence the equation can only be satisfied for all such arbitrary functions, if @-@. @.-@, @.-@). being properties which constitute the continuity of the three derivatives in value, on the passage through the free point of discontinuity in the direction of the characteristic curve. ‘The three characteristic equations are equivalent to two. 160. We note at once that, in virtue of the identity satisfied by the fanction g, the three characteristic equations in § 158 are really equivalent to only two independent equations. When the complete derivative of the identity with respect to ¢ is formed, we have aS (®) +h (2) +95 CD +egh +m = a 9, 0, H,, =a F442 2 7 oat tay tpg tay, TH160] EQUIVALENT 70 TWO EQUATIONS 238 and therefore Ea + By +E,= 0, a result which establishes the statement. But, owing to their convenience in the subsequent analysis, the three equations Z,=0, Zy=0, H,=0 will be retained for use. Further, when the three characteristic equations are combined with the identity satisfied by g, the fact that they are equivalent to two equations can also be indicated as follows, We have (2) But YG + A.Go— (Guo + Yaga t Gut Cut YoGie + £98) = th (Gea — Gx.) — 21 (Jue Jeo) ~ HJ — Yin — Feiss and therefore Eg = 245 (Gos Ga) — 4.2, (Gua ou) + £5 (Yin + #igs) — Baa — 2s = (29 Ger— gx) — (aya 204s) Gs} — {0141 (Yu — Jue) — (2s — 222), Similarly a and = {Ys x — 9a) — (Yee Yee) Ju} — {Ys (Goo Gx) — (Ys 4h) Jub 4B = (021 (Gu ~ ga) — rte 22%) Gus} — {Ys%i (Gu — Jo) — (Yaa Yori) Gus} ‘Thus, again, we have, on addition, By + By +E, Also, because Z,=0, H,=0, E,=0, there is a quantity T' such that 41 (GaGa) — (YF tact 448 Ga~ du) — (4% ~ 4%) a= TP: 2193 (Joa Ja) — (G1Ys— Ma) Ju=T relations which can be written in the form 4121 (a Gu— R(@iyo— nay) + Q (am —a14,)} 9G + P (ys yaar) — B20 — 2 20)} 21 (Jus — Gua) Ex, Show that the two characteristic equations of § 14 are Fg fs a9) 1181 ag) — (624-421) Gat Oi4-IH) 990, Fy, itn G98) 115 Ons Ie) ~ (Sir Ha) 984 1A Yet) HI =O.234, PRIMITIVE OF CHARACTERISTIC EQUATIONS [ou v Primitive of the three characteristic equations : ubsidiary equations. 161. The three equations E, = 0, Ey=0, E,=0 are equivalent to two independent equations when combined with’ the identity satisfied by the function g. Their primitive is of the form XE A, ay tay Oy, cs) PEF a4, M4, 3, ZEY EFA, Oy Oey ty, 4), where 4, 4, 4, as, % are arbitrary constants, one of them obviously occurring as additive to t because t does not occur explicitly in the characteristic equations. The constants a,, a,, 04, a in the primitive are essential ; but a is merely incidental. These equations are satisfied identically (for all values of the arbitrary constants) when «, y, 2 are substituted; and they remain satisfied when the arbitrary constants are changed. Let these be subjected to small variations, 80 as to become a+ Kal, a+ xa, (for r= 1, 2, 8, 4), where a’, a, dy, dy, dy are themselves arbitrary; and let the consequent small variations in 2, y, # change these variables to #+x&, y+ «n, 2+ xf respectively. Writing XO, $0, WO, for %, SB, HHO, $400, Held, for H, 3, oe 1,2,8,4); we have ‘ Faax (0 + 3 axe (0 aap + 3 aby. taay' +d ate 0| Further, we take 4 O=xX Ob O-8' O x04 %W=xX OW O-¥ OxO)” for r= 1, 2, 3, 4. We now form combinations of » and £, ¢ and £, such that = a1 — WE = a8; () + 00, (t) +0585 (0) + 0484 a Z mab - nF = a9 (+ Gs () +495) +a)” the arbitrary constant a’ no longer occurring in the magnitudes thus selected, With these changes effected upon the variables «, y, in the equations E,=0, E,=0, E,=0, the actual characteristic equations remain the same.161) SUBSIDIARY EQUATIONS 235 Thus the changed form of Z, is still equal to Hy, and the same holds for E, and for E,. We write 20 = Go + Gat? + Jul? + Wak + Woukl + Wank +2gukE' + 2akn! + QguskE + 2gank’ + 2gamm! + 2gunk” + gah’ + 2gakn’ + 2k" + gn E*+ gr! + Gab? + 2gisk'n' + 2gskS + 2gun's 5 and we note that ¢ does not occur explicitly in g, so that the change of @ into a+ nal does not affect g in the small variation except in so far as it enters into the expressions for fn, £ Then the changes in the first derivatives of 4g with respect to @, 2, ys Yu 4 #1, make them become a a ao aE" an, * "OF? 9,22, ow , 20 ay Tan” ay,” Gx” ag, 92 0% setae oa, * ag” respectively. ‘The three characteristic equations are unaltered by these changes: hence - y= 22 _ 4 (29) OE ale) og Rte ‘These three equations, which clearly are linear in £, », { and their deriva- tives and are of the second order, are called the subsidiary characteristic equations, Their variables £, 7, Care such as to keep #+ KE, y+ xn, 2+xb a8 integrals of the original characteristic equations. Moreover, we have @, B, + y Ey + 2, =0; and therefore (e+ 46) Be + Be) + (ys + 0m) (Hy + 0) + (21+ 06) (Bs + Eo) where only terms of the first order in x are to be retained. ‘Thus nBpt Ey + Bc=0, so that the three subsidiary equations are equivalent to only two independent equations, Accordingly, this relation must be satisfied unconditionally, what- ever forms be given to the coefficients gin. In placo of the three equations, we shall substitute two independent equations having Y and Z as their dependent variables; but, in order to secure guidance as to their form, we shall effect an initial modification in the expression of the ‘second’ variation in our integral Z.236 FIRST MODIFIED FORM OF [ou v The ‘second’ variation : first modified form. 162. Since the ‘first’ variation I, has been made to vanish, thereby pro- viding the characteristic equations as well as terminal conditions which settle the limits of Z, the governing magnitude of J—J is the ‘second’ variation te fs (@dz, where = guu?+ gud? + gow? + gawd’ + Iga’ + gar’! + 2yquu! + Agee’ + Qgaurs! + 2gnvu + 2gq0v + 2gq0" + 2gqwu' + Aggro! + 2ggwu! + 90%? + ga? + gut + 2gqguv + 2gu urn + 2pum0. It will be noticed that © is the same function of u, u,v, v, w, w' as 20 is (G161) of £, &', 9, 7, & {In both of them, it is assumed that the values of , y, #, which are determined by the characteristic equations, are substituted in the coeflicients Jim. It has been pointed out (§157) that the six coefficients gu, uy Jur J Ju» Ju can be expressed in terms of three quantities initially defined by means of some selected trio among them. No preferential dependence is accorded in the foregoing analysis to a selection of two of three variables @, y, 2 over the remaining variable, But as convenience can emerge from a choice of y and z, because we then recur to the earlier analysis of § 130—154 on making « =t, we shall select gu, gus, Jus a8 the three quantities of reference for the six coefficients. We take* Gn=23P, gu=07R, ge=2'Q; and then the other three coeficients are expressible by the relations Iu=— HYP HAR gum = ay,R- 029}. n= AP 2s, R+oQ In order to obtain the first modified expression for the second variation Is, we take a quantity Dm Aw + Bo + Cut + 2Fow + 2Guu+ 2Hwy, where A, B, C, F, G, H are functions of t at our disposal. As the limits of the integral J now are fixed, the quantity 7’ vanishes at each limit, because 1u,v, w vanish separately at each limit, + Tein eny to verity that ee ereP gpapaeeh Gane. to that, when 2=t, the quantities 7, R, Q are tho critic quantities which arise (§ 148) in ‘the discussion ofthe special weak variations163] SECOND VARIATION 237 With these values, the terms in @ of the second degree in w’, ’, w’ are, together, equal to Pla — yw + 2B ey — yu) (ew! — yt) +Q (aw au’). Accordingly, let Vand W denote the quantities so that ee rad Vi sav — yw +20— you ‘We proceed to choose the coefficients A, B, C, F, @, H in the quantity 7’, s0 that, if possible, we may have 042 prs 2Rv w+ QR +2KVV'+ 20 WV" 4 2MVW'+ 2NWW'+ DV? +280 W + EW, 163. In order that this relation may hold, the coefficients of the various combinations of u, x’, », v’, w, wo’ must be the same on the two sides, (® The coefficients of terms of the second degree in u,v, wu’ are the same, Decause of the expressions for gis, Gur Yor Ju Shes Jue (ii) In order that the terms in (w, », wifu', v', w’) may be the same on the two sides, we must have Gat 0 =Qam + at ) Jut+B ayy + aK Jot F = Raya, +a3h Gut F = Raye, + 03M Gat G =~ Qeyts— Rye — ayrb ~ 02,N Ju G =—Qu.2— Ranye—oy,M— 220 Yat H=— Pym Rae 2 y,K- eM Pay, — Roe, my K ~ a2, Pyset Byer + we) + Qt +y2k + y.5(L+M)+22N (iii) In order that the terms of the second degree in u,v, w may be the same on the two sides, we must have gat O' = Qe? +2n,2,N+03E Jat B’ = Pas +20,2,K +03D Sut F = Rap + 22, (L+M) +038 Gut GF =~ Qtat— Ra yn— mYy,L — yrtyM — (m2, + 2,0) N —a,y,S—0,2,E)- Gat Hl =— Pry, — Raat, — (yet he.) K — 2,0,L - 0,%M— myD—a2,8 Sot Al = Pyd+ Ryser + Qe + yy K + Qyyl + aM +2.%N FyeD + 2y28+ oA238, ‘TRANSFORMATION OF SECOND VARIATION [cu. v Here, there are fifteen equations in all; and they contain thirteen unknown quantities 4, B, C, F, G, H; K, L, M,N, D,8, E. One of the two equations, involving F in the set of nine, can be removed if we retain 9u—Go= 23 (M-L); one of the two, which involve @ in that set, can be removed if we retain Goa Ju Q (eres — 2144) + B (ysete— nye) — 9 (M— LZ); and one of the two, which involve HT in that set, can be removed if we retain Ga— Ga =P (Yrtta— tiye) + B (ey 2424) + 0,2 (ML). But any two of these retained equations are satisfied in virtue of the third, because of the two fundamental and independent characteristic equations as given in § 160. Hence the total of fifteen equations must, in number, be reduced by two units in counting the tale of independent equations, Thus there are thirteen independent equations, potentially sufficient for the deter- inination of the thirteen unknown quantities which occur linearly. 164. For convenience, all the fifteen will be used in the reductions, account being taken of those which ultimately are independent, viz. six out of the first set of nine, together with the one retained equation, and the six in the second set. In another aspect of the aggregate of equations, we may regard the set: of nine as defining quantities A, B, C, F, @, H and leaving one relation, free from these quantities so defined : the other set of six, together with this one relation, can (after substitution for A, B, C, F, G, H) be regarded as deter- mining the quantities K, L, M,N, D,8, B. Ibis, however, unnecessary to rosolve the equations, and obtain the explicit expressions for all these quantities A, B, ...: provided the one aim of the investigation is the construction, explicit and complete, of the final normal form for I; This form is obtained, in terms of the quantities P, R, Q, and in terms of the integrals of the subsidiary characteristic equation. Moreover, the primitive of this subsidiary equation is derived from that of the character- istic equation. ‘Thus the actual knowledge of these quantities A, B, ... is not required for the construction of the normal form; it is ancillary to the con- struction of an intermediate form, which itself is of transient importance. ‘The process adopted in the present case is, in fact, a fusion of the two processes in § 52 whereby, first, a form a aan Puss Gur F was constructed for @; and, next, a form P(w—u) +9 (0+ Hw) was deduced.165) SUBSIDIARY EQUATIONS REDUCIBLE IN NUMBER 239 If these separate processes were desired in the present case, the first stage is attainable by keeping the expression for 7’ (on p. 236) as simple as possible; for example, we could take K=0, N=0, M+L=0, initially, ‘The actual method adopted is more direct than the amplification of the earlier double process. Moreover, it has the added advantage that it can be extended, without further change, to the case of any number of dependent variables. The three subsidiary equations reducible to two, with modified variables. 165. Next, we modify the form of the three subsidiary characteristic equations, of which the full initial expressions are Ey=Guk + Gan+ GuS+ In€ + Gan + Gob a oe — Gut gun + galt gul’ + gun! +9sb")=0, Ey= Gat +9u0 + Gah + dak! + Gurl + Gut! a re jp Gut +907 + Gab + Iuk’ + 9u1! + Gub!)=0, Ee= gub + 9un+ Jub + Jak’ + gan + gab" a eG = Gab + 907+ 966+ uk’ +900! + gb!) =0. In the first place, these three equations are equivalent to only two inde- pendent equations (§ 161) in virtue of the relation a het Ey +nte=0, which now will be verified. Partial differentiations of the identity satisfied by g give relations ag fe edn t+ YG + Ago a Fyn dat dat eto, # = gat gor Jur ‘When the first of these identities is differentiated with regard to t, we have %Goot Yea AGort Gert Y2Gos + 229s Hagar t YaSeat 229+ HG + ges + Gus and therefore eee n ts © (Jou — Gu’) + Ys (Goa Gee) + #,(Gou~ Jo!) =O.240 MODIFICATION OF THE cu. v Similarly from the other two identities, we have © (Gu—Gu') + Hh (Gea— Jas) +21 (Yuu — Gu’) = 0, (Gor Go’) + Ys (Gu Go") + #5 (Gu—Ga') = 0. In virtue of these three relations, the coefficients of & of », and of f, in 2,Ey-+ yEy+ Ee vanish separately. Again, we have obtained the relations gat gut 29s @iG:a + iG t 219s Agut Ygut 29s In virtue of these, the coefficients of £”, of 9", and of £”, in a Be+y,By+4,Be vanish separately. Next, differentiating with regard to t the first of the three relations just quoted, we have Gul + Gia + 29s + 29u+ YG + Hs and therefore Bigu + Ys (Gis + Gos — Gu) +1 (Fu! + Jou Fa) =I: (Ga Gu) + #1 (Goo — Ja) ~ Lan — YaGhs— 209s 9 (Gea Gu) + 2s (Geo — Ga) — %a(yPP + 29,2, B+ 2°Q) ty(tipP + O%4R) +2 (HyR +2%Q) because of one of the fundamental characteristic equations as they are stated in §160. Similarly, from the second and the third of the foregoing relations, we have (gus + 9 —Ju) + YrGu' + (Yu' + In ~ Ju) =9, ® (Gx! + Ja—Jo) + Yi Gu! +Ja—Iu) + 29u' = 0. In virtue of these three relations, the coefficients of £’, of 7’, and of £, in 2, Hy-+ ysB, + %Be vanish separately. Consequently, we have a Bigt ly + Bg=0. 166. We therefore need retain only two (out of the three) subsidiary characteristic equations. We select the second and the third, and proceed to change their expressions, using the variables ¥ and Z (already introduced) instead of £, 7, & with the definitions Y=an-yb, Z=a,f-a6, Yi aanl — 9 + an — yk Ba ah— a8 +at— ab)" so that166] SUBSIDIARY EQUATIONS 241 where it will be noted that, if u=£ v=, w= then V=Y, W=Z, the quantities that occur in the modified expression for @ +97 in § 162, Substituting the expressions given for the various coefficients gim in @, we have Guak! + gan’ + gub' =— (ay. P + 0,2,R) E+ 0,Pr! +03RE =a ([P(an! —n€')+ RS —a€')) =a {PY'+ RZ’ P(an—yb)- R(@,6-48)), and Gok + 9un + Gah—— (HE + By + Fo)—§ (Pay + Reye,+ ayy K +a,2,L) tn( Paya + 02K) + (Raye, + 01); therefore GE + 907+ G0b+ Gul! + I97 +906 =-(HE+ By + Fo) +2,(PY' + RZ'+ KY +12). Again, Gak’ + gan’ + gut! =— (HE + Br + FC) —& (Pye, + Raya +y,K +2,2,M) to (Paty + 02K) +o (Reve + eM) =— (HE + Ba’ + Fe) + (Pay tn K) eal — 8") + (Ray tM) (a aE), and Guk + Gan + Gub= —(H'E + By + O'S) ~E(Pay. + Rant ny kK +0, K + 2,0,0+22,M +2,y,D+ 2,28) 40 (Paps 20.0,K + 22D)+ (Rast +02 +2,04M +08) (H'E+B'n +0) + (Pa, + ©) (00 ~ yo) + (Raat 2M) (08-26) 4K + aD) (9-8) + (@L +28) (mb-28); therefore Gab + Inn + Iub+ Ink + Jen! + Jub! = {HE + Bo + 06) 40,(PY' + R2'+ KY +12) +0,(KY' +M2Z'4+ DY +82), Substituting these values in the subsidiary equation #,~0, and removing a non-vanishing* factor a, we have By= 4 (PY'4 RY +KY + L2)—(KY'+ M2'+ DY +82) ~0. * As z varies uniformly through the range and ¢ increases throughoat the range, 2, can only. vanish for isolated places at the utmost.242, ‘TWO SUBSIDIARY EQUATIONS [cu v Proceeding in the same way with the subsidiary equation Z,=0 and again removing the non-vanishing factor «,, we have By= 4 (RY +.QU' + MY + N2)—(LY'+ NZ’ + SY+EZ)=0. We therefore take Ey=0, Ez,=0, being two independent equations, as the modified form of the subsidiary equations. Manifestly linear and of the second order, they can be regarded as determining Y and Z as functions of ¢; because the seven quantities K, L, M,N, D, 8, B, do not involve &, 7, §, that is, do not involve ¥ and Z. One selected integral of the two subsidiary equations. 167. One important property, in the form of an integral of these two simultaneous equations, will prove useful in the further reduction of ont to its final form. We write 20) =PY"+ 2RY'Z’+ QZ" +2KYY' + 2LZ¥'+2MYZ' +2NZZ’ + DY*+28YZ+ EZ, where it will be noticed that 2(] is the same function of Y and Z as of is of Vand W; and then the two subsidiary equations H,=0 and Hz=0 can be written in the form 20_ 4 (00).9 20_4 a SP aler)=% 92a (az) As each of these two equations is linear and of the second order, their complete primitive giving the expression of Y and Z—to its expression we shall return later (§ 170)—is compounded of linear combinations of four linearly inde- pendent sets of integrals Y, and Z,, Y, and Z,, Y, and Z, Y, and Z,. The property in question is that there exist two sets of integrals Y, and 2, Y, and Z, (where Y, and 4, are either one of the other three sets, or are linear com- binations of them in the form YrealitaNitaln Za akst alt ole the coofficients ¢ being constants) such that, if, and (), denote the values of C) when Y, and Z,, and Y, and Z,, are substituted for ¥ and Z, the relation en ao, vay? +n 0. Or a Nay t Baz, is satisfied ; and the choice is not unique. ‘The quantities (, and (], are homogeneous and quadratic in their respective arguments, and they have the same coefficients ; hence the relation 20h ta167) ONE SELECTED INTEGRAL 248 holds, the two sides being equal to the same expression lineo-linear in each set of arguments. But @ Os ax Gz)» o_O az) a0. ay, = Ye)" a _@ OY oY, dé a Gy : and so the relation can be written a (Mayet a) Hence where & isa constant. As ¥, and Z,, Y, and Z,,are individual and not general integrals of their equations, & is a specific (and not an arbitrary) constant. If k is zero, for a primarily selected pair of sets Y, and Z,, Y,and Z,, the property is established for that pai. If kis not zero, take another pair of sets ¥, and Z,, ¥, and Z,, where Y, and Z, are not a linear combination of ¥, and Z, Yz and Z,, Then, in the same way, we find oD: Noy; where J is a specific constant. If 1 is zero, the property is established for the pair of sets Y, and Z, ¥, and Z. If Lis not zero, take a set of integrals Y,=1¥,—kYs, 2Z,=12,—k2,, so that Y, and Z, are not merely Y, and Z,, nor merely ¥, and Z,, nor merely Y,and Z,. Then 2D 82, 12h 20) 20 lor? kay? ia * Oey owing to the linearity of the first derivatives of O:, Oh, a in their argu- ments, Then we have Oe ol a Nay, + 2ag7~ Pray; + Zag; establishing the property for the pair of sets Y, and Z,, Y; and Z,. Manifestly the property is not unique; because we can make Y, and Z, take the place of ¥, and Z as a set of integrals, and we still have the sets Y, and Z,, Y, and Z,, to associate with Y, and Z,, with the new Y, and Z, and with one another. For our immediate purpose, it is sufficient to possess the relation for two distinct sets of integrals of the equations. +5444 FINAL NORMAL FORM OF [on v Final norma form of second variation. 