A treatise on universal algebra, with applicationsFront MatterProduction Note Cornell University Library pro- duced this volume to replace the irreparably deteriorated original. It was scanned using Xerox soft- ware and equipment at 600 dots per inch resolution and com- pressed prior to storage using CCITT Group 4 compression. The digital data were used to create Cornell's replacement volume on paper that meets the ANSI Stand- ard Z39.48-1984. The production of this volume was supported in part by the Commission on Pres- ervation and Access and the Xerox Corporation. 1991.
A TREATISE ON UNIVERSAL ALGEBRA.
C. J. CLAY and SONS, CAMBRIDGE UNIVERSITY PRESS WAREHOUSE, AVE MARIA LANE. : 263, ARGYLE STREET. !Uip tfl: F. A. BROCKHAUS. gorfc: THE MACMILLAN COMPANY. fcombag: E. SEYMOUR HALE.
Title PageA TBEATISE ON UNIVEESAL ALGEBEA WITH APPLICATIONS. BY ALFRED NORTH WHITEHEAD, M.A. FELLOW AND LECTURER OP TRINITY COLLEGE, CAMBRIDGE. VOLUME I. CAMBRIDGE: AT THE UNIVERSITY PRESS. 1898 [All Rights reserved.]
Cambridge: PRINTED BY J. AND C. F. CLAY, AT THE UNIVERSITY PRESS.
Front MatterPBEFACE. XT is the purpose of this work to present a thorough investigation of the various systems of Symbolic Reasoning allied to ordinary Algebra. The chief examples of such systems are Hamilton's Quaternions, Grassmann's Calculus of Extension, and Boole's Symbolic Logic. Such algebras have an intrinsic value for separate detailed study; also they are worthy of a comparative study, for the sake of the light thereby thrown on the general theory of symbolic reasoning, and on algebraic symbolism in particular. The comparative study necessarily presupposes some previous separate study, comparison being impossible without knowledge. Accordingly after the general principles of the whole subject have been discussed in Book I. of this volume, the remaining books of the volume are devoted to the separate study of the Algebra of Symbolic Logic, and of Grassmann's Calculus of Extension, and of the ideas involved in them. The idea of a generalized conception of space has been made prominent, in the belief that the properties and operations involved in it can be made to form a uniform method of interpretation of the various algebras. Thus it is hoped in this work to exhibit the algebras both as systems of symbolism, and also as engines for the investigation of the possibilities of thought and reasoning connected with the abstract general idea of space. A natural mode of comparison between the algebras is thus at once provided by the unity of the subject-matters of their interpretation. The detailed comparison of their symbolic structures has been adjourned to the second volume, in which it is intended to deal with Quaternions, Matrices, and the general theory of Linear Algebras. This comparative anatomy of the subject was originated by B. Peirce's paper on Linear Associative Algebra*, and has been carried forward by more recent investigations in Germany. * First read before the National Academy of Sciences in Washington, 1871, and republished in the American Journal of Mathematics, vol. iv., 1881.
VI PREFACE. The general name to be given to the subject has caused me much thought: that finally adopted, Universal Algebra, has been used somewhat in this signification by Sylvester in a paper, Lectures on the P?*inciples of Universal Algebra, published in the American Journal of Mathematics, vol. vi., 1884. This paper however, apart from the suggestiveness of its title, deals ex- plicitly only with matrices. Universal Algebra has been looked on with some suspicion by many mathematicians, as being without intrinsic mathematical interest and as being comparatively useless as an engine of investigation. Indeed in this respect Symbolic Logic has been peculiarly unfortunate; for it has been disowned by many logicians on the plea that its interest is mathematical, and by many mathematicians on the plea that its interest is logical. Into the quarrels of logicians I shall not be rash enough to enter. Also the nature of the interest which any individual mathematician may feel in some branch of his subject is not a matter capable of abstract argumentation. But it may be shown, I think, that Universal Algebra has the same claim to be a serious subject of mathematical study as any other branch of mathematics. In order to substantiate this claim for the importance of Universal Algebra, it is necessary to dwell shortly upon the fundamental nature of Mathematics. Mathematics in its widest signification is the development of all types of formal, necessary, deductive reasoning. The reasoning is formal in the sense that the meaning of propositions forms no part of the investigation. The sole concern of mathematics is the inference of proposition from proposition. The justification of the rules of inference in any branch of mathematics is not properly part of mathematics: it is the business of experience or of philosophy. The business of mathematics is simply to follow the rule. In this sense all mathematical reasoning is necessary, namely, it has followed the rule. Mathematical reasoning is deductive in the sense that it is based upon definitions which, as far as the validity of the reasoning is concerned (apart from any existential import), need only the test of self-consistency. Thus no external verification of definitions is required in mathematics, as long as it is considered merely as mathematics. The subject-matter is not necessarily first presented to the mind by definitions: but no idea, which has not been completely defined as far as concerns its relations to other ideas involved in the subject-matter, can be admitted into the reasoning. Mathematical definitions are always to be construed as limitations as well as definitions;
PREFACE. Vll namely, the properties of the thing defined are to be considered for the purposes of the argument as being merely those involved in the definitions. Mathematical definitions either possess an existential import or are conventional. A mathematical definition with an existential import is the result of an act of pure abstraction. Such definitions are the starting points of applied mathematical sciences; and in so far as they are given this existential import, they require for verification more than the mere test of self-consistency. Hence a branch of applied mathematics, in so far as it is applied, is not merely deductive, unless in some sense the definitions are held to be guaranteed a priori as being true in addition to being self-consistent. A conventional mathematical definition has no existential import. It sets before the mind by an act of imagination a set of things with fully defined self-consistent types of relation. In order that a mathematical science of any importance may be founded upon conventional definitions, the entities created by them must have properties which bear some affinity to the properties of existing things. Thus the distinction between a mathematical definition with an existential import and a conventional definition is not always very obvious from the form in which they are stated. Though it is possible to make a definition in form unmistakably either conventional or existential, there is often no gain in so doing. In such a case the definitions and resulting propositions can be construed either as referring to a world of ideas created by convention, or as referring exactly or approximately to the world of existing things. The existential import of a mathematical definition attaches to it, if at all, qua mixed mathematics; qua pure mathematics, mathematical defi- nitions must be conventional*. Historically, mathematics has, till recently, been confined to the theories of Number, of Quantity (strictly so-called), and of the Space of common experience. The limitation was practically justified : for no other large systems of deductive reasoning were in existence, which satisfied our definition of mathematics. The introduction of the complex quantity of ordinary algebra, an entity which is evidently based upon conventional definitions, gave rise to the wider mathematical science of to-day. The realization of wider conceptions has been retarded by the habit of mathe- maticians, eminently useful and indeed necessary for its own purposes, of extending all names to apply to new ideas as they arise. Thus the name * Cf. Grassmann, Ausdehmmgslehre von 1844, Einleitung.
Vlll PREFACE. of quantity was transferred from the quantity, strictly so called, to the generalized entity of ordinary algebra, created by conventional definition, which only includes quantity (in the strict sense) as a special case. Ordinary algebra in its modern developments is studied as being a large body of propositions, inter-related by deductive reasoning, and based upon conventional definitions which are generalizations of fundamental conceptions. Thus a science is gradually being created, which by reason of its fundamental character has relation to almost every event, phenomenal or intellectual, which can occur. But these reasons for the study of ordinary Algebra apply to the study of Universal Algebra; provided that the newly invented algebras can be shown either to exemplify in their symbolism, or to represent in their interpretation interesting generalizations of important systems of ideas, and to be useful engines of investigation. Such algebras are mathematical sciences, which are not essentially concerned with number or quantity; and this bold extension beyond the traditional domain of pure quantity forms their peculiar interest. The ideal of mathematics should be to erect a calculus to facilitate reasoning in connection with every province of thought, or of external experience, in which the succession of thoughts, or of events can be definitely ascertained and precisely stated. So that all serious thought which is not philosophy, or inductive reasoning, or imaginative literature, shall be mathematics developed by means of a calculus. It is the object of the present work to exhibit the new algebras, in their detail, as being useful engines for the deduction of propositions; and in their several subordination to dominant ideas, as being representative symbolisms of fundamental conceptions. In conformity with this latter object I have not hesitated to compress, or even to omit, developments and applications which are not allied to the dominant interpretation of any algebra. Thus unity of idea, rather than completeness, is the ideal of this book. I am convinced that the comparative neglect of this subject during the last forty years is partially due to the lack of unity of idea in its presentation. The neglect of the subject is also, I think, partially due to another defect in its presentation, which (for the want of a better word) I will call the lack of independence with which it has been conceived. I will proceed to explain my meaning. Every method of research creates its own applications: thus Analytical Geometry is a different science from Synthetic Geometry, and both these sciences are different from modern Projective Geometry. Many propositions
PREFACE. IX are identical in all three sciences, and the general subject-matter, Space, is the same throughout. But it would be a serious mistake in the development of one of the three merely to take a list of the propositions as they occur in the others, and to endeavour to prove them by the methods of the one in hand. Some propositions could only be proved with great difficulty, some could hardly even be stated in the technical language, or symbolism, of the special branch. The same applies to the applications of the algebras in this book. Thus Grassmann's Algebra, the Calculus of Extension, is applied to Descriptive Geometry, Line Geometry, and Metrical Geometry, both non- Euclidean and Euclidean. But these sciences, as here developed, are not the same sciences as developed by other methods, though they apply to the same general subject-matter. Their combination here forms one new and distinct science, as distinct from the other sciences, whose general subject- matters they deal with, as is Analytical Geometry from Pure Geometry. This distinction, or independence, of the application of any new algebra appears to me to have been insufficiently realized, with the result that the developments of the new Algebras have been cramped. In the use of symbolism I have endeavoured to be very conservative. Strange symbols are apt to be rather an encumbrance than an aid to thought: accordingly I have not ventured to disturb any well-established notation. On the other hand I have not hesitated to introduce fresh symbols when they were required in order to express new ideas. This volume is divided into seven books. In Book I. the general prin- ciples of the whole subject are considered. Book II. is devoted to the Algebra of Symbolic Logic; the results of this book are not required in any of the succeeding books of this volume. Book III. is devoted to the general principles of addition and to the theory of a Positional manifold, which is a generalized conception of Space of any number of dimensions without the introduction of the idea of distance. The comprehension of this book is essential in reading the succeeding books. Book IV. is devoted to the principles of the Calculus of Extension. Book V. applies the Calculus of Extension to the theory of forces in a Positional manifold of three dimensions. Book VI. applies the Calculus of Extension to Non-Euclidean Geometry, considered, after Cayley, as being the most general theory of distance in a Positional manifold; the comprehension of this book is not necessary in reading the succeeding book. Book VII. applies the Calculus of Extension to ordinary Euclidean Space of three dimensions.
X PREFACE. It would have been impossible within reasonable limits of time to have made an exhaustive study of the many subjects, logical and mathematical, on which this volume touches; and, though the writing of this volume has been continued amidst other avocations since the year 1890, I cannot pretend to have done so. In the subject of pure Logic I am chiefly indebted to Mill, Jevons, Lotze, and Bradley; and in regard to Symbolic Logic to Boole, Schroder and Venn. Also I have not been able in the footnotes to this volume adequately to recognize my obligations to De Morgan's writings, both logical and mathematical. The subject-matter of this volume is not concerned with Quaternions; accordingly it is the more necessary to mention in this preface that Hamilton must be regarded as a founder of the science of Universal Algebra. He and De Morgan (cf. note, p. 131) were the first to express quite clearly the general possibilities of algebraic symbolism. The greatness of my obligations in this volume to Grassmann will be understood by those who have mastered his two Ausdehnungslehres. The technical development of the subject is inspired chiefly by his work of 1862, but the underlying ideas follow the work of 1844. At the same time I have tried to extend his Calculus of Extension both in its technique and in its ideas. But this work does not profess to be a complete interpretation of Grassmann's investigations, and there is much valuable matter in his Ausdehnungslehres which it has not fallen within my province to touch upon. Other obligations, as far as I am aware of them, are mentioned as they occur. But the book is the product of a long preparatory period of thought and miscellaneous reading, and it was only gradually that the subject in its full extent shaped itself in my mind; since then the various parts of this volume have been systematically deduced, according to the methods appropriate to them here, with hardly any aid from other works. This procedure was necessary, if any unity of idea was to be preserved, owing to the bewildering variety of methods and points of view adopted by writers on the various subjects of this volume. Accordingly there is a possibility of some oversights, which I should very much regret, in the attribution of ideas and methods to their sources. I should like in this connection to mention the names of Arthur Buchheim and of Homersham Cox as the mathematicians whose writings have chiefly aided me in the development of the Calculus of Extension (cf. notes, pp. 248, 346, 370, and 575). In the development of Non-Euclidean Geometry the ideas of Cayley, Klein, and Clifford have been
PREFACE. XI chiefly followed; and in the development of the theory of Systems of Forces I am indebted to Sir R. S. Ball, and to Lindemann. I have added unsystematically notes to a few theorems or methods, stating that they are, as far as I know, now enunciated for the first time. These notes are unsystematic in the double sense that I have not made a systematic search in the large literatures of the many branches of mathematics with which this book has to do, and that I have not added notes to every theorem or method which happens to be new to me. My warmest thanks for their aid in the final revision of this volume are due to Mr Arthur Berry, Fellow of King's College, to Mr W. E. Johnson, of King's College, and Lecturer to the University in Moral Science, to Prof. Forsyth, Sadlerian Professor to the University, who read the first three books in manuscript, and to the Hon. B. Russell, Fellow of Trinity College, who has read many of the proofs, especially in the parts connected with Non-Euclidean Geometry. Mr Johnson not only read the proofs of the first three books, and made many important suggestions and corrections, but also generously placed at my disposal some work of his own on Symbolic Logic, which will be found duly incorporated with acknowledgements. Mr Berry throughout the printing of this volume has spared himself no trouble in aiding me with criticisms and suggestions. He undertook the extremely laborious task of correcting all the proofs in detail. Every page has been improved either substantially or in expression owing to his suggestions. I cannot express too strongly my obligations to him both for his general and detailed criticism. The high efficiency of the University Press in all that concerns mathe- matical printing, and the courtesy which I have received from its Officials, also deserve grateful acknowledgements. Cambridge, December, 1897. W.
Table of ContentsCONTENTS. The following Books and Chapters are not essential for the comprehension of the subsequent pa?*ts of this volume: Book II, Chapter V of Book IV, Book VI. BOOK I. PRINCIPLES OF ALGEBRAIC SYMBOLISM. CHAPTER I. On the Nature of a Calculus. ART. PAGES 1. Signs............. 3 4 2. Definition of a Calculus.......... 4 5 3. Equivalence............ 5 7 4. Operations............ 7 8 5. Substitutive Schemes.......... 8 9 6. Conventional Schemes.......... 9 10 7. Uninterpretable Forms.......... 10 12 CHAPTER II. Manifolds. 8. Manifolds............ 13 14 9. Secondary Properties of Elements........ 14 15 10. Definitions............ 15 11. Special Manifolds........... 16 17 62
xiv CONTENTS. CHAPTER III. Principles of Universal Algebra. ART. PAGES 12. Introductory............ 18 13. Equivalence............ 18 19 14. Principles of Addition.......... 19 21 15. Addition............. 21 22 16. Principles of Subtraction......... 22 24 17. The Null Element........... 24 25 18. Steps............. 25 19. Multiplication............ 25 27 20. Orders of Algebraic Manifolds........ 27 28 21. The Null Element........... 28 29 22. Classification of Special Algebras........ 29 32 Note............. 32 BOOK II. THE ALGEBRA OF SYMBOLIC LOGIC. CHAPTER I. The Algebra of Symbolic Logic. 23. Formal Laws............ 35 37 24. Reciprocity between Addition and Multiplication .... 37 38 25. Interpretation............ 38 39 26. Elementary Propositions......... 39 41 27. Classification............ 41 42 28. Incident Regions........... 42 44 CHAPTER II. The Algebra of Symbolic Logic {continued). 29. Development............ 45 47 30. Elimination............ 47 55 31. Solution of Equations with One Unknown...... 55 59 32. On Limiting and Unlimiting Equations...... 59 60 33. On the Fields of Expressions........ 60 65 34. Solution of Equations with More than One Unknown .... 65 67 35. Symmetrical Solution of Equations with Two Unknowns . . . 67 73 36. Johnson's Method........... 73 75 37. Symmetrical Solution of Equations with Three Unknowns . . . 75 80 38. Subtraction and Division......... 80 82
CONTENTS. XV CHAPTER III. Existential Expressions. ART. PAGES 39. Existential Expressions.......... 83 86 40. Umbral Letters........... 86 89 41. Elimination............ 89 91 42. Solutions of Existential Expressions with One Unknown . . . 91 92 43. Existential Expressions with Two Unknowns..... 93 94 44. Equations and Existential Expressions with One Unknown. . . 94 96 45. Boole's General Problem.......... 96 97 46. Equations and Existential Propositions with Many Unknowns . . 97 98 Note............. 98 CHAPTER IV. Application to Logic. 47. Propositions............ 99 100 48. Exclusion of Nugatory Forms........ 100 101 49. Syllogism............. 101 103 50. Symbolic Equivalents of Syllogisms....... 103 105 51. Generalization of Logic.......... 105 106 CHAPTER V. Propositional Interpretation. 52. Propositional Interpretation......... 107 108 53. Equivalent Propositions.......... 108 54. Symbolic Representation of Complexes....... 108 55. Identification with the Algebra of Symbolic Logic .... 108 111 56. Existential Expressions.......... Ill 57. Symbolism of the Traditional Propositions...... Ill 112 58. Primitive Predication.......... 112 113 59. Existential Symbols and Primitive Predication..... 113 114 60. Propositions............ 114 115 Historical Note........... 115 116
XVI CONTENTS. BOOK III. POSITIONAL MANIFOLDS. CHAPTER I. Fundamental Propositions. AET. PAGES 61. Introductory............ 119 62. Intensity............. 119 121 63. Things representing Different Elements....... 121 122 64. Fundamental Propositions......... 122 125 65. Subregions............ 125 128 66. Loci............. 128 130 67. Surface Loci and Curve Loci........ 130 131 Note............. 131 CHAPTER II. Straight Lines and Planes. 68. Introductory............ 132 69. Anharmonic Ratio........... 132 70. Homographic Ranges.......... 133 71. Linear Transformations . . ........ 133 135 72. Elementary Properties.......... 136 137 73. Reference-Figures........... 138 -139 74. Perspective............ 139 142 75. Quadrangles............ 142 143 CHAPTER III. QUADRICS. 76. Introductory............ 114 77. Elementary Properties.......... 144 145 78. Poles and Polars....... 145 147 79. Generating Regions.......... 147 148 80. Conjugate Coordinates.......... 148 151 81. Quadriquadric Curve Loci......... 151 153 82. Closed Quadrics........... 153 155 83. Conical Quadric Surfaces.......... 155 157 84. Reciprocal Equations and Conical quadrics...... 157 161 Note............. 161
CONTENTS. xvii CHAPTER IV. Intensity. ^RT. PAGES 85. Defining Equation of Intensity........ 162__163 86. Locus of Zero Intensity.......... 163 164 87. Plane Locus of Zero Intensity........ 164 166 88. Quadric Locus of Zero Intensity........ 166 89. Antipodal Elements and Opposite Intensities..... 166 167 90. The Intercept between Two Elements....... 167 168 Note............. 168 BOOK IV. CALCULUS OF EXTENSION. CHAPTER I. Combinatorial Multiplication. 91. Introductory............ 171 172 92. Invariant Equations of Condition........ 172 173 93. Principles of Combinatorial Multiplication...... 173 175 94. Derived Manifolds........... 175 176 95. Extensive Magnitudes.......... 176 177 96. Simple and Compound Extensive Magnitudes..... 177 178 97. Fundamental Propositions......... 178 180 Note............. 180 CHAPTER II. Regressive Multiplication. 98. Progressive and Regressive Multiplication...... 181 99. Supplements............ 181 183 100. Definition of Regressive Multiplication....... 183 184 101. Pure and Mixed Products......... 184-185 102. Rule of the Middle Factor......... 185 188 103. Extended Rule of the Middle Factor....... 188 190 104. Regressive Multiplication independent of Reference-Elements . . 190 191 105. Proposition............ 191 106. Mailer's Theorems........... 192 195 107. Applications and Examples......... 195 198 Note............. 198
xviii CONTENTS. CHAPTER III. Supplements. ART. PAGES 108. Supplementary Regions.......... 199 109. Normal Systems of Points......... 199 200 110. Extension of the Definition of Supplements...... 201 202 111. Different kinds of Supplements........ 202 112. Normal Points and Straight Lines...... . . 202 203 113. Mutually normal Regions......... 203 204 114. Self-normal Elements.......... 204 206 115. Self-normal Planes........... 206 116. Complete Region of Three Dimensions....... 206 207 117. Inner Multiplication.......... 207 118. Elementary Transformations......... 208 119. Rule of the Middle Factor......... 208 120. Important Formula........... 208 209 121. Inner Multiplication of Normal Regions...... 209 122. General Formula for Inner Multiplication...... 209 210 123. Quadrics............. 210 212 124. Plane-Equation of a Quadric......... 212 213 CHAPTER IV. Descriptive Geometry. 125. Application to Descriptive Geometry....... 214 126. Explanation of Procedure . ....... 214 215 127. Illustration of Method.......... 215 128. von Staudt's Construction . . ....... 215 219 129. Grassmann's Constructions......... 219 223 130. Projection............. 224 228 CHAPTER V. Descriptive Geometry of Conics and Cubics. 131. General Equation of a Conic......... 229 231 132. Further Transformations.......... 231 233 133. Linear Construction of Cubics........ 233 134. First Type of Linear Construction of the Cubic..... 233 235 135. Linear Construction of Cubic through Nine arbitrary Points . . 235 237 136. Second Type of Linear Construction of the Cubic .... 238 239 137. Third Type of Linear Construction of the Cubic..... 239 244 138. Fourth Type of Linear Construction of the Cubic .... 244 246 139. Chasles' Construction.......... 246 247
CONTENTS. XIX CHAPTER VI. Matrices. ART. PAGES 140. Introductory............ 248 141. Definition of a Matrix.......... 248 249 142. Sums and Products of Matrices........ 250 252 143. Associated Determinant.......... 252 144. Null Spaces of Matrices.......... 252 254 145. Latent Points............ 254 255 146. Semi-Latent Regions.......... 256 147. The Identical Equation.......... 256-257 148. The Latent Region of a Repeated Latent Root..... 257 258 149. The First Species of Semi-Latent Regions...... 258 259 150. The Higher Species of Semi-Latent Regions...... 259 261 151. The Identical Equation.......... 261 152. The Vacuity of a Matrix.......... 261 262 153. Symmetrical Matrices.......... 262 265 154. Symmetrical Matrices and Supplements...... 265 267 155. Skew Matrices............ 267 269 BOOK V. EXTENSIVE MANIFOLDS OF THREE DIMENSIONS. CHAPTER I. Systems of Forces. 156. Non-metrical Theory of Forces........ 273 274 157. Recapitulation of Formulae......... 274 275 158. Inner Multiplication.......... 275 276 159. Elementary Properties of a Single Force...... 276 160. Elementary Properties of Systems of Forces...... 276 277 161. Condition for a Single Force......... 277 162. Conjugate Lines........... 277 278 163. Null Lines, Planes and Points........ 278 164. Properties of Null Lines.......... 279 280 165. Lines in Involution........... 280 281 166. Reciprocal Systems........... 281 282 167. Formulae for Systems of Forces........ 282 283
CONTENTS. CHAPTER II. Groups of Systems of Forces. 168. Specifications of a Group .... 169. Systems Reciprocal to Groups 170. Common Null Lines and Director Forces 171. Quintuple Groups...... 172. Quadruple and Dual Groups .... 173. Anharmonic Ratio of Systems 174. Self-Supplementary Dual Groujjs . 175. Triple Groups...... 176. Conjugate Sets of Systems in a Triple Group PAGES 284 285 285 286 286 287 287 290 290 292 292 294 295 298 298 299 CHAPTER III. Invariants of Groups. 177. Definition of an Iuvariant......... 300 178. The Null Invariants of a Dual Group....... 300 179. The Harmonic Invariants of a Dual Group...... 301 302 180. Further Properties of Harmonic Invariants...... 302 303 181. Formulae connected with Reciprocal Systems...... 303 304 182. Systems Reciprocal to a Dual Group....... 304 183. The Pole and Polar Invariants of a Triple Group .... 305 306 184. Conjugate Sets of Systems and the Pole and Polar Invariants . . 306 307 185. Interpretation of P \x) and P (Z)........ 307 308 186. Relations between Conjugate Sets of Systems..... 308 310 187. The Conjugate Invariant of a Triple Group...... 310 312 188. Transformations of G (p, p) and G (P, P)...... 312 315 CHAPTER IV. Matrices and Forces. 189. Linear Transformations in Three Dimensions..... 316 317 190. Enumeration of Types of Latent and Semi-Lateut Regions . . . 317 321 191. Matrices and Forces.......... 322 323 192. Latent Systems and Semi-Latent Groups...... 323 326 193. Enumeration of Types of Latent Systems and Semi-Latent Groups . 326 338 194. Transformation of a Quadric into itself....... 338 339 195. Direct Transformation of Quadrics........ 339 342 196. Skew Transformation of Quadrics........ 342 346 Note............. 346
CONTENTS. xxi BOOK VI. THEORY OF METRICS. CHAPTER I. Theory of Distance. ART- PAGES 197. Axioms of Distance...........349__350 198. Congruent Ranges of Points......... 350 351 199. Cayley's Theory of Distance.........351__353 200. Klein's Theorem...........353__354 201. Comparison with the Axioms of Distance ...... 354 202. Spatial Manifolds of Many Dimensions.......354__355 203. Division of Space........... 355 356 204. Elliptic Space............ 356 205. Polar Form............ 356 358 206. Length of Intercepts in Polar Form....... 358 361 207. Antipodal Form........... 361 362 208. Hyperbolic Space........... 362 363 209. The Space Constant.......... 363 364 210. Law of Intensity in Elliptic and Hyperbolic Geometry . . . 364 365 211. Distances of Planes and of Subregions....... 365 367 212. Parabolic Geometry.......... 367 368 213. Law of Intensity in Parabolic Geometry...... 368 369 Historical Note........... 369 370 CHAPTER II. Elliptic Geometry. 214. Introductory............ 371 215. Triangles............. 371 373 216. Further Formula for Triangles........ 374 375 217. Points inside a Triangle.......... 375 376 218. Oval Quadrics............ 376 378 219. Further Properties of Triangles........ 378 379 220. Planes One-sided........... 379 382 221. Angles between Planes.......... 382 222. Stereometrical Triangles.......... 382 383 223. Perpendiculars............ 383-385 224. Shortest Distances from Points to Planes...... 385 386 225. Common Perpendicular of Planes . . . . . . . . 386 226. Distances from Points to Subregions....... 387 388 227. Shortest Distances between Subregions....... 388 391 228. Spheres............. 391 396 229. Parallel Subregions........... 397 398
XX11 CONTENTS. CHAPTER III. Extensive Manifolds and Elliptic Geometry. ART. PAGES 230. Intensities of Forces.......... 399 400 231. Relations between Two Forces........ 400 401 232. Axes of a System of Forces......... 401 404 233. Non-Axal Systems of Forces......... 404 234. Parallel Lines............ 404 406 235. Vector Systems........... 406 407 236. Vector Systems and Parallel Lines....... 407 408 237. Further Properties of Parallel Lines....... 409 411 238. Planes and Parallel Lines......... 411 413 CHAPTER IV. Hyperbolic Geometry. 239. Space and Anti-Space.......... 414 240. Intensities of Points and Planes........ 415 416 241. Distances of Points........... 416 417 242. Distances of Planes........... 417 418 243. Spatial and Anti-spatial Lines........ 418 419 244. Distances of Subregions.......... 419 245. Geometrical Signification.......... 420 246. Poles and Polars........... 420 422 247. Points on the Absolute.......... 422 248. Triangles............. 422 424 249. Properties of Angles of a Spatial Triangle...... 424 425 250. Stereometrical Triangles.......... 425 426 251. Perpendiculars............ 426 427 252. The Feet of Perpendiculars......... 427 428 253. Distance between Planes.......... 428 429 254. Shortest Distances........... 429 430 255. Shortest Distances between Subregions....... 430 433 256. Rectangular Rectilinear Figures........ 433 436 257. Parallel Lines............ 436 438 258. Parallel Planes............ 439 440 CHAPTER V. Hyperbolic Geometry (continued). 259. The Sphere............ 441 444 260. Intersection of Spheres.......... 444 447 261. Limit-Surfaces............ 447 448 262. Great Circles on Spheres......... 448 451
CONTENTS. xxiii ART. PAGES 263. Surfaces of Equal Distance from Subregions...... 451 264. Intensities of Forces.......... 452 265. Relations between Two Spatial Forces....... 452 454 266. Central Axis of a System of Forces......, 454 455 267. Non-Axal Systems of Forces......... 455 CHAPTER VI Kinematics in Three Dimensions. 268. Congruent Transformations......... 456 458 269. Elementary Formulae.......... 458 459 270. Simple Geometrical Properties........ 459 460 271. Translations and Rotations......... 460 462 272. Locus of Points of Equal Displacement....... 462 463 273. Equivalent Sets of Congruent Transformations..... 463 274. Commutative Law........... 464 275. Small Displacements.......... 464 465 276. Small Translations and Rotations........ 465 466 277. Associated System of Forces......... 466 278. Properties deduced from the Associated System..... 467 468 279. Work............. 468 469 280. Characteristic Lines........... 470 281. Elliptic Space............ 470 471 282. Surfaces of Equal Displacement........ 472 283. Vector Transformations.......... 472 284. Associated Vector Systems of Forces....... 473 285. Successive Vector Transformations....... 473 476 286. Small Displacements.......... 476 477 CHAPTER VII. Curves and Surfaces. 287. Curve Lines............ 478 479 288. Curvature and Torsion.......... 479 481 289. Planar Formulae........... 481 482 290. Velocity and Acceleration......... 482 484 291. The Circle............ 484 487 292. Motion of a Rigid Body.......... 487 488 293. Gauss' Curvilinear Coordinates........ 488 489 294. Curvature of Surfaces.......... 489 490 295. Lines of Curvature........... 490 493 296. Meunier's Theorem........... 493 297. Normals............. 493_494 298. Curvilinear Coordinates.......... 494 299. Limit-Surfaces............ 494 495
XXIV CONTENTS. CHAPTER VIII. Transition to Parabolic Geometry. ART. PAGES 300. Parabolic Geometry........... 496 301. Plane Equation of the Absolute........ 496 498 302. Intensities............ 498 499 303. Congruent Transformations......... 500 502 BOOK VII. APPLICATION OF THE CALCULUS OF EXTENSION TO GEOMETRY. CHAPTER I. Vectors. 304. Introductory............ 505 506 305. Points at Infinity........... 506 507 306. Vectors............. 507 508 307. Linear Elements........... 508 509 308. Vector Areas............ 509 511 309. Vector Areas as Carriers......... 511 310. Planar Elements........... 512 513 311. Vector Volumes........... 513 312. Vector Volumes as Carriers ......... 513 514 313. Product of Four Points.......... 514 314. Point and Vector Factors......... 514 515 315. Interpretation of Formulae......... 515 516 316. Vector Formulae........... 516 317. Operation of Taking the Vector........ 516 518 318. Theory of Forces........... 518 520 319. Graphic Statics........... 520 522 Note.......,..... 522
CONTENTS. XXV CHAPTER II. Vectors (continued). ART. PAGES 320. Supplements............ 523 524 321. Rectangular Normal Systems......... 524 322. Imaginary Self-Normal Sphere........ 524 525 323. Real Self-Normal Sphere......... 525 526 324. Geometrical Formulae.......... 526 527 325. Taking the Flux........... 527 528 326. Flux Multiplication........... 528 327. Geometrical Formulae.......... 529 328. The Central Axis........... 529 530 329. Planes containing the Central Axis....... 530 330. Dual Groups of Systems of Forces ....... 530 531 331. Invariants of a Dual Group......... 531 332. Secondary Axes of a Dual Group........ 531 532 333. The Cylindroid........... 532 533 334. The Harmonic Invariants......... 533 335. Triple Groups............ 533 534 336. The Pole and Polar Invariants........ 534 535 337. Equation of the Associated Quadric....... 535 338. Normals............. 535 536 339. Small Displacements of a Rigid Body....... 536 537 340. Work............. 537-538 CHAPTER III Curves and Surfaces. 341. Curves............. 539 342. Osculating Plane and Normals....... 540 343. Acceleration............ 540 344. Simplified Formulge........... 541 345. Spherical Curvature.......... 541 542 346. Locus of Centre of Curvature........ 542 543 347. Gauss' Curvilinear Co-ordinates........ 543 544 348. Curvature............ 544 545 349. Lines of Curvature........... 545 546 350. Dupin's Theorem........... 546 547 351. Euler's Theorem........... 547 352 Meunier's Theorem........... 547 Note............. 547
XXVI CONTENTS. CHAPTER IV. Pure Vector Formula. ART. PAGES 353. Introductory............ 548 549 354. Lengths and Areas ........... 549 355. Formulae............. 549 550 356. The Origin............ 550 357. New Convention........... 550 551 358. System of Forces........... 551 359. Kinematics............ 551 552 360. A Continuously Distributed Substance....... 552 554 361. Hamilton's Differential Operator........ 554 555 362. Conventions and Formulae......... 555 557 363. Polar Co-ordinates .......... 557 558 364. Cylindrical Co-ordinates......... 558 560 365. Orthogonal Curvilinear Co-ordinates....... 560 562 366. Volume, Surface, and Line Integrals....... 562 367. The Equations of Hydrodynamics........ 562 563 368. Moving Origin........... 563 565 369. Transformations of Hydrodynamical Equations..... 565 370. Vector Potential of Velocity......... 565 566 371. Curl Filaments of Constant Strength....... 567 569 372. Carried Functions........... 569 570 373. Clebsch's Transformations......... 570 572 374. Flow of a Vector........... 572 573 Note............. 573 Note on Grassmann.......... 573 575 Index............. 576 586
Body of TextBOOK I. PRINCIPLES OF ALGEBRAIC SYMBOLISM. vv.
CHAPTER I. On the nature of a Calculus. 1. Signs. Words, spoken or written, and the symbols of Mathematics are alike signs. Signs have been analysed* into (a) suggestive signs, (ft) expressive signs, (7) substitutive signs. A suggestive sign is the most rudimentary possible, and need not be dwelt upon here. An obvious example of one is a knot tied in a hand- kerchief to remind the owner of some duty to be performed. In the use of expressive signs the attention is not fixed on the sign itself but on what it expresses; that is to say, it is fixed on the meaning conveyed by the sign. Ordinary language consists of groups of expressive signs, its primary object being to draw attention to the meaning of the words employed. Language, no doubt, in its secondary uses has some of the characteristics of a system of substitutive signs. It remedies the inability of the imagination to bring readily before the mind the whole extent of complex ideas by associating these ideas with familiar sounds or marks; and it is not always necessary for the attention to dwell on the complete meaning while using these symbols. But with all this allowance it remains true that language when challenged by criticism refers us to the meaning and not to the natural or conventional properties of its symbols for an explanation of its processes. A substitutive sign is such that in thought it takes the place of that for which it is substituted. A counter in a game may be such a sign: at the end of the game the counters lost or won may be interpreted in the form of money, but till then it may be convenient for attention to be concentrated on the counters and not on their signification. The signs of a Mathematical Calculus are substitutive signs. The difference between words and substitutive signs has been stated thus, i a word is an instrument for thinking about the meaning^ which it * Cf. Stout, * Thought and Language/ Mind, April, 1891, repeated in the same author's Analytic Psychology, (1896), ch. x. 1: cf. also a more obscure analysis to the same effect by C. S. Peirce, Proc. of the American Academy of Arts and Sciences, 1867, Vol. vii. p. 29L 1 2
4 ON THE NATURE OF A CALCULUS. [CHAP. I. expresses; a substitute sign is a means of not thinking about the meaning which it symbolizes*.' The use of substitutive signs in reasoning is to economize thought. 2. Definition of a Calculus. In order that reasoning may be con- ducted by means of substitutive signs, it is necessary that rules be given for the manipulation of the signs. The rules should be such that the final state of the signs after a series of operations according to rule denotes, when the signs are interpreted in terms of the things for which they are substituted, a proposition true for the things represented by the signs. The art of the manipulation of substitutive signs according to fixed rules, and of the deduction therefrom of true propositions is a Calculus. We may therefore define a sign used in a Calculus as 'an arbitrary mark, having a fixed interpretation, and susceptible of combination with other signs in subjection to fixed laws dependent upon their mutual interpretation j".' The interpretation of any sign used in a series of operations must be fixed in the sense of being the same throughout, but in a certain sense it may be ambiguous. For instance in ordinary Algebra a letter x may be used in a series of operations, and x may be defined to be any algebraical quantity, without further specification of the special quantity chosen. Such a sign denotes any one of an assigned class with certain un- ambiguously defined characteristics. In the same series of operations the sign must always denote the same member of the class ; but as far as any explicit definitions are concerned any member will do. When once the rules for the manipulation of the signs of a calculus are known, the art of their practical manipulation can be studied apart from any attention to the meaning to be assigned to the signs. It is obvious that we can take any marks we like and manipulate them according to any rules we choose to assign. It is also equally obvious that in general such occupations must be frivolous. They possess a serious scientific value when there is a similarity of type of the signs and of the rules of manipulation to those of some calculus in which the marks used are substitutive signs for things and relations of things. The comparative study of the various forms produced by variation of rules throws light on the principles of the calculus. Furthermore the knowledge thus gained gives facility in the invention of some significant calculus designed to facilitate reasoning with respect to some given subject. It enters therefore into the definition of a calculus properly so called that the marks used in it are substitutive signs. But when a set of marks and the rules for their arrangements and rearrangements are analogous to * Cf. Stout, ' Thought and Language,1 Mind, April, 1891. + Boole, Laivs of Thought, Ch. n.
2, 3] DEFINITION OF A CALCULUS. 5 those of a significant calculus so that the study of the allowable forms of their arrangements throws light on that of the calculus, or when the marks and their rules of arrangement are such as appear likely to receive an interpretation as substitutive signs or to facilitate the invention of a true calculus, then the art of arranging such marks may be called by an extension of the term an uninterpreted calculus. The study of such a calculus is of scientific value. The marks used in it will be called signs or symbols as are those of a true calculus, thus tacitly suggesting that there is some unknown interpretation which could be given to the calculus. 3. Equivalence. It is necessary to note- the form in which propositions occur in a calculus. Such a form may well be highly artificial from some points of view, and may yet state the propositions in a convenient form for the eliciting of deductions. Furthermore it is not necessary to assert that the form is a general form into which all judgments can be put by the aid of some torture. It is sufficient to observe that it is a form of wide appli- cation. In a calculus of the type here considered propositions take the form of assertions of equivalence. One thing or fact, which may be complex and involve an inter-related group of things or a succession of facts, is asserted to be equivalent in some sense or other to another thing or fact. Accordingly the sign = is taken to denote that the signs or groups of signs on either side of it are equivalent, and therefore symbolize things which are so far equivalent. When two groups of symbols are connected by this sign, it is to be understood that one group may be substituted for the other group whenever either occurs in the calculus under conditions for which the assertion of equivalence holds good. The idea of equivalence requires some explanation. Two things are equivalent when for some purpose they can be used indifferently. Thus the equivalence of distinct things implies a certain defined purpose in view, a certain limitation of thought or of action. Then within this limited field no distinction of property exists between the two things. As an instance of the limitation of the field of equivalence consider an ordinary algebraical equation, f(x, y) = 0. Then in finding -^ by the CtX formula, -^- J- / / , we may not substitute 0 for f on the right-hand dx dx: dy J J side of the last equation, though the equivalence of the two symbols has been asserted in the first equation, the reason being that the limitations under which /= 0 has been asserted are violated when / undergoes partial dif- ferentiation. The idea of equivalence must be carefully distinguished from that of
6 ON THK NATURE OF A CALCULUS. [CHAP. I. mere identity*. No investigations which proceed by the aid of propositions merely asserting identities such as A is A, can ever result in anything but barren identities "f\ Equivalence on the other hand implies non-identity as its general case. Identity may be conceived as a special limiting case of equivalence. For instance in arithmetic we write, 2 + 3 = 3 + 2. This means that, in so far as the total number of objects mentioned, 2 + 3 and 3 + 2 come to the same number, namely 5. But 2 + 3 and 3 + 2 are not identical; the order of the symbols is different in the two combinations, and this difference of order directs different processes of thought. The importance of the equation arises from its assertion that these different processes of thought are identical as far as the total number of things thought of is concerned. From this arithmetical point of view it is tempting to define equivalent things as being merely different ways of thinking of the same thing as it exists in the external world. Thus there is a certain aggregate, say of 5 things, which is thought of in different ways, as 2 + 3 and as 3 + 2. A sufficient objection to this definition is that the man who shall succeed in stating intelligibly the distinction between himself and the rest of the world will have solved the central problem of philosophy. As there is no universally accepted solution of this problem, it is obviously undesirable to assume this distinction as the basis of mathematical reasoning. Thus from another point of view all things which for any purpose can be conceived as equivalent form the extension (in the logical sense) of some uni- versal conception. And conversely the collection of objects which together form the extension of some universal conception can for some purpose be treated as equivalent. So b = b' can be interpreted as symbolizing the fact that the two individual things b and b' are two individual cases of the same general conception B J. For instance if b stand for 2 + 3 and V for 3 + 2, both b and V are individual instances of the general conception of a group of five things. The sign = as used in a calculus must be discriminated from the logical copula 'is/ Two things b and V are connected in a calculus by the sign =, so that b = b't when both b and V possess the attribute B. But we may not translate this into the standard logical form, b is b'. On the contrary, we say, b is B, and 6' is B; and we may not translate these standard forms of formal logic into the symbolic form, b = B, b' = B; at least we may not do so, if the sign = is to have the meaning which is assigned to it in a calculus. It is to be observed that the proposition asserted by the equation, b = ', consists of two elements; which for the sake of distinctness we will name, and will call respectively the ' truism ' and the ' paradox/ The truism is the partial identity of both b and b\ their common i?-ness. The paradox is the * Cf. Lotze, Logic, Bk. i. Ch. n. Art. 64. t Cf. Bradley, Principles of Logic, Bk. i. Ch. v. % Ibid. Bk. ii. Pt. i. Ch. iv. Art. 3 (/3).
4] EQUIVALENCE. 7 distinction between b and b\ so that b is one thing and b' is another thing: and these things, as being different, must have in some relation diverse properties. In assertions of equivalence as contained in a calculus the truism is passed over with the slightest possible attention, the main stress being laid on the paradox. Thus in the equation 2 4- 3 = 3 + 2, the fact that both sides represent a common five-ness of number is not even mentioned explicitly. The sole direct statement is that the two different things 3 + 2 and 2 + 3 are in point of number equivalent. The reason for this unequal distribution of attention is easy to under- stand. In order to discover new propositions asserting equivalence it is requisite to discover easy marks or tests of equivalent things. These tests are discovered by a careful discussion of the truism, of the common l?-ness of b and V. But when once such tests have been elaborated, we may drop all thought of the essential nature of the attribute B and simply apply the superficial test to b and V in order to verify b = V. Thus in order to verify that thirty-seven times fifty-six is equal to fifty-six times thirty-seven, we may use the entirely superficial test applicable to this case that the same factors are mentioned as multiplied, though in different order. This discussion leads us at once to comprehend the essence of a calculus of substitutive signs. The signs are by convention to be considered equiva- lent when certain conditions hold. And these conditions when inter- preted imply the fulfilment of the tests of equivalence. Thus in the discussion of the laws of a calculus stress is laid on the truism, in the development of the consequences on the paradox. 4. Operations. Judgments of equivalence can be founded on direct perception, as when it is judged by direct perception that two different pieces of stuff match in colour. But the judgment may be founded on a knowledge of the respective derivations of the things judged to be equivalent from other things respectively either identical or equivalent. It is this process of derivation which is the special province of a calculus. The derivation of a thing p from things a, b, c, ... , can also be conceived as an operation on the things a b, c, ... , which produces the thing p. The idea of derivation includes that of a series of phenomenal occurrences. Thus two pieces of stuff may be judged to match in colour because they were dyed in the same dipping, or were cut from the same piece of stuff. But the idea is more general than that of phenomenal sequence of events: it includes purely logical activities of the mind, as when it is judged that an aggregate of five things has been presented to the mind by two aggregates of three things and of two things respectively. Another example of derivation is that of two propositions a and b which are both derived by strict deductive reasoning from the same propositions c, d, and e. The two propositions are either both
8 ON THE NATURE OF A CALCULUS. [CHAP. I. proved or both unproved according as c, d, and e are granted or disputed. Thus a and b are so far equivalent. In other words a and b may be considered as the equivalent results of two operations on c, d and e. The words operation, combination, derivation, and synthesis will be used to express the same general idea, of which each word suggests a somewhat specialized form. This general idea may be defined thus: A thing a will be said to result from an operation on other things, c, d, e, etc., when a is presented to the mind as the result of the presentations of c, d and e, etc. under certain conditions; and these conditions are phenomenal events or mental activities which it is convenient to separate in idea into a group by themselves and to consider as defining the nature of the operation which is performed on c, d, e, etc. Furthermore the fact that c, d, ey etc. are capable of undergoing a certain operation involving them all will be considered as constituting a relation between c, d, e, etc. Also the fact that c is capable of undergoing an operation of a certain general kind will be considered as a property of c. Any additional speciali- zation of the kind of operation or of the nature of the result will be considered as a mode of that property. 5. Substitutive Schemes. Let a, a', etc., b, b'y etc.,......zy z', etc., denote any set of objects considered in relation to some common property which is symbolized by the use of the italic alphabet of letters. The common property may not be possessed in the same mode by different members of the set. Their equivalence, or identity in relation to this property, is symbolized by a literal identity. Thus the fact that the things a and m are both symbolized by letters from the italic alphabet is here a sign that the things have some property in common, and the fact that the letters a and m' are different letters is a sign that the two things possess this common property in different modes. On the other hand the two things a and a' possess the common property in the same mode, and as far as this property is concerned they are equivalent. Let the sign = express equivalence in relation to this property, then a = a\ and m = m. Let a set of things such as that described above, considered in relation to their possession of a common property in equivalent or in non-equivalent modes be called a scheme of things; and let the common property of which the possession by any object marks that object as belonging to the scheme be called the Determining Property of the Scheme. Thus objects belonging to the same scheme are equivalent if they possess the determining property in the same mode. Now relations must exist between non-equivalent things of the scheme which depend on the differences between the modes in which they possess the determining property of the scheme. In consequence of these relations
5, 6] SUBSTITUTIVE SCHEMES. 9 from things a, , c, etc. of the scheme another thing m of the scheme can be derived by certain operations. The equivalence, m = m, will exist between m and m', if m and m' are derived from other things of the scheme by operations which only differ in certain assigned modes. The modes in which processes of derivation of equivalent things m and m from other things of the scheme can differ without destroying the equivalence of m and m will be called the Characteristics of the scheme. Now it may happen that two schemes of things with of course different determining properties have the same characteristics. Also it may be possible to establish an unambiguous correspondence between the things of the two schemes, so that if a, a', b, etc., belong to one scheme and a, a', fi, etc., belong to the other, then a corresponds to a, a to a, b to # and so on. The essential rule of the correspondence is that if in one scheme two things, say a and a\ are equivalent, then in the other scheme their corresponding things a and a' are equivalent. Accordingly to any process of derivation in the italic alphabet by which m is derived from a, 6, etc. there must correspond a process of derivation in the Greek alphabet by which fi is derived from a, y8, etc. In such a case instead of reasoning with respect to the properties of one scheme in order to deduce equivalences, we may substitute the other scheme, or conversely; and then transpose at the end of the argument. This device of reasoning, which is almost universal in mathematics, we will call the method of substitutive schemes, or more briefly, the method of substitution. These substituted things belonging to another scheme are nothing else than substitutive signs. For in the use of substituted schemes we cease to think of the original scheme. The rule of reasoning is to confine thought to those properties, previously determined, which are shared in common with the original scheme, and to interpret the results from one set of things into the other at the end of the argument. An instance of this process of reasoning by substitution is to be found in the theory of quantity. Quantities are measured by their ratio to an arbitrarily assumed quantity of the same kind, called the unit. Any set of quantities of one kind can be represented by a corresponding set of quantities of any other kind merely in so far as their numerical ratios to their unit are concerned. For the representative set have only to bear the same ratios to their unit as do the original set to their unit. 6. Conventional Schemes. The use of a calculus of substitutive signs in reasoning can now be explained. Besides using substitutive schemes with naturally suitable properties, we may by convention assign to arbitrary marks laws of equivalence which are identical with the laws of equivalence of the originals about which we
10 ON THE NATURE OF A CALCULUS. [CHAR T. desire to reason. The set of marks may then be considered as a scheme of things with properties assigned by convention. The determining property of the scheme is that the marks are of certain assigned sorts arranged in certain types of sequence. The characteristics of the scheme are the conventional laws by which certain arrangements of the marks in sequence on paper are to be taken as equivalent. As long as the marks are treated as mutually determined by their conventional properties, reasoning concerning the marks will hold good concerning the originals for which the marks are substitutive signs. For instance in the employ- ment of the marks x, y, +, the equation, x + y = y + x, asserts that a certain union on paper of x and y possesses the conventional quality that the order of x and y is indifferent. Therefore any union of two things with a result independent of any precedence of one thing before the other possesses so far properties identical with those of the union above set down between x and y. Not only can the reasoning be transferred from the originals to the substitutive signs, but the imaginative thought itself can in a large measure be avoided. For whereas combinations of the original things are possible only in thought and by an act of the imagi- nation, the combinations of the conventional substitutive signs of a calculus are physically made on paper. The mind has simply to attend to the rules for transformation and to use its experience and imagination to suggest likely methods of procedure. The rest is merely physical actual inter- change of the signs instead of thought about the originals. A calculus avoids the necessity of inference and replaces it by an ex- ternal demonstration, where inference and external demonstration are to be taken in the senses assigned to them by F. H. Bradley*. In this connexion a demonstration is to be defined as a process of combining a complex of facts, the data, into a whole so that some new fact is evident. Inference is an ideal combination or construction within the mind of the reasoner which results in the intuitive evidence of a new fact or relation between the data. But in the use of a calculus this process of combina- tion is externally performed by the combination of the concrete symbols, with the result of a new fact respecting the symbols which arises for sensuous perception -f\ When this new fact is treated as a symbol carrying a meaning, it is found to mean the fact which would have been intuitively evident in the process of inference. 7. Uninterpretable Forms. The logical difficulty J involved in the use of a calculus only partially interpretable can now be explained. The * Cf. Bradley, Principles of Logic, Bk n. Pt i. Ch. in. t Cf. C. S. Peirce, Amer. Journ. of Math. Vol. vn. p. 182: ' As for algebra, the very idea of the art is that it presents formulae which can be manipulated, and that by observing the effects of such manipulation we find properties not otherwise to be discovered.' + Cf. Boole, Law* of Thought, Ch. v. 4.
7] UNINTERPRETABLE FORMS. 11 discussion of this great problem in its application to the special case of ( I)'2 engaged the attention of the leading mathematicians of the first half of this century, and led to the development on the one hand of the Theory of Functions of a Complex Variable, and on the other hand of the science here called Universal Algebra. The difficulty is this: the symbol ( 1)^ is absolutely without meaning when it is endeavoured to interpret it as a number; but algebraic trans- formations which involve the use of complex quantities of the form a + bi, where a and b are numbers and i stands for the above symbol, yield pro- positions which do relate purely to number. As a matter of fact the pro- positions thus discovered were found to be true propositions. The method therefore was trusted, before any explanation was forthcoming why algebraic reasoning which had no intelligible interpretation in arithmetic should give true arithmetical results. The difficulty was solved by observing that Algebra does not depend on Arithmetic for the validity of its laws of transformation. If there were such a dependence, it is obvious that as soon as algebraic expressions are arithmetically unintelligible all laws respecting them must lose their validity. But the laws of Algebra, though suggested by Arithmetic, do not depend on it. They depend entirely on the convention by which it is stated that certain modes of grouping the symbols are to be considered as identical. This assigns certain properties to the marks which form the symbols of Algebra. The laws regulating the manipulation of the algebraic symbols are identical with those of Arithmetic. It follows that no algebraic theorem can ever contradict any result which could be arrived at by Arithmetic; for the reasoning in both cases merely applies the same general laws to different classes of things. If an algebraic theorem is interpretable in Arithmetic, the corresponding arithmetical theorem is therefore true. In short when once Algebra is conceived as an independent science dealing with the re- lations of certain marks conditioned by the observance of certain conventional laws, the difficulty vanishes. If the laws be identical, the theorems of the one science can only give results conditioned by the laws which also hold good for the other science; and therefore these results, when interpretable, are true. It will be observed that the explanation of the legitimacy of the use of a partially interpretable calculus does not depend upon the fact that in another field of thought the calculus is entirely interpretable. The discovery of an interpretation undoubtedly gave the clue by means of which the true solution was arrived at. For the fact that the processes of the calculus were in- terpretable in a science so independent of Arithmetic as is Geometry at once showed that the laws of the calculus might have been defined in reference to geometrical processes. But it was a paradox to assert that a science like Algebra, which had been studied for centuries without reference to Geometry,
12 ON THE NATURE OF A CALCULUS. [CHAP. I. 7 was after all dependent upon Geometry for its first principles. The step to the true explanation was then easily taken. But the importance of the assistance given to the study of Algebra by the discovery of a complete interpretation of its processes cannot be over-esti- mated. It is natural to think of the substitutive set of things as assisting the study of the properties of the originals. Especially is this the case with a calculus of which the interest almost entirely depends upon its relation to the originals. But it must be remembered that conversely the originals give immense aid to the study of the substitutive things or symbols. The whole of Mathematics consists in the organization of a series of aids to the imagination in the process of reasoning; and for this purpose device is piled upon device. No sooner has a substitutive scheme been devised to assist in the investigation of any originals, than the imagination begins to use the originals to assist in the investigation of the substitutive scheme. In some connexions it would be better to abandon the conception of originals studied by the aid of substitutive schemes, and to conceive of two sets of inter-related things studied together, each scheme exemplifying the operation of the same general laws. The discovery therefore of the geometrical representation of the algebraical complex quantity, though unessential to the logic of Algebra, has been quite essential to the modern developments of the science.
CHAPTER II. Manifolds. 8. Manifolds. The idea of a manifold was first explicitly stated by Biemann*; Grassmann*f* had still earlier defined and investigated a particular kind of manifold. Consider any number of things possessing any common property. That property may be possessed by different things in different modes : let each separate mode in which the property is possessed be called an element. The aggregate of all such elements is called the manifold of the property. Any object which is specified as possessing a property in a given mode corresponds to an element in the manifold of that property. The element may be spoken of as representing the object or the object as representing the element according to convenience. All such objects may be conceived as equivalent in that they represent the same element of the manifold. Various relations can be stated between one mode of a property and another mode ; in other words, relations exist between two objects, whatever other properties they may possess, which possess this property in any two assigned modes. The relations will define how the objects necessarily differ in that they possess this property differently: they define the distinction between two sorts of the same property. These relations will be called relations between the various elements of the manifold of the property; and the axioms from which can be logically deduced the whole aggregate of such relations for all the elements of a given manifold are called the characteristics of the manifold. The idea of empty space referred to coordinate axes is an example of a manifold. Each point of space represents a special mode of the common property of spatiality. The fundamental properties of space expressed in terms of these coordinates, i.e. all geometrical axioms, form the character- istics of this manifold. * Ueber die Hypothesen, welche der Geometrie zu Grunde liegen, Gesammelte Mathematische Werke; a translation of this paper is to be found in Clifford's Collected Mathematical Papers. t Ausdehnungslehre von 1844.
14 MANIFOLDS. [CHAP. II. It is the logical deductions from the characteristics of a manifold which are investigated by means of a calculus. The manifolds of separate proper- ties may have the same characteristics. In such a case all theorems which are proved for one manifold can be directly translated so as to apply to the other. This is only another mode of stating the ideas explained in Chapter I. % 3, 4, 5. The relation of a manifold of elements to a scheme of things (cf. 5), is that of the abstract to the concrete. Consider as explained in 5 the scheme of things represented by a, a! etc., b, V etc., ......z, z' etc. Then these concrete things are not elements of a manifold. But to such a scheme a manifold always corresponds, and conversely to a manifold a scheme of things corresponds. The abstract property of a common J.-ness which makes the equivalence of a, a\ etc., in the scheme is an element of the manifold which corresponds to this scheme. Thus the relation of a thing in a scheme to the corresponding element of the corresponding manifold is that of a subject of which the element can be predicated. If A be the element corresponding to a, a etc., then a is A, and a is A. Thus if we write 2 + 3 = 5 at length, the assertion is seen to be (1 + 1)+ (1 + 1 + 1) = 1 + 1 + 1+1 + 1; this asserts that two methods of grouping the marks of the type 1 are equivalent as far as the common five-ness of the sum on each side. The manifold corresponding to a scheme is the manifold of the deter- mining property of the scheme. The characteristics of the manifold corre- spond to the characteristics of the scheme. 9. Secondary Properties of Elements. In order to state the characteristics of a manifold it may be necessary to ascribe to objects corre- sponding to the elements the capability of possessing other properties in addition to that definite property in special modes which the elements represent. Thus for the purpose of expressing the relation of an element A of a manifold to the elements B and G it may be necessary to conceive an object corresponding to A which is either ax or a2, or a3, where the suffix denotes the possession of some other property, in addition to the .A-ness of A, in some special mode which is here symbolized by the suffix chosen. Such a property of an object corresponding to A, which is necessary to define the relation of A to other elements of the manifold, is called a Secondary Property of the element A. Brevity is gained by considering each element of the manifold, such as A, as containing within itself a whole manifold of its secondary properties. Thus with the above notation A stands for any one of Alt A2, A3 etc., where the suffix denotes the special mode of the secondary property. Hence the object ctj, mentioned above, corresponds to Aj, and a to A2, and so on.
9, 10] SECONDARY PROPERTIES OF ELEMENTS. 15 And the statement of the relation between two elements of the original manifold, such as A and B, requires the mention of a special A, say A2 and of a special B, say B4. For example consider the manifold of musical notes conceived as repre- senting every note so far as it differs in pitch and quality from every other note. Thus each element is a note of given pitch and given quality. The attribute of loudness is not an attribute which this manifold represents ; but it is a secondary property of the elements. For consider a tone A and two of its overtones B and C, and consider the relations of A, B, G to a note P which is of the same pitch as A and which only involves the overtones B and G. Then P can be described as the pitch and quality of the sound produced by the simultaneous existence of concrete instances of A, B and G with certain relative loudnesses. Hence the relation of P to A, B, G requires the mention of the loudness of each element in order to express it. Thus if A2i B3, G4 denote A, By G with the required ratio of their loudnesses, P might be expressed as the combination of A2, B3, G4. The sole secondary property with which in this work we shall be concerned is that of intensity. Thus in some manifolds each element is to be conceived as the seat of a possible intensity of any arbitrarily assumed value, and this intensity is a secondary property necessary to express the various relations of the elements. 10. Definitions. To partition a manifold is to make a selection of elements possessing a common characteristic: thus if the manifold be a plane, a selection may be made of points at an equal distance from a given point. The selected points then form a circle. The selected elements of a partitioned manifold form another manifold, which may be called a submanifold in reference to the original manifold. Again the common attribute G, which is shared by the selected elements of the original manifold A, may also be shared by elements of another manifold B, For instance in the above illustration other points in other planes may be at the same distance from the given point. We thus arrive at the conception of the manifold of the attribute G which has common elements with the manifolds A and B. This conception undoubtedly implies that the three manifolds A, B and G have an organic connection, and are in fact parts of a manifold which embraces them all three. A manifold will be called the complete manifold in reference to its possible submanifolds ; and the complete manifold will be said to contain its submanifolds. The submanifolds will be said to be incident in the complete manifold. One submanifold may be incident in more than one manifold. It will then be called a common submanifold of the two manifolds. Manifolds will be said to intersect in their common submanifolds.
16 MANIFOLDS. [CHAP. II. 11. Special Manifolds. A few definitions of special manifolds will both elucidate the general explanation of a manifold given above and will serve to introduce the special manifolds of which the properties are dis- cussed in this work. A manifold may be called self-constituted when only the properties which the elements represent are used to define the relations between elements; that is, when there are no secondary properties. A manifold may be called extrinsically constituted when secondary properties have to be used to define these relations. The manifold of integral numbers is self-constituted, since all relations of such numbers can be defined in terms of them. A uniform manifold is a manifold in which each element bears the same relation as any other element to the manifold considered as a whole. If such a manifold be a submanifold of a complete manifold, it is not necessary that each element of the uniform submanifold bear the same relation to the complete manifold as any other element of that submanifold. Space, the points being elements, forms a uniform manifold. Again the perimeter of a circle, the points being elements, forms a uniform mani- fold. The area of a circle does not form a uniform manifold. A simple serial manifold is a manifold such that the elements can be arranged in one series. The meaning of this property is that some determinate process of deriving the elements in order one from the other exists (as in the case of the successive integral numbers), and that starting from some initial element all the other elements of the manifold are derived in a fixed order by the successive application of this process. Since the process is determinate for a simple serial manifold, there is no ambiguity as to the order of suc- cession of elements. The elements of such a manifold are not necessarily numerable. A test of a simple serial manifold is that, given any three elements of the manifold it may be possible to conceive their mutual relations in such a fashion that one of them can be said to lie between the other two. If a simple serial manifold be uniform it follows that any element can be chosen as the initial element. A manifold may be called a complex serial manifold when all its elements belong to one or more submanifolds which are simple serial manifolds, but when it is not itself a simple serial manifold. A surface is such a manifold, while a line is a simple serial manifold. Two manifolds have a one to one correspondence* between their elements if to every element of either manifold one and only one element of the other manifold corresponds, so that the corresponding elements bear a certain defined relation to each other. * The subject of the correspondence between the elements of manifolds has been investigated by G. Cantor, in a series of memoirs entitled, *Ueber unendliche, lineare Punktmannichfaltigkeiten,1 Math. Annalen, Bd. 15, 17, 20, 21, 23, and BorchardVs Journal, Bd. 77, 84.
11] SPECIAL MANIFOLDS. 17 A quantitively defined manifold is such that each element is specified by a definite number of measurable entities of which the measures for any element are the algebraic quantities , tj, f, etc., so that the manifold has a one to one correspondence with the aggregate of sets of simultaneous values of these variables. A quantitively defined manifold is a manifold of an algebraic function when each element represents in some way the value of an algebraic quantity w for a set of simultaneous values of , rj, , etc., where w is a function of , r), f, etc., in the sense that it can be constructed by definite algebraic operations on f, tj, f, etc., regarded as irresoluble magnitudes, real or imaginary *. A quantitively defined manifold in which the elements are defined by a single quantity is a simple serial manifold as far as real values of are concerned. For the elements can be conceived as successively generated in the order in which they occur as varies from x to + oo . If an element of the manifold corresponds to each value of f as it varies continuously through all its values, then the manifold may be called con- tinuous. If some values of f have no elements of the manifold corresponding to them, then the manifold may be called discontinuous. A quantitively defined manifold depending on more than one quantity is a complex serial manifold. For if the quantities defining it f, rj} f, etc. be put equal to arbitrary functions of any quantity r, so that =/i (t), tj =/2 (t), etc., then a submanifold is formed which is a quantitively defined manifold depending on the single quantity t. This submanifold is therefore a simple serial manifold. But by properly choosing the arbitrary functions such a submanifold may be made to contain any element of the complete manifold. Hence the complete manifold is a complex serial manifold. The quantitively defined manifold is continuous if an element corresponds to every set of values of the variables. A quantitively defined manifold which requires for its definition the absolute values (as distinct from the ratios) of v variables is said to be of v dimensions. A continuous quantitively defined manifold of v dimensions may also be called a z/-fold extended continuous manifold -f\ * Cf. Forsyth, Theory of Functions, Ch. i. 6, 7. t Cf. Riemann, loc. cit. section i. 2. W.
CHAPTER III. Principles of Universal Algebra. 12. Introductory. Universal Algebra is the name applied to that calculus which symbolizes general operations, defined later, which are called Addition and Multiplication. There are certain general definitions which hold for any process of addition and others which hold for any process of multiplication. These are the general principles of any branch of Universal Algebra. These principles, which are few in number, will be considered in the present chapter. But beyond these general definitions there are other special definitions which define special kinds of addition or of multiplication. The development and comparison of these special kinds of addition or of multipli- cation form special branches of Universal Algebra. Each such branch will be called a special algebraic calculus, or more shortly, a special algebra, and the more important branches will be given distinguishing names. Ordinary algebra will, when there is no risk of confusion, be called simply algebra; but when confusion may arise, the term ordinary will be prefixed. 13. Equivalence. It has been explained in 3 that the idea of equivalence requires special definition for any subject-matter to which it is applied. The definitions of the processes of addition and multiplication do carry with them this required definition of equivalence as it occurs in the field of Universal Algebra. One general definition holds both for addition and multiplication, and thus through the whole field of Universal Algebra. This definition may be framed thus: In any algebraic calculus only one recognized type of equivalence exists. The meaning of this definition is that if two symbols a and a! be equivalent in that sense which is explicitly recognized in some algebraic calculus by the use of the symbol =, then either a or a' may be used indifferently in any series of operations of addition or multiplication of the type defined in that calculus. This definition is so far from being obvious or necessary for any symbolic calculus, that it actually excludes from the scope of Universal Algebra the
12 14] EQUIVALENCE. 19 Differential Calculus, excepting limited parts of it. For if f(x, y) be a function of two independent variables x and y, and the equivalence 7) 7) f(x, y) = 0, be asserted, then ^-f(xf y) and ^-f(x, y) are not necessarily 7\ 7\ zero, whereas =- 0 and ~- 0 are necessarily zero. Hence the symbols f(x, y) and 0 which are recognized by the sign of equality as equivalent according to one type of equivalence are not equivalent when submitted to some operations which occur in the calculus. 14. Principles of Addition. The properties of the general operation termed addition will now be gradually defined by successive specifications. Consider a group of things, concrete or abstract, material things or merely ideas of relations between other things. Let the individuals of this group be denoted by letters a, b ... z. Let any two of the group of things be capable of a synthesis which results in some third thing. Let this synthesis be of such a nature that all the properties which are attributed to any one of the original group of things can also be attributed to this result of the synthesis. Accordingly the resultant thing belongs to the original group. Let the idea of order between the two things be attributable to their synthesis. Thus if a and b be the two things of which the synthesis is being discussed, orders as between a first or b first can be attributed to this synthesis. Also let only two possible alternative orders as between a and-6 be material, so as to be taken into explicit consideration when judging that things are or are not equivalent. Let the result of the synthesis be unambiguous, in the sense that all possible results of a special synthesis in so far as the process is varied by the variation of non-apparent details are to be equivalent. It is to be noted in this connection that the properties of the synthesis which are explicitly mentioned cannot be considered as necessarily defining its nature unambiguously. The present assumption therefore amounts to the state- ment that the same words (or symbols) are always to mean the same thing, at least in every way which can affect equivalence. This process of forming a synthesis between two things, such as a and b} and then of considering a and 6, thus united, as a third resultant thing, may be symbolized by a * 6*. Here the order is symbolized by the order in which a and b are mentioned; accordingly a ^ b and b * a symbolize two different things. Then by definition the only question of order as between a and b which can arise in this synthesis is adequately symbolized. Also a ~ b whenever it occurs must always mean the same thing, or at least stand for some one of a set of equivalent things. * Cf. Grassmann, Ausdehnungslehre von 1844, Preface. 2 2
20 PRINCIPLES OF UNIVERSAL ALGEBRA. [CHAP. III. Further a ^ b is by assumption a thing capable of the same synthesis with any other of the things a, b, ... x. Accordingly we may write p rs (a rs b) and (a ^b) * p to represent the two possible syntheses of the type involving p and a ^ b. The bracket is to have the usual meaning that the synthesis within the bracket is to be performed first and the resultant thing then to be combined as the symbols indicate. According to the convention adopted here the symbol a ^ b is to be read from left to right in the following manner: a is to be considered as given first, and b as joined on to it according to the manner prescribed by the symbol ^. Thus (a ^ b) ^p means that the result of a * b is first obtained and then p is united to it. But a * b is obtained by taking a and joining b on to it. Thus the total process may equally well be defined by a ^b * p. Hence, since both its right-hand and left-hand sides have been defined to have the same meaning, we obtain the equation a ~ b ^ p (a * b) ~ p. Definition. Let any one of the symbols, either a single letter or a com- plex of letters, which denotes one of the group of things capable of this synthesis be called a term. Let the symbol ^ be called the sign of the operation of this synthesis. It will be noticed that this synthesis has essentially been defined as a synthesis between two terms, and that when three terms such as a, by p, are indicated as subjects of the synthesis a sequence or time-order of the opera- tions is also unambiguously defined. Thus in the syntheses (a ^ b) * p there are two separate ideas of order symbolized; namely, the determined but unspecified idea of order of synthesis as between the two terms which is involved by hypothesis in the act of synthesis, and further the sequence of the two successive acts of synthesis, and this time-order involves the sequence in which the various terms mentioned are involved in the process. Thus a ^b * p and p * (a * b) both involve that the synthesis a * b is to be first performed and then the synthesis of a * b and p according to the special order of synthesis indicated. In the case of three successive acts of synthesis an ambiguity may arise. Consider the operations indicated in the symbols a b rs c^ d, c * (a ^ b) ^ d. No ambiguity exists in these two expressions; each of them definitely indicates that the synthesis a ^ b is to be made first, then a synthesis with c, and then a synthesis of this result with d. Similarly each of the two expressions d ~ (a ^ b ^ c), and d * {c ~ (a ^ b)} indicates unambiguously the same sequence of operations, though in the final synthesis of d with the result of the previous syntheses the alternative order of synthesis is adopted to that adopted in the two previous examples.
14, 15] PRINCIPLES OF ADDITION. 21 But consider the expressions (a * b) * (c * d) and (c * d) ^ (a * b). Here the two syntheses a ~ b and c * d are directed to be made and then the resulting terms to be combined together. Accordingly there is an ambiguity as to the sequence in which these syntheses a ^ b, c ^ d are to be performed. It has been defined however that a ~ b and c ~ d are always to be unambiguous and mean the same thing. This definition means that the synthesis ^ depends on no previous history and no varying part of the environment. Accordingly a ^ b is independent of c * d and these operations may take place in any sequence of time. The preceding definitions can be connected with the idea of a manifold. All equivalent things must represent the same element of the manifold. The synthesis a ^ b is a definite unambiguous union which by hypothesis it is always possible to construct with any two things representing any two elements of the manifold. This synthesis, when constructed and represented by its result, represents some third element of the manifold. It is also often convenient to express this fact by saying that a b represents a relation between two elements of the manifold by which a third element of the manifold is generated; or that the term a * b represents an element of the manifold. An element may be named after a term which represents it: thus the element x is the element represented by the term x. The same element might also be named after any term equivalent to x. It is obvious that any synthesis of the two terms a and b may be conceived as an operation performed on one of them with the help of the other. Accordingly it is a mere change of language without any alteration of real meaning, if we sometimes consider a * b as representing an operation performed on b or on a. 15. Addition. Conceive now that this synthesis which has been defined above is such that it follows the Commutative and Associative Laws. The Commutative Law asserts that a ^ b = b r\ a. Hence the two possible orders of synthesis produce equivalent results. It is to be carefully noticed that it would be erroneous to state the commutative law in the form that, order is not involved in the synthesis a ^ 6. For if order is not predicable of the synthesis, then the equation, a * b=b * a, must be a proposition which makes no assertion at all. Accordingly it is essential to the importance of the commutative law that order should be involved in the synthesis, but that it should be indifferent as far as equi- valence is concerned. The Associative Law is symbolized by where a * b ^ c is defined in 14.