168. We now can proceed to the final form of the quantity e+. In connection with any set Y and Z of integrals of the subsidiary equations, we introduce quantities @, 8, y, 8 which satisfy the two relations Y'+a¥+yZ=0, Z’+BY+8Z=0, these relations, as they stand, being insufficient to define the four quantities. ‘Then PY’ + RZ’ + KY+ LZ =(K - Pa-R8)¥ +(L —Py— Rd) Z, KY’ + MZ’ + DY + SZ =(D - Ka—M8) ¥+(S — Ky— M2) RY’ + Q2' +MY + NZ=(M—Ra—Q8) ¥+(W — Ry—Q8) Z LY’ +NZ’4SY +82 =(S -La-Nf) ¥+(E—-Ly-W8)Z; and the subsidiary characteristic equations Ey =0 and E,=0 are satisfied if K=Pa+ kB M= Ra+Q8 Py + BB N= Ry+Q Ka+ MB = Pat +2Ra8 + Qs S=Ky+ MB = [a+ NB = Pay+ R(ad-+ By) +Q88 B= Ly + Nb = Py + 2Ryb +. QE where it will be noted that the apparently different values of S, viz. Ky + M3 and La+ NB, agree. Thus far, the relations hold for any set Y and Z of integrals of the sub- sidiary equations. But a single set is not sufficient to define the quantities 2, 8,7 8. Accordingly, we take two independent sets Y, and %, Y,and Zs, and associate both sets with these four quantities, the association not affecting the values of K, L, M, W, D, 8, B. Then we have Yi + a¥ito% Yi +a¥,+9%,=0f" Z/ +BY, +82,=0| } 24 AY, 487,00)" é 7 =ViVi "20-2 ~ Y.ZI—¥, ¥,4,-¥.4," where ¥,Z— ¥,4Z, is not zero, because the two sets of integrals ¥, and Z,, Y,and Z,, are independent, in the sense that Y, and Z, are not merely the same constant multiples of ¥, and Z.168] SECOND VARIATION 245 The original characteristic equations are known to be satisfied in con- nection with the quantities K, Z, M, N, without this assignment of values. The subsidiary equations are satisfied in connection with the additional quantities D, 8, #, as well as K, L, M, N, without this assignment of values. All that has been done is to assign values to K, L, M, N, D, 8, B, associated with two independent sets of integrals Y, and Z,, Y, and Z,. Now we have seen that any two independent sets of integrals ¥, and %, Y, and Z,, are such that where & is a specific constant, and that an appropriate choice of two sets ean be made so as to secure that the specific constant is zero. Now re +0 (rR + 2.574) = ¥\(PYy+RZy + KY, +L) + Z,(RY; + Qh + MY + NZ) — ¥, (PY) + BZ! + KY, + LZ,) -Z( RY + QUi+ MY, + NZ) Y,4,— ¥,4)(L— Py — R8)—(¥,2,— ¥,Z,)(M —Ra— QB). Thus the values assigned to the various quantities, in particular to Z and Mf, require a zero value for the specific constant value of the left-hand side; and consequently the two sets of integrals Y, and Z,, Y, and Z,, of the subsidiary characteristic equations must be chosen (as they always can be chosen) so that 20, 20, ao, YS 4 2m vo 4 oh With these values now made definite for K, L, M, N, D, 8, B, we have 04 are PV4 2RV'W'+QW* +2KVV' + 2LWV'42MVW" +2 WW! + D4 2900 + EW = PV +aV ty WY + Q(W'+ BV 480) +2R(V' +aV+yW)(W'+ BV +8W) = amayep lh) Vs VW i+Qlw, Vv, Wye MH Ky A 4, % Yi Ya, & Bi, Va, By ls v, wey. wh wy Mh, 4M, Bj Yi, Ye ANA, B, A! which (with the foregoing limitation on the selection of the sets Y, and Z,, ar Y, and Z,) is the final reduced form for @-+ >.246 PRIMITIVE OF THE [ca v ‘The expression for 7 is T= Aut + Bo + Ou’ + 2F ow + 2Gwu + 2Huv (aut + gut? + gat + (Ju + Ju) Vw + (Gu + Gu) WH + (Jat Jun) WO} = PV+RW RV+QW footer |e Ys Weve yg ewonm Vs WY YW, hy & a, Ny wa} | Yi, Yn & | Zo, Yn Ai But e dt=[T}, taken at the limits, The limits have been settled by the conditions deduced from the first variation of the original integral J, so that u, ¥, w (and therefore V and W) vanish at each limit; and thus oF at=0, between the limits, ‘Thus the second variation of the integral has become equal to je fod Primitive of the subsidiary equations. 169. This reduction of the second variation of J manifestly depends upon a knowledge of the quantities Y and Z; and the significance of the normal form depends upon properties of two independent sets of integrals. ‘The dis- cussion thus requires a knowledge of the primitive of the subsidiary character- istie equations. Now these subsidiary equations arose from the original characteristic equations by the consideration of varied values ©-+ «8, y-+en, 2+K£ of @ y, 2, these values still satisfying the characteristic equations; and the values of £m £ were derived (§ 161) from those of «, y, z in the primitive of the characteristic equations. By means of these quantities, two other quantities ¥ and Z were construeted, such that Y= ain hE = 8, (1) + a2, (f) + 048, (1) +48, ni Laat 2£=49,()+a% +a ()+a9 0)" where a, ds, ds, @ are independent arbitrary constants. Also, Y and Z satisfy the linear equations ao D) =o, ao 4 (0) =0, ay Of ~ dt\aZ" the primitive of which is constituted by a set of values Y and Z, involving four arbitrary independent constants linearly. Clearly 8,(¢) and 9, (f), where r=1, 2,3, 4in turn, are a set of integrals; if, then, these four sets are linearly independent of one another, the foregoing expressions for Y and Z constitute the primitive of the subsidiary equations, We proceed to establish this linear independence.170) SUESIDIARY EQUATIONS 247 110. The primitive of the characteristic equations is Bm HCE A, i, Cay 05, 4) srateenmes moh FEY (E+G a, Oy, ay, 4) where the essential constants are a1, a, 4,4. The constants are determinable, by the assignment of initial values to ¢, y, 2 and of initial values to 2’, y, 2’; but as the variable ¢ is not specific and any function of ¢ may be taken in its place, the initial conditions become precise by the assignment of an initial value ¢ to 42(=2), and an initial value m to =( band c toy and 2, when «=a, and t=%, In this form, the primitive of the characteristic equations becomes y=y(z,, 1, 6, m) 2=2(e,b,|,0,m) ‘The set of essential arbitrary independent constants a, a, as, a is expressible in terms of the arbitrary independent constants 6, l, cm, by means of four independent equations 2). as well as of values a, = @, (6, 1, 0, m) a, = 02(6, J, ¢, m) = w4(b, 1, 6, m){” = 04(b, f, 6, m), As the four constants in each set are independent, these functional forms are such that neither of the Jacobians (tae a bbom vanishes. Now the quantities % and &, derived from the second form of the characteristic integral, must have the same values as those derived from the first form; hence dy 8) de vw ae XO & xO Again, from the equivalence of the two forms, we have By _ dy Be , dy ob , dy al, dy Be , dy Om a, - dada," 6 da, ” db da, dc da, Om da,” for r=1, 2, 3, 4; and therefore =vo%-¢o% (=x Os - #0 5 aye 24 2 4 2% Oy 20, By am) aX BE Bm 4 BE om a‘248 PRIMITIVE OF THE SUBSIDIARY EQUATIONS fon. v Similarly, proceeding from the equation in z, we have tin yan OF Sax’ in O75. (az ab dz al az de xO Ge Bl ba, * de Ba, * Gm ‘Now if there were a linear relation among the four quantities 8,(¢) with constant coefficients, and a linear relation among the four quantities 9, (t) with the same constant coefficients, of a type 3 8-0, 3 w9)=0, such relations would have the forms ay ay +,,3 Deedee lse ss te ma, a 25, 2, Smee Hp 4 oy, — But these have the form of linear ee * constant coefficients, between the set ay by dy ay a* a" a’ am together, and the set ae de de de a’ a0" am together ; and it has been proved (§ 140) that no such linear relations exist. ‘The foregoing apparent relations must be evanescent, so that the constant coeflicients must vanish ; hence ee) Pigg, * Hig, * a * Fe, al aa at a tm a #4, 2 bn tba, taaq, * 3a, am, mam am Mg, tM Gg, tM ag, tM Sa, ‘Phe determinant of the coefficients of the constants 1s, py Hay Ha is Blom oratere): snd s0 these equations can only be satisfied if which does not vanish ; 1=0, = 0, w= 0, y= 0.70) DEVIATION AND TANGENTIAL SLIP 249 Consequently, there is no linear relation between the integrals 6,(t), 0(t), 4, (t), 8,(t) with constant coefficients, and there is no linear relation between. the integrals 9, (t), 94(d), 94(¢), 94(t) with constant coefficients. Hence the primitive* of the subsidiary characteristic equations is Y=a,0, (t) + 0,0, (t) +448, (t) +. 4,8, a Z =a (6) + Q(t) +.4,% (t) + a,9,(0f" where a, ds, as, a, are independent arbitrary constants. Clearly 4, (t) and 3, (t); ,(t) and 9,(t); 0,(t) and 95(0; A(O and 3,(¢); are four sets of integrals, each linearly independent of the others. Note. In connection with the deviation from the characteristic curve for integrals involving only derivatives of the first order, it was pointed out (Wote, § 52) that an apparently arbitrary representation of the variation was taken by including only a quantity av yn (which measures the normal deviation), and ignoring or omitting a quantity mut yD (which, measuring the tangential slip, does not alter the shape of the curve). And it was seen that, in the reduction of the second variation, this omitted quantity cannot enter when w is retained—that, in effect, w gives the whole of the small variation which need be retained. The same considerations occur in the case of skew curves. There is an apparently arbitrary representation of the full expression of the variation, by retaining only the quantities Yean-né Z=a,f-26, and by ignoring any tangential slip along the skew curve. The justification of the result emerges in the same way as the justification for the simpler instance. Solely for the passing purpose, we take the are of the characteristic skew curve for the independent variable; and we resolve an arbitrary small variation «£, xn, xf into three reciprocally perpendicular components, giving a displacement «r along the tangent, a displacement « i » along the principal normal, and a displacement ae along the binormal. Then we have (with the customary notation for the direction-cosines of the principal lines of the curve at the point) rb = der + pa’ prt ee —#y) 628, * "The primitive of the subsidiary charactoristio equations has been dedueed from the primi- tive (supposed known) of the original characteristic equations. If, however, only a special set of integrals of the latter were known, the subsidiary equations would have to be integrated, leading to the same primitive.250 PROPERTY OF (cH. v that is, ae sa’ /e" — sy") B; eee Eaarta'vt ye" -2y)B aaytty’y + ea" — a's) 8, badrte’y +(e'y' -y¥e)8. Hence Yown-y=@y’—ya")\v— Be", Bndg—Z6 a (a2 — 20") + By’. The second variation has been proved to be expressible in terms of Y and Z alone—that is, of » and f alone (were these selected, to the ignoration of 7)— by means of expressions, into which 7 could not then enter. ‘The apparent ignoration of the displacement, due to a tangential slip, is justified alike in analysis and significance, Property of fundamental sets of integrals of the subsidiary characteristic equations, 171. The subsidiary characteristic equations are PY" + RZ’ -P'Y'+(M-L-R)2Z'+(D-K)Y+(S-L)Z, RY" +Q2" =(L—M-R)Y'— Q2' +(S—M)¥+(E- NZ; and they are satisfied by the four linearly independent sets of integrals 6,(t) and &, (b, for r=1, 2, 3,4. We have (PQ— Ry ¥" =~ (QP + R(L- MR); ¥'+(Q(M-L- R)+RQZ +terms in ¥ and Z, (PQ-R) 2" = (P(L-M-R’)+ RP} Y'—{R(M-L-R)+ PQ)’ + terms in Y and Z. A), Olt), A(t), 40) | HO, 2), MH, RO | | OO), OF), 8'(), 8 (t) WO, WO, WO, WO Denoting by J, we have Ta) AO, OW, WO, A) |+/ 40, BH, BO, BM) |. 30, HO, BO, YO | [RG BO, WO, HO O,'(6), Or" (t), y(t), A." (t) Od), At), A(t), O'() YO WOH XO XO! [WO WO 8O, YO When we substitute for 0,” (t) (that is, for Y”) its above value in the first determinant on the right-hand side, and for 9,” (t) (that is, for 2”) its above value in the second determinant there, we have (PQ Ry) Y= QP + RUL-M- RY} I~ [RM LR) + POT =—(QP' + PQ -2RR) J.172) FUNDAMENTAL SETS OF INTEGRALS 251 Hence (PQ-R) IT where A’ is a constant. The sets of integrals, 0, (t) and 9,(f), for r =1, 2, 8, 4, are linearly inde- pendent, so that no relations of the form X10 (Q=0, EaeS-(t)=0, where jay Hs) pay Ha are constants, can exist; hence J is not zero, Next, it will appear that, if a maximum or a minimum is to exist for our integral, the quantity PQ— R* must always be positive and may never vanish within the range. Hence 4’, the product of two non-vanishing factors, cannot be zero. Thus (PQ- BR) J=4, where A’ is a non-vanishing constant. Clearly A’ will be a specific constant, depending upon the specific selection of the four linearly independent sets of integrals. Any four linearly independent sets of integrals are linearly expressible in terms of any other four linearly independent sets; thus Yu= 3 cur Oil, Zam ¥ One (0, = 7 for m= 1, 2, 3, 4, where the determinant V of the coefficients cmr is not zero. ‘The determinant J of the quantities Y, and Z, is equal to the foregoing determinant J multiplied by the non-vanishing constant factor V: that is, J=JV. ‘Thus the determinant J is covariantive among the four linearly independent sets of integrals. Corollary. Moreover, as the quantity J is not zero, the following inferences hold: ‘The four quantities 8, (t), 6.(t), 0,(0), 61(2) do not have a common root in the range of the variable ¢, nor do their four derivatives; and, similarly, the four quantities 9, (¢), 92(¢), 95(¢),9.(¢) do not have a common root in that range, nor do their four derivatives, Discussion of the second variation : the Legendre test. 172 Tho second variation of the integral J has been expressed in the form 4x? [Udt, taken between fixed limits; and, here, U=P(V'+aV+ BWP + Q(W' +4V 430 +2R(V' +aV+BW)(W' +9 +307), with the values of a, 8, 7% 8 which have been determined. As the limits are fixed, 1, v, w (and therefore V and W) vanish at each limit: subject. to this property, the quantities V and W are arbitrary regular funetions of f.252 LEGENDRE TEST [ov If the original integral 7 is to provide a maximum, then in general we must have fo dt negative, for all variations; and, owing to the arbitrary character of V and W, this requirement can be met only if U is negative for all varia- tions, If that integral is to provide a minimum, a similar argument leads to the conclusion that U must be positive for all variations. If it could happen that Joa is zero, without V and W being everywhere zero, the second variation of I would vanish for such variations; and then the change of I would be governed by the third variation, ‘We shall set aside this possibility of a conditionally vanishing second variation, as being more restricted than is desirable in the quite general discussion; if and when instances occur, they can be discussed specially, the course of discussion having been illustrated in an earlier example (p. 33). ‘Accordingly, for our purpose, we now require that U shall be steadily positive (for a minimum) or steadily negative (for a maximum), so that it shall never vanish except for zero variations u, », w, and therefore for zero variations Vand W. Now so long as V’-+aV'+W and W’+yV+8W do not vanish simul- taneously, the conditions, which are necessary and sufficient to secure that U shall always have one sign, are: () the quantities P and Q must each have one, and only one, sign throughout the range; (ii) the persistent sign of P and the persistent sign of Q must be the same; (iii) the quantity PQ— Rt must be positive throughout the range, ‘The third condition implies the second, but the second condition does not cary the third. If PQ— R* could be negative, or P alone could change sign, or Qalone could change sign, U would not have always the same sign for all variations. If PQ—R* could be zero, while the first two conditions were satisfied, variations represented by P(V'+aV+BW)+Q(W'+yV4+8W)=0 would make U zero, a possibility to be excluded. Manifestly, if the persistent sign of P and of Q be positive, a minimum is admissible; if it be negative, a maximum is admissible. Other conditions may arise from other causes: these conditions must be maintained, as neces- sary conditions. ‘These conditions are called the Legendre test, in extension of the fact that the corresponding condition for the simplest problem was first stated by Legendre.174) JACOBI TEST: CONJUGATES 253 Conjugates on a characteristic curve: the Jacobi test. 173. Next, wemust consider the possibility that the quantities V’-+aV+8W and W’+V+8W, viz. ee le we, Wh &| |e. hy & Yi, Yu & lay & may vanish simultaneously for a pair of sets of integrals of the subsidiary characteristic, such that Y,Z,—Y,Z, is uot zero, and such that ach eee ect ond eee The only way in which this cam happen is that, for any pair of sots of integrals Y, and Z,, ¥, and Z,, which satisfy the conditions as stated, it may be possible to have variations V =aY +4F,) Wart,+ x2) where A and yw are non-zero constants, All variations V and W must vanish at each limit. All the requisite possibilities, that make failure admissible, will be attained, if we can have sets of integrals XY, + uY,, AZ, + uZ,, zero together at the lower limit and zero together at the upper limit, while AY, +uYy and AZ, + 2, are not otherwise zero together (a fortiors, if they fare zero at only isolated places) in the range of integration. In order to secure that U shall not become zero, through non-zero variations of such a type, we must secure the exclusion of the type. It therefore must be im- possible to have @ non-zero combination of any two conditioned pairs of sets of integrals, XY, +, and XZ, + 4Z, being the combination, which vanishes at the lower limit, and also at the upper limit, of the range of integration, ‘We may state the condition in a different form, by requiring that. the range, beginning at a place where any combination AY, + mY, and AZ, +2, is made to vanish, shall not extend so far as the nearest place where some such com- bination again vanishes, Such nearest place will, as before, be called the conjugate of the initial place. The required condition is called the Jacobi test, in extension of the fact that the corresponding condition for the simplest problem was first stated by Jacobi. Gritical equation determining conjugates. 