22 PRINCIPLES OF UNIVERSAL ALGEBRA. [CHAP. III. The two laws combined give the property that the element of the manifold identified by three given terms in successive synthesis is independent of the order in which the three terms are chosen for the operation, and also of the internal order of each synthesis. Let a synthesis with the above properties be termed addition; and let the manifold of the corresponding type be called an algebraic manifold; and let a scheme of things representing an algebraic manifold be called an algebraic scheme. Let addition be denoted by the sign +. Accordingly it is to be understood that the symbol a-ft represents a synthesis in which the above assumptions are satisfied. The properties of this operation will not be found to vary seriously in the different algebras. The great distinction between these properties turns on the meaning assigned to the addition of a term to itself. Ordinary algebra and most special algebras distinguish between a and a-\- a. But the algebra of Symbolic Logic identifies a and a H- a. The consequences of these assumptions will be discussed subsequently. 16. Principles of Subtraction. Let a and b be terms representing any two given elements of an algebraic manifold. Let us propose the problem, to find an element x of the manifold such that x -f- b = a. There may be no general solution to this problem, where a and b are connected by no special conditions. Also when there is one solution, there may be more than one solution. It is for instance easy to see that in an algebra which identifies a and a + a, there will be at least two solutions if there be one. For if x be one answer, then x + b = x + b +b= a. Hence x + b is another answer. If there be a solution of the above equation, let it be written in the form, a ^ b. Then it is assumed that a ^ b represents an element of the mani- fold, though it may be ambiguous in its signification. The definition of a ^ b is a ^ b + b = a.................................(1). If c be another element of the manifold let us assume that (a ^ b) ^ c symbolizes the solution of a double problem which has as its solution or solutions one or more elements of the manifold. Then a ^ b ^ c + (b + c) = a ^ b ~ c 4- (c + b) = a^b^c + c + b = a ~ b + b = a. It follows that the problem proposed by the symbol a ^ (b + c) has one or more solutions, and that the solutions to the problem flwjvcare included in them.
10] PRINCIPLES OF SUBTRACTION. 23 Conversely suppose that the problem a ^ (b + c) is solved by one or more elements of the manifold. Then by hypothesis a ^ (b + c) + (b + c) = a; and hence But if d 4- c 4- b = a, then d + c is one value of a ^ b and d is one value of a ^ b ^ c. Accordingly a ^ b^ c is a problem which by hypothesis must have one or more solutions, and the solutions to a ^ (b + c) are included in them. Hence since the solutions of each are included in those of the other, the two problems must have the same solutions. Therefore whatever particular meaning (in the choice of ambiguities) we assign to one may also be assigned to the other. We may therefore write a^(b + c) = a^b^ c.................................(2). Again we have a v (b 4- c) = a ^ (c + 6). Hence from equation (2), aw fcw c = a^ c ^ b..............................(3). It may be noted as a consequence of equations (2) and (3), that if a v (b + c) admit of solutions, then also both a ^ b and a ^ c admit of solutions. Hence if a ^ b and b ^ c admit of solutions ; then a ^ b = a ^ (b ^ c -f c) ; and it follows from the above note that a ^ (b ^ c) admits of a solution. Also in this case a^b + c = a,^(bvc + c) + c = av(bvc)^c + c, from equation (2). Hence a^b-\-c = a^(b^ c)..............................(4). We cannot prove that a^ b-\-c = a-\~c^b, and that a + (b ^ c) = a + b ^ c, without making the assumption that a ^ b, if it exists, is unambiguous. Summing up: for three terms a, b and c there are four equivalent forms symbolized by (a ^ b) v c = (a ^ c) ^ b = a ^ (b + c) = a ^ (c + b): also there are three sets of forms, the forms in each set being equivalent but not so forms taken from different sets, namely (a^b) + c = a ^ (b ^ c) = c + (a^b) ..................(a), (c w b) + a = c ~ (b v a) = a + (c ~ b) ..................(/3), c) w b = (c + a) ^ b .................................(7). Subtraction. Let us now make the further assumption that the reverse analytical process is unambiguous, that is to say that only one element of
24 PRINCIPLES OF UNIVERSAL ALGEBRA. [CHAP. III. the manifold is represented by a symbol of the type a ^ b. Let us replace in this case the sign ^ by , and call the process subtraction. Now at least one of the solutions of a +- b ^ b is a. Hence in subtraction the solution of a + b b is a, or symbolically a + b b = a. But by definition, a b + b = a. Hence, a -\-b b = a b + b = a...........................(5). We may note that the definition, a b + b = a, assumes that the question a b has an answer. But equation (5) proves that a manifold may always without any logical contradiction be assumed to exist in which the subtractive question a b has an answer independently of any condition between a and b. For from the definition, a b + b, where a b is assumed to have an answer, can then be transformed into the equivalent form a + b b, which is a question capable of an answer without any condition between a and b. But it may happen that in special interpretations of an algebra a b, though unam- biguous, has no solution unless a and b satisfy certain conditions. The remarks of 7 apply here. Again a+b c = a + (b c + c) c = a + (b c) + c c = a+(b-c).................................(6). 17. The Null Element. On the assumption that to any question of the type a b can be assigned an answer, some meaning must be assigned to the term a a. Now if c be any other term, c + a a = c = c + b b. Hence it may be assumed that a a = b b. Thus we may put a-a = 0.......................................(7); where 0 represents an element of the manifold independent of a. Let the element 0 be called the null element. The fundamental property of the null element is that the addition of this element and any other element a of the manifold yields the same element a. It would be wrong to think of 0 as necessarily symbolizing mere nonentity. For in that case, since there can be no differences in nonentities, its equivalent forms a a and b b must be not only equivalent, but absolutely identical; whereas they are palpably different. Let any term, such as a a, which represents the null element be called a null term. The fundamental property of 0 is, a + 0 = a.......................................(8).
17 19] THE NULL ELEMENT. 25 Other properties of 0 which can be derived from this by the help of the previous equations are, 0 + 0 = 0; and a 0 = a (b b) = a b + b =a. Again forms such as 0 a may have a meaning and be represented by definite elements of the manifold. The fundamental properties of 0 a are symbolized by and b (0- a) = 6-0 + a = 6 + a. Since in combination with any other element the null element 0 dis- appears, the symbolism may be rendered more convenient by writing a for 0 a. Thus a is to symbolize the element 0 a. 18. Steps. We notice that, since a = 0 + a, we may in a similar way consider a or + a as a degenerate form of 0 + a. From this point of view every element of the manifold is defined by reference to its relation with the null element. This relation with the null element may be called the step which leads from the null element to the other element. And by fastening the attention rather on the method of reaching the final element than on the element itself when reached, we may call the symbol + a the symbol of the step by which the element a of the manifold is reached. This idea may be extended to other elements besides the null element. For we may write b = a + (6 a): and 6 a may be conceived as the step from a to b. The word step has been used* to imply among other things a quantity; but as defined here there is no necessary implication of quantity. The step + a is simply the process by which any term p is transformed into the term p + a. The two steps + a and a may be conceived as exactly opposed in the sense that their successive application starting from any term p leads back to that term, thus p + a a=p. In relation to + a, the step a will be called a negative step; and in relation to a, the step + a will be called a positive step. The fundamental properties of steps are (1) that they can be taken in any order, which is the commutative law, and (2) that any number of successive steps may be replaced by one definite resultant step, which is the associative law. The introduction of the symbols + a and a involves the equations + (+ a) = + (0 + a) = 0 + a = + a = a, _ (+ a) = - (0 + a) = - 0 - a = - a, - (- a) = - (0 - a) = - 0 + a = + a = a.) 19. Multiplication. A new mode of synthesis, multiplication, is now to be introduced which does not, like addition, necessarily concern terms of a * Cf. Clifford, Elements of Dynamic.
26 PRINCIPLES OF UNIVERSAL ALGEBRA. [CHAP. III. single algebraic scheme (cf. 15), nor does it necessarily reproduce as its result a member of one of the algebraic schemes to which the terms synthe- sized belong. Again, the commutative and associated laws do not necessarily hold for multiplication; but a new law, the distributive law, which defines the relation of multiplication to addition holds. Any mode of synthesis for which this relation to addition holds is here called multiplication. The result of multiplication like that of addition is unambiguous. Consider two algebraic manifolds; call them the manifolds A and B. Let a, a, a" etc., be terms denoting the various elements of A, and let b, b', b" etc., denote the various elements of B. Assume that a mode of synthesis is possible between any two terms, one from each manifold. Let this synthesis result in some third thing, which is the definite unambiguous product under all circumstances of this special synthesis between those two elements. Also let the idea of order between the two things be attributable to their union in this synthesis. Thus if a and b be the two terms of which the synthesis is being discussed, an order as between a first or b first can be attributed to this synthesis. Also let only two possible alternative orders as between a and b exist. Let this mode of synthesis be, for the moment, expressed by the sign ^. Thus between two terms a, b from the respective manifolds can be generated the two things a ^ b and b ^ a. All the things thus generated may be represented by the elements of a third manifold, call it G. Also let the symbols a^6 and b^a conceived as representing such things be called terms. Now assume that the manifold G is an algebraic manifold, according to the definition given above ( 15). Then its corresponding terms are capable of addition. And we may write (a ^ b) + ( ' ^ a") + etc.; forming thereby another term representing an ele- ment of the manifold C. The definition of the algebraic nature of G does not exclude the possi- bility that elements of G exist which cannot be formed by this synthesis of two elements from A and Brespectively. For (a^b) + (6"^a) is by definition an element of G ; but it will appear that this element cannot in general be formed by a single synthesis of either of the types a^ ~ b^ or b^q) ^ aS*K Again a -f a! -f a" + etc., represents an element of the manifold A, and b -f b' + b" + etc., represents an element of the manifold B. Hence there are elements of the manifold G represented by terms of the form (a + a'+ a" + etc.) ^(6+ ' + " + ...), and (b + b' -f b" + etc.) ^ (a + a + a" +...). Now let this synthesis be termed Multiplication, when such expres- sions as the above follow the distributive law as defined by equations (10) below. For multiplication let the synthesis be denoted by x or by mere juxta-
20] MULTIPLICATION. 27 position. Then the definition of multiplication yields the following symbolic statements a (b + V) = ab + ab', (a + d) b = ab + db, \ b(a + d) = ba + bd, j ......" "'..............^ *' (b + V) a = ba + b'a. ) It will be noticed that the general definition of multiplication does not involve the associative or the commutative law. 20. Orders of Algebraic Manifolds. Consider a single algebraic manifold A, such that its elements can be multiplied together. Call such a manifold a self-multiplicative manifold of the first order. Now the products of the elements, namely aa, ad, da, etc., by hypothesis form another alge- braic manifold; call it B. Then B will be defined to be a manifold of the second order. Now let the elements of A and B be capable of multiplication, thus forming another algebraic manifold C. Let 0 be defined to be a manifold of the third order. Also in the same way the elements of A and C form by multiplication an algebraic manifold, D, of the fourth order; and so on. Further let the elements of any two of these manifolds be capable of multiplication, and each manifold be self-multiplicative. Let the following law hold, which we may call the associative law for manifolds. The elements formed by multiplying elements of the manifold of the mth order with elements of the manifold of the ?ith order belong to the manifold of the (m + n)t\i order. Thus the complete manifold of the mth order is formed by the multiplica- tion of the elements of any two manifolds, of which the sum of the orders forms m, and also by the elements deduced by the addition of elements thus formed. For instance ad, add', add'd", represent elements of the manifolds of the second, third, and fourth orders respectively; also aa represents an element of the manifold of the second order. Also a" {ad) is an element of the mani- fold of the third order; and (ad) {d'd") is an element of the manifold of the fourth order; and ad {ad d'd") is an element of the manifold of the sixth order; and so on. Such a system of manifolds will be called a complete algebraic system. In special algebras it will be found that the manifold of some order, say the mth, is identical with the manifold of the first order. Then the manifold of the m + lth order is identical with that of the second order, and so on. Such an algebra will be said to be of the m - lth species. In an algebra of the first species only the manifold of the first order can occur. Such
28 PRINCIPLES OF UNIVERSAL ALGEBRA. [CHAP. III. an algebra is called linear. The Calculus of Extension, which is a special algebra invented by Grassmann, can be of any species. It will save symbols, where no confusion results, to use dots instead of brackets. Thus a" (ad) is written d'.ad, and (ad)(dfd") is written ad. d'd", and so on. A dot will be conceived as standing for two opposed bracket signs, thus )(, the other ends of the two brackets being either other dots or the end or beginning of the row of letters. Thus ah . cd stands for (ab) (cd), and is not (ab) cd, unless in the special algebra considered, the two expressions happen to be identical; also ab . cde .fg stands for (ab) (cde) (fg). It will be noticed that in these examples each dot has been replaced by two opposed bracket signs. An ingenious use of dots has been proposed by Mr W. E. Johnson which entirely obviates the necessity for the use of brackets. Thus a [b (cd)} is written a ..b .cd, and a [b [c (de)}] is written a... 6.. c .de. The principle of the method is that those multiplications indicated by the fewest dots are the first performed. Thus a \b(cd)) (ef) is written a .. 6 .cd..ef, and a {b (cd)} ef is written a..b.cd..e ...f, where in the case of equal numbers of dots the left-hand multiplication is first performed. 21. The Null Element. Returning to the original general conception of two algebraic manifolds A and B of which the elements can be multiplied together, and thus form a third algebraic manifold 0; let Ox be the null element of A, 02 the null element of B, and 03 the null element of C. Then if a and b represent any two elements of the manifolds A and B respectively, we have a + Oi = a, and b + 02 = b. Hence (a + O1)b = ab = ab-h 0^. Accordingly, OJ) = 03. Similarly, 60x = 03 = aO2 = 02a. No confusion can arise if we use the same symbol 0 for the null elements of each of the three manifolds. Accordingly, Qa = a0 = 06 = 60 = 0..............................(11). It will be observed that a null element has not as yet been defined for the algebraic manifold in general; but only for those which allow of the process of subtraction, as defined in 16. Thus manifolds for which the relation a + a = a holds are excluded from the definition. In order to include these manifolds let now the null element be defined as that single definite element, if it exist, of the manifold for which the equation a + 0 = a, holds, where a is any element of the manifold. It will be noted that for the definite element a the same property may
21, 22] THE NULL ELEMENT. 29 hold for a as well as for 0; since in some algebras a + a = a. But 0 is defined to be the single element which retains this property with all elements. Then in the case of multiplication equations (11) hold. 22. Classification of Special Algebras. The succeeding books of this work will be devoted to the discussion and comparison of the leading special algebras. It remains now to explain the plan on which this in- vestigation will be conducted. It follows from a consideration of the ideas expounded in Chapter I. that it is desirable to conduct the investigation of a calculus strictly in connection with its interpretations, and that without some such interpretation, however general, no great progress is likely to be made. Therefore each special algebra will, as far as possible, be interpreted concurrently with its in- vestigation. The interpretation chosen, where many are available, will be that which is at once most simple and most general; but the remaining applications will also be mentioned with more or less fulness according as they aid in the development of the calculus. It must be remembered, however, in explanation of certain obvious gaps that the investigation is primarily for the sake of the algebra and not of the interpretation. No investigation of ordinary algebra will be attempted. This calculus stands by itself in the fundamental importance of the theory of quantity which forms its interpretation. Its formulae will of course be assumed throughout when required. In the classification of the special algebras the two genera of addition form the first ground for distinction. For the purpose of our immediate discussion it will be convenient to call the two genera of algebras thus formed the non-numerical genus and the numerical genus. In the non-numerical genus investigated in Book II. the two symbols a and a -f a, where a represents any element of the algebraic manifold, are equivalent, thus a = a + a. This definition leads to the simplest and most rudimentary type of algebraic symbolism. No symbols representing number or quantity are required in it. The interpretation of such an algebra may be expected therefore to lead to an equally simple and fundamental science. It will be found that the only species of this genus which at present has been developed is the Algebra of Symbolic Logic, though there seems no reason why other algebras of this genus should not be developed to receive interpretations in fields of science where strict demonstrative reasoning with- out relation to number and quantity is required. The Algebra of Symbolic Logic is the simplest possible species of its genus and has accordingly the simplest interpretation in the field of deductive logic. It is however always desirable while developing the symbolism of a calculus to reduce the inter- pretation to the utmost simplicity consistent with complete generality.
30 PRINCIPLES OF UNIVERSAL ALGEBRA. [CHAP. III. Accordingly in discussing the main theory of this algebra the difficulties peculiar to Symbolic Logic will be avoided by adopting the equally general interpretation which considers merely the intersection or non-intersection of regions of space. This interpretation will be developed concurrently with the algebra. After the main theory of the algebra has been developed, the more abstract interpretation of Symbolic Logic will be introduced. In the numerical genus the two symbols a and a + a are not equivalent. The symbol a -f a is shortened into 2a; and by generalization of this process a symbol of the form fa is created, where f is an ordinary algebraical quantity, real or imaginary. Hence the general type of addition for this genus is symbolized by ga + r}b + c + etc., where a, b, c, etc. are elements of the algebraic manifold, and f, rj, f, etc. are any ordinary algebraic quantities (such quantities being always symbolized by Greek letters, cf. Book III. Chapter I. below). There are many species of algebra with im- portant interpretations belonging to this genus; and an important general theory, that of Linear Associative Algebras, connecting and comparing an indefinitely large group of algebras belonging to this genus. The special manifolds, which respectively form the interpretation of all the special algebras of this genus, have all common properties in that they all admit of a process symbolized by addition of the numerical type. Any manifold with these properties will be called a ' Positional Manifold.' It is therefore necessary in developing the complete theory of Universal Algebra to enter into an investigation of the general properties of a positional manifold, that is, of the properties of the general type of numerical addition. It will be found that the idea of a positional manifold will be made more simple and concrete without any loss of generality by identifying it with the general idea of space of any arbitrarily assigned number of dimensions, but excluding all metrical spatial ideas. In the discussion of the general properties of numerical addition this therefore will be the interpretation adopted as being at once the most simple and the most general. All the properties thus deduced must necessarily hold for any special algebra of the genus, though the scale of the relative importance of different properties may vary in different algebras. Positional manifolds are investigated in Book III. Multiplication in algebras of the numerical genus of course follows all the general laws investigated in this chapter. There is also one other general law which holds throughout this genus. The product of fa and rjb (f and 7] being numbers) is defined to be equivalent to the product of grj (ordinary multiplication) into the product of a and b. Thus in symbols fa . 7jb = gqab, r)b. fa = frjba; where the juxtaposition of f and t) always means that they are to be multiplied according to the ordinary law of multiplication for numbers. If this law be combined with equation 10 of 19, the following general
22] CLASSIFICATION OF SPECIAL ALGEBRAS. 31 equation must hold: let eYi e2, ... ev be elements of the manifold, and let Greek letters denote numbers (i.e. ordinary algebraic quantities, real or imaginary), then a2e2 + ... + civev) (/8a + fa2 4-... + fav) #x + ... + avf3vevev. It follows that in the numerical genus of algebras the successive derived manifolds are also positional manifolds, as well as the manifold of the first order. In the classification of the special algebras of this genus the nature of the process of multiplication as it exists in each special algebra is the guide. The first division must be made between those algebras which involve a complete algebraical system of more than one manifold and those which involve only one manifold, that is, between algebras of an order higher than the first and between linear algebras (cf. 20). It is indeed possible to consider all algebras as linear. But this simplification, though it has very high authority, is, according to the theory expounded in this work, fallacious. For it involves treating elements for which addition has no mean- ing as elements of one manifold ; for instance in the Calculus of Extension it involves treating a point element and a linear element as elements of one manifold capable of addition, though such addition is necessarily meaningless. The only known algebra of a species higher than the first is Grassmann's Calculus of Extension; that is to say, this is the only algebra for which this objection to its simplification into a linear algebra holds good. The Calculus of Extension will accordingly be investigated first among the special algebras of the numerical genus. It can be of any species (cf. 20). The general type of manifold of the first algebraic order in which the algebra finds its interpretation will be called an Extensive Manifold. Thus an extensive manifold is also a positional manifold. In Book IV. the fundamental definitions and formulae of the Calculus of Extension will be stated and proved. The calculus will also be applied in this book to an investigation of simple properties of extensive manifolds which, though deduced by the aid of this calculus, belong equally to the more general type of positional manifolds. One type of formulae of the algebra will thus receive investigation. Other types of formulae of the same algebra are developed in Books V., VI. and VII., each type being developed in conjunction with its peculiar interpretation. The series of interpretations will form, as they ought to do, a connected investigation of the general theory of spatial ideas of which the foundation has been laid in the discussion of positional manifolds in Book III. This spatial interpretation, which also applies to the algebra of Symbolic Logic, will in some form or other apply to every special algebra, in so far as interpretation is possible. This fact is interesting and deserves investigation.
32 PRINCIPLES OF UNIVERSAL ALGEBRA. [CHAP. III. 22 The result of it is that a treatise on Universal Algebra is also to some extent a treatise on certain generalized ideas of space. In order to complete this subsidiary investigation an appendix on a mode of arrangement of the axioms of geometry is given at the end of this volume. The second volume of this work will deal with Linear Algebras. In addition to the general theory of their classification and comparison, the special algebras of quaternions and matrices will need detailed development. Note. The discussions of this chapter are largely based on the ' Ueber- sicht der allgemeinen Formenlehre' which forms the introductory chapter to Grassmann's Ausdehnungslehre von 1844. Other discussions of the same subject are to be found in Hamilton's Lectures on Quaternions, Preface; in HankeFs Vorlesungen ilber Complete Zahlen (1867); and in De Morgan's Trigonometry and Double Algebra, also in a series of four papers by De Morgan, ' On the Foundation of Algebra? Transactions of the Cambridge Philosophical Society, vols. vit. and viii., (1839, 1841, 1843, 1844).
BOOK II. THE ALGEBRA OF SYMBOLIC LOGIC. w.
CHAPTER I. The Algebra of Symbolic Logic. 23. Formal Laws. The Algebra of Symbolic Logic* is the only known member of the non-numerical genus of Universal Algebra (cf. Bk. I., Ch. in., 22). It will be convenient to collect the formal laws which define this special algebra before considering the interpretations which can be assigned to the symbols. The algebra is a linear algebra (cf. 20), so that all the terms used belong to the same algebraic scheme and are capable of addition. Let a, b} c, etc. be terms representing elements of the algebraic manifold of this algebra. Then the following symbolic laws hold. (1) The general laws of addition (cf. Bk. I. Ch. in, 14, 15): a 4- b = b -f a, a + b + c = (a + 6) + c = a + (b + c). (2) The special law of addition (cf. 22) : a + a = a. (3) The definition of the null element (cf. 21): a + 0 = a. (4) The general laws of multiplication (cf. 19): c (a + b) = ca + cb, (a + b)c = ac + be. (5) The special laws of multiplication : ab = ba, abc = ab.c = a.bc, aa = a. * This algebra in all essential particulars was invented and perfected by Boole, cf. his work entitled, An Investigation of the Laws of Thought, London, 1854. 3 2
36 THE ALGEBRA OF SYMBOLIC LOGIC. [CHAP. 1. (6) The law of' absorption ': a 4- ab = a. This law includes the special law (2) of addition. (7) The definition of the 'Universe/ This is a special element of the manifold, which will be always denoted in future by i, with the following property: ai = a. (8) Supplementary elements. An element b will be called supple- mentary to an element a if both a -f b = i, and ab = 0. It will be proved that only one element supplementary to a given element can exist; and it will be assumed that one such element always does exist. If a denote the given element, a will denote the supplementary element. Then a will be called the supplement of a. The supplement of an expression in a bracket, such as (a + 6), will be denoted by ~~ (a 4- 6). The theorem that any element a has only one supplement follows from the succeeding fundamental proposition which develops a method of proof of the equivalence of two terms. Proposition I. If the equations xy = xz, and x 4- y = x -f z, hold simul- taneously, then y = z. Multiply the second equation by x, where x is one of the supplements of x which by hypothesis exists. Then x (x + y) = x (x + z). Hence by (4) xx + xy = xx -f xz} hence by (8) and (3) xy = xz. Add this to the first equation, then by (4) (x + x) y = (x + x) z, hence by (8) iy = iz, hence by (7) y = z. Corollary I. There is only one supplement of any element x. For if possible let x and x be two supplements of x. Then xx = 0 = xx, and x -*-x = i = x-f x'. Hence by the proposition, x = x'. Corollary II. If x = y} then x = y. Corollary III. % = 0, and 0 = i. Corollary IV. ^ = x; where x means the supplement of the supplement of x. The proofs of these corollaries can be left to the reader.
24] FORMAL LAWS. 87 Proposition II. (x + y) (x + z) = x + yz. For {x + y) (x + z) = xx + xy + xz + yz = x + x (y + z) + yz = x + yz, by (6). Proposition III. #0 = 0 = Ox, and x + i = i = i + x. The first is proved in 21. The proof of the second follows at once from (6) and (7). 24. Reciprocity between Addition and Multiplication. A reci- procity between addition and multiplication obtains throughout this algebra ; so that corresponding to every proposition respecting the addition and multiplication of terms there is another proposition respecting the multi- plication and addition of terms. The discovery of this reciprocity was first made by C. S. Peirce*; and later independently by Schroder +. The mutual relations between addition and multiplication will be more easily understood if we employ the sign x to represent multiplication. The definitions and fundamental propositions of this calculus can now be arranged thus. The Commutative Laws are (cf. (1) and (5)) x + y = y + x,\ .............................. x x y = y x x.) The Distributive Laws are (cf. (4) and Prop. II.) xx(y + z) = (xxy) + (xx z),\ B x + (yxz) = (x + y)x (x+z).) ..................... The Associative Laws are (cf. (1) and (5)) x + (y + z) = x + y + z,\ ,~ x x (y x z) x x y x z.) The Laws of Absorption are (cf. (6)) x+(xxy) = x = x + x,) ..... ,-p, x x (x + y) = x = x x x.)........................... The properties of the Null element and of the Universe are (cf. (3), (7), and Prop. III.), i x x i = x. The definition of the supplement of a term gives (cf. (8) and Prop. I.) * Proc. of the American Academy of Arts and Sciences, 1867. f Der Operationskreis des Logikkalkiils, 1877.
38 THE ALGEBRA OF SYMBOLIC LOGIC. [CHAP. 1. There can therefore be no distinction in properties between addition and multiplication. All propositions in this calculus are necessarily divisible into pairs of reciprocal propositions; and given one proposition the reciprocal proposition can be immediately deduced from it by interchanging the signs + and x , and the terms i and 0. An independent proof can of course always be found : it will in general be left to the reader. Also any interpretation of which the calculus admits can always be inverted so that the interpretation of addition is assigned to multiplication, and that of multiplication to addition, also that of i to 0 and that of 0 to i. 25. Interpretation. It is desirable before developing the algebraic formulae to possess a simple and general form of interpretation (cf. 7 and 22). Let the elements of this algebraic manifold be regions in space, each region not being necessarily a continuous portion of space. Let any term symbolize the mental act of determining and apprehending the region which it represents. Terms are equivalent when they place the same region before the mind for apprehension. Let the operation of addition be conceived as the act of apprehending in the mind the complete region which comprises and is formed by all the regions represented by the terms added. Thus in addition the symbols represent firstly the act of the mind in apprehending the component regions represented by the added terms and then its act in apprehending the complete region. This last act of apprehension determines the region which the resultant term represents. This interpretation of terms and of addition both satisfies and requires the formal laws (1) and (2) of 23. For the complete region does not depend on the order of apprehension of the com- ponent regions ; nor does it depend on the formation of subsidiary complete regions out of a selection of the added terms. Hence the commutative and associative laws of addition are required. The law, a + a= a, is satisfied since a region is in no sense reduplicated by being placed before the mind repeatedly for apprehension. The complete region represented by a 4- a re- mains the region represented by a. This is called the Law of Unity by Jevons (cf. Pure Logic, ch. vi). The null element must be interpreted as denoting the non-existence of a region. Thus if a term represent the null element, it symbolizes that the mind after apprehending the component regions (if there be such) symbolized by the term, further apprehends that the region placed by the term before the mind for apprehension does not exist. It may be noted that the addition of terms which are not null cannot result in a null term. A null term can however arise in the multiplication of terms which are not null. Let the multiplication of terms result in a term which represents the entire region common to the terms multiplied. Thus xyz represents the
25, 26] INTERPRETATION. 39 entire region which is at once incident in the regions x and y and z. Hence the term xy symbolizes the mental acts first of apprehending the regions symbolized by x and y, and then of apprehending the region which is their complete intersection. This final act of apprehension determines the region which xy represents. This interpretation of multiplication both satisfies and requires the distributive law, numbered (4) in 23, and the commutative and associative laws marked (5) in 23. The law, aa = a, which also occurs in (5) of 23 is satisfied; for the region which is the complete intersection of the region a with itself is again the region a. This is called the Law of Simplicity by Jevons (cf. loc, cit.). The Law of Absorption (cf. (6) 23) is also required and satisfied. For the complete region both formed by and comprising the regions a and ah is the region a, and the final act of apprehension symbolized by a + ab is that of the region a. Hence This interpretation also requires that if x -f y = xt then y = xy. And this proposition can be shown to follow from the formal laws (cf. 26, Prop. vin.). The element called the Universe (cf. 23 (7)), must be identified with all space; or if discourse is limited to an assigned portion of space which is to comprise all the regions mentioned, then the Universe is to be that complete region of space. The term supplementary (cf. 23 (8)) to any term a represents that region which includes all the Universe with the exception of the region a. The two regions together make up the Universe; but they do not overlap, so that their region of intersection is non-existent. It follows that the supplement of the Universe is a non-existent region, and that the supplement of a non-existent region is the Universe (cf. Prop. I. Cor. 3). 26. Elementary Propositions. The following propositions of which the interpretation is obvious can be deduced from the formal laws and from the propositions already stated. Proposition IV. If x + y = 0, then x = 0, y = 0. For multiplying by x x (x + y) = 0. But x(x + y) = xf by (6) 23. Hence x = 0. Similarly, y = 0. The reciprocal theorem is, if xy = i, then x = i, y = i. Proposition V. x+y=x+ yx, and xy = x(y + x). For x + y = x + y (x -t- x) = x -f yx + yx = x + y%- The second part is the reciprocal proposition to the first part.
40 THE ALGEBRA OF SYMBOLIC LOGIC. [CHAP. I. Proposition VI. ~ (xy) = x + y, and ~ (x + y) = xy. For by Prop. V., x + y = x -f xy = x + xy. Hence xy + (x + y) = xy + xy + x = x (y + y) +x Also #y (x 4- /) = ##2/ + #^y = 0. Hence by 23 (8) x + y = - (xy). The second part is the reciprocal of the first part. Also it can be deduced from the first part thus : Taking the supplements of both sides " O + y) = = (xy) = xy. Corollary. The supplement of any complex expression is found by replacing each component term by its supplement and by interchanging + and x throughout. Proposition VII. If xy = xz, then xy = xz, and x + y = x -f z. For taking the supplement of both sides of the given equation, by Prop. VI., x + y = x + z. Multiplying by x, xy = xz. Again taking the supplement of this equation, then x + y = x + z. The reciprocal proposition is, if x + y = x + z, then x + y = x + z, and xy = xz. Proposition VIII. The following equations are equivalent, so that from any one the remainder can be derived: y = xy} x-\-y=x, xy = 0, x + y = i Firstly : assume y = xy. Then x + y = x + xy = x. And xy = xxy = 0. And x + y = x -f-~ (xy) =x-t-x + y = i. Secondly : assume x + y = x. Then Xy = (x+y)y = y. Hence the other two equations can be derived as in the first case. Thirdly: assume xy = 0.
27] ELEMENTARY PROPOSITIONS. 41 Then y = (x+x)y = xy-\-xy = xy. Hence the other two equations can be derived. Fourthly : assume x + y = i. Then taking the supplements of both sides Hence by the third case the other equations are true. Corollary. By taking the supplements of the first and second equations two other forms equivalent to the preceding can be derived, namely Proposition IX. If x = xyz, then x = xy = xz, and if x = x -f y + z, then x = x + y = x + z. For xy =xy (z -f 0) = xyz + xyz = x + xyz = xy from (6) 23. The second part of the proposition is the reciprocal theorem to the first part. Corollary. A similar proof shews that if z=z(xu+yv)y then z z(x-\-y)\ and that if z = z + (x + u) (y + v)y then z = z + xy. 27. Classification. The expression x -f y + z -f ..., which we can denote by u for the sake of brevity, is formed by the addition of the regions x, y, etc. Now these regions may be overlapping regions: we re- quire to express it as a sum of regions which have no common part. To this problem there exists the reciprocal problem, given that it stands for the product xyz..., to express u as a product of regions such that the sum of any two completes the universe. These problems may be enunciated and proved symbolically as follows. Proposition X. To express u(=x + y + z+ ...), in the form X+Y+Z+...; where X, F, Z have the property that for any two of them, Y and Z say, the condition YZ = 0, holds. Also to express u (= xyz...) in the form XYZ...; where X, F, Z have the property that for any two of them, F and Z say, the condition Y+Z = i, holds. Now from Prop. IV., if x (y + z) = 0, then xy = 0, xz = 0. Hence for the first part of the proposition the conditions that X, F, Z, etc. must satisfy can be expressed thus, X(F+^+T+...)=0, Y(Z+T+...) = (), Z(T+...) = 0yetc. Now by Prop. V.; u = x 4- y + z + ...
42 THE ALGEBRA OF SYMBOLIC LOGIC. [CHAP. i. and y + z + t+...=y+y(z + t + ...); and z + t + ... = z +z(t + ...) Proceeding in this way, we find u = x -f xy + ^^ + xyzt + ... . Hence we may write X = x, Y = xy, Z = xyz, etc. It is obvious that there is more than one solution of the problem. Again for the second part of the proposition, consider u=x+y+z+ ... , By the first part of the proposition, u = x + xy + xyz + #*/2 + .... Here any two terms, Y and Z, satisfy the condition YZ = 0. Taking the supplements of these equations, = xyz. = x (x + y) (x + y + 5) ( + 2/ + ^ + t).... Hence we may write XYZ... for xyz..., where X = x} Y = Z=x + y + z, etc. and any two of X, F, Z, etc., for instance Fand Z, satisfy the condition F-f Z = i. It is obvious that there is more than one solution of this problem. These problems are of some importance in the logical applications of the algebra. 28. Incident Regions. The symbolic study of regions incident (cf. 10) in other regions has some analogies to the theory of inequalities in ordinary algebra. These relationships also partly possess the properties of algebraic equations. Two mixed symbols have therefore been adopted to express them, namely ^ and =^ (cf. Schroder, Algebra der Logik). Then, y^x, expresses that y is incident in x; and x ^ y expresses that x contains y. Expressions of this kind will be called, borrowing a term from Logic, subsumptions. Then a subsumption has analogous properties to an inequality. The Theory of Symbolic Logic has been deduced by C. S. Peirce from the type of relation symbolized by ^=, cf. American Journal of Mathematics, Vols. ill and VII (1880, 1885). His investigations are incorporated in Schroder's Algebra der Logik. In order to deduce the properties of a subsumption as far as possible purely symbolically by the methods of this algebra, it is necessary to start from a proposition connecting subsumptions with equations. Such an initial proposition must be established by considering the meaning of a subsumption.
28] INCIDENT REGIONS. 43 Proposition XI. If y 4 # then y = xy; and conversely. For if y be incident in x, then y and xy denote the same region. The converse of this proposition is also obvious. It is obvious that any one of the equations proved in 26, Prop. VIIL, to be equivalent to y = xy is equivalent to y 4 x. In fact the subsumption y 4 x may be considered as the general expression for that relation between x and y which is implied by any one of the equations of Prop. VIIL It follows that an equation is a particular case of a subsumption. Corollary, xz 4 x 4 x + z. Proposition XII. If x $ y, and y $ z ; then x 4= z. For by Prop. XL and by 26, Prop. IX. z = zy = zxy = zx. Hence x ^ z. Proposition XIII. If x ^ y, and y $ x ; then x = y. For since x^y then y = xy. And since V^x then y = x + y. H ence y = xy = x (x -f y) = x. Proposition XIV. If x$y, and u^v\ then ux ^ vy, and u + x^ v + y. For y = y#, and y = vu ; hence vy = yxvu = vy . xu. Therefore ux ^ ^y. Also x = ^ + y, and w = u + v; hence a? + i* = 4-y + t* + v = ( +'w) + (y + v). Therefore x + u^y + v. Corollary. If x ^ y, and a = v ; then -?w? ^ vy, and + x ^ v 4- y. For v = v^, and i = ^ + v ; hence the proof can proceed exactly as in the proposition. The proofs of the following propositions may be left to the reader. Proposition XV. If x$y, then y^x. Proposition XVI. If z 4 xy, then z 4 #, ^42/? ^ 4 ^' + ?/ Proposition XVII. If s 4 XH tnen *2/ 4 ^ ^ + ^ 4 ^ Proposition XVIII. If z$x + y, then 2^^, Proposition XIX. If z ^ a? + y, then xy^z, Proposition XX. If xz 4 y, and x^y + z, then a? 4 V-
44 THE ALGEBRA OF SYMBOLIC LOGIC. [CHAP. I. 28. Proposition XXI. If z4xu + yv, then z^x + y. The importance of Prop. XXI. demands that its proof be given. By Prop. IX. Cor., z = z (xu + yv) =z (x + y). Therefore z = x + y. Corollary. If z = xu + yv, then # ^ a? + y; that is, a?w + yv ^ x 4- y. Prop.* XXII. If * (# + m) (y + v), then ^ ^ ary. Corollary, (a? + m) (y + v) ^ a?y. * This proposition, which I had overlooked, was pointed out by Mr W. E. Johnson.
CHAPTER II. The Algebra of Symbolic Logic (continued), 29. Development. (1) The expression for any region whatsoever may be written in the form ax + bx; where x represents any region. For let z be any region. Now x + x = i. Hence z = % + x = zx + zx. Now let a = zx + iixy and b = zx + vx, where u and v are restricted by no conditions. Then ax + bx = (zx + ux) x -f (zx + vx)x = zx + zx = z. Hence by properly choosing a and 6, ax + bx can be made to represent any region z without imposing any condition on x. Again the expression for any region can be written in the form (a + x)(b + x\ where x represents any other region. For by multiplication (a + x) (b + x) ab + ax -f bx = (a + ab) x + (b + ab) x = ax + bx. This last expression has just been proved to represent the most general region as far as its relation to the term x is concerned. (2) Binomial expressions of the form ax + bx have many important properties which must be studied. It is well to notice at once the follow- ing transformations: ax + bx = (b + x) (a + x); ~ (ax + bx) ={a + x)(b-\- x) (ax -f bx) (ex + dx) = acx + bdx; ax 4- bx + c = (a + c) x + (b + c) ^; ab.
46 THE ALGEBRA OF SYMBOLIC LOGIC. [CHAP. II. (3) Let f{%) stand for any complex expression formed according to the processes of this algebra by successive multiplications and additions of x and x and other terms denoting other regions. Theny*(#) denotes some region with a specified relation to x. But by (1) of this article fix) can also be written in the form ax + bx. Furthermore a and b can be regarded as specified by multiplications and additions of the other terms involved in the formation of f(x) without mention of x. For if a be a complex expression, it must be expressible, by a continual use of the distributive law, as a sum of products of which each product either involves x or x or neither. Since a only appears when multiplied by x, any of these products involving x as a factor can be rejected, since xx = 0 ; also any of these products involving x as a factor can be written with the omission of x, since xx ^= x. Hence a can be written in a form not containing x or x. Similarly b can be written in such a form. (4) Boole has shown how to deduce immediately from f(x) appropriate forms for a and b. For write f(x) = ax -f bx. Let i be substituted for x, then f(i) = ai + bl = ai + = a. Again let 0 be substituted for x, then Hence f(x) =f(i) x +/(0) x. For complicated expressions the rule expressed by this equation shortens the process of simplification. This process is called by Boole the develop- ment oif(x) with respect to x. The reciprocity between multiplication and addition gives the reciprocal rule (5) The expressions f[i) and f(0) may involve other letters y, z, etc. They may be developed in respect to these letters also. Consider for example the expression f(x, y, z) involving three letters. f(x, y, z) =f(i, y, z)x +/(0, y, z) x, f{i, i, z)=f{i, i, i)z+f{i, i, 0)0, /(*, 0, *)=/(*, 0, i)z+f{i, 0, 0)1, /(0, y, z) =/((), i, z) y +/(0, 0, z) % /(0, , z) =/((), i, i) z +/(0, i, 0) 0, /(0, 0, z)=f(0, 0, i)z +/(0, 0, 0) z. Hence by substitution f(w, y, z) =f(i, i, i) xyz +/(0, i, i) xyz +/(i, 0, i) xyz +/(i, i, 0) xyz i, 0, 0) xyz +/(0, i, 0) ^ +/(0, 0, i) xyz +/(0, 0, 0) xyz,
30] DEVELOPMENT. 47 The reciprocal formula, owing to the brackets necessary, becomes too complicated to be written down here. Let any term in the above developed expression for f(x, y, z)y say f(0, iy 0) xy~zy be called a constituent term of the type xyz in the develop- ment. (6) The rule for the supplement of a binomial expression given in subsection (1) of this article, namely ~ {ax + bx) = ax -f bx, can be extended to an expression developed with respect to any number of terms xy yy zy .... The extended rule is that if f{x,y,z, ...) = axyz... + ...+gxyz...y then ~f{xyy,zy ...) = dxyz ... + ... + gxyz.... In applying this proposition any absent constituent term must be replaced with 0 as its coefficient and any constituent term with the form xyz... must be written ixyz... so that i is its coefficient. For assume that the rule is true for n terms x, y, z... and let t be an (w + l)th term. Then developing with respect to the n terms x, y, z,... f(x, yy z, ... t) = Axyz...+ ... + Gxyz..., where the products such as xyz ..., ..., xyz ... do not involve t, and A =at + a't,..., G=gt + g't. Then the letters a, a, ..., gy gr are the coefficients of.constituent terms of the expression as developed with respect to the n+^rterms xy yy zy...t. Hence by the assumption -f(x,yyz,...t) = Axyz... + ... + Gxyz.... But by the rule already proved for one term, Hence the rule holds for (n + l)th terms. But the rule has been proved for one term. Thus it is true always. 30. Elimination. (1) The object of elimination may be stated thus: Given an equation or a subsumption involving certain terms among others, to find what equations or subsumptions can be deduced not involving those terms. The leading propositions in elimination are Propositions XXI. and XXII. of the last chapter, namely that, if z ^ xu 4- yvy then z^x + y\ and if z ^ (x + u) (y + v)y then z^xy; and their Corollaries that, xu + yv = x + yy and, (2) To prove that if ax + bx = ci then a + b ^ c ^ ab. Eliminating x and x by the above-mentioned proposition from the equation c = ax + bx, c^
48 THE ALGEBRA OF SYMBOLIC LOGIC. [CHAP. II. Also taking the supplementary equation, c = ax + bx. Hence from above c ^ a + b. Taking the supplementary subsumption, ab = ~ (a + b) ** Therefore finally a + b ^ c ^ ab. The second part can also be proved* by taking the reciprocal equation, c = (a 4- x) (b + x), and by using Prop. XXII. Corollary. The same subsumptions, written in the supplementary form, are a -h b ^= c 4= ab- (3) By Prop. XI. each of these subsumptions is equivalent to an equation, which by Prop. VIII. can be put into many equivalent forms. Thus a + b^c, can be written c = c (a + b); and this is equivalent to abc = 0, And c ^ ab, can be written ab = abc; and this is equivalent to abc = 0. (4) Conversely, if c ^ a + b,\ } then it has to be shown that we can write ax + bx = c; where we have to determine the conditions that x must fulfil. This problem amounts to proving that the equation ax + bx=c has a solution, when the requisite conditions between a, b, c are fulfilled. The solution of the problem is given in the next article (cf. 31 (9)). The equation ax + bx = c includes a number of subsidiary equations. For instance ax = ex ; thence a + x = c + x, and thence ax = ex. Similarly bx ex, and bx = ex. The solution of the given equation will satisfy identi- cally all these subsidiary equations. (5) Particular Cases. There are two important particular cases of this equation, when c = i, and when c = 0. Firstly, c = i. Then ax + bx = i. * Pointed out to me by Mr W. E. Johnson.
30] ELIMINATION. 49 Hence a + b ^= i. But the only possible case of this subsumption is a + b = i. Also ab = if which is necessarily true. Therefore finally, a + b = i, is the sole deduction independent of x. Secondly, c = 0. Then ax + bx = 0. Hence a + b ^ 0, which is necessarily true. Also ab = 0. But the only possible case of this subsumption is ab = 0. Therefore finally, ab = 0, is the only deduction independent of x. If the equation be written f(x) = 0, the result of the elimination becomes These particular cases include each other. For if ax + bx = i, then ~ (ax + bx) = 0, that is ax +bx = 0. And a + b = i is equivalent to ab = 0. (6) General Equation. The general form j (x) = |r (x), where f (x) and yfr (x) are defined in the same way as f(x) in 29 (3), can be reduced to these cases. For this equation is equivalent to (f (x)f(x) + 4 (x)yfr(x) = 0. This is easily proved by noticing that the derived equation implies that is f (x) that is $ (x) = sfr (x). Hence the equation f (x) = sjr (x) can be written by 29 (4) in the form \4 (i)^r(i) + $ (i) * (t)} x + { / (0)^(0) + (0) f (0)} x = 0. Hence the result of eliminating x from the general equation is {$ ( )+(i) + (t) ^ (0} f * (0) t (0) + * (0) * ( )1 = - This equation includes the four equations * (i) ) (0) f ( )+(0) = o, ^ (i) $ (0) ^ (t) ^ (0) = o, 4 (i) $ (0) t ( ^ (0) = o, $ (i) $ (0) t ( ) ^ (0) = o. The reduction of the general equation to the form with the right-hand side null is however often very cumbrous. It is best to take as the standard form ax + bx = ex + dx ..............................(1). This form reduces to the form, ax + bx = c, when c = d. For ex + dx = c(x + x) = c. W. 4
50 THE ALGEBRA OF SYMBOLIC LOGIC. [CHAP. II. The equation is equivalent to the two simultaneous equations ax = ex, bx = dx] as may be seen by multiplying the given equation respectively by x and x. Let the equation ax = cx be called the positive constituent equation, and the equation bx = dx be called the negative constituent equation of equation (1). Taking the supplements a + x = c + x, b + x = d + x. Hence multiplying by x and x respectively dx = ex, and bx = dx. So equation (1) can also be written ax -f bx = ex + dx; and the two supplementary forms give ax + bx = ex + dx, ax + bx = cx + dx. The elimination of x can also be conducted thus. Put each side of equation (1) equal to z. Then ax + bx = zt ex -f dx z. Hence the following subsumptions hold, a + b ^= z ^ ab; c -f d ^= z ^= cd. Therefore Also similarly from the form, ax + bx = ex + dx, we find the subsumptions a + b^cd, c + d ^= ab. These four subsumptions contain (cf. 31 (9), below) the complete result of eliminating x from the given equation (1). The two supplementary forms give the same subsumptions, only in their supplementary forms, but in- volving no fresh information. From these four subsumptions it follows that, abed = abed = abed = abed = 0. These are obviously the four equations found by the other method, only written in a different notation.
30] ELIMINATION. 51 From these equations the original subsumptions can be deduced. For abcd = 0 can be written and therefore ab = c + d. Similarly for the other subsumptions. Also it can easily be seen that the four subsumptions can be replaced by the more symmetrical subsumptions, which can be expressed thus, (The sum of any two coefficients, one from each constituent equation) ^= (The product of the other two). (7) Discriminants, All these conditions and (as it will be shown) the solution of the equation can be expressed compendiously by means of certain functions of the coefficients which will be called the Discriminants of the equation. The discriminant of ax = ex is ac + ac. Let it be denoted by A. The discriminant of bx = dx is bd + bd. Let it be denoted by B. Then A and B will respectively be called the positive and negative discriminants of the equation ax + bx = ex + dx ..............................(1). Now A = ~ (ac -f ac) = ac + ac, and B = - (bd + bd) = bd + bd. Therefore, remembering that y + z = 0 involves y = 0, z = 0, it follows that all the conditions between the coefficients a, 6, c, d can be expressed in the form This equation will be called the resultant* of the equation ax + bx = cx + dx. It can be put into the following forms, A^B, B^A, AB = A, BA = B, A+B = B} B+A=A. It is shown below in 31 (9) that the resultant includes every equation between the coefficients and not containing x which can be deduced from equation (1). The equation ax + bx = ex + dx when written with its right-hand side null takes the form (8) Again let there be n simultaneous equations a1x=c1x a^c=c^cy..., cnx} and let Al9 A2, ... An be the discriminants of the successive equations respectively; then their product AXA2... An is called the resultant * Cf. Schroder, Algebra der Logik, 21. 4 2
52 THE ALGEBRA OF SYMBOLIC LOGIC. [CHAP. II. discriminant of the n equations. It will be denoted by II (Ar), or more shortly A. Similarly let there be n simultaneous equations b\X = djX J)2x = d%x,... bnx d^x; and let Bu B2...Bn be the discriminants. Then BjB2... Bn is the resultant discriminant of the n equations. It will be called n (Br) or B. The n equations a^x + bjx = cYx + djX, b$c = c?x -f d^x, anx + bnx = cnx -f dnx, involve the 2n equations just mentioned and conversely. The functions A and B are called the positive and negative resultant discriminants of these equations. Now A =Aj + A2 + ... + Any 5 = ^+ 52H- ... + Bn. Hence AB = 2 ArB8. Now any equation djX = CrX may be joined with any equation bgX = dgX to form the equation a^x + bsx = c^ + d^c. Hence all the relations between the coefficients are included in all the equations of the type, ArB8 = 0. But these equations are all expressed by the equation 25=0. This equation may therefore conveniently be called the resultant of the n equations. This is the complete solution of the problem of the elimination of a single letter which satisfies any number of equations. The single equation is equivalent to the n given equations. It must be carefully noticed that in this algebra the distinctions of procedure, which exist in ordinary algebra according to the number of equations given, do not exist. For here one equation can always be found which is equivalent to a set of equations, and conversely a set of equations can be found which are equivalent to one equation. (9) More than one Unknown. The general equation involving two un- knowns, x and yy is of the type axy + bxy + cxy + dxy = exy +fxy + gxy + hxy. This equation is equivalent to the separate constituent equations, axy = exyy bxy ~fxyy etc. Let a constituent equation involving x (as distinct from x) be called a constituent positive with respect to x, and
30] ELIMINATION. 53 let a constituent equation involving x (as distinct from x) be called a constituent negative with respect to x. Thus, axy = exy is positive with respect both to x and y; bxy =fxy, is positive with respect to x. negative with respect to y, and so on. Let A, B, Gy 1) stand for the discriminants of these constituents. Thus A = ae + ae, B = bf-\-b/ C=cg + cg, D = dh + dh. Then the discriminant A is called the discriminant positive with respect to x and y, B is the discriminant positive with respect to x and negative with respect to y, and so on. The equation can be written in the form (ay + by)x + (cy + dy)x = (ey +fy) x -f (gy + hy) x. If we regard x as the only unknown, the positive discriminant is (ay + by) (ey +fy) + (ay + by) (ey +fy), that is Ay -f By. The negative discriminant is Gy + By. The resultant is (Ay + By)(Cy+ T)y) = 0; that is A Gy + BDy = 0. This is the equation satisfied by y when x is eliminated. It will be noticed that A and G are the discriminants of the given equation positive with respect to y, and B and D are the discriminants negative with respect toy. Similarly the equation satisfied by x when y is eliminated is ABx+ GDx = 0. The resultant of either of these two equations is ABGD = 0. This is therefore the resultant of the original equation. The original equation when written with its right-hand side null takes the form Axy + ~Banj + Gxy + Dxy = 0........................(1). Again suppose there are n simultaneous equations of the above type the coefficients of which are distinguished by suffixes 1, 2, ... n. Then it may be shown just as in the case of a single unknown x, that all equations of the type, ApBqGrDs = 0, hold. Hence if A stand for A^2... An, and B for B B2... Bn, G for G-fi^... Cn, and D for DiD2... Dn, the resultant of the equations is ABGD = 0. The n equations can be replaced by a single equation of the same form as (1) above. Also the equation satisfied by x, after eliminating y only is ABx+GDx=0}
54 THE ALGEBRA OF SYMBOLIC LOGIC. [CHAP. II. where A and B are the positive discriminants with respect to x and C and D are the negative discriminants. The equation satisfied by y is where a similar remark holds. (10) This formula can be extended by induction to equations involving any number of unknowns. For the sake of conciseness of statement we will only give the extension from two unknowns to three unknowns, though the reasoning is perfectly general. The general equation for three unknowns can be written in the form (az + a'z) xy -f (bz + b'z) xy + (cz + c'z) xy + (dz + d'z) xy = (ez + ez) xy + (fz +f'z) xy + (gz + g'z) xy + (hz + Kz) xy. Then, if A = ae + ae, A1 = ae + a'e', and so on, A is the discriminant positive with respect to x, y, and z} and A' is the discriminant positive with respect to x and y, but negative with respect to z; and so on. If x and y be regarded as the only unknowns, then the two discriminants positive with respect to x are (az + a'z) (ez + e'z) + (az + afz) (ez + e'z), and (bz + b'z) (fz +fz) + (bz + b'z) (fz +f*z\ that is, Az + A'z, and Bz + B'z. Similarly the two discriminants negative with respect to x are Gz + Cz, and Dz + D'z. Hence the equation for x after eliminating y is (Az + A'z) (Bz + B'z)x + (Cz + G'z)(Dz + D'z)x= 0, that is (ZB* + J'2?'0) 0 + (CDz + 07y ) = 0. The result of eliminating z from this equation is AA'BB'x + GG'DD'x = 0. Hence the equation for x after eliminating the other unknowns is of the form, Px + Qx = 0, where P is the product of the supplements of the discri- minants positive with respect to x, and Q is the product of the supplements of the discriminants negative with respect to x. The resultant of the whole equation is AA'BB'CG'DD' = Q, that is the product of the supplements of the discriminants is zero. The given equation when written with its right-hand side null takes the form Axyz + A'xyz + Bxyz + B'xyz + Gxyz + G'xyz + T)xyz + D'xyz = 0. The same formulae hold for any number of equations with any number of variables, if resultant discriminants are substituted for the discriminants of a single equation.
31] ELIMINATION. 55 (11) It is often convenient to notice that if ${x,y,z, ...)=^tey,s, ...), be an equation involving any number of variables, then any discriminant is of the form i, i, i, . ,o,o,. where i is substituted for each of the unknowns with respect to which the discriminant is positive and 0 is substituted for each of the unknowns with respect to which the discriminant is negative. (12) The formula for the elimination of some of the unknowns, say, u, v, w,..., from an equation involving any number of unknowns, x, y, z,... uy v, w,..., can easily be given. For example, consider only four unknowns, x, y, z, t, and let it be desired to eliminate z and t from this equation, so that a resultant involving only x and y is left. Let any discriminant of the 000 0/' wnere e*tner i or 0 is t0 ^e written ac- cording to the rule of subsection (11). The equation can be written {D (i, i, i, i) xyz + D (i, i, 0, i) xyz +1) (i, 0, i, i) xyz + D (0, i, i, i) xyz + 5(iy 0, 0, i) xyz + D (0, i, 0, i) xyz + D (0, 0, i, i) xyz + JD (0, 0, 0, i) xyz} t + [D (i, i, i, 0)ic^ + 5(i, i, 0, O)xyz+...}t = 0. Hence eliminating ty the resultant is 5 (i, i, i, i) D (i, i, i, 0) xyz + D (i, % 0, i) D (i, iy 0, 0) xyz + 2)(i, 0, i, i)5(i, 0, t, 0)^+ ... +5(0,0, 0, i)5(0, 0, 0,0)ap = 0. Again eliminating z by the same method, the resultant is D (t, , t, i) 5 ($, i, , 0) 5 (*, t, 0, t) B (i, t, 0, 0) ^ + 5 (t, 0, t, t) D (i, 0, i, 0) 5 (t, 0, 0, i) D (i 0, 0, 0) xy + 5 (0, %, i, t) 5 (0, i, iy 0) D (0, i, 0, i) D (0, t, 0, 0) xy + 5(0, 0, i t)B(0, 0, i, 0)5(0, 0, 0, t)5(0, 0, 0, 0)xy = 0. It is evident from the mode of deduction that the same type of formula holds for any number of unknowns. 31. Solution of equations with one unknown. (1) The solutions of equations will be found to be of the form of sums of definite regions together with sums of undetermined portions of other definite regions; for example to be of the form a + v1b + v2c, where a, 6, c are defined regions and vlt v2 are entirely arbitrary, including i or 0. Now it is to be remarked that u (b + c), where u is arbitrary, is as
56 THE ALGEBRA OF SYMBOLIC LOGIC. [CHAP. II. general as vj) -f v2c. For writing u = v1b + v2(b f 6c), which is allowable since u is entirely arbitrary, then u (b + c) = [vj + v.2 (6 + 6c)} {b + c] = vj + v2 (be + 6c) = vj) + v2c. Hence it will always be sufficient to use the form u (b + c), unless vx and v2 are connected by some condition in which case vj) + v^c may be less general than u (6 + c). (2) ax = ex. Then by 30, (7) Ax = 0. Hence by 26, Prop. VIII. x = Ax. But instead of x on the right-hand side of this last equation, (x + vA) may be substituted, where v is subject to no restriction. But the only restriction to which x is subjected by this equation is that it must be incident in A. Hence x + vA is perfectly arbitrary. Thus finally x = vA ; where v is arbitrary. (3) bx = dx. From subsection (2) ; x = uB. Hence cc=u + B. (4) ax + bx = ex + dx; where AB = 0. From the equation ax = ex, it follows that x = uA ; and from bx = dx, that x = v + B. Hence uA = v + B. Therefore v A = 0. Hence v = wA. Finally, x = B + wA ; where w is arbitrary. This solution can be put into a more symmetrical form, remembering that B + A=A. For x = B (w + w) + wA= Bw +w(A + B) = wA + wT$. Hence the solution can be written x = B + wA,\ x=A +wB.)
31] SOLUTION OF EQUATIONS WITH ONE UNKNOWN. 57 Or x = wA -f x wA +wB.) The first form of solution has the advantage of showing at a glance the terms definitely given and those only given with an undetermined factor. (5) To sum up the preceding results in another form: the condition that the equations ax = ex, bx = dx may be treated as simultaneous is The solution which satisfies both equations is The solution which satisfies the first and not necessarily the second is x uA. The solution which satisfies the second and not necessarily the first is x = uB, that is x = u + B. In all these cases u is quite undetermined and subject to no limitation. (6) The case ax + bx = c, is deduced from the preceding by putting d = c. Then A =ac + ac, B bc-\-bc. The solutions retain the same form as in the general case. The relations between a, b, c are all included in the two subsumptions The case ax -f bx = 0 is found by putting c = d = 0. The equation can be written ax + bx = 0# -f O#j. The positive discriminant is x0 + aO, that is a, the negative is b. The resultant is ab = 0. The solution is x = b + uci. (7) The solution for n simultaneous equations can be found with equal ease. Let x satisfy the n equations anx+bnx = cnx+dnx. Then x satisfies the two groups of n equations each, namely aYx = Ci#, a^x = c^x ... anx = cnx; and b^c = djjc, b jx = d^c ... bnx = dnx.
58 THE ALGEBRA OF SYMBOLIC LOGIC. [CHAP. II. From the first group x = uxAx = u2A2 = ... = unAn. Hence x = uA ; where u is not conditioned. Similarly from the second group x = v^ = v2B2 = ... = vnBn. Hence x = vB x = v + B; where v is not conditioned. Therefore uA = v 4- i?. Hence vJ. = 0, that is v = wJ.. So finally the solution of the n equations is x = B + wA = wA + wB,} x = A + wB = w A + wB j The group axx = CjX, a^x c^, etc. can always be treated as simultaneous, and so can the group of typical form b^c = d^x. The condition that the two groups can be treated as simultaneous is AB = 0. (8) It has been proved that the solution B + uA satisfies the equation, Ax+ Bx=0, without imposing any restriction on u. It has now to be proved that any solution of the equation can be represented by B + uA, when u has some definite value assigned to it. For if some solution cannot be written in this form, it must be capable of being expressed in the form mB + wA + nAB. But A x = 0, and AB = 0, hence, by substituting for x its assumed form, nAB = 0. Thus the last term can be omitted. Again, Bx=0; and AB = 0, hence B(m + B)(w + A) = 0; that is mwB = 0. Hence m = p (w + B), and therefore m=p +wB. Therefore the solution becomes x = mB + wA=(p + wB)B +w(A+B), = pB + B + wA = B + wA, Thus the original form contains all the solutions. (9) To prove that the resultant AB = 0, includes all the equations to be found by eliminating x from For x = B + wA satisfies the equation on the assumption that AB = 0, and without any other condition. Hence AB is the complete resultant. It easily follows that for more than one unknown the resultants found in 30 are the complete resultants.
32] SOLUTION OF EQUATIONS WITH ONE UNKNOWN. 59 (10) Subsumptions of the general type ax + bx^ cx+ dx can be treated as particular cases of equations. For the subsumption is equivalent to the equation ex + dx = (ex -f- dx) (ax + bx) = acx + bdx. Hence A = ac + c~ (ac) = ac + c = a + c, A = c (a + c) = ac, Therefore the resultant A B = 0 is equivalent to abed = 0. This is the only relation between the coefficients to be found by eliminating x. The given subsumption is equivalent to the two subsumptions ax $ ex, bx 4= dx; that is, to the two equations ex = acxt dx = bdx. The solution of ax ^ ex is x uA =u(a + c). The solution of bx ^ dx is x = u -f B u + bd. The solution of ax + bx$cx + dx is x = B + uA =bd + u(a + c) = uB + uA = u (a + c) + ul d. The case of n subsumptions of the general type with any number of unknowns can be treated in exactly the same way as a special type of equation. 32. On Limiting and Unlimiting Equations. (1) An equation $ (xy yy zt ... t) = \{r(x, y, z, ... t) involving the n unknowns x} y, z, ... t is called unlimiting with respect to any of its unknowns (x say), if any arbitrarily assigned value of x can be substituted in it and the equation can be satisfied by solving for the remaining unknowns y, z, ... t\ otherwise the equation is called limiting with respect to x. The equation is unlimiting with respect to a set of its variables x, y, z, ..., if the above property is true for each one of the unknowns of the set. The equation is unlimiting with respect to all its unknowns, if the above is true for each one of its unknowns. Such an equation is called an unlimiting equation. The equation is unlimiting with respect to a set of its unknowns simultaneously, if arbitrary values of each of the set of unknowns can be simultaneously substituted in the equation.
60 THE ALGEBRA OF SYMBOLIC LOGIC. [CHAP. II. It is obvious that an equation cannot be unlimiting with respect to all its unknowns simultaneously, unless it be an identity. (2) The condition that any equation is unlimiting with respect to an unknown x is found from 30 (10). For let P be the product of the supple- ments of the discriminants positive with respect to x and Q be the product of the supplements of the discriminants negative with respect to x. Then the equation limiting the arbitrary choice of x is, Px -f Qx = 0. Hence if the given equation be unlimiting with respect to x, the equation just found must be an identity. Hence P = 0, Q = 0. (3) The condition that the equation be unlimiting with respect to a set of its unknowns is that the corresponding condition hold for each variable. (4) The condition that the equation is unlimiting with respect to a set x, y, z, ... of its unknowns simultaneously is that the equation found after eliminating the remaining unknowns t,uyvy ... should be an identity. The conditions are found by reference to 30 (12) to be that each product of supplements of all the discriminants of the same denomination (positive or negative) with respect to each unknown of the set, but not necessarily of the same denomination for different unknowns of the set, vanishes. (5) Every equation can be transformed into an unlimiting equation. For let the equation involve the unknowns x, y, z, ... t: and let the resultant of the elimination of all the unknowns except x be, Px + Qx = 0. Then x= Q 4- uP, and if u be assigned any value without restriction, then x will assume a suitable value which may be substituted in the equation previous to solving for the other unknowns. Thus if all the equations of the type Px +Qx = 0, be solved, and the original equation be transformed by substitution of, x=Q+uP, y=S + vR, etc., then the new equation between u, v, ... is unlimiting. (6) The field of an unknown which appears in an equation is the collection of values, any one of which can be assigned to the unknown consistently with the solution of the equation. If the equation be un- limiting with respect to an unknown, the field of that unknown is said to be unlimited; otherwise the field is said to be limited. Let the unknown be x, and with the notation of subsection (5), let the resultant after eliminating the other unknowns be Px + Qx = 0. Then x = Q -f uP. Hence the field of x is the collection of values found by substituting all possible values for u, including i and 0. Thus every member of the field of x contains Q; and P contains every member of the field, since PQ = 0. The field of x will be said to have the minimum extension Q and the maximum extension P. 33. On the Fields of Expressions. (1) Definition. The 'field of the expression f (x, y, z, ... t)' will be used to denote the collection of
33] ON THE FIELDS OF EXPRESSIONS. 61 values which the expression f (x, y, z, ... t) can be made to assume by different choices of the unknowns x, y, z, ... t. If p (x, yy z, ... t) can be made to assume any assigned value by a proper choice of x, y, z, ... t, then the field of f (x, y, z,... i) will be said to be unlimited. But if p(x, y, z,... t) cannot by any choice of xf y, z,... t, be made to assume some values, then the field of (j (x, y, z, ... t) will be said to be limited. (2) To prove that axyz ... t + bxyz ...t + ... kxyz ... i, is capable of assuming the value a + b +... +k. This problem is the same as proving that the equation axyz ... t + bxyz ... t +... 4-kxyz ... i = a + b + ... + k, is always possible. The discriminants (cf. 30 (11)) are A =a + ab ... k, B=b4-ab ...k, ...K Hence Z = a(6+c + ... k), B = b(a + c + ...+k\ ...K=k(a Hence the resultant AB ... K = 0 is satisfied identically. It is obvious that each member of the field of the expression must be incident in the region a + b +c+ ...+k : for a + b+ c + ... + k is the value assumed by the expression when i is substituted for each product xyz ...t, xyz ... t, ... xyz ...t. But this value certainly contains each member of the field. (3) To prove that any member of the field of axyz ... t + bxyz ...t + ... + kocyz ...i contains the region abc ... k. For let (x, y, z,... i), stand for the given expression. Then the region containing any member of the field of p (x, y} z,... t) by the previous subsection is d + b-\-c-r ... -h h. Hence the region contained by any member of the field of f (x,y,z, ... t) is abc.k. Hence combining the results of the previous and present subsections a + b -f c + ... H- k $ f (x, y, z,... t)^= abc ...k. The field of (x} y, z ... t) will be said to be contained between the maximum extension a + b + ... -j- k, and the minimum extension ab ... k. (4) The most general form of p, where a + b + c + ... + k^p^abc.k, is p = abc ... k + u(a + b + c + ... +k). In order to prove that the fields of (p(x, y, z, ...i) and abc ... k + u(a + b + c + ...k),
62 THE ALGEBRA OF SYMBOLIC LOGIC. [CHAP. II. are identical, it is necessary to prove that the equation f (x, y, z, ...t) = abc ... k + u (a + b + c + ... + k), is unlimiting as regards u. The equation can be written axyz... t + bxyz ... + ...+ kxyz ... t = ate ... ku -f (a + 6 + c + ... k)u. The discriminants positive with respect to u are (cf. 30 (11)) a(a-f + c...+ )-f-(X. abc ... A?, that is, a + a c... Tc, and b+ abc ...k, c + dbc ...k, ...k + abc ...k. Their supplements are a(6 + c+...+ ), 6(a + c + ... + ), c(a+ 6+ ... + ), ... A( Hence the product of the supplements is identically zero. Similarly the discriminants negative with respect to u are abc... k + a(a + 6 + ... + k), abc... + 6 (a + b + ... and so on. Their supplements are a(b~\-c + ...+ ), and so on. The product of the supplements is identically zero. Hence (cf. 32 (2)) the . equation is unlimiting with respect to u. Thus* the fields of (f (oc, y,2,...t) and of abc ...k + u(a + b + c+...+k) are identical and therefore without imposing any restriction on u we may write (j (xt y} z,... t) = abc ... k + u(a + b+ c 4-... + ). (5) The conditions that the field of / (#, y, z,... t) may be unlimited are obviously abc ... = 0, a + b + c+ ... +k = i. The two conditions may also be written abc ... k = 0 = abc ... Tc. (6) Consider the two expressions axyz... t + bxyz ... t + . 4- hxyz ...i, and OiUvw ...p-t- bYuvw ...p+ ... h^vw ...p} not necessarily involving the same number of unknowns. Call them $(x,yiz... ) and y{r(utv,w...p). The conditions that the field of (a?,y,z...t) may contain the field of ^{u, v, w ...p), i.e. that all the values which yfr may assume shall be among those which f may assume, are abc ... h = a^Cj... Aj, and a + b + c ... +k^a1-\-b1 + c1+ ... + Ax. The two conditions may also be written abc ... k4 ihCi ...h. * Cf. Schroder, Algebra der Logik, Lecture 10, 19, where this theorem is deduced by another proof.
33] ON THE FIELDS OF EXPRESSIONS. 63 (7) The conditions that the fields of (f (a, y, z ... t) and yjr (u, v, w ... p) may be identical are obviously abc ...k = a1b1c1... h^, dbc ...k = obxb^x... hx. (8) To find the field off(x, y,z...t)f when the unknowns are conditioned by any number of equations of the general type f r(x, y} z...t) = yfrr(x, yy z ...t). Write p = f (x, y,z ...t); and eliminate x,y, z ... from this equation and the equations of condition. Let the discriminant of the typical equation of condition positive with respect to all the variables be Ar, let the dis- criminant positive with respect to all except t be BTi and so on, till all the discriminants are expressed. Then the resultant discriminants (cf. 30 (8) and (9)) of these equations are A = II (Ar\ B = U {Br\ etc. Also let f(x, y, z ...t) be developed with respect to all its unknowns, so that we may write p = axyz ... t + bxyz ... t + ... . The discriminants of this equation are pa+pa, pb+pb, etc. Hence the resultant after eliminating x, y} z... t is __-{(pa+pa)A} -[(pb + pb)B}...=0, that is, {p(a + A) +p(a + A)} [p(b + B) +p(b + B)}... = 0. Hence p(a + A)(b+~B)...+p(a+A)(b + B)... = 0. Thus (cf. 32 (6)) the field of p is comprised between (a + A)(b + B)... and aA + bB+.... But apart from the conditioning equations the field of p is comprised be- tween abc... and a + b + c+__ Thus the effect of the equations in limiting the field of p is exhibited. The problem of this subsection is Boole's general problem of this algebra, which is stated by him as follows (cf. Laws of Thought, Chapter ix. 8): 'Given any equation connecting the symbols x, y,...w, z,..., required to determine the logical expression of any class expressed in any way by the symbols x, y... in terms of the remaining symbols, w, z} etc/ His mode of solution is in essence followed here, w, z,... being replaced by the coefficients and discriminants. Boole however did not notice the distinction between expressions with limited and unlimited fields, so that he does not point out that the problem may also have a solution where no equation of condition is given. A particular case of this general problem is as follows: Given n equations of the type a^ + b^x = c^x + drX, to determine zy where z is given by z = ex +fx. Let the discriminants of the n equations be A and B, those of the equation which defines z are ez + ez, fz+f%
64 THE ALGEBRA OF SYMBOLIC LOGIC. [CHAP. II. Hence the resultant is ~ (eAz + eAz) ~ (fBz +fBz) = 0, that is (ef+fA + eB)z + (ef+/A + eB)z = O; where AB = 0. Hence z = (ef+fA +eB) + u (eA +/B). Another mode of solution, useful later, of this particular case is as follows: The solution for x of the equations is x = B + vA, x = A+ vB. Substitute this value of x in the expression for z. Then z = eB +/A + veA -f l/B = (eA +fl) v + (/B + eB) v. It is easy to verify by the use of subsection (7) that this solution is equivalent to the previous solution. (9) An example of the general problem of subsection (8), which leads to important results later (cf. 36 (2) and (3)), is as follows. Given the equation Axy -\-Bxy -f Cxy -\-Dxy = 0, to determine xy, xy, xy, xy, xy -f xy, xy + xy. Put z = xy, then by comparison with subsection (8) a = i, b = 0 = c d. Hence (a + A)(b + B)(c + C)(d + D) becomes BCD, and aA + bB + cG + dD becomes A, Thus, remembering that ABCD=0, xy = BCD + uA=A (BCD + u). Similarly xy = ACI)+uB = B (A CD + u), xy = ABD + uG= G (ABD + u), Also xy + xy = BC +u(A+D) = (A + D){BC +u}, xy + xy = Z5 + u(B + G) = (B + G) [AD + u}. It is to be noticed that the arbitrary term u of one equation is not identical with the arbitrary term u of any other equation. But relations between the various us must exist, since xy + xy + xy +xy = i. (10) It is possible that the dependence of the value of an expression f(x, y,z...t)on the value of any one of the unknowns may be only apparent. For instance if f(x) stand for x + x, then f(x) is always i for all values of x. It is required to find the condition that, when the values of y, z, ...t are given, the value of f(x, y, z ...t) is also given. For let f(x,y,z...t) = xf1 + xf2, where fx and /2 are functions of y, z ... t only. Then on the right-hand side either i or 0 may by hypothesis be put for x without altering the value of the function. Hence fx =f(x, y...t)=f2. Thus fx =/2 is the requisite condition.
34] ON THE FIELDS OF EXPRESSIONS. 65 Let f(x, y, z ... t) be written in the form x(ayz ...t + byz ... t + ...) 4-x{a'yz ... t + b'yz ... + ...), then the required condition is a = a\ b = b\ etc. 34. Solution of Equations with more than one unknown. (1) Any equation involving n unknowns, x y} z ... r, s, t can always be transformed into an equation simultaneously unlimiting with respect to a set of any number of its unknowns, say with respect to x, y, z.... For let Pj be the product of the supplements of the discriminants positive with respect to x, and QY the product of the supplements of those negative with respect to x. Then (cf. 30 (11)) the resultant after the elimination of all unknowns except x is, PlX + Qxx = 0. Hence we may write, x = QT +Pi#i = Q^ +P1xli where xY is perfectly arbitrary. Substitute this value of x in the given equation, then the transformed equation is unlimiting with respect to its new unknown xx. Again, in the original equation treat x as known, and eliminate all the other unknowns except y. Then the resultant is an equation of the form (Rx + 8x)y + (Tx +Ux)y = 0, where iJ, S, Ty U can easily be expressed in terms of the products of the supplements of discriminants of the original equation. The discriminants in each product are to be selected according to the following scheme (cf. 30(12)): R, S, T, U +, -, +, - y +, +, -, - Now substitute for x in terms of x1} and the resultant becomes P*y + Q2V = 0, where P2 and Q2 are functions of xlt Solving, y = Q-2 + P^ = Q22/2 + Ay2; where y2 is an arbitrary unknown. If this value for y be substituted in the transformed equation, then an equation between xl9 y2, z ... r, s, t is found which is unlimiting with respect to x1 and y2 simultaneously. Similarly in the original equation treat x, y as known, and eliminate all the remaining unknowns except z: a resultant equation is found of the type ( Vxxy + Voxy -f V3xy + V4xy) z + ( W y + W y 4- W3xy -\- W4xy) z = 0; W. 5
66 THE ALGEBRA OF SYMBOLIC LOGIC. [CHAP. II. where the F's and W's are products of the supplements of discriminants selected according to an extension of the above scheme. Now substitute for x and y in terms of x1 and y2, and there results an equation of the type iV + Q-3? = 0 ; where P3 and Q3 contain w1 y2. Solving, z = Qs 4- Psz3 = Qszs + Pss3 where zz is an arbitrary unknown. Then substituting for z, the transformed equation involving xlt y2, z3... r, s, is unlimiting with regard to xly y2, z3 simultaneously. Thus by successive substitutions, proceeding according to this rule, any set of the unknowns can be replaced by a corresponding set with respect to which the transformed equation is simultaneously unlimiting. (2) If this process has been carried on so as to include the n \ un- knowns x, y, z ... s, then the remaining unknown t is conditioned by the equation Pnt+Qnt = 0', where Pn and Qn involve wly y2... sn^ which are unlimited simultaneously. Solving for t, t = Qn + Pntn = Qjn + Pntn; where tn is an arbitrary unknown. Thus the general equation is solved by the following system of values, where scl9 y2... tn are arbitrary unknowns. (3) The generality of the solution, namely the fact that the field of the solution for any variable is identical with the field of that variable as implicitly defined by the original equation, is proved by noting that each step of the process of solution is either a process of forming the resultant of an equation or of solving an equation for one unknown. But since the resultant thus formed is known to be the complete resultant (cf. 31 (9)), and the solution of the equation for one unknown is known to be the complete solution (cf. 31 (8)), it follows that the solutions found are the general solutions. It follows from this method of solution that the general solution of the general equation involving n unknowns requires n arbitrary unknowns. (4) Consider, as an example*, the general equation involving two un- knowns, axy + bxy + cxy + dxy = exy + fxy + gxy + hxy. Let A, B, G, D be its discriminants. Then w = CD + -(AB)x1 = - (IS) x, + ODxlt * Cf. Schroder, Algebra der Lofjik, 22.
35] SOLUTION OF EQUATIONS WITH MORE THAN ONE UNKNOWN. 67 Also (Ax + Cx) y + (Bx + Dx) y = 0. Hence y= Bx + Dx + (Ax + Cx) y2 = (Ax + Cx) y2 + (Bx + Dx) y2 = {(4 + ABC) x, + (C + A CD)x,} y2 +{(AB + A~BD)x1+(CD + BCD) x,} y2. As a verification it may be noticed that the field of y as thus expressed is contained between A 4- C and BD. This is easily seen to be true, re- membering that ABCD = O. (5) The equation involving two unknowns may be more symmetrically solved by substituting (cf. 32 (5)) x =~C~D + ~(AB)u = CDu + ~(AB)u, y = BT)+ -(AC) v = BDv -f ~(AC)v. Then u and v are connected by the equation*, ABCuv + ABDuv + ACDuv + BCDuv = 0. This is an unlimiting equation: thus either u or v may be assumed arbitrarily and the other found by solving the equation. Thus v = ABDu + BCDu + ~(ABCu + ACDu)p, or u = ACDv + BCDv + ~(ABCv + ABDv) q ; where p and g are arbitrary. 35. Symmetrical Solution of Equations with two unknowns. (1) Schroder*|* has given a general symmetrical solution of the general equation involving two unknowns in a form involving three arbitrary un- knowns. The following method of solution includes his results but in a more general form. (2) Consider any unlimiting equation involving two unknowns. Let A, B, C, D be its four discriminants. Then the equation can be written in the form A xy + Bxy + Cxy + Dxy =0.....................(a). Now put x = axuv + biiiv + cfiv + d^uv ........................(/3), y = a2uv + b2uv + c2uv + d2iiv........................(7). Since the equation (a) is unlimiting (cf. 32 (2)), * This equation was pointed out to me by Mr W. E. Johnson and formed the starting-point for my investigations into limiting and unlimiting equations and into expressions with limited and unlimited fields. As far as I am aware these ideas have not previously been developed, nor have the general symmetrical solutions for equations involving three or more unknowns been previously given, cf. 35 37. t Algebra der Logik, Lecture xn. 24. 5 2
68 THE ALGEBRA OF SYMBOLIC LOGIC. [CHAP. II. Also since the fields both of x and y are unlimited, then (cf. 33 (5)) ajb di = 0 = a^c^! = a2b2c2d2 = a2b2c2d2. Substitute for x and y from (ft) and (7) in (a), and write f (p, q) for the expression Apq -f- Bpq + Cpq 4- Dpq. Then the equation between u and v is found to be j (alf a2) uv 4- ( i, b2)uv 4- (f (c1} c2)uv + f (dly d2)uv = 0......( ). Equation (8) is the result of a general transformation from unknowns x and y to unknowns u and v. (3) If the forms (ft) and (7) satisfy equation (a) identically for any two simultaneous values of u and v, then f ( i, O = 0 = (6lf b2) = f)(clt c2) = / (dlf d2). Thus if the pairs (a^, a2), (bly b2), (clf c2), (rfj, d2) be any pairs of simul- taneous particular solutions of the original equation, then (ft) and (7) are also solutions. (4) Assuming that (alf a2) ... (d1} d2) are pairs of simultaneous particular solutions of (a), it remains to discover the condition that the expressions (ft) and (7) for x and y give the general form of the solution. This condition is discovered by noting that the solution is general, if when x has any arbitrarily assigned value, the field of y as defined by equation (a) is the same as the field of y as defined by (7) when u and v are conditioned by equation (ft). Now equation (a) can be written (Ax + Gx) y + (Bx + Dx) y = 0. Hence the field of y as defined by (a) is contained between the maximum extension (cf. 32 (6)) Ax+ Ox and the minimum extension Bx + Dx. Now let AX} BXi CX) Dx be the discriminants of (ft) considered as an equation between u and v. Then Ax = aYx + cijX, Bx bYx + b^x, Gx = cYx + C\X, Dx = dxx + d^c. But by 33 (8) the field of y as defined by (7), where u and v are conditioned by (ft) is contained between the maximum extension a2Ax + b2Bx + c2Cx + d^Dgc, and the minimum extension (a2 + Ax) (b2 + Bx) (c2 + Cx) (d2 + Dx) : that is, between the maximum extension (axa2 -h bj)2 + c^c2 + dxd2) x and the minimum extension 6 + c2
35] SYMMETRICAL SOLUTION OF EQUATIONS WITH TWO UNKNOWNS. 69 If the field of y be the same according to both definitions, then Oja2 + bxb2 + dc2 + dYd2 = A ..................(e), bxb2 + c$2 + dxd2 = G..................(f), (S + d2) = B ..................(v), d2) = 5 ..................(0). These equations can be rewritten in the form axa2 + bj)2 + C 2 + dxd2 = A ..................(e^, axd2 + ft^ a^ + 6^2 + GjC2 + rf^a = Z)..................(fli). It follows from their symmetry that if y be given, the field of x as defined by (/3) and conditioned by (7) is the same as the field of x as defined by (a). By adding 6! and r)1} a1 + b1 +c1 + d1 = A+B. Hence a^c^ = A B. By adding ( ) and (0,1 Hence aj)^^ = GD. Similarly, d2b2c2d2 = AC, a2b2c2d2 = j Thus if the conditions between A, Bt G, D of subsection (2) are fulfilled, then the conditions between a1} bL, c1} dx and a2) 62, c2, d2 of subsection (2) are also fulfilled. Hence finally if (ab a2), (blf b2\ (c1} c2), (dly d2) be any pairs of simul- taneous solutions of (a) which satisfy equations (ej), (^), fa), (0^, then the expressions (/3) and (7) for x and y form the general solution of equation (a). (5) Now take one pair of coefficients, say Oj and a2, to be any pair of particular simultaneous solutions of the equations A xy + Bxy + G'xy + Dxy = Q........................(k), and xy = A....................................(X). These two equations can be treated as simultaneous; for the discriminants of (X) are A, Ay A} A. Hence the complete resultant of the two equations is that is ABGD = 0; and this equation is satisfied by hypothesis. Thus (tc) and (\) can be combined into the single equation
70 THE ALGEBRA OF SYMBOLIC LOGIC. [CHAP. II. that is, since A B = A G = 0, Axy -f- Axy -f Axy + (D + A)xy = 0. Any solution of this equation gives xy = Ay xy^B, xy^C, xy ^D\ and hence any solution is consistent with equations (ej), ( ), (t/j), (0^. This equation is a limiting equation. By 34 (5) it can be transformed into an unlimiting equation. Put x = A+k, y = A + l Then the equation becomes T Let another pair of the coefficients, say bx and b2, be chosen to be any particular solutions of the equations Axy -f Bxy + Gxy + Dxy = 0, #?/ = . . These equations can be treated as simultaneous; and are equivalent to the single equation _ Bxy + Bxy + (C + B)xy + Bxy = 0. Any solutions of this equation give xy = Ay xy = B, xy = C, xy = D. To transform into an unlimiting equation, put x = B + m, y = Then the equation becomes 5 Let another pair of the coefficients, say Cj and c2, be chosen to be any particular solutions of the equations Axy + .B#?/ 4- Gxy + D#j?/ = 0, xy = G. These equations can be treated as simultaneous; and are equivalent to the single equation Gxy + (B+G)xy + Gxy + Gxy = 0. Any solutions of this equation give xy ^ A, xy^ B, xy = G, xy ^= D. To transform into an unlimiting equation, put x = C+p, y = C + q. Then the equation becomes BCpq + Gpq = 0. Let the last pair of coefficients, namely dx and d2, be chosen to be any particular solutions of the equations Axy + Bxy + Gxy + Bxy = 0, xy = D. These equations can be treated as simultaneous; and are equivalent to the single equation (A + D) xy + Dxy + Dxy +Dxy = 0.
35] SYMMETRICAL SOLUTION OF EQUATIONS WITH TWO UNKNOWNS. 71 Any solutions of this equation give xy^A, xy^B, xy = 0, xy = D. To transform into an unlimiting equation, put x D -\-r, y = Then the equation becomes If the coefficients ctj, a2... d2, have these values, then the equations (e), (f), (77), (0) are necessarily satisfied. Hence finally we have the result that the most general solution of the unlimiting equation A xy 4- Bxy + Gxy + Dxy = 0, can be written # = (A + ) mv + (B + m) rw + Gpuv 4- Druv, y = (A + I) uv + .Bniiv + (0 + g) mv 4- Dsuv; where w, v are arbitrary unknowns, and k and I, m and w, p and g, r and 5, are any particular pairs of simultaneous solutions of + BGmn=0, Brs + ADrs = 0.) Let these equations be called the auxiliary equations. The auxiliary equations can also be written, n) = 0, C (p, ?) = 0, (6) As an example, we may determine k, I, m, n, p} qt r, s so that the general solution has a kind of skew symmetry; namely so that x has the same relation to A as y has to D. Thus put ]fc = 0, l = A) m = Bi n = C; q = G, p = B; 5 = 0, r = 5. These satisfy the auxiliary equations. Hence the general solution can be written, remembering that BC = B, GB = C, x = Auv + Buv + Guv, x = Auv + Biiv + Guv + iLv, y = uv-\- Buv + Guv + Duv, y = Buv + Cwv + Duv. Again, put k = i, 1 = 0) m = 0, ?i = i; p = 0, q = i) r = i, 5 = 0. The solution takes the skew symmetrical form w = uv + Buv + Guv, x = Buv + Guv + wv; y = ^Lwy + uv + DuVy y = Auv + uv + Duv. As another example, notice that the auxiliary equations are satisfied by k = w, I = w, m = w, n = w, p = w, q = w, r = w, s = m
72 THE ALGEBRA OF SYMBOLIC LOGIC. [CHAP. II. Hence the general solution can be written x = (A -+ w) uv + (B + w) uv -f Gwuv -f- Dwuv, y {A -\-w)uv + Bwuv + (G + w)uv + Dwuv; where u, v and w are unrestricted, and any special value can be given to w without limiting the generality of the solution. (7) The general symmetrical solution of the limiting equation can now be given. Let Axy + Bxy + Cxy + Dxy = 0 be the given equation. By 34 (5), put w = GD + {A+B)X, y=BD + (A + G)Y; where X and Y are conditioned by ABCXY + ABDXY+ AGDXY + BCDXY= 0. The general symmetrical solution for X and Y is therefore by (5) of this section, X = (A + B + C + k) uv + (A + B + D + m) uv + AGDpuv + BCDruv, Y=(A + B + C+l)uv + ABDTmv + (A + C+D +q)uv + BGD'suv ; where k, I; m, n; p, q; r, s are any simultaneous particular solutions of the auxiliary equations ABGkl + ABCDkl=0}y ABJDmn + ABGDthn = 0, A~CDpq + A BCBpq = 0," BCDrs + ABCDrs = 0.j (8) As a particular example, adapt the first solution of subsection (6) of this section. Then a general solution of the equation is x = CD + (A + BG) uv + (B + AD)uv + AGDuv, y = BD + (A + G) uv + A BDuv + {C + AD) uv + BGDuv. (9) If a number of equations of the type, ^i (x y) = Xi to #) ^2 (x, y) = %2 (#, y) etc., be given, then (assuming that they satisfy the condition for their possibility) their solution can be found by substituting their resultant discriminants (cf. 30, (8), (9)) for the discriminants of the single equation which has been considered in the previous subsections of this article. (10) The symmetrical solution of an equation with two unknowns has been obtained in terms of two arbitrary unknowns, and of one or more unknowns to which any arbitrary particular values can be assigned without loss of the generality of the solution. It was proved in 34 (3) that no solution with less than two unknowns could be general. It is of im- portance in the following articles to obtain the general symmetrical
36] Johnson's method. 73 solution with more than two arbitrary unknowns. For instance take three unknowns, u} vy w (though the reasoning will apply equally well to any number). Let the given unlimiting equation be Axy + T$xy + T)xy + Bxy = 0.....................(a). Put x = a2 uvw + ! uvw 4- d uvw 4- dx uvw } 4-Or[uvw 4- biuvw4- d'uvw + d^uvw, I y = a2 uvw 4-..... + d2 uvw j 4- a2uvw 4-...... 4- d2uvw. ) Consider x as a known, then the maximum extension of the field of y as defined by (a) is Ax 4- Gx, and its minimum extension is Bx + Dx. Also the maximum extension of the field of y as defined by (/3) is SaA # + 2aia2. x, and its minimum extension is II (a2 + a2) x + II (a2 4- a2). ^. Hence, if (/?) is the general solution of (a), the following four conditions must hold ^ZaYa2 = A, Sa^ = 5, SajCia = G, Sa2a2 = i). Also alf a2; blyb2... d/, d2'; must be pairs of simultaneous solutions of the given equation (a). 36. Johnson's Method. (1) The following interesting method of solving symmetrically equations, limiting or unlimiting, involving any number of unknowns is due to Mr W. E. Johnson. (2) Lemma. To divide a 4- b into two mutually exclusive parts x and yf such that x ^ a and y 4 b. The required conditions are x 4- y = a 4- by ax = 0, by = 0, xy = 0. These can be written #3/ 4- axy 4- fey + (a + 6)a^ = 0. Hence by 34 (5), x = ab 4- au = a (6 4- bv = b(a + v);) ........................ where a (uv 4- wv) =0..............................(2). Solving (2) for v in terms of u, by 31 (5), v = abu + (a + b + u)w. Substituting for v in (1) and simplifying, x = a (b 4- w), y = b (a 4- u). (3) Let the equation, limiting or unlimiting, be Z#y 4- Bxy 4- C% 4- Bxy = 0.......................(3). The resultant of elimination can be written A+B+G + D = i. Also xy 4- xy and #y 4- % are mutually exclusive, their sum and obviously from the given equation xy 4- xy = A 4- D, xy + xy ^ B + G.
74 THE ALGEBRA OF SYMBOLIC LOGIC. [CHAP. II. Hence by the lemma xy + xy= (A 4 D)(B() + u), xy + xy = (B + G) (AD + ii)......(4). The course of the proof has obviously secured that u does not have to satisfy some further condition in order that equation (4) may express the full knowledge concerning xy + xy and xy -f xy, which can be extracted from equation (3). Also, as an alternative proof of this point, 33 (9) secures that equation (4) represents the complete solution for these expressions. Again, by equations (4) xy -\-xy^ BG + u, hence xy = BG + u. Also by equation (3), xy^ A. Hence by 28, Prop. XIV., xy ^ A (BG -h u). Similarly xy ^ D (BG + u). Therefore by the lemma and equations (4) and simplifying xy = A (BG+u) (D + p),\ Also, as before, it follows that equations (5) are a complete expression of the information respecting xy and xy to be extracted from equation (3). Similarly xy = B (AD + u)(C +q),) xy=C(AD + u)(B + q))........................( '' Adding appropriate equations out of (5) and (6), x = A (BG + u) (5+p) + B (AD + u) (G_ + q),\ AD + u)(B + q).\ y = A (BG + u)(D + p)+ G(AD + u)(B + q). This symmetrical solution with three arbitraries is the symmetrical solution first obtained by Schroder (cf. loc. cit.). (4) A simplified form of this expression has also been given by Johnson. For A (BG + u) (D+p)=A (BGu + u) (D + p\ and B (AD + u) (G + q) = B (ADu + u) (G + q). Hence x = u [AD + Ap + BAD(G + q)} + u [ABC(D+p) + BC + Bq} Similarly y = A (B + u) (D + p) + G (D + u) (B + q), (5) This method of solution can be applied to equations involving any number of unknowns. The proof is the same as for two unknowns, and the headings of the argument will now be stated for three unknowns. Consider the equation Axyz + Bxyz + Gxyz + Dxyz + A'xyz + B'xyz + G'xyz -f- D'xyz = 0.. .(1).
37] SYMMETRICAL SOLUTION OF EQUATIONS WITH THREE UNKNOWNS. 75 The resultant is A + B + 0 + D + A' + B' + C 4- D' = i. Also from (1) #?/2 4- a??/3 + xyz + ?/ : il 4- D + 2J' -f C", xyz ^ B + C + A' + D'. Hence by the lemma, cf. subsection (2) xyz 4- xyz + xyz + xyz = (A+D+B' + C)(BCAD' 4- ),) xyz 4- ^ + ^0 4- xyz = (B + C + A' + D') (ADB'C' 4- s).) ......^' Again, from (2) and (1), xyz 4- xyz ^ (5' 4- C") (BCA'D' 4- *), a^s 4- xyz ^(A+D) (BCA'D' 4- s). Hence from the lemma, cf. subsection (2), and simplifying, xyz 4- xyz = (E 4- (?) (BCA'D' 4- s) (AD 4- m),) J = (A+D) (BCA'D' + s) (BV + Yn).\............(3)' Similarly xyz 4- xyz = (B+C) (ADB'C' 4- s) (I'D' 4- n),) xyz + xyz = (A' + D')(ADB'C' + s)(BC + v).\............^ Again, from equations (3) and (1), xyz 4 A (BCA'D' 4- s) (BC 4- m)t xyz = D (BCA'D' 4- s) (B'C' 4- /). Hence by similar reasoning to that above i =A (BCIB+ s)(B'C'+ w) (D 4- q), xyz = D (BCA'D'- Similarly, ? = B' (A'BCD' 4- s) (AD 4- m)(G 4- p\ xyz = 0' (ABCD' - xyz = 0(IB'C'D 4- sXZ'ir 4- w)(5 + t\ xyz = B (AB'C'D 4- s)(A'D' + n) ^0 = ^' (AB'C'D 4- *) (504- 77) (5' 4- 0, *yz = -D' (AB'C'D 4- )(5(7 4- w) (2' + 7). By adding the appropriate equations we determine x, y, z. This method is applicable to an equation involving n unknowns, and in this case the solution will involve 2n 1 arbitraries. 37. Symmetrical Solution of Equations with three unknowns. (1) Consider the unlimiting equation Axyz 4- Bxyz 4- Cxyz 4- Dxyz 4- A'xyz 4- Bxyz 4- C'xyz 4- D'xyz = 0 ...............(a). The conditions that the equation is unlimiting are (cf. 32 (3)), ai)Oi/ = U = jl Jj kj U Jx JL 00 = JjJj JLJ U = Jx xL X) x =OO U U . Let the left-hand side of (a) be written 0 (x, y, z) for brevity.
08) 76 THE ALGEBRA OF SYMBOLIC LOGIC. [CHAP. II. Let the general solution of (a) be x = a^uvw + biuvw + cjuvw + d{avw } + diuvw + bi'uvw + CiUvw + d^uvw, y = a2uvw + b2uvw + c2uvw + d2uvw + a2uvw + b2uvw + c2uvw + d2'uvw, z ~ a3uvw + b3uvw + (^uvtv + d3uvw + cix'uvw -f fr/uyw + c3'uvw + dz'uvw\\ where w, v, w are arbitrary unknowns. By substituting for x, y, z from equations (0) in equation (a) the con- ditions that (/3) should be some solution of (a) without restricting u, v, w are found to be, cf ( !, a2, a3) = 0 = f (b1} b2, b3) = j (c1} c2, c3) = f (dx, d2, d3) = f fa'; a2', a3) = f (6/, b2y b3) = f (c/, c/, c3) = f fa', d2, d3'). Thus the corresponding triplets of coefficients must be solutions of the given equation. (2) It remains to find the conditions that (j3) may represent the general solution of (a). Eliminate z from (a), the resultant is A A' xy + BB'xy + GG'xy + DD'xy = 0. By 35 (10), the conditions that the first two equations of (/3) should be the general solution of this equation are %a,a2 = A + A', Sc^o, = B + Bf, 2aA = C + C", ta.a, = D + D'. Similarly eliminating y from equation (a), the resultant is A~Bxz + A'B'xz + 7Ife + C'L'xz = 0. The conditions that the first and third of equations ($) should form the general solution of this equation are X(ha3 = A+B, Za t = A'+ B\ ldxa3 = G + D, So^ = C + B'. Lastly, eliminating x from equation (a), the resultant is ACyz + A'G'yz + BDyz + B'T)fyz = 0. The conditions that the second and third of equations (/3) should form the general solution of this equation are (3) Again, if y and z be conceived as given, the field of x as defined by equation (a) is contained between the maximum extension Ayz + A'yz + Byz + B'yz, and the minimum extension Gyz + G'yz + Byz + B'yz. But (cf. 33 (8)) the field of x as defined by equations (#) is contained
37] SYMMETRICAL SOLUTION OF EQUATIONS WITH THREE UNKNOWNS. 77 between the maximum extension 2cra (a2y + a2y) {a^z 4- a3z), and the minimum extension II \ax 4- a^y + a2y + a3z + a3z); that is, between the maximum ex- tension 'ZoqOjh. yz + Xaja2a3. yz -f Sa^ag. yz + Sc^a/^. ^ and the minimum extension II (ax + a2 + d3) . yz + U (ax + a2 + a3) . yl + II ( ! + a2 + a3) . 2/2 + II (ax + a2 4- a3) . p?. Hence since the extensions as defined by (a) and (0) must be identical, we find by comparison = B, Xa^/i^ = 0, The symmetry of these equations shows that, if z and x be conceived as given, the field of y as defined by (a) is the same as that defined by (#), and that if x and y be conceived as given, the same is true for z. By adding the appropriate pairs of this set of equations it can be seen at once that these eight conditions include the twelve conditions of sub- section (2). Hence finally equations (/?) form the general solution of equation (a), if the triplets alf a2) a%\ 6i, 62 63; ...; d/, d2', d3'; are any simultaneous sets of solutions of the given equation which satisfy the eight conditions above. (4) Now following the method of 35 (5), let alf a2i a3 be any particular simultaneous solutions of the equations cj (x, y, z) = 0, and xyz = A. These equations can be treated as simultaneous and are equivalent to the single equation Axyz + (B + A)xyz + (C + A)xyz + (D + A)xyz + (A' + A)xyz + (B' -h A) xyz + (C' + A)xyz + (5' + A)xyz = 0. This equation is in general a limiting equation. It can be transformed into an unlimiting equation by writing (cf. 32 (5)) y = (B + A)(S + A)(D + A)(D'-+ A) + -(AA'BB')p2, The conditions in subsection (1) that the original equation may be un- limiting reduce these formulae of transformation to x = A-i-ply y = A+p2) z = A+ps. Then pup2i p3 satisfy the unlimiting equation + A BpYp2p3 + A Cpiptfz + A Dp^ps i + A B'p.pzps -f A C'p^ps + A D%p2p., = 0......(1).
78 THE ALGEBRA OF SYMBOLIC LOGIC [CHAP. II. Similarly put ^ = B + qlt 62 = B 4 q2i bz = B + q3, where q1} q2, q3 satisfy the unlimiting equation ABqxq2q3 + Bq^q^ 4- 'BGql q2q3 + BDq1q2q3 + BA'q^q, + BB'qiq2q3 + BC'q^q, + BD'q^q, = 0......(2). Similarly put c1=(7 + r1, c2= G + r2, c3=(7 + r3, where rn r2, r3 satisfy the unlimiting equation AGr1r2r3 + BGi\r2r3 + Cr^r^ 4- CBr^j^ + CA'rws + CB'rffo + GG'f^n + CD%rars = 0......(3). And so on for the remaining triplets of coefficients, putting = D + s2, d3 =D+s3 =A' + p2', a./ = A' + p3, K = B + qif, K = ^ + ?/, 6/ = S' + ?/, c/ = C + r/, c2' = C" + r/, c/ = G' + r/, 5/ = D' + ^/ 5/ = U + 5/, *' = I? + V. And the sets s1} 52, s3; pi\p2\ }h'', ; /, 53', 53'; satisfy unlimiting equa- tions formed according to the same law as (1), (2), (3). These other equations will be numbered (4), (5), (6), (7), (8). Let the equations (1)...(8) be called the auxiliary equations. When the coefficients aly a2, a3; blt b2, b3; ...; d/, d2, d3 have the values here assigned, the eight equations of condition of subsection (3) are identically satisfied. (5) Hence the general solution of the equation A xyz -f B xyz 4- G xyz + D xyz + A'xyz + B'xyz 4 G'xyz 4- D'xyz = 0, is given by x = (A 4- Pi) uvw 4- (B + qj) uvw 4 Cr^ivw 4- Dsjuvw 4- (^4/ + i') wv^4- (Br 4- ?/) ^^ + G'riuvw 4-D's^uvw, y = (A 4- jp2) w/yw 4- Bq2uvw 4- ((7 4- r2) wvw 4- Ds2uvw + (4' 4-^2') ^^^ + B'qjuvw 4- (Cr 4 r2') wv^; 4 D's2'uvw, z = (A + p3) uvw + (B + q3) uvw 4- (C 4- rs) wvw 4- (i) 4- sz) uvw A fp3'uvw + B'q3nvw + G'r3uvw 4 Dfs3uvw; where plfp2ip3] qYi q2, q^', ; ^V,*/; respectively are any sets of par- ticular solutions of the auxiliary equations (1), (2)...(8) These equations can be written, 3^r,#, = 0 ...........................(1), ?a,j,)=0 ...........................(2), ra,r8) = 0 ...........................(3), 8,*,) = 0 ...........................(4),
37] SYMMETRICAL SOLUTION OF EQUATIONS WITH THREE UNKNOWNS. 79 p2/,p.;) = 0 ...........................(5), q ;,q3') = 0 ...........................(6), rf/,r,/) = 6~...........................(7), ,8%',89') = 0 ..........................(8), where f (x, y, z) stands for the left-hand side of the given equation. It will be observed that we may put where t1} t2, tz form any particular solution of the given equation, t to y, z) = - (6) The general solution of a limiting equation involving three un- knowns is found, first by transforming it into an unlimiting equation accord- ing to 32 (5), and then by applying the solution of subsection (5) of the present section. (7) The method of reasoning of the present section and the result are both perfectly general. Thus the general equation involving three unknowns can be solved with a redundant unknown by the method of 35 (10). Then by the method of the present section the equation involving four unknowns can be solved in a general symmetrical form. And the auxiliary equations will take the same form: and so on for any number of unknowns. (8) As an example consider the equations yz a, zx = by xy = c. These equations can be combined into a single equation with its right- hand side zero by finding their resultant discriminants (cf. 30 (9)). The discriminants, cf. 30 (11), of the first equation positive with respect to x and y are a and a. The discriminants of the first equation positive with respect to x and negative with respect to y are a and a. The discriminants negative with respect to x of the first equation are the same as those positive with respect to x. Hence the following scheme holds for the discriminants: Constituent...... xyz xyz xyz xyz xyz xyz xyz xyz 1st Equation ... a a d a a a a a 2nd Equation... b I b b b b b b 3rd Equation... c c c c c c c c Resultant \ Discriminants/ abc al)c abc abc abc abc abc abc
80 THE ALGEBRA OF SYMBOLIC LOGIC. [CHAP. II. The resultant is (a + b + c) (a + b +c) (a + b + c) (a + b + c) (a + 6 + c) = 0; that is abc -f ate + ate = 0. This equation can be solved for c in terms of a and 6. The positive discriminant is -(ab + ab), that is ab + db, the negative one is a + 6. The resultant is at (a + db) = 0, which is identically true. The solution is c = ab + u (ab + a"6) = a6 + waft. Hence the solution for x is found from x = (a + b + c) (a -I- b + c) + u (abc + abc -f abc + ate) = 6 + c + wate = b + c + wa. Similarly y = c + a + vb, z = a + b + ivc; where u, v, w satisfy three unlimiting equations, found by substituting for x, y, z in the original equations. Thus substituting in yz = a, we obtain a + be 4- abv + acw + 6mo = a; that is a + bc + bcvw = a. that is abc + abcvw = 0. But ate = 0. Hence the equation becomes abevw = 0. Similarly a b civu = O = db cuv. Thus u, v, w satisfy the equation ate (uv + vw + wu) = 0. Comparing this with the typical form (a) given in 37, Also a particular solution of the given equation is u = v = w = 0. Hence from subsection (5) u =(a + b + c)(UVW + UVW+ UVW)+UVW, v =(a + b + c)(UVW+UVW+ UVW) + w = (a + b + c)(UVW+UVW+ UVW) + Hence the general solutions for x, y, z are + aUVW, bUVW, where U, V, W are arbitrary unknowns. 38. Subtraction and Division. (1) The Analytical (or reverse) pro- cesses, which may be called subtraction and division, have now to be discussed,
38] SUBTRACTION AND DIVISION. 81 Let the expressions a b and a -f- b satisfy the following general conditions: I. That they denote regions, as do all other expressions of the algebra; so that they can be replaced by single letters which have all the properties of other letters of the algebra. II. That they satisfy respectively the following equations : (a b) -f b = a; (a -f- b) x b = a. (2) Let x stand for a b. Then x is given by the equation x + b = a. The positive discriminant is ai + al, that is a, the negative is ab + ab. The resultant is a (ab + ab) = 0, that is ab = 0; hence b = a. The solution is x = ab + ab + ua = ab + ua. Hence for the symbol a b to satisfy the required conditions it is necessary that b = a. Furthermore negative terms in combination with positive terms do not obey the associative law. For by definition, b + {a - b) = (a - b) -f b = a. Also since 6 = a, and therefore b + a a, it follows that (Jb-\-a) b = a b al) + ua. Therefore b + (a b) is not equal to (6 + a) 6. This difficulty may be evaded for groups of terms by supposing that all the positive terms are added together first and reduced to the mutually exclusive form of 27, Prop. X. Such groups of terms must evidently be kept strictly within brackets. It is to be further noticed that the result of subtraction is indeterminate. (3) Again, for division, let x a + b. Then bx = a. The positive discriminant is ab +a6, the negative is aO + aO, that is a. The resultant is a {ab + ab) 0 ; that is ab = 0. Hence a^b. The solution is x = a + v (ab + ab) = a + vdb = a + vb. Factors with the symbol of division prefixed are not associative with those with the symbol of multiplication prefixed (or supposed). For b (a -i- b) = (a -r- b) b = a, by definition. Also since ab = a, (ba) -r-b = a-i-b = a + vb. This difficulty can be evaded by suitable assumptions just as in the case of subtraction. The result of division is indeterminate. w, 6
82 THE ALGEBRA OF SYMBOLIC LOGIC. [CHAP. II. (4) Owing to these difficulties with the associative law the processes of subtraction and division are not of much importance in this algebra. All results which might depend on them can be obtained otherwise*. They are useful at times since thereby the introduction of a fresh symbol may be avoided. Thus instead of introducing x, denned by x + b = a, we may write (a 6), never however omitting the brackets. Similarly we may write (a -f- b) instead of x, denned by bx = a. But great care must be taken even in the limited use of these symbols not to be led away by fallacious analogies. For (a b) = ab + ua ; with the condition b ^a. But {{a + c) - (b + c)} = (a + c)~ (b + c) + u(a + c) = abc + u(a + c). These two symbols are not identical unless abc = ab, and a + c = a. From the first condition c = a + 6, that is ac = ab ^ b ; since ab = b. From the second condition ac = c. Hence from the two conditions c = b. Again, (a -s- b) = a + vb ; with the condition a 4 b. But {(ac) -T- (6c)} =ac + v ~(bc) = ac + v(b + c). These two symbols are therefore not identical unless ac=a, and b+c = ab. From the second condition be = a + b = b, hence b ^c. This includes the first condition which can be written a^c. But a 4 b. Hence the final condition is b ^ c. (5) It can be proved that -(a-6) = (a-f- 6), "(a^6) = (a-6). For both (a b) and (a -f- b) involve the same condition, namely b ^ a, or as it may be written a^b. Again, a-b = ab + va. Therefore ~ (a b) (a + v) (a + b) = a + vb. But (a-f- b) = a + ub. Therefore the two forms are identical in meaning. Similarly - (a + b) = (a b). * First pointed out by Schroder, Der Operationkreis des Logihkalkuls, Leipsic, 1877.
CHAPTER III. Existential Expressions. 39. Existential Expressions. (1) Results which are important in view of the logical application of the algebra are obtained by modifying the symbolism so as to express information as to whether the regions denoted by certain of the terms either are known to be existent (i.e. the terms are then not null), or are known not to include the whole of space (i.e. the terms are then not equal to the universe). If this information is expressed the terms, besides representing regions, give also the additional information, that they are not 0, or are not L When this additional existential information is being given let the symbol = be used instead of the symbol =; and let the use of = be taken to mean that, in addition to the regions respectively represented by the combinations of symbols on either side of it being the same, the exist- ential information on the right-hand side can be derived from that on the left-hand side. The symbolism wanted is one which will adapt itself to the various transformations through which expressions may be passed. If all regions were denoted by single letters, it would be possible simply to write capital letters for regions known to exist, thus X instead of x, and then the information required, namely that X exists, would be preserved through all transformations. Thus X at once tells us that X exists and that X does not embrace all the universe i. But this notation of capitals is not sufficiently flexible. For instance it is not possible to express by it that the region ab exists : this requires that a exists, that b exists, and in addition that they overlap, and this last piece of information is not conveyed by AB. The merit of the symbolism now to be developed is that the new symbols go through exactly the same transformations as the old symbols, and thus two sorts of information, viz. the denotation of regions and the implication of their existence, are thrown into various equivalent forms by the same process of transformation. (2) Any term x can be written in the form xi. Now when the fact has to be expressed that x is not null, let i be modified into j; so that xj expresses that x exists, the j being added after the symbol on which it operates. 6-2
84 EXISTENTIAL EXPRESSIONS. [CHAP. III. Furthermore any term x can be written in the form x + 0. Now when the fact has to be expressed that x does not exhaust the whole region of discourse, that is to say is not i, let the 0 be modified into o . Then x + expresses that x is not equivalent to i. Let any combination of symbols involving j or eo be called an existential expression. Thus j may be looked on as an affirmative symbol, giving assurance of reality, and co as a limitative symbol restraining from undue extension. They have no meaning apart from the terms to which they are indissolubly attached, the attachment being indicated by brackets when necessary, i.e. by (xj) and by (x + co). It is to be noted that xj or x -f co can be read off as assertions: thus xj states that x is not 0, x -f co that x is not i. (3) The symbol xy .j will be taken to mean that j operates on xy, so that xy exists. Thus xy .j implies xj and yj; but the converse does not hold. The mode of attachment of j to the term on which it operates has some analogy to multiplication as it obtains in this algebra. Thus though xj.yj is not equivalent to xy.j as far as its existential information is concerned. Again, if x} y, z, uy ... represent any number of regions, then (xyzu...) j = (xj. yj. zj...) j ; but the final j cannot be omitted, if the existential information is to be the same on both sides. (4) The distributive laws have now to be examined as regards the mul- tiplication and addition of existential expressions. Consider in the first place the expression (x + y)j. Now if x = 0 and y = 0, then x -f y = 0. Hence (x -f y)j implies either xj or yj or both. Thus we may adapt the symbolism so as to write (x + y) j = xjx + yjj; where the suffixes of the j's weaken the meaning to this extent, that one of the j's with this suffix is to hold good as to its existential information but not necessarily both. We define therefore wji + yji + zji+ to mean that one of the terms at least is not 0. The other formal properties (cf. subsection (3)) of j evidently hold good, retaining always this weakened meaning. The only point requiring notice is that xjxj = xj; for jx has the same meaning as j in a weakened hypothetical form. Further (xj + y) j = xjjx + yj1 = xj + y; for the jx can be omitted, since it is known that x exists.
39] EXISTENTIAL EXPRESSIONS. 85 In using the multiplication of existential symbols the dots (or brackets) must be carefully attended to. For instance But 0 + y) z .j = (xz 4- yz)j = xz .j, 4- xz .jY. In the first expression (x 4- y). zj, the j simply asserts that z exists; in the second expression it asserts that (pc + y)z exists. Again, xy + z = (x 4- z) (y + z). Also xy .j 4- z implies (x + z) (y + #) .J. But though #jy -|- = (x 4- ^) (y 4- z) .j ; the left-hand side gives more definite information than the right-hand side. For (x + z)(y + z) .j = xy. jx + zjx. Also xy .j implies xj, yj. Hence xy .j + z = (xj + z) {yj + z)j = xj. yj ,jx + zjx. But still the right-hand side does not give as much information as the left-hand side ; for xj. yj .j1 is not equivalent to xy .j. Hence the distributive power of addition in reference to multiplication to some extent has been lost. It cannot be employed in this instance without some loss of existential information. (5) The symbol (x + y + co) will be taken to mean that a operates on x + y, and therefore that x + y is not i. Thus x + y + implies x -f co and y+oo] but the converse does not hold. The mode of attachment of cd to the term on which it operates has some analogy to addition as it obtains in this algebra. Thus (x + ) + (y + co) + a) = x 4- y + o , though (x 4- ) 4- (y 4- o) is not equivalent to (x + y + q ) as far as its exist- ential information is concerned. Again, if x, y, zyu, ... denote any number of regions, then but the final co cannot be omitted if the existential information is to be the same on both sides. The distributive law of addition in relation to multiplication (cf. 24, equation B) does not hold completely. Consider the expression xy+co. Now xy can only be i, if both x and y are equivalent to i. Hence xy 4- m implies either x 4- co or y + co or both. Thus we may adapt the symbolism so as to write xy 4- w = (x 4- ft i) (y 4- ft i) ; where the suffixes of the g)'s weaken the meaning to this extent, that one of the )'s is to hold good but not necessarily both. We define therefore (x 4- ft i) (y + fth) (z 4- ft)i)... to mean that one at least of the terms oc,y,z,... is not i.
86 EXISTENTIAL EXPRESSIONS. [CHAP. III. It is obvious that (x -f coY 4- co) (y + co ) (z + cox) ... = (x 4-co) y-z..., since # -f co -\- co ensures definitely that # is not i. For example, (# 4- co) y + ) = (x 4- ) + Wi) (y + ft)j) = (a? 4- co) y. Let the symbols such as jx or cox be called weak symbols in contrast to j or a) which are strong symbols. Then a strong symbol absorbs a weak symbol of the same name (j or o ) when they both operate on the same term, and destroys all the companion weak symbols. Thus xjij + yji = j + y, (a?+ i + co)(y + wj) = (# + (c)) #. (6) The chief use of this notation arises from its adaptation to the ordinary transformations owing to the following consideration. If x exists, then x cannot be i ; and conversely if x be not i, then x exists. Hence ~~ (xj) = ~x + co, and ~ (x + co) = xj. But by analogy to 26, Prop. VI. - (xj) = x +], and ~(x + a ) = xco. Hence we may write j = , and = j, corresponding to I = 0, and 0 = i. Thus the original existential information can be retained through any transformations of the algebra. 40. Umbral Letters. (1) This existential notation can be extended. Let the letters of the Greek alphabet be taken to correspond to the letters of the Roman alphabet, so that a corresponds to a, /3 to b, and so on. Let xol mean that the regions x and a overlap ; in other words xa implies xa .j, but the symbol xa in itself denotes only the region x\ it only implies this extra information. Also let x + a, while denoting only the region x, imply that x does not include all the region a; in other words x + a implies xa .j, that is, it implies x + a + co. Thus xa implies aj and xj and xa.j; while x + a implies aj, xj and xa. j. Also xa does not necessarily exclude act, and x + a does not necessarily exclude x 4- a. (2) Now if x includes some of a, it follows that x cannot include all a. Hence if xa, then x 4- a. This can be expressed by the equation - (xa) =x + a. Thus for instance, ~ (xa) . y = (x + a) y. Also it follows that = (xa) = ~ (x 4- a); and hence ~ (a; 4- a) = xa. But by analogy to 26, Prop. VI. ( ) = Wa = xd.
40] UMBRAL LETTERS. 87 Hence we may write BE = a; though as a matter of fact the Greek letters have no meaning apart from the Roman letters to which they assign properties, and therefore should not be written alone. (3) Let these Greek letters be called shadows or umbral letters ; and let the Roman letters denoting regions be called regional letters. Then the umbral letters essentially refer to some regional letters or groups of letters and are never to be separated from them. Thus a (b -f 7) cannot be transformed into ab + ay; the symbol (6 + 7) is essentially one whole, and the bracket can never be broken. Similarly a. by cannot be transformed into ab . 7; since by is one indivisible symbol. But with this limitation that brackets connecting regional and umbral letters are never to be broken it will be found that the umbral letters follow all the laws of transformation of regional letters. (4) In accordance with our previous definitions it may be noted that x(a + ft) implies x (a + b) .j, and (x + a + ft) implies that x does not include all "(a+ 6). Also xaft implies xab ,j and (x + aft) implies that x does not include all ~(ab), that is, all (a + 6). It is further to be remarked that x(a + ft) is not identical in meaning with xa + xft. For x(a + ft) implies x (a + b) .j, that is either xa .j or xb .j or both, while xa + xft implies both xa .j and xb .j. Now xaft implies xab .j, that is both xa.j and xb .j as well as xab .j. Hence xaft implies all that xa + xft implies and more, and xa + xft implies all that x (a + ft) implies and more; while all three expressions represent the same region, namely x. (5) The shadows follow among themselves all the symbolic laws of ordinary letters. For x (a+ff) = x(JJ + a), x + (a + ft) = x + (ft + a), xaft = xfta, x+a$ = x + fta, x (a + a) = xa, x + (a + a) = x + a, xa (ft + y) = x (aft + 7), x + a (ft + 7) = x + (aft + a7), x-(a + p) = xaft, x + ~( Apart from this detailed consideration it is obvious that the same laws must hold; for the shadows also represent regions, though these shadowed regions are only mentioned in the equations for the sake of indicating properties of other regions in reference to them. It should also be noticed that since xaft implies xab.j, it also implies ab .j.
88 EXISTENTIAL EXPRESSIONS. [CHAP. III. Other transformations are [x + (a + fi)} = x ~ (a + fi) = = x + " (a/3) = a + (a a/3} = x"(a/3) = x(a It is to be noted that with the symbol x(a + ft), we may not transform to xa + xft, and thence infer xa and xft; the true transformation is x (a + /3) = ai + a?A, where ax and ^ are weak forms of a and /3. Similarly we may not transform x + (a + j$) into (# + a) + (# + /3) and thence infer x + a and x + ft. (6) Each complex umbral symbol should be treated as one whole as far as symbolic transformations are concerned. Thus the laws of unity and simplicity (cf. 25) have to be partially suspended. For instance xa+xj3 denotes only the region x, but for the purposes of the existential shadow letters xa and x/3 must be treated as distinct symbols. Similarly xa. xfi denotes only the region x, but it does not mean the same as xa/3; for xa. xft denotes the region x and implies xa.j and xb.j, whereas xaft denotes the region x and implies xab .j. The second implication includes the first, but not the first the second. Hence for the purposes of multiplication xa and xfi must be treated as different symbols. The suspension of these laws of unity and simplicity causes no confusion, for the symbols are only to be treated as different symbols (although denoting the same region) when they are so obviously to the eye; thus xa and xft are obviously different symbols. (7) When the same regional letter is combined with various umbral letters, the same result is obtained whether the expressions are added or multiplied *. Thus xa + xft = xa . x/3, xa + 13 =(x + /3)a. (8) This notation enables existential expressions to be transformed. Thus if ( corresponds to x, rj to y, and f to z, xy .j==X7).yi;. Hence xy . j + z = (xV + z) (yf + z); and in this case the connotation is exactly the same on both sides. Hence the distributive power of addition in reference to multiplication has now * This remark is due to Mr W. E. Johnson.
41] ELIMINATION. 89 been retained. It may be noticed that the right-hand side might have been written (xrj + z) (y + z) without alteration of connotation; for xrj implies xj, yjj xy -j an(i ^ne f affixed to y implies do more. Again, x + y + o = (x + rj) + (y + ), where (x -f ?/) implies that x does not include all y and y + implies that y does not include all x. Thus (x + y + (o) z = (x + rj) z + (y + ) z, the connotation of both sides is the same. Thus the distributing power of multiplication in reference to addition has now been retained. It is to be noticed that symbols like x + rj and xt] are to be treated as indivisible wholes. Again as examples consider the transformations ~(xv -j) = ~(^v- y 1 and - (x + y + a ) = ~ \(x 41. Elimination. (1) It is in general possible to eliminate x, y, zy ... from existential expressions of the forms f(x, y,z,...t) j and f(x, y, z, ... t) +co. Consider first the form f{x, y, z,... t) j. Let f(x, y, z,... t) be developed and take the form axyz ... + #;?/ ... ^+ ... + gxyz ... t. By 33 (2) the maximum extension of the field of this expression is Hence if f(x, y, z,... t) j, the maximum extension cannot be null. Thus is the resultant expression when x, yy z, ... t have been eliminated. (2) Consider the form,/(#, y} z,... t) + co. This is equivalent to / (x, y, z, ...t) .j. If f(x, y, zy...t) be developed as in (1), then the existential expression becomes (axyz ... t+ bxyz ... t+ ... +gxyz ... i)j. Hence by (1) (a + b + ... + g)j, that is a ... g + . This result might also have been deduced by noticing that ab ... g is the minimum extension of the field off(x, y, z, ...t); and therefore is necessarily not i, if f(x, y, z, ...t) is not i.
90 EXISTENTIAL EXPRESSIONS. [CHAP. III. (3) As particular cases of the above two subsections, note that (ax -f bx) .j yields (a 4- b) j, (au + bv).j yields (a + b)j, (ax + bx) + o yields ab -f co. Also note that (au + bv) + co yields no information respecting a and b; for when the formula of (2) is applied to its developed form the resultant becomes 0 + w, which is an identity. (4) To eliminate x, y, z, ... t from f (x, y, z, ... t) j and from n equations involving them. Let f(xyy,z, ...t) be developed as in (1), and let the corresponding resultant discriminants of the equations be A, B, G, ... G. Then the maximum extension of the field of/(#, y, z, ...t) as conditioned by the equations is a A + bB + ... + gG. Now f(x, y, z...t)j requires that the maximum extension shall not be null. Hence the complete existential expression* to be found by elimina- tion is Let this be called the existential resultant. The resultant found by elimination of x, y, z,... t from the equations is IB... G=0. The existential resultant and the resultant of the equations contain the complete information to be obtained from the given premises, after the elimination of x} y, ... t (5) An allied problem to that of the previous subsection is to find the condition that the existential expression may not condition x, y, z, ... t any further than they are already conditioned by the equations. The minimum extension of the field of f (x, y, z,... t) as conditioned by the equations is by 38 (8), Hence if (a + A) (b + B) ... (g + (?) j, then f(xy y, z,... t)j} for all values of xt y, z, ... t; and thus f(x, y, z,... t)j does not condition x, y, z, ... t. The condition can also be written (aA +bB+ ...gG + co). (6) A special case of (5) arises when there are no equations; the exist- ential expression does not condition the unknowns, if abc.g.j. * This expression found by another method was pointed out to me by Mr W. E. Johnson.
42] SOLUTIONS OF EXISTENTIAL EXPRESSIONS WITH ONE UNKNOWN. 91 (7) If the existential expression be f(x, y, z, ...t) + co, then by reasoning similar to that in subsections (4) and (5) the existential resultant is The condition that the unknowns are not conditioned by the existential expression is (aA + bB + ... + gG + ). These conditions may respectively be written and (a + A)(b + B)...(g+~G)j. 42. Solutions of Existential Expressions with one unknown. (1) Solution of ax.j. The form of solution for x can be written in two alternative forms by using symbols for undetermined regions: thus x = wa . j + ua = pa. The first form states explicitly that x is some (not none) undetermined part of the region a together with some (or none) of a. The second form states the same solution more concisely but perhaps less in detail: it states that x may be any region p, so long as p is assumed to include some (not none) of the region a. There is no reason in future to write p for the undeter- mined region denoted by x. Thus we shall say that the solution of ax .j is x = xa. (2) Solution ofbx.j. From the preceding proposition x = wb.j + u = x/3. Hence x = ~ (wb .j + u) = ~ (x/3) = u (w + b + co) = x + /?. The form u (w + b + ay) states that x must be some (or none) of a region which is composed of all b and of any other region, except that the total region must not comprise all the Universe. The form x + 0 states that x may be any region so long as it does not comprise all b. (3) Solution of any number of expressions ax .jy ax.j, ... anx.j. The required solution is obviously x = xa + xo! + ... + xan = 2#a (say). It may be noticed that x (a -f a' + ... -f an) is not the required solution, since it is only equivalent to the weakened form xoly + xa-[+ ... + xaf1; also that xaa ... ctn implies aa ...an.j and xaa' ... an .j, which is more than is given by the equations. By 40 (7) the solution can also be written x = xa. xa ... xan = II (xa) (say).
92 EXISTENTIAL EXPRESSIONS. [CHAP. III. (4) Solution of any number of expressions of the types bx. J, b'x. j, ... bnx. j. The required solution is (say). (5) Solution of any number of expressions of the types ax. j, dx . j9 ... anx . j, bx . j, b'x. j,... bmx . j. The solution is obviously ( a J3 If there are only two such expressions, namely ax. j and bx. j, the solution becomes x = (x + ft) a. (6) Solution of (ax + 6^) .;'. Now (a# 4- bx)j = ax. ^ + bx. j\. By subsection (5) ax,jx and bx.jx imply ^ = (x + ft) !; where a! and /32 are alternative weakened forms of shadows. But this expression does not necessarily imply any restriction on x. For ax + bx can only vanish if ab = 0. Hence (ax + bx)j either implies ab. j and x entirely unconditioned, or ab = Q and (7) Solution of ax + a. Now ax + co implies ~ (ax + ), that is ( x + x)j. But ( z + x)j=(ax + x)j. This implies either ay and x entirely un- conditioned, or a 0, that is a = i, and # + . (8) Solution of bx + co. Now te 4- implies ~ ( c -f cd), that is (6 + x)j. But (6 + ^)^'= (6^4-5?)J. This implies either bj and ^ entirely uncon- ditioned, or b = 0, that is = i, and xj. (9) Solution of ax + bx + g . Now a# -f ai + a) implies ~ (ai 4- bx + ). But ~ (ax + bx + co) = (ax 4- fe)i. Hence either ab . j (that is, a + b + a) and a? is entirely unconditioned, or ab = 0 and x = (x + ) ax; where ax and are weak forms.
43] EXISTENTIAL EXPRESSIONS WITH TWO UNKNOWNS. 93 43. Existential Expressions with two unknowns. (1) The general form of the existential expression involving two unknowns, x and yy is (axy + bxy + cxy + dxy)j. Let f(x, y) stand for the expression axy -f- bxy + cxy + dxy. If abed. j, the above existential expression does not condition x and y in any way (cf. 41 (6)). But if abed =0, then f(x, y) vanishes (cf. 34 (5)), if x = cd-l-u(a + b), y bd + v(a-\-c)..................(1); where u and v satisfy the unlimiting equation abcuv + ~dbduv + acduv + beduv = 0. Thus if f(x, y) is to vanish the minimum extension of the field of x is cdt its maximum extension is a 4- b, the minimum extension of the field of y is bd, its maximum extension is a 4- c. Accordingly,/(#, y)j and abcd = 0, yield three cases: (a) x lies outside its above-mentioned field, and y is unrestricted: (ft) y lies outside its above-mentioned field, and x is unrestricted : (7) both x and y lie within their respective fields, but do not occupy simultaneous positions within their fields. That is to say, x and y can both be expressed by equations (1), but (abcuv+ abduv + aeduv + icduv)j. If f(x, y) = 0 be an unlimiting equation for x and y, then cases (a) and (ft) necessarily cannot be realized; and the existential expression in case (7) becomes f(uy v) j, where u and v are written instead of x and y. Case (a) is symbolized by x = {^ + ~(%S)1j (ay8)j, where % is the umbral letter of c and the suffixes denote alternative weak forms. This existential expression for x implies that either x does not include all cd, or x does include some region not (a + 6). Case (ft) is symbolized by y = {y -f ~ (/SS^} (a%X. Case (7) requires that the problem of the next subsection be first con- sidered. (2) To solve for x and y from the expression,/(#, y)j\ where f(x, y) = 0 is an unlimiting equation. No expression for x or for y can be given, which taken by itself will satisfy f(x, y)j: for since the equation, f(x, y) = 0, is unlimiting any value of x or of y is consistent with its satisfaction. Thus to secure the satisfaction of /(#, y)j, either x or y must be assumed to have been assigned and then the suitable expression for the other (i.e. y or x) can be given. Thus write f(x, y) j = {(ay + by)x+ (cy + dy) x] j. Then by 42 (6), if y be conceived as given,
94 EXISTENTIAL EXPRESSIONS. [CHAP. III. Similarly if x be conceived as given Both these expressions for x and y hold concurrently, and either of them expresses the full solution of the problem. (3) Returning to the general problem of the solution of (axy + bxy + cxy + dxy) j, where abed = 0; the different cases can be symbolized thus: (a) ={* + (8) y={y+ (7) x = [x + (xv + lv)i] ( V where x and y have the forms assigned in equations (1) of subsection (1). 44. Equations and Existential Expressions with one unknown. (1) Let there be n equations of the type arx + brx = crx + drx; and an existential expression of the type ex . j. Let A and B be the resultant discriminants of the n equations. Then the total amount of information to be got from the equations alone is (cf. 30), AB = 0, and x = B + uA. The full information to be obtained by eliminating x is (cf. 41 (4)), AB = 0, eA .j. In considering the effect of the existential proposition on the solution for x two cases arise. For x = B + uA} where u is conditioned by Hence either (1) eB . /, x = Be + uA, in which case u is entirely uncon- ditioned (cf. 41 (5)); or (2) eB = 0, and ueA .j. If the coefficients such as e, ar, bri etc. be supposed to be known, then any result not conditioning u may be supposed to give no fresh information. Thus in case (1), where eB .j, this result must be supposed to have been previously known, and therefore the existential expression ex . j adds nothing to the equations. But in the case (2), ueA . j gives u = ueay where a is the umbral letter of A. Hence the solution for x is x = B -f uA . eoL. Here the existential expression ex. j has partially conditioned uy and thus has given fresh information.
44] EQUATIONS AND EXISTENTIAL EXPRESSIONS WITH ONE UNKNOWN. 95 (2) Let there be n equations of the type arx -\-brx = crx + drx, and an existential expression of the type ex .j. The resultant of the equations is A B = 0, and their solution is x = B + uA. Hence x = A + uB. Hence e (A + uB) . j. The resultants AB = Q, eB.j contain the full information to be found by eliminating x (cf. 41 (4)). The solution for x falls into two cases; either (1) eA.j, and u is not conditioned (cf. 41 (5)); or (2) eA = 0, and ueB, j. If the coefficients be assumed to be known apart from these given equations, then the solution in case (1) must be taken to mean that the existential expression adds nothing to the determination of x beyond the information already contained in the equations. But in case (2) u is partially determined; for from ueB ,jy we deduce u = U fi, where /3 is the umbral letter of B. Hence u = (u + + /3). Therefore if eA = 0, x = B + (u + ~e + fi)A. In this case the existential expression has given fresh information. (3) Let there be n equations of the type a^x + b^c = v# + dyXy and an existential expression of the type (ex + gx)j. The resultant of the equations is AB 0, and their solution is x = B 4- uA, x = A + uB. Hence {eB 4- gA + eAu+ gBu) j. The resultants .4i? = 0, (eA + gB)j contain the full information to be found by eliminating x (cf. 41 (4)). The solution for x falls into two cases, according as the existential expression {eB + gA + eAn -f gBu}j does not or does condition u. Case (1). If (eB -f- gA +egAB)j, then the above existential expression does not condition u at all (cf. 41 (5)). Hence if the coefficients are assumed to be known apart from the in- formation of the given equations and existential expression, then the exist- ential expression must be considered as included in the equations. Case (2). If (eB -\-gA +egAB) = 0, then the existential expression for u reduces to (eAu + gBu)j, where egAB = 0. Hence (cf. 42 (5)) the solution for u is
96 EXISTENTIAL EXPRESSIONS. [CHAP. III. where the suffix 1 to the brackets of the umbral letters implies that they are alternative weak forms. Hence the solution for x is in this case In this case the given existential expression is to be considered as giving fresh information. 45. Boole's General Problem. (1) This problem (cf 33 (8)) can be adapted to the case when existential expressions are given, as in the following special case. Let there be given n equations of the type a^x + brx = CrX + d^x, and an existential expression of the type gx .j; it is required to determine z. where z is given by z = ex -f fx. By 33 (8), z=eB+fA + veA + ufB, where x = B + uA. Hence from above either (1) gB.j, and u is unconditioned by the existential expression, or (2) gB = 0, gAu .j. In the second case w= uya. Hence if gB. j the existential expression adds nothing to the solution, assuming that the coefficients are already known ; if gB = 0, then z = eB +f~A + eA . uya +fB (u + y + ) It is to be noticed that even in the second case the existential expression gives no positive information as to z, and that it only suggests a possibility. For the solution asserts that u contains some of gA, but eA need not overlap that part of gA contained in u. Similarly the umbral letters in the ex- pression fB(u + 7 + a) give no definite information as to the nature of the term. (2) If the existential expression in this problem be of the type gx .j, then if gA .j, it is included in the equations. But if gA = 0, the solution for z is z = eB +/Z+ eA (u + /8 + 7) +/B.uj3y. Similar remarks apply to this solution as apply to that of the previous form of the problem. (3) If the existential expression be of the type gz .j or gz .j, then more definite information can be extracted. Take the first case, namely gz.j, as an example. The solution for z from the equations is (cf. 33 (8)) z = (ef+ eB +fA)4-u {eA +fB\ where ef+ eB +/A 4 eA +fB.
46] EQUATIONS AND EXISTENTIAL PROPOSITIONS WITH MANY UNKNOWNS. 97 The existential expression requires the condition g(eA+/B).j. If the coefficients are assumed to be well-known, then if g(ef+eB+/l).j, no information is added by the existential expression. But if g(ef+eB+fA) = 0, then z = (ef+ eB +/A ) + (eA+fB) u (7ea + ycf p), where f is the umbral letter off. The solution for gz .j is similar in type. 46. Equations and Existential Propositions with many unknowns. (1) A more complicated series of problems is arrived at by considering the set of n equations involving two unknowns of the type drxy -f brxy + c^cy + d^cy = ar'xy 4- b/xy + cr'xy + dr'xy...... (1) ; combined with the existential expression of the type (eny +fxy+ gxy + Ixy) .j........................ (2). The various discriminants of the typical equation are Ar = ar V 4- ara,f, Br = brbr' 4- 1jbr\ Gr = crcr' + crcr\ Dr = drdr' + drd/. Also the resultant discriminants are a = n (Ar), B=n (Br), c - n (cr\ n = n (Dr). Then from 30 (9) the resultant of the equations is ABCD = 0, and from 41 (4) the existential resultant is (eA +/B+gC+W).j. If {e + I)(f+5)(g+C)(l + B).j, then by 41 (5) the existential expression (2) adds nothing to the equations (1) as regards the determination of x, assuming that the coefficients are well- known. Assume that (e + A)(f+B)(g+ C) (l + D) = 0. The solutions of the equations for x and y can be written x = (A+B)u+ CDu, y = {A+C)v+ BDv, where u and v satisfy ABCuv + ABDuv + ACDuv+BCDuv = O..................(3). Substituting in (2) for x and y, [{e(A+BC)+flBC + gABC+lABC}uv + {eABD+f(B+AD)+gABD + lABD}uv + {eA CD +fZ CD+g (G + AD) + IA CD) uv + {eBCD + /BCD + gBGD + l(D + BG)}uv] .j............(4). w. 7
98 EXISTENTIAL EXPRESSIONS. [CHAP. III. The equation (3) is unlimiting and the problem now becomes that of the next subsection. (2) Given an unlimiting equation (5) and an existential expression (exy+fxy+gxy + lxy).j ..........................(6) to find the solution for x and y. Let A, By C, D be the discriminants of the equation (5). Then, as before, the condition that x and y are conditioned by (6) is Since the equation (5) is unlimiting, this equation can be written efgl +fglA + gleB + lefC + efgD = 0. Let a symmetrical solution of the equation (5) according to the method of 35 be x axuv -f huv + cxuv -f dYuv, y a2uv + b2uv + c2uv + d2uv. Let the expression (6) be written/(#, y) .j for brevity. Then substituting in (6) for x and y, as in 33 (2), the expression becomes i/( i, h) uv +f(blt b2) uv +f(cly c2) uv +f(d1} d2) uv} .j. But this expression has been solved in 43. Note. In this discussion of Existential Expressions valuable hints have been taken from the admirable paper, * On the Algebra of Logic,' by Miss Christine Ladd (Mrs Franklin) in the book entitled Studies in Deductive Logic, by Members of the Johns Hopkins University But Mrs Franklin's calculus does not conform to the algebraic type considered in this book ; and the discussion of Existential Expressions given here will, it is believed, be found to have been developed on lines essentially different to the discussion in that paper.
CHAPTER IV. Application to Logic. 47. Propositions. (1) It remains to notice the application of this algebra to Formal Logic, conceived as the Art of Deductive Reasoning. It seems obvious that a calculus beyond its suggestiveness can add nothing to the theory of Reasoning. For the use of a calculus is after all nothing but a way of avoiding reasoning by the help of the manipulation of symbols. (2) The four traditional forms of proposition of Deductive Logic are All a isb........................................(A), No a is b.......................................(E), Some a is b ....................................(I), Some a is not b ..............................(0). Proposition A can be conceived as stating that the region of a's is included within that of 6's, the regions of space being correlated to classes of things. It is unnecessary to enquire here whether this is a satisfactory mode of stating the proposition for the purpose of explaining the theory of judgment: it is sufficient that it is a mode of expressing what the proposition expresses. (3) Accordingly in the notation of the Algebra of Symbolic Logic proposition A can be represented by a^b ....................................(A), where a symbolizes the class of things each a, and b the class of things each b. By 26, Prop, vin, and 28, this proposition can be put into many equi- valent symbolic forms, namely a = ab, b = a + b. Also into other forms involving i, a and b; namely, b^a, ab = O, a = a + b, ci + b=i, b = ab. Also into other forms involving the mention of an undetermined class u; namely a ub, b = a -f u} a = b + u. 7 2
100 APPLICATION TO LOGIC. [CHAP. IV. (4) According to this interpretation i must symbolize that limited class of things which is the special subject of discourse on any particular occasion. Such a class was called by De Morgan, the Universe of Discourse. Hence the name, Universe, which has been adopted for it here. (5) Proposition E can be construed as denying that the regions of a's and 6's overlap. Its symbolic expression is therefore ab = 0 ....................................(E). This can be converted into the alternative forms a^b, b= d, a = ab, b = bd, a + 6 = i, d d + 6, 6 = 6-f-a. Thus, allowing the introduction of i, there are eight equivalent symbolic forms of the universal negative proposition, as well as eight forms of the universal affirmative. But if the introduction of i be not allowed, there is but one form, namely, ab = 0; remembering that the supplement of a term by its definition [cf. 23 (8)] implies i. On the other hand if the introduction of an undefined class symbol (u) be allowed, then four other forms appear, namely, a = ub, a = u + b, b = ud, b = u + a. (6) Proposition I can be construed as affirming that the regions of the a's and 6's overlap. Hence it affirms that the region ab exists. This is symbolically asserted by t -5.......................................(i)- Equivalent forms are (cf. 40) a/3 . 6a; a -f b + a ; (a + /3) + (b + a). Also if the introduction of undefined class symbols be allowed, then other equivalent forms are, a = wb.j + u] b = wa.j-\-u; d = u(w + b + co)] b = u(w + a + o)). (7) Proposition O affirms that the regions of a's and b's overlap. This is symbolically expressed by ab.j.......................................(O). Equivalent forms are a$ . ba; d + b + a ; (a + /3) + (6 + - ). Also using undefined class symbols, a = wb.j + u; b = u {tv + a + ); a = u (w + / + ); b = wa .j + u. 48. Exclusion of Nugatory Forms. (1) It is sometimes necessary to symbolize propositions of the type A, so as to exclude nugatory forms; for instance when it is desired to infer symbolically a particular proposition from two universals. (2) In order to avoid the form of nugatoriness which would arise from a = 0, in a 4 ft, we can write or aj = ab .j.................................(2).
49] SYLLOGISM. 101 The series of other forms can be deduced by mere symbolical reasoning from this form. Thus b = b + ab; also bj, ab .j, and ab .j = aj\ hence bJ = bJ + aJ ................................(3). Again, by taking the supplement of bj, we deduce b + co. Multiplying (2) by (6 + co), we find aj .(6 + a ) = 0.................................(4). By taking supplements of (1), b + co ^a + co..................................(5). By taking supplements of (2) a + co = a + b + co ..............................(6). By taking supplements of (3) b + o) = (b -f to) (a -f co)...........................(7). By taking supplements of (4) (a + G ) + bj = i ..............................(8). Thus the eight forms of the proposition (A) (excluding those with un- determined class symbols) have been symbolized so as to exclude the nugatory form which arises when a = 0. (3) Another nugatory form arises when b = it this form can be excluded by the forms a -f co = b + (o, or (a + co) = (a + ro) (b + co). By comparing these forms with equations (5) and (7) in subsection (2) it is easy to write down the remaining six forms. It is also possible to combine the symbolism of both cases and thus to exclude both forms of nugatoriness, viz. a = 0, or b = i But it is rarely that reasoning requires both forms to be excluded simultaneously, so there is no gain in the additional complication of the symbolism. 49. Syllogism. (1) The various figures of the traditional syllogisms are as follows, where a is the minor term, b the middle term and c the major term: First Figure. A, All b is c, E, No b is c, A, All b is c, E, No b is c, A, All a is b, A, All a is 6, I, Some a is b, I, Some a is b, therefore therefore therefore therefore A, All a is c. E, No a is c. I, Some a is c. O, Some a is not c. Second Figure. E, No c is 6, A, All c is b, E, No c is b, A, All c is 6, A, All a is b, E, No a is 6, I, Some a is b, 0, Some a is not b, therefore therefore therefore therefore E, No a is c. E, No a is c. 0, Some a is not c. O, Some a is not c.