174. The analytical expression of this test requires the fuller consideration of the characteristic curve. As the argument follows almost exactly the earlier argument in the case of the special variation (§ 152), it ean be ex- plained rather briefly.254 CRITICAL FUNCTION FOR [cn. v We take, adjacent to the characteristic curve, another consecutive curve represented by a set of integrals Y, and Z, of the subsidiary equations, the deviation being represented by its projections «Y, and eZ, upon the axes of y and of ¢ respectively. The conditions, assigned for the unique determination of this consecutive curve, are that, at Y,=0, Z=0, Vy pT (t), Zi where p is a non-zero constant, Then constants bj, b,, bs, b,, such that 0,0, (t)=0, Eb,S-(q)=0, Eb6,'(t)=pT(h), b-9-' (t) =0, give Y,=B,0,(t)+0,0,(1)+0,0,()+6.0.()=p| (t), Olt), 8), Alt) |, (ts), Bolte)» s(t) » Balto) BiH), ME), B(6), Sal) BG), BH), 4 (H), (4) Za=bAO+OIO+EAWO +I =P | HO y HO, WO, KO - to), a(t)» Pate), Bete) | Sil), Beh), MG), SaCb) | Sh’ (bo), Seb), Bui), SW) We take a second consecutive curve represented by a set of integrals Y, and 2, 0f the subsidiary equations, with «Y, and «Z, a8 the projections, upon the axes of y and of 2, of the corresponding deviation. ‘The conditions, assigned for the unique determination of this curve, are that, at t= t, ¥,=0, %=0, ¥i=0, Z/=-oJ(t), where o is a non-zero constant. Then constants ¢,, ¢, ¢, ¢, such that J aete=0, TeBe(t)=0, Ee,8/'(4)=0, Ley (te) =— (te), ive Tre eBiradOeabl+eed=e | 1D AC), BO), Bt) |, | OCt), Ct) Balto), Balt) | Silt), I(t), Bolt), Bel) | 85 (t), Oto), 05 (bs), 8: Ct) | =o (t) +aM(H+a%wO+o% =e] W(), M), WO, XO |. (0), all), Oat)» O4(4) SG), Sab), Seb), BeC6) | (te), Be (te) 8, (ta), 86 Ct) \ In order that these two sets of integrals Y, and Z,, Y, and Z,, may be suited for our purpose, they must satisfy two requirements.174) JACOBI TEST 255 In the first place, they must make the quantity 20, 30, ao, OD: Yay Baz - (ay + 2 az)) zero. As Y, and Z,, Y, and Z, are two sets of integrals, this quantity is bound to be constant; and its constant value can be taken as its value anywhere. When t=, we have ¥,, Z, Ys, Z each equal to zero: so the constant is zero. The first requirement is satisfied. In the second place, the two sets of integrals are to be independent, in the sense that the magnitude ¥,2,-¥.2, must not be zero in general. It vanishes when t=; but Vand W vanish then also, so that (as in § 64) no infinity occurs in the expression for U at t=t. But after the initial value of ¢, the quantities V and W must not vanish simultaneously until the conjugate of t, in the range is attained. Moreover, the quantity (¥,Z— Y_4)-* ocenrs in the expression for U, so that ¥,2,~ ¥,Z, ought not to vanish in the range after the initial point. Now V2, — Yad, = [2b,4,(0)]} (ZeeBr O} — {Zo,6-(8)} {Eb-9-(O)} rea bats) {A.M ( (D)- Writing | A(t), Be), (D, Blt) | HO, 2O, HO, UO | Y(t), BL (t), BS (), BE (8) |) = ats as (ts Lat), belt), laa ts bat), NO, WO WO, WO | AC), A(t), BCE), At) | = AG, 1), 2O,%O, WO, WO (ts), Balto), Os(te), Oa(ts) 3G), Sate), Bs (to), Se (te) we have a(t), Fist), kus), hes (t), ka (Os ba (t), by = p {les te) Se’ (ta) + Kea (to) is (te) + His (to) 9 (ts — Ba= p hau lo) Sx’ (to) + hea (be) Ss (be) + kas (bo) Set)» 61 =o {hag (to) By (ta) + (to) Oy’ (t) + Kes (te) 8. (4) — C= a {lis (te) By (te) + Hear (to) 8, (la) + Bis (to) 8, (t0)} 5 and similarly for the other constants 6 and o. Then, as before (§ 152), i a (BxC— bac) = {Uh (t) ea (ts) — Kea (to) has (t)} baa (to) Hest) (a(t) Lan (t) + Hen (te) ba ba) + Ha to) b(t) + a(t) Ua te) + Ha () ae ()} = halt) I (bas256 CRITICAL FUNCTION FOR JACOBI TEST (on. v and similarly for the other combinations of the coefficients bem — Paci. Thus Via Yoh pod) Fas to) Bn (+ an) Balt) + Fat) Rat) + Bat) t) + Bs (bs) Bea) +B) at) = Alte t)5 and therefore ¥,2,— VaZ,=— pol (be) M(t, t). Here p and ¢ aro non-zero constants; J (f,) does not vanish ; and A (t,t) does not vanish identically, Hence ¥,Z,— Y,%, though it vanishes when t=, does not vanish identically along the range; and thus the second requirement, for the two sets of integrals is satisfied. Now consider a variation V and W, given by Vaar+uY, WarZ+nly, where X and 4 are non-zero constants, At t=%, we have V=0 and W=0, so that the variation vanishes (as it is bound to vanish) at the lower limit, ‘The conjugate of f is the first place after the lower limit at which two integrals XY; + w¥, and XZ, + 4Z,, vanishing at the lower limit, again vanish, Denoting this conjugate by ¢,, we have AN (h) + Vat) =0, Z(G) + w(t) =0; as X and p are not zero, we must have Ya(t) Z2(t) — Z,(4) ¥o(t) = 0, A(t 4)=0. Thus the conjugate of any point t, on the characteristic curve is given by the first root t,, greater than ty, of the equation A(t, 4) = 05 and the range of the integral, in order to satisfy the Jacobi test, must not extend as far as the conjugate of the lower limit. that is, 175. We take, in illustration, the question of a particular curve in space, viz, Lat it be required to find a massimym or minimum for th integral Va fade, where x is any continuous function in space, and the integral is taken from one surface—that ia, from any unspecified point on that surface—to another non-intersecting surface—that is, to any unspecified point on that other surface, (The problem relates to the Principle of the minimum ‘reduced path? of a ray in a heterogeneous medium, where 4 is the index of refraction at any point in the medium according to this Principle, the actual path of a ray between any two points of its course provides the minimum length of reduoed path between them, Newton enunciated thoorem, according to which tho path of the ray ie thatof a particle moving with a velocity115] MINIMUM REDUCED PATH OF A RAY 257 # under a system of forces with a potential ; the problem associates this theorea, s0 interpreted, with the Principle of least action, Further, for such a system of rays, the quantity 7, when regarded as a function of position in space represented by the upper limit of the integral, is Hamilton's Characteristic Funetion®,) (A) The integral is Jutsrsyeecnh at so thatthe function g in yw (o?-+yi?-+a8, while is any function of , ys When subati tution in made in the characteristic equations, they become am a2 {on Betrntrayh-Z }-2 ast . ie Sareea with two others. When we take the langth of are as the independent variable in order to simplify the equations already formed, the characteristic equations become Mm d( de) 9 me d/o ‘=o, (ZB) -0, (4 ae aa" aa) = ya ade a ae » in fall, de ds de * at - ays, Gi +@y+ for, on multiplying them by 2, 4, % respectively, and addi =o, de dy de fu der, iu dy . Be da dy ds Neat, eliminating “# and y determinantally, we have & in me de ee a de’ dt bac ay de de a ae 7) ‘that is, the three lines whose direction-cosines are proportional to Mm mw ds dy di dr Py ds ae? Oy? We da? ds da dal? at lie in one plane, Hence the osculating plane of the curve at any point contains the normal to the level surface wm constant at that point: or the direction of the path just outside that surface and its direction just inside are in one plane with the normal to the surface, * For the discussion of these matters, reference may be made to R. A. Herman's Treatise om Geometrical Optics, eb. x, eh. xi.258 MINIMUM REDUCED PATH OF [on. v wna) +) +@Y- G9 where p in he aio of cua carat of tho curve the poi I be the an Soiree eon of ect ad that ote sutvant-daws normal to aos, 1 ay de, 1 Qe dy, 1 ap de covey teas t Way det Wo de Again, wo have or, taking socount of the value of 17, Boos pa 4 sin y. we HH Wo have 3, iis ie direction, that is, of incroasing =, y, 2; if, then, RPN=y and fas the direction-cosines of the outward-drawn normal—in the P'PN'=y-+dy), the angle of contingence de at P is —dy. But pt ds Gi i ; el in the figure p is positive, and dis postive: thus } om y= din 7 | that is fon woos y. d= —dy sin y, and therefore sin yaconetant, a first integral of the characteristic equations. ‘These two results are the expression of the customary laws of refraction, originally due to Snell. (B) Specific knowledge of the form of p, expressing ite value in terms of its position in space, is required before a second integral of the characteristic equations can be obtained. Moreover, in the absence of this specific knowledge, the construction of the tubsidiary characteristic equations remains merely formal analysis; and the detailed form of Jacobi’s rango-test cannot be obtained precisely. ‘As regards the Legendre test, we have ‘ta! dP, (etayteant % ya! - tented Hence P is always positive, @ is always positive, and PQ~ R* is always positive. Con- sequently, the obtained path provides a minimum value for the integral V, within a range that must satiafy the Jacobi test,175) A RAY IN A MEDIUM 259 (C) To take a specific example, suppose that and (yee where yp is a constant. (It is easy to see that no generality is provided by taking wi 4¥ (24+ 92-429), where A is a constant, for the value of y2) The characteristic equations de, dr nat Gre (on multiplying throughout by j, and writing +? for 2%+y*+2%), with two others. Multi plying this equation by 2S and integrating, we have (nn (By aa (ud -1) S50, and, simitary, : cnn (Pao, cen (BYaaoa from the other two equations, with the relation BHD mt ‘among the constants, when account is taken of the pe de\t | (ayy? (ds? Gi) + @)+@)— ‘These three equations manifestly are satisfied by smasinim x(), ymbsing+A)=9(0, smesin (Cy) =¥(O: no advantage arises from taking #=a sin (t+), because the explicit absence of ¢ from the ‘equations shews than an unessential constant (euch as y) can occur as additive tof. ‘The quantities arising as integrals of the subsidiary equations are 4(O=x’ © $1(0-# (© xs (= — bain 008 (t+) 4)" x ObO-8 Ox O= aeostsin(¢+8) XO b30- $0) x= aboos teos(e+8)[” (bu 4'0) xe =0 were x1 (0 2 (€s x8 Oh x4 @) denote 2X, 3X, x, 2 and likewise for @ and y: and HO=X OWO-W Ox) & cos tsin 47) ~e sin con (C4) a ~ 2 co sin (4-7) ein £008 (6+) K=XOWO-¥ Ox na Secos tsin(t+y) = thesia IsD=y’ OVs)- HO xx(=0 I O=xO Val) — HO) x4 (0) ac 008 £008 (t+)260 HAMILTON'S CHARACTERISTIC FUNCTION AND [ox. v It is clear that there are no simultaneous relations Ay6s (0) + 4260(0)+ As8 (t)+ Aee()=0) Ard (O4AsS2 (0+ 459s OF Aad ( | with constant coefficients 41, 43, 45, 44; and that therefore the combinations 6, and 3, 6, and 9, 63 and 93, 6, and %, constitute four linearly independent integral-sets of the subsidiary characteristic equations. ‘The clement of are is given by (ti) =steotestteot coer ceot 0 that «is expressible by an elliptic intogral; the explicit value of # is imelevant to the solution. ‘The actual path of the ray, as given by the equations eaasing y=bsin(¢+8), emcsin(t+y), is algo given by the equations 2 #2 cops faants Be Boon o#. E22 omy + Senin y, Zain o—7) +¥}siny—$ sin deo; and therefore it is an ellipae, the section of either of the elliptic cylinders by the plane, The function, which determines f, (the conjugate of the initial place t), is Alloy H)=| (Gs (Gs Os), 0 Allo), Aalto» Astley 0 S(t), Hla 9» HCG) MXto Malte 5 (60) | = (x(t) 8 (to) — 82 (10) As (A) £91 (4) 94 (to)— 9s (a) HCH} ~ {8s (4) Os (ts) ~ 64 C0) 80 (4)} £92 (6) 9a (te) — 92 (60) (ED. Now 8, (61) 85(ta)— 82 (4) 85(t)= a*Be0s 4, co8 4 sin (4 — 4), 4 (44) 8560) ~ 8 (te) 8 (4) = — a 008 (4, +8) C08 (to+8) sin (ty — he) 94) 9 (lo) Bu (Co) 94 (4) = ~ a fa 008 £ 005 fy +62 605 (4, +7) 008 (fy 7} sim (41 f)s 9a (64) 9 (4) ~ 92(4) 9 (64) ~ 24% 008 f, 008 sim (4,—f0) 5 and therefore = (lay 1) =2%bA 008 ¢, £08 fo sin® (f,~), =a cost, 008 fy +5? 008 (+8) 608 (fy +8) +08 08 (t,-+7) £08 (lo): We are to have Af ‘The merely algebraical possibilities {y=} and f1=4 are illusory, because ¢ is subject to an additive unessential constant : moreover, the conjugate (if there is a limit) must be related to f. Hence we must have A=05 and thorofore, at conjugate points, the tangents to the elliptic path are perpendicular to one another.175) MINIMUM REDUCED PATH. 261 (D) As regards the original integral ae v= de, ce ‘wo still have to consider the terms at the limits. By our general result (§ 198), whero the path is to end at a terminal surface, we have 2 enton to bo anise a ech extremity, hat aes of rend rey zoo: consequently, the cures to eu each terminal surface orthogonally ‘The foregoing integral V, when it is exprossible solely and definitely in terms of the final position x, y, 2 (and of the initial position z., yoy %) without reference to the mode of passage from that initial position, is ealled the Characteristic Function®. We then have, for Variations at the upper limit, Vm ude, that is, a : BY a4 ays U aemry (dra. Y ayy Ht wate dy TE demy (Gi ter M dys S de) for all variations dz, dy, de at that uppor limit, because Vis a function only of the variables of that limit (and of the lower limit, which does not affect the upper analytically). Hence av_ de aV_ dy Ov ds Bede? By" Mae? Beh de ‘hick, expressed in the vocabulary of goometrical optics, shews that the ray is normal to the eurfaces given by the level of the characteristic function, ‘The persistence of these level surfaces, normal to a gystem of rays, was fitst stated by Malus. ® Again, wo have | (y+) + Gy from the foregoing equations. Hence, when pis given as a function of postion, the fore going is a partial diferential equation ofthe fst order satistod by the charactrit function V. Its general integral, obtainable by Jacob's method for the integration of such ‘equations in more than two independent variables, is of the form Vase 4210, B46 shore a, 2, ¢ ate arbitrary constants, ‘Thocquationissatisfed for ll values ofthe arbitrary constants, andi therefore satisfied when they are subjected toa small variation so as to become @-+e4, B+f, in the one direction, and also <4, in the other direction, P, lying within the range from 1, to 4. Hence the178) ‘A GIVEN CHARACTERISTIC 267 equation IM (to, &)=| xh) Fe- 9" (th) Xe, 92(t), 8b), Olt) |, XG)-H'(b)Xe, M(t), (4), I(t) 0 + alte), s(t), x(t) 0 + Salto), Salt), Salto) and three similar equations, determine g,, gs, gs, 9. uniquely as four finite constants. As their values are now known, we infer the value of 7, and the value of 7; and thus all the elements of the contiguous characteristic curve are known, Hence a unique contiguous characteristic curve can be drawn as required. Corollary. If we were dealing with a characteristic as given in the earlier form $ (@, Ay, Ay As, Ad, ¥ (a, Ai As, As, Ad, the corresponding conditions are easily seen to be Omg, BD 4 9, HE (x) 5, Op ( to, 0), ar a) OA,” +H 0 HE 9, AE 4 LD a mn) a) Yang ta oan Aa g A BBLS 9 ABO, ’ (2) a a a ) =e Ke AOD 9, Hg AH, and the inference, as to the existence of a unique contiguous characteristic curve, is made as before. +96 +9 Integrals involving any number of dependent variables, and their first derivatives, 178. The following results, analogous to those which have been obtained for integrals involving two dependent variables and their first derivatives, apply to integrals involving any number of such variables and their first derivatives. Proofs can be obtained by analysis exactly similar to that used for the simple instance, when there are two dependent variables. An integral is given, involving n—1 dependent variables ys, -.., yx and their first derivatives p., ..., Pq with respect to the independent variable «, in a form fre. Yes Pas Yor Pay vv0y Yous Pu) dee268 TESTS FOR INTEGRALS INVOLVING (cu. ¥ (A) In order that the integral may have a maximum or a minimum, the n —1 dependent, variables must satisfy the characteristic equations af _ 4 (af). ue )"° for 7=2, 3, 0.) n. (B) When the limits of the integral are fixed by assigned data, those data determine the 2n—2 arbitrary constants that occur in the primitive of the characteristic equations. When a limit of the integral is not fixed, so that a variation at that limit represented by £, ms, ns, -.-. 