102 APPLICATION TO LOGIC. [CHAP. IV. Third Figure. A, All b is c, A, All b is a, therefore I, Some a is c. E, No b is c, A, All b is a, therefore O, Some a is not c. O, Some 6 is not c, A, All 6 is a, therefore O, Some a is not c. I, Some b is c, A, All b is a, therefore I, Some a is c. E, No 6 is c, I, Some b is or, therefore O, Some a is not c. A, All 6 is c, I, Some 6 is a, therefore 1, Some a is c. A, All c is 6, A, All b is a, therefore I, Some a is c. Fourth Figure. A, All c is b, I, Some c is 6, E, No 6 is a, A, All 6 is a, therefore therefore O, Some a is not c. I, Some a is c. E, No c is 6, I, Some b is a, therefore O, Some a is not c. E, No c is 6, A, All b is a, therefore 0, Some a is not c. (2) The first mood of the first figure can be symbolized thus : b 4 c, a^by therefore a = c : or thus : b = bc, a = ab, therefore a = ab= abc = ac : or thus : be = 0, ab = 0, therefore ac = a (b -f b) c = a. be + ab. c = 0 : or thus : c^b, b^a, therefore e^a: or thus : c = b+ c, b = a + b, therefore c=b + c = a + b + c = a+c: or thus: b = b + c, d = d+b, therefore a = a + i = d+b+c = d + c: or thus : b -f c = iy a + 6 = {, therefore a + c = x + 66 + c = (a + b 4- c) (a 4- b + c) = i: or thus: c = bc, b = dby therefore e = bc = dbe = ac. One half of these forms can be deduced from the other half by taking supplements. In each case the two premises, which are each of the type A, have been written down in the same form. By combining two different methods of exhibiting symbolically propositions of the type A many other methods of conducting the reasoning symbolically can be deduced. It is unnecessary to state them here. (3) The second mood of the first figure can be symbolized thus: be = 0, a = ab} therefore ac = abc = 0 : or thus : b = c, a = b, therefore a ^= c : or thus: c-^b, b^d, therefore c^d: or thus : b = bc, a = ab, therefore a = abc = ac : or thus : c cb, b = db, therefore c = cb= cab = cd :
50] SYMBOLIC EQUIVALENTS OF SYLLOGISMS. 103 or thus: 6 + c = iy ab = 0, therefore a = a (b + c) = ab + ac = ac : or thus: 6 = b -f c, a6 = 0, therefore a (b + c) = ac = ab = 0 : or thus: c = c -f 6, a = a , therefore ac = a (c + b) = ac + a = a. Eight forms have been given here but many others could be added by combining otherwise the modes of symbolizing propositions of the type A and E. 50. Symbolic Equivalents of Syllogisms. (1) It is better however at once to generalize the point of view of this symbolic discussion of the syllogism. It is evident that each syllogism is simply a problem of elimina- tion of the middle term, and the symbolic discussions can be treated as special cases of the general methods already developed. Also the symbolic equivalence of all the forms of a proposition makes it indifferent which special form of a proposition is chosen as typical. (2) Consider the first mood of the first figure: the term b is to be eliminated from b = bc, a = ab. The positive discriminant of b = bc, is c, the negative discriminant is i. The equation, a = ab, can be written ab + ab = ab. The positive discri- minant is i ; its negative, discriminant is a. Hence all the information to be found by eliminating b is ~ (ci) x ~ (id) = 0 ; that is ac = 0. (3) Consider the second mood : the term b is to be eliminated from be = 0, a = ab. The discriminants of the first equation are c and i; and of the second equation are i and a. Hence the elimination of b gives ~ (ci) x ~ (id) = 0 ; that is ac = 0. It is obvious that the first and second moods of the second figure are symbolically the same problem as this mood. (4) The third mood of the first figure is symbolically stated thus : b = bc, ab .j. Hence eliminating b by 41 (4), the existential resultant is ac .j. This is symbolically the same problem as the third and fourth moods of the third figure, and the third of the fourth figure. (5) The fourth mood of the first figure can be symbolized thus : be = 0, ab .j. Hence eliminating 6 by 41 (4), the existential resultant is ac .j. This is symbolically the same problem as the third mood of the second figure, the sixth of the third figure, and the fifth of the fourth figure.
104 APPLICATION TO LOGIC. [CHAP. IV. (6) The only mood in the second figure not already discussed is the fourth ; it can be symbolically stated thus: cb = 0, ab .j. Hence eliminating b by 41 (4), the existential resultant is ac.j. (7) In the first mood of the third figure a particular proposition is inferred from two universal premises. It is necessary therefore in order to symbolize this mood that universal propositions as symbolically expressed be put on the same level as particular propositions in regard to the ex- clusion of nugatory forms. The syllogism can be symbolized thus, bj = bc.j, bj=ba.jy hence bj= be .j = bac .j, hence ac .j. (8) It is immediately evident that the premises assume more than is necessary to prove the conclusion, thus b = be, instead of bj= be .j, and ab .j, instead of bj = ab.j, would have been sufficient. This is not a syllogism with what is technically known as a weakened conclusion, since no stronger conclusion of this type could have been drawn. It might be called a syllogism with over-strong premises. The syllogism of the same type with its premises not over-strong is the third of the first figure. Hence the symbolic treatment of that mood would serve for this one. (9) The second mood of the third figure can be symbolized thus, be 0, bj = ab .j, now ab .j = ab (c f c) .j = abc .j, hence ac .j. This is obviously a syllogism with over-strong premises, since be = 0, ab .j, would have been sufficient for the conclusion. The syllogism of the same type with sufficient premises is the fourth of the first figure. (10) The fifth mood of the third figure can be symbolized thus: be.j, a =0. Hence eliminating b by 41 (4), the existential resultant is ac .j. (11) The first mood of the fourth figure is symbolized thus, cj = be .j, bj = ab .j, hence be .j = abc .j, hence ac .j. This is a syllogism with over-strong premises, the corresponding syllogism with sufficient premises is the third of the first figure. (12) The second mood of the fourth figure is symbolized thus, ej = be .j, ab = 0, therefore be .j = be (a + a) .j = bca .j, hence ca .j. This is a syllogism with over-strong premises; the corresponding syllogism with sufficient premises is the fourth of the first figure. (13) The fourth mood of the fourth figure is symbolized thus, be = 0, bj = ab . j therefore ab ,j= ab (c + c) .j = abc .j, hence ac .j. This is a syllogism with over-strong premises; the corresponding syllogism with sufficient premises is the fourth of the first figure.
51] GENERALIZATION OF LOGIC. 105 (14) Since the conclusion of any syllogism can be obtained from the premises by the purely symbolic methods of this algebra, it follows that the conclusion of any train of reasoning, valid according to the formal canons of the traditional Deductive Logic, can also be obtained from the premises by the use of the algebra, using purely symbolic transformations. 51. Generalization of Logic. (1) This discussion of the various moods of Syllogism suggests1 that the processes of elimination and solution as applied to a system of equations and existential expressions developed in the preceding chapters of this Book can be construed as being a generaliza- tion of the processes of syllogism and conversion of common Logic. It will be seen by reference to 47 that a universal proposition is symbolized in the form of an equation, and a particular proposition in the form of an existential expression. Hence the most general form of equation may be conceived as a complex universal proposition, and a set of equations as a set of universal propositions. Also the most general form of an exis- tential expression is the most general form of a particular proposition, and a set of such expressions is a set of particular propositions. (2) The most general form of a system, entirely of universal proposi- tions and involving one element to be determined, is given in Chapter II, % 29, 30. It is arx 4- brX = ct-x + drx, anx + bnx = cnx + dnx. Here x is supposed to be the class to be further determined, and the other symbols all refer to well-known classes. Then the information wanted is found by forming n functions of the type, Ar = arcr + a7er, and n of the type, Br = brdr + brdr, and by forming the products A = AxA2... An, B = BVB2... Bn. Then x = B + uA ; with the con- dition that AB = 0, which is probably well-known. (3) The essential part of this process is the formation of the two regions A and B out of the well-defined regions involved in the system of proposi- tions. This composition of the two discriminants is a process of rearranging our original knowledge so as to express in a convenient form the fresh infor- mation conveyed in the system. Formally it is a mere picking out of certain regions defined by the inter-relations of the known regions which are the coefficients of the equations: but the process in practice may result in a real addition to knowledge of the true definition of x. For instance rationality and animality may have been the characteristics of two regions among the 1 Cf. Boole, Laws of Thought, chapter IX. 8, chapter XV.
106 APPLICATION TO LOGIC. [CHAP. IV. coefficients in the system ; but in A and B the common part of the regions may only occur: then it is at once known that x only involves the ideas of rationality and animality in so far as it involves those of humanity a very real addition to knowledge, though formally it is only a question of better arrangement as compared to the original system. (4) The undefined nature of the information given by particular pro- positions makes it usually desirable not to deal with such propositions in a mass, but to sort them one by one, comparing their information with that derived from the known system of universal propositions. Thus let the above system of universal propositions be known, and also the proposition of the type I, viz. ex . j. Then from 41 the full information to be found by eliminating x is, AB = 0, eA .j; and the solution for x is, either (1) eB.j, x = ~Be + uA, or (2) eB = 0, b=B + uA. 0l Now propositions including a common term x are in general accumulated in science or elsewhere just because information concerning x is required. Also the propositions will as far as possible connect x with thoroughly well- known terms. If we conceived this process as performed with ideal success, then the coefficients of x and x in the above equations and existential expres- sion must be conceived as completely known, and no information concerning their relations will be fresh. Hence in case (1), when eB .j, the particular proposition {ex .j) is included in the universal propositions; but in case (2) the particular proposition has added fresh information. But this sharp division between things known and things unknown is not always present in reasoning. In such a case the universals and the particular perform a double function, they both define more accurately the properties of things already fairly well-known, and determine the things x which are comparatively unknown. The discussion of this typical case may serve to exemplify the logical interpretation of the problems of the previous chapters.
CHAPTER V. Propositional Interpretation. 52. Propositional Interpretation. (1) There is another possible mode of interpreting the Algebra of Symbolic Logic which forms another application of the calculus to Logic. Let any letter of the calculus represent a proposition or complex of propositions. The propositions represented are to be either simple categorical propositions, or complexes of such propositions of one or other of two types. One type is the complex proposition which asserts two or more simple propo- sitions to be conjointly true ; such a proposition asserts the truth of all its simple components, and the proposer is prepared to maintain any one of them. The verbal form by which such propositions are coupled together is a mere accident: the essential point to be noticed is that the complex proposition is conceived as the product of a set of simple propositions, marked off from all other propositions, and set before the mind by some device, linguistic or otherwise, in such fashion that each single proposition of the set is stated as valid. Hence if one single proposition of the set be disproved, the complex proposition is disproved. Let such a complex of propositions be called a conjunctive complex. (2) The other type of complex proposition is that which asserts that one at least out of a group of simple propositions, somehow set before the mind, is true. Here again the linguistic device is immaterial, the essential point is that the group of propositions is set before the mind with the understood assertion that one at least is true. Let such a type of complex of propositions be called a disjunctive complex. (3) Furthermore we may escape the difficult (and perhaps unanswerable or even unmeaning) question of deciding what propositions are to be regarded as simple propositions. The simplicity which is here asserted of certain propositions, is, so to speak, a simplicity de facto and not de jure. All that is meant is that a simple proposition is one which as a matter of fact for the purpose in hand is regarded, and is capable of being regarded, as a simple
108 PROPOSITIONAL INTERPRETATION. [CHAP. V. assertion of a fact, which fact may be indefinitely complex and capable of further analysis. Thus a conjunctive or a disjunctive complex may each of them be regarded as a simple proposition by directing attention to the single element of assertion which binds together the different component propositions of a complex of either type. (4) To sum up: all propositions symbolized, actually or potentially, by single letters can be regarded as simple propositions: and the only analysis of simple propositions is to be their analysis either into conjunctive or disjunctive complexes of simple propositions. Also a simple proposition is a proposition which can be regarded as containing a single element of categori- cal assertion. 53. Equivalent Propositions. Two propositions, x and y, will be said to be equivalent, the equivalence being expressed by x = y, when they are equivalent in validity. By this is meant that any motives (of those motives which are taken account of in the particular discourse) to assent, which on presentation to the mind induce assent to x, also necessarily induce assent to y and conversely. 54. Symbolic Representation of Complexes. (1) Let the disjunctive complex formed out of the component propositions a, b, c... be symbolized by (a + 6-f-c ...). This symbolism is allowable since the disjunctive complex has the properties of addition : for (1) the result of the sjmthesis of the propositions is a definite unique thing of the same type as the thing synthesized, namely another proposition: (2) the order which is conceivable in the mental arrangement of the propositions is immaterial as far as the equivalence of the resulting proposition is concerned: (3) the components of a disjunctive complex may be associated in any way into disjunctive complexes; so that the associative law holds. (2) Let the conjunctive complex formed out of the component proposi- tions a, b, c... be symbolized by abc... This symbolism by the sign of multiplication is allowable: (1) since the result of the synthesis of a number of component propositions into a conjunctive complex is definite and unique, being in fact another proposition which can be regarded as a simple propo- sition; (2) since the conjunctive complex formed out of the proposition a and the complex b -f c is the same proposition as the disjunctive complex formed by ab and ac; in other words a (b + c) = ab + ac. 55. Identification with the Algebra of Symbolic Logic. (1) It now remains to identify the addition and multiplication of propositions, as here defined, with the operations of the Algebra of Symbolic Logic. The disjunctive complex x + x is the same as the simple proposition x.
55] IDENTIFICATION WITH THE ALGEBRA OF SYMBOLIC LOGIC. 109 For x + x means either the proposition x or the proposition x, and this is nothing else than the proposition x. Hence x + x = x. (2) The conjunctive complex obeys the associative law: for to assert a and b and c conjointly is the same as asserting b and c conjointly and assert- ing a conjointly with this complex assertion. Hence abc = a . be. (3) The conjunctive complex also obviously obeys the commutative law: thus abc = acb = bac. (4) The conjunctive complex formed of a and a is the same as the simple proposition a; hence aa = a. (5) The null-element of the manifold of the Algebra corresponds to the absolute rejection of all motives for assent to a proposition, and further to the consequent rejection of the validity of the proposition. Hence x = 0, comes to mean the rejection of x from any process of reason, or from any act of assertion. In so far as they are thus rejected all such propositions are equivalent. Thus if x = 0, y = 0, then x = y = 0. Furthermore if 6 = 0, the proposition a + b is equivalent to the proposition a alone; for the motives of validity of b being absolutely rejected, those for the validity of a alone remain. Hence if b = 0, a + b = a. Again, if b = 0, then x = 0; for ab means that a and b are asserted conjointly, and if the motives for b be rejected, then the motives for the complex proposition are rejected. The class of propositions to be thus absolutely rejected is best discussed later, after the discussion of the other special element. (6) The Universe. The other special element of the manifold is that which has been called the Universe. Those propositions, or that class of perhaps an indefinite number of propositions, will be severally considered as equivalent to the Universe when their validity has acquired some special absoluteness of assent, either conventionally (for the sake of argument), or naturally. This class of propositions may be fixed by sheer convention: certain propositions may be arbitrarily enumerated and to them may be assigned the absolute validity which is typified by the element called the Universe. Or some natural characteristic may be assigned as the discriminating mark of propositions which are equivalent to the universe. For instance, propositions which while reasoning on a given subject matter are implied in reasoning without rising to explicit consciousness or needing explicit statement at any stage of the argument might be equated to the Universe. The laws of thought as stated in Logic are such propositions. Again in a discussion between two billiard markers on a game of billiards the propo- sition, that two of the balls were white and the third red, might be of this
110 PROPOSITIONAL INTERPRETATION. [CHAP. V. character. For billiard markers such a proposition rises to the level of a law of thought. Again, in legal arguments before an inferior court the judgments of the Supreme Court of Judicature might be considered as propositions each equivalent to the Universe. In this interpretation the name of the Universe as applied to this element is unfortunate: the Truism would be a better name for it. Let all propositions equivalent to the Universe be termed self-evident. (7) The properties assigned to the Universe (i) in relation to any proposition x are (cf. 23 (6) and (7)) x -f { = i, xi = x. The validity of any proposition equivalent to the Universe being taken as absolute, the validity of the disjunctive complex formed of this proposition and some other proposition x cannot be anything else but the absolute validity of the Universe. Hence the equation x + i = i is valid for the present interpretation. Again, in the conjunctive complex formed of any proposition and a proposition equivalent to the Universe, the validity of the second proposition being unquestioned, the validity of the whole is regulated by that of the first proposition. Hence the equation xi = x is also valid. (8) This conception of a class of propositions either conventionally or naturally of absolute validity gives rise for symbolic purposes in this chapter to an extension of the traditional idea of the conversion of propositions. If the Universe be narrowed down to the Laws of Thought, then all the propositions which can be derived from any given proposition x taken in connection with the propositions of the Universe are those propositions which arise in the traditional theory of the conversion of propositions. Hence if we extend the Universe of self-evident propositions either by some natural or conventional definition, we may extend the conception of conversion to include any proposition which can be derived from a given proposition x taken in connection with the assigned propositions equivalent to the Universe. Thus if i be any proposition equivalent to the Universe, xi will be considered to be simply the proposition x in another form. (9) The supplementary proposition, x, of the proposition x is defined by the properties, xx = 0, x 4- x = i. Whatever the propositions of the Universe may be, even if they are reduced to the minimum of the Laws of Thought, the logical contradictory of x satisfies these conditions and therefore is a form of the supplementary pro- position. But by the aid of the propositions of the Universe there are other more special forms into which the contradictory can be f converted/
56, 57] SYMBOLISM OF THE TRADITIONAL PROPOSITIONS. Ill Any such form, equivalent to the contradictory, is with equal right called the supplement of x. Thus to the billiard markers cited above the supplement to the proposition, the ball is red, is the proposition, the ball is white; for one of the two must be true and they cannot both be true. (10) It is now possible clearly to define the class, necessarily of indefinite number, of propositions which are to be equated to the null element. This equation must not rest merely on the empirical negative fact of the apparent absence of motives for assent; but on the positive fact of inconsistency with the propositions which are equated to the Universe. If the Universe be reduced to the Laws of Thought, then all propositions equated to null are self-contradictory. With a more extended Universe, all propositions equated to null are those which contradict the fundamental assumptions of our reasoning. Let all propositions equated to the null-element be called self- condemned. (11) The hypothetical relation between two propositions x and y, namely, If y be true then x is true, implies that the motives for assent to y are included among those for assent to x. Hence the relation can be expressed by y^x, or by any of the equivalent equational forms of 26, Prop. vin. And y may be said to be incident in x. We have now examined all the fundamental principles of the Algebra of Symbolic Logic and shown that our present symbolism for propositions agrees with and interprets them all. Hence the development of this symbolism is simply the development of the Algebra which has been already carried out. 56. Existential Expressions. The symbol x.j denotes the pro- position x and implies that it is not self-condemned. The symbol x + o denotes the proposition x and implies that it is not self-evident. Hence, - (xj) = x + o , implies that the supplement of a proposition not self-con- demned is itself not self-evident. Umbral letters. The symbol xtj denotes the proposition x and implies that xy is not self-condemned: the symbol x + rj implies that x + y is not self- evident (cf. 40 (1)). The whole use of umbral letters therefore receives its interpretation. 57. Symbolism of the Traditional Propositions. (1) This system of interpretation, which in its main ideas is a modification of that due to Boole1, has perhaps the best right to be called a system of Symbolic Logic. It assumes the existence of an unquestioned sphere of knowledge, and traces generally the consequences which can be deduced from any categorical proposition or set of categorical propositions taken in connection with this sphere of knowledge. The former mode of interpretation, by class inclusion 1 Cf. Lmcs of Thought, chap. xi.
112 PROPOSITIONAL INTERPRETATION. [CHAP. V. and exclusion, only applied to propositions of the subsumption type : the present mode applies to any categorical proposition, that is to any proposition depending on a single element of assertion. Further it can symbolize any relation in which two such propositions can stand to each other, namely, (1) the disjunctive relation, in either of the two forms, namely, either when the propositions can be both true or when only one can be true (i.e. by the forms x + y and x + yx); (2) the conjunctive relation ; (3) the hypothetical relation (i.e. by the equation y = xy). (2) A defect of the method at first sight is that it seemingly cannot exhibit the process of thought in a syllogism. Thus if x and y be the two premises, and z be the conclusion, then z is true if xy be true : hence xy = xy. z, or xy ^ z are two of the forms in which an argument from two propositions to a third can be exhibited. But this symbolism only exhibits the fact that z has been concluded from xy, and in no way traces the course of thought. (3) The defect is remedied by McColl (Proc. London Math. Soc, Vols. IX., X., XI., XIII.), by means of the device of analysing a proposition of one of the traditional types, A, E, 7, 0, into a relation between other propositions thus instead of, All A is B, consider the propositions, It is A, It is jB; then, All A is B, is the same thing as saying that the proposition, It is A, is equivalent in validity to the conjunctive complex, It is A and It is B. Hence if one proposition is a, and the other 6, the original proposition is symbolized by a = ah. In other words, the hypothetical relation mentioned in 55 (11) holds between the propositions a and b. This analysis is certainly possible; and it is not necessary for the symbolism that it should be put forward as a fundamental analysis, but merely as possible. It requires however some careful explanation in order to understand the possible relations and transformations of such propositions as, It is A. 58. Primitive Predication. (1) Let a proposition of the type, It is Ay be called a primitive predication. In such a proposition the subject is not defined in the proposition itself; it is supposed to be known, either by direct intuition, or as the result of previous discourse. In the latter case the proposition must not be considered as an analytical deduction from previous propositions defining the ' it.' The previous discourse is simply a means of bringing the subject before the mind : and when the subject is so brought before the mind, the proposition is a fresh synthetic proposition. A primitive predication necessarily implies the existence of the subject. The proposition may be in error; but without a subject, instead of a proposition there is a mere exclamation. (2) If the predicate be a possible predicate, either because it is not self- contradictory, or further because its possibility is not inconsistent with the
58, 59] EXISTENTIAL SYMBOLS AND PRIMITIVE PREDICATION. 113 rest of knowledge, primitive predication can only be tested as to its truth or falsehood by an act of intuition. For a primitive predication is essentially a singular act having relation to a definite intuition; and it is only knowledge based on definite intuitions having concrete relations with this intuition which can confirm or invalidate it. The propositions taken as equivalent to the Universe in the present symbolism must be propositions deducible from propositions relating universal ideas or be such propositions themselves. Hence if x stand for a proposition which is a primitive predication, then x can only be self-condemned if the predicate be self-contradictory or inconsistent with the propositions equivalent to the Universe. Also x can only itself be equivalent to the universe, if there be the convention that during the given process of inference the ultimate subject of every proposition shall have certain assigned attributes. Then an act of primitive predication attributing one of these attributes to a subject is equivalent to the Universe, that is, is self-evident. (3) If x be a primitive predication, x is not a primitive predication; it may be called a primitive negation. Thus if x stands for, It is man, then x stands for, It is not man; that is to say, the subject may have any possible attribute except that of man. If x be self-condemned, then x states that the subject may have any possible attribute; thus x = i, since it is an obvious presupposition of all thought that a subject undefined except by the fact of an act of intuition may have any possible attribute. If x = i, then x is a denial that the subject referred to has a certain attribute, which by hypothesis all subjects under consideration do possess; hence x is self-condemned : that is, x 0. (4) A primitive negation does not necessarily occur merely as the denial of a primitive predication. The relations of the two types of proposition may be inverted. The fundamental proposition may be the denial that a certain predicate is attributable to the subjects within a certain field of thought. If this proposition, which relates universal ideas, be included among propositions which are self-evident, then any primitive denial which denies the certain predicate is also self-evident; and its supplement, which is a primitive predication, is self-condemned. 59. Existential Symbols and Primitive Predication. (1) If x stand for a primitive predication, then xj implies that the predicate is a possible predicate of a subject in so far as the self-evident propositions regulate our knowledge of possibility. Now xj implies x + co; this last expression implies that the denial of the primitive predication cannot be deduced to be true for all possible subjects of predication by means of the self-evident propositions. This deduction is an obvious consequence of xj. w. 8
114 PROPOSITIONAL INTERPRETATION. [CHAP. V. (2) Also xj implies that the denial of the primitive predication is con- sistent with the self-evident propositions as far as some possible subjects of predication are concerned. Now xj implies x -f co, and this implies that the primitive predication is not self-evident for all possible subjects of predi- cation. (3) If x, yy z, etc., all stand for separate primitive predications, then in any complex, either conjunctive or disjunctive, which comprises two or more of these propositions, the propositions are to be understood to refer to the same subject. Otherwise, since the propositions are singular acts, the pro- positions can have no relation to each other. Thus xy, i.e. x with y, stands for the combined assertions, It is X and it is F, or in other words, It is both X and F. Also x 4- y stands for, it is either X or F or both. Similarly primitive denials occurring together in a complex must both refer to the s^me subject; so also must primitive predications and primitive denials occurring together in a complex. (4) The symbol xrj stands for the proposition, It is X, and also implies the consistency with the self-evident propositions of the proposition, It is F, as applied to the same subject as x. The umbral letter r? affixed to x is in fact a reminder that xy is consistent with the self-evident propositions for some possible subjects of predication. 60. Propositions. (1) It is now possible to symbolize the traditional forms of logical proposition. Proposition A. All X is F, takes the form, if x then y, where x and y are the primitive predications, It is X, It is F. Hence the proposition takes the symbolic forms x = y} x = xy, or any symbolically equivalent form. (2) Proposition E. No X is F, takes the form, If x then y. Hence the proposition takes the symbolic forms x = y, x = xy, or any symbolically equivalent form. (3) Proposition I. Some X is F, takes the form that the conjunctive complex xy is not self-condemned; if the denial of all predicates or combina- tions of predicates, which do not actually occur in subjects belonging to the field of thought considered, be included among the self-evident propositions. Hence the proposition can be-put in the symbolic form, xy.j, or in any symbolically equivalent form. It must be carefully noticed that it is the connotation of xy.j which expresses the Proposition I and not the conjunctive complex xy, which stands for, It is X and F. Thus the supplement of xy.j, namely, ~{xy.j), or x + y -f , does not express the contradictory of the Proposition I, but the contradictory of the conjunctive complex xy. On the contrary the connotation of x + y + ) still expresses the same Proposition I.
60] PROPOSITIONS. 115 (4) Proposition 0. Some X is not Yy takes the form that the con- junctive complex xy is not self-condemned; where the same hypothesis as to the self-evident propositions is made as in the case of Proposition I. The symbolic form is therefore* xy .j, or any equivalent symbolic form. (5) The universal Propositions A, E as symbolized above give no existential import to their subjects. But the symbolism as there explained has the further serious defect that there is no symbolic mode of giving warning of the nugatoriness of the propositions when the subject is non- existent. But this can be easily remedied by including among the self- evident propositions the denial of any predicates which do not appear in an existent subject in the field of thought. This is the same supposition as had to be made in order to symbolize I and 0. Hence in the proposition x = xy, if there be no X's, then x = 0. Also if it be desired to exclude this nugatory case, then the proposition can be written (6) It has now been proved that the present form of interpretation includes that of the preceding chapter as a particular case. Thus all the results of the previous chapter take their place as particular cases of the interpretations of this present chapter. Historical Note. The Algebra of Symbolic Logic, viewed as a distinct algebra, is due to Boole, whose 'Laws of Thought7 was published in 1854. Boole does not seem in this work to fully realize that he had discovered a system of symbols distinct from that of ordinary algebra. In fact the idea of ' extraordinary algebras' was only then in process of formation and he himself in this work was one of its creators. Hamilton's Lectures on Quaternions were only published in 1853 (though his first paper on Quaternions was published in the Philosophical Magazine, 1844), and Grassmann's Ausdehnungslehre of 1844 was then unknown. The task of giving thorough consistency to Boole's ideas and notation, with the slightest possible change, was performed by Venn in his 'Symbolic Logic,' (1st Ed. 1881, 2nd Ed. 1894). The non-exclusive symbolism for addition (i.e. x+y instead of x + yx) was introduced by Jevons in his 'Pure Logic,' 1864, and by C. S. Peirce in the Proceedings of the American Academy of Arts and Sciences, Vol. vn, 1867. Peirce continued his investigations in the American Journal of Mathematics, Vols. in. and vn. The later articles also contain the symbolism for a subsumption, and many further symbolic investigations of logical ideas, especially in the Logic of Kelatives, which it does not enter into the plan of this treatise to describe. These investigations of Peirce form the most important contribution to the subject of Symbolic Logic since Boole's work. Peirce (loc. cit. 1867) and Schroder in his important pamphlet, Operationskreis des Logikkalkiils, 1877, shewed that the use of numerals, retained by Boole, was unnecessary, and also exhibited the reciprocity between multiplication and addition; Schroder {loc. cit.) also shewed that the operations of subtraction and division might be dispensed with. Schroder has since written a very complete treatise on the subject, 'Vorlesungen iiber die Algebra der Logik,' Teubner, Leipsic, Vol. I, 1890, Vol. n, 1891, Vol. in, 1895; Vol. in. deals with the Logic of Relatives. 8 2
116 PROPOSITIONAL INTERPRETATION. [CHAP. V. A small book entitled ' Studies in Deductive Logic,' Boston 1883, has in it suggestive papers, especially one by Miss Ladd (Mrs Franklin) ' On the Algebra of Logic,' and one by Dr Mitchell * On a new Algebra of Logic' A most important investigation on the underlying principles and assumptions which belong equally to the ordinary Formal Logic, to Symbolic Logic, and to the Logic of Relatives is given by Mr W. E. Johnson in three articles, * The Logical Calculus,' in Mind, Vol. i, New Series, 1892. His symbolism is not in general that of the Algebraic type dealt with in this work. The propositional interpretation in a different form to that given in this work was given by Boole in his book: modifications of it have been given by Venn (Symbolic Logic), Peirce (loc. cit.), H. McColl in the Proceedings of the London Mathematical Society, Vols. ix, x, xi, xiii, ' On the Calculus of Equivalent Statements.' The latter also introduces some changes in notation and some applications to the limits of definite integrals, which are interesting to mathematicians. A large part of Boole's ' Laws of Thought' is devoted to the application of his method to the Theory of Probability. Both Venn and Schroder give careful bibliographies in their works. These two works, Johnson's articles in Mind, and of course Boole's ' Laws of Thought,' should be the first consulted by students desirous of entering further into the subject. There is a hostile criticism of the utility of the whole subject from a logical point of view in Lotze's Logic.
BOOK III. POSITIONAL MANIFOLDS.
CHAPTER I. Fundamental Propositions. 61. Introductory. (1) In all algebras of the numerical genus (cf. 22) any element of the algebraic manifold of the first order can be expressed in the form axex + a2e2 + ... + avev, where e1} e2)... ev are v elements of this manifold and alf a2,... olv are numbers, where number here means a quantity of ordinary algebra, real or imaginary. It will be convenient in future invariably to use ordinary Roman or italic letters to represent the symbols following the laws of the special algebra considered: thus also each group of such letters is a symbol following the laws of the special algebra. Such letters or such group of letters may be called extraordinaries * to indicate that in their mutual relations they do not follow the laws of ordinary algebra. Greek letters will be strictly confined to representing numbers, and will in their mutual relations therefore follow all the laws of ordinary algebra. (2) The properties of a positional manifold will be easily identified with the descriptive properties of Space of any number of dimensions, to the exclusion of all metrical properties. It will be convenient therefore, without effecting any formal identification, to use spatial language in investigating the properties of positional manifolds. A positional manifold will be seen to be a quantitively defined manifold, and therefore also a complex serial manifold (cf. 11). (3) The fundamental properties which must belong, in some form or other, to any positional manifold must now be discussed. The investigation of 62 63 will be conducted according to the same principles as that of 14 18, which will be presupposed throughout. The present investigation is an amplification of those articles, stress being laid on the special properties of algebraic manifolds of the numerical genus. 62. Intensity. (1) Each thing denoted by an extraordinary, repre- senting an element of a positional manifold, involves a quantity special to it, to be called its intensity. The special characteristic of intensity is that in general the thing is absent when the intensity is zero, and is never absent * This name was used by Cayley.
120 FUNDAMENTAL PROPOSITIONS. [(JHAP. I. unless the intensity is zero. There is, however, an exceptional case discussed in Chapter IV. of this book. (2) Two things alike in all respects, except that they possess intensities of different magnitudes, will be called things of the same kind. They repre- sent the same element of the positional manifold, the intensity being in fact a secondary property of the elements of the manifold (cf. 9). (3) Let any arbitrary intensity of a thing representing a certain element be chosen as the unit intensity, then the numerical measure of the intensity of another thing representing the same element is the ratio of its intensity to the unit intensity. Let the letter e denote the thing at unit intensity, then a thing of the same kind at intensity a, where a is some number, will be denoted by ae or by ea, which will be treated as equivalent symbols. (4) Let the intensity of a thing which is absent be denoted by 0. Then by the definition of intensity, 0e = eO = 0. (5) Further, two things representing the same element at intensities a and j3 are to be conceived as capable of a synthesis so as to form one thing representing the same element at intensity a + /3. This synthesis is un- ambiguous and unique, and such as can be symbolized by the laws of addition. Hence ae + fie = (a + fi) e = (fi + a) e = fie + ae. The equation involves the formal distributive law of multiplication (cf. 19). Accordingly in the symbol ae, we may conceive a and e, as multiplied together. (6) Conversely a thing of intensity a + fi is to be conceived as analysable into the two things representing the same element at intensities a and fi. Then it is to be supposed that one of the things at intensity fi can be removed, and only the thing at intensity a left. This process can be con- ceived as and symbolized by subtraction. Its result is unambiguous and unique. Hence (a + fi) e fie = ae. (7) If corresponding to any thing ae there can be conceived another thing, such that a synthesis of addition of the two annihilates both, then this second thing may be conceived as representing the same element as the first but of negative intensity a [cf. 89 in limitation of this statement]. Thus ae + ( ae) = ae ae = Oe = 0. Complex intensities of the form a + ifi can also be admitted (i being s/ 1). It was explained in 7 that the logical admissibility of their use was altogether independent of the power of interpreting them.
62, 63] INTENSITY. 121 (8) Thus finally, if a, /3, 7, be any numbers, real or complex, and e an extraordinary, we have aBe + fihe ySe = (ctS -h /3S 78) e = S (a + /3 7) e = 8 (ae + /3s ye); also Oe = 0. All the general laws of addition and subtraction (cf. 14 18) can be easily seen to be compatible with the definitions and explanations given above. (9) It must be remembered that other quantities may be involved in a thing ae besides its own intensity. But such quantities are to be conceived as defining the quality, or character, of the thing, in other words, the element of the manifold which the thing represents; as for instance its pitch defines in part the character of a wrench. If any of these quantities alter, the thing alters and either it ceases to be capable of representation by any multiple of e, or e can represent more than one element [cf. 89 (2)]. 63. Things representing different elements. (1) Let e1} e2...ev denote v things representing different elements each at unit intensity. Let things at any intensities of these kinds be capable of a synthesis giving a resultant thing; and let the laws of this synthesis be capable of being sym- bolized by addition. Let a be the resultant of a^, a2e2, ... oivev; then a = a1e1 + a2e2 + a3e3+ ... + ctv ev. (2) By these principles and by the previous definitions of the present chapter, 2a = a + a = (a^ + a2e2 + ... + oivev) + (a^ + ... + oivev) = 2a1e1 + 2 x2e2 + ... + 2avev. Similarly if /3 be any real positive number, integral or fractional, /3a = /3a! eY + fi z2e2 + ... + avev. Let this law be extended by definition to the case of negative and complex numbers. Hence for all values of 0 fta = ft (a^i + a2e2 + ... VLvev) = $VLxex +/3a2e2+ ... + fiavev. Then 0a=0e1 4- 0e2 + ... + 0ev = 0. (3) The resultant of an addition is a thing possessing a character (in that it represents a definite element) and intensity of its own. The character is completely defined (cf. Prop. 11. following) by the ratios i : a2 : a3 ... : . Hence the intensities are secondary properties of elements according to the definition of 9. The comparison of the intensities of things representing different elements
122 FUNDAMENTAL PROPOSITIONS. [CHAP. I. may be possible. The whole question of such comparison will be discussed later in chapter IV. of this book. But it is only in special developments that the comparison of intensities assumes any importance: the more general formulae do not assume any definite law of comparison. (4) Definition of Independent Units. Let ely e2... ev be defined to be such that no one of them can be expressed as the sum of the rest at any intensities. Symbolically this definition states that no one of these letters, ely say, can be expressed in the form a2e2 -f *%ez + ... + avev. Then eY, e2... ev are said to be mutually independent. If e1} e2...ev are all respectively at unit intensity, then they are said to be independent units. Any one of them is said to be independent of the rest. 64. Fundamental Propositions. (1) A group of propositions* can now be proved; they will be numbered because of their importance and fundamental character. Prop. I. If ely e2... ev be v independent extraordinaries, then the equation, 0L + a2e2 + ... + avev = 0, involves the n simultaneous equations, ai = 0, a2 = 0 ... av = 0. Suppose firstly that all the coefficients are zero except one, a, say, then afa = 0. And by definition this involves ax = 0. Again assume that a number of the coefficients, including alf are not zero. Then we can write But this is contrary to the supposition that el9e2...ev are independent. Hence finally all the coefficients must separately vanish. Prop. II. If the two sums axeY + a2e2 +... + avev and ft^ + fte2 4-... + fivev are multiples of the same extraordinary, where ely e2... ev are independent extraordinaries, then aj/ = or2//32 = ... = av/l3v. For by hypothesis J3 + ft2e2 + ... + fivev = 7 (a^ + a2e2 + ... + avev). Hence (ft - 7^) eY + (ft - 7 2) e2 + ... + (ft - 7^) ev = 0. Therefore by Proposition I, ft 7^ = 0, ft 7a2 = 0 ... ft yotv = 0. Hence ft/o^ = ft/a*, = ... = $v\olv = 7. It follows [cf. 62 (2)], as has been explained in 63 (3), that the ratios of the coefficients of a sum define the character of the resultant, that is to say, the element represented by the resultant. Only it must be remembered that the extraordinaries have to be independent. (2) These propositions make a few definitions and recapitulations desirable. If two terms a and b both represent the same element, but at different intensities, then a and b will be said to be congruent to each other. The * Cf. Grassmann, Ausdehnungslekre of 1862; also De Morgan, Transactions of the Cambridge Philosophical Society, 1844.
64] FUNDAMENTAL PROPOSITIONS. 123 fact that the extraordinaries a and b are congruent will be expressed by a = b. This relation implies an equation of the form a = \b, where X is some number. The symbol = will also be used to imply that an equation, concerned solely with the quantities of ordinary algebra, is an identity. (3) The extraordinary O. + vl$2 + ... 4- avev will be said to be dependent on the extraordinaries e1} e2... ep; and the element represented by i0i + a2e2 + + aiA will be said to be dependent on the elements represented by elt e2...ev, An expression of the form 0L + a2e2 + ... + avev is often written Sae. (4) Let the v given independent extraordinaries be called the original defining extraordinaries, or the original defining units, if they are known to be at unit intensity. They define a positional manifold of v 1 dimensions (cf. 11). Any element of the type 2ae belongs to this manifold. This complete positional manifold, found by giving all values (real or complex) to alt a2,... av will be called the complete region. Any p of these v defining extraordinaries define a positional manifold of p 1 dimensions. It is incident in the complete region, and will be called a subregion of the complete region. A region or subregion defined by e2, e2... ep will be called the region or subregion (e1} e2... ep). (5) As far as has been shown up to the present, the v defining units represent elements which appear to have a certain special function and preeminence in the complete region. It will be proved in the succeeding propositions that this is not really the case, but that any two elements are on an equality like two points in space. (6) If letters a,b,c... denoting elements of the region be not mutually independent, then at least one equation of the form, aa + /?6-f 7C+ ... =0, exists between them, where a, /3, 7 ... are not all zero. Let such equations be called the addition relations between the mutually dependent letters. (7) Prop. III. An unlimited number of groups of v independent extra- ordinaries can be found in a region of v 1 dimensions. Let the region be (^, e2 ... ev). It is possible in an unlimited number of ways to find v2 numbers, real or complex, aly a,... aV) fa, j32... fiv... /c1} /c2... kv, such that the determinant ! alt a2 ... olv ' is not zero.
124 FUNDAMENTAL PROPOSITIONS. [CHAP. 1. Let a = ai k = Kxex + K2e2 4-... + tcvev. Now let |, ?;... % be numbers such that (if possible) Then substituting for a, 6 ... k, we find 17 A + ... + x 2) e* + Hence by Proposition I, the z coefficients are separately zero. But since the determinant written above is not zero, these v equations involve = 0, 17 = 0 ... 2 = 0. Hence the v letters a, b ... k are mutually independent. Prop. IV. No group containing more than v independent letters can be found in a region of v 1 dimensions. Let the region be defined by ely e2...ev, and let alt a2...av be v in- dependent letters in the region. Then by solving for elt e2... ev in terms of Oi, a2... av we can write a12a2 e2 = ev = x Now any other letter b in the region is of the form = 0A + Aft + ...+ * ; hence substituting for elf e2 ...ev in terms of a1} a2... avy b = 7^ + 72a2 + ... + 7^^. Thus 6 cannot be independent of a1? a2... a,,. Prop. V. If a1? a2... ap be p independent extraordinaries in a region of v 1 dimensions, where v is greater than p} then another extraordinary can be found in an infinite number of ways which is independent of the p independent extraordinaries. Let the complete region be defined by elf e2,... ev. Then the expressions for aly a2... ap in terms of the units must involve at least p of the defining extraordinaries with non-vanishing coefficients in such a way that they cannot be simultaneously eliminated. For if not, then the defining extra- ordinaries involved define a region containing a1? a2,... ap and of less than p 1 dimensions. But since al7 a,2) ... a9 are independent, by Prop. IV. this is impossible.
64, 65] FUNDAMENTAL PROPOSITIONS. 125 Let the extraordinaries e1} e2,... ep at least be involved in the expressions for alt a2} ... ap. Then by solving for elt e2... ep, we have Let b be any other letter, defined by b= % + % +...+ %vev. Substituting for e^ e2... #p, 6 = ^Oi + ... + rjpap + ?p+iep+1 + .. where any one of the "s, say ,, is of the form Thus there are i undetermined numbers, gl9 lj2... gv, and i; p coefficients of ep+1, ep+2... ev (viz. fp+1,... fv). Hence it is possible in an infinite number of ways to determine the numbers so that all these coefficients do not simul- taneously vanish. And in such a case b is independent of the group ax... ap. Corollary. By continually adding another independent letter to a group of independent letters, it is obvious that any group of independent letters can be completed so as to contain a number of letters one more than the number of dimensions of the region. (8) By the aid of these propositions it can be seen that any group of v independent extraordinaries can be taken as defining the complete region. The original units have only the advantage that their unit intensities are known (cf. chapter IV. following). Any group of v independent elements which are being used to define a complete region will be called coordinate elements of the region, or more shortly coordinates of the region. 65. Subregions. (1) The definition of a subregion in 64 (4) can be extended. A region defined by any p independent letters lying in a region of v 1 dimensions, where p is less than v, is called a subregion of the original region. The original region is the complete region, and the subregion is incident in the complete region. (2) A region of no dimensions consists of a single element. It is analogous to a point in space. A subregion of one dimension, defined by a, bt is in its real part the collection of elements found from fa + rjb, where f/77 is given all real values. Hence it contains a singly infinite number of elements, which will be called real elements when a and b are considered to be real elements. It is analo- gous to a straight line. And like a straight line it is given an imaginary extension by the inclusion of all elements found by giving f/77 all imaginary values. A subregion of two dimensions, defined by a, b, c, is the collection of
126 FUNDAMENTAL PROPOSITIONS. [CHAP. I. elements found from %a + r]b + c, where / , t?/ are given all values. Hence its real part contains a doubly infinite number of elements. It is analogous to a plane of ordinary space. Similarly a subregion of three dimensions contains in its real part a trebly infinite number of elements, and is analogous to space of three dimensions; and so on. (8) A subregion is called a co-ordinate region when, being itself of p 1 dimensions, p of the co-ordinate (or reference) elements of the complete region have been taken in it. For example, if eu e2...ev be the co-ordinate elements, then the region defined by elt e2... ep is a co-ordinate region of p 1 dimensions. (4) The complete region being of v 1 dimensions, to every co-ordinate subregion of p 1 dimensions there corresponds another co-ordinate subregion of (y p 1) dimensions, so that the two do not overlap, and the co-ordinate elements of the two together define the complete region. Such co-ordinate regions will be called supplementary, and one will be said to be supplementary to the other. (5) If two co-ordinate subregions of p 1 and r 1 dimensions do not overlap, then there is in the complete region a remaining co-ordinate sub- region of {y p cr 1) dimensions belonging to neither; where of course p + a is less than v. If the co-ordinate regions do overlap and have a common subregion of r 1 dimensions, then the remaining co-ordinate subregion is of (y + t - p a 1) dimensions; also the common subregion of t-1 dimensions must be a co-ordinate region. If p + a is greater than v, then the subregions must overlap and have in common a subregion of at least (p + a v 1) dimensions. (6) Let two regions of p 1 and a 1 dimensions overlap and have in common a subregion of t 1 dimensions. Let the subregion be defined by the terms alt a2... aT- Then the region of p 1 dimensions can be defined by the p terms Oi, a2... aTy 6T+1... 6P; and the region of cr 1 dimensions by the cr terms d1} d2 ... Qjj , CT_j_! ... Cff'y where the terms 6T+1... bp are independent of the terms cT+1 ... ca. For if the 6's and the c's do not together form a group of independent letters then another common letter aT+1 can be found independent of the other a's and can be added to them to define a common subregion of t dimensions. The region defined by the letters alt a2... aT} bT+1... 6P, cp+1... ca is called the containing region of the two overlapping regions; and is of r 1) dimensions.
65] SUBREGIONS. 127 It is the region of fewest dimensions which contains both regions as sub- regions, whether the regions do or do not overlap. (7) Every complete region and every subregion can be conceived of as a continuous whole. For any element of a subregion can be represented by and by a gradual modification of the values of the coefficients x can be gradually altered so as to represent any element of the region. Hence x can be conceived as representing a gradually altering element which successively coincides with all the elements of the region. The region can always there- fore be conceived as continuous. (8) Also the subregions must not be conceived as bounding each other. Each subregion has no limits, and may be called therefore unlimited. For any region is an aggregation of elements, and no one of these elements is more at the boundary or more in the midst of the region than any other element. Overlapping regions are not in any sense bounded by their common subregion. For any subregion of a region may be common to another region also. Regions, therefore, are like unlimited lines or surfaces, either stretching in all directions to infinity or returning into themselves so as to be closed ; two infinite planes cutting each other in a line are not bounded by this line, which is a subregion common to both. (9) Consider the one-dimensional subregion defined by the two elements ! and a2. Any extraordinary x which represents an element belonging to the subregion is of the form ^^4- %2a2. As f2/f-i takes all real positive values from 0 to + oo , # may be conceived as representing a variable element travelling through a continuous series of elements arranged in order, starting from ax and ending at a2. Again, as 2/ i takes all real negative values from oo to 0, x may be conceived as travelling through another continuous series of elements arranged in order, starting from a2 and ending in ax. It is for the purposes of this book the simplest and most convenient sup- position to conceive a one-dimensional subregion defined by two elements as formed by a continuous series of elements arranged in order, and such that by starting from any one element a1 and proceeding through the continuous series in order a variable element finally returns to ax. This supposition might be replaced* in investigations in which the object was to illustrate the Theory of Functions by another one. The element x starts from ax and passes through the series of elements given by 2/ffx positive and varying from 0 to x , and thus reaches a2; then as ^/^ varies from oo to 0, a? passes through another series of elements and finally reaches an element which in our symbolism is a2. But a1} as thus arrived at, may be conceived to be a different element from the element ax from which x started. * Cf. Klein, Nicht-Euklidische Geometrie, Vorlesungen, 1889 1890.
128 FUNDAMENTAL PROPOSITIONS. [CHAP. I. This conception has no natural symbolism in the investigations of this work, and therefore will not be adopted; but in other modes of investigation it is imperative that it be kept in view. Call this second ax the aY of the first arrival, and denote it by ^ Similarly we might find an a, of the second arrival and denote it by 2ax, and so on. Finally the analysis might suggest the identification of the ax of the pth arrival with the original aY. Thus a,! = pax. It is sufficient in this treatise simply to have noticed these possibilities. 66. Loci. (1) A locus is a more general conception than a subregion; it is an aggregation of a number (in general infinite) of elements deter- mined according to some law. Thus if x denote the element f^i + 2e2 + ... %vev, in the region defined by e1} e2...ev, then the equation j ( 1? f2 ... ) = 0, where f is a homogeneous function, limits the arbitrary nature of the ratios fi 2 : (v- The corresponding values of x form therefore a special aggre- gation of the elements out of the whole number in the region. But these elements, except in the special case of flat loci (cf. subsection (5) of this section), do not form themselves a subregion, according to the definition of a subregion given in this book; they are parts of many subregions. (2) A locus may be defined by more than one equation: thus the equations fa( ... ,) = 0, fa( ... ) = 0, ..., P (ft ... fv) = 0, where the left-hand sides are all homogeneous, define a locus when treated as simul- taneous. If there be v 1 independent equations, they determine a finite definite number of elements; and more than v 1 equations cannot in general be simultaneously satisfied. (3) A locus defined by p simultaneous equations will be said to be of v p 1 dimensions when the case is excluded in which the satisfaction of some of the equations secures that of others of the equations. In a region of v 1 dimensions there cannot be a locus of more than v 2 dimensions, and a locus containing an infinite number of elements must be of at least one dimension. Hence such a locus cannot be defined by more than v 2 equations. In a locus of one dimension the number of elements is singly infinite: in a locus of two dimensions it is doubly infinite, and so on. (4) Let p + r equations define a locus of v p r 1 dimensions. These equations may be split into two groups of p equations and of a equations respectively. The group of p equations defines a locus of v p 1 dimensions, and the group of a equations defines a locus of v a 1 dimen- sions. The original locus is contained in both these loci. Hence the locus of v p a 1 dimensions may be conceived as formed by the intersection of two loci of v p 1 dimensions and of v a 1 dimensions respectively. Similarly these intersecting loci can be split up into the intersections of other loci of higher dimensions. So finally the locus of v p 1 dimensions
66] loci. 129 may be conceived as the intersection of p loci of v 2 dimensions; each of these loci being given by one of the simultaneous equations. (5) The locus corresponding to an equation of the first degree, namely, is also a subregion of v 2 dimensions, as well as being a locus of the same number of dimensions. For if x1} x2... xv be v elements in the locus, given by then the v equations of the type, ttifiii + 2 2 + ... + CLvgpv = 0, involve the vanishing of the determinant S2 vl v2 v\ Hence an addition relation of the form, \1x1 + A^ + ... + \vxv = 0, exists between the elements. The v elements are therefore not independent. But v 1 independent points can be determined. Again, since the equation of the locus is linear, if xx and x2 be two elements in it, then A,^ + fxx2 also lies in the locus. Hence the whole region of the v 1 independent elements is contained in the locus, and vice versa; the region and locus coinciding. A locus defined by p simultaneous independent linear equations can in the same way be proved to contain groups of v p independent elements. It therefore in like manner can be proved to be a subregion of v p 1 dimensions. Let such a locus be called ' flat.' Then a flat locus is a sub- region. (6) A locus defined by p + a equations, of which p are linear, can therefore be treated as a locus of v p a 1 dimensions in a region of v p 1 dimensions. (7) There is a great distinction between the region defined by the com- bined elements which define subregions and the flat locus determined by the simultaneous satisfaction of the equations of other flat loci. Consider for instance, in a complete region of two dimensions, the two subregions defined by eu e2 and e3, e4 respectively. The region defined by the four elements e1} e2, e3y e4 includes not only the elements of the subregion e y of the form i0i + a and of the subregion e3ei} of the form ^3e3 + ^e4; but also it includes all elements of the form w.
130 FUNDAMENTAL PROPOSITIONS. [CHAP. I. which includes elements not lying in the subregions. But the equations of the loci taken together indicate a locus which is the region (in this case a single element) common to the two regions e y 03e4. 67. Surface Loci and Curve Loci. (1) Let a locus which is of one dimension less than a region or subregion containing it be called a surface locus in this region. Let this region, which contains the surface locus and is of one dimension more than the surface locus, be called the containing region. In other words [cf. 66 (6)] a surface locus can be defined by p +1 equations, of which only one at most is non-linear and p (defining the containing region) are linear. A surface locus defined by an equation of the fith degree will be called a surface locus of the /ith degree. For example, in this nomen- clature we may say that a surface locus of the first degree is flat. (2) In reference to a complete region of v 1 dimensions a flat surface locus of v 2 dimensions will be called a plane. A flat locus of v 3 dimen- sions will be called a subplane in a complete region of v 1 dimensions. Subplanes are planes of planes. (3) A subregion is either contained in any surface locus of the complete region or intersects it in a locus which is another surface locus contained in the subregion. Let the complete region be of v 1 dimensions and the subregion of v p \ dimensions. Then the subregion is a flat locus defined by p equations. The intersection of the subregion and the surface locus is there- fore defined by p 4-1 equations and is therefore of v p 2 dimensions. Also it lies in the subregion which is of v p 1 dimensions. Hence it is a surface locus; unless the satisfaction of the p equations of the flat locus also secures the satisfaction of the equation of the surface locus. In this case the subregion is contained in the original surface locus. (4) A locus of one dimension either intersects a surface locus in a definite number of elements or lies completely in the surface locus. For if the complete region be of v 1 dimensions, the locus of one dimension is determined by v 2 equations; and these together with the equation of the surface locus give v 1 equations which, if independent, determine a definite number of elements. If the equation of the surface locus does not form an additional independent equation, it must be satisfied when the equations of the one dimensional locus are satisfied; that is to say the one dimensional locus lies in the surface locus. A locus of one dimension will be called a curvilinear locus. A flat curvi- linear locus is a region of one dimension, and will be called a straight line. (5) A locus of t 1 dimensions which cannot be contained in a region of t dimensions (i.e. which is not a surface locus) will be called a curve locus. Thus if the locus be determined by p 4- a equations, of which p are linear and a non-linear, then a 1; and the locus is of z p r 1 dimensions.
67] SURFACE LOCI AND CURVE LOCI. 131 (6) A curve locus formed by the intersection of p surface loci, each in the same containing region, and such that it cannot be contained in a region of fewer dimensions than this common containing region, will be called a curve locus of the (p l)th order of tortuosity contained in this region. Thus the order of tortuosity of a surface locus is zero. Each of the surface loci, which form the curve locus by their intersection, will be called a containing locus. (7) In general the intersection of a curve locus of any order of tortuosity with a subregion is another curve locus contained in that subregion and of the same order of tortuosity. For the curve locus may be conceived as denned by p equations of the first degree which define the containing region of v p 1 dimensions; and by a equations each of a degree higher than the first degree, which define the tortuosity. Now the subregion which intersects the curve locus is defined by r equations of the first degree in addition to the equations defining the containing region. These p + r equations of the first degree and the a equations $! = 0,... f a = 0, in general define a curve locus of the (a l)th order of tor- tuosity in the subregion. (8) A plane curve in geometry is a surface locus; a plane curve of the first order of tortuosity consists of a finite number of points. In three dimensions an ordinary surface is a surface locus, an ordinary tortuous curve is a curve locus of the first order of tortuosity; and a finite number of points not in one plane form a curve locus of the second order of tortuosity. There are of course exceptional cases in relation to the tortuosity of curve loci when the equations are not all independent. It is not necessary now to enter into them. Note. The analytical part of this chapter follows closely the methods of Grass- mann's Ausdehnungslehre von 1862, chapter I. 1 36. This theory of Grassmann is a generalization of Mobius' Der Barycentrischer Calcul (1827), in which the addition of points is defined and considered. Hamilton's Quaternions also involve the same theory of the addition of extraordinaries (the number of independent extraordinaries being however limited to four). This theory is considered in his ' Lectures on Quaternions' (1853), Lecture I, and in his ' Elements of Quaternions,' Part I. The idea in Hamilton's works was a generalization of the composition of velocities according to the parallelogram law. Hamilton's first paper on Quaternions was published in the Philosophical Magazine (1844); De Morgan in his last paper, ' On the Foundation of Algebra' (loc. cit. 1844) writes of it, ' To this paper I am indebted for the idea of inventing a distinct system of unit-symbols, and investigating or assigning relations which define their mode of action on each other.' Some simple ideas which arise in the study of Descriptive Geometry of any number of dimensions have been discussed in 66, 67 as far as they will be wanted in this treatise. On this subject Cayley's ' Memoir on Abstract Geometry,' Phil. Trans. Vol. clx. 1870 (and Collected Mathematical Papers, Vol. vi. No. 413), should be studied. It enters into the subject with a complete generality of treatment which is not necessary here. 9 2
CHAPTER II. Straight Lines and Planes. 68. Introductory. The theorems of Projective Geometry extended to any number of dimensions can be deduced as necessary consequences of the definitions of a positional manifold. Grassmann's 'Calculus of Extension' (to be investigated in Book IV.) forms a powerful instrument for such an investigation; the properties also can to some extent be deduced by the methods of ordinary co-ordinate Geometry. Only such theorems will be now investigated which are either useful subsequently in this treatise or exemplify in their proof the method of the addition of extraordinaries. 69. Anharmonic Ratio. (1) Any point p on the straight line aa can be written in the form %a + %'a, where the position of p is defined by the ratio / '. If px be another point, ^a + i'a', on the same line, then the ratio f '/lr'fi is called the anharmonic ratio of the range (aa\ ppi)- It is to be carefully noticed that the anharmonic ratio of a range of four collinear elements is here defined apart from the introduction of any idea of distance. It is also independent of the intensities at which a and a happen to repre- sent their elements. For it is obviously unaltered if a, a are replaced by aa, aV, a and a' being any arbitrary quantities. (2) If the anharmonic ratio of (aa} ppY) be 1, the range is said to be harmonic ; and p and px can then be written respectively in the forms and a- V. (3) Let p1} p2, p3, p4 be any four points, ^a + iW, etc. Then where fogs') stands for the determinant Similarly a' = fop2 - Hence ps = + V = {( /) Pl - and Pi = {( ) Pl - ( /) p2}/( 2'). Hence the anharmonic ratio of the range (pxp2, pspt) is
68 71] HOMOGRAPHIC RANGES. 133 70. Homographic Ranges. (1) Let ( Apil sPs---) and (c q^qz...) be two ranges of corresponding points such that the anharmonic ratio of the four points {bYb y pppp+1), and that of the corresponding points (dc2, qPqP+i) are equal, where p is any one of the suffixes 1, 2, 3, etc. (2) It can now be proved that the anharmonic ratio of any four points (P^Pm PpPo) of the first range is equal to that of the corresponding points (q\qn qPq*) of the second range. For let pp = p62 + ff/fta, qp = i)pcY + Vp'c*. Then by definition fP?'p+i/fp+ifi ' = VpVp+i/Vp+iVp- Now replace p in turn by /o + l, p + 2, ...cr 1, and multiply together corresponding sides of the equations, so obtained. Finally we deduce and hence by subtracting 1 from both sides, It follows that the anharmonic ratio of any four points (p^p^, pPpo) of the first range is equal to that of the four corresponding points (q^q^, qPq r) of the second range. Such corresponding ranges are called homographic. 71. Linear Transformations. (1) Let el and e2 be any two points, and let be three given points and any fourth point on one range of points; also let be the corresponding points on a second range homographic to the first range. Then (aX) (?o/)/0W) (? ) = 08.A') W*')l{Pz ) ivfr'). Therefore / ' and rj/rj' are connected by a relation of the form X +/ifr'+A*T^ + *TV = O.....................(A), where X, /a, /a', \' are constants depending on the arbitrarily chosen points (h % as, bu b2, b3. This equation can also be written in the form ttug + ttiag7 _ Qf2ig + Oaaf _ ^ ../x ^ " v ~P........................{ )j where an, a12, a21, a^ are constants which determine the nature of the trans- formation, and p must be chosen so that the point q = V ' may have the desired intensity.
134 STRAIGHT LINES AND PLANES. [CHAP. II. Such transformations as those represented algebraically by equations (A) or (A') are called linear transformations. Only real transformations will be considered, namely those for which the coefficients an, a12, a^, i22 of equation (A') are real. (2) There are in general two points which correspond to themselves on the two ranges. For by substituting , ' for 77, rj in the above equations and eliminating we find ( *n- p) (a^- p) - ct12ct21 = 0...........................(B), an equation which determines two values of p real and unequal or real and equal or imaginary; and each value of p determines : ' and rj : tj' uniquely. (3) Let p1 and p2 be the two roots of this quadratic, and first let them be assumed to be unequal. Then by substituting p1 in one of the equations (A') a self-corresponding point dx is determined by / ' = a12/(p! - an) = (/h - a^/asa. Similarly a self-corresponding point d2 is determined by / ' = au/(p* - n) = (P2 - ofe)/**. Let these self-corresponding points be the reference points, so that any point is determined by gdi + %'d2. Then the equations denning the transformation take the form px p T T p By putting v for pi/p2, this equation can be written (4) Linear transformations fall into three main classes, according as the roots are (1) real and unequal, (2) imaginary, (3) equal. In transformations of the first class the two points dx and d2 are real. Then v is real, and is positive when any point in the first range and its corresponding point on the second range both lie between d1 and d2. (5) In transformations of the second class the two points dY and d2 are imaginary. Then v is complex, and it can be proved that mod. v 1, assuming that real points are transformed into real points. For p1 and p2 are conjugate complexes, and can be written ere** and re~ . Accordingly v = pY\p2 = e2^. Hence mod. v = 1, and log v 2i$; where 8 is real. (6) The linear transformations of the first class are called hyperbolic ; and those of the second class elliptic.
71] LINEAR TRANSFORMATIONS. 135 (7) The third special class of linear transformations exists in the case when the roots of the quadratic are equal, that is to say when the two points ctj and a2 coincide. Linear transformations of this class are called parabolic. The condition for this case is that in equation (A), modified by substituting | and f for rj and r{ respectively, the following relation holds between the coefficients, Let u be the double self-corresponding point and e any other point, and let these points replace e1 and e2 in subsection (1) above. Then the modified equation (A), regarded as a quadratic in : ', must have two roots infinite: hence X Therefore if a point P( be transformed into q (= tju + rfe), equation (A) takes the form that is ( /( ' = rj/rjf -f constant..............................(D). (8) By a linear transformation a series of points pu p2i p3 ... can be determined with the property that the range (plt p2i p3...) is homographic with the range (p2,p3,p4l ...) Firstly, let the linear transformation be elliptic or hyperbolic and let the co-ordinate points e1} e2 be the pair of self-corresponding points of the two ranges. Let p1 = fa + %'e2. Then if v be any arbitrarily chosen constant, the points Pi = vfa + Z'e2, p3 = ififa + ?e2,... pp = vp-lfa + Z'e2 satisfy the required condition. (9) Secondly, let the linear transformations be parabolic. Let u be the double self-corresponding point, and let e be another arbitrarily chosen reference point. Let pi = gu + e} and let 8 be the arbitrarily chosen constant of the trans- formation. Then by equation (D) the other points of the range are successively given by These results will be found to be of importance in the discussion in Book vi., chapter I., on the Cayley-Klein theory of distance.
136 STRAIGHT LINES AND PLANES. [CHAP. II. 72. Elementary Properties. (1) Let the v independent elements e1,e2,...ev define the complete region of v 1 dimensions. Then the p elements el9 e2, ... ep (p v) define a subregion of p 1 dimensions, Any point in this region can be written in the form Thus any point on the straight line defined by e1} e2 can be written and any point on the subregion of two dimensions (or ordinary geometrical plane) defined by el9 e2, e3 can be written 51^1 "t" b2^2 "f" ?3^3" (2) Any p + 1 points a^, x2...xp+1 in the subregion e1 e2...ep can be connected by at least one equation of the form Let such an equation be called the addition relation between the de- pendent points oolf #2 #p+i Thus any three points wlf x2i xs on a straight line satisfy an equation of the form i#i + ^2^2 + 3#3 = 0; and similarly for any four points on a two-dimensional subregion. (3) If 0i, e2...ev be the independent reference units of the complete region, and any point be written in the form 2 e, then the quantities are called the co-ordinates of the point. The locus denoted by is a plane (i.e. a subregion of v 2 dimensions). The intersection of the p planes (p v) is a subregion of v p 1 dimensions. (4) The intersection of v 1 such planes is a single point which can be written in the form 'lj 02 , . . . 0|/, L12 ^22 * * ^^2 Ll, v 1 j ^-2, v 1 ",v, V J
72] ELEMENTARY PROPERTIES. 137 For instance, in the region e^e^ of two dimensions, the two straight lines intersect in the point (^ 21^-32 ~- ^31^22) #1 " " (^31^-12 ~" ^-11^-32.) #2 + (^11^-22 "" ^21^12) #3- Returning to the general case of v 1 planes, it is obvious that their point of intersection lies on the plane i?i+aif2 + ...+arfi, = ( if the determinant !, a2, ... aVi 11 y ^21 "-vD vanishes. (5) To prove that it is in general impossible in a complete region of v 1 dimensions to draw a straight line from any given point to intersect two non-intersecting subregions of p - 1 and a 1 dimensions respectively, where p and r are arbitrarily assigned; and that, when it is possible, only one such straight line can be drawn. Since the subregions are non-inter- secting [cf. 65 (5)] p + a v. Of the reference elements let p be chosen in the subregion of p 1 dimensions, namely ji,j2,... jp, and let r be chosen in the subregion of a 1 dimensions, namely k1} h2i... ka, and v p cr must be chosen in neither region, namely ely e2}... e,,_p_ff. Let the given point be p = Xaj + %/3k + 2ye. Let 2 ?' be any point in the subregion ji^-jp- Then jp + XS0 can be made to be any point on the line joining p and X%j by properly choosing X. But if this line intersect the subregion kx... kO9 then for some value of X, say Xx, p + \^j depends on hY... ha only. Hence either yx = y2 = ... = yv-p-a = 0, in which case p cannot be any arbitrarily assigned point; or p + a = v, and there are no reference points of the type Hence we find the condition p + a = v. Again p + X^^j = S/3 . Hence also f2 = al9 2 = a2, ... ( = a,,, and X2 = 1. Thus the line through p intersecting the two regions intersects them in Xu.j and S/9 respectively. Accordingly there is only one such line through any given point p.
138 STRAIGHT LINES AND PLANES. [CHAP. II. 73. Reference-Figures. (1) The figure formed in a region of v 1 dimensions by constructing the straight lines connecting every pair of v independent elements is the analogue of the triangle in plane geometry and of the tetrahedron in space of three dimensions, the sides of the triangle and edges of the tetrahedron being supposed to be produced indefinitely. Let such a figure be called a reference-figure, because its corner points can be taken as reference points to define the region. Let the straight lines be called the edges of the figure. (2) Let ex, e2,... ev be the corners of such a figure, and let them also be taken as reference points. Consider the points in which any plane, such as cuts the edges. The point in which the edge e^ is cut is found by putting , = () = =... = ,. Hence the point is axex a2e2. Similarly the point in which the edge epea is cut is apep aoea. (3) Consider the points of the typical form a9e? + aoe9. Then the range ep, e r, apep-aaea a.?ep + a. ,ev is harmonic. Also any point on the plane defined by the point a^ and the remaining corners of the reference-figure, namely by e3 e4i... ev is Hence all such planes pass through the point ai^i + 262-f a3e3+... + avev. And conversely the planes through this point and all the corners but two cut the edge defined by the remaining two corners in the points The harmonic conjugates of these points with respect to the corresponding corners are of the typical form apep a^, and these points are coplanar and lie on (4) The point %ae may be called the pole of the plane 2 /a = 0 with respect to the given reference-figure, and the plane may be called the polar of the point. These properties are easily seen to be generalizations of the familiar properties of triangles (cf. Lachlan, Modern Pure Geometry, 110). (5) Let points be assumed one on each edge, of the typical form ap rep -f aapea. It is required to find the condition that they should be coplanar. Consider the v 1 edges joining the corner eY to the remaining corners. There are v \ assumed points of the typical form
73, 74] REFERENCE-FIGURES. 139 on such edges, and these points define a plane. It remains, therefore, to determine the condition that any other point of the form (where neither X nor /jl is unity) lies on this plane. It must be possible to choose 2, 3,,.. and rj so as to fulfil the condition V 2 a {oi1 Je1 + aalea) + y (aAfAeA + a^) = 0. r=2 This requires that the coefficients of elt e2,...ev should be separately zero. Hence if a be not equal to X or //,, we find a = 0; and also the three equations Hence But this is the condition that three points on the edges joining elf eKy e^ should be collinear. Hence since elt eky e^ are any three of the corners of the given reference-figure, the necessary and sufficient condition is that the assumed points lying on the edges which join any three corners in pairs should be collinear. (6) It follows from (3) of this section that the condition for the con- currence of the planes joining each assumed point of the form with the corners, not lying on the edge on which the point itself lies, is for the three points on the edges joining in pairs Hence if any three edges be taken forming a triangle with the corners as vertices, the three lines joining each assumed point with the opposite vertex are concurrent. 74. Perspective. (1) The perspective properties of triangles can be generalized for reference-figures in regions of v 1 dimensions *. Let ej j... ev and ei'e2'... ej be two reference-figures, and let the v lines be concurrent and meet in the point g. Then it is required to show that the corresponding edges are concurrent in points which are coplanar. Since g is in e^', e2e^t ... evev\ it follows that A i#i H~ A.j 6\ == ^2^2 i Arfj #2 == == A*y#|/ "T A/y v == g. Hence X^ X^e2 = X2V~ V^i' w^n similar equations. But X1e1 X2e2 is on the edge eYe2, and X/e/ X2/e2/ is on the edge e*[e*. But these are the same point. Hence the edges eYe2 and e-[e2 are concurrent in this point. * The theorems of subsections (1) to (4) of this article, proved otherwise, were first given by Veronese, cf. "Behandlung der projectivischen Verhaltnisse der Raume von verschiedenen Dimen- sionen durch das Princip des Projicirens und Schneidens," Math. Annalen, Bd. 19 (1882).
140 STRAIGHT LINES AND PLANES. [CHAP. II. But taking exe2... ev as reference points,it has been proved that the points of the typical form X^ X2e2 are coplanar and lie on the plane, 2 /X = 0. Hence the theorem is proved. (2) Conversely, if the corresponding edges, such as eYe2 and e/e/, intersect in coplanar points, then the lines ^e/, e2e2', ... evej are concurrent. For let two edges such as eYe2, e-[e2, intersect in the point d12. Let the plane, on which the points such as d12 lie, have for its equation referred to p p p and referred to ex'e2' ...ej let it have for its equation Then any such point d12 can be written \1ex-'\2e2 or X/e/ \2e2. But it has not yet been proved that these alternative forms can be assumed to be at the same intensity. Now consider any corresponding triangles with corners such as eYe2ez and e^e2e3. Write d12 = Xi^ \2e2 = k12 (X/e/ X/e/), (X23 = X2#2 X363 = /C23 (X2 e2 X3 63), ^31 = X3#g Xj^i = /C31 y\ 2 S Xi 1 ). Hence d12 + d23 + d31 = 0, cZ12//tr12 + d^/tc^ + d31/fc31 = 0. Bat if these relations are independent, the three points d12, d2Sy d31 must coincide, which is not true. Hence #12 =/% = #31; and by altering the in- tensity of all the points e(e2 ...ej in the same ratio, each factor such as kp x can be made equal to 1. Hence d12 = \e1 \2e2 = X sV XxV ........................(A). Thus X1e1 + X1V=X262 + X2/62/ = -..=X,eI/ + X|/V = i7............(B). Hence the point g is the point of concurrence of (3) Let the point g be called the centre of perspective and the plane of the points, dp(r, the axal plane of perspective of the two reference-figures. The equation of the axal plane referred to e e2... ev is with the previous notation 2 /X = 0: its equation referred to e/e2'... evr is XgfX' = 0. Let g, the centre of perspective be expressed in the form Then by eliminating e1} e2... ev by means of equations (B) above, X/flie/ + \2a2e2 + ... + Xvfavev' = ( i + a2 + ... 4 av - 1) g. Hence g} though of different intensities, can be expressed in the two forms and 2XW.