1m must satisfy one or more equations initially prescribed, the relation Arndt a Pugh) + mg ben +mge must be satisfied at each such limit. When conditions (A) and (B) are satisfied, the first variation of the integral vanishes, (©) Let the primitive of the characteristic equations be Ye= be (% Ay Arye A, where A,, Ay, ...) Aoy-s are arbitrary constants, and r=2,...,n. Take varied arbitrary constants Ap + eit instead of Aq, for m=1, 2, -., 2n—2; and write ay Oe Also, let af 20=25 Fan, 2 ay 2, poo +a ef + 2250p where the summations are for r and s,=2, ...,n, independently of one another. Then the quantities 7, ..., m Satisfy the equations 20) d (20 bye de (Br) =° n; and they constitute the primitive of these equations (called the subsidiary characteristic equations), when @,,..., my» are taken as arbi- ‘trary constants. (D) Let the small variations applied to the original integral be y,+ «tp in place of y,, for all the variables, ‘Then the second variation of the integral178] ANY NUMBER OF DEPENDENT VARIABLES 269 can be expressed in the form ay Be PeZapip the double summation being taken as in (C); where Yp=0y + 3 demas = and the equation tra! + 2 dentine =O holds for n — 1 linearly independent integral-sets ay Make oy of the subsidiary characteristic equations given by k= 2, ..., m (E) In order that the second variation may have a persistent sign, either always positive (for a minimum) or always negative (for a maximum) for all weak variations, two tests must be satisfied: the extended Legendre test, relating to the derivatives £& and the extended Jacobi test, relating to the range of the integral. (F) The Legendre test may be stated in various forms, The form, that seems simplest for inspection, is to require that all the inequalities Of of (OF apni pe (Grd) shall be satisfied for all combinations of r and m, = 2, Op? Opt most all have the same sign, and are not to vanish anywhere in the range. When the sign is positive, the integral admits a minimum when the sign is negative, the integral admits a maximum. (G) The Jacobi test limits the range of the integral so that, beginning with a lower limit a, it shall not extend as far as a (greater) upper limit «,, where 2 is the stallest root «,, greater than z,,of an equation ® (2, ¢,) =0; and this quantity © (z,, 2) is | ao), dao), | ba) baler, | | Palm), bua(a), $u(%), Gala), hn (%), han(a)y Pama(m) |, Ps, mn-a(m) ‘Pn, maa (mi) > $s,e-2(%) $5, maa (ts) bm (%), Pan (a)s Pn, na (ta)270 ‘TESTS FOR GENERAL INTEGRAL [ca v = 2bm(@, Ar, eed eee (H) There are corresponding theorems for the integral, when it is taken in the form f Fis 4:1, Ys Yes os Yas Yu), where the integrand F satisfies the identity OP ar Pay Tit te FeCHAPTER VI. INTEGRALS INVOLVING TWO DEPENDENT VARIABLES, WITH DERIVATIVES OF THE SECOND ORDER: SPECIAL WEAK VARIATIONS, First variation of an integral. 179. We proceed to consider (but with merely brief developments) integrals which involve one independent variable, two dependent variables, and derivatives of the latter of the first order and the second order. The typical integral can be expressed by [Rrens.xinecde with the customary limitations on the function f. The enquiry is as to the conditions under which this integral can assume a maximum or minimum. In the present chapter, we shall deal almost entirely with special variations. Thus, we except « from variation, save at the limits; there, variation may have to be imposed on «, in order to admit the possibility of mobile limits, unfixed specifically by express data, We suppose y and 2 respectively changed to y+«Y and 242, where Y and Z are regular (thongh arbitrary) functions of «, independent moreover of x. ‘To include the instances of mobile limits, we suppose the upper limit to become a, +«U; and the lower to become 2, +U,, where U, and U, will govern variations at a boundary curve or on a boundary surface: these boundaries being supposed given when the extremities are not fixed. ‘The completely varied value of the integral is jeetet jes S(O YH RY, +HY' "$0", 24 el, E+ KL, +4") de. By arguments precisely analogous to those in the preceding investigations (such as in §§ 32, 82), we can express the variation of the integral in the form ef (d+ vier) + (Lea Len Bhac +e (Of Uf) + Ks, where f, and f, are the values of f(z, y, y's y’, 2, 2, 2") at a and at a respectively, and where K, denotes the aggregate of terms of the second and272 CHARACTERISTIC EQUATIONS [ow. vt higher degrees in x. Now [rv Bane [PRT -f eG) fede eral and similarly for the terms in Z’ and 2” in the foregoing integral. Hence the full variation of the integral is LE g- a(a)* B@)+ aff z@) +E) a sal oper (Z- 2 (Her G+2i8- a Lire ze +Ky Characteristic equations: terminal conditions. 180. In order that the integral may possess a maximum or a minimum, the first variation must vanish for all admissible variations; consequently, each portion of that first variation must vanish, when it arises from variations absolutely independent of all the others. ‘Thus, first of all, ile-s RE B-a@) GP) 8a @)+ eG and therefore, as «Y and xZ are independent of one another and are arbitrary, we must have aye x) ieliy)* da Gy) -° f-a (2) iG _ These are the characteristic equations, current along the whole of the skew curve which is to produce a maximum or a minimum ; and they deter- mine the nature of the characteristic curve. 181. Next, we bave [are rae ea) +e elo It is possible to have a variation at a mobile limit, quite independently of all variations (if any) at the other limit. Hence the relation Uf+ +e rr he iy £@}+ +a tn 0 is satisfied at each limit. ‘The quantities Y, ¥’, Z, Z' belong to a terminal variation of the characteristic curve along the #-ordinate alone, while the actual variation181] TERMINAL CONDITIONS 273 at the limit takes place along the boundary curve or on the boundary surface. If then V and W be quantities along this boundary as determined by the quantity U, the former analysis (used in § 86, ii) shews that Ye=V-yU, Z=W-¥zu, YaV-yU, Z=W-/0, When these values are inserted, the condition at each limit becomes Lo £@)-“- £88] . viz ae f+ wiZ-E B+ rbewd io (j) When the limit lies on a boundary curve, the equations of which are y=9(2), z=S(2), VaG@)U, W=9@U. Also, for this boundary, we have y¥=6@), #=9@), Vi =6"(2)U, W'=9" (aU. When these values are inserted in the foregoing relation, and when the terminal point is not fixed so that U is not zero, we have ’ {af _ a . ey @ ae) ee eG }- 9" ea (& ” 9 -e- 9% Conditions of an intermediate type might be assigned. Thus, if the extreme point were fixed in position but free in direction, we should have U=0, V=0, W=0, from U=0, ¥=0, Z=0, while ¥' and ” are arbitrary; then the resulting conditions are and so which must be satisfied at the limit, If the extreme point is fixed both in position and direction, then Y¥=0, Z=0, Y’=0, Z'=0, at that rigorously fixed extremity. (ii) When the limit lies freely on a boundary surface, the equation of which is S(@y,2)=0,214 PRIMITIVE OF THE [ox. vr then for any direction on the surface at a point, 78, ay f+ Consequently, we have - as sy BU + 5s W=0, rs #8, BS, eS #8 7S, , OS +u(Stys mayt? # Be) tT snag + Vat oe) SOS | OS +w (Sete aye * a) = [At the limit of the range, we therefore have ¥_ 4%) _ 4 (Hp 9 ale) ae) 9 ap oP es (7S eS as (+9 geay +” game) * ae af aa as ,28), a8 £-aly) a +2 Fae) Fay? a d (a as , gos os BiG) Gare Y jyae* 2 ree x yoy of 28 ae” = ae° Similarly for conditions of an intermediate type. Form of the primitive of the characteristic equations. 182. ‘The primitive of the characteristic equations F_ ay, @ (#9 ¥_ 4% ay de (#) . zeley) % ae FAC) ie required, When we denote derivatives of f with respect to y, y', ’,2, 2” by subscripts 4, 5, 6, 7. 8,9 respectively, a to the conventions -te nha, gbgem foo and so on, the characteristie equations take the form Saal” + Seat” + feb "+ (Fol + fafa) 2" + 1 = 0, Soy” + fot” + (foo + fa —So) y+ fol 2” +. = 0,182] (CHARACTERISTIC EQUATIONS 215 where the unexpressed terms do not involve y””, y”, 2”, 2”. For reasons which will appear later, it becomes requisite (i) that fy and fy shall have one and the same uniform sign throughout the range of the variable «, and Gi) that the quantity fufo— fut shall be uniformly positive throughout that range, Denoting this positive quantity by o, the equations (resolved for y’”” and 2"”) give oy” =— 9" fufes — foa Fal + fa —Fu)} 2" | So fo! + fa fa) ~ Safad + os on = — "| fea Sal + fa —Ju) — fafe') ~ 2" (Safe! — Su fu + Sa—Ju)} + os where the unexpressed terms do not involve y", y”, 2, 2". When « and its derivatives are eliminated between these equations in the manner used in § 136 for a simpler ease, the final equation in y is of the form dy dy ). a7 function of ( Its primitive involves eight arbitrary independent constants; and, if these are such that, when 2=a, a5 = br, (for r=0, 1, ..., 7), then Yeh ah (oar Fienay tn th ena teayR, where F is a function of by, .... b;, and is a regular function of 2—a. After this value of y has been inserted in the equations that lead to the elimination of 2, they provide a value of « which is « rogular function of «—a, involving the eight arbitrary cooflicionts by, ..., by. Similarly, when the equations are combined so as to lead to the elimina- tion of y, we are led to an equation of the eighth order in z alone, the primitive of which involves eight arbitrary independent constants, If these were such that, when 2=a, ae er, (for 7=0, 1, 0.4.7), then Feqten(e—a)+ 3 (e— a} +...+ F(e— ay + (o— ay, where $ is a function of ¢, ..., ¢ and is a regular function of «— a. By using ‘the equations that lead to the elimination of y, the value of y is derived as a regular function of «— a, involving the eight arbitrary coefficients ¢y, ..,¢ ‘The two simultaneous values of y and 2 in the former case, and the two simultaneous values of y and z in the latter case, constitute separately the primitive of the characteristic equations.” They are, therefore, equivalent to ‘one another. But there is another form of the primitive, evenly balanced as276 PRIMITIVE OF CHARACTERISTIC EQUATIONS fo. vr between two sets of constants associnted with the arbitrary values, assigned to y, to s, and to their derivatives, when e=c; it is as follows. We take a dp, pd, wae, ds dq, a=% gp and then the second form of the characteristic equations is ap, dn, =P 8G Ys 2 Pas Pas Pas Gs Gor > Hm Que, Ys 4 Par Pas Par Dr Ges )- We thus have eight equations. There exists a primi equations such that, when 2= Y=, PAG, Pr=dy Pras, 25%, h=A, GM, =A, ve of these eight where ay, ds, day 45, 4, 4%, &, &y are arbitrarily assigned constants, indepen- dent of one another, provided the functions P, and Q, are regular in their immediate vicinity; and the primitive expresses the eight variables y, p,, Ps, Ps» 2, 91 Ya» 4s 88 regular functions of «—a. The expression of this primitive is Y= y+ a (@— 0) + $3 (@— aP + F (@— a) + (2—a)G, s=aq+m(e—a)+ Fi @- art Fe- ay +@- ayy, where @ and H are functions of the eight constants a, dy, day ds, @, &, 4, %» and are regular functions of « — a; and the values of p,, Pas Pas drs qos 4x ATC derivable from those of y and z, by the first six equations of the set of eight above. Each of the two foregoing primitives is equivalent to this new primi- tive, Manifestly we have %, b=, b=a,, byes, =O, =H, =O, =A} also by, by, By, by are functions of do, dy, day My, Mo» &y My as; and similarly for C4 Cay Cas Ope Properties of the primitive of the characteristic equations. 183. Some properties of this primitive may be established at once. In the first place, there are no linear relations of the form °, 5 4.6.(0)-0, 3 Ave183] INDEPENDENCE OF PARAMETRIC DERIVATIVES 207 ‘with constant coefficients A,, ..., Ay, where ov = pel =2Y) HO, =H, d=, o (=H t= 2, do=%, doa, paz a ae ae : Ho-rZ, Wo-%, work, wo-¥ HORE, Wo, woO-%, wo-¥ We have earth , (om 07 By — Se ay * ee ee Bat a get a Se (e@- aa 4 Ea aF by ert ay aby 4@z or a a ea rear 1a Da eG eect mt ye (w-ay' ab, , (a eee at 7 tei (=a) a, | (w= a) abe eran a ay "41a, ST 8a,* 61 da, * 71 BOO : $@)= wae + (on ap $(o)= (yards eae e973 Fi He= erar the ( oar os + (e— ap 3h If an identical linear relation 3 reGe(a)=0 could exist, in which the coefficients \ are constants, manifestly M=0, m=O, =O, =O, 80 that the terms (e— a, (e—a), (2—a)', (@—a)' may disappear from the identical relation. And Re a NRE ge ge te ° Ob, a, & ret a rd rnZeno, by ab, a Dedat ad +B ta mB an atte tah278 st sIDIARY (cm. vt so that the terms in (e—a), (e—a), (#—a), (@—a¥ may disappear from that relation. The determinant of the cocfficients of As, Ae, Xr, Ax on the left- hand sides is by, Ds, be, by oerare) If this could vanish, we should be able to eliminate a, a,, a, 4 among the relations: bg, bs, bs, by = functions of ay, dy, My, ds, My, Oy, Gay Ms, and have a resulting relation Qty aay ey Bay Bos Bas By) = 0, that is, a relation a Qos Bas bay Bas Day Bas Bay B= contrary to the hypothesis that by, 0s, stants. Thus the Jacobian J( , , are independent arbitrary con- by, bay bes by Cra foregoing relations can be satisfied only if As=0, A=, =O, =O. Hence all the constants 4 are zero. No linear relation among the functions $1 (@), «++, ds(a) can exist, with constant coefficients, Similarly, no linear relation among the functions ¥(2), ... Yu(2) can exist, with constant coeficients. }) does not vanish; and s0 the ‘The proposition is thus established. Subsidiary characteristic equations. 184. The curve—manifestly a skew curve—which represents the primitive of the characteristic equations is called the characteristic curve. ‘The differ- ential equations are satisfied by the primitive for all values of the arbitrary constants; and therefore they are satisfied when a small variation is effected ‘upon these constants, the variation being as arbitrary as we please. Accord- ingly, we take a small variation Ay + HDs» Gt KGay Gat KGsy s+ KGe, FOF dy, dy day ds» GF HGs, AKG, Ay AKGr, Ast KYss +++ Moy My May Ooy in the constants, where g,, .... gs are themselves arbitrary constants. When we denote the changed values of y and z by y +n and z+«f, we have a2 oda t= Sov, when « is made very small, But the curve given by yan and +f, as arising out of y and s through the small variations of the arbitrary constants, is also a characteristic curve. The differential equations satisfied by y +n and 2+ «fare the same as those184] CHARACTERISTIC EQUATIONS 279 satisfied by y and z: and therefore parts of those equations, which appear as increments through the changes of y and s into y+en and 2+, must disappear. To express these increments, let 22 = far? + Yan's" + fal +2 fala + fo 5" + foo" + fob'S") +2 (fan + fond” + forth" + fa") t+ fan + Bar's! + faa +2 fant! +fand + fab + fatty that + York + fad The new value of Lis Fe (LyrZ c4 Digs Boe te Oh Sth apes tayay Tape © + apie’ + oe’), on neglecting squares and higher powers of «: that is, it is = Let fant Seb fal +fak! + fa” + Sab’) Ff, 2 eB. The new value of ¥ is and the new value of & ‘The first of the characteristic equations now is a and fa an @ a an a Z H+ 0S) + ga (tees) a0, and it is to be the same as the first of the characteristic equations in its original form, Hence an d san @ 722 Sn” de (Be) as (a) =° Similarly, from the fact that the second of the characteristic equations is unaltered, we have and fa @& 7a ae ae Ge) + ae (er) = In these equations, the twenty-one coefficients fig, (for m, n= 4, 5, 6, 7, 8, 9) initially are functions of y, y, y", 2, ¢, 2", As we are deducing another characteristic from the curve initially supposed known, the values of these280 PRIMITIVE OF SUBSIDIARY EQUATIONS [on. vt six arguments as functions of # are substituted in the said coefficients, which thus become functions of « alone, Consequently, the two equations £2). £(@2)=) gle) ar) ~°) are linear equations in and £ as the dependent variables, with coefficients that are functions of «, As the equations have been derived from the original characteristic equations by small variations, they are called the subsidiary characteristic equations, Primitive of the subsidiary equations. 185. When the equations are expressed more fully by substituting the explicit values of the derivatives of 2, they become Fa” + foss” + fo” + (Sul + fafa) 6" +» Fast” + faa + Sea + Safa) 0" + for 6 + the unstated terms being linear in 7, 7, 9, £¢', §". The earlier discussion shows that the primitive of these two equations contains eight arbitrary independent constants, ‘The equations are linear in the dependent variables ; hence these arbitrary constants occur linearly in the primitive, Thus we may take the primitive in the form 8 & n= = gm f= = arte, where 7, and f, are an integral-pair for r= 8, and where gh, +... gs are arbitrary independent constants, provided the eight integral-pairs are linearly independent : that is, provided linear relations £ Anno, $ 4-f-=0, a m with constant coefficients A, do not exist. But we have seen that, when y and 2 are known integrals of the character- istic equations, other known integrals of those equations can be derived by small variations of the arbitrary constants in y and z, so as to give yte abe (@ 240 > mre) Moreover, it has been proved that simultaneous linear relations $ xde=0, E rvela)=0, with constant coefficients A, do not exist. Hence we may take n= br(a), b= Yr (es186) PROPERTY OF GROUP OF INTEGRAI-SETS 281 for r=1,..., 8; and therefore the primitive of the subsidiary characteristic equations is given by aa Sader t= 3 one where g,, ..., gs are eight independent arbitrary constants. Invariant property of « fundamental group of integral-sets of the subsidiary equations. 