74] PERSPECTIVE. 141 Since au a2...a,, can be assumed in independence of \1? X2, ... \v it follows that, given one reference-figure, it is possible to find another reference- figure in perspective with it having any assigned centre of perspective and axal plane of perspective. (4) Suppose that the corresponding edges of three reference-figures 00 o 0f0 0 0' 0 0 l C-2 t?y j %j\ 2 ... t?|/ y K/\ C?2 * * ^V intersect in coplanar points, so that each triad of corresponding edges is concurrent; and let g, g\ g" be the three corresponding centres of per- spective. Consider the three edges e?ea, epfej, ep"ej\ Then we may assume that \ep \ Tea = \pep \JeJ \"ep" ^J'eJ' = dp T\ and hence that g = W - VV' g =" W7 ~ \ep^ 9" = \ep- \PV Hence g + g + g"' 0. Hence the three centres of perspective are collinear. (5) Let there be v reference-figures such that each pair is in perspective, all pairs having the same centre of perspective g. It is required to show that all the axal planes of perspective are concurrent. Let the reference-figures be Consider the v 1 pairs of figures formed by taking the first reference- figure successively with each of the remainder. Let g be the given centre of perspective, and let the equation of the axal plane of perspective of the pair comprising the first and the pth figure be, referred to the first figure, ?i/l^lp + |2/2^lp+...+^/Alp=0 .....................(1), and referred to the pth figure, fi/iV + f^V + - + f^V = o.....................(2)- Hence two typical sets of equations are 2\lp 12_l_ \ pl p2 \lp lv \pl pv ~lpu\ ' #(^)' From equation (1) the point (p) of concurrence of the v 1 planes of this type is, when referred to the first figure, given by But by the first of the set of equations (3),
142 STRAIGHT LINES AND PLANES. [CHAP. II. Hence substituting in the expression for p, and noticing that the co- efficient of g vanishes, we obtain P j l^pl^pl/l^lpi 2^pl^p'2/^lpi ) v^pl^pvfv^py I This is the point of concurrence of the v 1 planes, referred to the pth figure. Now by eliminating en, e12, ... elv from equations (3), we obtain _ Hence the equation of the axal plane of perspective of the pth and ath figures is, referred to the pth figure, I Klp V ^ V^^pl 2Xpi 2^-lor/ Now by 72 (4) the point p lies on this plane, if the determinant formed by substituting the coefficients of x %2, ... in this equation for ^plj ^p2 pv in the determinant, which is the expression for p} vanishes. The determinant so formed can be expressed as the sum of two determinants, one with k1p as a factor, the other with tc1(T as a factor. The determinant with k1p as a factor vanishes because it has two rows of the form The determinant with tc1(T as a factor vanishes because it has two rows of the form 1/Al r 1/2^10- * l/^lo" Hence all the axal planes are concurrent in the same point. The particular case of this theorem for triangles in two dimensions is well-known. 75. Quadrangles. (1) As a simple example of this type of reasoning, let us investigate the properties of a quadrangle in a two-dimensional region. Any four points a, b, c, d are connected by the addition relation Hence act + fib and yc + Bd represent the same point, namely the point of intersection of the lines ab and cd.
75] QUADRANGLES. 143 (2) Consider the six lines joining these four points. Let the three pairs which do not intersect in a, 6, c, d, intersect in e, f, g. Then e = yc + aa = - (ftb + Sd), g = 6+ yc = (Sd + aa). Hence f g = aa yc\ and f+g=/3b-8d. From the form of these expressions it follows that/ $r is the point where /^ intersects ac. Also it follows that e and f g are harmonic conjugates with respect to a and c. Similarly f+g is the point where fg intersects bd; and / + # and e are harmonic conjugates with respect to b and d. Furthermore f g and f+g are harmonic conjugates with respect to The points g e, and e f, have similar properties. Thus the harmonic properties of a complete quadrilateral are immediately obvious. (3) Again the six points f g,g e,e flie by threes on four straight lines. For identically In the accompanying figure h and k stand for /+ g respectively, I and m for e +/respectively, n and j? for g e respectively.
CHAPTER III. Qtjadrics. 76. Introductory. (1) Let a surface locus of the second degree be called a quadric surface. Let a curve locus which can be denned as the inter- section of p (p v) quadric surfaces be called a quadriquadric curve locus. If it is impossible to define the locus as the intersection of p a quadric sur- faces and of a plane subregions (a p\ then the quadriquadric curve locus is said to be the (p l)th order of tortuosity [cf. 67 (6)]. (2) Let the v reference elements be e1} e2, ... ev, and let any point x be denned by + ... + %vevt which is shortened into e. Let the quadric form an i2 + 2a12|:1f2 + ... be written (a$V)2. Then (a} #)2 = 0 is the equation of a quadric surface. Let the lineo-linear form ctnZiVi + ai2 ( i??2 + f^i) + ... be written where x = S e, and y = 77. Elementary Properties. (1) If the element z be of the form \x + fjuy, then (a^zy = \\a y + 2\p(a xJy) + S(a yy...............(A). If more than two elements of a subregion of one dimension lie on a quadric, the whole subregion lies on it. This follows evidently from equation (A). (2) If a quadric contain one plane subregion of the same dimensions as itself, it must consist of two plane loci taken together. For if there is one linear factor of a quadric form (c($#)? the remaining factor must be linear. (3) A subregion of any dimensions either intersects a quadric surface in a quadric surface locus contained in that subregion as its containing region or itself lies entirely in the quadric. For if the subregion be of p 1 dimensions, it may be chosen as a co-ordinate region containing the p reference elements e1} e2,... ep. Hence any element in the region has the v p co-ordinates
76-78] ELEMENTARY PROPERTIES. 145 respectively zero. Thus the intersection of ( $#)* = 0 with the subregion is found by putting these co-ordinates zero in the quadric equation. Thus either the equation is left as a quadric equation between the remaining p co-ordinates; or the left-hand side vanishes identically. (4) It follows as a corollary from the two previous subsections that a subregion, which intersects a quadric in one subregion of dimensions lower by one than itself, intersects it also in another such subregion; and that these two flat loci together form the entire intersection of the subregion with the quadric. 78. Poles and Polars. (1) The equation, (ot^x^x') = 0, may be con- ceived as defining the locus of one of the two elements x or xf, when the other is fixed. If x be fixed, the locus will be called the polar of x with respect to the quadric surface, (a$V)2 = 0. The polar of an element is obviously a plane. The element x will be called the pole of the plane ( $#$#') = 0. (2) The ordinary theorems respecting poles and polars obviously hold. If x be on the polar of x\ then xr lies in the polar of x. For in either case the condition is (a$#$y) = 0. Two elements for which this condition holds will be called reciprocally polar with respect to the quadric. If a pole xf lie in its polar (a][#$y) = 0, then {a\xJx') = {a\xJ^0. Hence the element x' lies on the quadric. Thus all elements on the quadric may be conceived as reciprocally polar to themselves: they may be called self-polar. The polars of all elements lying in a plane must pass through the polar of the plane. (3) By means of these theorems on poles and polars with respect to any assumed quadric a correspondence is established between the elements of a region of v 1 dimensions and the subregions of v 2 dimensions. Corresponding to an element there is its polar subregion: corresponding to elements lying in a plane of v 2 dimensions there are polars all containing the pole of this plane: corresponding to elements lying in a subregion of v p 1 dimensions there are polars all containing a common subregion of p 1 dimensions. (4) Again, consider the elements in which the linear subregion through two reciprocally polar elements x and x intersects the quadric. Let Xx + jjlx be one of these elements, then from equation (A), The two points of intersection must accordingly be of the form Xx fix'. It follows that two reciprocally polar elements and the two elements in which w. 10
146 QUADRICS. [CHAP. III. the straight line containing them intersects the quadric together form a harmonic range. (5) When the element x is on the quadric, its polar, viz. (ol\x x) = 0, will be called a tangential polar of the quadric. Let x lie on the polar of any point x' on the quadric and let \x -f fix' be on the quadric. Then substituting in the equation of the quadric, the equation to determine \//jl becomes X2(a #)2 = 0. Now in general (a$#)2 is not zero. Hence X is zero and both roots of the quadratic are zero. Thus all straight lines drawn through an element x on the quadric and lying in the polar of x intersect the quadric in two coincident elements at x\ (6) Let any plane be represented by the equation The condition that this plane should be a tangential polar of the quadric is obviously Xly X2, .. Xj,, 0 This condition can be written in the form ... = 0. Let A stand for the determinant then = -y , etc. dot12 Now let the plane be denoted by L, then the condition that this plane may be a tangential polar of the quadric may be written by analogy Hence corresponding to the condition, (a$[x)2=0, that the element x lies on the quadric there is the condition. (OffaL)2 = 0, that the plane L is a tangential polar of the quadric. (7) It will be found on developing the theory of multiplication of Grassmann's Calculus of Extension (cf. Book IV. ch. I.) that, analogously to the notation by which an element can be written ^ex 4- ^2^ + + f^ where
79] POLES AND POLARS. 147 elt e2, ... ev are extraordinaries denoting reference elements, the plane L can be written in the form \lE1 + \.2E2 + ... + \VEV where E1,E2... Ev are extra- ordinaries denoting reference planes. Thus the theory of duality will receive a full expression later [cf. 110 (4) and 123] and need not be pursued now, except to state the fundamental properties. (8) The equation, (a] #)2=0, will be called the point-equation of the quadric, and the equation, (fl$Z)2 = 0, will be called the plane-equation of the quadric. The two equations will be called reciprocal to each other. (9) It is possible in general to find sets of v independent elements reciprocally polar to each other. For let ex be any point not on the quadric. Its polar plane is of v 2 dimensions and does not contain eY. The intersection of this plane with the quadric is another quadric of v 3 dimensions contained in it. Take any point e2 in this polar plane not on the quadric. Again take any point e3 on the intersection of the polar planes of eL and e2: then e4 on the intersection of the polar planes of eXi e2, e3; and so on. Thus ultimately v independent points are found all reciprocally polar to each other. If such points be taken as reference elements, the equation of the quadric becomes *ifi2 + a2 22 + .. + ^2 = 0. If the elements lie on the quadric it will be proved in the next article that v/2 or (*/ l)/2, according as v is even or odd, independent elements can be found reciprocally polar to each other. 79. Generating Regions. (1) A quadric surface contains within it an infinite number of flat loci, or subregions, real or imaginary, according to the nature of the quadric. Let such contained regions be called generating regions. If the complete region be of 2/n or of 2/jl 1 dimensions, the subregions, real or imaginary, contained within any quadric surface will be proved to be of /ju 1 dimensions*. If ! be any point on the quadric, it lies on its polar (aQb^oc) = 0. Let b2 be another element on the quadric lying in the polar of b . Then (cc][ i5 2) = 0, and each point lies in the polar of the other. Hence any element AA + XA lies in both polars and on the quadric. But the polars of 6X and b2 intersect in a subregion of v 3 dimensions, where v is put for 2fx +1 or 2/jl as the case may be. Take a third point 63 on the intersection of this subregion with the quadric. Then the three points b1} b2, b3 are reciprocally polar, and any point of the form \bi -f \J)2 + X363 lies in the intersection of the three polars and on the quadric. * This theorem is clue to Veronese, cf. loc. cit. 10 2
148 QUADRICS. [CHAP. III. Proceed in this way till p points bly 62, ... b9 are determined such that each lies on the polars of all the others and on the quadric, and therefore on its own polar. But the p polars, if bl9 62,... bp be independent, intersect in a region of v p 1 dimensions, which contains the p independent points. Hence v p p. Hence the greatest value of p is the greatest integer in \v. If v = 2/jl, or 2/m +1, then p = jjl ; there are therefore fi independent points and these define a subregion of fi 1 dimensions contained in the quadric. This proposition is a generalization of the proposition that generating lines, real or imaginary, can be drawn through every point of a conicoid. (2) If v be even, then each generating region of a quadric is defined by \v independent points. Hence by 72 (5) from any point one straight line can be drawn intersecting two non-intersecting generating regions. If the point from which the line be drawn be on the quadric and do not lie in either of the generating regions, the line meets the quadric in three points, and therefore lies wholly on the quadric. Hence from any point on a quadric one line and only one line can be drawn meeting any two non-intersecting generating regions and thus lying wholly in the quadric. (3) If v be odd, then each generating region is defined by \ (y 1) inde- pendent elements. Hence from 72 (5) it is in general impossible to draw a line from any point, on or off the quadric, intersecting two non-intersecting generating regions. 80. Conjugate Co-ordinates. (1) Let the v co-ordinate elements be a reciprocally polar set. Let the equation of the quadric be The elements ** (- )** are on the quadric. They can be assumed to be any two points on the quadric, not in the same generating region, since e1 can be any point not on the quadric and e2 any point on the polar of ex. (2) Firstly, let v = 2fi. The set of jjl elements, are all on the quadric and reciprocally polar to each other. Hence they define a generating region on the quadric.
80] CONJUGATE CO-ORDINATES. 149 Similarly the set, A**-(-ft)**, etc. define another generating region on the quadric. Also the jjl elements of the first set are independent of the fi elements of the second set. Therefore the two generating regions do not overlap at all. (3) Let the elements of the first set be named in order jltja, ... j^., and of the second set klt k2} ... k^. Then any element x of the form lies in the generating region jij2-.- j*, and any element y of the form lies in the generating region lc1ki...ktL. (4) Again, jx is reciprocally polar to all the k's except kx. Hence is a generating region, and \Jl)j2i ^21 K4 *V) is another, and so on. Accordingly given one generating region (jlf j2,... j^) including a given element, other generating regions including that element can be found which either overlap the given region in that element only, or in regions of 1,2. ..,4-2 dimensions respectively. Also regions can be found which do not overlap the given region at all. (5) The 2/jl elements, Jl 3% Jfif "1 ^2) "V can be taken as co-ordinate elements. Let them be called a system of conjugate co-ordinate elements. The properties of such a system are that they are all on the quadric, and that any pair of elements, with the exception of pairs having the same suffix, are reciprocally polar, namely^ not with kltj2 not with k2, and so on. Let jx and klt j9 and k2t etc. be called conjugate pairs. It can easily be seen by the method of subsections (1) and (2) of this section that in any two non-intersecting generating regions of a quadric a conjugate set of elements can be found, so that jlf ju, ... j^ are in one region, and klf 2, ... k^ in the other. Let any element be written in the form iji + J2+ .. + f eM 4- Vih + V2h + ... +v^- Then, from the definitions of the conjugate elements in subsections (2) and (3) above, the equation of the quadric takes the form
150 QUADRICS. [CHAP. III. (6) The polar of the element jx is ^ = 0, that is to say, is the region defined by the elements Jl j2 Jh- ^2, A73, ... tCp. The intersection of the polar of jx with the quadric is and the 2/jl 1 co-ordinate elements J\ ,hi Jn) k2, k3y "- ten) define its containing region. This quadric is contained in a region of 2^ 2 dimensions. In such a region quadrics in general have generating regions of /i 2 dimensions. But in this quadric all regions of the type (ji Jz -" Jr *V+i fop) are generating regions, being of /jl 1 dimensions. (7) The coefficient of jx in the expression defining an element does not appear in the equation of the quadric. Hence all one dimensional regions defined by jx and any point on the quadric lie entirely in the quadric. Such a surface will be called a conical quadric; the point with the property of jx will be called its vertex. (8) Accordingly the intersection of the polar of any element of a quadric in a region of 2/i 1 dimensions with the quadric is a conical quadric of which the given element is the vertex. Thus in three dimensions, the inter- section of a tangent plane with a quadric is two straight lines, that is to say a conical quadric in two dimensions. (9) Secondly, let v = 2/l+\. Then the system of 2/jl conjugate elements can be found by the same process as in the first case; but do not define the complete region. In forming^, ...,; , kly ...Jcv from the elements ex... e2(l+1 the last element e2/A+1 was left over. This element, which will be called simply e, leaving out the suffix, is reciprocally polar to all the other elements jx ... k^, but does not lie on the quadric. (10) Let any element in the region be denoted by i ji + ... + fijV + Vih + + Vh-K + fe Then the equation of the quadric becomes (11) The polar of the element j\ is given by the equation % = 0. The intersection of the polar and the quadric is another quadric which lies in the region j1}j2 ...jp,k2,... kv e. All these co-ordinate elements are reciprocally polar and all, except e, He on the quadric.
81] CONJUGATE CO-ORDINATES. I5i This quadric is contained in a region of 2/j. 1 dimensions, and its generating regions passing through jx are of ft 1 dimensions, not more than the number of dimensions of the generating regions of any quadric in this region. (12) Since the coefficient x of jY does not appear in the equation of the quadric, if any point x be on the quadric then the region (jlt x) lies entirely in the quadric. Hence the quadric is a conical quadric. So finally we find the general proposition that the intersection of the polar of an element on a quadric with the quadric is a conical quadric with its vertex at the element. (13) The reduction of the equation of a quadric contained in a region of 2/j, dimensions to the form, is a generalization of the reduction of the equation of a conic section to the form, LM + R2 = 0 (cf. Salmon's Gonic Sections). Applying to space of four dimensions the above proposition on the inter- section of polars with quadrics, we see that if our flat three-dimensional space be any intersecting region, it intersects the quadric in some conicoid. But if the space be the polar of some element of the quadric, it intersects the quadric in a cone with its vertex at the element on the quadric which is the pole of the space. A quadric in five-dimensional space has two-dimensional flat spaces as generating regions. (14) The co-ordinates jj ... j^, kY ... M, e of a complete region of 2//, dimensions, giving the equation of some quadric in the form are such that e is the pole of the region jx ... k^. Now e may be any point in the region. Hence the polar of any point (i.e. any plane) can be defined by two not overlapping generating regions of the quadric, viz. jx ... j^ and kx ... kp, which are also generating regions of the quadric formed by the intersection of the polar with the original quadric. This includes the case of space of two dimensions. If, however, the complete region be of 2/n 1 dimensions, the polar of any point intersects the quadric in another quadric which only contains generating regions of /jl 2 dimensions; any two such regions cannot serve to define the polar which is of 2/jl 2 dimensions. This includes the case of space of three dimensions. 81. Quadriquadric Curve Loci. (1) Consider the general case of the curve locus formed by the intersection of the p quadric surfaces, (a xy = 0, (ofix? = 0, .. (aP5^)2 = 0.
152 QUADRICS. [CHAP. III. Let ! be any point on the locus. Then the polar planes of bx are = 0, (a b oe) = 0, ... (aP$6i$^) = 0. The intersection of the p polar planes forms a subregion of v p 1 dimensions; where the complete region is of v 1 dimensions. The intersection of this region with the curve locus is another quadri- quadric curve locus of the same order of tortuosity, namely p 1 [cf. 67 (7)]. Now find another point b2 in this second quadriquadric curve locus, then all the p polars of b2 contain bx. Also the 2p polars of bj)2 form by their intersection a region of v 2p 1 dimensions, and this region intersects the quadriquadric curve locus in another quadriquadric curve locus of the same order of tortuosity, namely p 1. Also it can easily be seen, as in 79 (1), that the region (blt b2) lies entirely in this last curve locus. Continuing in this way and taking a points bly b2}... bVi successively, each in the quadriquadric curve locus lying in the region of the intersection of the polars of all the preceding points, we find a subregion defined by b1} b2 ... K, lying entirely in the original quadriquadric curve locus. Also it must lie in the region of dimensions v ap 1 formed by the intersection of the polars. Hence we must have T ^ V Tp \ that is a vj{p + 1). Now let / (X) denote the greatest integer in the number X. Then we have proved that it is always possible to proceed as above till r = i /0 + l)}. Hence a quadriquadric curve locus, apart from any special relation between the intersecting quadric surfaces, of tortuosity p 1, in a complete region of v 1 dimensions contains subregions (real or imaginary) defined by / {v/(p + 1)} elements, that is to say, of / {(v p l)/(/ + 1)} dimensions*. (2) Hence the least dimensions of a complete region such that a curve locus, of order of tortuosity p 1, apart from special conditions must contain a region of one dimension is 2p +1. For example, space of five dimensions is of the lowest dimensions for which it is the case that the intersection of two quadric surfaces (a curve locus of order of tortuosity 1) must contain straight lines. * This generalization of Veronese's Theorem, cf. 79 1), has not been stated before, as far as I am aware.
82] QUADRIQUADRIC CURVE LOCI. 153 Also space of eight dimensions is of the lowest dimensions for which it is the case that a quadriquadric curve locus, of order of tortuosity 1, must contain subregions of two dimensions. 82. Closed Quadrics. (1) A quadric will be called a closed quadric if points not on the quadric exist such that any straight line drawn through one of them must cut the quadric in real points. Such points will be said to be within the quadric: other points not on the quadric which do not possess this property will be said to be without the quadric. (2) Let a straight line be drawn through any point p cutting the quadric in two points yY and y2 real or imaginary; and let x be any other real point on this line. Also let y1-\1p f- fixxt y2 = then \/fii and X^/f^ ar^ the roots of the equation \2 (a$|))2 + 2X/i (a$j)^) + /^2 (a$ )2 = 0..................(1). The roots of this equation are real or imaginary according as is negative or positive. (3) Now choose as the co-ordinate elements a reciprocally polar system with respect to the quadric, and let p be one point of this system. Let the system bep, e2, es,... eV} and let Then (a$#)2 takes the form ap-f 2k2+... + , ,2. Hence the roots of equation (1) are real or imaginary according as is negative or positive. (4) If all lines through p meet the quadric in imaginary points, then the expression is positive for all values of 2, f3, ... . Hence a*2, a3) ... must be all positive; and therefore a, a2,... av must all be of the same sign. The equation of the quadric takes the form, *2 2 + A^f22+... + /c,2 ,2 = 0; and the quadric is therefore entirely imaginary. (5) Again, if all lines drawn through p meet the quadric in real points, then aa2, aa3,... oliv must all be negative.
154 QUADR1CS. [CHAP. III. Hence a2, or3, ...nv are of one sign and a is of the other. The equation of the quadric can therefore be written in the form ^" a+^f, +... + ^ f, -^p = O .....................(2); where the co-ordinate point p is within the quadric and the remaining v - 1 co-ordinate points can easily be proved to be without the quadric on the polar of p. (6) It also follows that the polar of a point inside the quadric does not intersect the quadric in real points. For the polar of p is = 0, and its points of intersection with the quadric lie on the imaginary quadric of v 2 dimensions given by *22 22+... + ,,2 ,2 = 0. It has been proved by Sylvester that if a quadric expression referred to one set of reciprocal co-ordinate elements has p positive terms and v p negative terms, then when referred to any other set of reciprocal elements it still has p positive terms and v p negative terms (or vice versa). Hence if the given quadric of equation (2) be referred to any other reciprocal set of co-ordinates, it still takes the form of (2) as far as the signs of its terms are concerned. Thus if the quadric considered be a closed quadric, one element of a reciprocal set of elements is within the quadric and the remaining elements are without the quadric. (7) The polar of a point without a closed quadric necessarily cuts the quadric and contains points within the quadric. For considering the quadric of equation (2) of subsection (5), the polar of e2 is, 3 = 0. Its in- tersection with (2) is the quadric lying in the plane f2 =0, that is in the region of p, e3) ... ev. Now if e2 be first chosen, p may be any point in this region and within this quadric, which is a real closed quadric. Hence the polar of any point without a closed quadric necessarily cuts the quadric in real points and contains points within the quadric. (8) It may be noted that no real generating regions exist on closed quadrics. (9) Again, choosing any reference points whatsoever, (a$#)2 and are of the same sign if both x and y be inside the closed surface, or if both be outside the surface, but are of opposite signs if one be inside the surface and one be outside. For let x and yf be two points respectively on the polars of x and y. Then Xx + W and fiy are two points on the lines xx' and yy'.
83] CLOSED QUADRICS. 155 The points where these lines cut the surface are given by and f Firstly, let x and y be both within the surface. Then their polars intersect in a region of v 3 dimensions without the surface. Let x and y' both denote the same point z in this region. Then since the roots of the quadratics for \/\' and fi/fjf are both real, (a$/)2 has opposite signs to both (a xf and (a Hence (a$V)2 and (a$y)2 have the same sign. (10) Secondly let x and y be both without the quadric. Then their polars both cut the quadric, hence x and y may both be chosen within the quadric. Hence (a$V)2 and (a^yj have both the same sign. Also the straight lines xx and yy' both cut the quadric in real points, since x and y' lie within it. Hence (a$V)2 has the opposite sign to (ol$x')2 and (a$y)2 has the opposite sign to (ofay')2. Accord- ingly (oi xf and (orfjiy)2 have the same sign. (11) Thirdly, let x be within the quadric and y without the quadric. Then any point x on the polar of x lies without the quadric. Also ( x$x)2 and (ajx')2 have opposite signs, and (a$V)2 and (a$y)2 have the same sign because both lie without the quadric. Hence (a$a)2 and ( x.][y)2 have opposite signs. 83. Conical Quadric Surfaces. (1) To find the condition that S = (a #)2 = 0, should be a conical quadric. Let b be the vertex and x any point on the surface. Then \x + fib lies on the surface for all values of X and fi. Hence (a] 6$V) = 0; where x is any point on the surface. Therefore, if b = 2#e, there are v equations of the type, ... + avp/3v = 0, found by putting p equal to 1, 2,... v, in turn. It follows that the equation (a$ 5#) = 0 holds for all positions of x. And eliminating the /3's, the required condition is found to be, A = an, a12, ... alv j = 0.
156 QUADRICS. (2) Also the vertex b is the point \y e2, ... ev 12 0-22 *1 [chap. hi. (3) Again, consider the quadriquadric curve locus of the first order of tortuosity, defined by (a$:z)2 = 0, and ( '$ ) = 0. Any quadric surface intersecting both surfaces in this curve locus is This surface is a conical quadric if , + \a2l,' = 0. Hence in general v conical quadric surfaces can be drawn intersecting two quadrics in their common curve locus. (4) Let b be the vertex of one of these conical quadrics. Take v 1 independent points in the quadriquadric curve locus, so as to make with b an independent set of elements. Join b by a straight line with any one of these points; the straight line cuts the quadrics again in another common point. Hence by the harmonic property proved in 78 (4) it cuts the two polars of b with respect to the two quadrics in a common point. Hence these polars have v 1 independent common points. Hence they are identical. Hence the equations (ct^b^x) = 0, and (^^b^x) = 0, are identical. (5) Let the reference points ex and e2 be the vertices of two such conical quadrics. Then the equations (aJtfxJV) = 0, and ( '' e$x) = 0, are identical: that is nfi + ai2 2+ awfa +. . + !,?,= 0,) . 6++ ;foj...........'- ......w are identical. Similarly the equations are identical.
84] CONICAL QUADRIC SURFACES. 157 From equations (1) it follows that ^1,= ^, = !?-,= ... = ^..........................(3), from equations (2) it follows that a^ = a^= o^="* = ^..........................^ Hence either a12 = 0 = a12'; or else if p be any element \ex + fie2 on the line e y then the two polars I Q O jj II Q(j \ ^sr \j cLU \JL \ OL w X/ w 0 f ~ v/ are identical. Excluding the second alternative, which is obviously a special case, it follows that any two of the v vertices lie each on the polar of the other with respect to either quadric. Thus the v vertices form v independent elements and can be taken as reference elements. (6) It follows that in general any two quadrics have one common system of polar reciprocal elements, and that these elements are the v vertices of the v conical quadrics which can be drawn through the intersection of the two given quadrics. (7) Let this system of polar reciprocal elements be taken as co-ordinate elements. The equations of the quadrics become and 7ll' 2 + yvV+ - + yJV = 0. And the ratios Yn/71/, 722/722? e^c. are the roots, with their signs changed, given by the above equation [cf. subsection (4)] of the * th degree determining X. (8) One means of making the properties of conical quadrics more evident is to take the vertex as one of the co-ordinate elements of the complete region. Let e1 be the vertex of the quadric. Then if x be any element on the quadric, by hypothesis 6eY + is on the quadric, 6 being arbitrary. Hence if x = Xge, the element is also on the quadric. It follows that cannot occur at all in the equation of the quadric. Accordingly the expression (^x)2 reduces to 84. Reciprocal Equations and Conical Quadrics. (1) When the quadric (a$#)2 = 0 is conical, the reciprocal equation of the quadric, namely, (OL^Ly = 0, has peculiar properties. It has been proved that if b be the vertex, then $ #)=0, whatever element x may be. Hence the polar of any element x passes through the vertex 6.
158 QUADRICS. [CHAP. III. Let L be the plane \ + X2 2 4- + X*(v = 0. Then it is proved in Salmon's Higher Algebra, Lesson v. and elsewhere that, when A = 0, But we may write by 83 (3) b = flu i + O^ + + OLlvev. Hence, {OL\Lf = 0, reduces to, 2/3X = 0; that is to the condition that L pass through the vertex. But this is the property of all polars and not merely of tangential polars. Thus in this particular case of conical quadrics the reciprocal equation to the ordinary point-equation, deduced as in the general case, merely defines the vertex of the quadric. (2) In order to find the nature of the condition which L must satisfy in order to be a tangential polar of the conical quadric, suppose that e^ has been chosen to be the vertex of the quadric. Then (o xf = a!B?22+ 2a23 2 ; + ... = 0 is the equation of the quadric. And ( 3Mfc0 = Hence, as before, the conditions that L, which is the locus defined by should touch the quadric are Xx = 0, and a22/x22+2a23/\2x3 + ... = 0; where OL^, OL^',... are the minors of o^, o^,... in the determinant I a2Vi UZvy ... CLVV Let this determinant be called A'. The first condition, Xj = 0, is the condition given by the ordinary reciprocal equation, namely that the polar should pass through the vertex. Let the second condition be called the conical reciprocal equation. (3) Now if we transform to any co-ordinate elements whatever, so that any point b is the vertex, these equations become, 2#X = 0, which is the condition that b should be the vertex ; and (OOfcL)2 = 0, where the coefficients 0Lu, #i2 satisfy the condition an, a12, ... alv, =o. OLlv
84] RECIPROCAL EQUATIONS AND CONICAL QUADRICS. 159 In accordance with the notation explained 78 (6) this determinant will be called Aj. (4) Suppose now that we are simply given the equation, We have to determine what it is to be conceived as denoting when the above determinant vanishes. If we had the two equations (a$i)2 = 0, and 2 \ = 0, a conical quadric of vertex b would be determined. Hence it is possible to conceive (OCQL)2 = 0 as denoting a conical quadric with an undetermined vertex. This, however, is not satisfactory. Let the reciprocal point-equation be formed. It follows from the previous investigation that this equation is where jttn, fiL12i... are the minors of 0Ln, 0L12, 0Lu, ... in the determinant Aj. This locus is two coincident planes forming a quadric. If we choose v 1 co-ordinate elements in this region and any co-ordinate element e1 outside it, then (OCQL)2 = 0 takes the form Hence it is best to consider, (OLlfcLy = 0, as denoting in the reciprocal point-form the region, taken twice over, and as denoting in the original plane-form a quadric sur- face of v 3 dimensions lying entirely within this region. (5) If any vertex b be assumed and all the one dimensional regions joining b to elements of this quadric of v 3 dimensions be drawn, then a conical quadric of v 2 dimensions is obtained. The vertex b should not be in the region, Ai +Astra + etc. = 0, which contains the quadric surface of v 3 dimensions, if a true conical quadric is to be obtained. Hence we may call the region, the ' non-vertical region'. Also call the quadric of v 3 dimensions lying in it the * contained quadric'. (6) Accordingly, summing up, given the equation, (OOfcL)2 = 0, where Aj = 0, we derive the reciprocal equation lOiifi + Aaft + eto. = 0.
160 QUADRICS. [CHAP. III. These two equations taken together represent a non-vertical region and a contained quadric. This may be considered as the degenerate form of a quadric defined by either of its two reciprocal equations. Accordingly the equation, (OC^L)2 = 0 (when Ax = 0), gives the condition to be satisfied by the co-ordinates of all regions of v 2 dimensions whose intersection with the non-vertical region is a tangential polar of the contained quadric. (7) Let us consider as a special case of the above investigations Geometry of two dimensions. Let ex, e2, e3 be the co-ordinate points, and Ely E2) Ez the corresponding symbols denoting the straight lines e2e3, e3elt e^. Then any point can be denoted by and any straight line L by the equation Also the equation, denotes either that x lies on L or that L passes through x. Any quadric (a] #)2 = 0 is a conic ; the determinant A is an #22 0(33 + 2ot12a23O 31 tfnCt232 ^ha^si2 tt33#i22 A conical quadric, for which A = 0, is two straight lines. The reciprocal equation is the tangential equation. Conversely given the tangential equation (OL^L)2 = 0, the point-equation is formed from it by the same law. Also if Aj = 0, then splits up into two factors. In this case let Let p be the point and let r be the point Then (fl] Z)2 = 0 is the condition to be satisfied by all lines which pass through either p or r. Also 0tn = 'ur1pli etc.; and #2* = i 0 2 3"3 + / S'cr2), etc. Hence ^ = QLJ1* - fl a = - and
84] RECIPROCAL EQUATIONS AND CONICAL QUADRICS. 161 Accordingly the non-vertical region is the straight line " Pl^s) %2 + (P1 2 - Pi^l) ft = 0. This is the straight line joining the points p and q. Thus the non-vertical region is a straight line, and the contained quadric is two points in it. This agrees with Cayley's statements in his 'Sixth Memoir on Quantics*/ respecting conies in two dimensions. Note. For further information in regard to what is known of the projective geometry of many dimensions, cf. Veronese's treatise, Fondamenti di geometria (Padova, 1891), translated into German under the title, Grundziige der Geometrie von mehreren Dimensionen und mehreren Arten gradliniger Einheiten in elementarer Form entwickelt (Leipzig, 1894). * Cf. Phil. Trans. 1859 and Collected Mathematical Papers, vol. 11., no. 158. W, 11
CHAPTER IV. Intensity. 85. Defining Equation of Intensity. (1) Let a complete region of v 1 dimensions be denned by the units elt e2, ... ev. Then by hypothesis the intensity of the element represented by apep is xp, since the intensity of ep is by definition unity. But no principle has as yet been laid down whereby the intensity of a derived element a can be determined; where a = aft + a2e2 + ... + avev = 2,ae, say. (2) Let a be the intensity, then we assume that a is some function of alt x2, ... av. Hence we can write =/( !, a2, ... ). Now the intensity of /jua is fia; therefore we have the condition /(/ !, i*CL2, ... fia¥) = fif(al9 o^, ... a,). Accordingly f(oL1} a2, ... a,,) must be a homogeneous function of the first degree. (3) If a be at unit intensity, then the coefficients must satisfy the equation /( !, 2 p) = l..............................(A). This equation will be called ' the defining equation'; since it defines the unit intensities of elements of the region. It will be noticed that the equation does not in any way limit the ratios aa/ i, s/ai a"/ai which determine the character, or position, of the element represented by a. Furthermore this equation essentially refers to the v special units e^ e2) ... ev which have been chosen as defining (or co-ordinate) units of the region. (4) Let 0a( i, ... /) denote a rational integral homogeneous function of the Xth degree, and /v( i, ... ) a rational integral homogeneous function of the /*th degree. Then the most general algebraic form of/(a1, ... ) is
86] DEFINING EQUATION OF INTENSITY. 163 If there be only one term in this expression, then equation (A) can be written in the form (5) Let Ci Cs... Cy be any other group of independent letters which can be chosen as co-ordinates of the region. Let a = 7iCi + 72c2 + . + VvCy = 27c. Then the 7's are homogeneous linear functions of the a's. Hence the defining equation (A') with reference to any other co-ordinate elements becomes ^a(7i 72 ... 7^) = ^(7]? 72. 7*); where ^A and yfr^ are homogeneous functions of the Xth and /ith degrees respectively. (6) Again, when c^, cc3)... av all simultaneously vanish, the intensity is unity when aY = 1. But in this case equation (A') reduces to an equation of the form Therefore for this to be satisfied by i = l, we must have ^1 = %. So the coefficients of the highest powers of a1} a2,... xv on the two sides of the equation are respectively equal. (7) In the subsequent work, unless otherwise stated, we will assume the defining equation to be of the form { f)h(au a2, ... )}'* = 1; where (f ^(oclf a^ ... *v) is a rational integral homogeneous function of the /*th degree of the form af + of + ... + */ + 2/wr^'a/2... a , / being an arbitrary coefficient and 0"i + cr2 + ... + o-,, = /i. Raising each side of the defining equation to the /-tth power it becomes, M*i ^ a^) = 1- CoeflScients which satisfy the defining equation will be called the co- ordinates of an element. They define the element at unit intensity. The co-ordinates of an element must be distinguished from the co-ordinate elements of a region, which have been defined before [cf. 64 (8)]. 86. Locus of Zero Intensity. (1) It is obvious that there is one locus of v 2 dimensions with exceptional properties in regard to the intensities of its elements. For the equation M i ... ,) = 0, is a relation between the ratios a^l^i, ots/ai, *v\*\\ and it, therefore, determines a locus which is such that all the elements of it are 11 2
164 INTENSITY. [CHAP. IV. necessarily at zero intensity, according to this mode of defining the in- tensity, and yet do not themselves vanish, since the coefficients of the co- ordinate extraordinaries do not vanish separately. Therefore in relation to the given definition of unit intensity, elements of this locus are all at zero intensity. (2) Some other law of intensity is necessarily required in the locus of zero intensity, at least in idea as a possibility in order to prevent the intro- duction of fallacious reasoning. For if two terms a and a! represent the same element at the same intensity, then a = a\ and the coefficients of the co- ordinate elements in a and a' are respectively equal in pairs. But if the element be in the locus of zero intensity a and pa are both zero intensity according to the old definition. Hence from the above argument a = pa, and therefore (p l)a = 0; and since p 1 is not zero, a=0, which is untrue. Therefore in the locus of zero intensity in order to preserve generality of expression some other definition of intensity is to be substituted, at least in idea if not actually formulated. An analogy to this property of points in the locus of zero intensity is found in the fact that two zero forces at infinity are not therefore identical in effects, and that for such forces another definition of intensity is substituted, namely the moment of the force about any point, or in other words the moment of the couple. (3) If the properties of the region with respect to the intensity are to be assumed to be continuous, at any point of the locus of zero intensity one or more of the co-ordinates must be infinite. For the equation M i, ... *) = 1, viewed as an equation connecting the absolute magnitudes of the co-ordinates, can only be satisfied simultaneously with Mai ... 0=0, viewed as an equation connecting the ratios of the co-ordinates if one or more of the a's are infinite. In this case we can write the first equation in the form 1 Then if ap becomes infinite, the equation, between the ratios of the co-ordinates is simultaneously satisfied. 87. Plane Locus of Zero Intensity. (1) There are two special cases of great importance, one when the locus of zero intensity is plane, the other when it is a quadric.
87] PLANE LOCUS OF ZERO INTENSITY. 165 Considering the case of a plane locus, by a proper choice of the unit intensities of the co-ordinate elements of the complete region the equation of the locus can be written in the form + ?, + ... + , = 0. (2) Let elf e2, ... ev be these co-ordinate elements, and let aly a2, ... av be another set of independent elements at unit intensity to be used as a new set of co-ordinate elements. Let ctj = auei + a12e2 + ... 4- alveV) a2 = + ... +avvev. Then by hypothesis an -I- a12 + ... + a^ = 1, with v 1 other equations of the same type. Let any element x at unit intensity be given by Xge and also by Then + ?.+ . But by comparison = a^ + o^ + ... + avl r]v with j/ 1 other equations of the same type. Hence substituting for the 's in the defining equation, we get (an + 012 + ... + ! ) Vi + ( 2i + 022 + ... + Ok,) 172 + ... = 1. Therefore using the defining equations for the a's, there results as the defining equation for the new co-ordinates. It follows that if this special type of defining equation of the first degree hold for one set of co- ordinate elements it holds for all sets of co-ordinate elements. (3) Any one-dimensional region meets the locus of zero intensity in one element only, unless it lies wholly in the locus. Let a and b be two elements at unit intensity defining a one-dimensional region. Then by subsection (2) the intensity of any element fa -f- rjb in the region ab is f + v- Hence b a is of zero intensity. Let 6 a = u; then u is the only element in oh at zero intensity according to the original law of intensity, but possessing a finite intensity according to some new definition. (4) If p 1 of the co-ordinate elements, where p v, be assumed in the region of zero intensity and the remaining v p +1 outside that region, then the defining equation takes a peculiar form. For let ^, u2i... up^ be the co- ordinate elements in the region of zero intensity, and ep, ep+1,... ev the remain- ing co-ordinate elements. Then any element can be expressed in the form
166 INTENSITY. [CHAP. IV. Now any element of the form %\u lies in the region of zero intensity. Hence the denning equation must take the form fP + fP+1 + ...+f, = l. If p = v, then the co-ordinate elements ulf ... uv^ completely define the region of zero intensity. Let e denote the remaining co-ordinate element. Any element can be written %Xu + !je. The defining equation becomes, = 1. 88. Quadric Locus of Zero Intensity. (1) Let the intensity of the point x be + {( $#)*}*. Then the locus of zero intensity is the quadric surface (a$#)2 = 0. (2) Let us assume this quadric to be closed, or imaginary with real coefficients [cf. 82 (4)]. If x lie within this quadric and y lie without it (the quadric being real and closed), then by 82 (11) (a$ 2 and (a y)2 are of opposite sign. Suppose for example that (a$#)2 is positive for elements within the quadric. And let {ajxf = ^ and (ajy)2 = - v2; where fi and v are by hypothesis real, since the co-ordinates of x and y are real. Then the intensities of x and y as denoted by the symbols x and y are fju and V(~ v2) respectively. (3) The symbols which denote these points at unit intensity are x/fi, and yl*J( v2). Hence although the element y is defined by real ratios, its co- ordinates at unit intensity are imaginaries of the form irjly ir)2, ..., where rji, r)2... are real. Such elements will be called 'intensively imaginary elements.' If the element be defined by real co-ordinates, its intensity is imaginary. Those elements such that real co-ordinates define a real intensity will be called ' intensively real elements.9 (4) If intensively real elements lie without the quadric of zero intensity, then intensively imaginary elements lie within it, and conversely. It is to be noted that both sets of elements are real in the sense that the ratios of their co-ordinates are real. (5) If the quadric of zero intensity be imaginary, then all real elements are intensively real. 89. Antipodal elements and opposite intensities. (1) Since ( $ ) = ( $-*) , the intensities of x and x are both positive and equal, when the locus of zero intensity is a quadric. An exception, therefore, arises to the law that if 2fe and 2f 'e denote the same element at the same intensity, then i = ', 2 = etc.
90] ANTIPODAL ELEMENTS AND OPPOSITE INTENSITIES. 167 Let the generality of this law be saved by considering the intensities denoted by x and x, though numerically the same, to differ by another quality which we will call oppositeness. (2) Another method of evading this exception to the general law is to regard x and a? as two different elements at the same intensity. This is really a special case of the supposition alluded to in 65 (9). Let x and x be called antipodal elements. In this method the quality of oppositeness has been assigned to the intrinsic nature of the element denoted, whereas in the first method it was assigned to the intensity. When the quadric locus of zero intensity is real and closed, the first method is most convenient; when it is imaginary, either method can be chosen. (3) Antipodal elements have special properties. If any locus include an element, it also includes its antipodal element. If two one-dimensional regions intersect, they also intersect in the anti- podal element. Hence two one-dimensional regions, if they intersect, intersect in two antipodal points. A one-dimensional region meets a quadric in four points, real or imaginary, namely in two pairs of antipodal points. (4) The sign of congruence, namely = [cf. 64 (2)], connects symbols representing antipodal points as well as symbols representing the same point. 90. The Intercept between two elements. (1) The one-dimensional region which includes ex and e2 may be conceived as divided by the elements 6^2 into two or more intercepts. For the element f^ + 2#2 maybe conceived as traversing one real portion of the region from ex to e2y if it takes all positions expressed by the continuous variation of ^2/^i from 0 to + oo. Similarly it may travel from ^ to e2 through the remaining real portion of the region by assuming all the positions expressed by the continuous variation of 2/fi from 0 to 00 . A one-dimensional region may, therefore, be considered as unbounded and as returning into itself [cf. 65 (9)]. (2) Assume that the expression for the intensity is linear and of the form 2 . Then the locus of zero intensity cuts the region e^e2 at the element defined by / = 1. Hence as fa/f^ varies continuously from 0 to + 00 , the element x does not pass through the locus of zero intensity, and its intensity cannot change sign, if 2 and ^ do not change sign. Let this portion of the region be called the intercept between e1 and e2. Also let the other portion of the region be called external to the portion limited by e1 and e2, which is the intercept.
168 INTENSITY. [CHAP. IV. (3) An element on the intercept between eY and e2 will be said to lie between e^ and e2. Also the external portion of the region is divided into two parts by the element of zero intensity, e2 ex. Let the continuous portion bounded by e1 and e2 e1 and not including e2 be called the portion beyond ex, and let the portion bounded by e2 and e2 ex and not including ex be called the portion beyond e2. (4) Assume the intensity to be {(a$#)2} . Let the locus of zero intensity be the real closed quadric, (a$#)2 = 0. Firstly, assume that x denote the same element at opposite intensities. Let the two elements eY and e2 both belong to the intensively real part of the region. Then x moves from e1 to e2, as 2/ i varies from 0 to oo or from 0 to oo. Now [cf. 82 (9)] since (a^e^2 and (a$e2)2 are both of the same sign, as x moves from ex to e2 by either route it must either cut the surface of zero intensity twice or not at all. Call the latter route the intercept between eY and e2. The intercept only contains intensively real elements. (5) If the quadric (a$V)2=0 be imaginary, then the one-dimensional region eYe2 does not cut it at all in real points. Hence there is no funda- mental distinction between the two routes from ex to e2i and both of them may be called intercepts between ex and e2. Also all real elements are inten- sively real. Hence a one-dimensional region is to be conceived as a closed region, such that two elements exe2 divide it into two parts. (6) Secondly, assume that x denote two antipodal elements. Assume that the quadric (ol][x)2 = 0 is entirely imaginary. The two routes from ex to e2 are discriminated by the fact that the one contains both antipodal points ex and e2, and the other contains neither. Let the latter portion of the region be called the intercept, and the former portion the antipodal intercept. The case when (ol x)2 = 0 is real is of no practical importance, and need not be discussed. Note. Grassmann does not consider the general question of the comparison of intensities. In the Ausdehnungslehre von 1844, 2nd Part, Chapter L, 94 100, he assumes in effect a linear denning equation without considering any other possibility. In the Ausdehnungslehre von 1862 no general discussion of the subject is given; but in Chapter v., ' Applications to Geometry,' a linear denning equation for points is in effect assumed, and a quadric denning equation for vectors assumptions which are obvious and necessary in Euclidean Geometry. It should also be mentioned that the general idea of a denning equation, different for different manifolds, and the idea of a locus of zero intensity do not occur in either of these works. Also v. Helmholtz in his Handbuch der Physiologischen Optik, 20, pp. 327 to 330 (2nd Edition) apparently assumes that only a linear denning equation is possible.
BOOK IV. CALCULUS OF EXTENSION.
CHAPTER I. Combinatorial Multiplication. 91. Introductory. (1) The preceding book has developed the general theory of addition for algebras of the numerical genus (cf. 22). The first special algebra to be discussed is Grassmann's Calculus of Extension*. This algebra requires for its interpretation a complete algebraic system of mani- folds (cf. 20). The manifold of the first order is a positional manifold of v 1 dimensions, where v is any assigned integer; the successive manifolds of the second and higher orders are also positional manifolds (cf. 22); the manifold of the i/th order reduces to a single element; the manifold of the (z/ + l)th order is identical with that of the first order. Hence (cf. 20), when the manifold of the first order is of v 1 dimensions, the algebra is of the i/th species. (2) It follows from the general equation for multiplication of algebras of the numerical genus given in 22, that if two points a(=Xae) and b(= 2j8e) be multiplied together, where eXi e2...ev are any v reference points, then ab = 2ae. 2fie = 22 (aP/8a6PO; where j = 1, 2 ... v, and ap/3a are multiplied together according to the rules of ordinary algebra. (3) Thus the products two together of the reference elements ^, e2... ev yield v2 new elements of the form (e^), (e2e2) (e^), (e \) etc. These v2 elements (which may not all be independent) are conceived as defining a fresh positional manifold of v2 1 dimensions at most, and ab is an element of this manifold. This is the most general conception possible of a relation between any two elements of a positional manifold which may be symbolized by a multiplication. (4) No necessary connection exists between the symbols (e^), (e^), ( A) (^aX etc.: they may therefore, as far as the logic of the formal symbolism is concerned, be conceived as given independent reference elements * Cf. Die Ausdehnungslehre von 1844, and Die Ausdehnungslehre von 1862, both by H. Grassmann.
172 COMBINATORIAL MULTIPLICATION. [CHAP. I. of a new positional manifold. But on the other hand we are equally at liberty to assume that some addition equations exist between these v2 products, whereby the number of them, which can be assumed as forming a complete set of independent elements, is reduced. These products of elements are then interpreted as symbolizing relations between the elements of the manifold of the first order which form the factors; and thus the mani- folds of orders higher than the first represent properties of the manifold of the first order which it possesses in addition to its properties as a positional manifold. Let any addition equations which exist between products of the reference elements ely e2... ev be called 'equations of condition' of that type of multiplication which is under consideration. 92. Invariant Equations of Condition. (1) The equations of con- dition will be called invariant, when the same equations of condition hold whatever set of v independent reference elements be chosen in the manifold of the first order*. (2) For products of two elements of the first order, there are only two types of multiplication with invariant equations of condition, namely that type for which the equations of condition are of the form (^ff) + (6^p) = 0, (epep) = 0........................(1); and that type for which the equations of condition are of the form For assume an equation of condition of the most general form possible, namely (^ () (^ ...=0..................(3). Then if xly x2.,.xv be any v independent elements, this equation (3) is to persist unchanged when xlf x2... xv are respectively substituted for ely e2...ev. Thus in equation (3) change e into ^ely where is any arbitrary number, not unity. Subtract equation (3) from this modified form, and divide by 1, which by hypothesis is not zero. Then {alp (e^) + apl (e^)} = 0. Hence since is arbitrary, V an(^i) = 0, 2 {ctiP(e1ep) + apl(epe1)}=0..................(4). p=2 Therefore by hypothesis these forms are to be invariant equations of condition. Hence the second of equations (4) must still hold when ge2 is substituted * The type of multiplication is then called by Grassmann (cf. Ausdehnungslehre von 1862, 50) * linear.' But this nomenclature clashes with the generally accepted meaning of a ' linear algebra' as denned by B. Peirce in his paper on Linear Associative Algebra, American Journal of Mathe- matics, vol. rv. (1881), The theorem of subsection (2) is due to Grassmann, cf. loc. cit.
93] INVARIANT EQUATIONS OF CONDITION. 173 for e2i being any number not unity. Thus, as before, by subtraction and division by 1, Since this equation is invariant, it must hold when ex and e2 are interchanged, thus by subtraction, (au - a) {Oi02) - (e2e1)} = 0. (3) Firstly assume, ai2 = a21. Then, if ex and e2 are any two of the reference elements, Now since this equation must be invariant, put e2 + %eY for e2, where ff is any number not unity; then by subtraction we find the typical form (**) = ....................................(6), and this satisfies the first of equations (4). It is evident and is formally proved in 93 (3) that equations of condition, of which equations (5) and (6) are typical forms, actually are invariant. (4) Secondly assume, (e^) = (e2e^), as the typical form of equation of condition. Then it is immediately evident that (xy) = (yx)} where x stands for %%e and y for %rje. Thus this form of equation of condition is invariant. Also substituting x instead of ex in the first of equations (4), which is invariant, it is obvious that au = 0. Hence this equation introduces no further equation of condition. Thus there are only two types of multiplication of two elements of the manifold of the first order which have invariant equations of condition. 93. Principles of Combinatorial Multiplication. (1) Let the multiplication be called ' combinatorial' when the following relations hold: ........................^ '' (e1e2...ep)(ep+1ep+2...eir) = {e1e2...epep+1...ea)...............(2). (2) The second of equations (1) follows from the first, if the first equation be understood to hold in the case when p = a. For then (epea) = (epp), and therefore (epea) + {eaep) = 2 (epep) = 0. (3) Equations (1) and (2) as they stand apply to one given set of independent elements, elye2...ev. Now if a = Xae and b = %/3e, then the product ah takes the form 2 (afiv o^/Sp) (epea); and the number of indepen- dent reference elements of the type e^^ in the new manifold of the second order created by the products of the reference elements of the first order is \v (y 1). Similarly the product ba becomes 2 (a r/8P ap/3o-) (^o-) Thus for any two elements a and b of the manifold of the first order equations of the same type as equation (1) hold, namely
174 COMBINATORIAL MULTIPLICATION. [CHAP. I. (4) Equation (2) expresses what is known as the associative law of multiplication. It has been denned to hold for products of the v independent elements ely e2... ev. It follows from this law and from equation (1) that exe2 ... epep+1... ev = {exe2 ... e^)(epep+1)(ep+2 ...e^) . ep-i) (ep+1ep)(ep+2 ...ej Accordingly any two adjacent factors may be interchanged, if the sign of the whole product be changed. By a continually repeated interchange of adjacent factors any two factors can be interchanged if the sign of the whole product be changed. Again, if the same element appear twice among the factors of a product, the product is null. For by interchanges of factors the product can be written in the form {e^eY. e2e3...), where eY is the repeated factor. But (e^j) = 0. Hence by 21 {exe^e2ez...) = 0. It follows from this that in a region of v 1 dimensions products of more than v factors following this combinatorial law are necessarily null, for 5tte factor at least must be repeated. (5) It remains to extend the associative law and the deductions from it in the previous subsection to products of any elements. In the left-hand side of equation (2) let any element, say epy be replaced by ep =ep + deky where 6 is any arbitrary number. Then by the distributive law of multiplication (cf. 19), (eYe2... ep)(ep+1ep+2... O = (exe2... ep)(ep+1ep+2... ea) + 0(exe2 ...ek)(ep+1ep+2 ...ea) = (eLe2... epep+1ep+2... ea) + 6{e 2... ekep+1ep+2 ...e^) = (- l)p^(eP exe2... ep+1ep+2... e^ + i-1)*-1^ e,e2... ep+1ep+2... ev) = (- ly-1 (ep . e#2... ep+1ep+2... ea). Also by a similar proof exe2... epep+1ep+2 ...e(r = (- ly-1 (ep . exe2... ep+1ep+2... e^). Hence fae*... ep)(ep+1ep+2 ...e^ = exe2... epep+1ep+2 ...e*. Hence the associative law holds when ep has been modified into ep; and by successive modifications of this type elye2...ev can be modified into alfa2...av, where aly o^... av are any independent elements. (6) The only deduction in subsection (4) requiring further proof to extend it to any product is the last, that in a region of v 1 dimensions products of more that v factors are necessarily null. This theorem is a special instance of the more general theorem, that products of elements which are not independent are necessarily null. For let alta2...ap be in- dependent, but let ap+1 = axax + ... + apap. Then (a^ ... apap+1) = ax (a^... apai) + ... + aP (a^ ... apap).
94] PRINCIPLES OF COMBINATORIAL MULTIPLICATION. 175 Thus every product on the right-hand side has a repeated factor and is therefore null. (7) Conversely, let it be assumed that a product formed by any number of reference elements is not null, when no reference element is repeated as a factor. 94. Derived Manifolds. (1) There are v\/(v-p)l p! combinations of the v independent elements ely e2...ev taken p together (p v). Let the result of multiplying the p elements of any one combination together in any arbitrary succession so as to form a product of the pth order be called a 'multi- plicative combination' of the pth order of the elements elt e2... ev. There are obviously v\j{y p)\ pi such multiplicative combinations of the pth order. (2) It is easy to prove formally that these multiplicative combinations are independent elements of the derived manifold of the pth order (cf. 20 and 22). For let Elt E2,... Eay... be these multiplicative combinations. Assume that !#! + 0L2E2 + . . . + OLcEa + . . . = 0. Then if E1 denote the multiplicative combination (eLe2... ep), the v p ele- ments ep+1, ep+2 -" ev do not occur in Elt and one at least of these elements must occur in each of the other multiplicative combinations. Now multiply the assumed equation successively by ep+1, ep+2...ev, then by 93 (4) all the terms become null, except the first term. Accordingly a2 {eYe2... ev) = 0. But (^...e,) is not zero by 93 (7). So ^ = 0. Similarly O2 = 03 a3=0; and so on. It follows that the sum of different multiplicative combinations cannot itself be a multiplicative combination of the same set of reference elements. (3) The complete derived manifold of the pth order is the positional manifold defined by the v\j{v p)\p\ independent multiplicative combinations of the pth order formed out of the v reference elements of the first order. Thus the manifold of the second order is defined by reference elements of the type (e^a), of which there are \v(v 1); the manifold of the {v l)th order is defined by reference elements of the type {eYe2... ^_i), of which there are v\ the manifold of the i/th order reduces to the single element (4) The product of any number of independent elements of the first order is never null, no factor being repeated. For let alf a2,... aa, (a v) be a independent elements of the first order. Then by 64, Prop. v. Corollary, v - a other elements aa+li ... av can be added to these elements, so as to form an independent system of v elements.
176 COMBINATORIAL MULTIPLICATION. [CHAR I. Let v equations hold of the typical form aP = alp i Then by 93, . a r){a T+1a T^2... av) = (a1a2... aaaa+1... av)= A (exe2 ... ev); where A is the determinant 2 olu 22 ... . Now A is not null, since all the elements aly a2,... av are independent, and (0x02... e,,) is not zero by 93 (7). Hence {axa2 ... aa) cannot be null (cf. 21). 95. Extensive Magnitudes. (1) Consider a product of /j, elements, where /jl v: let all these elements, namely o^ a2...afl, be assumed to be independent. Then they define a subregion of /jl 1 dimensions, which we will call the subregion A^. Let a/, a2f... a/ be /a other independent elements lying in the same subregion A^. Then fju equations of the following type must be satisfied, namely Hence by multiplication we find, remembering the law of interchange of factors, where A stands for the determinant ^11 i ^12 A-21, A^2, If the elements a/, a2'... a/ be not independent, then {a{a2 ... a/) = 0, and A = 0 ; and hence in this case also (oxV ... a/) = A (a^a... ). Thus if a/, a2'... a/ and a1? a2... a^ be respectively two sets of independent elements defining the same subregion, then [cf. 64 (2)] (2) Conversely, if (a/a/ ...a/) = (ox^ ...a^), where neither product is zero, then a/, a2'... a/ and a1} a2... a^ define the same region: or in words, two congruent products respectively define by their factors of the first order the same subregion of the manifold of the first order. For we may write (a/a/... aAt') = X(a1a2... a^). Multiply both sides by Oiy then (a/a/... a/c^) = 0. Hence by 93 (6) and 94 (4) a^ lies in the region (a/, a/... a/). Similarly 02 lies in the same region, and so on. Thus the two regions are identical.
96] EXTENSIVE MAGNITUDES. 177 (3) A product of fi elements of the first order represents an element of the derived manifold of the /xth order (cf. 20) at a given intensity; two congruent but not equivalent products represent the same element but at different intensities. Now an element of the manifold of the /xth order, which is represented by a product, may by means of subsections (1) and (2) be identified with the subregion of the manifold of the first order defined by that product. Thus the product is to be conceived as representing the subregion at a given intensity. Then we shall, consistently with this con- ception, use the symbol for a product, such as A^ (where A^ = axa2... aM), also as the name of the subregion represented by the product. (4) This symbolism and its interpretations can have no application unless a subregion is more than a mere aggregate of its contained elements. It is essentially assumed that a subregion can be treated as a whole and that it possesses certain properties which are symbolized by the relations between the elements of the derived manifold of the appropriate order. Thus a subregion of the manifold of the first order, conceived as an element of a positional manifold of a higher order, is the seat of an intensity and the term which symbolizes it always symbolizes it as at a definite intensity. (5) A positional manifold whose subregions possess this property will be called an extensive manifold. Let a product of p (p v) elements of the first order (points) be called a regional element of the pth order, and also a simple extensive magnitude of the pth order. Let regional elements of the first order be also called points, as was done in Book III.: let regional elements of the second order be also called linear elements or forces: let regional elements of the(V l)th order be called planar elements. Also it will be convenient to understand (regions' to mean regions of the manifold of the first order, unless it is explained otherwise. (6) Elements of the extensive manifold of the first order (i.e. points) will be denoted exclusively by small Roman letters. Elements of the derived manifolds, when denoted by single letters, will be denoted exclusively by capital Roman letters. 96. Simple and Compound Extensive Magnitudes. (1) There is one difficulty in this theory of derived manifolds which must be carefully noted. For example let the original manifold be of three dimensions defined by reference elements ely e2, e3,e4. The reference linear elements of the manifold of the second order are (e^), (e3e4\ (e^s), (e4e2), {eYe4\ (e2e3). Then any element P of the positional manifold defined by these six elements is expressed by P = 7r12 (e1e2) -f 7734 (e3e4) + tt31 (e ) + it^ (e2e4) + tt14 (e^4) + tt^ (e ). w, 12
178 COMBINATORIAL MULTIPLICATION. [CHAP. I. But if an element of this manifold of the second order represent the product of two elements 2 e and Xrje of the original manifold, it can be expressed as where (^o-) stands for ^rj^ - %a qp. But the following identity holds (frfo) (^4) + (^3) ( %) + ( W (ft*) = 0. Accordingly P does not represent a product of elements of the original manifold unless ^12^34 + ^13^42 + 7rl47r23 = 0. Thus only the elements lying on a quadric surface locus in the positional manifold of five dimensions (which is the manifold of the second order) represent products of elements of the original manifold. (2) Let a derived manifold of the pth order be understood to denote the complete positional manifold which is defined by the v \jp! (v p)! in- dependent reference elements. Let those elements of this derived manifold which can be represented as products of elements of the original manifold be called ' simple': let the other elements be called ' compound.' Let the term regional element [cf. 95 (5)] be restricted to simple extensive magni- tudes ; and let compound elements be termed compound extensive magnitudes or a system of regional elements. The latter term is used since every compound element can be represented as a sum of simple elements. Thus an extensive magnitude of the pth order is an element of the derived manifold of the pth order, and may be either simple or compound. (3) The associative law of multiplication identifies the product of two simple elements (Ep and E^) of derived manifolds of the pth and o-th orders (p + a v) with the simple element of the derived manifold of the (p + cr)th order formed by multiplying in any succession the elements of the original manifold which are the factors of E9 and Ev. Thus the product of any two elements, simple or compound, respectively belonging to manifolds of the pth and crth orders yield an element, simple or compound, of the manifold of the (p + cr)th order. But the product of two compound elements may be simple. In the case of simple elements, Ep and Eay the subregions Ep and Ea of the original manifold may be said to be multiplied together. 97. Fundamental Propositions. Prop. I. If Sp be an element (simple or compound) of the derived manifold of the pth order, and if (aSp) = 0, where a is a point [cf. 95 (6)], then Sp can be written in the form (aS^) ; where Sp-i is an element of the derived manifold of the (p l)th order. For the reference elements of the original manifold may be assumed to be v independent elements a, b, c__ Let Aly A2:,.BXi Z?2... be the
97] FUNDAMENTAL PROPOSITIONS. 179 multiplicative combinations of the pth order formed out of these elements. Let Alf A2... be those which do contain a, and let Bly B2... be those which do not contain a. Then we may write But by hypothesis (aSp) = 0 = (aAJ = (aA2) = etc. Hence multiplying the assumed equation by a we deduce Now (aB^, (aB2), etc. are different multiplicative combinations of a, b, c, etc. of the (/o + l)th order. Hence they are independent, and by hypothesis they do not vanish. Accordingly the above equation requires = 0 = /32 = etc. Hence Sp = aYAY + a2A2 + etc. = Corollary. If (e^ ... ea)(a p) be a simple element of the o-th order, and if a equations hold of the type eK8p = 0 (X = 1, 2... a), then Sp = {exe2... e^S^); where /S^ is an element of the (p cr)th order. Prop. II. If A denote a regional element of the o-th order, and B denote a regional element of the pth order (p a) such that the subregion A contains the subregion B, then A can be written (BG); where G is a regional element of the ( r - p)th order. For let the subregion B be defined by the a inde- pendent elements a1,a2...ap. Then to these independent elements a p other independent elements ap+1, ap+2... a , can be added such that the a elements a1? a2... ac define the region A. But A = A Oa ... apap+1... aa) = A (BG') = (BG); where Gr stands for the product (ap+1ap+2... aa), and G= AC Corollary. It follows from the two foregoing theorems that the com- binatorial product of two subregions is zero if they possess one or more elements in common. If they possess no common subregion their product is the region which contains them both. Prop. III. If Ap and Aa be two regional elements of orders p and a respectively, and if p + r = v + y, then we can write Ap = ((7Y5p_y) and Aa= (GyBa-y), where Gy is a regional element of the 7th order and Bp_y and Ba-y are regional elements of the (p y)th and (a y)th orders. For the subregions Ap and Aa must contain in common a subregion of at least 7 1 dimensions. Hence we are at liberty to assume the regional element Gy as a common factor both to Ap and Aa. 12 2
180 COMBINATORIAL MULTIPLICATION. [CHAR I. Prop. IV. All the elements of the derived manifold of the (v l)th order are simple. For let A and B be two simple elements of the (i/ l)th order. Then, since (y 1) + (y 1) = v + (y 2), we may assume by the previous proposition a regional element C the {v - 2)th order which is a common factor of A and B. Hence A = {aC)y and 5 = (bC), where a and b are of the first order. Thus A + J5 = (a + 6) C But a + 6 is some element of the first order, call it c. Hence A + B = cC. But cO is simple. Hence the sum of any number of simple elements of the (v l)th order is a simple element. Note. All the propositions of this chapter are substantially to be found in the Ausdehnungslehre von 1862. The application of Combinatorial Multiplication to the theory of Determinants is investigated by R. F. Scott, cf. A Treatise on the Theory of Determinants, Cambridge, 1880. Terms obeying the combinatorial law of multiplication are called by him 'alternate numbers.'
CHAPTER II. Regressive Multiplication. 98. Progressive and Regressive Multiplication. (1) According to the laws of combinatorial multiplication just explained the product of two extensive magnitudes Sp and Sa respectively of the pth and o-th order must necessarily be null, if p + cr be greater than v, where the original manifold is of v 1 dimensions. Such products can therefore never occur, since every term of any equation involving them would necessarily be null. In the case p + a v we are accordingly at liberty to assign a fresh law of multiplication to be denoted by the symbol 8p8 r. Let multiplication according to this new law (to be defined in 100) be termed ' regressive/ and in contradistinction let combinatorial multiplication be called progressive. Thus if p + a v, the product $p$a is formed according to the progressive law already explained. Such products will be called progressive products. If p -f a vy the product SpS^ will be formed according to the regressive law. Such products will be called regressive products. If p + a = v, the product SfiSa may be conceived indifferently as formed according to the progressive or regressive law. (2) In this last case it is to be noted that 898V must necessarily be of the form a (e^ ... ev), where elye2...ev are v independent reference elements of the original manifold. Since therefore such products can only represent a numerical multiple of a given product, we are at liberty to assume them to be merely numerical. Thus for example we may assume {e .. ev) = 1, and (SpSa) = a; where it is to be remembered that p + a = v. Let a product which is merely numerical be always enclosed in a bracket, as thus (e^ ..ev); other products will be enclosed in brackets where con- venient, but numerical products invariably so. 99. Supplements. (1) Corresponding to any multiplicative combination E^ of the /i,th order (fi v) of the elements elt e2... eVi there exists [cf. 65 (4)]
182 REGRESSIVE MULTIPLICATION. [CHAP. II. a multiplicative combination Ev_^ of the (y /z)th order which contains those elements as factors which are omitted from E^. Let it be assumed that (e1e2...ev) = 1. Hence (E^E,^) = {exe2 ...ev) = l. We may notice that if E'' _,,. be any other multiplicative combination of the {v - fji)th order, then (E^E^^) = 0. (2) The 'supplement' of any multiplicative combination E^ of the reference elements e1} e2...ev and of the fith order is that multiplicative combination Ev_^ of the (v fjb)th order which contains those reference elements omitted from E^ multiplied in such succession that (EnEv-n) 1. Let the supplement of E^ be denoted by \Etk. (3) Then if E^^ contain the same elements as \EIL but multiplied in any succession, Ev_^ will be called the multiplicative combination supple- mentary to E^. Then since (EfliEv_fl)= 1, we see that \Ef/, = (EllEv_fJi)Ev_ll] where {E^EV_^) is treated as a numerical multiplier of Ev_^. The fundamental equations satisfied by | E^ are (^ | Ep) = 1, and (#/1 E ) = 0 ; where E^ is any multiplicative combination of the /^th order other than E^. (4) The analogy of the above definitions leads us in the extreme cases to define |(e es ...ev)=l, and 11 = (e^ ... ev). Since (e^ ... ev) = 1, it follows from these definitions that, 11 = 1. (5) Let the supplement of a sum of multiplicative combinations of a given order be defined to be the sum of the supplements. This definition is consistent with that of subsection (2), since [cf. 94 (2)] the sum of different multiplicative combinations is not a multiplicative combination of the reference elements. Thus \(E* + E* +...)= |^+I-Bi/+- Let this definition be assumed to apply to the special case where E^ is repeated a times. Thus |(^ + ^M+... to a terms) =1^+1^+... to a terms = a {E^ Hence \(aEll) = a\Efl. Now let fj = v, and Efl = Ev=l. Then the above equation becomes |a = a. Also finally | (aE^ + olE^ + etc.) = a \ E^ + a' \ EJ + etc. (6) The symbol | may be considered as denoting an operation on the terms following it. It will be called the operation of taking the supplement.
100] SUPPLEMENTS. 183 This operation is distributive in reference to addition and also in reference to the product of a numerical factor and an extensive magnitude. For \(A+B)=\A + \B, and \(a.A)=\a.\A. (7) Let the symbol \\A denote the supplement of the supplement of A. If A be an extensive magnitude of the fith order, then For with the notation used above, E^ = \Jit^liiV^p) Ji/p fif and. Hence from (5) \\Efl = (EfJLEv_fl)\Ev_fl=(ElxEv__fJi)(Ev_tlEll)Etl. But (Ev-h.E^ = (- 1)*{v~^(EpE^)', and (E^E^) = 1. Hence \\E l = ("iy^^Etl. But A = taE^ and therefore | \A = (- Vf- ""* A, (8) It must finally be noted that the supplement of an extensive magnitude must be taken to refer to a definite set of reference elements of the original manifold, and that it has no signification except in relation to such a set. (9) The following notation for the operation of the symbol | on products will be observed. The symbol will be taken to operate on all succeeding letters of a product up to the next dot; thus a \ bed means that j (bed) is to be multiplied into a; and a | be. d means that | (6c) is to be multiplied into a and d into this product. Also a second | will be taken to act as a dot in limiting the operation of a former |: thus | A | B means that | B is multiplied into |Jl, and it does not mean \(A \B). Again, | placed before a bracket will be taken to act only on the magnitude inside the bracket: thus \(AB)C means that G is multiplied into \(AB). Johnson's notation with dots might be employed [cf. 20] : thus | ab. c.. d would mean that c is multiplied into \(ab) and d into this product. 100. Definition of Regressive Multiplication. (1) If A9 and Aa be extensive magnitudes of the pth and rth orders, where p + a v; then Ap and | Ao are extensive magnitudes of the (v p)th and (v cr)th orders, and (j/ p) + (v a) v. Hence \A9 and \Aa can be multiplied together according to the progressive rule of multiplication, already explained. Now the regressive product of Ap and Aa will be so defined that the operation of taking the supplement may be distributive in reference to the factors of a product. (2) Let the regressive product ApAa be defined to be an extensive magnitude, such that its supplement is 1^1-4*. In symbols, | ApAa =\AP\A T.