186. Consider the determinant constructed from the eight functions $r(a) and the eight functions y,(2), and from their first three derivatives, the diagonal leading term being H:" (2) Ba" @) $s (@) BA(2) Wu" (2) Ya! (2) Hy (@) al). Let the value of the determinant be denoted by V. Then a is the sum of two determinants of eight rows. ‘The first of them has G1" (@) he" () ds (2) ba @) Ys" (@) Pe" (Ve (@) Ye), for its leading term ; and the second has Bi" (@) 1" (2) by (@) ba@) Ye" (2) We" (2) We (@) Ya(@) for its leading term. Now ¢,(2) and (2), for r=1, ..., 8, are an integral- pair satisfying the subsidiary characteristic equations: and therefore, for each such integral-pair, Fabe” + fase” + fod Ge" + (fo + faa) Ye" + 10 = 0, Sabet fase + Sas + fa~ Fo be" + Sal Yr + = 0. Consequently ae” + | fas Fn— Seal So’ + Fra — Sea)) $e" + { fal + Sra Sn) fasfoo | Wo" +=. =, ob + | Sa Sal + Fea Jea) —SraSe\ $e" + | far’ — Ser fo + Far fu)} Yo where o denotes the quantity fu/a—fe% When these values of $,"” and y,/ (for r= 1, ..., 8) are substituted in the two determinants that occur in the 0, “on for & expression for =, we have ° a =~ (ot fu— Fou ful +- fa fu)} V - (Sahu! ~ fo ful + fafa) Y de --v and therefore Wo = constant. ‘Now ¥ cannot be zero, because linear relations 3 4,4-(2)=0, Save) =0,282 PROPERTIES OF COMBINATIONS OF [on. vr with constant coefficients 4, do not exist. For reasons that will appear sub- sequently, o is always positive throughout the range of integration. Thus the constant, which is the value of Yo, cannot be zero; let it be denoted by 0. On the other hand, Cis not an arbitrary constant; for, when the character- istic curve is settled, no new arbitrary element enters into ¢ ; and the functions $,(a) and ¥, (2) are specific. Thus C is a specific constant, different from zero, and dependent upon the specific choice of the integral-sets ¢,(2) and -(z) of the subsidiary characteristic equations. Hence we have Wo=0, where Cis a specific constant, Consequently, the function W(z) does not become zero or infinite within the range of integration; it keeps the same sign throughout that range, Moreover, this result is substantially independent of the particular choice of the fundamental system of eight integral-sets $,(2) and ,(2), of the subsidiary characteristic equations, If another fundamental system of eight integral-sets, x, (2) and «, (2), were chosen, we should have relations aee)= 3 habe) rte) = 3 kevvr(a), for r=1,..., 8, where the determinant |r| of the constant coefficients k is different from zero. ‘Then ¥ Le), wr a] = Leal ¥ (0), Yo) =lknl 2, so that the result of changing from one fundamental system of eight integral sets to another such system is merely to change the specific constant value of the funetion Vo, In fact, the function V (2) is covariantive for all forms of primitive of the subsidiary characteristic equations, Particular properties of combinations of integrals of the subsidiary equations, 187. A fow properties of the subsidiary — equations Ge can be established at this stage, with a view to their use in the later reduction of the second variation to its normal quadratic form,187] ‘TWO INDEPENDENT INTEGRAL-SETS 283 We denote by Q, the result of substituting 9, and & (for r=1,...,8) in place of 7 and £ in the quantity ©. Then we have, identically, he 1 2Me 4 BD | Oe, 4, 2%y 9 By 1 Fat ane TO ange Sag, + 8 age + 8" see 20, 20), 8, 20 |, BM IGG EM Fal tM Girt Segg +O Bett Eset because each side is equal to one and the same quantity, lineo-linear in the two sets of quantities m, and &, n- and f, and their derivatives. But 1 a Brg 2 sf ZH wale de eh mee: with similar modifications for the other three combinations of terms. Hence one de) + By fee ad (op) + 8a -of- 2 pen Bre A LDL eM, where ©, is a constant, obviously specific and not arbitrary. When G, is not a zero constant, we can always take another integral-set, 7 and f, such that the new constant for the new functions is actually zero. Let the foregoing relation be written 90, NHAC, 1) +0, Take another integral-set, n and , linearly independent of m, and &, and linearly independent also of 9, and £,; combining it similarly with 7, and £,, we have g(Q, s)=A(1, 8)+0,, where C, is a constant, The unfavourable cirumstance would be that C, should be zero. In that event, we take m= Cane Cone b= Orb — Orbe, s0 that m and £; is an integral-set, linearly independent of m, and £,, of and £,, and of m, and {,. Then, for and £; thus defined, gL, J=h(, 0, a combination of n and £; with , and £,, such that the specific constant Cis actually zero. ‘As any integral-set is linearly expressible (with constant coefficients) in terms of the eight sets in a fundamental system, we clearly may make n, and $+ take the place of 7, and £, in the fundamental system. Moreover, », and £, were taken to be any integral-set, linearly independent of 7, and {, and of n- and {,; hence, even if C, were not zero, six choices of integral-sets are possible,284 SIX RELATIONS BETWEEN [ow vi after m and {, n, and £,, have been chosen, such that the specific constant C for the combination is zero. Hence we have the theorem* that an integral- set m and f, can be chosen in a number of ways so that the relation “(= 2 Gojoe 2 CB a h h 7 2D, an, a 0a, no -£ Bow Bbc LGB] oe holds, where the set n, and &, is distinct from the set n, and &. Six fundamental relations. 188. The foregoing analysis shews that it is certainly possible to make six selections of integral-sets »- and f,, linearly independent of m, and &, such that the foregoing relation holds for each selection : let these sets be denoted by t=2, 8, 4, 5, 6, 7, s0 that, for these values 9, )=h0, 0. Now take m, and & as an initial set, and let m denote any one of the integers 8, 4, 5, 6,7. As before, we have 92, m)=h(2, m) + Br, where By is a specific constant. Should By not be zero—and we shall allow the least favourable circumstances, by assuming that no one of the constants By. is zero—we take new combinations Byrn — Buin = Your BuSn— Bubn = ms for m=3, 4, 5,6. Combining (i) m and & with yp and zm, and (ii) m and & with y,, and s,,, we have g(1,m)=h(1, m), 9 (2, m)=h(2, m), for m=3, 4,5, 6. Lastly, take y, and s, as an initial set, and let n denote any one of the integers 4, 5, 6, ‘Then, as before, combining y, and z, with yq and zq, we have 9 (3,n)=h(3, 2) + Oy, where C, is a specific constant. Should Q, not be zero—and again we shall allow the least favourable ciroumstances by assuming that no one of the constants C,, is zero—we take new combinations Cogn —Cnye= Vn, Crtn—Cnte= Zn, for n=4, 5. Combining (i) 9, and & with Y, and Z,, (ii) ye and 4 with Yq and Z,, and (iii) yy and z, with Y, and Z,, we have 90, n)=h(1, 2), 9 (2, n)=h(2, n), 9 (8, n)=h (3, ), for n= 4, and n=5. “In ite present form, it is an extension of the theorem, proved in § 148 and originally due to Clebech,188] FOUR SELECTED INTEGRAL-SETS 285 Alll these integral-sets 7, and &, , and &, ys and z, Y, and Z,, Y, and Z, are linearly independent of one another. The fundamental set of integrals, for the linear expression of ali particular integral-sets, can be modified so as to include, in an altered aggregate, these five sets; and this composition of the fundamental set has been made in the circumstances least favourable to our present aim, ‘We now take any four of the five sets, which certainly possess the specified property. Let them be 7, and &,, for r=1, 2, 3, 4: we have six relations AM, _ d (ADn’ 1 8D, AQ, _ d (20, ' +1 82, 1 Gna (it) BB ed ~ de (Se dz Oh de AQ H 1 32, (285 a (@ } 22, [bm de \Em”) , 2, 3,4. We take them in turn for (Bett tm Spee + bo +h ae for the six arrangements out of m, 7, the arrangements m,n, =1,2; =2,8; =8,1; =1,4; =2,4; =3,4. First, multiply them in respective succession by |™ mi, [ae My 1 Te a | » ms [ee . Io &l 1h & be & bar bs b& & b, & and add. Then Seem &l4| Bo ms & [=0, ee (oe See me BFE Be we & Beam 6) | eo & an, ’ ” an? Te be age? bm G which may be briefly stated in the form Bye mm Ble Bo Secondly, multiply them in respective succession by bi be i, & , 8; = 3,1; = 1,2; and add. Then 60, d an, - nt ae (Ggt) ‘Thirdly, multiply them in respective succession by m, S|, ar 1,4; =2,4; mes we [+286 RELATIONS BETWEEN INTEGRALS OF SUBSIDIARY EQUATIONS (CH. VI for the same combinations of | and r; and add. Then Fons te le] BRE), os me t [a0 Fourthly, multiply them in respective succession by | oom I oom for the same combinations of / and r; and add. Then Bos ee ne] BE (Ee), & won [=0 Fifthly, multiply them in respective succession by wm l ms Te for the same combinations of | and r; and add. Then an, amd @) ag fm a+ ‘9G ~ dela) om Sixthly and lastly, multiply them in respective succession by ms oF | nes be for the same combinations of ! and r; and add. Then oo de ee aye Bee). mi Bo m+] Fe dese) mH & [=O First variation, by the method of Weierstrass. 189. ‘The second variation of the original integral will be considered only for the special variations y +«Y, 2+#Z, of the variables y and s, while 2 is loft unvaried save possibly at the limits. ‘The whole problem will not be dis- cussed for general weak variations, We shall therefore now deal with weak variations which affect all the three variables «, y, 2, only so far as the first variation is concerned. The investigation is desirable, not least because the customary analysis for skew curves gives no preference to one of the variables, over the other two, as the independent variable. ‘The characteristic equations and the terminal conditions will be identified, for the special weak variations and the general weak variations: and a proposition as to the continuity of certain quantities through a free discontinuity will be obtained, But that will mark the limit of the detailed use of the Weierstrass method. (See § 206, for statements of results.)189] GENERAL WEAK VARIATIONS 287 We make a, y,2 to be functions of a new independent variable t, which can remain unspecified intrinsically; after characteristic (and other) equations have been formed, we frequently can select s, the length of the are of the curve from a fixed point, as an intrinsic independent variable. Moreover, this mode of proceeding enables us to deal with strong variations where, as will be seen, the ordinary ‘first’ variation and the ordinary ‘second’ variation of the original integral do not occur. (‘The subject of integration does not admit of expansion in powers of x for strong variations, because these imply abrupt changes of the arguments.) Derivatives of 2, y, 2 with respect to ¢ are denoted by suffix numbers; thus de a or and similarly for y and for z, Hence a eee 1 P= BG Ne) 2” =F (ee— a0); and the integral f [fe y, v9". 6 ¢, 2") de becomes fre, By as Ys hs Yor Fu He) db, where FE, 1H Yrs Yor BA AaB (EY YY, 2 2,2"). As in earlier instances (§§ 38, 97, 156), the funetion F satisfies identities, expressible in the form of partial differential equations. We have ar Oa, ar ay ar_ z.288 IDENTITIES SATISFIED BY INTEGRAND {ew vr Consequently, two identities are to be expected after the climination of f, be af af of we j and they arc* or or oF a an +h May, +as- Da oO OF Pa 5 OF a on 4 oy OF 4 00, OF F= ag tM ay, tHe, + aa + Hay, 7 ns, Moreover, we have and therefore and, similarly, ~h)-te- ae): Further, we have (by the second identity) ar OF 2 0F grok Be, 7 a Oy, 8, Biey ,oF oF ee OF oF aF. on substituting for 5, Fy ge,3 aso so that, vin a a ug) -* oi we 4 ( Wh 1 Of --1a%, EG) Vee "If they are regarded as simultaneous partial equations of the first order for the deter. rmination of an unknown quantity ¥, their most general integral is (ey AMEpe, 4 a, aan) wu) ~ Ras le BYE arbitrary fonction of its arguments,190] CHARACTERISTIC EQUATIONS 289 consequently aP a (aR) _ Fd (aP OF dry .% a baa (in) “SY fon ~ su)h~* fie a }-¥ ay ~ a” = af _ a (¥)_,(er_a Apa 1 ble? B- EO} -v Eek. ‘These results will be used in connection with the terms at the limits, arising out of the conditions for the evanescence of the first variation of the integral. a2, Transformation of the first variation: characteristic equations. 190. The varied value of the integral fray, when @, y, 2 are changed into u+nX,y +e, 2442, while « is a small arbitrary constant, is frcetex, a teXy, wale, yteP, wreaks, yt e¥, Btnd, thy, t+%B,) dt; and therefore, when the function in the latter integral is expanded in powers of x, the increment of the integral is [(y oF oF oF OF aF or el(x Ge Migg, thie ty Mgt age OF or OF” FATA +L) de K, where K represents the aggregate of terms which are of the second and higher degrees in x. (As the variable ¢ is parametric for the curve, and account is taken of variations of « as well as of y and of z, the limits in the two integrals are the same,) Now oa [ee] Fa Ge) oF oF d (aF\ @ (aP . aoe [2-74 Q)] +{¥ ge (Ze) a hence the part of the foregoing integral, dependent upon the quantities X, X,, %, is equal to (ar d (oF) @ yy @F d /aFy) oF [efi aG) +a Gt a+ [a ES (eene]. where the terms outside the integration are to be taken at the limits of the original integral. The parts of the integral, dependent upon Y, Yj, ¥, and upon Z, Z,, Z, respectively, are to be modified similarly; and so the integral,200 [TERMINAL CONDITIONS [ou vr expressing the first variation, is fo, + YE, +2ZE,) dt Fd (Fy ard (OF)|, , {OF _ a (a + E-a@) Be -2G)} tea Ga)} aP , y aF , 72 +Xige + Ni5y + aay, where the terms in the second line are taken at the limits, and oF d (aF\, d aF Bene Gn) * ae Ge) oF d (aFy , d& (aF By 55 a (ag) +a Gn) F OF _d aF\, @ (oF) ear +2@)| In order that the original integral may possess a maximum or a minimum, its first variation under weak variations must vanish ; and therefore the fore- going expression thust vanish for all admissible variations. The part of the expression, comprised by the integral, must vanish by itself; because there are variations, zero at the limits and not zero within the range. For such variations, the terms at the limits vanish; while, subject to the acquisition of zero values at the limits, X, Y, Z are arbitrary quantities within the range, varying independently of one another. In order that the integral may vanish for all such values of X, Y, Z, we must have E,=0, By=0, &, which (as usual) are the characteristic equations, Terminal conditions. 191. Next, for the terms at the limits when variations, other than those {just indicated, are imposed, For these, the portion of the first variation, com- ‘prised by the integral, vanishes because the characteristic equations must be satisfied. Among the variations, the set must be included which impose no change at the lower limit; for them, the terms at the upper limit must vanish, that is, the condition xe ae +x fF SiGe)} + fig aelon)) +2 eae Gam must be satisfied for such variations as are possible at the upper limit.192] ‘TWO EQUIVALENT CHARACTERISTIC EQUATIONS 291 We similarly infer that the same condition must be satisfied for such variations as are possible at the lower limit, It thus appears that, in order to secure the compulsory zero value for the first variation, (i), the three characteristic equations must be satisjied; and (ii), the prescribed condition must be satisfied, at each limit separately, for such variations as are possible at the respective limits. The three characteristic equations are equivalent to two. 192. ‘This investigation shews that three characteristic equations arise which must be satisfied: the former investigation (§ 180) required only two such equations. But for the form of integral now under discussion, the func- tion F satisfies two identities, whereas the function f of the former integral was not thus restricted. It is easy to prove that, in virtue of these identities, a single relation connects the three equations, We have aF_ OF oF, oF OF oF, aF, OF oF ar De TM ay TH Ge te, TMG, t Hae, t GG, tM Te whatever be the form of a function F as involving arguments 2 ri, Y, Yin Yor 2,214, Because of the second identity satisfied by the function F, Genie ae Ge) +8 Gp) to) (ak + 2nge tong + nF 4208 BE) + ang) +245 (2) Let the two values of 4 be equated: then ofS) +n SCD] fE-400} not 42nd (2) + 5p +20 (Se) tage +208 (MP) wte(Oam tae) -8 dG) ae Ge) ae): and therefore, because of the first identity, 2B, + yEy +E, =0. ‘That is, the three characteristic equations are connected by one relation; and 80, because of the two identities satisfied by F, they are equivalent to only two independent equations, the number of characteristic equations when the tn- restricted function f is the subject of integration.292 COMPARISON OF TWO SETS OF TERMINAL CONDITIONS (CH. VI Comparison of terminal conditions with conditions in § 181. 198. Further, the form of condition at the limits, which (§ 191) has been obtained, can be harmonised at once with the earlier form ($181). The new form of the condition is Fy WF, F MeN tage OF_d QP), -QF_a ary), ,F_a +2 ae Ear PF a a ES Now Bn) SY fp ~ ae (op ar 2 @ -¥-2(%), j-th ay, dE Gy) ~ dy’ dey’ oF 4 (2) =-4-2(%): G2, dt \de,/~ 2 da \ae")? also or F_ 4 0F ar aa, yew, Be, Y By,” Oe,” so that oF oF ar oF KEM tags ny For comparison with the earlier form, we have X=U, Y=V, Z=W. ‘The change in the value of y/ is nteh_w ateX, wy =5@M- 2X) + neglecting powers of « higher than the first: thus Vine L(Y -¥X), so that Yi-yXaav, and therefore ar ar 7 (= YK) aE Similarly, r 2x) e194] DERIVATIVES CONTINUOUS THROUGH FREE DISCONTINUITY 298 When these values are substituted, the agreement between the expressions for the conditions at the limits of the integrals in the two forms is complete. ‘Tho detailed inferences from the conditions, according as an extremity of the characteristic curve (corresponding to a limit of the integral) (i) is fixed in position, with or without a fixed direction; (ii) is obliged to lie on a boundary skew curve, with or without restrictions as to direction; or (iii) is obliged to lie on a boundary surface, with or without restrictions as to direc- tion: are, all of them, the same as in § 181, Continuity of certain derivatives through a free discontinuity on the characteristic curve. 