184 REGRESSIVE MULTIPLICATION. [CHAP. II. Since \A9 \Aa is of the (2v p r)th order, the regressive product ApAa is an extensive quantity of the (p + a z )th order. (3) If p + r=zpy then ApAa can be indifferently conceived either as progressive or as regressive. For if Ep, EPi etc. are the multiplicative com- binations of the reference elements of the pth. order, we can write Ap = %apEpi and A r=%a T\Ep. Hence (ApAa) = apaa + a^aj + etc., since (Ep \EP)=1} and (Ep \Epf) = 0. Also \AP = X*P\EP and \AO= 2a, \\EP = (- l)'^ ,# . Hence (\Ap\Aa) = (- iy *-p {apaa ( \EP . Ep) + apV ( \EP'. E;) + etc.}. Now (\EP.EP) = (- 1)p "-p (^p |^p) = (- iy . Thus finally (\AP \Aa) = apaa + apV + etc. = | {apaa + apV + etc.} = | (ApAa). (4) Again if p + a v, the product -4p^(r is progressive and the product |-4p |-4 r is regressive. But by definition of the regressive product \AP \A9 we have Now \\Ap = (-iyi"-riAp) and \\A r = (-iy^A ,. Hence | {\AP \Aa} = (- iy v-^+v v-^A/tA1f. Therefore, taking the supplement of each side, || {\AP \Aa} = (- i)pe-p +* *-" | ApAff. Now \AP \Aa is an extensive magnitude of the (v p o-)th degree. Hence || {\Ap\Aa] = (- 1) -*-* p+o) l^l^. AISO ( 1) (v-P-*) (P+ r) = ( 1 )P (V-P) +* ("- ) Hence |4p4 r=|4p|4 r. (5) Finally therefore in every case, whether the product ApAa be pro- gressive or regressive, we deduce | APA T = \AP \AO. 101. Pure and Mixed Products. (1) A product in which all the multiplications indicated are all progressive or all regressive (as the case may be) is called pure; if the multiplications are all progressive the product is called a pure progressive product; if all regressive, a pure regressive product. Thus if A, B, G and D be extensive magnitudes, the product AB. CD is a pure regressive product if the product of A and B is regressive, and that of G and D, and that of AB and CD. (2) A product which is not pure is called ' mixed/ Thus if the product of A and B is regressive, and that of G and D is progressive, then the product AB. GD is mixed.
102] PURE AND MIXED PRODUCTS. 185 (3) A pure product is associative. This proposition is true by definition, if the pure product be progressive. If the product (P) of magnitudes A, B, C, etc. be a pure regressive pro- duct, then the product of \A, \B, \C, etc. is a pure progressive product. But this product is associative. Hence 1^ = 1^- \B \G ... = \A \B \G.... Taking the supplements of both sides hence P = A .BC ... = ABC.... For instance, if the product AB. CD be pure, we may write AB.GD = ABGD. A mixed product is not in general associative. 102. Rule of the Middle Factor. (1) We have now to give rules for the identification of that extensive magnitude of the (p + a z/)th order which is denoted by the regressive product ApAa. This will be accomplished by the following theorems. (2) Proposition A. Let Ep, Ea, ET be three multiplicative combinations of the reference units of the pth, ath, and Tth orders respectively, and let p + cr + t = v. To prove that EpEa. EpEr = (EpEaEr) Ep. It will be noticed that the products EpEff and EpET are progressive, while the final product of EpEa and EPET is regressive; and also that (EpE^E^ is either zero or 1. Case I. Let (EpEaET) = 0. Then since in this product there are only v factors of the first order, one of the v reference elements must be absent in order that another one may be repeated. Let #! be the absent element. Then e1 is contained neither in EpEa nor in EPET. Hence it is contained both in {(EpE^) and in \(EPET). Therefore | {Efi9. EPET) = | {EPEC) \ (EPET) = 0. Hence EpEa . E,ET = 0 = (EpEaET) Ep. Case II. Let (EpEaET) = 1. In this case no factor of the first order is repeated in the product Hence [cf. 99 (3)] \E. = (EaE,ET) EfiET, \Er = Hence \{E,E^\Er\Em = {ErEJS,){E^9E^EtE..E9Er .........(1). But from 99 (3) \(ETEa) = (E^E,,) Ep...........................(2).
186 REGRESSIVE MULTIPLICATION. [CHAP. II. Also (ErEpE ) = (- l)*(p+* (EpEaET), {E EpEr) = (- iy (EpEaET)} and (ErEaEp) = (- 1)p ^(p+") {EpEvEr). Therefore from equations (1) and (2), and remembering that (EpE rET) = 1, it follows that EpE r m EpEr = (EpE rEr) Ep. (3) Proposition B. If -4P, J.^, J.T be any simple extensive magnitudes of the pth, o-th and Tth orders respectively, such that p + r + r = v, then ApA j. ApAT = (ApAffAr) Ap. For let us assume that this formula holds for the case when the factors of the first order of Ap, Aa and Ar are composed out of a given set of v inde- pendent elements a1} a2... av. Then we shall show that the formula holds for products formed out of the set a/, a2... av, where a/ = Sa^a^ and replaces ax. Now a2 may occur in Ap, Aa, or AT. Firstly assume that it occurs in Aa. Let Aa = a^A^^', and let A9'= a(Aa_Y. Then AJ is what Aa becomes when a/ is everywhere substituted for a1; and Ap and J.T are unaltered by this substitution. Thus AJ = a/J-a.j = Sa^ (a^Aa^). Hence ApAJ. ^.P^1T = 2a^ (^.pa^Jl^!. -4PJ.T). But each product of the type A9a^A9^. APAT is such that the factors of the first degree in Ap, a^A^^, and Ar are composed of the set of elements Ox, a2... av. Accordingly by our assumption ^LpOj^xLff__i . xLp^LT ^ y^LpQjfljcL r__ixj.tj -o.p. Hence APAJ. APAT = Sa^ (^.pa^^.!J.T) J.p = (Ap. SaMa^. Aa^xAr) Ap = (^-Aptt-j -o.(r_1-AT^ ^4p = (^-o-p^lo- -"-T) -"-p. Secondly, it can be proved in exactly the same manner that if aY occurs in AT, so that when a/ is substituted for a1} AT becomes Ar' and Ap and A* are unaffected, then ApAa . APAT' = (APA9AT')AP. Thirdly, assume that a2 occurs in Ap. Let aY be changed into a/ = a^ + a2a2 and let ^lp = (aYA9^y and If a2 occur in Ap^, then Accordingly ^.p' is merely a multiple of j4p, and we deduce immediately that ApfAo. AP'AT = a2J.p^la. ApAt = a? (ApAaAT) Ap \Ap A^At) Ap. If a2 does not occur in Ap^, suppose that it occurs in Aa. Let Aa = Then AJA^ = a2 (oaAp^^A^^) + And
102] RULE OF THE MIDDLE FACTOR. 187 Hence ApAa. Ap'Ar = a^ApA*. ApAr + a^ApA^ . a%A9^AT. Now ApAa. APAT = (ApAaAr) Ap. And ApAa. Therefore by substitution, ^4p ^1^ . Ap Ar = ! But (^p'J.aJ.T) = o^ (J.pila^lT), and a Hence finally ApAa . AP'A T = {A?'AoAr) Ap'. But by repeated substitutions for al5 a/, etc. of the type a1/ = a1a1-\- a,2a2, ai" = ft^i + /93a3, and so on, ax is finally replaced by any arbitrary element Thus if any element of the set a1} a2... av be replaced by an arbitrary element, the formula still holds. Hence by successive substitution the v elements o^, a2... av can be replaced by v other elements blf b2...bv. But if Ep, Ea ET be simple magnitudes formed by products of the reference elements elf e2...ev, the formula, EpE r.EpET = (EpE rET)Epi has been proved to hold by proposition A. Therefore if Ap, Aa, AT be simple magnitudes formed by products of any set of elements ax, a2... av the formula, r)Ap, holds. *(4) Corollary. It is easy to see that the formula still holds if Aa and AT be compound. But it does not hold if Ap is compound. Proposition B is the foundation of all the formulae in this algebra. The following important formulae given by propositions C and D can be deduced from it. (5) Proposition C. ApAa. APAT = {ApAaAr) Ap, when p + r + t = 2v. In this case the products ApAa and ApAr are both regressive. Hence the products \AP \Ao and \AP \AT are both progressive, and {v p) + {v o-) 4- (v t) = v. Hence by proposition B, \AP \Aa. \AP \AT = (\AP \Aa \AT) \AP. Therefore by taking supplements of both sides ApAff. ApAT = (ApAffAr) Ap. (6) Proposition D. A9Aa. AoAr (ApAcAT) Aa, and ApAr. AaAT = (ApAaAT) AT; where p + r + r = v or 2v. These formulae follow immediately from propositions B and C.
188 REGRESSIVE MULTIPLICATION. [CHAP. II. For ApAa. AaAr = (- iy AaAp. AaAr = (-iy*(A rApAT)A T = (ApA TAr)A(T. Similarly for the other formula. (7) These formulae may all be included in one rule, which we will call the ride of the middle factor, given in the following proposition. Proposition E. Let Ap and Aa be two simple extensive magnitudes of the pth and ath. order such that p + r = v + 7. Then the regions Ap and Aa have a common region of at least 7 1 dimensions. Let Gy be this common region. Then we may write (cf. 97 Prop. III.) Ap = Bp-yCyy and Aa = (775a_y. And it is easy from the foregoing propositions to prove that Ap Aa = Bp.yCy . Aa = (Bp_yAa) Gy = Ap . CyB T^y = \ApBa-y) Gy. These formulae embody the rule of the middle factor. 103. Extended Rule of the Middle Factor. (1) But this rule in its present form is not very easily applicable in most cases. Thus sup- pose that the complete manifold be of three dimensions, so that v = 4, and let Ap=pqr, and A T = st\ where p, q, r, s, t are elements of the complete manifold. Then to find the product pqr. st, the rule directs us to find the element x wrhich the line st must have in common with the plane pqr and to write either pqr in the form uvx or st in the form xz; and then pqr. st = uvx. st = (uvsi) x, and pqr . st= pqr . xz = {pqrz) x. But no rule has yet been given to express x in terms of p, q, r, s, t This defect is remedied by the following proposition embodied in equa- tions (1) and (2) of the next subsection, which we will call the * extended rule of the middle factor.' (2) Proposition F. Let Ap and Ba be simple extensive magnitudes of the pth and o-th. orders respectively, and let p + a- = v + 7, where 7 must be less than v. Let G{y, G(y, etc. denote the multiplicative combinations of the 7th order which can be formed out of the factors of the first order of Ap. Then we may write Ap = A -yC = A?-yC* = etc., where A{p-y, A{p-y, etc. are extensive magnitudes of the (p 7)th order. Then according to the extended rule of the middle factor ArB^iAtlyBJCn + iAf-yBJCy + ete................(1). Similarly let D{y, Dy\ etc. be the multiplicative combinations of the 7th order formed out of the factors of the first order of Ba. Then we may write JB, = D*Bf2-y = 2) B Ly = etc., where B{^_y) 2C_y, etc. are extensive magnitudes of the (a 7)th order.
103] EXTENDED RULE OF THE MIDDLE FACTOR. 189 Then according to the extended rule of the middle factor ApB(r = (ApB%)Dy + (Ap ?_y)D +etc................(2). Equations (1) and (2) form the extended rule of the middle factor which has now to be proved. Let alt a2... ap be the p factors of Ap and let blt b2 ...ba be the a factors of Ba. Let v p other elements ap+1, ap+2... av be added to a1...ap, so as to form a set of v independent elements. Then we may write Ba in the form where B , B , etc. are the multiplicative combinations of the o-th order of the elements aly a2...av] and any number of the coefficients /3lf fi2, etc. may be zero. Also, remembering that (p y) + a = v% let the index-notation be so arranged that B contains those a's which do not appear in AlpLy} and B contains those which do not appear in Af_y, and so on. Then it may be noted that to every magnitude A^}y there corresponds a magnitude B^\ but not necessarily conversely. Furthermore it is obvious that when X 4= f1 Then ApBa = X Now Cy^ must represent a subregion contained in the subregion jB^}; since Q^ is a product of 7 of the a's which do not appear in Ap*lv and B[^ is a product of all those a's which do not appear in Ap*}y. Hence by the rule of the middle factor Also since A^yBlK) = 0, we deduce Hence finally ApBa = 2 (A^yBv) f ; which is the equation (1) of the enunciation. An exactly similar proof yields equation (2). (3) The following formulae are important special examples of this ex- tended rule of the middle factor. Let v = 3, the complete manifold being therefore of two dimensions. Then pq .rs = (prs) q (qrs) p = (pqs) r (pqr) s...............(3). Let v = 4, the complete manifold being therefore of three dimensions. Then pqr ,st = st. pqr = (pqri) s (pqrs) t r + (rpst) q + (qrst)p.....................(4).
190 REGRESSIVE MULTIPLICATION. [CHAP. II. And pqr . stu = stu . pqr = (pqrs) tu + (pqru) st + (pqrt) us = (pstu) qr + (rstu)pq + (qstn) rp ........................(5). (4) Take the supplements of these formulae. When v = 3, the supplement of a magnitude of the first order is a magnitude of the second order. Let P, Q, R, S be magnitudes of the second order such that P = \p, Q=\q, etc. Then by taking the supplement of (3) we deduce PQ.RS = (PRS) Q - (QRS) P = (PQS)R- (PQR)S......(3'). Again let v = 4; then the supplement of a magnitude of the first order is one of the third order. Let P, Q, R, S, T, U be put for \p, \qy |r, \s, \t, \u\ and let the supple- ments of equations (4) and (5) be taken. Then PQR . ST = ST. PQR = (PQRT) S - (PQRS) T = (PQST) R + (RPST) Q + (QRST) P .........(4'). And PQR . STU= - STU. PQR = (PQRS) TU + (PQRU)ST + (PQRT) US = (PSTU)QR + (RSTU)PQ + (QSTU)RP..................(5'). In fact by taking supplements any formula involving magnitudes of the first order is converted into one involving planar elements, i.e. magnitudes of the (v l)th order; where the complete manifold is of v 1 dimensions. 104. Regressive Multiplication independent of Reference Ele- ments. (1) The rule of the middle factor and the extended rule disclose the fact that the regressive product of two magnitudes A and B is inde- pendent of the special reference elements in the original manifold which were chosen for defining the operation of taking the supplement. Accordingly regressive multiplication is an operation independent of any special reference elements or of their intensities, though such elements are used in its defini- tion for the sake of simplicity. Also it is independent of the fact that the product of the v reference elements was taken to be unity for simplicity of explanation. Thus the product may be assumed to have any numerical value A which may be convenient [cf. 98 (2)]. It would have been possible to define regressive multiplication by means of the rule of the middle factor. It would then have been necessary to prove that it is a true multiplication, namely that it is distributive in reference to addition. (2) It is useful to bear in mind the following summary of results re- specting the multiplication of two regions Pp and Pffi of the pth and crth orders respectively:
105] REGRESSIVE MULTIPLICATION INDEPENDENT OF REFERENCE ELEMENTS. 191 If p + 7 vy then PpPa is progressive and represents the containing region [cf. 65 (6)] of the two regions Pp and Pa ; unless Pp and Pa overlap, and in this case the progressive product PpPa is zero. If p + a vy then PpPa is regressive and represents the complete region common both to Pp and Pff; unless Pp and Pa overlap in a region of order greater than p + cr vy and in this case PpPa is zero. If p + a- = Vy then {PpPa) is a mere number and can be considered either as progressive or regressive. The only formulae which in practice it is necessary to retain in the memory are the extended rule of the middle factor [cf. 103] and propo- sition G of 105. 105. Proposition G. If a1} a2...ap be p points in a region of v 1 dimensions (v p), and if Blt B2... Bp be p planar elements, then ,...BP) = ), (a2B2) ...(a2Bp) For assume that the formula is true for the number p 1 respectively of points and of planar elements, to prove that it is true for the number p; where p v. Let A rT denote the minor of the element (aa5T) of the above determinant. Now the product of the p+1 regional elements of the pth order (o1a2...ap), Bl9 B2... Bp is a pure regressive product and is therefore associative. Hence (aYa2... ap. BXB2... Bp) = {(a^ ... ap. J52)B2BZ... Bp} = (aA) (a2a3... ap. 5^3... Bp) + Os^) Oa ... ap. 52S3... Bp) + ... + (a^) (axa2... ap_2. 5253 ... Bp). But since the theorem holds for the number p 1, a2a,... ap.i?2i?3 ...Bp = An, with /3 similar equations. Hence ...Bp) = (a^) An + (a,^) A21 4- ... + (a^) Apl ... ap. (aA), (a2B2) ... (a2Bp) But when p = 2, . BXB2) = Therefore the theorem is true universally.
192 REGRESSIVE MULTIPLICATION. [CHAP. II. 106. Muller's Theorems*. (1) Let A, B and G be three regional elements of the /eth, pth and crth orders respectively. Also let /c+p + o-=v+r, where r is positive and not zero. Then the products A . BC, ABC and AGB are mixed products. Hence in general [cf. 101] A . BG, ABC, and AGB are not congruent. The conditions will now be investigated which are necessary that an addition relation of the form, A.BC=XABC + nACB ........................(i), may exist. (2) It will be only necessary to consider the proofs for the cases when BG is a progressive product, and consequently when the product of BC into A, namely A . BC, is regressive. For if it has been proved in this case that under certain conditions the above equation (i) holds, then by taking supple- ments \A.\B\C = \\A\B\C + fi\A\C\B. Let A'=\A, B' = \B, C = \C. Then B'C is regressive, and the product of B'C into A', namely A'. B'C\ is progressive. Also the conditions which hold between A, B, C can be interpreted as conditions between A', B'} C. Hence it follows that, when B'C is regressive, under certain conditions A'. B'C' = XA'B'C (3) Let the p points which are the factors of B be bl9 62,... 6P, so that B= bj)2 ...bp. Also let the a points which are the factors of C be clt c2 ... ca, so that C=c1c2... ca. Let the multiplicative combinations of the /juth order (//, p) formed out of b1} b2)... bp be written 2?M, 2?/, B/, and so on. Also let Bp-p, B'p_^, B^p^f,., and so on, be the complementary multiplicative combina- tions of the (p fi)ih order formed out of b1} b2, ... bP) so that B = (BpBp^) = (B^B'p^) = (B^'B"^) = etc. Let this convention hold for any number //,; and also a similar convention for the multiplicative combinations formed out of c1} c2, ... ca, which are the factors of C. (4) Assume that the product BC is progressive. Then the multiplication symbolized by the dot in A . BC is regressive. Also BC = bj 2... bpCtfz ...c r = Dy say. Let the convention of the previous subsection apply to the multiplicative combinations formed out of the p + cr factors of the first order which compose D. Then by the extended rule of the middle factor [cf. 103] A.BC= AD = Z(AD .^)DT .....................(ii), * Cf. Emil Miiller, * Ueber das gemischte Product,' Mathematische Annalen, vol. xlviii. 1897,
106] muller's theorems. 193 where DTJ Z)T\ etc., are the multiplicative combinations of the Tth order formed out of blf 62, ... bp, cly c2, ... V If t be less than both p and r, some of the multiplicative combinations DT, DT', etc., contain only 6's, some only c's, and some both 6's and c's. If t be less than p and greater than a (assuming p a), then some of the DT's contain only 6's, and some contain both 6's and c's ; but none contain only c's. If r be greater than both p and (T, then all the i)T's must contain both 6's and c's. (5) Let the products BC, AB, AC be progressive. Then p + G v, k + p v, k + a v. By the extended rule of the middle factor ABC = 2 (ABCa_T) GTi ACB = %(ACBP_T) BT. Hence if a relation of the form of equation (i) holds, no DT's must exist which contain both 6's and c's. But this condition can only hold when t = 1. Hence the condition is that /c + p + r = v + l. Also, remembering that cIJLBC(fll = (-1)pBchC(Jtll = (-iyBC, equation (ii) becomes A. BC = (-lY%(ABCl?l1)Cp + 2(AB/fl1C)bll. And ABC = 2 (ABCUd c*, ACB=t (AGB^\) ^ = (-1)- p- 2 (AB^ G) 6M. Hence A . BC = {-iyABC + {-1) ^ ACB...............(iii). This is the required equation of the form of equation (i). (6) Let the products BC, AB, AC be regressive. Then p +a v, tc+p v, tc-\-a v. Taking the supplements, \A, \Bt \C are of orders (y k), (v p), (v a) respectively, and \B \ C, \A \Bt \A \C are progressive. Hence in order that a relation of the required form may hold, by the previous subsection, (v-K) + (v-p) + (v-v) = v + li that is, k + p + a = 2i 1. Also from equation (iii) \A . \B\C = {- iy-p \A \B \G + (- \yv~^ {v-o-l} \A \C\B. Hence A .BC= (- iy- ABC + (- 1 ) "-') ^-^ ACB............(iv). (7) Let the products BC and AB be progressive ; and let the product A C be regressive. Then k -hp + a = v -f t, k + p vy p + cr v, k + a v. Hence r t, k t, p t. w. 13
194 REGRESSIVE MULTIPLICATION. [CHAP. II. By the extended rule of the middle factor ABC = 2 (ABCa_T) CT, AC = 2 (ACV_K) CK+a_v. Hence ACS = 2 (ACM) 0 + r 5. Accordingly if a relation of the form of equation (i) holds, the DT's [cf. subsection (4)] must consist of two classes only, namely those composed only of c's, and those which contain all the 6's. But this is only possible if P = i. In this case B is of the first order and will be written b. Then remembering that cj c^=(- iy be, cu cff_T+1 = be, and that k + a v = t 1, cr r -\- \ = v k, equation (ii) takes the form A . BC = (- \y 2 (A6Ca_T) CT + 2 (4CMH) 6CU, = (-l)TA50 + (-l)T-MCS...........................(v). This is the required equation of the form of equation (i). (8) Let the products BC and AB he regressive and the product AC he progressive. This case can be deduced from subsection (7) by the method of subsection (6). The necessary condition for the existence of the required addition relation is p = v 1. Then from the assumptions it follows that Also A . BC = (- 1)"-T ABC + (- I)"-*-1 ACB...............(vi). (9) Let the products BC and AC he progressive, and the product AB he regressive. Now A . BC= (- iy A . CB. Hence this case can be deduced from that of subsection (7). The necessary condition is that r = l. Then A . CB = (- iy ACB + (-1)7"1 ABC Hence A.BC = (- l)*^"1 ABC + (-iy+rACB...............(vii). (10) Let the products BC and AC he regressive, and the product AB he progressive. Then from the previous subsection a- = v 1, And A . BC = (- l)^-p-r-i ABq + (_ ^-p-r ^(7^ ^(7S..................(viii).
107] muller's theorems. 195 (11) Let the product BG be progressive, and the products AB and AC be regressive. Then AB = X(ABV_K)BP+K_V) AC=t(ACv_K)Cff+K_v. Hence ABC=t {ABV_K) Bp+K_v C, AGB=t(ACV_K) Ca+K_v B. Thus the DT's of subsection (4) equation (ii) must either contain all the 6's or all the c's; and thus the Dp+(T_T's of the same equation must contain only b's or only c's. Hence the Dp+(r_T's are of the first order, that is to say, p + o- t = 1. But tc -\-p + 7 = v + t. Hence the required condition is that K V 1. Then AB^ZiAbJB^, AC = (AcJ C ; where J3J b = 5, C^lx cM = G. Thus ABC= 2 (46,.) BJ C, ^1(7B = t (AcJ G^lx B. Now b{;\ cb^=(-iy Bf^;\ c=(-1 * ^c, and OtMii 5^ = (- iy Cfclx c^ = (-1 GB = (- 1)^+p 5C/. Hence by comparing with equation (ii) of subsection (4) A.BC=(-iyABC + (-iy( +VACB ...............(ix). (12) Let the product BG be regressive, and the products AB and AC be progressive. Then from the previous subsection the condition is Also A .BC = (-iy~* ABC+(-lY?-ti y- +v AGB.........(x). (13) It has nowhere been assumed in the foregoing reasoning that A is simple. Accordingly A may be compound. 107. Applications and Examples. (1) The condition that an ele- ment x may lie in a subregion Pp of p 1 dimensions is, xPp 0. This equation may therefore be regarded as the equation of the subregion. (2) The supplementary equation is, \x jPp = O. The product of \x and \PP is regressive, and the equation indicates that \x and \PP overlap in a regional element of an order greater than the excess of the orders of | x and \PP above v. Now the order of \x is v 1, and the order of \PP v p. Hence the order of the common region is greater than P that is, is greater than v p 1. But the subregional element \PP is only of order v p. Hence | Pp must be contained in the plane | x. This is the signification of the supplementary equation. 13 2
196 REGRESSIVE MULTIPLICATION. [CHAP. II. (3) The supplementary equation can be regarded as the original equation and written in the form where Xv_x is a planar element, and Pp is a subregional element of the pth order. The preceding proof shows that this equation is the condition that the plane Xv-X contains the subregion Pp. The supplementary equation is now \XV^ |Pp = 0, and signifies that the point \Xy_i lies in the region |PP of v p l dimensions. (4) The theory of duality also applies, and xPp = 0 can be regarded as the condition that the subregion Pp contains the given point x\ and the equation, XV-1PP = O, as the condition that the subregion Pp is contained in the given plane Zr_lt (5) In the previous subsection it has been assumed that Pp is a regional element, that is to say, is simple. Now let Sp be a compound extensive magnitude of the pth order. Then in general it is impossible to satisfy the equation xSp = O, except by the assumption that x = 0. For xSp is an extensive magnitude of the p + lth order; but this manifold is defined by 7----------zr-fr------r^-, independent units (cf. 8 94). Hence if (v-p-l)\{p + l)l r v a / ccSp 0, the coefficient of each of these units, as it appears in the expression xSp, must vanish. Thus there are 7 ------ ^ ----- equations to be satisfied. But in x(= 2 e) there are only v 1 unknowns, namely, the ratios f ?i, fa--- fi" But if /o be any one of the numbers 2, 3 ... v 2, In these cases the requisite equations cannot be satisfied. If p = l, then Sp is a point and must be simple: the equation xSp = 0 then means that x = Sp. If p = v 1, then #p is a planar element and must be simple (cf. 97, Prop. IV). (6) Let P,_i be the planar element Then irlf tt2 ... irv are the co-ordinates of the planar element Pv-X with respect to the reference elements e1 e2 ... ev. Also if x be 2f^, then v) (e^e2... ev). Hence the equation (xP^j) = 0, is equivalent to the usual equation of a plane, namely, = 0.
107] APPLICATIONS AND EXAMPLES. 197 And conversely Pv-lf as defined in this subsection, is a planar element in the plane which is denned by the equation (7) Another simple method of obtaining a slightly different form of a planar element corresponding to the plane 71-^+... +7^=0, is found by means of 73 (2). The point in which the plane cuts the straight p p line eYe9 is by that article-------- . Hence by multiplying the v 1 such points which lie on the v \ such straight lines meeting in e1, a planar element in the plane is found to be Hence p=i /g ^ IT ( ------* ) = irxe ... ew- ir.e^ ... ev Therefore by multiplying out the left-hand side and comparing the coefficients of the term e2e3... ev on the two sides, (-l)"-'7r17r2...7r/ff (^ - ^ e3... ev + ...+(- If 17rl^1e2... ev_A. This factorization of the right-hand side of the above equation into a product of v 1 points forms another proof of 97, Prop. IV. (8) Among special applications of these theorems we may notice that the condition that x may lie on the straight line joining a and b is xab = 0; the condition that x may lie in the two dimensional region abc is xabc= 0; the condition that x may lie in the three dimensional region abed is xabed = 0. (9) Let the complete manifold be of more than two dimensions so that the multiplication of linear elements is progressive. The multiplication of a planar and linear element together is necessarily regressive. Then two lines ab and cd intersect if abed = 0. For this is the condition that a, b, c and d lie in the same subregion of two dimensions. The point where a line ab intersects a given plane Pv-i is Pv-Y. ab. But by 103 (2) (the extended rule of the middle factor) P,_2. ab = (Pv-Jb) a - (P^a) b.
193 REGRESSIVE MULTIPLICATION. [CHAP. II. 107 If the line lie entirely in the plane, (P(/_16) = 0, and (P^a) = 0; hence Pv_1.ab=Q. If the planar element be written as the product cxc(i ...(; _,, then the point of intersection of the line ab with it can be written c^... cv-.Y. ab. And by 103 (2) cxc2... c_i. ab = {cxc2... cv^ab) cu-Y + (- \)v (c^a ... cv_3ab) c_3 4- ... The last form exhibits the fact that the point of intersection lies in the plane c1c2...cp-1] while the form (PV_J)) a (Pv^la)b exhibits the fact that the point of intersection lies on the straight line ab. (10) Two planar elements Pv^ and Qv^ must intersect in a region of v 3 dimensions, or in other words the extensive magnitude Pv^ . Qv-X is a regional element of the (v 2)th order. Let such subregions be called sub- planes. The magnitudes denoted by \PV-Y + /^Qv-i for varying values of the ratio X/fi are planes containing the subplane Pv-Y. Qv-X, common to Pv-Y and ft,-i. In regions of three dimensions straight lines and subplanes are identical. (11) If four given planes Pv-lt Qv-lf Rv-i, Sv^ contain a common sub- plane Z.,-2, then the four points of intersection of any straight line with these planes form a range with a given anharmonic ratio. For let Pv i = Lv-2 a, Qv-i Lv_2 b, Hv i = Lv-i v $v-i = Lv_2 d. Let pq be any line, and assume that p lies in Pv^ and q in Qv^. Then Lv_2p = ^Pv^1} and Lv-2q = pQv_1. Also let pq intersect Rv^ and _! in r and s. Then r = R^ . pq = (R^ q)p- (Rv-! p) q = - (Lv_2qc)p + (Lv_2pc)q=-p (Q,-i c)p + w (P^c) q. Similarly s = p ((^-i d)p + vs {Pv-^ d) q. Hence the anharmonic ratio (pq, rs) = ^-1 VeT"1 \ - (V a)(PC) This ratio is the same for all lines pq; it can also be expressed as ^ a). (12) If Rv_, = XP^ + /aQ^, and Sv^ = VP^ + m'Q,-i, then c = \a + fib, and c? = \'a + /a'6. Also since (aP,_j) = 0 = (6QV_!), we have ^l9 \ Qv_u \ Rv_ly and j Sv^ are four collinear points with the same anharmonic ratio, XfjujX'/j,, as the four planes. Note. In developing the theory of Regressive Multiplication the Amdehnungslehre von 1862 has been closely adhered to.
CHAPTER III. Supplements. 108. Supplementary Regions. (1) The supplement of a regional element Pp of the pth order is a regional element | Pp of the (v p)th order [cf. 65 (4) and 99]. The two subregions Pp and | Pp are called supple- mentary. In particular \ x is the supplementary plane of the point x, and x the supplementary point of the plane | x. (2) If Pp be expressed as the product of p points ply p2, ...pp, then taking the supplement I^P=l^l|P2...|iV Hence if Pp be the containing region of the p independent points, then Pp is the common region of the p supplementary planes of those points. (3) If Pp and Pp be two regional elements both of the pth order, then (Pp | Pp) is merely numerical. Hence (P,|P/) = |(PP|P/) = (|PP||P/) = (- 1)p "~ (|Pp.P/) = (P/|PP). (4) Thus if y lies in the supplementary plane of x, then (y \ x) = 0 =(x \ y). Hence x lies in the supplementary plane of y. (5) Definition. Points which lie each in the supplementary plane of the other will be called mutually normal points. If the points x(= 2 e) and y (= S^e) be mutually normal, then (x\y) = (%lVl + 2?/2 + ... fvi/,) = 0. (6) A point x does not in general lie in its own supplementary plane, unless it lies on the quadric (^1^ = 0 = ^ + 1/+... + ^. Let points which lie in their own supplementary planes be called self- normal ; and let the quadric which is the locus of such points be called the self-normal quadric. 109. Normal systems of points. (1) All the points normal to a given point x1 lie in the plane |^. Let x2 be any such point, and let x3 lie in the subregion \xx \x2, and xA in the subregion |#i|#2|#3; and so on; and finally let xv be the point \xi \x2 ... # -!. Then assuming that none of these points
200 SUPPLEMENTS. [CHAP. III. are self-normal, we have deduced a system of v independent mutually normal points, starting with any arbitrary point xx. (2) Definition. Let a system of v independent mutually normal elements be called a normal system. (3) The intensities of the normal system of points as denoted by xx, x2... xv are arbitrary. Definition. Let any point p be said to be denoted at its normal intensity when (p \p)= 1. Note that the normal intensity of a point is not neces- sarily its unit intensity. (4) Then if xly x2, ... xv be a normal system of elements at their normal intensities, the following equations are satisfied (#i |#i) = (#21^) = etc. = 1, and (xp \xa) = 0, where p =f= a. If xp = ^lpe1 -f ^2pe2 + %vsfiv these equations can be written with v 1 other similar equations, and lp 1 T -f f^ + ... -f ,pf w = 0, with \v{y 1) 1 other similar equations. (5) Also by 97, Prop. I., the following equations hold .. xv, xv= Hence (xY \ x ) = Xj {xxx2... xv) = 1. Therefore X2 = ------- = - X2 = X3 =...=(- I)*'-1 V \XjX.2 ... Xv) Also since (^ 1^) is merely numerical, then by 99 (5) (xx \x1)=\(xl\xl)=(\xl\\x1)=------------ {x s ...xv\x *... xv). (x r,... xvy Hence by 105 and by the previous subsection of the present article \x2 ... xv)2 (x3 \x2), (x3 \xs), ... (x31 1 Hence (x x2 ... xv)2 = 1, and therefore {xYx2... ) = + 1. Now if xp be at its normal intensity, then (cf. 89) xp is also at its normal intensity. Hence by properly choosing the signs of xly x2, ... xv, we can make (xxx2... xv) = 1. Thus finally with this convention x1 = x2x3 ...xv, Ix2 = xxx3 ...xv, ... !# =( l)1^"1 ^j^2... # _!. (6) Hence the operator | bears the same relation to the normal system x1, ip2... xv at normal intensities as it does to the original reference-elements. Accordingly in the operation of taking the supplement the original reference- elements may be replaced by any normal system at normal intensities.
110J EXTENSION OF THE DEFINITION. 201 110. Extension of the definition of Supplements. (1) This possi- bility of replacing the original reference-elements by other elements in the operation of taking the supplement suggests an extended* conception of the operation. In the original definition the terms e1} e2...ev represent the reference- elements at their normal intensities as well as at their unit intensities. But suppose now, as a new definition which is allowable by 109 (3) and (G), that the normal intensities of these reference-elements are elt e2... ev. Then by hypothesis [cf. 109 (3)] and so on. Also it must be assumed that e^ ... ev(eLe2... e^ 1. Let ! 2... ev = A = - ------- Then \ex e2es... evy \e2 ^eLe3... evy and so on. (2) This extended definition in no way alters the fundamental properties of the operation denoted by |. For this operation has been proved to be referred to an indefinite number of normal sets of points and cannot therefore be dependent on the symbolism by which we choose to denote one set of them. Thus it follows that the symbol | obeys the distributive law both for multiplication and addition. Also \\PP = ( l)p^~p)PP7 where Pp is of the pth order. But (e11 eY) = -o, (e | e2) = , ... (ev \ ev) = ~ . ^1 ^2 v Also it is not necessary that e2, e2... should all be real; thus any number of their squares may be conceived as being negative. (3) The self-normal quadric is defined by the equation (x \x) = 0; that is by 7^ 4- 2/e22 + ... frfeS = 0. If l7 e2... ev be all real, this quadric is purely imaginary: but if some of them be pure imaginaries, this quadric is real. Since only the ratios of 1 2... v are required for defining the self-normal quadric, it is allowable when convenient to define, e^ ... = 1. Hence in this case (e^ ... ev) = 1. (4) The equation of the supplementary plane of any point x is {y \x) = 0; that is, if x = 2 e and y = Xye, the equation W*i2 + IWv + + V*/eS = 0. But this is the equation of the polar plane of x [cf. 78 (1)]. Hence the method of supplements is simply a symbolic application of the theory of reciprocal polars and its extension to linear elements and to other regional elements in manifolds of more than three dimensions. * This extension is not given by Grassmann.
202 SUPPLEMENTS. [CHAP. III. (5) Normal sets of elements are obviously sets of polar reciprocal elements forming a self-conjugate set with respect to the self-normal quadric. In future it will be better to speak of taking the supplement with respect to an assumed self-normal quadric, rather than with respect to a particular set of normal elements. 111. Different Kinds of Supplements. (1) It may be desirable to take supplements with respect to various quadrics. The operation of taking the supplement with respect to one quadric is different from the operation of taking it with respect to another. If one operation be denoted by the symbol |, let another be denoted by the symbol I. Then \P and IP denote different extensive magnitudes. But the operator I possesses all the proper- ties which have been proved to belong to the operator |. Also if the supplement is taken with respect to a third quadric, the operator might be denoted by Ix and so on. (2) Confining ourselves to two operations of taking the supplement, denoted by | and I, we see that the two self-normal quadrics are denoted by (x\x) = 0, and (x Ix) = 0. But [cf. 83 (6)] in general two quadrics possess one and only one system of v distinct self-conjugate points. Let ely e2... ev be these points and let elt e2... ev be their normal intensities with respect to the operation |, and e/, e./... e/ those with respect to the operation I. Then eA ... ,= (_1_ } = eiV ... ej = A. Hence ( |,)= A (g + g+ ... + g) , and (a,.Ix) = Also ki = -^e2e3...ev, and Ie1 = -y2e2e3... ev. 1 1 112. Normal Points and Straight Lines. (1) The following propo- sitions can easily be seen to be true for mutually normal points with respect to any quadric. On any straight line one point and only one point can be found normal to a given point, unless every point on the line is normal to the given point. If a be the given point and be the given line, this point is be \a = (b \a)c (c \a)b, unless be \a 0 = (a \b) = (a \c).
Ill 113] NORMAL POINTS AND STRAIGHT LINES. 203 (2) There are two exceptional self-normal points on every straight line (viz. the points in which the line cuts the self-normal quadric), but in general these self-normal points are normal to no other points on the line. If however these two self-normal points coincide, so that the line is tan- gential, then this double point is normal to every other point on the line. (3) It follows from the harmonic properties of poles and polars that the pairs of normal points on a line form a system of points in involution, with the self-normal points as foci. (4) This harmonic theorem can be proved thus: let aYi a2 be the two self-normal points of any line; then (ax \a1) = 0 = (a2 |a2). Let \Oi 4- /LtOa and \rax + fia2 be any pair of normal points. Then (Xa! + /jia2) {(X'a,! + fi'a2) = 0. Hence (\p'+ \'p) (a* |a2) = 0. Hence \jfi = \'//j,'. 113. Mutually normal regions. (1) Two regions Pp and Pff, where p and a denote the orders of Pp and Po respectively, are called mutually normal, or normal to each other, if every pair of points pp and pa respectively in Pp and Pa are mutually normal. (2) Let Pp be denned by the points p(p\ pf\ ... p{pp\ and Pa by the points Pa\p(a\ '"P^- Then any point on Pa must lie on the intersection of the supplementary planes of p(p\ pf\ ... ppp\ Similarly any point on Pp must lie on the intersection of the supplementary planes of pQ\ p^\ ... p^. Thus the condition that Pa and Pp should be mutually normal is that Pa should be contained in | Pp, or that Pp should be contained in \PO. Either condition is sufficient to secure the satisfaction of the other. (3) If Pp and Pa be mutually normal, then a ^ v p, that is v ^ p + cr. (4) If v = p + cr, then Pa = |Pp. Hence the supplementary regions are mutually normal. The supplementary region of Pp will be called the complete normal region of Pp, or (where there is no risk of mistake) the normal region of Pp. Thus the supplementary plane of a point is its normal region. (5) In any subregion Pp (of the pth order) p mutually normal points can be found, of which any assumed point in Pp (which is not self-normal) is one. For [cf. 78 (9)] take xY to be any point in Pp, then \xY intersects Pp in a region of the (p l)th order. Take x2 to be any point (not self- normal) in this region. Then | oc2 intersects this region in a subregion of the (p 2)th order; take x3 (not self-normal) in this subregion of the (p 2)th order, and so on.
204 SUPPLEMENTS. [CHAP. III. If however Pp lie in the tangent plane to the self-normal quadric at one of the self-normal points lying in Pp, then this self-normal point must be one of any set of p mutually normal points in Pp. For the supplementary plane of the self-normal point by hypothesis contains Pp, hence the supplementary plane of any point in Pp contains the self-normal point. Thus proceeding as above in the choice of xly w2, etc., the last point, xpy chosen must be the self-normal point. (6) To find the subregion (if any) of the highest order normal to Pp which is necessarily contained in Po; where p and a are respectively the orders of Pp and Pa. Any region normal to Pp is contained in |PP. Now |PP and Pa do not necessarily intersect unless (v p) + a v, that is, unless a p. Assume a p. Then [cf. 65 (5)] |PP necessarily overlaps Pa in a subregion of the {(v p) -her v}th order, that is, of the (cr p)ih order. Every point in this subregion is necessarily normal to Pp; and hence this subregion of the (cr p)th order, contained in Pa, is normal to Pp. If the intersection of Pa and |PP is not of a higher order than (a p)} the regional element Pa\Pp defines it; thus if Pa \PP be not zero, it is the subregion of Pa normal to Pp. (7) By subsection (5) p mutually normal points can be found in Pp and (cr p) mutually normal points can be found in the intersection of Pa and | Pp (cr p). Also by the previous subsection each point of the one set is normal to each point of the other set. Thus the a points form a mutually normal set. Hence it is easy to see that, given two subregions Pff and Pp ( r p), a mutually normal points can be found in them, and of these p (or any less number) can be chosen in Pp and the remainder in Pa; also that any one point in Pp can be chosen arbitrarily to be one of these points or (if p points are to be taken in Pp) any one point in the inter- section of Pa and | Pp. 114. Self-normal Elements. (1) Every element in a subregion defined by p independent self-normal elements mutually normal to each other is itself self-normal. For if xl5 a2, ... ap be such elements, (fh K) = 0 = (a2 \a2) = etc. = (ax \a2) = etc. Hence {(AaC^ + \2a2 + ... + Xpap) KA^ -f \2a2 4- ... 4- Xpap)} = 0. Also any two elements of such a subregion are normal to each other. For {(\cti + A2a2 + ... + \pap) \(fii( (2) Accordingly such a subregion is itself a complete generating region [cf. 79] of the quadric, (x \x) = 0; or is contained in one.
114] SELF-NORMAL ELEMENTS. 205 But from 79 the generating regions of this quadric are, in general, of v/2 1 or (v l)/2 1 dimensions according as v is even or odd. Hence sets of v/2 or (v l)/2 (as v is even or odd) self-normal and mutually normal elements can be found. (3) Also by 80 a set of conjugate co-ordinates jlt j2... , klf k2... can be found all self-normal and all mutually normal except in pairs, i.e. (j11^) is not zero, nor (j2 \k2) and so on. But (jj Ij^ = (jj \j2) = etc. = (jj \k2) = etc. = 0. If v be even, v such co-ordinates can be found which define the complete manifold; but if v be odd, v 1 can be found, and one co-ordinate element e remains over, which can be assumed to be normal to the v 1 other elements, but not self-normal. (4) Firstly let v be even. Let ely e2,... ev be a set of normal elements, elt 2... ev being their normal intensities according to the notation of 110. Then by 80, we may assume ji = \ (^i + ie2e2), kY = \Y (e^ - ie2e2), j2 = A,2 (e3e3 H- i 4e4)f k2 = X2 (e^ i 4e4), v), Av = X* {ev^.1ev_l ievev). 2 2 2 2 Hence jJcY = 2i\12 le2e1e2, with - 1 other similar equations. V Thus (jiki.hh - jX) = (- 2if \{2X2- ... X,2. 2 2 2 Again \jY = ^ ( x \el + ie21e2) = -1- (e2e2e^... ev- ie^e^ ... ev\ 6162 . ""I But jjkijzkz.,. jje^ = ( 2i)a e3e4 ... e^V ... X,,2^ ... ev; 2 2 2 hence jjohjih ... A\, = (- 2i)2 X^2^2 ... V (e^ea... ev + i*62^3... ev) 2 2 6l62 2 (5) Now let X1? X2, ... X,, be so chosen that 2 Xj = X2 = ... = \v = . a v2 Then j2 = -^ (62g2 - ie^), ^ = -^ (^2 + left), 2 v
206 SUPPLEMENTS. [CHAP. III. And {jAfrh jjc,) = ( 2 2 V V Rencej1j2k2j,k,...j ,kv = i2\j1. Similarly j2j1k1j.,k3...jlLklL=i21 j2, and so on. 2 2 2 2 When Xl9 A2, ... Xv have been chosen as above, the conjugate self-normal 2 elements will be said to be in their standard normal form. When the self-normal elements are in this form 2 2 (6) Secondly, let v be odd, and let e, e1} e2 ... ev_1 be the set of normal elements with normal intensities e, e^ ... _!. Let the standard normal forms of the conjugate self-normal elements be ji = -J2 (e ~ ie^ *i = 7/2 h = -ft (64^4 - i 36s), k2 = - ^ 2 VZ "2 V^ Then J^ = ie^ae^a, with similar equations. v-l 1/-1 Hence (ej^j^ ... j^-i^^-i) = * 2 ^eg... v^ {eeY... ev^) i 2 e"1. 2 2 Also ej-ijjcz ...jV-i^-i = i2 c^lji, with similar equations for the other T ~¥ elements. And (i;1|AI) = (fc1|j1) = -i. 115. Self-normal Planes. (1) Let a be a self-normal element; now the region \a contains all the self-normal elements which are normal to a. Hence \a contains all the generating regions of the quadric which contain a. Therefore | a is the tangent plane to the quadric at a. (2) The plane-equation of the quadric is, (X \X) = 0, where X is any planar element. For this equation is the condition that the region X contains its supplementary element \X. A tangent plane X, for which (X|X) = 0, will be called a self-normal plane. 116. Complete Region of Three Dimensions. (1) The application of these formulae to a manifold of three dimensions is important. Consider a
115 117] COMPLETE REGION OF THREE DIMENSIONS. 207 skew quadrilateral j^hh formed by generators of the self-normal quadric; so that jxj2 and k k2 are two generators of one system, and jxk2 and jJcY are two generators of the other system. A self-conjugate tetrahedron exe fi$A can be found such that if x be the point 2fe, the self-normal quadric is 2M2 + hi + \* + f?\e? = o, the normal intensities of el9 e2) e3, e4 are then ex, 2, e , e4, and (2) Assume j, = (e2e2 - ie^), kx = -^ (e2e2 + ie^) j2 = ,2 (6^4 Hence ( jYkY j2 k2) = %2 = l. Thus | j2 j2 =jj2k2. j2jA = - {jAh^hh =hh \ and similarly | ^2 = ^^2. Also lji 2 = -ji 2, \J2k1=-j2k1. (3) Hence for a generator (Cr) of one system of the self-normal quadric \G=zG, and for a generator Gf of the other system \G' = G'. Let the system of generators to which G belongs be called the positive system, and that to which G' belongs be called the negative system. 117. Inner Multiplication. (1) The product of one extensive mag- nitude (such as Pp) into the supplement of another extensive magnitude (such as | Pa) is of frequent occurrence; and the rules for its transformation deserve study. These rules are of course merely a special application of the general rules of progressive and regressive multiplication, which have been explained above. (2) This product Pp\Pa may also be regarded from another point of view. Since Pp | (P* + PJ) = Pp | Pff + Pp | Pa\ we may conceive [cf. 19] the symbol | not as an operation on Pa but as the mark of a special sort of multiplication between Pp and Pff. Let this species of multiplication be called ' Inner Multiplication,' and let the product P9\PV be termed the inner product of Pp and Pa. In distinction to Inner Multiplication Progressive and Regressive Multiplication are called Outer Multiplication. (3) It is obvious that inner products and inner multiplication must be understood to refer to a definitely assumed self-normal quadric ; and further that, corresponding to different self-normal quadrics, there can be different sorts of inner multiplication. But general formulae for the transformation of such products can be laid down.
208 SUPPLEMENTS. [CHAP. III. 118. Elementary Transformations. (1) Let Pp and Pa be extensive magnitudes, simple or complex, of the pth and rth orders respectively. Then v r is the order of \PV. The product Pp \PO is progressive if p -b (v a) vy that is, if p r; and is regressive, if p a. (2) Ifp r; Pp|Pa = (- and hence | (Pp | Pa) = (- But by 99 (7), ||^=(- Therefore finally, | (Pp | Pa) = ( (8) If p a- then|(Pp|P,) = |Pp.||P, = (-l)'C-')(|Pp.P,) (4) If p = r ; then (Pp | P,,) is merely numerical: write Pp' instead of Pa. Then (Pp | P/) = |(PP | P/) = (P/1 Pp). 119. Rule of the Middle Factor. (I) The extended rule of the middle factor can be applied to transform Pp\Qa) where Pp and Qa are simple magnitudes of the pth and crth orders respectively. In the first place assume that p a. Let the multiplicative combinations of the crth order formed out of the factors of the first order of Pp be P , P , etc., and let PP = P{?P(l]- r = P?P?-. = etc. Then from the extended rule of the middle factor, we deduce Pp| = (P?| W. + (i ?| )PpB-,, + ete................(1). (2) Secondly, assume that p cr. Then (Pp | Qa) = (- 1) p+ (- | (Q. \ 1\). Now let Qlp, Qp\ etc. be the multiplicative combinations of the pth order formed out of the factors of the first order of QO} and let Then by equation (1) of the first case + (Q? I S-, + etc. = (P, I Off) Qi'-P + (P, IQ?) -, + etc. Hence (- l)0 +")(-p (Pp | Q.) = (Pp | Q?) \ Q?_p + (Pp | Qf) | Q _p + etc... .(2). The formulae of equations (1) and (2) will be called the rule of the middle factor for inner multiplication. 120. Important Formula. (1) The rule of the middle factor does not apply when both factors are of the same order. But the transformation in this case is given by 105. For if each factor be of the pth order, then ?... aD (a2 |6j), ( z2 (ap \bx\ (ap (o.|W ! .........(3). (a. I ftp)
118 122] IMPORTANT FORMULA. 209 Important special cases of this formula are ;... ao\ axa2... ap) = ), O2|a2), ...... (a2\ap) (a^di), (ap\a2),...... OP | = (a, I bj) (a2 | 62) - (^ \ b2) (a2 \ b^); (a.a., I aYa2) = ( ! | ax) (a21 a2) - (ax | a2)2. (2) Also if the complete region be of v - 1 dimensions, the products {axa2... a,.) and (bj)2 ... bv), although merely numerical, may each be conceived as progressive products. The proof of 105 still holds in this case, and therefore {axa2 ...av)(bj)2 ...bv) = {axa2 ...av\bj 2... b,) (Oil61), (Oi| ), ...... ( i|^) (a2 \bY\ (a2 \b2), ...... (a2\bv) (a, I^X (av\b2), ...... (av\bv) This is the ordinary rule for the multiplication of two determinants. 121. Inner Multiplication of Normal Regions. If A, B, 0 be three mutually normal regions, (so that the multiplication ABO must be pure progressive), then (ABC\ABC) = (A\A)(B\B)(C\C) = (AB\AB)(C\C). This theorem can easily be proved independently; but we will deduce it at once from the formula for {axa2... ap \aYa2... ap) of 120. For let A = axa2... ap, B = bj)2 ...bv,C= cYc2...c,; then each of the groups (a1} a2... ap), (6^ 62... ba), (c1} c2... cT) may be conceived [cf. 113 (5)] to consist of mutually normal elements. But since A, B, G are mutually normal regions, it follows that the whole set of p + a + t elements are mutually normal. Hence {ABC I ABC) = (a^2.. ...cT 9) (c, \ cT) . (cT | cT). and = (ax I ax) (a2 \a2)... (h {b,)... Also (A\A) = (a1\a1)(a2 \a2) ... (ap |ap), and are equal to similar expressions. Hence the theorem follows. 122. General Formula for Inner Multiplication. (1) Equations (1) and (2) of 119 can be extended so as to prove two more general formulae which include both them and equation (3) of 120. Consider the product Pp+a\QpQry where Pp+a, Qp, QT represent simple magnitudes of the (p + c)th, pth, and rth orders respectively. In the first place assume that a t. w. 14
210 SUPPLEMENTS. [CHAP. III. Then PP+.\Ql,Qr = Pe But since a t, the product is a pure regressive product and is therefore associative. Hence PP+o\Q(,Qr = PP+ ,\QP.\Qr- Now let Pp\ P , etc. be the multiplicative combinations of the pih order formed out of the factors of the first order of Pp+ ,, and let P%\ P \ etc. be the multiplicative combinations of the crth order formed out of the factors of Pp+tTy so that Pp+a = P P2 = P P{? = etc. Hence by equation (1), Pp+a | Qp = 2 (P{pK) \ Qp) P}\ A Therefore finally Pp+a\. QPQT = 2 (P A | Qp) P A) \QT .....................(4). (2) Secondly, let t t. Let QpQt = Qp+t, and let ($\ Qp2\ etc. be the multiplicative combinations of the pth order formed out of the factors of the first order of Qp+T, and let Q , Q be those of the -rth order, so that Qp+r= W = Wf = etC. Also let Pp+* = PpPa, where Pp is of the pth order, and Pa of the crth order. Then Pp+ff |Qf+r = (- 1)( +* -p- \(Qp+T \Pp+a). But by equation (4) Qp+T \PpPa = t ( * |PP) Q \Pa. Hence \(Qp+r \P9Pa) = 2^ |PP) \(QlK) |P.) = ( - i)(-+r (r-.) S(P/) \QM)P, |qW Therefore finally, PPPO | Qp+r = (- 1)" *+t) 2 (Pp | Q A)) Pff | Q x)......(5). A (3) Equations (4) and (5) are more general than equations (1), (2) and (3) but the readiness with which the equations first found can be recovered from the extended rule of the middle factor makes them to be of the greater utility. The theory of Inner Multiplication and the above formulae are given in Grassmann's Ausdehnungslehre von 1862. 123. Quadrics. (1) The theory of quadrics can be investigated by the aid of this notation. Let the quadric be chosen as the self-normal quadric for the operation |. Then the equation of the quadric is (x \%) 0. Let the reference points el9 e2... ev be any v independent elements, not necessarily mutually normal. Then if x = 2 e, the equation of the quadric according to the notation of Book ill., Chapter ill. is written (a\xf = an * + ... + 2a1 + ... = 0. But ( k) = (eik)fia+..- + 2( i|%)f1f,+ .... Hence we may write, (eY 1^) = aUt (e2 \e2) a^, etc., (eY \e2) = a12, etc.
123] QUADRICS. 211 (2) Since by 120 (e,e2... evf = (e ...e¥\e,e2 ...e¥) 1 |*i), (e1\e i)i..,{e1 \ev) 2 |O, (e2\e2\ ...(e2\ev) it follows that (e^... e,,)2 is the discriminant of the quadratic expression (x \x). Since (e^ ... ev) cannot vanish (ely e2y ... ev being independent), it follows that the quadric cannot be conical. (3) The equation of the polar plane of any point x becomes (x \y) = 0. The plane-equation of the quadric is (Xv_1 \XV_^) = 0; where J^ is a planar element. The equation of the polar point of any plane Xv_lf becomes (4) Let bx be any line drawn through a given point b; and let this line intersect the quadric in the point \b + fix. Then \2(b \b)+2\fi(b \x)+tf{x \x) = 0.....................(1). This quadratic for Xjfi in general gives two points on the quadric. (5) These points coincide if (b \b)(x \x) (b \x)2 = 0. This is the equation of the tangent quadric cone with vertex b. But (b | b) (x | x) - (b | xf = (bx | bx). Hence this cone can be written (bx | bx) = 0. (6) The identity (b \b)(x \x) - (b \x)2 = (bx \bx), can be written (ajbf (ajxf - {(a$b%x)Y = 2 ( , - /3 ) (J3 Also (epea {e^) = (eP \ek) (ea \e^) - (ep le^) (e9 \eK) = (7) The roots of the quadratic equation (1) are X, - (b \x) + V{( \xf - (x \x) (b \b)} (8) If ax and a2 are the points XX6 + fax, and X26 + fi^, then the an- harmonic ratio (aYa2i xb) is - (b \x) ^/{(b]xy- (c^x)^b\b)\ = (6|s) + V{-(telfaQ) - (6 | ) + V{(6 k)2 - ( la?) (6 16)} Q k) - V{- (^ l *)}' (9) Firstly let (b \xf (x \x) (b \b). (b \x)2 Let 6 be such that cos20 = , , ' 7 .,. ; and let p = (aYa2i xb). (x \x)(b \o) r cos 0 + i sin 0 Then _ cos 0 i sin _ 2ie 14 2
212 SUPPLEMENTS. [CHAP. III. Also sin20 = V - , ' . O 1*0 ( 16) Therefore we deduce the group of equations = cos (10) Secondly, let (b \xf (x \x)(b \b). Put ^ ( r \ r\ (h \h\ ' yu/ \ii j yu \uJ Then p = CQS, n 55.._. - #* r cosh 6 si Also sinh2 6 = , ,\/y ,,\ . (x \x)(b \b) Hence we deduce the group of equations -(6a? | bx) , . (6 1^?) . , . / -(6a? | bx) ^ gp = cosh"1 -jj- \ ' / = sinh"1 X / V/ , wl .^ ...(3). (11) If (bx \bx) be positive for every pair of elements 6 and x, it follows from 82 (4) that the quadric, (x\x) = 0, is imaginary. If the quadric be a closed real quadric and 6 and x both lie within it, or if both lie without it and the line bx cut the quadric in real points, then it follows from the same article that (bx \bx) is necessarily negative. 124. Plane-equation of a Quadric. (1) Taking the supplement of the equation, (bx \ bx) = 0, and writing B instead of 6 and X instead of | x, we find the equation (BX | BX) = 0, which can also be written (B\B)(X\X)-(B\Xy = 0. This equation [cf. 84 (4)] is the plane equation of the degenerate quadric enveloped by sub-planes lying in the plane B and touching the quadric. (2) Again, by a proof similar to that in 123, let B and X be any two planes, and let the two planes through their intersection BX which touch the quadric be Aj and A2. Also let p be the anharmonic ratio of the range X, AXA*). Then if (B \ Xf (B \B) (X | X), BX) And if (B | X)" (B | B) (X \ X\ 61 = 9 logP = cosh ~iuyTyYTWYl^Vl = Smh A/ |/v I y\/Di mi (3) Again, let Lp be any subregion of p 1 dimensions which touches the quadric. This condition requires that Lp should lie in the tangent plane
124] PLANE-EQUATION OF A QUADRIC. 213 to the quadric at some point b, and should contain b. We can prove that the condition to be satisfied by Lp is, (Lp | Lp) = 0. For let l1} l2, ... lp, be assumed to be p mutually normal points on Lp, which is possible according to 113 (5). Then by 120 (1) (Lp\Lp) = (l1\ k)(l2\l2) ... (lp\lp). Hence if (Lp\ Lp) = 0, then one at least of the points llt l2, ... lp must be self-normal. Assume that lp is the self-normal point b. Then the remaining points l1} l2i ... lp-x all lie on the plane | b, which is the tangent plane of b. Thus Plucker's conception of the line equation of a quadric in three dimen- sions can be generalized for any subregion in any number of dimensions*. (4) Consider the four subregions Bp, Xp, Apy Ap, of p 1 dimensions which lie in the same subregion of p dimensions. Then considering this containing subregion as a complete region we see that Bp, Xp, APi Ap have the properties of planes in this region. Let Ap and Ap both contain the subregion of p 2 dimensions in which Bp and Xp intersect. So that Ap = \BP + pXpy and Ap = \'BP + i*!Xp. Then the four subregions Bp, Xp, Ap, Apf form a range with a definite anharmonic ratio \fi!j\ffju\ let this ratio be called p. Let Ap and Ap touch the self- normal quadric. Then A,///, and X'/p are the roots of the quadratic Hence, as before, if (Bp \ Xp)2 (Bp \ Bp) (Xp | Xp\ (Tt I K \ 1 6 = -. log p = cos"1 (Bp | X9f (Bf | B,) (X, \ X,), tf -1 log , = cosh- then And if then It is to be noticed that the formulae for sin 6 and sinh 6 do not hold unless p be unity or v 1. * As far as I am aware this generalized form of Plucker's line-equation has not been given before.
CHAPTER IV. Descriptive Geometry. 125. Application to Descriptive Geometry. An extensive manifold of v 1 dimensions is a positional manifold of v 1 dimensions with other properties superadded. These further properties have in general no meaning for a positional manifold merely as such. But yet it is often possible conveniently to prove properties of all positional manifolds by reasoning which introduces the special extensive properties of extensive manifolds. This is due to the fact that the calculus of extension and some of the properties of extensive manifolds are capable of a partial interpreta- tion which construes them merely as directions to form * constructions * in a positional manifold. Ideally a construction is merely an act of fixing attention upon a certain aggregate of elements so as to mark them out in the mind apart from all others; physically, it represents some operation which makes the constructed objects evident to the senses. Now an extensive magnitude of any order, say the /oth, may be interpreted as simply representing the fact of the construction of the subregion of p 1 dimensions which it represents. This interpretation leaves unnoticed that congruent products may differ by a numerical factor, and that, therefore, extensive magnitudes must be con- ceived as capable of various intensities. Accordingly, in all applications of the Calculus to Positional Manifolds by the use of this interpretation it will be found that the congruence of products is the sole material question, and that their intensities can be left unnoticed; except when the products are numerical and are the coefficients of elements of the first order which have intensities in positional manifolds. The sign of congruence, viz. = [cf. 64 (2)], rather than that of equality is adapted to this type of reasoning. Also supplements never explicitly appear, since they answer to no mental process connected with this type of reasoning. 126. Explanation of Procedure. (1) In the present chapter and in the succeeding one some applications of the calculus to positional manifolds are given. Except in 130 on Projection, the manifolds are of two dimensions,
125 128] APPLICATION TO DESCRIPTIVE GEOMETRY. 215 and the investigations form an example of the application of the Calculus to Descriptive Geometry of Two Dimensions. Other applications of this type have already been given in 106, 107. (2) In two dimensional complete regions the only products are of two points which produce a linear element, of two linear elements which produce a point, and of a point and a linear element which produce a numerical quantity. If a product yields an extensive magnitude, the act of using such a product is equivalent to the claim to be able to construct that subregion which the magnitude represents. Thus the product ab represents the indefi- nitely produced line joining ab, and the use of the product is the equivalent to drawing the line. Similarly if L and L' are two linear elements in a plane, the use of LL' is equivalent to the claim to be able to identify the point of intersection of the two lines L and L', which by hypothesis have been constructed. Thus the representation of a point by a product of certain assumed points is the construction of that point by drawing straight lines joining the assumed points and is the point of intersection of lines thus drawn. 127. Illustration of Method. The method of reasoning in the application of this algebra to Descriptive Geometry is exemplified by the proof of the following theorem*. If abc and def be two coplanar triangles, and if s be a point such that sd, se, sf cut the sides be, ca, ab respectively in three collinear points, then sa, sb, sc cut the sides ef,fd, de respectively in three collinear points. For by hypothesis (sd. 6c) (se . ca) (sf. ab) = 0. Hence by the extended rule of the middle factor {(sdc) b (sdb) c] {(sea) c (sec) a] {(sfb) a (sfa) b] = 0. Multiplying out and dividing by the numerical factor (abc), (sdc) (sea) (sfb) - (sdb) (sec) (sfa) = 0. The symmetry of this condition as between the triangles abc and def proves the proposition. 128. von Staudt's Construction. (1) Let a, c, b represent any three points in a two-dimensional region, which are not collinear. In ac assume any point d arbitrarily, and in cb assume any point e. Since the intensities of a, c and b are quite arbitrary, we may assume that d = a + hc, e = b + ec, where 8 and e are any assumed numerical magnitudes. Then it is to be proved that any point x on ac (i.e. of the form a 4- fc) can be exhibited as a product of the assumed points, or in other words can be constructed. This construction to be given is due to von Staudtf. * Due, I believe, to H. M. Taylor. t Geometrie der Lagey 1847.
216 DESCRIPTIVE GEOMETRY. [CHAP. IV. (2) Firstly, consider the following products, or in other words make the constructions symbolized by them : q = ae. db, qY = qc. de, pY = qj). ac. Thus px = ae .db.c .de.b . ac. Then px is the point a + 28c. For q = (ab -f eac) (ab + 8cb) Sab . cb + eac . ab + eSac. cb = (acb) b + e (acb) a+eB (acb) c = - (abc) [ea +8b + eSc] = ea + Sb 4- eSc; where the numerical factor (abc) has been dropped for brevity. This will be done in future without remark. qx = (eac + Sbc) . de = (eac + $bc) (ab + eac + Sc6) = eac . a + Seac. cb + Sbc. ab + Se6c. ac = e (acb) a + Be (acb) c + 8 (acb) b + Be (acb) c = ea + 8b + 28ec. px = qj). ac= e{a + 2Bc] b.ac = e (abc) {a + 28c} =a + 28c. Hence pY is the point a + 2Sc. (3) Again, substitute d for a and p1 for d in the above product. The new lines in the figure are represented by dotted lines. Then since px=d + 8c, we obtain the point p2= de.pxb . c ,pxe.b.dc = d-\- 28c = a + 38c. Similarly by substituting p for d and p2 for pY in this construction, we find p3 = a + 4Sc, and so on successively. Thus finally if v be any positive integer, we find pv = a+(v + l) 8c.
128] von staudt's construction. 217 (4) Secondly, consider the point e' qp . bc = b ec (cf. fig. 2). Make the following construction, r2 = ae'.de p, = rj). ac = ae .de.b. ac. Now rx = (ab eac) (ab + eac + Scb) = eab . ac + Bab. cb eac . ah Seac . cb = 2eac. ab + Sab . cb Seac .cb = 2e (acb) a + S (acb) b Se (acb) c; and Pi = (acb) {2eab + Secb}. ac = a 4- ^Sc. (5) Similarly by substituting pL instead of d in this construction we obtain 22 Fig. 2. and by continually proceeding in this manner we finally obtain if v be any positive integer (6) Then if fi be any other positive integer, the construction of the first figure can be applied /j, times starting with p instead of with d. Thus the point ^ can be constructed.
218 DESCRIPTIVE GEOMETRY. [CHAP. IV. (7) Thirdly, make the following construction (cf. fig. 3) : q = ab . de', df = q'e . ac = ab. de . e. ac. I'IG. 3. Now q = ab {ab eac 4- Bcb} = eab. ac + Sab .cb = (abc) [ea + 36}, d' = (eae + ^6e) ac = (ea-6 4- e2ac 4- Sebc) ac = eab . ac 4- Se c. ac = e (a6c) {a 8c} = a Sc. (8) In this construction if we substitute for d any constructed point of the form we obtain , 2* Thus all points of the form f a | Sc J can now be constructed. (9) Fourthly, let p, p' and p" be three constructed points, and let p = a 4- -crSc, p' = a 4- r'Sc, p/7 = a 4- ^"Sc. Then p' = p 4- (^r - r) Sc.
129] von staudt's construction. 219 Now in the first construction substitute p for a and p' for d. Then we obtain Pi = p 4- 2 (wf w) Sc = a + (2s/ tv) 8c. Similarly by substituting _p/ for a, and p" for d, we obtain p2' = a + (2*x" - 2tsr' + isr) 8c; and so on by successive substitutions. (10) But any positive number , rational or irrational, can be expressed to any approximation desired in the scale of 2, as the radix of notation, in the form where /90 is the integer next below f and fjlf fi2, etc. are either unity or zero. If the series is finite any point of the form a %8c can be constructed in a finite number of constructions; if the series is infinite it can be con- structed in an infinite number of constructions; and (since the series is convergent) this means that in a finite (but sufficiently large) number of constructions we can construct a point a %'Sc, where f ~ f' is less than any assigned finite number however small. Thus any point a+fcoo the line ac can be constructed, and similarly any point on the line be can be constructed; and it is sufficiently easy to see that any point a + l;b + rjc can be constructed. This type of construction can easily be extended to a protective manifold of any dimensions. 129. Grassmann's Constructions. (1) Grassmann's constructions* in a complete region of two dimensions have for their ultimate object to construct the point tt + ^(^1, ^2)(^i+ 2), where t|t( i, 2) is any rational integral homogeneous function of jjlt 2, and a, alt a2 are any three given not collinear points, provided also that the point a + ^aj + 2a2 and also certain points of the form a + a^ + oc^ are given, where the a's are known coefficients. In order to accomplish this end the constructions are given for the following series of points, a + ( i + a2), a-f- 2Oi + a2), a + K + a2), a + %\ (ai + (h) {where v is any positive integer}, a 4- / ( i + a2)y a + 2^i ^M2 (ai f a^ {where y1 + y2 is not zero}. Yi + 72 Then finally a construction is deduced for when 7 + y + ... is not zero, and fi, v, /*', z/, ... are positive integers, and = fA +V = .... * Cf. Ausdehnungslehre von 1862, 325 329.
220 DESCRIPTIVE GEOMETRY. [CHAP. IV. (2) Let a, aly a2 denote any three elements forming a reference triangle in the two dimensional region ; let x = a + i i + 2 2; and let d = a + aY + a2- Fm. 4. Firstly, make the following constructions (fig. 4), yY = xa2. ad, y2 = xaY. ad. Then yY = (aa2 + ^a^) (aax f aa2) = (aa^) {a + ^ {ax -f a2)} = a + f i (aj + a2). Note that in future numerical factors which do not involve f^ or 2 will be dropped without remark. Similarly y2 = a + |2 (aj + a2). (3) Secondly, make the following constructions (fig. 4), 0i = 2/2^1 ^d, x = exa . y1a2y y = xfaY. ad. Then e1 = (aaj + fj^hOj) ( + 2 i) () a (aa^) ax ^2 (aa^) a2 And And y = (a Xx 4- ljif; i) (aax + aa2) = (aa^o) {a + f^ (ax + a2)} = a + f^2 (a2 + a2) = y2ax. a^d. a. yxa2. ax. ad. (4) Now substitute y2 for y2 in the above construction (fig. 5). We obtain a*d. a . y^a* ,a1.ad = a + ^12 (ax + a2). aa2 a
129] GRASSMANN S CONSTRUCTIONS. 221 Similarly we construct y2 = . a. 22 (a-j 4- a2). Fig. 5. It is also obvious that in the constructions aY and a2 can be interchanged. Thus y = y2a2. M. a. yYaY. a2. ad, and 2// = yYa2. c^d . a. y1a1. a2 ad9 and y2' = 2/2^2 ^i^ a 3/2^1 ^2 d. Also in the construction in subsection (3) for y from y1 and y2,2/x and y2 can be interchanged, thus giving two fresh forms of the construction, namely, y = yxaY. a2d. a. 2/2 z2. aa. ad, and 3/ = yYa2. aYd. a . y2ax. a2. ad. (5) Let the symbol ( ") stand for the point a + ^ ( i + a2), and similarly let (|/) stand for the point a + / (ax + a2), and let (f1^/) stand for the point a + \"U Oi + as)- Now substitute the point (^v) for y2 in the first construction given for y, then we obtain (^"+1) = (f/) aj. a2rf . a. y1a2. a2. ad. (6) Again, let ^iJj denote that the point p has been multiplied succes- sively by the factors alt a2d, a, yxa2, aly ad, so that pRx stands for the point pax. a2d. a . yYa2. aY. ad. In order to avoid misconception it may be men- tioned that i?! is not the product aY. a2d . a . yYa2 .aY.ad) for in pRx the first factor aY is multiplied into the point p. Also pRY is itself a point : let i be denoted by pR 2, and so on.