194. It has been assumed that no point of discontinuity in direction, or in curvature, or in both direction and curvature (but not of actual currency), of the curve can occur in the range. Now suppose that some such place occurs; the course of the argument will be seen to allow that a finite number of such places may exist. At each of them (it will be sufficient to prove the properties at one of them), the quantities ar aF_ ag) oF OF_d Py aF oF ¢ dm’ Gx, dé\Ber)" dy," dy, dé Bye)” B25" de, ~ dt NB)’ are unchanged in passing through an isolated free place of discontinuity. Let any such place be given by the value 7’ of the independent variable t. Let ¢’ be any place just before 7 and ¢” any place just after 7, so that t arb’, ayn, % arb, va re ma a vanishing simultaneously in the range. Therefore the range must not extend 80 far as the first value of x, greater than the initial value 2, for which these “four quantities simultaneously vanish. Thus we have a possible upper limit for the range, and denote it by 2, when there exists such a value of «; and then the Jacobi test makes ©, the upper limit of a range which has 2, for its lower limit. ‘The range of the integral, consequently, is not to extend so far as the conjugate of the lower limit. We must obtain the analytical determination of the upper limit «,, to be taken as the conjugate of z, the lower limit. Critical equation, in connection with the Jacobi test. 202. The four integral-sets, 7, and {, 7 and & m and &, mand £4, are linearly expressible (§ 184) in terms of the eight fundamental intogral-sets br(2) and y, (2), for r=1, ..., 8. Let n= 2 bb b= 3 bevel) 2h6@ b- Zt ee so Emo a = 1, r(2), ‘The quantities ey and xf, in each instance, denote a small variation from the characteristic curve under consideration, chosen so as to give a contiguous characteristic curve, For every curve, the constants are determined by initial values chosen at the lower limit. Accordingly, we choose 8 = m Yr (2), =, &=0; m=0, Sm for r =1, 2, 3, 4, at the lower limit z,—a selection which makes each of the four contiguous curves touch the fundamental characteristic curve at that lower limit, Among these four contiguous curves, we discriminate by the assignment of other initial conditions at that place # =<), as follows: Gi" =0, a" =0) hi" =o, a" =0 : g"=0, cee w= op308 CRITICAL FUNCTION IN (cx. vi where p, 0, 7, # ate finite non-zero constants, Manifestly, the curves are distinet from one another; and each of them is distinet from the character- istic eurve which (§ 182) would arise, if either p or ¢ or + or # were to vanish. All these assigned conditions are in accord with the six relations of § 188, satisfied by the four integral-sets », and £, (for r=1, 2,8, 4). ‘Those relations require that certain constants are to be zero: when the foregoing values are substituted, they do, in fact, make each of the said constants equal to zero. ‘The eight constants hj, ..., ks in the expressions for n, and f, are thus subject to the eight linear relations Lhrbr (%)=0, Sher (@)=0, TkrG (@)=0, Lkeyr’ (4) =0, hye” (a) = py Ser" (as)= 0, Serge" (a) = 0, Shep” (m) = 0. The determinant of the coefficients of the constants & on the left-hand sides 1b (oo), Yoo), os’ (xa), Pu (ods Ba (to), Yo" (ads i” (ae), Yo" (a) | is the value, at ay, of the quantity V of § 186, a quantity there proved not to vanish anywhere in the range: we denote this value by W.. Let [£] denote the minor of any constituent & in the determinantal expression for Vs, s0 that, for example, [Gx” Co) = |e e), Yu wo), ui (eds Woe! (wads Ye" (we), 1” (eo), Yu” (He) | [Gs (ood) =| Gu (0), Yow), be (ad, Yu! ods Ya" (aa)s fr” (eds Ws” (ao) |» and so on. ‘Then we have kWo= p[ di" (a)], ba ¥o and so on: generally Lda" (aa) Ka o= p [spa" (ae)], ba¥o= ps" (as)), ke Mo= ploy” (a), ., 8. Similarly, for the same values of r in succession, — LWo=o[hr" (a), mVy= 1 [$e (wa), — me Vo= ely" (a) We therefore have unique finite non-zero values for all the coefficients in the four integral-sets 7 and f, m and &, m and %, and m and £.. for r=1, 203. Now consider, once more, the quantity @, where @=|%, tm m % Gs Sar bs me Ws ms me oo Bw Tt vanishes at the lower limit 2, of the integral; but no difficulty is thereby caused in the second variation, because La Y pa 40 +7V +82, Sak" + OV 492 +yV + xh208) ‘THE JACOBI TEST 309 It must not again vanish within the range of integration; and that range must extend 90 far as the first value «, of e which is greater than 2, at which the quantity could vanish. But (p07) Y= Zhe @)[e" (wd), Ede LY" (], Ede (a) [be CH), Shr (Ye Cod] | Eabr (2) Ube" CW) Se Le" (wal, rr (@) Lbs” Ceo], Er EY” (#o)] Ege’ (©) Ee" (oY Zhe! (w) Le” (ae)], Ede’ (a) (Ge (a) Zhe! (@) Le” (@o)] | Eee [Ge ed], Se’ Ye" (aod), Ed’ Ube” od], Ee) Le” ‘The determinant on the right-hand side can be evaluated by the repeated use of the known theorem relating to determinants made up of constituents which are the complementary minors of determinants, themselves made up of the constituents of a determinant such as Vo. Thus, in the expansion of the foregoing determinantal expression for @, the coefficient of dH @), br) $s (@), br (2) WE), HO), Ww, We) Gi (2), by (@), di(e), di (@) WO IO WO, We [d."” GL [oe” Cd, [4s Gd] [he Co] | - Gh" @) Be Gl, Ge" @L, WH @)) [i (em) C6" od], [ps [40 @)) IC) a CC AC) RC By the theorem quoted, this last determinant (made up of first minors of Y.) is equal to We] be (a), he (@), gr (a), he (We) | Vs (2), Yo), Vr (me), Ye (aa) $e (0), ge (i), (ao), s' (aed | We (ea), We (od, Yee), es (a) | Similarly for other terms in the expansion : the coefficient of Ide(2), vale), oF (@), Wi ()\, where p, g, r, &, are any combination from 1, 2, 3, 4, 5, 6, 7, 8, is We | peal), Ver) Fra(m) Yor (m)|, by the same theorem. Consequently, when all the evaluated terms are added together, we have Hime, Yale), G5), WE), bo (ee), Pole), 1’ (Be) Ys (2)| =A(@, 2),310 EXCLUSIVE CONJUGATE-BOUNDED RANGES [on v1 where A(#, 2) denotes the determinant of eight rows, made up of $(2), ¥(@), $(@), («,), and their first derivatives. Now ®, does not vanich, nor become infinite, The quantities p, ¢, r, p are finite quantities, respectively determining the four characteristics contiguous to the characteristic under consideration. The quantity is not to vanish along the range; henee the function A(e,, 2) must not vanish along the range for any value of # within the range greater than a. The conjugate of a, is given by that value of «, say , for which a i Zam=0, Sato, Zan'=0, 3 at'=0, without a, dy, ds, 4 themsolves vanishing: that is, for which @ vanishes. Hence the conjugate, given by «,, of the lower limit of the range of integration given by ay, is the first root (greater than 2,) of the equation Ale, a) =0. If the original integral is to possess a maximum or a minimum, the range of that integral, beginning at #,, must not extend as far as the value =, which defines the conjugate of ~ on the characteristic curve. We thus have the third test for the possession of a maximum or minimum; afver the analogy of preceding instances, it is called the Jacobi test. A range, bounded by two conjugates, does not enclose any similarly bounded range. 204. In enunciating this result, one property has been tacitly assumed by anticipation: viz. a complete range along a characteristic curve, between point and its conjugate, cannot contain within itself another complete range of the same type: in other words, the conjugate of any point within a range lies without the range, The establishment of this property can be effected as in the preceding cases (&§ 119, 176), by proceeding from the function A (2, 2), the outline being as follows, This function A(e%, 2) does not vanish anywhere that lies actually between «, and ite conjugate «,. Let two points be taken in the range, one near #, given by a+£, (where X and £, are positive, and X is small), and one near «, given by 2, ~ e£, (where ¢ and f, are positive, and ¢ is small); the two magnitudes Ay, e+ dfs), A (a, a — ef) have the same sign. But, approximately, for sufficiently small values of 2, we have A (oa, tot Mh) = EME Wo, while Ae, 8h) =A (es, a) ef 8 SH)206) SUMMARY OF TESTS ail so that ©) has a persistent sign, opposite to the persistent sign of ¥ a whatever initial point be chosen. Also, as A(#, #,—ef,) does not vanish, a Gu8) does not vanish. When a point a, +a£ is taken, with a and & positive, and ite conjugate 4, +a€' is wanted, a being small, we have A(e+ab, 2, +08’) = 0; so that, when a is sufficiently small, we have 20. (a, 21) pO (ae, eat If £ and £’ have opposite signs, the new range would lie within the old; and the distances of the new ends from the old ends would be of the same order of magnitude. ‘This diminished range would similarly lead to a more restricted range, because of the persistence of the sign of V. Gradually, as before, we could wear down the range (on the hypothesis that £ and £’ had opposite signs) until the fanetion A(z’, 2”) would palpably cease to vanish when 2” is sufficiently near a’. ‘That hypothesis must therefore be abandoned. Con- sequently, & and £ have the same sign; and they are of the same order of magnitude. Proceeding gradually from 2,, and taking the new conjugate for each new initial point—which conjugate must lie beyond each immediately earlier con- jugate—wo infor the property that the conjugate of any point within a complete range lies outside the range. =0. Summary of tests. 205. The conditions, when satisfied, secure the existence of a maximum or a minimum for weak variations. The Euler test and the terminal con- ditions make the first variation vanish. The Legendre test and the Jacobi test preserve one sign for the second variation—either positive (when there is a minimum) or negative (when there is a maximum)—through all weak variations that are possible. General weak variations : statement of results concerning the ‘second’ variation. 206. When general weak variations are considered, so that we have to deal with the integral [Pee 215 Ye Yes se DME, and the variations are «+ «X, y+#Y, 2-+Z, the results connected with the first variation of that integral have already been obtained. Much laborious algebra is required for the reduction of its second variation to a normal form.312 SYNOPSIS OF RESULTS FOR [en. vi The results are significant, not least in connection with the discussion of a characteristic curve, of characteristics contiguous to that curve, and of a range along the curve bounded by conjugate points, with the definition of conjugates already used. The discussion follows, in more general shape, the discussion already given ; and the algebraical calculations, in themselves and in their process, are the extension—an elaborate extension—of calculations of the same kind already effected for simpler cases. It may, therefore, suffice to give an outline only, arranged in successive statements, ‘Three fundamental quantities are definitely established, in (II); the other inferences are stated, without detailed proofs being given. (D The function F satisfies two identities 2F yy FoF PF. ak sy Ben homade Fs oy oe Pte ge Derivatives of F with regard to the variables 2 By Bey Ys Yr Yoo Bs fav Bn are denoted by numerical suffixes 1, 2, 8, 4,5, 6,7, 8, 9, respectively; and the variable quantities X, XX ¥,V, Va, Z 4 Zn are denoted respectively by 8, Ox, Bs, Oey B55 Bo, Ory Bar Bs ‘We shall want combinations of Y and X, of Z and X, and the derivatives of these combinations; we write b=V =m¥-yX=28- 4, bs Vi = 2,0, 91004 (00, —Y201) = ,0,— Ya + 2 (08.— 204) + (0, ~ 8), 2 = 5,X=2,0,-2,8,, 8, — £18, + (048, — 2,84), — 2,0, + 2 (0, — 2405) + (040, ~ 248,)- (ID) Differentiating the first of the identities satisfied by F with respect to a, to y, and to z,, separately, we have Fat Fut 4Fa=0, Fg + iF t Fy =0, 2,F at \Fat iFa=0. Consequently, there exist three quantities gx, ga, gu such that Fugu, Fo= gm, Po= 2m Fy =—2trga-%%9e, Fe=- GYgu— 22 Go, Fes= y'gut 2yr29u0t 2790 a206] (GENERAL WEAK VARIATIONS 318 ‘The three quantities gy, ga» Jw are fundamental; they are, save as to a power of z,, equal to the quantities fr, fy fis of & 182—5. (AIL) The first variation of the integral has been made to vanish, thereby leading to the chateeteristic equations and utilising the terminal data so that the limits of the integral ean be considered as fixed. ‘The second variation of the integral is ie fo dt, taken between the fixed limits, where = EE FanOnOns the full expression of © being Fy0°+ 2F 40,0, +...+ 2Fixh6.+ F,8% We take a quantity (C= AO, + BO: + 08; + 20,0, + 20,0, + 20,0, +00 + 502+ 00; + 2/00, + 2980, + 20,0, + 20, (a0, + 0,-+ 70.) +28,(NO,+ 105+ v0,) + 26, (pb, +08, +76,). ‘The first aim of the analysis is to modify the second variation, by expressing @+ a as equal to ®, where and, in ®, as few terms are retained as possible. For then le +90) a= fou +(a)= [ow because the limits are now fixed and therefore every variable 8 in [] vanishes at each limit (0,, 0,, 0, are absent from []): that is, fou=foa. ‘Thus © can be regarded as an intermediate normal quadratic form for the integral, ‘As regards the retention of as few terms in ® as possible, we have m6 W+ 2g WW" +2941 W" +g + 296 W' 4 gy? =G0W" + Gea Gu’ — 2m) W? + (Gn — Gs + gn’) W? +H iggW+ 20m WW’ + (oun — gn’) Wh The term in the last line, with its sign changed, can be moved into 2, and then only the squared terms in the first line will be left: that is, by a change of the coefficients gnn which initially are at our disposal, and without loss of generality, we can choose 9o=% ga=0, gn=0.314 SYNOPSIS OF RESULTS FOR cu. vi Similarly, without loss of generality, we can take ga=0, ga=, ga=0, in the form initially postulated for &. We therefore can take, in umbral notation, D= (he + rede + 11h + Mobs + Hobs + Yeh)? ‘The six umbral symbols 7%, 76, Ye» Y Ye Ye combine quadratically into the significance of reality, according to the laws Sous (the three fundamental quantities already introduced) ; Hr=0, HH=9, NH=0, YW N=, WE (according to the foregoing initial simplifications of ); and, for all other combinations, Yn Yn = Irons these non-zero coefficients gaa remaining. The consequent modifications, by the absorption of = [G0 + 2900 W’ + (Gr— go’) WH into [, will be supposed made, so that the coefficients in [] (as yet unused) will be regarded as having absorbed the necessary changes, (IV) Owing to the expressions for the real variables ¢ in terms of the real variables 0, we have = Yor Bo + (Zroite + yet) 85 + (Yates + Yetta + 11%) By 3m F 6,85 + (Quite + Yes) Os + (Ys + Yatta+ Yet) A — (m2 + Yeh) Bs — Qyots + Wey + W621 + Ys) Oe = (nts t Ys + V620+ Toye t He +1) Or ‘Then we are to have , os ae-{ nel) where, on the right-hand side, we substitute the equivalent expression which is quadratic in 6, (V) Equating the coefficients of the various combinations of the variables 64, --+, 8 on the two sides of the last relation in (IV), we have relations between the known coofficients Fis, the unknown coofficients gman, and the coefficients in C]. ‘The total number of relations in this aggregate is forty-five, being the number of coefficients Fn,206) GENERAL WEAK VARIATIONS 315 (i) The terms, involving (6,, 64, 6) in @ +29 and &, agree without jt any relation, owing to the expressions for Fg, (with m, n=3,6, 9) in terms of ‘the three coeflicients gun, Jer Joe: Gi) The terms, involving (6,, 0., 8,%0., 6, 64) agree, if Fat C= %wo.t% } Jays FotF= %at.te+ Gut? Pat =~ 2gnt22— 2getiyYs Fat P= 2gux%+ gar Fut B= gure, i Fug + H= — get 24 2G Yo ~ Joa 5 Fat G = — 2gnti4— Bayi —Gatyy — yen &s— Wears — Joss gotta + 2a (sige t ther) + 2gariss |” + 90h t Inrith In these relations, F occurs twice, G twice, H twice; and their repeated values must be the same. But from the fundamental identities, we have oF Fat Fut Fam — Se Fy + yiPrg + 2 Fu + 20, Pre + 2ysFun + 2e4Fs z. which give aA tyHtnGa ee, and I: Fas + Fos) + 2, Fo — Fs) = 2 (t4Fes + YsFrs + #2F x). Similarly Hay B+ Pa Se and 41 (Fa~ Fa) + (Fru Fy) =2 (0 F gt oP + 4F, and G+ y P4200, and @, (Fy — Fa) + ys (Fes — Fea) = 2 (Fon + Ys Fen + 2 Fy). Of these necessary equations, the three which are free from A, B, 0, F, @, H, are satisfied by the foregoing nine postulated equations, if only Fea— Fa= (Guu) 2? ‘Therefore the whole set of equations defines A, B, C, F, @, H, gus gu, always subject to the latter equation,316 ‘TRANSFORMATION OF (on. vr (iii) The terms, involving (6,, 8, 836,, 6, 6) agree, if Fett = gutsy Fotv = Gat %+ Got\het gat! | > Fat =—Gutiks— Jo%Ys— Jaye Jas Fate Ja X; + Yeu, Xs + Jos? Fat h= gute > Fig t B= Grates ~ Gui Ys— Gute Jo%e GooFs%s — Jeoyf%s— Joti Xe — JarTiYr GeoFaks— Gstiits — oFikra— Ju Goti2s+ Jo (Ys + is) +IaYiYs + Gato Jathls + Ioyit%i+ Jair which imay be regarded as defining the nine quantities a, ..., 7, in terms of new quantities gun, Jus Jer Jo (iv) ‘The terms, involving (6, 6, 6)* agree, if Fret OC +27 = Aguas + gua? Pot PF ty to= Agee + putts + gate + gut? Fat @' +4 +p Agmitata— AgeeteYs 2x Yo— Wu Ys ite — Pours — Jest Aguas + gut? gees — 49a — gusts &a— 2s Fe — Jour 2 — Jessa FatA' 42a = 4gnz + Bgetitet Sguye | + Agents Ys + Agayits + Inti? + Wayizi + Ieys? ‘The quantities A, ..., H, a, ..., 7 have been defined in the earlier group (ii) and the earlier group (iii) of relations. ‘These six conditions therefore define $us Jus Joss and they leave three relations to be satisfied. (v) The terms in (@,, 0, 8:16, 64, 6) agree, if Fatt $6= 2gutets+ Qutits Pat VAS = 2g e+ 2st + 2Gat is + Gusts + Joab + Gace Fist 1 +9 = —2gmtty2s— 2guitays— 2gut — 2a Gite = Ieatrea~ JesTrYs— IsTrYo— Jats Fat ot f= Yutatts + 29+ 2atits + 2a 2+ Getta Int? Fete +b = gutter t Jost, | Fist B+ h= — 2get.25— 2geaitays — 2Gertata — gents | : —Getsts— Guat te~ Goi Ys — Gorter Fut B +p Fat H+ B+206] SECOND VARIATION 317 Fat 0! +9 =—2gmtats— 2a yets — Ways — Waa state Foss — Gui %s— Gertiyy Fy th +h — 29a 2sits — 2grotrts — 2a te 2WostsYs — Gusts Junta Gast Juha Fatal +a= 2gntrts + 2ga (tryst Y%s) + 2geateYs + Gea 2yata+ Wits) + Gua (2yate + 414s) + GentiFet Jos (Ye + Yi22) + Gade + 2gathte + WatiYet+ Get In) Y% (vi) The terms in (6,, 6, 6,)* agree if Fate = goot? + gat! + guts Fit f= gate + (gout Jor) Cats + (Gea + Ga) trees + Pout + (Jaa Gx) tra + Gat? Frat J =~ Gnitsts— Jo%sYs— GnYe%s— Joye — GIeatsta—Jesta— Jetafe— Jayate = Gni%s81— Joris — Jato Iaith Fy t B= gets + Gott + guts Fu th! =—gaitsts— Gu%sYs— Jeaa%s— Jai % = Geos£s— Jua¥s— Jer? Bs Jutate = Gutits— Gu2a— Gosh ~ Jur ta = Gute +2GaYs2st Poor + Gust + 2GrsZYot Ia + Inzt + Wait tIay? F 2geeYats + Aue + WosyYs2s + Waza + Woyszr + Weyer (vii) Various relations are satisfied by the quantities Fn, which arise as derivatives of the two identities satisfied by the subject of integration, ‘Thus, from the first of these identities, we have Fy + Fat 4iFn Oy Py + y Fut AF 2 Fat yFat Fs and thence 204 yB+27=0, mrt Wet AY =O, aptyotar=0. Again, from that same identity, we have iF an Fat late han- ab — uF —20, =a Pgt Pot Fe aH—yB-aF, ena Fat Fat Fann d— nH —aG; 2G +P +50 +aG+yF +20 =F yt eka t %Pot Fat oka t Pot Fn t Fat km, with two similar relations.318 SYNOPSIS OF RESULTS (on. vi From the second identity, we have 0 = 0, Fat ¥: Fest 2 Fg + 20, F es + 2yeF eu + 2eF re, Oma Fn t Yi Fot Feet WeF s+ 2yeFugt 26.F oy, Oma Fn t Fat 2: Feat laa Fn + 24 Fest 25,Feo. When the values of Fy, ..., Fin, Faso» Fare substituted here, these relations are satisfied identically, Similarly with the other relations, that can be derived from the two identities in the same manner as those in Chapter II and Chapter III: it can be verified that each of them is identically satisfied when the values of Fyn are substituted, (VI) There are subsidiary characteristic equations, associated with the fandamental characteristic equations. Let 20 denote the same function of BEL En 0,60, 0", a8 @ is of @,, ..., 0, with the coefficients Fyn. The subsidiary equations axe a _ d (an ae ~ aloe’) * de aaa) * 2¢f)-. and em ad (8) ea oe) +aelie)=2 which are really equivalent to two equations in virtue of the relations among the coefficients. (VID) Corresponding to V and W, we introduce the combinations yan nk, fab We denote by 2T, the same function of y, yi, Ys, 2, %1, %, a8 ® is of v,V’,¥", W, W, W". Then y and z satisfy the equations rid @& at oy ~ (op) * ae Gye) =® ar det), & er de ~ at (ie) +a (fer) = which are the modified subsidiary equations. ‘The primitive of the subsidiary equations, and therefore also the primitive of the modified subsidiary equations, can be derived directly from that of the fandamental characteristic equations. Every integral-set is made up of linear combinations, with constant coefficients, of eight linearly independent integral- sets which constitute a fundamental system. ‘There are at least five such integral-sets, any four of which (in pair- combinations) provide six relations, identical in form with the six relations of § 188: a selected four are denoted by y, and 2,, yz and z,, y,and z, y, and &,. ao206) LEGENDRE TEST: JACOBI TEST 319 (VIE) By means of the relation Iv" fou= (e+) a—(v1 = fae, where V is a homogeneous quadratic expression in V, V’, W, W’, we can (by analysis exactly similar to the analysis in § 200) obtain an expression (g0W* + 29. W¥ + Ju"), where wi 8 Yn By Ye, Bs Yay Be Ws Bs Yan Be BL Ya ee w=| ee Yo 3 WB, Yh Yoo te Wer By Yar Be Yo", ss Bs Yar Be Yo, Bi, Yu My VW. Ye, Ry Yu We Rs Yo Be where @ is a quantity that does not vanish within the range of integration. This form {tat is the final normal form, as {dt was the fist or inter- mediate normal form, of the second variation. (IX) The discussion of this result proceeds in the same way as the dis- cussion in § 200—204, The Legendre test requires that, everywhere in the range, Guu > Ga" When this is satisfied, a common positive sign for gu and gy admits a minimum for the original integral, and a common negative sign for gu and gy admits a maximum. The Jacobi test requires that the range of integration, which begins at an initial place t on the characteristic curve, shall not extend as far as the place 4, the conjugate of t: where @; is the first value of (greater than t)at which the determinant [yi (G), 2'(f), Yolts Bt) Yel (to), Be (tds ¥e(4), a(t) | vanishes. (X) A complete range on the characteristic curve, bounded by a place and its conjugate, does ot contain within itself another complete range.CHAPTER VIL ORDINARY INTEGRALS UNDER STRONG VARIATIONS, AND THE Wererstrass Test: Soup OF LEAST RESISTANCE: ACTION. General notion of a small variation. 207. The systematic analysis dealing with irregular small variations, which can be represented by a jagged line resoluble into small straight pieces, is due to Weierstrass, He introduced it (and developed it, thereby meeting some general objections of Steiner) so as to discuss possible maxima and minima of integrals, which involve one origioal dependent variable and its first derivative. Jagged lines were recognised by Legendre* as possibly leading—in the case of Newton’s problem of the solid resistance, he obtained jagged lines which do lead—to results, reduced below the minimum} provided by the characteristic curve under small variations that now are called ‘weak.’ Such lines of a zig-zag type were excluded from consideration in their analysis by Legendre and succeeding generations of writers, through @ definition which was implicit at first and was made explicit only after- wards. It had been assumed, usually without any statement or only on a later definite statement kept unobtrusive, that, when the variation of the dependent variable y is maintained small, the variation of its derivative y/ will also be small. The assumption seemed to be that the smaliness of y’, consequent upon the smallness of y, was too obvious to merit even mention. ‘There was the further assumption, also implicit and ignored in statement, that the variation of y’ is to be of the same small order as the small variation of y. All such variations were represented by by, y, Sdy, and so on; the mere presence of the Lagrange symbol 8, in this association, apparently ensured the requirement of smallness. Moreover, in spite of restrictions in the range of the integral to small variations dy of the ordinate alone, circumstances compelled the consideration of small variations of « at the limits of the range (as at mobile limits, in § 81), though these were not admitted elsewhere ; and a compromise had to be framed, expedient for the discussion of such limits, * “Mémoire sur la maniére de distinguer lea maxima des minima dans le caleal de variations,” Hist, de V Acad. Roy. des Sciences (1788), ‘f Legendre (1c.) called # maximom or a minit208] GENERAL SMALL VARIATION 321 Manifestly the general notion of the slight deformation of a curve, into a contiguous curve, merely requires in essence that the distance, between the position of a point on the curve and the displaced position of the corre- sponding point on the deformed curve, must always be small, whatever point be chosen, If this requirement be met, the contiguous curve gives a small variation of the original curve, whatever be its shape in detail. We are accustomed, over a stretch of continuous curve, to very rapid small changes which, while of full importance in detail, would be deemed of slight statistical account as compared with the smoothness of an averaging curve over that stretch. Restrictions at every point, as to changes of direction, of curvature and the like, that all of these changes shall be small when the variation is small, are essential limitations upon the completeness of the variations admitted or imposed. A minimum for restricted (weak) small variations, which fails to be a minimum for unrestricted (strong) small variations, cannot be claimed as a true minimum, any more than the summit of a mountain pass could be regarded as a true minimum, because there is no higher point to which a pedestrian need rise when passing from the valley on one side of the pass to the valley on the other side, or as a true maximum, becanse there is no lower point to which a pedestrian need descend when passing from the ridge on one side of the pass to the ridge on the other side. A true minimum and a true maximum must possess their distinctive properties, in all circumstances of variation, and not merely for a selected group of such circumstances, Accordingly, we proceed to the consideration of strong variations, indi- cating by the term specially the types that can be resolved into an aggregate of variations of a specifically simple character. The character of these com- ponent simple strong variations must now be explained. Strong variations : fundamental constituent type. 208. In all the preceding investigations, the small variations of the characteristic curve have been of the type styled weak. ‘They are such that, if the point © +f, y +47 be the varied position of a point @, y on a curve, for which « and y are expressed as functions of a continuously increasing parameter ¢, the quantities £ and are arbitrary functions of ¢, independent of and limited by the requirement of being everywhere regular functions, so that they possess their due succession of derivatives which also are regular functions, We now proceed to consider small variations which are not limited to this class. But the mere exclusion of such limitation would be mainly a negative description. For the purposes of mathematical representation, it is convenient to select particular types of small variation from among those which hitherto have been excluded from consideration. Thus we have already (§ 47, 102) had small variations, where isolated discontinuities of322 NON-REGULAR SMALL VARIATIONS (ow. vir direction within a range have been admissible; though the number of such discontinuities in any finite range has been limited, owing to the require- ment of isolation. We have mentioned, though only by the way of illus- tration of possible small variations that are not of the definite weak type, the instance = wasin(te"), Such a quantity & is of course finite, continuous, and small when « is small (vis a constant), But the first derivative of & is of an order smaller than the order of £ when n is positive. When n is negative, the derivative of £ is of an order not so small as that of £; it may be finite, or it may even be Jarge and oscillate rapidly, when « is small. In all such instances, the expansion of the subject of integration in powers of «, after the small variation has been imposed, would not be effective to the end for which it was to be used. Again, it is conceivable that we could have small variations over a range continuous, not merely in are but also in direction, yet not continuous in curvature; an integrand of the second order would suffer a dis- continuity in passing through such a place, even if the place were isolated ; and difficulties would arise over the curvature properties of the contiguous varied curve. Similarly, if the subject of integration involves derivatives of order higher than the second, we should be faced with the possibility of abrupt changes in the rate of curvature. 209. An indication of one method, of taking some of such non-regular small variations into account, is given by the consideration of an integral 2. ae ‘ with a jagged variation, Let AMPRS... be a characteristic curve, on which A’ (the conjugate of A) lies beyond 8. Let any point P in this range AS be displaced to a neighbouring place Q. Because P lies within the range bounded by A and its conjugate A’, a consecutive characteristic AQT... can be drawn (§ 70) from A to pass through Q; and this consecutive curve does not meet AMPRS... until A’. Suppose that the conditions for a minimum are required. (The alter- native conditions for a maximum follow by changing, throughout the ex- planation, a greater inequality into a lesser inequality.) Accordingly, if the integral is to be a minimum along a characteristic curve AMP from A to P, and if we denote its value by J,» when taken along that curve, we must have Luxe Toe > Lanes210) ‘REQUIREMENT FOR CONSTITUENT JAGGED VARIATION 323, where (Lexa is the integral along ANQ from A to Q; (ii) Zgp is the integral along QP, the value of the independent variable of integration along QP increasing from Q up to P; and (iii) Typ is the integral along AMP from A to P. Now ANQT... equally is a characteristic curve, determined by slightly different initial conditions. Equally it provides a minimum: so that, taking AMPQ as a variation of ANQ, we have Laue + Inq > Lana where the integrals I4yp and I4yq are the same as before, and Iyq is the integral along PQ, the value of the independent variable of integration along PQ increasing from P up to Q (The integral Zpg is not the integral Top reversed: along PQ in Ipg, the independent variable of integration is increasing and likewise, along QP in Igp, it is increasing.) Next, consider ANQ and QR as a variation of the continuous length AMPR. Owing to the requirement of a minimum, we have Laxa+ Ton > Laure, where Igyq is the same as before, Igy is the integral from Q to R along QR, and I,ypx is the integral along the characteristic are AMPR. Thus Laur + Inq + Ton > Lava + Ton > Laser: On the left-hand side, the sum of the integrals is the whole integral, first along AMP from A to P, next along PQ from P to Q, and then along QR from @ to R. Thus the integral along AMPQR is greater than the integral along AMPR, Hence if the minimum condition is obeyed for a broken line AMPQ as a variation of a characteristic are AQ, and for a broken line ANQR as a variation of a characteristic are AR, it is obeyed for a line AMPQR, with a jagged portion PQR, as a variation of the characteristic are AMPR. It follows that the minimum condition will be secured for any jagged variation of a characteristic arc, composed of portions such as PQR which may be taken on either side of the are AMPR..., when a requirement is satisfied, that the minimum condition holds along any characteristic curve, separately for each broken curve, suck as ANQP for AMP, and such as ANQR for AMPR. In other words, the requirement is to be satisfied, whatever point P bbe chosen on a characteristic arc, and whatever small are be drawn from that chosen point to any contiguous point Q. 210. It should be noted that considerable freedom, thus far, is allowed to the form of the small ares, such as PQ and QR. No assumption has been made, tacitly or overtly, that they are straight lines: but we have assumed324 CONSTITUENT TYPE OF STRONG VARIATION [ow. vir (tacitly) that such an are PQ is continuous in arc-length, and that it does not eut the characteristic curve except at P. The assumption of continuity of are-length in the variation remains an assumption. ‘The other assumption is not essential: for, if PQ erossed the curve at points U, V,..., we should consider PQ resolved into portions PU, UV,..., each similar to PQR in the figure; and then we should treat each portion in the same way as PQR. We do not therefore need to modify the requirement of the minimum con- dition for a broken line such as AQP for AP, where P is any point on the line and PQ is any direction through P. Further, there need be no assumption as to continuity of direction, or of curvature, and so on, in the broken or jagged variation. ‘Thus consider acurve such as PQSTR. We can draw characteristics AQ, AS, AT, consecutive to APR. When the minimum condition holds for a single broken line such as AQP taken asa variation of AP, for every point on any characteristic curve and for every direction through the point, it holds generally. For, with the preceding notation, we have Laue + Tra > Lanes Taxa + Tos > Lass Lys +1 en> Lar Lap +Iyn> Lascen’ and therefore Laan Ira Los + Tor + Ten > Lowen Hence the minimum condition will hold for any accumulation of broken jagged small variations such as PQSTR, as it will hold for even the simplest variation such as the broken line AQP for AP, if the minimum condition be obeyed for each such simplest variation from every point P on the general characteristic curve to any contiguous point Q off the are. Note. As already stated, all the explanations and all the results hold as regards the conditions requisite for a possible maximum, provided the greater inequalities are changed into lesser inequalities. ‘Strong variation of ordinary integral with first derivatives. 211. Variations of the nature indicated are called strong variations. It is to be noted that they are only one type of the variations, distinct in their character from those which have hitherto been considered. The explanation just given shews that some of these strong variations can be compounded of ‘variations of the simple broken-line type, while each individual component of212) STRONG VARIATION OF INTEGRAL OF FIRST ORDER 325 the composite variation is itself a possible variant of the central curve. Further, when the condition for a minimum (or a maximum) holds at every place on that curve and for every direction at that place, the effect of the variation becomes cumulative. ‘The condition for a minimum (ora maximum) is satisfied for a quite general composite strong variation, if it is always satisfied for the quite general ultimate constituent of fundamental type. We therefore proceed to consider the analytical expression of this con- dition, initially individual for @ broken Jine, and scoure in its cumulative effect. As the introduction of strong variations, and the expre condition for the simplest and most frequent type of such variations, are due to Weierstrass, the analytical requirement for the fundamental constituent strong variation may fitly be called the Weierstrass test, It ean be obtained as follows for the simplest kind of integral. 