222 DESCRIPTIVE GEOMETRY. [CHAP. IV. By applying this notation to the construction for ( 1"1"1) in terms of |y, we see that when v is a positive integer, ( /) = yiRip~\ Since yxa2 = xa2} R1 may be conceived to stand for the set of factors alf a^d, a, xa2) alf ad successively multiplied on to a point. Also yx = xa2. ad. Hence (|V) = xa2. ad . Rf'1. Thus the point a + %i ( i + 2) is exhibited as a product in which x occurs v times. (7) Similarly interchanging the suffixes 1 and 2, let R2 stand for the set of factors a2, aYd, a, xalf a2, ad successively multiplied on to a point. Also y2 = xa1. ad. Hence (( /) = a^ . ad . i^"1. Thus the point a + f/ (a2 + a2) is exhibited as a product in which a? occurs fjb times. Fig. 6. (8) Again, in any of the constructions for y (say the first) substitute and ( /*) for ^ and y2, say, for example, ( /) for y^ and (^a*1) for 2/2-
129] grassmann's constructions. 223 Then (frfr) = (ff)ai.a2d-a.( ")a^a,. ad. Hence the point a + fi1^ ( i + (h) is represented as a product in which x occurs (fi + v) times. (9) Finally, let p and p' be any two points a + nr (ar + a2) and a + r' (ax + a2) on the line ad; and let c denote any point 71a1 ry2a^ on the line a^a2. Make the constructions (fig. 6) q = pa1. pa2f r = qc ,ad = pal. pa2 .cad. Then q = j9ax. ^'aa = (aa1a2) {a + 'ST/a1 + ^a2)y r = [ /1aa1 - y2aa2 + (7^ 4- y2vr') ^J {aa! + aa2] a + (7^ + y2vr') (ax + a2)] 71 Similarly, let pr/ denote a third point a + 'gt// (aa + a2), and let c' denote the point (7! + 72) aY y3a2. Make the construction r = to, . p"a2. c. ad = a + ^ + ^ + ^ (a, 71 + 72 + 73 And so on for any number of points p, p\ p"t etc. Thus if any number of points of the form {% ?%/) have been constructed, then the point can be exhibited in the form of a product. (10) Hence the numerical product It will be observed that x occurs in this product {(/x. + v)-\-(/j ' + v) + ...} times, and that therefore if x be written in the form rja + rj^ + %a2, then the product is a homogeneous function of 77, %, r}2 of degree {(/x -|- v)+(jl +v) + ...}. But let p be the greatest of the numbers (fi + v)y(/jl'-f v), etc., then it is easily verified that the homogeneous function represented by the product is any required homogeneous function of degree p multiplied by 77 to the power: [{^ _[_ v} -|_ (^/ -|_ |/) + ... j p], and also by some constant numerical factor. If however we keep x in the form a + ^A + f2a2, then the most general rational integral algebraic function (not necessarily homogeneous) of x and 2 can be exhibited in the form of a product; or if / ( i, f2) be the function, it can be represented by a point which is constructed as a product of the point x and fixed points, partly arbitrarily chosen and partly chosen to suit the special function.
224 DESCRIPTIVE GEOMETRY. [CHAP. IV. 130. Projection. (1) Definition. Let the complete region be of v 1 dimensions, and let x, yy etc., be elements on any given plane A of this region. Let e be any given point not on this plane and let B be any other given plane. Then the lines ex, ey, etc., intersect the plane B in elements af9 y', etc.: the assemblage of elements x, y\ etc., on the plane B is called the projection on B from the vertex e of the assemblage of elements x, y, etc., on the plane A. Definition. Two assemblages of elements x, y, etc., on the plane A and x', y\ etc. on the plane A\ which is not necessarily distinct from A, are called mutually projective, if one assemblage can be derived from the other by a series of projections. If one figure can be deduced from another by a single projection, the two figures are obviously in perspective. (2) These constructions can be symbolized by products: thus the pro- jection of x on to the plane B from the vertex e is x = xe . B. Let the projected points always be assumed to be at intensities which are deduced from the intensities of the original points according to this formula. Since (\x + fiy) e. B = \xe. B -f fiye. B, it follows that any range of ele- ments on a line is projected into a homographic range. (3) Proposition I. Let any subregional element in the plane A be denoted by the product xly x2... xpy where p is less than v 2; also let Xi, x2' ... xp be the projections of the points xly x2, ... xp on to the plane B from the vertex e, so that for instance xi = x^e. B; then it will be proved that* xix2f ...xp' = (eBy-1 XjX2... xpe. B = xxx2... xpe.B. In other words, if Xp be any subregional element of the pth order, and Xp be the corresponding subregional element formed by the projected points, then X; = (eB)f -* Xpe.B= Xpe. B. Thus Xp will be called the projection of Xp, and the above equation forms the universal formula for the projection of elements of any order. (4) In order to prove this formula the following notation will be useful. Let xYx2... (xa) ... xpe denote that the elements atlt x2... xp, e, with the excep- tion of xa, are multiplied together in the order indicated. Then the extended rule of the middle factor gives the transformations p x^x2 ...xpe.B= 2 ( I)0""1 (xaB) XjX2... (xa) ... xpe + ( 1)p(eB)x^x2 ...xp. T=1 Also 'p+i = #p+i e. B= (xp+1 B)e (eB)xp+1. But x1x2... (xa)... xpe .e = 0, and xYx2... (xa) ...xpe. xp+1 = xYx2... (xa) ... XpX^e. * This formula has not, I think, been stated before.
130] PROJECTION. 225 Hence by multiplication and rearrangement of factors it follows that {#!#2... xpe. B} %'p+T, = ( iy (eB) (xp+1 B) x^2... xpe + V (- I)*"1 (eB) (xaB)x,x2 ... (xa) ... xp+1e- (- 1)p(eB)'x,x2... xp+1 r=l 2 (- I)""1 (x*B) x,x2 ...(xa)... xp+l e + (- 1)p+1 (eB)x,x2... xp+l = (eB) xYx2... xp+1e . #. Hence by successively applying this theorem we deduce x^x2... xp = (eB)^1 xYx2... xpe . B = xYx2... xpe. B. (5) It is obvious that the relation between a point x and its projection x is reciprocal; that is, if x be the projection of x on B from vertex e then # is the projection of xf on A from vertex e. For x'e.A = {(xB)e-(eB) x}e . A=-(eB) xe.A = (eB) (eA) x = x, since (xA) = 0, by hypothesis. Thus two figures are protective if they can both be projected into the same figure. (6) Proposition II. Any three collinear points are projective with any other three collinear points. This is the same as the proposition that any two homographic ranges are projective. Firstly, let the two lines L and L', on which the points respectively lie, be intersecting, so that the complete region is of two dimensions. Let a, 6, c and a, b\ c be the two sets of three points on L and U respectively. Take e and e any two points on aa'. Construct the points eb. erV and ec . e'c : call them the points 6", c". Construct the point ad . b"c" = a!'. Then we have evidently a = a"e . L, b = b"e. L, c = c"e. X, and a' = aV. L\ V = b"e . L\ c' = c'V . L\ Thus the collinear points a", b'\ c" can be projected both into a, 6, c and a, b\ c'. Hence a, 6, c and a, b\ c' are projective. (7) It may be noticed that if the regressive multiplications are defined for a complete region of three dimensions and the ranges abc and db'd be coplanar, then the above results must be written a = a"e. Ld, b = b"e . Ld, c = c"e. Ld, and a'=a"J.L'd, V = b"e'.L'd, c=d'ef.Ld\ where d is any point not in the plane of the straight lines L and L\ and e and e being both on aa' are in the plane of L and L'. w. 15
226 DESCRIPTIVE GEOMETRY. [CHAP. IV. And more generally, if the regressive multiplication be defined for a complete region of v 1 dimensions, let Dv__z be any extensive magnitude of the (v 3)rd order which does not intersect the two dimensional regions containing L and L', then LDV_Z and L'DV_3 can be taken as the planes of the two projections, so that a = a!'e . LDV_Z, etc., and a = a!'e'. L'DV_Z etc. (8) Secondly, let the lines containing a, b c and of a, b\ c' be not inter- secting. Take any point p on abc and pr on a' V. Construct the line pp\ and on it take any three points a", b'\ c". Then a, 6, c and a", b", c" are protective, also a!. bf, c and a", V\ c" are protective. Hence a, 6, c and a', ', c' are projective. (9) Proposition III. If any p points in a subregion of p 2 dimen- sions (with only one addition relation) are projective with any other p points in another subregion of p 2 dimensions (with only one addition relation), then any p +1 points (with only one addition relation) in a subregion of p 1 dimensions are projective with any other p + 1 points (with only one addition relation) in another subregion of p 1 dimensions. For let aly a2,... ap+1 and blt b2,... 6P+1 be any two sets of p + 1 points in regions of p 1 dimensions. Since the p + 1 points ax... ap+1 are contained in a subregion of /o 1 dimensions, apap+1 must intersect the subregion of the independent points aly a2i ... ap_! in some point c; and similarly bpbp+1 must intersect the sub- region of the independent points b1} b2 ... bp_1 in some point d. Now by hypothesis a series of projections can be made which transforms b1} b2... 6p_i, d into a1} a2... p_;l, c. Assume that such a series transforms bp and p+1 into b?f and 6rp^i. Let J.p_! stand for the subregional element axa2... a^, and let D^ denote the product of any v p independent points which do not lie in A^, where v 1 is the number of dimensions of the complete region. Then Ap-j Dv_p is a planar element. Again, c, 6/, b'p+1 are collinear and so are c, ap, ap+1, hence bp\ b'p+1, ap, ap+1 lie in the same two dimensional region. Therefore apbp and ap+16'p+1 intersect in some point e. Let Dv-P be so chosen that e does not lie in the plane AP^1DV_P: also let Dv-P contain ap and a^. Then it cannot contain bp and b'p+1, since it does not contain e. Project on to the plane J4P_1DJ/_P from the vertex e. The points alf a2, ... ap_i, c are unchanged, being already in that plane, also bp is projected into ap, and 6'p+1 into ap+1. Hence the proposition is proved. (10) It has already been proved that three collinear points can be pro- jected into any other three collinear points ; it follows that any p points in a
130] PROJECTION. 227 subregion of p 2 dimensions are projective with any other p points in another subregion of p 2 dimensions. (11) Proposition IV. The least number of separate projections required can also be easily determined. For we have proved in the course of subsec- tion (9) that if f (p) projections are required for two sets of p points in subregions of p 2 dimensions, then f (p) + 1 projections are required for two sets of p +1 points in subregions of p 1 dimensions. We have there- fore only to determine the least number requisite to project three collinear elements a, 6, c into three other collinear elements a, b\ c'. The construction given above in the second and general case may be simplified thus. Join ab'. Project from any point e on bb\ Then a is unaltered, b becomes V and c becomes some point c". Now project a, V c" from the point of intersection of aaf and c'c". Then a becomes a', bf is unaltered, c" becomes c'. Hence two projections are in general requisite. Thus three projections are requisite for four points in a two dimensional region, and p 1 projections for p points in a region of p 2 dimensions. (12) These constructions still hold if the two sets of p points are both in the same subregion of p 2 dimensions. In such a case the same series of projections which transforms one set of p elements into another set of p elements may be conceived as applied to every point of the subregion. Thus every point of the subregion is transformed into some other point of the same subregion. (13) Proposition V. It will now be proved that the most general type of such a projective transformation is equivalent to the most general type of linear transformation which transforms every point of the given subregion into another point of that subregion. If x be the point into which any point x is finally projected, the relation between x' and x can be written in the form x1 = xex. BY. e2 B2. e3. B3... e?-1. -Bp_i. It is obvious therefore that x can be conceived as transformed into x by some linear transformation. The only question is, whether it is of the most general type. Now in the most general type of linear transformation, as applied to a region of p 2 dimensions, p 1 elements must remain unchanged. Let alf a2, ... ap_! be these elements, and let any other point x be represented by Then if x1 be the transformed element corresponding to x, we have #i = ai i + o^^Aj + + p-if:p-i*V-i where alt o^, ... ap_! are the constants which in conjunction with the fixed points define by their ratios the linear transformation. Hence if a given point g is to be transformed into a given point d, where q %fyay and d = 2Sa, we must have, 0L1 = 81/y1, ok^ty^, p-i = Sp-iA/p-i 15 2
228 DESCRIPTIVE GEOMETRY. [CHAP. IV. Accordingly if the p 1 unchanged points are arbitrarily assumed, it is not possible by a linear transformation to transform more than one arbitrarily assumed point into another arbitrarily assumed point. But it is possible by a series of projections to transform the p points Oj, a2, ... ap_i, c into the p points alt a2, ... CLp-.ly d. Hence the general type of projection is equivalent to the general type of linear transformation. (14) It is to be noticed that a linear transformation can be conceived as transforming all the points of the complete region of (say) v 1 dimensions. But these points cannot be projectively transformed without considering the region of v 1 dimensions as a subregion in a containing region of v dimensions. This fresh conception is of course always possible without in any way altering the intrinsic properties of the original region of v 1 dimen- sions. The Theory of Linear Transformation in connection with this Calculus is resumed in Chapter VI. of this Book.
CHAPTER V. Descriptive Geometry of Conics and Cubics. 131. General Equation of a Conic. (1) The following investigation concerning conics and cubics is in substance with some extensions a repro- duction of Grassmann's applications of the Calculus of Extension to this subject*. In places the algebra is handled differently and alternative proofs are given for the sake of illustration. A quadric surface in a complete region of two dimensions will be called a conic. It will also in this chapter be called a curve in order to agree with the usual nomenclature of Geometry. (2) The complete region is of two dimensions: the product of three points or of three linear elements or of a point and a linear element is purely numerical. Also the product of three linear elements, being a pure progressive product, is associative; thus if Llt L2 L3 be the linear elements, (L^LJL^) = (Li. L2L3). Also if p and q be points, then, since L L2 is a point, {LxL2pq) is the product of the three points LYL2i p} q. Hence (L^pq) = (Z^ . pq). (3) The equation, (xaBcDex) = 0, where a, c, e are any points and B and D are any linear elements, is evidently, since x occurs twice, of the second degree in the three co-ordinates of x. For let e1} e2, es be the three reference points, and let x = f^ 4- Also let the fixed points and lines be written in the form a = oa + a^a + a3e3, B = $ 4- and so on; where alf a^ a3, etc., are given numerical coefficients. Then the given equation, after multiplying the various expressions for the points and lines, takes the form (eAfrY (\ +... + 2X^3 + ...) = 0. Hence, dividing out the numerical factor (e^e-s)3, the given equation is equivalent to a single numerical equation of the second degree defining a quadric locus. * Cf. Ausdehnungslehre von 1862, and Crelle, vols. xxxi, xxxvi, lii.
230 DESCRIPTIVE GEOMETRY OF CONICS AND CUBICS. [CHAP. V. Write zrx for the expression (xaBcDex), then the following transformations by the aid of subsection (2) are obviously seen to be true: urx = {xaBcD . ex} = {xaBc (D . ex)} = - {xaBc. exD} [c . axB . exD} = {c . exD . axB} = {xeDcBax}; where it is to be remembered in proving the transformations that xa is a linear element, xaB is a point, #ai?c is a linear element, xaBcD is a point. (4) From zjx = {c . a#i?. e#D} = 0, it is obvious that a and e are points on the conic. In general c is not on the conic, for the points c, (acB), and (ecD) are not in general collinear. (5) Points in which B and D meet the curve. Suppose that B meets the curve in the point p, and let B = pq. Also substitute j9 for x in the expression vrx. Now a )i? = ap . pq = (apq)p = (aB)p. Therefore -ro^ = {c . opi?. epD} = (a5) (cp . epD) = 0. This involves either (i) that (aB) = 0, or (ii) that (cp . epD) = 0. (i) Let (aB) = 0. Then axB = (aB) x - (xB) a = - (xB) a. Hence x = (#i?) (ca . z?2 ) = 0. Therefore the curve splits up into the two straight lines (xB) = 0, and (ca . exD) = 0. Similarly if (eD) = 0, the curve becomes the two straight lines (xD) = 0, and (ce . axB) = 0. These special cases in which the conic degenerates into two straight lines will not be further considered. (ii) Let (cp . epD) = 0. Then {cp. epD} = {c(p. epD)]. But p.epD^p {(eD)p - (pD) e} = - (pD)pe. Hence (cp. epD) = (pL) (cep) = 0, so that p lies in D or in the line ce. Accordingly the two points in which B intersects the curve are B. JD and B.ce. Similarly the points in which D intersects the curve are B. D and D . ca. (6) Let g = B.D, b = B.ce} d = D.ca. Then b = (Be) c - (Be) e, d = (Da) c - (Dc) a. Hence eb.ad = (Be) (Da) {ec. ac} = (Be) (Da) (eac) c = c. Also we may write B bg} D = dg. Hence the equation becomes {(xa . bg) (eb . ad) (dg. ex)} = 0; where a, 6, d} e, g are five given points on the curve and x is a variable point. (7) Conversely, if we take any five points a, b, d} e g and write, {(xa. bg) (eb. ad) (dg. ex)} = 0,
131, 132] GENERAL EQUATION OF A CONIC. 231 then the above reasoning shows that the five points are on the curve which is the locus of x. But only one conic can be drawn through five points; therefore by properly choosing the five points this equation can be made to represent any conic section, and is therefore the general equation of the second degree. (8) If we perform the constructions indicated by the products on the left-hand side (cf. fig. 1), we see that the equation is a direct expression of Pascal's theorem, which is thereby proved. Fig. l. (9) Perform the operation of taking the supplement on the equation, and write X for the linear element \x, A for \a, and so on. Then | {(xa . bg) (eb . ad) (dg . ex)} = {(XA . BO) (EB . AD) (DO. EX)\ = 0. This is the general tangential equation of a conic [cf. 107 (4)]: hence from subsection (7) it follows that A, B} D, Ey G are tangents; and the equation is a direct expression of Brianchon's Theorem. 132. Further Transformations. (1) These results can be obtained by a different method which forms an instructive illustration of the algebra. The following series of transformations follow immediately from the extended rule of the middle factor: axB = (aB) x (xB) a; hence, axBcDx = {(aB) xc . D (xB) ac . D) x = (aB) (xD) ex - (xB) (aD) ex + (xB) (cD) ax. Now (aB) (xD) - (xB) (aD) = x[(aB) D - (aD) B]=x[a. DB] = (xa . DB). Hence, axBcDx = (xa . DB) ex + (xB) (cD) ax, and, (axBcDxe) = (xa . DB) (cxe) + (xB) (cD) (axe).
232 DESCRIPTIVE GEOMETRY OF CONICS AND CUBICS. [CHAP. V. Thus the equation of the curve, vrx = 0, can be written (xa. DB) (ewe) + (xB) (axe) (cD) = 0. (2) To find where B meets the curve, put (xB) = 0. Then either (xa . DB) = 0, or (cxe) = 0. Thus either x is the point BD or it is the point ceB; therefore these are the points where B meets the curve. Similarly the points where D meets the curve are BD and caD. (3) Obviously the points a and e lie on the curve. (4) If (cD) = 0, the curve degenerates into the two straight lines (xa . BD) = 0, (xce) = 0. Similarly if (cB) = 0, the curve becomes the two straight lines, (xe.BD) = 0, (xca) = 0. (5) To find the second point in which any line through the point a cuts the curve. Let L be the line, then (aL) = 0. Let x be the required point in Z, then xa = L. Hence (xaBcDex) = (LBcDex) = 0. Hence x is incident in the linear element LBcDe, also x is incident in L. Therefore x = LBcDeL. (6) It is to be noticed that a apart from L does not appear explicitly in this expression for x. Hence the theorem can be stated thus: If a be any variable point on the line L, the conic through the five points a. BD, ceB, e, caD passes through the fixed point LBcDeL. (7) The conditions that T should be the tangent at a are (aT) = 0, and a = TBcDeT. (8) The general expression ibx is susceptible of a very large number of transformations of which the following is a type: xa.bg = (xbg) a (abg) xy eb.ad = (ebd) a - (eba) d, dg .ex = (dgx) e (dge) x. Hence {(xa. bg)(eb. ad)(dg. ex)} =(eba)(dge)(xbg)(adx) - (eba)(ade)(xbg)(dgx) -h (abg) (ebd) (dgx) (aex) (abg) (eba) (dgx) (dex). (9) The equation, (xalB1a2B2... an-1Bn_-1anx) = 0, represents a conic. Hence the following theorem due to Grassmann: 'If all the sides of an n-sided polygon pass through n fixed points respectively, and n 1 of the corners move onw-1 fixed lines respectively, the nth corner moves on a conic section/
133, 134] LINEAR CONSTRUCTION OF CUBICS. 233 133. Linear Construction of Cubics. The first linear constructions satisfied by any point of a cubic were given by Grassmann* in 1846; and the theory was extended and enlarged by him in 1848 and 1856 f. An indefinite number of such linear constructions of increasing complexity can be successively written down by the aid of the calculus. * The simplest types are given by (xaAa1.xbBWb1.xc) = O...........................(1), (xaAax. xbBbx .xc) = Q...........................(2), (x. xaBcD . xaji-fij)^ = 0 (xaBcDxD B^x^ (xaA . xbB. xcG) = 0...........................(4). The two equations, marked (3), are alternative forms of the same equa- tion. It is to be noted that none of these constructions give a method of discovering points on a cubic; but that, given a point x on a cubic, the constructions can be made. Thus a point x on the cubic will be said to satisfy the corresponding construction, but not to be found by it. 134. First Type of Linear Construction of the Cubic. (1) To investigate the construction {xaAax. xbBkCbi. xc) = 0. This equation asserts that if the three lines xaAalt xbBkCbly xc are concurrent, the locus of x is a cubic. Let y be the point of concurrence; then the construction is exemplified in figure 2. Fig. 2, (2) It has now to be proved that any cubic can be represented by this construction. This will be proved by shewing that by a proper choice * Cf. Crelle's Journal, vol. xxxi. t Cf. Crelleh Journal, vols. xxxvi. and lii.
234 DESCRIPTIVE GEOMETRY OF CONICS AND CUBICS. [CHAP. V. of the fixed lines and points of the construction the cubic may be made to pass through any nine arbitrarily assumed points. Thus we proceed to investigate the solution of the following problem : Given any nine arbi- trarily assumed points in a plane to find a linear construction satisfied by any point of the cubic passing through them. But previously to the direct solution of this problem in 135 some properties of the expression {xaAaY. xbBkCb-^. xc) must be investigated. (3) Let thx stand for the product (xaAa^. xbBkGbx. xc). Then urx = (xaA at. xc . xbBkCbJ. Now put p = xaAa^ .xc, q = xbB. Then mx = - (p . qkCb,) = It is easily proved that {pbYCkq) = (4) To find the particular positions of x for which p = 0, or q = 0. Now p = 0, when x a, and when x = c. Also by 105 p = {xaAaxc) x (xaA^x) c = {(xA) (aaic) (aA) (xaic)} x (xA) (actix) c. Hence all the points x for which p = 0 (except x = c) must satisfy (in order to make the coefficient of c zero) either (xA) = 0, or (xaa^ = 0. If (xA) = 0, then, since the coefficient of x must also be zero, (xatf) = 0. Hence x = aYcA ; and thus a^cA is another of the required values of x for which p vanishes. If (xaa1) = Q, then (xA)(aa1c) (aA)(xa1c) = Q. The only point on the line aa which satisfies this equation is the point a. For if Xa + fxa^ be substituted for x, the equation reduces to fi (alA)(aalc) = 0; and hence, /jl = 0. Hence the three values of x for which p = 0 are a, c, a^cA. The only value of x for which q = 0 is x = 6. (5) To investigate the values of x for which p = x. These are included among the points satisfying the equation px = 0. Though this equation for x is also satisfied by the points just found which make p = 0. Now px = (xA) (aaxx) ex. Hence if x lie in A, i.e. if (xA) = 0, or if x lie in aalf i.e. if (xaa1) = 0 then p = x. But the points a and a^cA must be excluded, as involving p 0. (6) The points for which q = x are given by qx = 0, excluding the point b for which q = 0. Now qx = #6ifo = (#J9) xb. Hence if (xB) = 0, then q = x. Thus either of the points AB or aaYB substituted for x in the expressions p and gr make p=x = q.
135] FIRST TYPE OF LINEAR CONSTRUCTION OF THE CUBIC. 235 (7) Hence if x be either of these points -urx {pb1Ckq) = (xbfikx). Now (xb Gkx) = (xbjc) (xG). Therefore (xb Gkx) = 0, implies either (xbjc) = 0, or (xG) = 0. Hence if the points AB and aaxB lie on the cubic they must lie either on bje or G. Thus if A, B, G be concurrent, the point of concurrence lies on the cubic. This analysis of the equation will enable us easily to follow Grassmann's solution of the problem. 135. Linear Construction of Cubic through nine arbitrary points. (1) Let the nine given points be a, b, c, d, e f g, h, i\ and let the cubic be (xaAa^. xbBkGbx. xc) = 0. Then the curve obviously goes through the points a, b, c. Let a1cAJ which lies on the cubic [cf. 134 (4)], be the point d; and let A, B, G be concurrent in the point e, which is therefore on the cubic by 134 (7). Hence we may write A = de. Let the point aaxB lie on bjc and therefore be on the cubic by 134 (7): let it be the point / so that (fbjc) = 0. Hence both e and / lie on B; there- fore we may write B = ef. Also axcA = d now becomes a^ . de = d; hence d is the point of intersection of a-fi and de and therefore (a^d) = 0. And {axaB) =f becomes aYa. ef=f; hence (^af) = 0. Therefore, since ax lies both in cd and in af, we may assume a1 = af. cd. It has been assumed that no three of a, c, d, e, f, are collinear; for other- wise some of these equations become nugatory. (2) We have k, G, 62 still partially at disposal: the conditions to be satisfied by them being only, (Ge) = 0, (A*) = . Now let #! = gaAax ,gc=ga (de) (af. cd). gc, hx = haAax ,hc=ha (de) (af. cd) . he, 11 = iaAa1. ic = ia (de) (af. cd) . ic, =gb.ef, =hb.ef 12 = ibB = ib. ef Thus the six points glt hly ilt g2, h2 i2 can be obtained by linear construc- tions from the nine given points a, b, c,... i. We proceed to choose k, C, 6X, so that the following equations hold [cf. 134 (3)] = O, (h2kCb1hl) = 0, (i2kCb1i1) = O.
236 DESCRIPTIVE GEOMETRY OF CONICS AND CUBICS. [CHAP. V. (3) Let G and k be chosen, if possible, to satisfy the equation (i Ckit) = 0 without conditioning blt Then for this purpose we must write i C = iY; that is to say, C must be assumed to pass through iY. But e lies in C, hence we must assume G = eilt Hence k is given by (kifa) = 0. Further, except in the special case in which (fei1)=0, k and bx are also [cf. subsection (1)] related by (fkb1) = O. Thus h = i1i2.fbi where bx is as yet any arbitrarily assumed point. (4) The remaining equations can be written Hence k must also be such that the three lines kf, kg2Cg1} kh2Ghx intersect in the same point 62; also k lies in ixi2. Therefore k is one of the points in which ixi2 intersects the cubic curve, (xf. xgJJgx. xh2Gh1) = 0. (5) But this curve is formed of three straight lines. For if x be any point in Cy then xgJJ = x, xh2G = x, and hence (xf, xg2Gg1. xhJJh^) = (xf. xgx. xhx) = 0. Thus G is part of the locus. Now g2 (=gbB) and h2 (= hbB) both lie on B, also / lies on B. Thus if x be any point on B} xf=B, xgjOg1 = BGglf xh.Ch^BCh,. Hence (xf. xg2Cg,. xh.Ch,) = (B . BCg1. BGh,) = 0. Thus jB is part of the locus. (6) Hence the remainder of the locus is another straight line. To find this required line, let y = xf. xg2Cg1. Then (y. xh^Gh,) = (xk.Gh^) = - (yKCK.2x). Hence the equation of the three straight lines is (xf. xg2GgY. xh2Ch^ = {xf'(xgfig^ hfih^x] = 0. This equation is satisfied by any value of x for which that is, by any value of x making xf and xg2Ggx intersect in /^; that is, if x satisfies (xfh^) = 0, and (xgJJgJi^ = 0. But (xg2Gg1h1) = (h^Gg^); hence x must lie in the intersection of fhx and h1g1Cg2. Therefore x = h1g1Gg2.fh1. Similarly another point on the third line is gJi1Ch2.fgl. Hence the required line which completes the locus is 2 ,f\
135] LINEAR CONSTRUCTION OF CUBIC THROUGH NINE ARBITRARY POINTS. 237 (7) Put K^(h1g1Cg2Jh1)(g1h1Ch2.fg1). Then k must lie in iYi2 [by subsection (3)] and in B or G or K. Now the assumed equation of the cubic is (waActj. xbBkCbj. xc) = 0. Assume that k lies in B. Then oobBk = (xbk) B. Hence the equation of the cubic becomes (xbk) (xaAax. BGbY. xc) = 0. Accordingly the cubic degenerates into the straight line bk and a conic section; and cannot therefore be made to pass through any nine arbitrarily assumed points. Assume that k lies in G. Then xbBkC = {xbBG) k. Hence the equation of the cubic becomes (xbBC) (xaAax. kbx. xc). Thus in this case also the cubic degenerates in a conic section and a straight line, namely, BGb. Therefore the only possibility left is that k lie in K. It will be shown that this assumption allows the cubic to be of the general type by showing that the cubic passes through the nine arbitrarily assumed points. Hence let it be assumed that k = i{L2K. Accordingly with these assumptions the equations are satisfied. Again, fcj has been determined, for by subsection (4) it is the point of intersection of kf, kg2Ggly M20Aa; hence 6X = kg2Cgx. kf. (8) Finally, therefore, it has been proved that the equation, xaAaY. xbBkCbY. xc = 0, denotes a curve passing through the nine arbitrarily assumed points a, b, c, d, e,f, g, h, i, provided that A, B, C, aa, bl9 k are determined by the linear constructions, A = de, B ef, G eiv a2 = af'. cd, k = iji2K, 62 = kg2GgY. kf\ where g1 = ga (de) (af. cd). gc, h^ = ha (de) (af. cd) . he, iY = ia (de) (af. cd). ic, g2=gb.ef, h2 = hb.ef i2 = ib.ef, and K = (Kg^ .fh,) (gACh2. fg,). (9) This linear construction satisfied by any point x on the cubic repre- sents the general property of any ten-cornered figure x, a, b, c, dt e, f g, h, i, inscribed in a cubic. It is the analogue for cubics of Pascal's Theorem for conies.
238 DESCRIPTIVE GEOMETRY OF CONICS AND CUBICS. [CHAP. V. 136. Second Type of Linear Construction of the Cubic. (1) Equation (2) 133, namely (xaAa-L. xbBb,. xc) = 0 is a simplified form of (1), which has just been discussed. It can be derived from (1) by putting k =b . For in this case xbBb.Cb, = (xbBbA) 0 - {Ob,) xbBb, (2) Hence as in 134 (4) and (7) the points a, b, c, a^cA, AB, aa^B are seen to lie on the curve. Similarly, from the symmetry of the equation, b^B, bbxA are seen to be points on the curve. Also it is easy to see that aaY. bb, is a point on the curve. Let these nine points be denoted by a, b, c, d, e, f, g, h, k respectively; so that , e = ABy f=aaxBt g^ (3) To prove that the cubic denoted by this equation is of the general type. Take any cubic, and inscribe in it any quadrilateral khef as in the figure 3. Fm. 3. Let the side kh meet the curve again in J, the side he meet the curve again in d, the side ef in g, the side fk in a. Assume c to be any other point of the curve not collinear with any two of the other points. Then
136, 137] SECOND TYPE OF LINEAR CONSTRUCTION OF THE CUBIC. 239 the assumed points on it determine the cubic. Join cd cutting kfin alt and eg cutting hk in 6a. Then iffe = B,he=A, the equation xaAax. xbEbx . #c = 0, has been proved to represent a cubic through the nine points. Hence by a proper choice of constants the equation can represent any cubic. (4) The construction represented by this equation is exemplified in figure 4. Fig. 4. 137. Third Type of Linear Construction of the Cubic. (1) The equation (3) of 133 is The points a and a2 obviously lie on the curve. To discover other points on the curve, notice that by 132 (1) axBcDx = (xa. DB) ex + (xB) (cD) ax. Hence isx = (xa. DB) (cxD B^a^ + (xB) (cD) (axD^B^a^.........(A). But (cxD B^xa^) = 0 is, by 131 (4) and (5), a conic through the five points c, a1} B1D1, c^D^ c^cB^ Also (axDtfJixxa^) = 0 is, by 131, a conic through the five points a, aly Hence the points B^ and c1a1D1 lie on both conies and therefore also on the cubic. But tbx Hence BD and caD are also points on the curve.
240 DESCRIPTIVE GEOMETRY OF CONICS AND CUBICS. [CHAP. V. (2) As a verification notice that, if x BD, then xaB = x, xaBcD = xcD = x, xaBcDx = xx = 0: also, if x = ca . D, then xa = ca, xaBc = ca.Bc = ca, xaBcDx = caD# = ## = 0. (3) To find where D cuts the curve a third time; note that axBcD is a point in D; hence if # be also in D, axBcDx = D, excluding the case when axBcDx is zero. Hence, by substituting D for axBcDx in the equation of the curve, we see that x satisfies (DD B^x) = 0, and (Dx) = 0 : therefore x = DD B^D. Hence D cuts the curve in the three points BD, caD, DD^JB^D, and similarly Dj cuts the curve in the three points BYDly c1a1Dl, DJ)cBaDlt (4) The two conies (cxD^B^a^ = 0, and (axD^Bxa^) = 0, have been proved to intersect in the three points aly B^D^, c^JDi. The fourth point of intersection is caD B^. ca; since by 132 (5) this is the point in which the line ca meets either conic. Hence the three points in which the line ca meets the cubic are a, caD, caD B^ . ca. Similarly the three points in which the line c^ meets the cubic are a11 CidiD1} CiOjDcBa . c^. (5) It is easy to obtain expressions for the three points in which the line BDa cuts the cubic. Two of the points are already known, namely, a and BD. To find the third notice that from equation (A) of subsection (1), the required point is the point, other than a, in which the line cuts the conic (xaD B^x) = 0. By 132 (5) this is the point (BDa) D B^ (BDa), which can also be written BDclD Bm . BDa. Thus the three points in which BDa meets the cubic are a, BD, BDaD1c1B1a1. BDa. Similarly the three points in which B1D1a1 meets the cubic are a1} B1D1, (6) To find the three points in which B cuts the curve, notice that if (xB) = 0, then from equation (A) of subsection (1) the equation of the curve reduces to (xa . DB) (cxD B aJ = 0. Hence either xa. DB = 0, and x= BD, which has been already discovered; or (xcD B^x) = 0. Therefore the two remaining points in which B meets the cubic are the points in which B meets the conic (xcD B^x) = 0. These points can be immediately expressed, if B = Bl. In this case the cubic becomes (xaBcDxD^Ba^) = 0; and it will be proved [cf. subsection 13] that the equation still represents any cubic curve.
137] THIRD TYPE OF LINEAR CONSTRUCTION OF THE CUBIC. 241 The points where B meets the conic, (xcD Ba^x) = 0, have been proved in 131 (5) to be BDX and ccxB. Hence B meets the simplified cubic in the three points BD, BB1} c^B. (7) The transformation gtx = x . xaBcD . xalB1c1D1 is established as follows. Let x = xaBcDxD B^, then ^x = xla1x= x1.xa1\ since the product of three points is associative. Let X2 = xaBcDxD^y then ?nx X2BY. xaY = X2. xaYBY; since the pro- duct of three linear elements is associative. Let x3 xaBcDxDx, then mx = x$x. xaJ$Y = xz. xa-^B^. Let X4 = xaBcDx, then t*tx = X4D1. xa^B^ = X . xaYB1cxD1, Hence H7X (xaBcD) x. xa1B1c1D1 = x . xaBcD. xa1B1c1D1. The previous results can be easily obtained by means of this form of the equation. (8) The geometrical meaning of the equation is that x, xaBcD, and xa1B1c1D1 are collinear. This property is shown in the annexed figure 5. Hence if two variable triangles have a common variable vertex, and two sides, one of each triangle, which meet in the common vertex lie in the same straight line, and if also the four remaining sides pass respectively through four fixed points, and the four remaining vertices lie respectively on four fixed straight lines, then the locus of the common vertex is a cubic. (9) The four lines A (= ca), D, AY (= c^), A have a special relation to the cubic (xaBcDxD B^x) = 0, in addition to the fact that the points caD and c1a1B1 both lie on the curve [cf. subsection (1)]. For suppose that the lines A, D, Aly A are arbitrarily assumed. Then the points AD (= e) and A1D1(= e2) are determined. w. 16
242 DESCRIPTIVE GEOMETRY OF CONICS AND CUBICS. [CHAP. V. Also suppose that the remaining points in which A and A, cut the curve are arbitrarily assumed on these lines, namely [cf. subsection (4)], /(= AD,c,B,a,A), f, (= A,DcBaA,), a and a,. Thus a, a,,ff, are supposed to be known, and the equations /= AD,c,B,a,A and f, = A1DcBaA1 partially determine c, and B,, and c and B, which are the remaining unknowns. Again the arbitrarily assumed lines D and D, are supposed to meet the curve in two arbitrarily assumed points e(=AD) and e,(= A,D,). Let two other points k and k, in which D and A respectively meet the curve be arbitrarily assumed, so that [cf. subsection (3)] we may assume k = DD,c,Ba,D, and k, = D,DcBaD,. Then the remaining points in which D and X^ respectively meet the curve are [cf. subsection (3)] BD and B,D,. Call these points g and g,. It will now be shown that g and g1 are both determined by the previous assumptions of the eight points a, a1; e, e,,ff,y k, k,\ and that accordingly the group of four lines A, D, A,, D, must bear some special relation to the cubic curve which passes through the eight assumed points. (10) For if L, and L2 are linear elements and p, and p2 are points, the extended rule of the middle factor gives, L,L2p, = (L,p,)L2-(L2p,)L,, and p1p2L1 = (p,L,)p2-(p2L,)p,. Remembering these formulae we see that fa, = AD B^Aai = (Aa^) AD B^ = AD1c1B1a1; faxBx = AD1clB1a1B1 = - (aA) AD.cA = AD1c1B1; Hence AD-fi-y. fax .B1 = 0. Similarly DD . kax. Bx = 0. Hence BY passes through the points AD .fax (=p) and DD . fcc^ (= q). Therefore we may write Bx = (AD .fa^) (DD . kaY) =pq. Hence 9l = 5XA = (J.A^ . fa,) {DDlGl. A?ai) A = pqD1 = (pB,) q - Now (pA) = (^Aci .A . A) = - (ADlCl. A A) And (?A) = (^ i iCi. kaY. A) = - (i Aci. A . ha*) = (cxA) (DDJca,). Hence ^x = (AD,fa,) q - (DDJca,) p. But p = ADlCl .fa, = (AD,fa,) c, - (c,fa,)AD,; and q = D^d . A;^ = (DD,ka,) c, - (c,ka,) DD,. Also (c,fa,) = - (A,f), and (c,ka,) = - (AJc) by subsection (9). Thus g, = {AJc) (AD,fa,) DD, - (A,f) (DDJca,) AD,.
137] THIRD TYPE OF LINEAR CONSTRUCTION OF THE CUBIC. 243 Hence the position of gx is completely determined by the arbitrarily assumed elements. Similarly the position of g is completely determined. (11) Hence ten points on the cubic are now known. The cubic is there- fore independent of the positions of c and c2 on A and AY; except that c must not coincide with a or AD, nor cx with a or AxDly in which cases some of the previous equations become nugatory. (12) We will now prove that the specialized form of equation introduced in subsection (6), namely (xaBcDxD^Ba^x) = 0, where (cA) = 0 = (^D) represents the most general form of cubic. The three points in which D cuts the curve are [cf. subsection (3)], BD, caD, DD1c1Ba1D. But since (c.D) = 0, DD^Ba.D = DBa.D = DB. Hence D touches the curve at BD and cuts it again in caD. Similarly A touches the curve at BDY and cuts it again in c^A- Also [cf. subsection (6)] B cuts the curve in the points BD, BD1} Fig. 6. (13) Now (cf. fig. 6) take any cubic curve and draw the lines D and Dx tangents to it at any two points g and gx. Join gg1 by the line B which cuts the curve in another point L Through h draw any line cutting D in cx and A in c. The tangents D and A cut the curve again in two points e and eY. 16 2
244 DESCRIPTIVE GEOMETRY OF CONlCS AND CUBICS. [CHAP. V. Now join ec; this line cuts the curve in two points. Call one of the two a. Similarly call one of the two points, in which e^ cuts the curve, ax. Then by construction h = cCjB, e = caD, eY = c^JJ^ Now the tangents D, Dx at g and gY and the points h, e, e1} a, ax completely determine the cubic. But (xaBcDxD Ba^x) = 0 is a cubic satisfying these conditions. Hence this equation represents the assumed cubic. 138. Fourth Type of Linear Construction of the Cubic. (1) The equation (4) of 133 is {xaA . xbB. xcC) = 0; and it represents the fact that the points xaA, xbB, xcC are collinear. The construction is shown in figure 7. It will be shown that any cubic can be thus described. Fig. 7. (2) To find where A cuts the cubic, note that if x lies in A, xaA = x. Hence (xaA . xbB . xcC) = (x . xbB . xcG) = (xB) xb . xcG = (xB)(xG)(xbc); where the sign of congruence means that only constant factors have been dropped. Therefore the three points in which A cuts the cubic are AB, AC, bcA. Hence by symmetry, BG, caB, abC also lie on the cubic. Also obviously a, b, c lie on the cubic. Thus the two triangles respectively formed by a, b, c as vertices, and by A, B, G as sides are both inscribed in the cubic and their corresponding sides, namely A and be, B and ca, G and ab, intersect also on the cubic.
138] FOURTH TYPE OF LINEAR CONSTRUCTION OF THE CUBIC. 245 (3) We have to prove that, given any triangle abc inscribed in a cubic, a triangle A, B, G always exists with these properties relatively to abc and the cubic. Take a, b, c any three points on a given cubic, not collinear. Let be cut the cubic again in /, ca in g, ab in h. Let a, b, c be the reference triangle, and let , rj, f be the co-ordinates of any point x. Then we can write x = fa + rjb + c. Let A, B, C be any straight lines through f, g, h. Then, since any numerical multiples of A, B, and G can be substituted for them, we may write A = Xbc + 7lca + foab, B = yjbc + fxca + c^ab, G = ft3bc + asca + vab; where X, jjl, v are at our disposal and /3ly yl9 y2y a2, A as are known from the equations, f= bcA, g = caB, h = abC and from the fact that one of the letters with each subscript can be assumed arbitrarily without affecting anything except the intensities of A, B, G, which are immaterial. Now xaA = (xA) a (aA) x = (abc) {(Xf* + ^77 + A?) ^ ~~ X ( a + r)b + fc)} = (a c) {(71^ -f A?) a ~ ^vb ~~ ^?cl- Similarly #6J9 = (a6c) { /xfa + (a2f + 72f) 6 /x c}, ^c(7 = (a c) {- v%a vrjb + (j3d% 4- a3^)}. Hence, (^a^l . # i?. ,tcC) = 0 can be written as the ordinary algebraic equation, = 0. This becomes on expanding the determinant (w + A?) ( ?+ 72?) (Af + asi?) - ^^? (w + A?) - *X?f (oa? + 7 f) - V^ (Af + ^) - 2\/u iy? = 0. This is the equation to a cubic through the six points a, b, c, f, g, h: it is required to determine X, fi, v so that it may be the given cubic through these points. The given cubic is determined by any other three points on it flf g1} hx forming another triangle. Now X, fi, v can be so determined that the above equation is satisfied by the co-ordinates of these points. For by substituting successively the co-ordinates we find three linear equations to determine X, fju, v, each of the form X fju v Xfiv where Plf P2,..., P5 do not contain X, jjl, v.
246 DESCRIPTIVE GEOMETRY OF CONICS AND CUBICS. [CHAP. V. Now put a for Xfiv, and solve these three linear equations for X"1, /a"1, v~l. Then we may assume X (T fJL T V a where H1} H2,... K3 do not contain X, fi, v. Hence multiplying and replacing Xpv by cr, an equation of the form, p0a3 + p a2 + p2a + pz = 0, is found; where pOy plf p2, p3 do not contain X, fi, v. Thus there are three values of a, one of which must be real. Hence there are three systems of values of A,, fi, v; and one system must consist of real values. Thus three systems of values can be found for A, B, G; and one of these systems must make A, B, C to be real lines. Thus three triangles, of which one must be real, can be found related to a, b, c and to the given cubic in the required manner. Let A, B, G be one of these triangles. Then {xaA . xbB . xcG) = 0 is the given cubic. The above proof of the required proposition is different from that which is given by Grassmann*. 139. Chasles' Construction. (1) Another construction for a cubic given by Chaslesf, without knowledge of Grassmann's results or methods is represented by xeDpEdF . xfB . xdG = 0, where (Ff) = 0 = (Bd). (2) It is easy to proveJ the following relations: The points d, e, f, BG, GF lie on the curve. The third point in which de cuts the curve is deDpEdF (deG) Bf{de). The third point in which ef cuts the curve is efDpEdF{efB) Gd(ef). The third point in which BGf cuts the curve is FCdEpDe {BGf). * Crelle, vol. lii. t Comptes Rendus, vol. xxxvi., 1853. X Cf. Grassmann, loc. cit.
139] CHASLES' CONSTRUCTION. 247 The third point in which BCd cuts the curve is BFdEpDe(BCd). Also if we put a = deDpEdF(deC) Bf(de), b = e/DpEdF (e/B) Cd (ef), c = FG, A = BCf, a, = cdEpDeAc (de), bx = BFdEpDeBc (ef) ; then the given cubic can be expressed by the construction xaAaY . xbBb-L . xc = 0.
CHAPTER VI. Matrices. 140. Introductory. The leading properties of Matrices, that is of Linear Transformations, can be easily expressed by the aid of the Calculus of Extension. A complete investigation into the theory of Matrices will not be undertaken in this chapter: the subject will only be taken far enough to explain the method here employed and prove the results required in the subsequent investigations in the theory of Extensive Manifolds. Grassmann treated the subject in his Avsdehnungslehre of 1861 apparently in ignorance of Cayley's classical memoir on Matrices*. An exposition and amplification of Grassmann's methods was given by Buchheim-f\ The present chapter is in its greater part little more than a free translation of Grassmann's own writing, amplified by the aid of Buchheim's paper; except that Grassmann and Buchheim do not explicitly consider the case of a matrix operating on an extensive magnitude of an order higher than the first; and that the treatment here given of symmetrical matrices is new, and also that of skew matrices. I have also ventured in 146 to distinguish between latent regions and semi-latent regions: in the ordinary nomenclature both would be called latent regions. 141. Definition of a Matrix. (1) Let eli e2... ev be any v reference elements in a complete region of v 1 dimensions. Let the symbol (l\, CL-2 ... Ctv prefixed to any product of some or all of these elements, be defined to denote the operation of replacing the element e1 by aly the element e2 by a2, and so on. Thus if eK, eK ... ep be any of the original reference elements, alt a2 ... av _ 6K6\ ... p CLKCl\ ... p y * Phil. Trans, vol. cxlviii., 1858; and Collected Mathematical Papers, vol. n., no. 152. t Proc. London Math. Soc. vol. xvi. 1885.
140, 141] DEFINITION OF A MATRIX. 249 so that in this instance the symbol of operation has transformed the product eKe\ ... ep into the product aKaK ... ap. Let p be put for the symbol ^-^ -. ?i, 62 6V The convention with respect to the operator f will be the same as that with respect to the operator | which is stated in 99 (9). Then $ex alt f e2 = a2, = c^Oa, and so on. (2) It follows from this definition that cf is distributive in reference to multiplication. For f e1 f e2 = i 2 = feify, and so on. (3) Furthermore let f be defined to be distributive in reference to addition, so that if Elt E2...EP be regional elements of the crth order formed by the multiplicative combinations of the crth order formed out of the reference elements [cf. 94 (1)], then f (a^ + ol2E2 + ... + cxpEp) = a1 f E1 + 2 f E2 + ... + ap(f Ep. For example, if x = %xeY + f2e2 + . + %vev, then (4) The operator cf called by Grassmann a quotient may be identified with Cayley's matrix. For assume i = oLn^i + a21e2 + ... + avlev, a2 = al2e^ + a22e2 + ... + av2ev, av = tnxvex + a2^2 + + avvev. Then J x = (an^ + a12f2 + ... 4- a1(,| ) ^ f 1 + ^22^2 + . + OqvZv) e2 + (a,i?i + a^2?2 + . 4- ! ) ^,. Hence if we put f % = rj^ + rj2e2... +rfvev, then with the usual notation for matrices, a21, vl, av2 ... a^ Thus we may identify f Avith the matrix ||ap T||. (5) It will be convenient to call the elements a1} a2...av the elements of the numerator of the matrix, and elt e2...ev those of the denominator. The elements of the denominator must necessarily be independent, if the matrix is to have a meaning.
250 MATRICES. [CHAP. VI. 142. Sums and products of Matrices. (1) If Eo denote any regional element of the ath order, say exe2... e r, then Xau \a2...Xav If j denote the matrix ^^ ", then the matrix ------* v will 6\, 62 ... ev 6i, 62 . 6V be said to be the matrix cj multiplied by X, and will be written symbolically \cj . This convention agrees with the ordinary notation, and will cause no confusion when the matrix is operating on elements of the first order, but must be abandoned when the matrix operates on regional elements of order greater than unity. (2) If two matrices, operating on v independent elements of the first order, give the same result in each case, then they give the same result whatever extensive magnitude they operate on. For let Ci, c2...Cg, be any v independent elements, / and % the two matrices. Assume cx %cl3 / c2 = %c2,... j cv = ycv. Then any extensive magnitude Aa of the crth order can be written as the sum of terms of which \cxc2... c^ is a type. Hence f Aa = SX / C!C2... c0 = SX. c2. cj c2. f c3... 4 ca = SX. x i - Xc* X^-- Xc = X^*- Thus the two matrices j and ^ must be considered as equivalent, and their equivalence may be expressed by / = %. (3) If Cj, c2...cv be any v independent elements, and if the matrix / , originally given as ^ ^ ~, give the results (f d = d1} j c2 = d2... j cv = dvy 6if 62 ... 6V then 6 can also be written in the form ^-^ ". For if A be any exten- V-l, C/2 (" sive magnitude, it follows that $A = ^ ^ v A. Cj, C2 .. Cv Hence any matrix can be written in a form in which any v independent elements form its denominator. (4) The sum of numerical multiples of matrices operating on any element of the first order can be replaced by a single matrix operating on the same element. For it can be seen that qK b2... bv } ^ = ce 6ee J ex, e2... ev 6\, e2... ev But if the extensive magnitude operated on be of order greater than the first, then this theorem is not true. For example consider the product e^. Then \ex, e2 ... el9 e2...)
142] SUMS AND PRODUCTS OF MATRICES. 251 But in general axa2 -f bj)2 is not a single force, and cannot therefore be derived from e^ by the operation of a single matrix. (5) A numerical multiplier can be conceived as a matrix. For if x = 2 e, then Xx = 2f \e, where X is some number. Hence X may be con- ceived as the matrix-----------------. If A a be an extensive magnitude of the crth order, then 1 @2 v Also from subsection (4) if be any matrix, X any number, and x an element of the first order, then ( f + X) x can be written %%, where % is a single matrix. (6) Let f and ^ be two matrices and A any extensive magnitude. Then the expression f xA is defined to mean that the transformation x A = B is first effected and then the transformation j B. The combined operation f x can itself be represented by a single matrix. For let e1} e2 ... ev be the independent reference elements, and let % ?i = fli, X a = fla X^ = ^; and let ! = tj, f a2 = 62 / ai/ = ^f Then the matrix which replaces ^ by b1} e2by b2... # by , is equivalent to the complex operation / ^;. (7) The operator 0^, when operating on an element of the first order, may be conceived as a product [cf. 19] of two matrices. For let ^ be a third matrix, and a any extensive magnitude of the first order. Then (x + id a = 4 (xa + ira)== 4 xa + ^fa=( bc + ^)a- Hence the two operators / (% + ^) and ^ + ^-v/r are equivalent. It is to be noticed that the sum of the matrices is another matrix and the product of matrices is another matrix. It will be convenient to speak of the product of two matrices when the matrices are operating on a magnitude of an order greater than the first. In this case the matrices have not a sum [cf. subsection (4)], and therefore strictly speaking have not a product [cf. 19]. The product of three matrices is associative; that is f x . -\{rA = f . %^^4. For the meaning of f % . yfrA is that a single matrix fa is substituted for the product / %, and the meaning of f . %^-4 is that a single matrix j^i is substi- tuted for the product x^'* an ^ then the equation asserts that fayjrA = / %p4. Now let e1} e2...ev be the v reference elements; then, taking a typical element only, let ^rep = ap, xa? p 0^p = V Hence (j idf = cP) and
252 MATRICES. [CHAP. VI. Therefore i"^eP = f idP = cp, and $ %iep = j)bp = cp. Thus fayjrep = (f XieP 5 an(i, since this is true for every reference element, = (f X\^ where J. is any extensive magnitude. 143. Associated Determinant. If the matrix can be written in ,, , . r aly a2 ... av , dly d2... dv . , . (a^... av) the alternative forms------------and-------------, then the ratios ~----------^ and . dv) Yi C2 ... Cv ... Cv) are equal. For let cx = Yh^ + 712^ + ... + Yi,A, with i; 1 other similar equations. Then dx = / d = 7n^i + 712^2 + + 7i, / ^ = 7n i + 712^2 + - + Yi ^ with v 1 similar equations. Hence (c^o,... cv) = A (e^g... ^.,), where A stands for the determinant 721, 722 Similarly Finally therefore, 7*1 7,2 7^. xd2... dv) = A {axa2... av). ^2... dv) _ (a^ ... av) ... cv) (2) If with the notation of 141 (4) the matrix be \ ^11 ^12 ^"\v / j ! I !, av2 *v then the ratio {a^a2... av)\(eYe2... ev) is the determinant an, a12 ... alt ... CLV 144. Null Spaces of Matrices. (1) If the v elements which form the numerator of a matrix are not independent, so that one or more relations exist between them, then the matrix can always be reduced to the form in which one or more of the elements of the numerator are null. For let the matrix f be ^ ^ -; and let alf a2... aM be independent, e1, e2... ev while the remaining v fju elements of the numerator are expressible in terms of the preceding //, elements ; so that we may assume v jjl equations of the form
143, 144] NULL SPACES OF MATRICES. 253 Let c^+1} c^+2... cv be defined by (v fi) equations of the type, where ^d^+p = cv+p- Then it is easily seen that elf e2... e^, c^+i, cM+2 ... cv are z/ independent elements. Hence these elements can be chosen to form the denominator of the matrix. But $ V+p = 0^+p ~~ fan+p = ah+p ~ ah+p = 0* tt ,i , , i .i r ax, a2... aM, 0, 0 ... 0 Hence the matrix takes the form------------------------------. i, e2... #/* c^+i, c^+2 Cf (2) In this case the associated determinant is zero. The region of ^ fi 1 dimensions, is called the null space of the matrix; and the matrix is said to be of nullity v jjl. Thus if the associated determinant vanish, the matrix is of nullity other than zero. Any point in the null space is said to be destroyed by the matrix, and will be called a null point of the matrix. Any point x is transformed by the matrix into a point in the region (ely e2... e^). This region (e1} e2... eM) is said to be the space or region preserved by the matrix. (3) Sylvester* has enunciated the theorem that the nullity of the product of two matrices is not less than the greater of their nullities, but not greater than the sum of the two nullities. The following proof is due to Buchheim-f*. Let cf be a matrix of nullity a and let Na be its null space; and let % be a matrix of nullity /3 and let iV^ be its null space. Also let P,_a and Pv-p be the spaces preserved by cj and % respectively. Then if Na and Pv_p intersect in a region T$ of S 1 dimensions, the nullity of the matrix ^ is fi + S. For to find the nullity of ^ we have only to find the most general region which ^ transforms into T8, since any point in this latter region is destroyed by 0. Now if T and Np be taken as co-ordinate regions [cf. 65 (3)], any point in the region of fi + S 1 dimensions, defined by the co-ordinate elements lying in T and iVp, is transformed by ^ into a point in T . Thus the nullity of (f x is y8 + S, and the null space of / % is the region defined by the co-ordinate points lying in Ts and Np. Hence the nullity of % is not less than fi, being equal to /3 if Na and Pv-p do not intersect. Also it is immediately evident that the nullity of f x is at least equal to the nullity of f : for if yx lie in 2Ta, then 0%^ = O. Hence the nullity of / % is not less than a. * Cf. 3 Johns Hopkins Circulars 33, " On the three Laws of Motion in the World of Universal Algebra." t Cf. Phil. Mag. Series 5, vol. 18, November, 1884.
254 MATRICES. [CHAP. VI. Again, to prove that the nullity of (f x is less than a-f/3, note that if a + ft v the theorem is obvious. For a matrix of nullity v would destroy all space. Assume therefore a + ft v. Now S is greatest when N* is con- tained in Pv-p, since a v ft; hence the greatest possible value of 8 is a. Thus the greatest possible value of the nullity of jy% is a + ft. (4) Buchheim extends Sylvester's theorem. For if a + (v ft) v, that is, if a ft, then in general Na and Pv_p do not intersect. In this case there is no region T8. Thus if a ft, the nullity of f x is in general ft. Again, if a + (v ft) v, that is, if a ft} then in general Na and Pv_p intersect in a region of a ft - 1 dimensions; thus h = a- ft, and in general the nullity of f X is Thus in general the nullity of / % is equal to the greater of the two nullities of f and ^; but if special conditions are fulfilled, it may have any greater value up to the sum of the two nullities. 145. matrix, Latent Points. (1) If a point x is such that, f being a given then x is called a latent point of the matrix, and the ordinary algebraic quantity p is called a latent root. The transformation due to the matrix does not alter the position of a latent point x, it merely changes its intensity. (2) Let the latent point x be expressed in the form Xge. Also let ( / p) e1 = c1} (cf p) e2 = c2, and so on. Hence ($ p)x = 0 = %j;( f) p)e = H!;c. Therefore clf c2... cv are not independent, and thus {cYc2... cv) = 0. This equation can also be written n [( / p) e] = 0, that is {( #! pe^) ( f e2 pe2)... ( f ev pev)} = 0. This is an equation of the j/th degree in p, of which the first term is ( l)v(e1... ev)pv, and the last term is ((j)e1. cj e2 ... f ev). The roots of this equation in p are the latent roots of the matrix. (3) From 142 (4) and (5) with the notation of 141 (4), j p is the matrix ... alv p ti* , O-V2 ... avv p Hence the equation for the latent roots is ^21 y ^ 22 P 2v (t vl , CLV2 d-vv p = 0.
145] LATENT POINTS. 255 (4) If all the roots of this equation are unequal, then v, and only v, latent points exist, one corresponding to each root, and these points form an independent system. These propositions are proved in the following three subsections. (5) There is at least one latent point corresponding to any root px of the equation giving the latent roots. For let ( j - p^)e1 = d, ( / - p^ e2 = c2 ...( / - pi)ev = cv. Then since (c^ ... cv) = 0, a relation holds such as 7^ + 72c2-h ... -f yvcv = 0. Hence jj ( j -pOe1 + y2( f -p1)ea+...+ yv( / -pjev = 0 ; this becomes f {7^ + y2e2 + ... 4- yvev) = px {7^ + y2e2 + ... + yvev\. Hence the point 7^ + y2e2 + + Y A is a latent point corresponding to the root pY of the equation. (6) A system of v such points, one corresponding to each root, form a system of independent elements. For let a1} a2... av be the v latent points; then, if they are not all independent, at least two of them are independent, otherwise the v points could not be distinct. Assume that the //, points alf a2 ... a^ are independent, and that another point aa can be expressed in terms of them, by the relation aa = axax + a2a2 + ... + V V Then (/ . = a^dj + or2 / a2 + .. + a^a^, that is, paCta = !/?! ! + a.2p2(i2 + ... + a^/o^a^. Multiply the first equation by pa and subtract from this equation, then 0 = (p1- pa) dOx + (p2 - p0) CL2a2 + ... + {pt*. ~ pa) W Since none of the latent roots, plt p2 ... po are equal, this forms one relation between a1} a2... a^ contrary to the hypothesis. But at least two of the latent points must be independent, hence they are all independent. (7) Two latent points cannot belong to the same latent root. For assume that ax and a/ are two distinct points such that f a1 = pYaly a/ = /?1Oi/. Let a2, a3... av be latent points corresponding to the remaining v 1 roots. Then alf a2... av form an independent system. Hence a/ can be written in the form a^ + a2a2 + ... + W Hence a/ = a^a^ + a2 a2 + ... + av \ av = /o^^i -h p2ct2a2 + ... + pvavav. But / a/ = p l = pYaxav + pxa2a2 + ... + pYavav. Therefore (p2 - pj a2a2 + (p3 - pj a3a3 + ... + (pv - pi) avav = 0. Accordingly there is a relation between a2, a3... av, which has been proved to be impossible. Hence there is only one latent point corresponding to each latent root, when the latent roots are all unequal.
256 MATRICES. [CHAP. VI. 146. Semi-Latent Regions. (1) Let the region denned by fju latent points of a matrix with unequal latent roots be called a semi-latent region of the (fju l)th species. The number indicating the species of a semi-latent region is thus equal to the dimensions of the region when all the latent roots are unequal. (2) Let e1} e2... ev be the v latent points of a matrix with unequal latent roots p1} p2... pv. Then the region denned by elt e2... e^ (/x v) is a semi- latent region. The characteristic property of a semi-latent region is that if x be any element in the region, then j x is an element of the same region; for if x = %YeY + ... -f f^, then And if X be any regional element incident in the semi-latent region, then cf)X is a regional element incident in the same semi-latent region. In particular if L = \exe2... e^, then f L = \pYp2... p e* ...etL = pYp2. (3) It is also important to notice that x, where x is a point in the semi-latent region e2... e^, excluding ex. Thus x lies in a semi-latent region of the (p, 2)th species, whereas x lies in a semi- latent region of the (jju l)th species. 147. The Identical Equation. (1) If p1} p2... pv be the latent roots of a matrix, no two being equal and none vanishing, and if a1} a2...av be the corresponding latent points, then it follows from above that the matrix can be written in the form alya2...av (2) If f be the matrix, let f 2 denote the matrix fxf t 3 the matrix and so on. Also any point x can be written %YaY -I- 2a2 + ... + %vav. Hence f x = p^i^ + p2%2a2 + ... + pv%vav. Hence f)X pYx = (p2 px) %2a2 + ... + (pv p^ %vav. Again / ( f x - p ) = (f 2x - p^x = p2 (p2 - p ) g2a2 +...+/? (pv - p^ %va Hence f 2x - (/Oj + p2) f x + plR2x = (p3 - p2) (p3 - p + Proceeding in this way, we finally prove that (0 - Pi) (0 - Pa) ( - P.) = 0, whatever element # may be.
146 148] THE IDENTICAL EQUATION. 257 (3) The equation may be written that is, p - (Pl + p2 + ... pv) 4T-1 + ...(- l)* / 2... Pv = 0. This is called the identical equation satisfied by the matrix cj . A similar equation is satisfied by any matrix, though the above proof has only been given for the case when all the roots are unequal and none vanish. 148. The Latent Region of a repeated latent root. (1) In the case when the equation giving the latent roots has equal roots, assume that aj of the roots are p1} % are p2, ... a^ are ph, where px, p2 ... p^ are the jju distinct roots of the equation. Then i + o + + a* = v. (2) Then subsections (5) and (6) of 145 still hold, proving that at least one latent point corresponds to each distinct root, and that the /u latent points which therefore certainly exist are independent. (3) Consider now the root p1 which occurs aY times, where a2 is greater than unity. Let elt e2... ev be v reference elements, and for brevity write p\Q\ M = e/, Pi^2 l e2 = e2, ... pYev f ev = ej. Then since p1 is a latent root of the matrix / , {e^e2 ... ej) = 0. Hence e/, e2 ... ej are not independent [cf. 145 (2)]. (4) Assume that v /3j of them and no more are independent, so that there are fti relations of the type K^ + K ' + +X*A/ = 0.....................(1), where a is an integer less than or equal to fa and equation (1) denotes the o-th relation of that type. Let a1 r = Xo-j^j + A,^ + ... + \ jVev. Then (Pl - f ) al T = \ale^ + ... + K,eJ = 0. Hence f ala = pYala. Thus corresponding to each relation of the type (1) existing between ei, e2'... ev\ there is a latent point, such as a1(r, corresponding to the root p1. Hence, since fa relations have been assumed to exist, there are fa latent points of the type a1 r. Furthermore all these points are independent. For if not, the relations of the type (1) are not independent. (5) The region of fa 1 dimensions defined by an, a12... alfii is such that if x be any point in it, f {x) = pxx. This region is therefore such that every point in it is a latent point, corresponding to the root px. Let it be called the latent region of the matrix corresponding to plt w. 17
258 MATRICES. [CHAP. VI. (6) The number ft cannot be greater than a2. For let On, al2...alPxy defining the latent region corresponding to ply be chosen to be ft of the reference elements el9 e2... ev. Thus let au = e1, a12 = e2 and so on. Let e9 stand for (p f )ea, ej for (pi f )ea. Then the equation (e^ ... ev) = 0, contains the factor (p p^)a\ But (p - f ) ala = (p- pi) ala = (p- pi) e*, when a ^ ft. Hence the equation becomes (p p^ fae*... eptefil+1 ...ev) = 0. Therefore ft : i. 149. The first species of semi-latent regions. (1) If/91 a1, then (exe2... epepi+1... ev) = 0, is satisfied by the root p1 which occurs aa ft times. Hence (exe2... ePlefp+1... ej) = 0. Thus the v elements el9 e2... e$o e'p+1... ej are not independent. It is known that the ft elements elt e.2 ... epx at least are independent. Assume that v y1 only are independent (v ^ ft). Then ^ relations hold of the type ^ 1^1 + K T2e2 -f ... + Kvjfih 4- (/^.Pj+i^+i + ... 4 /J rVeJ) = 0......(2), where a is put successively equal to 1, 2 ... y1. Since e1? e2... ePl are independent, in each relation of type (2), all the coefficients fiffip1+i ... fiO]V cannot vanish, nor can it be possible to eliminate all the elements e'pi+1 ... ej between these relations and thus to find a relation between e1 ... ePl. Thus if we assume yx elements of the type bla = /fcir.fc+i^+i + ... 4 /Um then these elements of the type b1 r are mutually independent, and are also independent of ely e2... epr Also (Pi ~ ^ ) ha = /xa ^1+1e^l+14 ... 4 /^(r^/ = - ^i^i - ... - ^/i^j. The coefficients tcal ... /c r^] cannot all vanish: for otherwise blir would belong to the latent region corresponding to plt which by supposition is only of ft 1 dimensions. Let a'1(T stand for KQ1ex 4 ... K.o^e^x, then a\a is a point in the latent region corresponding to pY: and / 61 r = pAa 4 a'1 r. (2) Thus y1 independent elements, 6n,... 61Yi, satisfying an equation of this type [cf. 146 (3)] have been proved to exist, defining a region of yl 1 dimensions. Also by the same reasoning as in 148 (6) it is proved that ft4 7i ai- (3) The ft independent points of the latent region of the type aYa corresponding to the root px and the y1 points of the type bla just found, together define a region of ft 4 yx 1 dimensions, which will be called the
149, 150] THE FIRST SPECIES OF SEMI-LATENT REGIONS. 259 semi-latent region of the first species corresponding to the root px. This definition is in harmony with the definition of semi-latent regions given in 146 for the case where all the latent roots are unequal. For let x be any point in this semi-latent region, then f x is another point in the same region; let X be any regional element incident in this region, then cf X is a regional element incident in the same region. And if L be a regional element denoting the semi-latent region itself, then f L p/^y^L. Also we can write, px = pxx + y, where y belongs to the latent region (that is, to the semi-latent region of the zero species). It should be noticed that by definition the semi-latent region of the first species corresponding to any given repeated root contains the latent region corresponding to that root. (4) The region defined by the points of typical form bla in subsection (1) is contained within the region defined by epi+1 ... ev\ while the latent region is defined by elt e2 ... e^. Hence the region defined by the points M r = l,2...7l) does not intersect the latent region. But from subsection (1), f bliT = pjbla + a\a, where a'ltr lies in the latent region. Now it can be proved that the 7l points a'1 T( r= 1, 2 ... 7l) are independent. For if a relation of the type, %% '** = 0, holds between them, then by writing ( j px) b1 T for a'1(T, we have 2 r(0-/ i) r = O, that is (4 -p1)XgJh,r=O. Hence the point %%J)1O lies in the latent region, and therefore the region defined by 6lo.(cr=l, 2 ...71) must intersect the latent region, contrary to what has been proved above. Hence the 7l points a'1 r( x = 1, 2 ... 7l) are independent. But they all lie in the latent region which is defined by ft points. Hence ft = 7l. 150. The higher species of semi-latent regions. (1) Semi-latent regions of the second and of higher species can be successively deduced by an application of the same reasoning as that of 149 (1). Thus to deduce, when ft + 7i oti, the semi-latent region of the second species, corresponding to the repeated root plf take as before ft of the refer- ence elements, namely elt e2...eplt in the latent region, which is assumed to be of -1 dimensions, and take 7l of the reference elements, namely in the semi-latent region of the first species (but not in the latent region), so that the ft + 7l reference elements thus assumed define the complete semi-latent region of the first species. Then, if a ^ ft, pea = p\eo, and hence (p- f )e r = (p-p1)e(r. Also, if oft and ^ft + 7l, ^ r = /o1^ + a r, where aa lies in the latent region and is therefore dependent on elf e2... ePl. Hence (p - f))e r = (p pY) ea - a*. 17 2
260 MATRICES. [CHAP. VI. Thus {(p - f ) el. (j - / ) e2... (p - / ) efil+yi] =(p- Pif1+y* e,e2... ^1+Yl. v Hence the equation for the latent roots, namely, II f(p ^ )e(r} =0, can 7 = 1 V be written (p - p^1"1"* [^3... ty1+Yl II {(p - ) e*}] = 0. o- = /31+Vl+l But the equation for the latent roots has by hypothesis the factor (p piYl, where fa -f yx alf thus the expression V [e1e2...ePl+yi U {(p - f ) ea}] o-=/3i+yi+i contains the factor (p pi)ai~^~Vl. Thus writing p1 for p we see that the v points eu e2... epl+yi, (p1- p)e^+yi+i,..., {pl-^ )ev are not independent. Assume that v Sx only are independent (i S1 fa + 71), so that there are c^ independent relations of the type ~ W^i+Yi+i (Pi - $) e^+y.+i + + P*" (Pi ~ / ) ^} = 5 where a is put successively 1, 2 ... Si. All the fis cannot vanish simul- taneously in any typical relation; and all the terms of the type (pi ~ t ) er (t fa + yO cannot be simultaneously eliminated between the Bx relations, so as to leave a relation between the independent elements e1, e2... epl+yr Now assume cla = /A r^i+Yi+1^I+Yl+1 + ... + fivve¥. Also note that the point Koxex + KiT2e2 + ... -f /ta^1+yie/3l+Yi (= b1 r say) lies in the semi-latent region of the first species. Hence the above typical relation takes the form 4 cla = piC1 r + bla. (2) Also by the same reasoning as in 148 (6), it follows that (3) Also by the same reasoning as in 149 (4) it follows that the region defined by the S1 points c1 r( r = 1, 2 ... S2) does not intersect the semi-latent region of the first species. Also, as before, the S2 points of the type bla are independent and the subregion defined by them (lying in the semi-latent region of the first species) does not intersect the latent region ; for otherwise some point of the type 2fiOc1 r lies in the semi-latent region of the first order, contrary to the assumption that this semi-latent region is only of the fa + 7i ~ 1 dimensions. Thus S1 ^ 7^ (4) If fa = aly then only a latent region exists corresponding to the repeated root ax and no semi-latent region. If fa ax and fa + yx aly then no semi-latent region of a species higher than the first exists. If fa -h 7X al7 and fa + 71 + Sx = au then no semi-latent region of a species higher than the
151, 152] THE HIGHER SPECIES OF SEMI-LATENT REGIONS. 261 second exists. If /3X + ^ -f 8j a1, then by similar reasoning a semi-latent region of the third species exists, and so on till ax independent points have been introduced defining the complete series of semi-latent regions corre- sponding to the root pY. Also from subsection (3) and from 149 (4) it follows that if aY fi^ly where fM is an integer, then in addition to the latent region at least fi semi-latent regions of the successive species must exist*. (5) It follows from (3) and 149 (4) that a matrix can always be written thus KT) Uifj , Ciff , ... where only those typical terms are exhibited which correspond to the latent root px. 151. The Identical Equation. (1) Suppose that the number of different groups of points of the types altr, bla, c1 r, and so on corresponding to a latent root p is t1b Then (4 - fh) a*r = o, ( / - Ply b1 r = o, ((/ - Ply clv = o, and if pla be a point in the Txth group, that is in the semi-latent region of the (tj l)th species (but not in that of the (t 2)th species), then ( / pi)Tlp1(r = 0. Let the region defined by all these points be called the semi-latent region of the matrix corresponding to plt (2) Now all the points of the different types thus found, corresponding to all the latent roots, are independent, and may be taken as a reference system. Hence if t2, t3 ... r^ be the corresponding numbers relating to the other latent roots, and x be any point, then ( - ptf ( /, - pj* ...( /,- prf* x = 0. Thus any matrix satisfies the identical equation (0 - Plr ( t - p,r.. (^ - pj =o- (3) Since tx aly t2 a2... t^ aM, it follows that any matrix satisfies the equation (* - ,0* (^-p,)"... ($-?,)"*= ). Thus the equation of 147 is proved for the case of equal roots. But in this case the matrix satisfies an equation of an order lower than the yth. 152. The Vacuity of a Matrix. (1) A null space [cf. 144] can only exist if a matrix has a zero latent root. The null space, or null region, is the latent region corresponding to the zero latent root. * This theorem does not seem to have been noticed before : nor do I think that the relations yx jgj, 52 ^ 7X, etc. have been previously explicitly stated.