212. Let a point Q, with coordinates «+ x£, y-+1, be the displaced Position of a point P on the central characteristic curve. The are PQ joining Pand Q is not restricted, save that it shall be continuous in the intrinsic qualities (such as tangency, curvature, rate of curvature, and so on) repre- sented by the derivatives of the dependent variable which occur in the subject of integration. If, and when, there should come a cessation of eon- tinuity in any of those properties, the are in question would be held to cease at such a point 7’; a new consecutive charactoristie A’ would be drawn through 7, and a new are would be considered to begin at 7. Thus within the limited range PQ the quantities £ and » will be regarded as regular fanctions of some parameter along that arc, valid only up to @. For variations beyond Q, new functions £ and », similarly regular along a new limited are, will be required, and so on: the test will be demanded in connection with each such elementary are. It is to bo noted that all such variations are of a restricted type; thus they do not cover variations such as xcos(nte~4), resin (nte”}), to take simple functions in terms of which, by Fourier series, some less simple variations can be represented. We shall deal, in the first instance, with the simplest kind of integral ‘ T= fF yaya already considered in Chapter II; and we shall assume that the same con- ditions, as before, attach to the integral. Also,all the tests, there found sufficient and necessary to secure a maximum or a minimum for weak variations, will be supposed to be already satisfied. We now have to consider the combination Taxa + Top — Tames326 ‘STRONG VARIATION OF [en. vir where AMP is the central characteristic; ANQ is the consecutive charac- teristic through Q a point contiguous to P, provided (§ 70—72) P is a point within a range AA’ bounded by the initial point A and its conjugate A’ g — x By definition, Zgp is the integral taken from Q to P along any curve joining Pand Q, continuous in are, direction, curvature and the like properties, but not meeting AMP save at P. Denoting the small weak variation from any point Mf on AMP to the cor- responding point Non AQ by xu and ev, and regarding the integral Luo as the variation of I4yp, we have (§ 41, 45), or ayy ? te [Few- smears [« (Ree +l where [1] denotes the aggregate of terms of the second and higher powers of small quantities. Now u=0, v=0,at A; and u=£,v=m, at P: also everywhere along the characteristic curve, the characteristic equation @ = 0 is satisfied. Hence, neglecting terms of the second and higher orders as com- pared with terms of the first order, we have cr) Tayo ~ Lane = « (ERE 213. The function F(z, y, 2, y:), in its initial form which involves 2, y, %, y, a8 unknown functions of f, is not a perfect differential with respect tot; and therefore the integral Igp, being [Pexewe, 7 cannot depend solely upon the values at P and at Q, but will depend partly upon the path from Q to P. ‘The simplest curve to select is the straight line joining Q to P. We shall therefore begin with the case when the are QP is a straight line. Let 4 and be the direction-cosines of the line QP in the direction from Q to P; and let U denote the length of PQ, so that eE=—M, eg =— pl ‘Also, let be any point on QP, and let QS=o, so that ¢ may range from 0 to Lalong the path QP from Q to P. The coordinates of 8 are a-dl-e) y-w(l-e).213) INTEGRAL OF FIRST ORDER 327 As the integral |F(e, y, %, :)d¢ is invariantive (§ 38), so far as concerns gral nH change of the independent variable ¢ of integration, we shall take «as the current independent variable of integration for Jp. AS NOW 2,=2, y,=1, along QP, we have Zap |Fle-N(t=0), y~n(l—o), ® ph de, while 1 is a small quantity of the same order of magnitude as x. The function F is regular over the range of integration which, moreover, is small. Hence, by the first theorem of mean value, there is a positive proper fraction ¢ such that Top = UF (a— nel, y ~ pel, , u) =IF (ay, % w)+[P), where [J*] denotes an aggregate of quantities of the second and higher orders in |, that is, in , Consequently, up to small quantities of the first order, we can take Top =1F (2, ys >, p). ‘Thus OF (@, Lara + Tor ~ Tasen= 1 {P(g uw) —92F accurate up to the first order. Now aF (ey, a my oy FE ya yaa FEMA yy Mess identically ; and therefore F@ydm=r hyde) Hace Tavat Tor Lane = UE (e, y, 5 Yrs wy where Ey Yu w= Fe y, dw) —V OE P(e. yu) _IF(e, y, ew) =a ee +f This quantity E(x, y, #., 1,2, w) depends, partly upon the position «, y on the characteristic curve and the direction «,, y, of the tangent to the characteristic curve at that position, and partly upon the direction 2, 4 of the short line PQ through that position: so that, if there is a variation off the curve, PQ must not coincide in direction with the tangent at P. Further, F(a, y, %1, yy) is homogeneous of the first degree in 2, and y,; and therefore PY YW) ang OF Ys Hs th en, ay, order zero. ‘They are unchanged in value when «, and y, are multiplied are homogeneous in those quantities of328 WEIERSTRASS TEST fox. vin by the same quantity; hence, when we multiply them by : so that they vecune istic eurve, the derivatives become 22, P(e, yay’) and é F(ey,2,y). We now have EO 2 JW =F BAZ PH 8 HP H HY) ae ay (=a) and w (=y), where s is the are-length along the character- of reyan renin te remw-g reser} The Weierstrass E-function, and Weierstrass test. 214. The integral J is to be a maximum or minimum for all small variations. The conditions for weak variations have been obtained and are satisfied. If the integral is to be a minimum under the strong variations composed of variations such as ANQP for AMP, then E(@, Yo YB) must be positive, for all positions «, y on the characteristic curve, and for all directions 2, through every position giving a variation, that is, for all directions off the tangent ; while, if the integral is to be a maximum, the same function must be negative for all positions 2, y on the characteristic curve, and for all non-tangential directions through every position. ‘The function 14 measures the excess over @ minimum or the deficiency from a maximum. Conversely, if Eis always positive, an excess always arises, so that there isa minimum ; while, if Z is always negative, a deficiency always arises, so that there is a maximum. ‘The function is often called the excess- function, though it is a deficiency when a maximum exists. As it is due to Weierstrass, it will be called the Weierstrass E-function. The test—that it is to be always positive for a minimum, and always negative for a maximum, along the characteristic curve—will be called the Weierstrass test. Manifestly, it is distinct in character, as in source, from the Buler test, the Legendre test, and the Jacobi test. The requirement is that the quantity US Ferm wrens nh i +H 1S Few mn BP ewe, av} for all places on the characteristic curve, and for every direction X and pw through each place distinct from the tangential direction, shall have the same sign: the sign being positive for a minimum, and negative for a maximum. Tt isa necessary test. It is a sufficient test, for the type of strong varia- tion imposed. But that type of strong variation is only a selected type.215) EXAMPLES OF THE WEIERSTRASS TEST 329 Eaamples of the E-function as a test. 215. The Weierstrass test is simple, in its application to all integrals which are of the form [ (2, y) de or, what is the equivalent, ate of the form [ren ers ynk ae For such integrals, the H-function is Xn a fread a fons at jee ava telfen shay sen wel which is equal to Fey) (L-Qa' + py), because (A + w)t=1, (w+ y%)#=1. If x be the inclination of QP to the tangent to the curve, so that x lies between 0 and 2m for any direction that is non-tangential, we have E=f(@, y) (1 — cos x). If f (2, y) does not change its sign along the characteristic curve, then— as the sign of f(x, y) may be taken positive—we have H always positive. ‘Therefore a minimum is admissible, and does occur, so far as concerns the requirement of strong variations of the selected type. Es, 1. ‘Thus the testis satisfied for a catenoid of revolution. “The integral ix ae fae; and 20, a8 the tests are satisfied—with a limited rango—for weak vati we still have a minimum, Ex. 2. Similarly, the testis satisfied for a gravity brachistookrone, ‘The integral effect ively (§ 34, Ex. 3) is ions (§ 30, Ex. 1), fo ta) bas, ‘where the positive sign is given to the radical. ‘The Weierstrass test, established (as above) for the case when PQ is a straight line, can be extended to all integrals [rends=[reyerrynta, when P and Q are joined by any are (and not alone by a straight line). As before, we have iF Taso Lan « (Er +030) =f ey) Ey Grey ot Fe») Oat + wy’) If (@ y) 008 x,330 WEIERSTRASS TEST SATISFIED BY GEODESICS [ou vir where 1 is the length of the straight line PQ. For Jap, we have Top= [f%, Ya, where X, Y are the coordinates of any point on the are PQ as now drawn. Hence, if the total length of the arc be L, we have Ter= If, P), where f(X, Y) represents a mean value of f(X, Y) along the are, while X differs from 2, and Y differs from y, by quantities of the same order as Z at least. Thus FE, Y=f@ y+ (1, where L represents the aggregate of the terms in Z, I’, ...; and so Tap = Lf (@, y) + L(L} =f (2,9) +(e], that is, to the first order of small quantities, = If sy). Hence, to the first order of small quantities, Lana + Toe - Laue =(L — boos x) f(a, y). Now L is always greater than J, the rectilinear distanee between P and Q; and therefore L — cos x is positive for all values of x. Consequently Taxa Tor — Tours for the integrals in question, is always positive, whatever form the continuous are PQ may take, Hence the Weierstrass test, with the less specialised type of aro, is satisfied for integrals [/(e, y) ds, thus admitting a minimum. Ex. 3. The Weierstrass testis satisfied for all geodesics, With the customary notation of differential geometry*, and when p, gare the parameters, on a surface, any are on a surface is freees28 pat ogres and the geodesic quality is determined by the charactaristicequationt. For the eonstruction of the Weierstms test, we take a point P given by p, q,and a neighbouring point @ given by p-ted, q-+07; and then, asin the text, we have Lam Taree {0 BB py Bet (Bpe+2Fpg+ Gq" Ept+2Fp.g + Ga pt we {e(0%f +4) +9(rlE+0%)h. Now if 4 %2e th spcecortintas Pandy Fy th of X-Xametont(e} Ni- Yenttyntl’) A-Zeattant(e]s * See my Lectures on Diferentiat Geometry, cbep. i. + 1b. chap. v.215] EXAMPLES 381 snd therefore . Pa E+ 2Rby + Gr, ee dr_, dp wy y iq Goagend. E wih goaded Thos ann n$e-o%. Beer (647 2) + 0nd. But the left-hand side =leos yy hor isthe length PQ, and ¥ isthe angle betweon PQ and the tangent tothe character. istic ourve, If then, as before, x ix the angle between @P and this tangent, x=r 4; — Lay Lane —1 008 y. 1 therefore the Weierstrass fancton is 1(1~co8y), ‘hich in always positive, ‘The Weierstrass tot is satis. Also Jpp=U5 Ez. 4. Discuss the possible minimum of the integral Se taken between two given values of ‘The characteristic equation is 4 (ay zd ® =0, 0 that the primitive is yaartb, where a and & are constants determined by the end-points (#,, 91) and (¢,, 93). ‘The Legendre tost gives 54, =2; itn ants fora minimum. As the characteristic curve is a straight line, there is no finite conjugate of any initial point ; 0 the Jacobi teat i satisfied unconditionally. (@) When the integral is taken in the Weierstrass form, it is Either of the (equivalent) characteristic equations leads to yrartb, For the Legendre test, we take eo ap yt a” which is positive if x i positive, There is no conjugate of a value &; for if sao, ymacttb, With the notation of §§ 61, 63, we have HO=0, HO=0 W=ch Wi=1 smo ¥ Omac, x=c e@=4 and the function Z(t, f9)=68(¢—%), which has no zero except f. Thus for general weak variations, the integral admits a minimum, and 20332 EXAMPLES NOT SATISFYING (ou. vi But the subject of integration is rational (and so the Weierstrass test cannot be satisfied : soo § 217). The Z-function is =( (-Be)er( (2f-24) aero, ‘and so changes sign with 2. ‘The straight line does not providea true minimum, that is, « minimum also under strong variations, (ii) We here have taken the change of sign of X as dominating the Z-funetion; and it is sufficient, for our purpose, to determine whether H can change its sign. In this case, other curves, not regular in direction but jagged, can easily be constructed which lead to values smaller than the minimum under weak: variations. Let AB be the straight line; PQ any element. Take PZ’ parallel to the axis of x, where P7’ may be large compared with PQ, if we please; and let R7Q=«, RPO: . When for the portion PQ of AB, we sub- é state the boken line PZT, along PO we have — a Wao and along 79 we have tan e; thus the value of the intogral for P77¢ 0, for PT; + TRtan® «, for 7Q; =TRtante - QRtane =detanatane. ‘The value of the element of the integral for PQ along AB is detente; 8 ¢ ‘Thus both the Legendre test and the Jacobi test admit the possibility of a minimum (necessarily for ‘weak variations alone, because only weak variations have been imposed) if , beginning at 1 48 value go, does not attain a value % (v) It therefore is desirable to trace the curve, We have ae Gr pora-w, Po -Za+na-wy; ‘and therefore a cusp (or some higher singularity) may be expected at g=—1.. We take 1 Inte where » is small in the vicinity of the place. We find, for the values of and y in that vicinity, -e 2H (te wetfiee..), ‘showing that the place C is a cusp, the tangential direction at which is given by an i Ya 1. The two branches of the cure, op sitber |g Sis af Un cup ar sven: C4 by a postive vaun of . tad Oe ys aan Re gle OO sear : tng 04, ¢ wich =-{f) incre, 2 tat gs © iy ese ae ion ee a yowbotetblons) +e (set + wt ae 7+ 1ep) 89h fa hence _ Along C4, ¢ is positive and g, is positive; while along CB, q ia positive and g, is negative, ‘Thus along C4, the quantity P is positive ; the Legendre test admits a rainimum for the branch CA. And clearly ¢ is a negative constant. ‘Thus, for weak variations, the characteristic curve is G04 y=wtalogg- a1,344 SOLID OF REVOLUTION (ox. vir on changing the negative ¢ into a positive a. ‘The Legendre test is satisfied. ‘The Jacobi test imposes no limit on a range, that begins anywhere on C/A and extends in the direction Ctowards 4. Hence we have a minimut for weak variations, (vi) But the Weierstrass test, for the simplest strong variation, was proved to be not satisfied. What is a minimum, under weak variations, is not a minimum under the simplest strong variation, ‘Therefore it is not a true minimum. (Vii) Te was stated (§ 207) that Legendre obtained, by means of jagged lines a curve (for the generation of the solid of revolution) which leads to an integral less than the quasi-minimum provided by the characteristic curve for weak variations. ‘The following is « simplified form of Legendre’s construction* for the purpose. Let PQ be any arc of the characteristic curve. Let QT be ‘the tangent at the lower extremity; through P, draw PC . ‘making with the axis of « the angle y equal to 70 ; and through @ draw the ordinate VQC. Along the arc PQ, we have pi [i pia Along the line PC, we have Tee ee re [ay Along the line CQ, we have p==, #,=0: thus Toa Lrot Tog da ” yn Se) Rov ty, so that pre” + pp'x" +a’ =0, pty” + pp'y +y =0, equations that are useful for the expansions of a and y’ (and therefore of # and y) in terms of s. As the curve is arbitrarily assumed, we can imagine it defined by means of the curvature, and of initial values of a’ and y’ at P; and so we take a am, fad, 2 =—%, y c at P. We proceed to make 6 the variable in the expansion; and we make the small quantity in terms of the magnitude of which all the small quantities are expressed. We take, at any point, p=oR,220) SMALL CURVED ARC 349 0 that R, being a function of S, is a function of 28. Now, along PTQ, a dp @X aX Pas +P as ast ds ax Rik @X aX ae te apt ie ‘To express X as a series in powers of @ satisfying this equation, we have, when @ is zero, and therefore R ax_, ay. 7% 95 @X__b VY _ ay, aan eo” or if c=0l, 0 that, for the radius of curvature, R= C+ OG +44 ..., the initial values are a 4, OX an, Gaon, Oe ‘The expansions take the form aX a G77 (ger ee...), ay ab B77 (M+ G+ HOt), and therefore, as X and Y vanish when @=0, Kno (96-7048 0+...) = 08,, .)=28, the values of the coefficients a,, b,,... being definite: they are linear in @, and b,, and they involve the coefficients in R. be oy Va o(b0+ S64 Also at Q, we have x€ and «7 as the values of X and Y, while @ there is unity; thus ebmo(u— pat wpm a(bs tHe be. )=om, using J and m to denote the two expressions. Further, if 8’=Q' so that S” is measured from Q towards P, we have S + 8’ =o, and so ax__ ax a ay __ de, aS" dd d6’ dS dd a’350 ‘MODIFIED WEIERSTRASS FUNCTION FOR fon. vn ‘so that, . . an (aa) + Ga)» Pe L ay 7 Modified E-function. 221, With these values, we have ; Tep=| F(a+X,y+Y, wo [Florey anof rer nyt oe af) 28 nef F(x +0, y+0®,, —L, —M)d0 wef Fle y—L,— Mao + [ot} where [o!] is an aggregate of quantities of the second and higher orders in @. Now a, and 6 denote any direction at P; for that direction, substitute the opposite direction — a, and -b,. Then —L and —M become L and 2f, while 1 and m become —1 and —m, all four expressions being linear in a, and b,. Hence - i f FoyL, a doo (12 m3). In the definite integral, the only quantities involving @ are L, M, quantities connected with the are PQ; let 6 be a mean value of 6 (with Zand if as the corresponding values of L and Mf) such that 0<@<1, and [Fey L, Mdo=F yD, ih. Also F(a,y, LE, =L $F yD, M+ HS eh oyl M Let E’ denote the 3 of integrals, s0 — E'= Lag bene — Lars then Fey lL M-12 Fe ye v9} { aL soft 5 rey Tiy-m Zreane, v} In this expression, Z and Mf are the direction-cosines of a tangent to Q7P at the point having the mean value 4; also lo = doos (x+y), mo=dsin (y+ ¥),221) STRONG VARIATION WITH CURVED ARC 351 where d is the length of the line PQ, y is the inclination of the tangent of the characteristic curve to the axis of , and x is the angle between PQ and this tangent, The function E' must be positive at all points of the curve for all values of and m (and so of Z and M) if there is to bea ‘minimum ; it must be similarly negative if there is to be a maximum. Bes.1, Varify that Bis positive (40 that « minimum is admnisible) for al intageas of the type [re ncrayoh a, Ez. 2, The arc PT@ is taken to be circular, an arbitrary radius ¢ being chosen, as in § 222 (post). Provo thatthe vatiation ofthe intogral can he expressed in the form D sin go (OF OP) ar. ar. DG R HG) PGonoesyand) where ¢ isthe (arbitrary in F(e, 9, XB) X= cos (p+e), =! Exp. ation of the chord QP to the axis of x, where F denotes sin (+e), and where ¢ is a mean value of x such that iP 2F (2, 95%, sin 2 — ge ‘pF (2, 9, 008 (+) sin ($+x)} dy. Zs 3: Consider the caso when the arc PQ is parabolio, point (X, Y) on it being given by He, Youte2 Xan-$e, vans de At P, lett=05 at Q,let t= 7. Then we take the straight line QP=D, wi direction-cosines J, m: so that xf= WD ee 7-£ 7%, sqemDap Pe 7, Thus + # » ie ae a-£7 ‘and therefore dma am Let 2, m=cos $, sin 80 that = @PH, a=A'PY; then, if xe gran QPs’, oh Also, if P/Q is a right angle, 0 that Pif= 7,