262 MATRICES. [CHAP. VI. (2) If the zero latent root occur a times, then the matrix is said to be of vacuity a. Thus by definition the vacuity of a matrix is not less than its nullity. Let the semi-latent regions corresponding to the zero root be called also the vacuous regions of the matrix. Thus if b be a point in a vacuous region of the first species, c/ b = z, where a is a point in the null region; also if c be a point in the vacuous region of the second species, f c = b, where b is a point in the vacuous region of the first species; and so on. (3) Assume that 8 independent points, and no more, can be found in the vacuous regions of the first species defining a subregion which does not intersect the null region. Let dlf d2... ds, be these points, and let the /3 points bi, b2... bp define the null region. Then any point x in the vacuous region of the first species can be written Xf;d + 2?/ . Also by 150(3), S /3; and we may assume consistently with the previous assumptioos, f dx = AA, f d2 = \2b2 ... f d = Hence f x = (j X^d + j %r)b = Xgcfrd = 2 P=i Thus any point in the vacuous region of the first species is transformed into a point in the subregion of the null region defined by bly b.2 ... b8. Call this subregion the subregion of the null region associated with the vacuous region of the first species. 153. Symmetrical Matrices*. (1) In general, if x and y be any two elements and j any matrix, (x \ f y) is not equal to {y | / #). In order to obtain the conditions which must hold for these expressions to be equal, let the matrix be 2jL1^ -, where, according to the notation #1 #2 v of 141 (4), ap = altfix + c^fa + ... + avpev. In other words the matrix is ( an, o^, ... alv ). Then, supposing that el7 e2, ... ev are a set of normal elements at unit normal intensities [cf. 109 (3) and 110 (1)], Op I $eo) = Op I r) = pa (ep | ep) = a^, and (ea \ f ep) = (eo-1 ap) = au-p (e^ | ea) a^. Hence, if the required condition holds, a^ = aap. (2) Thus the matrix with the desired property is a matrix symmetrical about its leading diagonal when the elements of the denominator form a * Symmetrical matrices are considered by Grassmann [cf. Ausdehnungslehre von 1862, 391]; but his use of supplements implicitly implies a purely imaginary, self-normal quadric. Hence his conclusions are limited to those of subsection (7).
153] SYMMETRICAL MATRICES. 263 normal system (at unit normal iotensities) with respect to the quadric chosen as the self-normal quadric. Let such matrices be called symmetrical with respect to the normal systems, or, more shortly, symmetrical matrices. (3) If fi out of the v latent roots of a symmetrical matrix be distinct and not zero, so that at least jjl points clf c2, ... c^, can be found with the property 4 cP==rYpcP then the ft points c1? c2)... Cy, corresponding to different latent roots 7i. 72, 7/* are mutually normal. For let x = f^ + a + ... + f^, and y = tj^ + r}2c2 + ... + v^- Then (y \ j x) = (ZVc \ tfyc) = 2 (^pVvyP + ^P7 r) (cP |ca), and (x j j y) = (2?c | 2t?7c) = 2 (1^7* + *?p7p) Op ! c*)- Hence (y | / #) = (x \ f y) gives 2 {% - f^p) (7p - 7 r) (cp | ca) = 0. Now let all the ^s, except fp, and all the ^'s, except rjay vanish ; and it follows that (cp | Co) = 0. Hence c1? c2... c^ are mutually normal. (4) Let c/, c/', etc., be other points in the latent region of the root 7lJ so that ^ Ci' = 71c1', etc.: then the same proof shows that c/ is normal to all of c2, ... c^, and so on. Hence the latent region corresponding to 7l is normal to the latent region corresponding to p2, and so on. (5) In the same way it can be proved that the whole semi-latent region corresponding to any latent root 7l is normal to the whole semi-latent region corresponding to any other latent root 7.2. For let c?j be any point in the semi-latent region of 7l of the first species. Then / dj = 7lc^ + X^, / c2 = 72c2. Hence (c21 (jxh) = 7l (c21^), by (3) and (4). Also (dx | / c2) = 72 (dx | Ca) = 72 (c21 d,). But (c2\4 d1) = (d1\(j)C2)7 by hypothesis. Hence (7l y2)(c2 [d^ = 0, and 71+72 by hypothesis. Therefore (c2 |cZ1) = O. Hence the semi-latent region of the first species corresponding to 7l is normal to the latent region correspond- ing to y2. Similarly the semi-latent region of the first species corresponding to 72 is normal to the latent region corresponding to 7l. Again h and d2 lying respectively in the semi-latent regions of the first species corresponding respectively to 7l and to 7a are normal to each other. For (d, | f d2) = (d, | (y2d2 + X,c2)) = 72 (d, I d2), and (d2 \0^) = 7l (d2 \d,). Thus (d11 / d2) = (d21 $di) gives (7l - 72) (d11 d2) = 0 ; and hence (^ | ci2) = 0. Similarly if fx be another point in the semi-latent region of the second species of the root 7l, such that f fi= 7i/i + Midi, then the same proof shows thatyi is normal to c2, d2 and/2; and so on. Hence the semi-latent regions of different roots are mutually normal.
264 MATRICES. [CHAP. VI. (6) Again consider the equation This becomes y1 (d \di) + Xj (c2 jd) = 7i(d l^i)- Hence Xx (d d) = 0. Thus if d does not lie on the self-normal quadric, Xa = 0. Now suppose that the latent region defined by cly d' d" -- does not touch the self-normal quadric. Then it is always possible in an infinite number of ways to choose cl9 d', c/'... to be mutually normal and none of them self- normal. Also the most general form for d1 is such that f)d1 = 7^ + Xxd + X/d' ~H__ Then (d 14 di) = 7i (d I ^i) + ^i (ci | d) = (^i | / d) = 7i (d I ^i)- Hence Xx = 0, similarly X/ = 0, X/' = 0, and so on. Hence dY lies in the latent region, and no semi-latent regions of the first or higher species exist corresponding to the root 7^ (7) It is a well-known proposition that the roots of the equation are all real; provided that apa = xapi where all the quantities apv are real. Hence it follows that the latent regions of symmetrical matrices are all real. For if yx be one of the real roots, the equation f x = y-iX, determines the ratios of the co-ordinates of x by real linear equations. If the self-normal quadric be imaginary owing to all the normal intensities being real [cf. 110 (3)], a latent region, being real, cannot touch it. Hence in this case there can be no latent self-normal point, such that f cp= ypcp. Hence from above there are no semi-latent regions. Thus finally in this case a complete real normal system of the type cly c(, d"... c2, c2'..., c3..., ch, c/, c/'... can be found defining the latent regions of 7^ y2i etc.; each element being at unit normal intensity. (8) If the latent region defined by c1} d' c/'... touches the self-normal quadric (assumed real), but is not part of a generating region, take d to be the point of contact, and take d' Ci" to be mutually normal elements on the tangent plane at clt but not self-normal [cf. 113 (5)]. Then the general form of d1 is such that, f d1 = 7^ + Xid + X/c/ +__ Hence (d | M) = 7i (ci I ^1) = (rfi I #0 = 7i (ci I di)- Also (d' 4 dd = 7i (Ci I *) + V (d' IO = (^ I fa') = 7i W I* Hence X/ = 0, similarly \" = 0, and so on. Thus d1 is such that f d1 =7^ + X2d-
154] SYMMETRICAL MATRICES. 265 There can only be one independent point d1 satisfying this equation. For if di be another point such that f)di =y1d1/ + ^0^ then it was proved in 149 (4) that if cly dly dxr are independent, then Xjd is independent of X/d, whereas here they are the same point; which is impossible. (9) If the latent region of the root y2 contain a real generating region of p 1 dimensions of the self-normal quadric, let the points cn c12,... clp be chosen to be mutually normal points in this generating region [cf. 79 and 80], and let the remaining points of the latent region be mutually normal and normal to cu ... clp, but not self-normal. Let these remaining points be Ci} Ci Ci .... Let d1 be any point in the semi-latent region of the first species, but not in the latent region. K=p Then j dY = y^ + 2 X^ + /^d + /^V + .... K = l Hence (cx | di) = yx (d j dY) + /^ (d j d) = (^11 fa) = ji (d | dx). Hence ^ = 0. Similarly fi-l = 0, /-i/' = 0, and so on. K=p Hence (f d1 = y^ + S XKclK. *c = l Thus in the semi-latent region of the first species the subregion of highest dimensions not necessarily intersecting the latent region cannot be of higher dimensions than the real generating region contained in the latent region [cf. 149 (4)]. Similarly for the semi-latent regions of higher species. 154. Symmetrical Matrices and Supplements. (1) A one to one correspondence of points to planes is given by the operation which transforms the reference elements e1} e2... ev into the planes A1} A2... Av, where [cf. 97 Prop. IV.] Ax = oluEj + a2lE2 + aslE3 + ... + avlEVi and so on. Then the element x (= e) is transformed into the plane X (= 2J^4). Now let e1} e2 ... ev be a normal system, so that Ex = \e1} and so on; then A1 = |(and + Oaea +...) = l i say. Similarly A2 = \{al2ex + a22e2 + ...)= \a2, and so on. Also let f denote the matrix -1'- 2---" v. 6i, 62 ... 6V Then the type of one to one correspondence of points to planes, which we have been considering, can be denoted by X = \ cj x. Similarly this type of correspondence could be denoted by f)\x; but \cf and $| are in general different operations. (2) If to every point there corresponds a plane and to every plane there corresponds a point, then the matrix 6 has no vacuity. In this case 6lf e2'"lv alfa2...au
266 MATRICES. [CHAP. VI. is a determinate matrix; denote it by c^"1. Then if X = \ j x, \X $x, arid In general the transformations different from and ^r1] are different: thus \ j x is x. (3) If the latent roots of f are all unequal, then the operations \ f and 0"1! can only be identical when the v latent points of (j form a normal system, that is, when the matrix is symmetrical; and when, in addition, the product of the latent roots of the matrix is unity. For let d, c2... cv be the latent points of , so that f can be written that is Then | c1 = 71|c1. Hence | c/ d = c/)"1 I d, becomes 711 cx = / ~ cv C2 = cxcz... cv, ..., C^ = c^ ... ? |cx = \1G1 + A^Og 4-... + \VCV. 1 4- y1\2 ] G2 + ... + ^ etc. Let Gx = c.2c3. Assume that Then But by 141 (1) (/ (?! = 7^3... ry^d, Hence 7^ | d = 7^2... yv\G1 + 7^73 = c1 = \1C1+\2C2+...+\VCV. Hence since 7^ y2...yv are all unequal,7^ ... 7^=1, A2 = 0, X3=0,...\I/ = 0. Thus \c1 = \1C1] and similarly for |c2, |c3, etc. Accordingly the latent points of the matrix form a normal system, and the product of the latent roots is unity. (4) Conversely if the matrix be a symmetrical matrix with unequal latent roots of which the product is unity, then \ f and ^^l are the same operations. For let d, c2... cv be the latent points, y1, y2 ... yv the latent roots of cf . Then cx, c2 ... cv are the latent points and 7^, y2~* ... y^1 are the latent roots of cf)"1. Also ! / d = 7ilciJ an(i 0| / ^i = 7i0Ici = 7i72--- 7^lci = lci- Hence | / Ci = cfr^1 |d- Similarly for the other latent points. Thus finally \$x = fy-1 \x. (5) It is obvious that in this case the operation \ f is equivalent to the operation of taking the supplements with respect to some quadric with respect to which cl9c2...cv form a self-normal system. Let I denote this operation; let e1? e2 ... ev be the normal intensities of d c2... cv with respect to this operation; and let B1} S2...SV denote the normal intensities of d, c2... cv with respect to the operation |. Also put
155] SYMMETRICAL MATRICES AND SUPPLEMENTS. 267 A A7 X ildi I l/j r\ - v 2v3 l^y j -*-^l o ^-/2^'3 ^v But Id = | / d = 7i | i = 5\ 2 A ' ^ 2 A ' Thus ^ = -^ . Similarly 72 = -^ -r- , and so on. Hence y^... 7,, = 1 = kv_2 , therefore A = A'. Hence y1 = ~} y2= ~ , ... yv = -^ . Thus the symmetrical matrix / , with unequal roots of product unity, has been expressed in the form |I; so that tf x = \Ix. The latent points of the matrix are the one common system of self- normal points of the two self-normal quadrics corresponding to | and I; and the relations between the latent roots and normal intensities are given above. 155. Skew Matrices. (1) The matrix -1-' ^-^ (= ) has important properties in the special case when ax = * + a21e2 + asle3 + ... + avlevy etc., where a12 + or2i = 0, ..., p r + aP = 0 ..., and e1} e2... ev form a normal system at unit normal intensities. Let such a matrix be called a skew matrix. Then (ep \ j ep) = 0, (eo \ (j ep) = aap = - apa = - (ep \ $ea). Thus (x \ f x) = 0 (A), and (x \ f y) + (y \$x) = 0 (B)3 whatever points x and y may be. (2) Any latent point cx of this matrix, such that cf)c1 = y1c1) where y1 is not zero, is self-normal. For from equation (A) (d | / d) = 7i (d \cx) = 0. Again, putting cx and c2 for x and y in equation (B), where c2 is another latent point, Hence either 71 + 72 = 0, or (d | c2) = 0. (3) Assume that there are no repeated roots. The self-normal quadric contains generating regions of dimensions -~1 or -------1, according as v Zt Zi be even or odd (cf. 79). If v be even, mutually normal elements jlt j2, ... jv can be found on the ^ " 2 quadric, defining one generating region, and k1? k2 ... kv mutually normal
268 MATRICES. [CHAP. VI. elements defining another generating region. Also any element such as jp can be made normal to all the k's, except kp, and conversely kp is normal to all the/s, except jp (cf. 80). Then ji,j2-- jv k1)k2...kv can be chosen as the latent points of the 2 2 matrix. If yly y2 ... yv be the latent roots corresponding to jlfj2 ...jv, then 2 2 by subsection (2) 71, y2 Ju are the latent roots corresponding to 2 A/i *^/* #t'|/ 2 Hence if x = jj + 2h + + %vj* + vA + vJh + + Vrfa then / # = 7^! jj + 72 2 j2 + ... - jiVih Thus (^\4^) = (jiSiVi-"YiSiVi)(ji \k)+ -.. =0. (4) If 1/ be odd, - mutually normal elements of the type jp can be found, and of the type kp, and an element e, not on the quadric, normal to all the /s and all the k's. Let these be the latent points of the matrix, then the element e must be a null point of the matrix. If x = e + X%j + 2?7 , then f x = ^y^j ^yr\k ; and {x \ f x) = 0. (5) Assume that there are repeated roots. Let the roots 72 and y2 be both repeated, and neither zero. Let dx and d2 be in the semi-latent regions of the first species (and not in the latent regions) corresponding to y1 and y2 respectively, fY and f2 in the semi-latent regions of the second species (and not in the semi-latent regions of the first species), and so on. Let clf c/,... be in the latent region of 7^ and c2, c2, ... in that of y2. Then we may assume [cf. 150 (5)], / Ci = 7A M = Jidi + \c1} f fx = yYfx + /Mi, and so on. Hence by equation (B), (c2 |0c/) + (c/ 1^) = 27j (d |c/) = 0. Hence c2 and c/ are mutually normal as well as being self-normal. Thus the latent region of a repeated root is a subregion of some generating region of the self-normal quadric. (6) Again from equation (B), (c/ 10^) + (di |0c/) = 272 (c/ | rfx) = 0. Hence the semi-latent region of the first species corresponding to 7! is normal to the latent region corresponding to ylm Also by equation (A), (d2 l^c^) = 7! (dj |c?j) = 0. Hence {d1 |di) = 0. Therefore each point in the semi-latent region of the first species is self-normal. Further if d/ be another point in this semi- latent region, = 0.
155] SKEW MATRICES. 269 Thus (d1\d1') = O. Hence the semi-latent region of the first species is a subregion of a generating region of the self-normal quadric; and therefore the latent region and semi-latent region of the first species are together contained in the same generating region. (7) The same proof applies to semi-latent regions of higher species. Hence the complete semi-latent region (which contains the latent region) corresponding to a repeated root is a subregion of a generating region of the self-normal quadric. (8) The same proof shows that the complete semi-latent region of one repeated root y1 is normal to the complete semi-latent region of another repeated root y2 unless y1 + y2 = 0. (9) Again assume that the matrix is of vacuity a and of nullity /3. Let c be any null point of the matrix, and cx any latent point corresponding to the non-vanishing root 7l. Then f c = 0, hence (ex \ f c) = 0. Thus by equation (B) (c, \ f c) + (c \cfyc,) = 7l (c |d) = 0. Hence the null region is normal to the latent regions of all the other latent roots. Similarly the null region can be proved to be normal to all the semi- latent regions of the other latent roots. (10) Let d be a point in the vacuous region of the first species: assume f d = \c, where c is a null point. Then (c | j d) + (d \ f c) = \ (c \ c) = 0, by equation (B). Hence either X = 0, and d is in the null space; or (c |c) = 0, that is to say, c is self-normal. Hence the subregion of the null region associated with the vacuous region of the first species is self-normal. Also from equation (A), (d | f d) = \ (c \ d) =0. Hence, assuming that X is not zero, (c \d) = 0, that is to say, d is normal to c. Again let c' be any other null point, then (c' | f d) + (d \ / c') = X (c \ cr) = 0. Hence, assuming X4=0, c is normal to every other null point. (11) Similar theorems apply to vacuous regions of higher species.
BOOK V. EXTENSIVE MANIFOLDS OF THREE DIMENSIONS.
CHAPTER I. Systems of Forces. 156. Non-metrical Theory of Forces. (1) The general theory of extensive manifolds, apart from the additional specification of the Theory of Metrics, has received very little attention. It is proposed here to investigate the properties of Extensive Manifolds of three dimensions, thereby on the one hand illustrating the development of one type of formulae of the Calculus of Extension, and on the other hand discussing properties which are important from their connection with Geometry*. (2) Since in this case four independent points define the complete region the simple extensive magnitudes are only of three orders, the point, the linear element, the planar element. Also the only complex extensive magnitudes are systems of linear elements. A linear element, in that (a) it is an intensity associated with a straight line, (y8) it is directed along the line, so as to be capable of two opposite senses, (7) it is to be combined with other linear elements on the same line by a mere addition of the intensities [cf. 95 (1)], has so far identical properties with a force acting on a rigid body. Only in an extensive manifold no metrical ideas with respect to distance have been introduced. The other properties of a linear element, whereby it is defined by two points and is combined with other linear elements on other lines form a generalization of the properties of a force so as to avoid the introduction of any notion of distance. It will be noticed that the theorem respecting the combination of Forces known as Leibnitz's theorem expresses the aspect of the properties of forces which are here generalized. The parallelogram of forces is without meaning at this stage of our investigations: for the idea of a parallelogram depends on the Euclidean (or equivalent) axioms concerning parallel lines, and such axioms presuppose metrical conceptions with respect to distance which have not yet been enunciated. * The formula and proofs of propositions in this book are, I believe, new. Many of the propositions are well-known; but I believe that they have hitherto been obtained in connection with Metrical Geometry, either Euclidean or non-Euclidean. w, 18
274 SYSTEMS OF FORCES. [CHAP. I. (3) We shall therefore use the term force as equivalent to linear element, meaning by it the generalized conception here developed apart from metrical considerations. It will be found that very few of the geometrical properties of ordinary mechanical forces are lost by this generalization. Also, when no confusion will arise, plane will be used for planar element. The context will always shew the exact meaning of the term. 157. Recapitulation of Formulae. (1) It will be useful to re- capitulate the leading formulae of the Calculus of Extension in the shape in which they appear, when the complete manifold is of three dimensions. (2) The product of four points is merely numerical. The product of a linear element and planar element is the point of intersection of the line and plane. The product of two planar elements is a linear element in the line of intersection of the two planes. Thus a linear element can be conceived either as the product of two points or as the product of two planar elements. The product of three planar elements is a point. The product of three points a planar element. The product of a linear element and a point is a planar element. The product of two linear elements is merely numerical. (3) The formulas for regressive multiplication are [cf. 103 (3) and (4)] abc ,de=de. abc = (abce) d (abed) e = (abde) c + (cade) b + (bede) a.. .(1). Thus five points ay b, c, dy e are connected by the equation (bede) a - (acde) b + (abde) c - (abce) d + (abed) e = 0........(2). Again abc. def= (abef) de + (abed) ef+ (abce)fd = (adef) be + (bdef) ca + (cdef) ab = - def. abc ..... (3). By taking supplements, we deduce that these formulae still hold when planar elements A, By C, D, Ey F are substituted for the points a, 6, c, d, etf. (4) Also from 105 there come the group of formulae, Bly B2y B3, B4being planar elements, (a1a2. BXB2) = (aA) (aA) - (aA), . BXB2BZ) = , ( A) (aA), (aA), (aA), ( A), (aA), (aA), (aA), (aA), (aA), .(6).
157, 158] RECAPITULATION OF FORMULA, 275 (5) Also from equation (4) a useful formula may be deduced by putting = bcclt B2 = bcc2. Then from equation (4) axa2. (bcCi) (bcc2) = {axbcc^ (a2bcc2) - (aficc.) (ajbccj). But from 102, (bcc^) (bcc2) = (bcc^) be. Therefore a^. (bcCi) (bcc2) = (a^jbc) (bcc^). Hence finally, (a^a^c) (bec^) = (aficc^) (a2bcc2) (aj)cc2) {a2bccx) ......(7). This equation can be written in another form by putting F for the force Then Clc2F) = (alClF) (a2c2F) - (a F) (a F)............(7'). be. 158. Inner Multiplication. (1) If a be any point, then \a is a planar element; and if A be any planar element, then | A is a point. If F be a simple linear element, then \F is a simple linear element; and if S be a system of linear elements, then | S is a system of linear elements. .............(8). ..............(9). (2) Also (cf. 118), and hence .(10). .(11). \(abc \de) = \\de. \abc =(de \abc), \ \(de \abc) = \\(abc \de)= (abc\de)) .......'" Also \(abc\d) = \abc. d = (d\abc),} and hence \(d\abc)= \\(abc \d) = (abc \d)\............... Also \(ab\c) = -\ab.c = -(c\ab)i\ hence |(c|a6)=-||(o6|c) = (a6|c) j ............... Finally (a 16) = (b \ a), and (ab | cd) = (cd \ ab),\ and (abc\def)=(def\abc) J ..............( ' (3) Again from the extended rule of the middle factor (cf. 119), abc \de = (ab \de) c + (be \de) a + (ca \de) b. ) And de\abc=\(abc \de) = (de\bc) \a + (de \ca)\b+(ab\de)\c) .(13). Again abc \d = (a\d)be + (b\d)ca + (c\d) ab. \ And d \abc = \(abc \d)=(d\a) \bc + (d \b) \ca + (d\c)\ah\ Again ab|c = (a c)b (b\c)a. c\ab = -\(ab\c) = (c \b)\a-(c \a) \b (14). And (4) .(15). Again from 120, (ab\cd) = (a\c)(b]d)-(a\d)(b\c)............(10), (abc\de/)= (o|d), (a \e), (a \f) (b\d), (b\e), (b\f) (e\d), (c\e), (c\f) (a\e), (a|/), (a\g), (a\h), (b\e), (b\f),(b\g), (b\h), (c \e), (c |/), (c \g), (c \h), (d\e), (d\f),(d\g), (d\h), (17), (abed | efgh) = .(18). 18 2
276 SYSTEMS OF FORCES. [CHAP. I. (5) It is unnecessary to reproduce the special forms of the more general but less useful formulae in 122. These eighteen formulae of the present and the preceding articles are the fundamental formulae which will be appealed to as known. They are all immediate consequences either of the extended rule of the middle factor or of the formula of 105. 159. Elementary Properties of a Single Force. (1) A force can be represented as a product of any two points in its line. This is a simple corollary of 95. (2) A system of forces lying in one plane is equivalent to a single force. This is a corollary of 97, Prop. IV. (3) A force can be resolved into the sum of two forces on lines concurrent with it and coplanar with it. For let a be the point of concurrence, then ab can be chosen to represent the given force. Two points c and d can be found on the other lines respectively, such that b = \c + fid. Hence ab = Xac + pad. Thus ab is resolved as required. (4) Any force can be resolved into the sum of two forces, of which one passes through a given point and one lies in a given plane, which does not contain the point. For consider the plane P through the given force and the given point. It cuts the given plane in a line concurrent with the force, and through the point of concurrence a line can be drawn in P through the given point: then two forces can be found by (3) along these lines of which the sum is equivalent to the given force. Thus if a be any given point, A any given plane, F any given force, then we can write, F=ap+AP. 160. Elementary Properties of Systems of Forces. (1) The letter S will only be used to denote a system of forces. Two congruent systems of forces (i.e. of the types S and \S) will be spoken of as the same system at different intensities. If Flt F2, etc. be any number of forces, then S = %F represents the most general type of system. (2) If a be any given point and A any given planar element not containing a, any system of forces (S) can be written 8 = ap + AP. For by 159 (4), F, = aPl + APlt F2 = ap2 + AP2 etc. Hence S = Fx + F2 + ... = a(p, + p2 + ...) + A (P, + P2 + ...) = ap + AP. Hence any system can always be represented by two forces of which one lies in a given plane, and one passes through a given point not lying in the plane,
159 162] ELEMENTARY PROPERTIES OF SYSTEMS OF FORCES. 277 (3) The mention of p and P can be avoided by means of the formula This can be proved as follows. From (2) of this article 8 = ap + AP. Multiplying by a, we have aS = a . AP = (aP) A (a A) P. Hence aS. A = - (aA) PA = (aA) AP. Again multiplying by A, we have AS = A .ap (Ap)a (Aa)p. Hence a. AS = (aA) ap. The required formula follows at once. (4) It follows from (2) that any system S can be expressed in the form S = ab + cd. For we may write cd instead of AP in the expression for S. It will be proved in 162 (2) that one of the two lines, say ah, can be assumed arbitrarily. (5) If ely e2, e3, e4 be any four independent elements, then [cf. 96 (1)] S can be written S = 7r12eie2 + Wvfi s + Wa^i + TiA^ + ttm^A + Ww^A Hence any system can be represented as six forces along the edges of any given tetrahedron. When ely e2y e3, e4 are unit reference elements, 7r12, etc. will be called the co-ordinates of the system S. 161. Condition for a Single Force. (1) If 8 be any system of forces, (SS) is not in general zero. For by 160 (4), S may be written ab + cd; hence (SS) = 2 (abed). Thus (SS) only vanishes when (abed) = 0, i.e. when ab and cd intersect. But in this case ab + cd can be combined into a single force. Thus (SS) = 0, is the required condition that S may reduce to a single force. (2) If 8=ap + AP, then (SS) = 2 (op . 4P) = 2 (ail) (pP) - 2 (aP)(pA). If S = TTvfi^ + 7723^2^3 + ^3^3^ + ^4^64 + TT'ii^i + ^34^4^ then (SS) = tt^tt^ + 7733^4 + 773,77-24. (3) If S reduce to a single force, \8 reduces to a single force. For if (SS) = 0, then |(SS) = 0, that is (\S \S) = 0. 162. Conjugate Lines. (1) When a system S is reduced to the sum of two forces ab and cd, then the lines ab and cd are called conjugate lines, and the forces ab and cd are called conjugate forces with respect to the system. Also ab will be called conjugate to cd, and vice versa.
278 SYSTEMS OF FORCES. [CHAP. t. (2) To prove that in general any line ab has one and only one conjugate with respect to any system S, not a single force. For if 8 = Xab + pcd, then S Xab is a single force. Hence {(8 - Xab) (S - Xab)} = 0; that is (88) - 2X (abS) = 0. Therefore A, = -= j-t-^ \ and hence 8 h / , v 6 represents the force conjugate to Xab. Since only one value of \ has been found, there is only one such force; and if (abS) be not zero, there is always one such force. Similarly if any line be symbolized by AB, its conjugate with respect to 8 a 1 (88) lsS"p (3) If two lines ab and cd intersect, their conjugates with respect to any system S intersect. For by multiplication 1 (88) h) ( 1 (88) ] __ 1 (SSy (abed) _ 2 (abS) j { 2 (cdS) } ~~ 4 (abS) (cdS) ' since by hypothesis (abed) = 0. 163. Null Lines, Planes and Points. (1) If L be any force, and (LS) = 0, then the line L is called a null line of the system S. Note that L can be written in the two forms ab and AB; the product (abS) is a pure progressive product; the product (ABS) is a pure regressive product. If F be any force, then (FS) is called the moment of 8 about the force jF. (2) The assemblage of null lines of any given system 8 will be called the linear complex* defined by the system 8. (3) If a be any point, then the planar element aS defines a plane containing a} which is called the null plane of the point a with respect to the system 8. If A be any plane, then the point AS lies in A and is called the null point of the plane A with respect to the system S. * Linear Complexes were first invented and studied by Pliicker, cf. Phil. Trans, vol. 155,1865, and his book Neue Geometrie des Raumes, 1868. The theory of Linear Complexes is developed in Clebsch and Lindemann's Vorlesungen iiber Geometrie, vol. 2, 1891; also (among other places) in Kcenig's La Geometrie Reglee, Paris, 1895, and in Dr Rudolf Sturm's Liniengeometrie9 3 vols., Leipzig, 1892, 1893, 1896. The chief advances in Line Geometry, since Pliicker, are due to Klein. Buchheim first pointed out the possibility of applying Grassmann's Ausdehnungslehre to the investigation of the Linear Complex, cf. On the Theory of Screws in Elliptic Space, Proc. of London Math. Soc. vols. xv, xvi, and xvn, 1884 and 1886.
163, 164] NULL LINES, PLANES AND POINTS. 279 164. Properties of Null Lines. (1) All the null lines of S which pass through any point a lie in the null plane of a; and conversely all the null lines which lie in any plane A pass through its null point. For if ab be any null line of S through a, then (abS) = 0 = (aS. 6). Hence b lies on the plane aS. Similarly if AB be any null line of S in A, then (ABS) = 0 = (AS. B). Hence B contains the point AS. (2) If a lie on the null plane of b, then b lies on the null plane of a. For (bS . a) = 0 = - (aS. b). It is obvious that in this case ab is a null line. (3) If any null line L of a system of forces intersect any line ab} it intersects its conjugate. For by hypothesis, (8L) = 0 = (abL). Hence Also obviously any line intersecting each of two conjugates is a null line. (4) The conjugates of all lines through a given point a lie in the null plane of a. For let ab be any line through a. Then the plane through a and the conjugate of ab is denned by a \8 ^ r~rJ\ ^\ It follows as a corollary that aS. bS represents the line conjugate to ab. For this conjugate lies in the line of intersection of the null planes of a and b. Thus 2 (abS) (5) If the system do not reduce to a single force, no two points have the same null plane and no two planes have the same null point. For if x and y be two points such that xS = yS, then putting x = y + z, zS = 0. Hence by 97, Prop. I., S = zp; and therefore S reduces to a single force, contrary to the assumption. Thus no two points with the same null plane exist. If X and Y be two planes with the same null point, then XS = YS. Hence by taking supplements \X \S = \Y\S. But \S is a system of forces, and hence the points \X and \Y cannot have the same null planes with regard to it unless \S reduce to a single force. Hence from 161 (3) X and Y cannot have the same null points with regard to S, unless S reduce to a single force.
280 SYSTEMS OF FORCES. [CHAP. I. (6) The relations between planes and their null points and between points and their null planes can be expressed in terms of ordinary algebraic equations involving their coordinates*. For let and X = X1e2e3e4l Then the equation, either of a plane through x, or of a point on X, is (xX) = (\ x + X2 2 + X3 4- \ ) teeAfc) = 0. Also [cf. 160 (5)] let be the system Then by simple multiplication ## = (a^ a13 2 Hence the co-ordinates \lf Xa, X3, X4 of the null plane of x can be written, f3 + * , where we assume a12 4- flfo = 0 = a13 + a31 = etc. Again by simple multiplication, we find XS = ( * + OnXg + asi^3 + 4i^4) (^^3^4) ex + etc. Hence the co-ordinates gl9 %2t f3, f4 of the null point of X are given by c'fi = * + a2i^2 + ol31\3 + a41X4, * + QC43X4 , 334X3 + * . (7) Thus, if the reference elements be normal points at unit normal intensities, a skew matrix [cf. 155] in a complete region of three dimensions operating on x can be symbolized by | xS. 165. Lines in Involution. (1) A system of forces can always be found so that five given lines are null lines with respect to it. But if six lines are null lines with respect to some system, their co-ordinates must satisfy a condition. For let Llt L2) L3i L4, L5, L6 be any six independent lines. Then [cf. 96 (2)] we may write any system 8, Assume that (^S) = 0 = (L2S) = (L3S) = (LJ3) = (LJ3). * Cf. Clebsch and Lindemann, Vorlesungen fiber Geometrie, vol. 11. pp. 41 et seq.
165, 166] LINES IN INVOLUTION. 281 Then the five ratios : 2 3 '- '- ?5 ' e are determined by the five equations * + (LJL,) +13 (ZA) + 4 (L^) + (AZ6) + f, (A J ) = 0.. Hence S is completely determined. Therefore one and only one system of forces can in general be found such that the five lines Llt L2...L5 are null lines with respect to it. (2) If L6 be also a null line with respect to the same system then eliminating , 2, etc. from the six equations of condition, we find (AA), (AA),... OW *, CW,... (LJL.) = 0; (Z.A), (Z.A), where it is to be noticed that (LxLa) = (3) Definition. Six lines which are null lines with respect to the same system are said to be in involution; and each is said to be in involution with respect to the other five. Thus the propositions of the preceding article can be stated thus : The lines through a given point in involution with five given lines lie in a plane, cf. 164 (1). The lines in a given plane in involution with five given lines are con- current, cf. 164 (1). Again, a linear complex may be conceived as defined by five independent lines belonging to it. 166. Reciprocal Systems. (1) Two systems of forces S and ' are said to be reciprocal* if (S8') = 0. It is obvious that a force on a null line of any system is a force reciprocal to the system. (2) If two systems be reciprocal, the null lines of one system taken in pairs are conjugates with respect to the other system. For let S and fif be the two systems. Then (S8') = 0. Let ab be a null line of 8, its conjugate with respect to 8' is 8' , , ^J ab. * Reciprocal systems of mechanical forces were first studied by Sir R. S. Ball, cf. Transactions of the Royal Irish Academy, 1871 and 1874, vol. 25, and Phil. Trans. (London), vol. 164, 1874, and his book Theory of Screws (1876), ch. in. The theory of systems of forces for non-Euclidean Geometry was first worked out by Lindemann in his classical memoir, Mechanik bei Projectiven Maasbestimmung, Math. Annal. vol. vn, 1873. The most complete presentment of Sir R. S. Ball's Theory of Screws is given by H. Gravelius, Theoretische Mechanik, Berlin, 1889.
282 SYSTEMS OF FORCES. [CHAP. I. But Hence the conjugate of ab with respect to 8' is a null line of S. It is to be noted that there are conjugates of either system which are not null lines of the other. 167. Formulae for Systems of Forges. (1) The following formula are obvious extensions of the standard formulae of 157, remembering the distributive law of multiplication. From equation (I), 157, abc.S = S.abc = (abS) c + (caS) b + (bcS) a.) Also Sc. de = (See) d (Scd) e J ............ From equation (3), abc. dS = (ad8) be + (bdS) ca + (cdS) ab .........(20). By taking supplements, and replacing \S by 8, we see that the formulae hold when planar elements replace the points. (2) To prove that, if a be any point and S any system of forces SS)A) ....................K h For let 8 = bc+de. Then 8. aS = de. abc + be . ade = - (abed) e + (abce) d - (abde) c + (acde) b = (bcde)a; from 157, equation (2). Also (SS) = 2 (bede). Hence 8. aS = (SS) a. The second formula follows by taking supplements. (3) To prove that a8.bS=(ab8)8-l{88)ab9\ AS.BS = {ABS)S-1 (SS)AB\ .....................v '' For let 8 = Xab + cd. Then a8.b8 = acd. bed = (abed)cd. But (abed) = (abS), and cd = 8- Hence a8. b8 = (abS) S - J ( S) ab. This forms another proof of the corollary to 164 (4). (4) From equations (21) and (22) it is easily proved that aS . bS . cS = i (SS) {(bcS) a + (caS) b + (abS) e) = Also from equation (22), aS .bS.S= (SS) (abS),\
167] FORMULAE FOR SYSTEMS OF FORCES. 283 (5) To prove that if a be any point and 8 and 8' any two systems of forces, then 8. aS' +S'. aS=(SS')a,\ ]........................{ ' For in equations (21) write 8+8' instead of 8. Then (S + S').a(S + ') = \ {(S + S')(S + S')} a. Hence by multiplying out both sides, But 8. aS = \ {88) a, and 8'. aS' = | (#' ') a. Hence the required result. Similarly from equation (22) we can prove aS. bS' + aS'. bS = (abS) 8' + (abS') S-(SS') ab, \ AS. BS' + AS'. BS = (ABS) S' + (^5^) S - (S8T) 4 j ......{
CHAPTER II. Groups of Systems of Forces. 168. Specifications of a Group. (1) If SlyS2,...S6 be any six independent [cf. 96 (2)] systems of forces, then any system can be written in the form X^ + X-A 4- ... 4- \S6. Let X1} X2, ... \ be called the co-ordinates of S as referred to the six systems. Definitions. The assemblage of systems, found from the expression X^ + ^2^2 by giving the ratio \ : X2 all possible values, will be called a ' dual group' of systems. The assemblage of systems, found from the expression X^! + \J32 + Xg j by giving the ratios X2: X2: X3 all possible values, will be called a ' triple group ' of systems. The assemblage, found from X^ -f X2S2 + \S3 + X4$4 by giving the ratios Xj: X2: X3: X4 all possible values, will be called a ' quadruple group/ The assemblage, found from X^ + 7^S2 + X3$8 + X4 4 -f X5S5 by giving the ratios Xi: X2: X3: X4: X5 all possible values, will be called a ' quintuple group/ (2) A dual group will be said to be of one dimension, a triple group of two dimensions, and so on. It is obvious that a group of p 1 dimensions (p = 2, 3, 4, 5) can be denned by any p independent systems belonging to it; and also that not more than p independent systems can be found belonging to it. (3) Again, if the co-ordinates X^, Xa,... X6 of any system S satisfy a linear equation of the form, fltjXj + a2X2 + a3X3 + a4X4 + a5X5 + a^ = 0, then S belongs to a given quintuple group. For by eliminating Xg, we can write a6S = Xx (oeSi - * ) + X, (a^2 - a2S6) + X3 (aj39 - 3#6) + X4 (a A - a4S6) + X5 (a6S5 - a5S6). Hence a^ a^, a6S2 a2S6} etc., define a quintuple group to which S belongs.
168, 169] SPECIFICATIONS OF A GROUP. 285 Similarly it can be proved that if the co-ordinates X2... X6 satisfy two linear equations 2aX = 0, S/3X = 0, then the system must belong to a certain quadruple group: if the co-ordinates satisfy three linear equations, the system must belong to a certain triple group: and if four linear equations, to a certain dual group. (4) Hence a dual group may be conceived as defined by two systems belonging to it, or by four linear equations connecting the co-ordinates of any system belonging to it. And generally, a group of p 1 dimensions (p 2, 3, 4, 5) is defined by p independent systems belonging to it, or by 6 p linear equations connecting the co-ordinates of any system belonging to it. 169. Systems Reciprocal to Groups. (1) Definition. A system of forces, which is reciprocal to every system of a group, is said to be reciprocal to the group. If a system 8' be reciprocal to p independent systems, 8l9 S2, ... Sp, of a group of p 1 dimensions, it is reciprocal to the group. For any system of the group is 8 = X^ + ... 4- Xp p. Hence (88') = X2 (Sfi') + ... + Xp (flyS'). But by hypothesis (Sfl) = 0 = (S2S') = ... = (Sfi1). Hence (88') = 0. (2) All the systems reciprocal to a given group of p 1 dimensions form a group of 5 p dimensions. For let Ely E2, ... E6 be any six independent reference forces. Then any system can be written If this system be reciprocal to the p independent systems Sl9 S2, ... $p which define the given group, then the following p equations hold: Si) = 0, 1 = 0, X, (E ) + X2 (E2SP) + ... + X6 (E6SP) = 0. Hence by 168 (4) the group of reciprocal systems is of (5 p) dimensions, and is therefore defined by any (6 p) independent systems belonging to it. (3) Definition. Let this group of reciprocal systems be called the group reciprocal to the given group; and let the two groups be called reciprocal. It is to be noted that there is only one system reciprocal to a quintuple group; or in other words, the reciprocal group is of no dimensions.
286 GROUPS OF SYSTEMS OF FORCES. [CHAP. II. 170. Common Null Lines and Director Forces. (1) Definition. A line which is a null line of every system of a group is called a ' common null line of the group.' It is obvious that if a line be a null line of p independent systems of a group of (p 1) dimensions, it is a common null line of the group. Definition. Those systems of forces of a group which are simple, that is, which reduce to single forces, are called ' director forces of the group'; and the lines, on which they lie, are called ' director lines of the group.' (2) Since the null lines of a system are the lines of forces reciprocal to the system, it follows that the common null lines of a group must be the director lines of the reciprocal group; and conversely. (3) Let 8lt 82, ... Sp define a group of p 1 dimensions, and let S'p+i $'p+2, ... 86f define the reciprocal group. Call the first group Gy the second group G'. Then if X^ + X^ + ... + ApSp be a director force of G, we must have (XA+ ... + \PSP) (X + ... + V?p) = 0. Hence Xx2 (S ) + 2X1X2 (S ) + ... + Xp2 (8J3,) = 0. Let this equation be called the director equation of the group G. If ax: a2:...: ap be a system of values of the ratios Xj: Xa: ...: Xp which satisfy this equation, then a^ + a^ + ... + dpSp is a director line of G and a null line of (?'. Similarly if \p+1^'p+1 + ... + X^Sg' be a director line of (?', the \'s must satisfy the equation \2P+1 (S^i SVhH 2Xp+iXp+2 (S'p+iS'p-h) + etc = 0; and the director line of Gf is a null line of G. (4) A common null line of the group G is a null line of any one of its director forces F. But the null lines of a single force are the lines inter- secting it. Accordingly each common null line of a group intersects all the director lines and conversely. 171. Quintuple Groups. (1) Let a quintuple group be defined by the five systems jS^, S2, 83, 8it S5 and let / be the system which forms the reciprocal group. The director equation is V (SA) + 2X^2 (8A) + .. + V (8A) = 0. If otj: ol^ : a3: a4: a5 satisfies this equation, then aA + Q^ + a3$3 + A + s$5 is a director line of the quintuple group; and accordingly is a null line of 89'. Hence the director lines of a quintuple group form a linear complex defined by the system 86' [cf. 163 (2)]. Thus conversely a linear complex may be said to be defined, not only by
170 172] QUINTUPLE GROUPS. 287 any five independent lines belonging to it [cf. 165 (3)], but also by any five independent systems of the group reciprocal to S6'. (2) Also if eie^e^ be the four co-ordinate points and any system S be denoted by irY2exe2 + Tr^e^-\-ir^s^ + 7r426462 + ^e^ + ir^e^y then a linear complex is defined by the two equations 2cc7r = 0....................................(1), and ^12^34 + 7T13T42 + ^14,^23 = 0........................(2), where the a's are given coefficients. For the first equation secures that the variable system S belong to a given quintuple group, and the second that it be a director force of the group. Then by subsection (1) the lines, on which these director forces lie, form a linear complex. (3) The system reciprocal to the quintuple group given by equation (1) can easily be expressed. For let this equation be written at length in the form, a347rl2 + ^12^34 + ^W^IS + ^13^42 + Ct23 7ri4 + 0t147r23 = 0- Then the system, 8S' = OL12exe2 + 01346364 + a^^a + *42e4e2 + a^i^ + a^e2ez, is reciprocal to any system 8, whose co-ordinates satisfy equation (1). Therefore 8t is the required system. All the lines of the linear complex are null lines ofS6'. (4) In general a quintuple group has no common null line. But if the reciprocal system reduce to a single force, then this line is the common null line of the group. The linear complex is in this case called a special linear complex. It consists of the assemblage of lines which intersect the line of the reciprocal force. 172. Quadruple and Dual Groups. (1) Let 81 and 82 define a dual group and 83\ S4\ 85\ S the reciprocal quadruple group. Let the dual group be called G and the quadruple group G'. The director equation of G is V (SA) + 2\\ (S ) + V (TO = 0. This equation is a quadratic in \i/\2t and has in general two roots, real or imaginary. Let cixja2 and ft/j82 be the roots, assumed unequal [cf. subsection (9) below]. Then a^ + a2S2 and % + @282 are the only two director forces of the dual group G. Thus a dual group has in general two and only two director forces; and a quadruple group has two and only two common null lines. Another statement of this proposition is that two systems of forces have one and only one common pair of conjugate lines. (2) Also the common null lines of a dual group are the lines intersecting the two director lines of the group; and the director lines of a quadruple group are the lines intersecting the two common null lines of the group.
288 GROUPS OF SYSTEMS OF FORCES. [CHAP. II. (3) Definition. The assemblage of common null lines of a dual group is called the * congruence' defined by the group. Thus the lines of a congruence are lines intersecting two given lines. The lines indicated by the director equation of the group G\ namely V (iS/iSf.O + 2\3X4 (flf/flf/) +... + V ( / ') = 0, form the congruence defined by the group G. (4) Through any point one and only one line of a congruence can in general be drawn. To find the line through any point x of the congruence defined by the group (?, notice that it must lie in the null planes of x with respect to any two systems Sx and S2 of the group. Hence x8l. xS2 is the common null line through x. Similarly in any plane X one and only one line of the congruence lies. This line is XS1. XS2. (5) The equation, xS1.xS2=0) implies that x is on one of the two director lines of G. For if a1a2 and bj)2 are the director lines, and 81 = \1a1a2 +fabj)2, 82 = X2a1a2 -f /-lAfc,, then x8x. xS2 = (X^ X^i) xaYa2. xbj)2. Hence, assuming that the director lines are not co-planar, either xa1a2=Q, or xbj)2 = 0. Similarly the equation, X8X .X82 = Q, implies that the plane X contains one of the director lines. If x81 . xS2 = 0, and X81 . XS2 = 0, then the theorems of subsection (4) do not hold. (6) If the congruence be defined as the assemblage of the director lines of the quadruple group G'y the line belonging to it which lies in any plane or passes through any point can be determined thus: Lemma. If L denote a single force the two equations, (abL) = 0, (bcL) = 0, imply the equation (caL) = 0 and that L lies in the plane abc. But if L denote a system which is not a single force then the three equations cannot coexist. For the equations (abL) = 0 and (bcL) = 0 imply that b is the null point of the plane abc with respect to L. Hence ca cannot be a null line (assuming that abc is not zero), unless L represent a single force lying in the plane abc. Now let abc represent any given plane, and let X^SJ -f X4^4' + \fS5' + X^Sq represent any system of the group G'. Then it follows from the Lemma that the three equations, X3 (bc8/) + X4 (bcS/) + X5 (bc85r) + \6 (bcS6') = 0, X3 (caS/) + X4 (caS/) + \5 (caS/) -f X6 (caSe) = 0, X3 W) + X4 (afcS/) + \5 (abS5') + X, (aftS/) = 0, are the three conditions that this system may represent the director line in the plane a be,
172] QUADRUPLE AND DUAL GROUPS. 289 Hence the system of the group Qt' which can be written in the form Os O4, O5, O6 (bcS3% (bcSfi, (bcS5'), (bcS6') (caSf), (caS:), (coS,'), (caS6') (abSs'), (abSS), (abS5'), (abSe') is the director force of the group which lies in the plane abc. (7) Similarly the line of the congruence, which passes through any point ABC, where A, Bt 0 are planes, is found by substituting A, B, C for a, b, c respectively in the above expression. (8) Again, if the plane abc contain one of the two common null lines of G\ then every line lying in it and passing through its point of intersection with the other common null line must be a director line. Hence the above expression for the single director line lying in the plane abc must be nugatory. Accordingly the conditions, that the plane abc may contain one of the two common null lines of G\ are (bcS3'\ (bcSS), (bcS5'), (bcS6') = 0. (caS3')} (caS:), (caS5r), (caS6') (abS3') (abS4'), (abS5% W) Similarly the conditions, that the point ABO may lie on one of the common null lines, is found by replacing the points a, b, c by the planes A, By C in the above conditions. (9) An exceptional type of dual group arises, when the director equation has two equal roots. In this case, with the notation of subsection (1), if Sl and $2 be any two systems of the group, A group of this type will be called a parabolic group. There is only one director force in the group. Let it be Z), and sub- stitute D for Si in the above equation. Then, since (DD) = 0, the equation reduces to (DS2) = 0. Hence the director line is a common null line of all the other systems of the group; in other words, the director force is reciprocal to every other system of the group. The null plane of a point on the director line is the same for each system of the group, and contains the director line. For, if S be any system of the group and D the director force, any other system of the group can be written \D + fiS. Hence, if x be any point on the line D, x (\D + fiS) = fixS = xS. Since the director line is a common null line of the group, the plane xS contains the director line. Similarly the null point of a plane containing the director line is the same for each system of the group, and lies on the director line. w, 19
290 GROUPS OF SYSTEMS. [CHAP. II. The theorems of subsection (4) still hold. For, if x be any point not on the director line, the common null lines of the group through x must intersect the director force D; and therefore must pass through the common null point of the plane xD. Hence there is only one such line through x, and there is always one such line. Also, if Sx and 82 be any two systems of the group, the common null line through x is xSi. xS2* The theorem of subsection (5) still holds. For, if ab be the director force, any system of the group can be written in the form ac + bd. Now xab . x (ac + bd) = (xabc) xa + (xabd) xb. Hence, xab .x(ac + bd) = 0, implies (xabc) = 0 = (xabd). Therefore x must lie on the line ab. Now, if $! = ac -f bd, any other system S2 of the group can be written in the form Xab + fiS^ Hence xSi. xS2 = xSx. x (Xab + jjuS^ = XxSx. xab. Now X is not zero, if $2 be different from 8^ Hence, x8i . #$2 = 0, implies, xab. xSx = 0. (10) If N be any line not intersecting the director force D of a parabolic group, then one and only one system of the group can be found for which N is a null line. For let 8 be any system of the group. Then XD + fiS is any other system. If N is a null line of this system X (ND) + fi (N8) = 0. Now by hypothesis (ND) is not zero. Hence the system D (NS) S (ND) has N for a null line. And no other system has N for a null line. If D = exe2, and i\^= e3e4, then the conjugate with respect to D(N8)-S(]SrD) of the line exe2 must intersect both D and N. Hence D (NS) 8 (ND) can be written in the form 4- fiaby where e^ is any given line intersecting D and N, and a lies on D and b on N. 173. Anharmonic Ratio of Systems. (1) The null points of any given plane with respect to the systems of a dual group are collinear. For let the two systems 8j and 82 define the group, and let 8 be any third system of the group. Also let A be any plane. Then S = X1S1 + 'X2S2, also the null point of A with respect to'8 is AS=XlA81 + X2AS2. Hence AS, A8lt AS2 are collinear.
173] ANHARMONIC RATIO OF SYSTEMS. 291 (2) The anharmonic ratio of the four null points of any plane with respect to four systems of a dual group is the same for all planes and depends only on the four systems. For let 81} S2, X1S1 + \2S2) /aA + A*A be the four systems. The four null points of any plane A are A8ly AS2, XiASj-{-7^AS2i fiiA81 + fi2AS2. The anharmonic ratio of these four points, taking the first two and the last two as conjugates, is \1/jl2/\2jul1. This ratio is independent of A. (3) Similarly the four null planes of any point a with respect to the four systems have the same line of intersection, and their anharmonic ratio is also (4) Definitions. Let this ratio be called the anharmonic ratio of the four systems. If the anharmonic ratio be 1, the four systems are said to be harmonic; and one pair are harmonic conjugates to the other pair. Pairs of systems, harmonically conjugate to the two systems $j and S2, are said to form an involution, of which Si and S2 are the foci. The anharmonic ratio of the four systems X1S1-\-\282, \i81 + \2'S2i X/"Si + \'"82 is - \\") (x/x2'" - \X"')/(W" - W) Oi V - x2'V). (5) There is one and only one system belonging to a dual group which is reciprocal to a given system of the group. For if aA + a2$2 be any given system, and XA + X2 2 a system of the dual group reciprocal to it, then fci { i (SA) + a, (flfA)} + X2 K (SA) + a2 OSA)} = 0. And this equation determines X2: \2 uniquely. Thus a dual group can be divided into pairs of reciprocal systems. Each director force is its own reciprocal system. But if the group be parabolic [cf. 172 (9)], the director force is the only system of the group reciprocal to any of the other systems. For, if S be any system and D the director force, any other system can be written \D 4- jjlS. If this system be reciprocal to 8, \(DS) + /i(8S) = 0. But (DS) = 0, and (88) is not zero. Hence fi = 0. (6) A pair of reciprocal systems of a dual group are harmonic conjugates to the two director forces of the group. For let A and D2 be the two director forces, and \YDY + X2D2 and AhA + f^A be the two reciprocal systems. Then (XlAt2 + \^) (AA) = 0. Hence (assuming that the director lines do not intersect), The two reciprocal systems can therefore be written XjA A \D2t and are harmonic conjugates to A and A- 19 2
292 GROUPS OF SYSTEMS. [CHAP. II. (7) Hence systems Sly S2, 83 etc., belonging to one dual group form an assemblage of systems in involution with their reciprocal systems #/, S2', Ss\ etc., belonging to the same dual group. The foci of the involution are the director forces. The dual group will be called elliptic or hyperbolic according as these foci are imaginary or real. (8) Since a single system uniquely defines a linear complex, we can also speak of the anharmonic ratio of four linear complexes which have the same congruence in common. An assemblage of complexes with the same con- gruence in common contains two and only two special complexes. These are the foci of an involution in which each complex corresponds to its reciprocal complex, that is, to the complex of the assemblage which is denned by a system reciprocal to its own. These theorems respecting linear complexes are merely other statements of the theorems proved above. 174. Self-Supplementary Dual Groups. (1) Let the operation of taking the supplement be assumed to refer to any given quadric. The system | $ will be called the supplementary system of 8, where S is any system. Also S and \8 define a dual group. This dual group has the property that the supplement of any system belonging to it also belongs to the group. For if 8' = \S + n \S, then \S' = X \S + fiS. Let the group be called ' self-supplementary/ (2) A self-supplementary group is obviously in general determined by any one system belonging to it. For if S be known, S and \S in general determine the group. But if 8' be of the form \8 \ \S, then \S' = 8', hence 8' and |S' do not determine the group. A system $', such that \8'= S', is called a self- supplementary system. (3) If two generators of the same system of any quadric are conjugate lines with respect to any system of forces, then the generators of that system of generators taken in pairs are all conjugate lines with respect to that system of forces. Let S be the system of forces, Dl and D2 the two generators which are conjugate with respect to S; and let G be any third generator of the same system of generators. We require to prove that \8 ^/'7TJr G) is also a generator of the quadric. Now let the operation of taking supplements be performed in reference to this quadric. Then \DX= Dlt and |D2 = D%, where both the upper signs or both the under signs are to be taken [cf. 116 (3)]. Hence since 8 can be written XjDj + X2D2, we have \S= S. Also \G= G.
174] SELF-SUPPLEMENTARY DUAL GROUPS. 293 Therefore 1(SS) Accordingly the conjugate of G is a generator [cf. 116 (3)]. (4) Conversely it is obvious that if S be self-supplementary, that is, if | S = + S, then the conjugate of any generator G belonging to one of the two systems of the self-normal quadric is another generator of the same system of generators as G. It is obvious that if \S = S, the generator must be of the positive system ; if \8 = 8, the generator must be of the negative system. (5) In general the director lines of a self-supplementary group are supplementary to each other. For if the group be defined by S and | 8, the director equation is Let the roots of this equation be a//3 and /3/a, then the director forces A and A are A = fif + | S, D2 = /38+a\8. Hence |A = A, and |A = A- (6) But if we choose two director forces so that each lies on the self- normal quadric, that is, so that | A = A and | A = A (making the same choice of both ambiguities), then any system 8 = XA + ^A belonging to the group is self-supplementary. Hence these exceptional groups cannot be defined by two systems of the form S and \8. Therefore the above reasoning fails. Also if (8S) = (S\8), the roots of the director equation are equal; and the group is parabolic [cf. 172 (9)]. If (88) = (S\8), the director force is 8 \8, and is self-supplementary, and belongs to the negative system of generating lines : if (SS) = (S \ S), the director force is S + | S, and belongs to the positive system. This is the most general type of self-supplementary parabolic group, in which each system is not self-supplementary. (7) In general there is one and only one self-supplementary system of each type (positive and negative) in each self-supplementary dual group. For if the group be defined by S and \8t where 8 is any system, or by 8' and \8', where \8f is any other system of the group, then any two pairs of self-supplementary systems of the two types belonging to the group are S \8, and S' \S'. But i{8' = \8 + it\8, then \S'= \\S+ fjuS; and hence Thus all such pairs of systems are identical.
294 GROUPS OF SYSTEMS. [CHAP. II. (8) Any system S which is not self-supplementary has in general two and only two conjugate lines which are supplementary. The system obviously has one pair of such conjugate lines, namely, the director lines of the group 8 and \S. It has no more, for if possible let D and \D be two such lines which are not the director lines of the group 8, | S. Then S = hence \8 = Accordingly D and \D must be director lines of the group 8, \8, which by hypothesis is not the case. This proposition does not hold, if the group (S, \S) be parabolic. (9) This proposition may also be stated thus: Any system has in general one and only one pair of conjugate lines which are polar reciprocal to each other with reference to a given quadric. Let ab and cd be this pair of conjugates for any system 8. Let ab and cd meet the quadric in a, b and c, d. Then ad ac and bd, be are generating lines of the quadric. But these lines are also null lines of the system 8. Hence in general [cf. 175 (12) and (13)] any linear* complex has four lines which are generators of any given quadric, two belonging to one system of generators and two belonging to the other system. (10) The proposition can easily be extended to self-supplementary systems with respect to the given quadric. For if 8 be any system, then 8 \8 is the general type of a self-supplementary system. But the director lines of the group 8 and \S are supplementary, and they are conjugate lines of 8 \8 which belong to the dual group. The discussion of self-supplementary systems, and of systems such that (88) (8\8) = 0, is resumed in 175 (8) to (13). * Cf. Clebsch and Lindemann, Vorlesungen iiber Geometrie, vol. n.
175] TRIPLE GROUPS. 295 175. Triple Groups. (1) The reciprocal group of a triple group is another triple group. Let Sly 82) 83 define any triple group G, and let 84, S5\ Sq define the reciprocal group G\ The director equation of G, namely, the condition that X^ -f A2#2 + Ag#3 reduce to a single force, is V (SA) + X22 (8A) + V (8A) + 2XXX2 (818fl) + 2X2X3 (S2S3) + 2X3X2 (8A) = 0. This equation is also the condition that the line \A + \282 + \3S3 be a common null line of the group G'. (2) The condition that x may lie on a common null line of G is, For xSi. xS2 is a common null line of 81 and #2, and the given condition secures that it be also a null line of S3. (3) But the equation, (xSx. xS2. S3) = 0, is the equation of a quadric surface. Hence the common null lines of a triple group G are generators of a quadric surface. The director lines therefore, which are null lines of the triple group G', must also be generators of a quadric surface. Furthermore every null line intersects every director line, and conversely. Thus it follows that the quadric surfaces, on which the null lines of G and of G' lie, must be the same surface; and that the null lines of G are generators of one system on the surface, and the director lines of G (i.e. the null lines of (?') are generators of the other system on the surface. Let the two systems of generators be called respectively the null system and the director system with respect to the given group. (4) Hence a triple group G defines a quadric surface. The only other triple group which defines the same surface is the reciprocal group G'. The director system of generators with respect to G is the null system with respect to (?', and vice versa, (5) Conversely, any quadric surface defines a pair of reciprocal triple groups. For take any three generators of the same system belonging to this quadric. Let G1} G2, G3 be forces along them. Then G1} G2 G3 define a triple group, and its associated quadric must contain the three lines Gly G2, G3. But there is only one quadric which contains three given lines. Hence the associated quadric is the given quadric. (6) The condition that the plane abc may contain a director line of the group G is = 0. (bc8%), (caS2), (caS3) (abS2), (abS3)
296 GROUPS OF SYSTEMS. [CHAP. II. For assume that X^ + \2S2 + X3 3 is a single force lying in the plane abc Then, by the lemma of 172 (6), the three following equations are the necessary and sufficient conditions, Xi (bcSJ + X2 (bcS2) + X3 (bc83) = 0, \ (caSJ + X2 (caS2) + X3 (caS3) = 0, \* (abSi) + X2 (abS2) + X3 (a 3) = 0. But these equations require the given condition. Accordingly this is also the condition that abc may touch the associated quadric and contain a common null line of the group. (7) Similarly the condition that the point ABC, where A, B, G are planar elements, may lie on the associated quadric is found by replacing a, b, c in the above condition by A, B and C. (8) If the supplements of G and Gf be taken with respect to the asso- ciated quadric, then from 174 (3) and (4) every system belonging to G or G' is self-supplementary; and conversely all self-supplementary systems with respect to a given quadric must belong to one of the two associated groups of the quadric. For any system 8 of group G we may assume \S = S', then for any system 8' of G' we have \8' = - 8'. (9) Corresponding to each director line of a triple group, one parabolic dual subgroup can be found with that line as director line. For let Flf F2i F3 be any three director lines of the triple group, and let FY be the given director line. Then any system 8 of the triple group can be written Now, if the subgroup (Fu 8) is parabolic, (F1S) = 0. Hence the required condition is Thus the subgroup denned by i^ and (FjFs) F2 {FYF2) F3 is parabolic with Fj as director line. (10) Let the quadric defined by the triple group be self-supplementary. Hence by the previous subsection, if S1=(F1F3) F2- (F1F2)F3f the dual group defined by F1 and S1 is parabolic and such that each system 8 is self- supplementary. If |Fx = F1, then \8=S; and if \F, = -F,, 18 = - S. Corre- sponding to each generator of either system there is one such parabolic self-supplementary dual group [cf. subsection (12), below], (11) The most general type of self-supplementary parabolic group, in which each system is not self-supplementary, is the type defined by a generator, Gy of the self-normal quadric and a self-supplementary system S; such that, either | G = (?, and | S = - 8, or, | G = - (?, and J 8 = 8 [cf. 174 (7)].
175] TRIPLE GROUPS. 297 For firstly let G, the director line of the parabolic subgroup, be such that \G = G; and let 81 be any other system of the group. Then by 174 (6) \8j~aG-8!. Also by hypothesis, (GSi) = 0. Then any system S of the group can be written \G + m#i. Hence \S = X \G + p \8= (X + fia) G-fiS^ Thus, if 8 be self-supplementary, that is, if \8 = S, then, X-t-/xa = X; that is, X = \fiOL. Hence the system 8 = Sx ^aG} is such that \8= 8. Accordingly the self-supplementary parabolic group can be defined by G and 8; where \G = G, | S = - 8. Similarly if | G = (?, then the self-supplementary system 8 belonging to the group is such that | S = 8. Thus corresponding to any generator G of the self-normal quadric there are an infinite number of such parabolic self-supplementary groups, since any self-supplementary system 8 of the opposite denomination (positive or negative) to G will with G define such a group. (12) It is evident [cf. 174 (3) and (4)] that any self-supplementary system 8 has as null lines all the generators of the self-normal quadric of the opposite denomination. It also has as null lines two generators of the same denomination. For we may write S = olJ)1 + oc2D2, where A and A are two generators of the same denomination as S. Let A be a third such generator. Then any self-normal system of the same denomination as 8 can be written in the form This system is a generator (D) if x2x3 (A A) + Vw (AA) + x2x2 (A A) = 0. Also D is a null line of 8, if (DS) = 0, that is, if (Xlttl + X2a2) (AA) + X3 {^ (AA) + a, (AA)j = 0. These two equations give two solutions for the set of ratios of Xx to X2 to X3. Hence S has two null lines among the generators of the same denomi- nation [cf. subsection (10) above, and also 174 (10)]. (13) Now let if be a generator of the opposite denomination to the self-supplementary system S; and let D and D' be the two generators of the same denomination as S, which are null lines of 8 according to the previous subsection. Then D and D' necessarily intersect N. Also the parabolic group defined by N and 8 is of the type discussed in subsection (11). But D and D' and N must be common nulls of this group. Also no other generators of
298 GROUPS OF SYSTEMS. [CHAP. IT. the quadric can be null lines of any system of the group, other than N and 8. For consider the system \N + pS. Then every generator of the D type intersects JV, but only D and D' are null lines of 8. Accordingly only D and D' of the generators of this type are null lines of \N+fA,8. Again, all the generators of the N type are null lines of 8; but no generator of this type, except JV, intersects N. Hence N is the only null line belonging to the generators of this type. Hence any system S', not self-supplementary, which is such that (8'8r) (S' \8') = 0, has two generators of one system and one generator of the other system as null lines. This proposition should be compared with that of 174 (9). (14) Thus, summing up and repeating, any quadric has in general two generators only of one system and two generators only of the other system, which are null lines of any system of forces S. But, as exceptional cases, either all the generators of one system and two only of the other system are null lines of S; or one generator only of one system and two only of the other system are null lines of S. 176. Conjugate sets of Systems in a Triple Group. (1) Any two systems 8lt S2 of the triple group G define a subgroup. It is possible to find one and only one system S belonging to G which is reciprocal to the whole subgroup Sj, 82. For let 8 be such a system and let 83 be any third independent system so that S1} S2, 83 define G. Then we may write S = X + \2S2 Hence by hypothesis X, (SA) + A, (8,8,) + \ OW = 0, x, (5A) + x, ( ) + x, (S. ) = o. Thus the ratios \x: \2: X^ are completely determined, and therefore 8 is completely determined. The reciprocal system 8 can be written in the form 8ly 82i S3 (82S3) This system does not belong to the dual subgroup (Slt S2), it the coefficient of 83 does not vanish; that is, if be not zero; that is, if the subgroup (8l9 S2) be not parabolic. In subsections (2), (3), (4), following, the subgroups will be assumed to be not parabolic. (2) Also in the subgroup defined by 8l9 S2 we may choose 8X and 82 so as to be reciprocal [cf. 173 (5)]. Thus three systems Slt 82, S3 can be found, belonging to the triple group (?, such that- each system is reciprocal
176] CONJUGATE SETS OF SYSTEMS IN A TRIPLE GROUP. 299 to the subgroup formed by the other two. And one of these systems, say 8lt can be chosen arbitrarily out of the systems of the group G; and then S2 and 83 can be chosen in a singly-infinite number of ways out of the dual subgroup of G which is reciprocal to jSx. Definition. Let such a set of three mutually reciprocal systems of a group G be called a 'conjugate' set of the group. (3) If 81} 82 S3 be a conjugate set of systems, then XSlf XS2, X83 are three conjugate points lying in the plane X with respect to the associated quadric of G. For the director lines of the group {8ly 82) are generators of the quadric G; and it has been proved [cf. 173 (6)] that the line joining the points Z^ and XS2 intersects these director lines in two points d1 and d2 such that the range formed by (d1} d2 XSlt XS2) is harmonic. But di and d2 are on the quadric G. Hence by the harmonic properties of poles and polars XSi is on the polar of XS2, and XS2 on the polar of XSX. Similarly for XS2 and X83, and for X8X and XS3. Hence the three points XSU X82, XS3 are three mutually conjugate points on the plane X. (4) An analogous proof shows that xS1} oo82y xS3 are conjugate planes through the point x.
CHAPTER III. Invariants of Groups. 177. Definition of an Invariant. (1) Let 8lf S2, ... 8P define a group G of p - 1 dimensions, and let /, 82, ...Sp be any p systems belonging to this group G. Then there must exist p equations of the typical form Also let A denote the determinant Then, if A be not zero, the systems #/, 82)... 8P are independent [cf. 96 and 63 (4)] systems. (2) Let 4 (S1} 82y ... Sp) be any function of the p systems 8lt S2, ... Sp formed by multiplications and additions of 8i, 82, ... Sp and of given points, forces, and planar elements. Let / ($/, 82i ... Sp) denote the same function only with #/, 82, ... Sp substituted respectively for 8lf S2, ... 8P. Then if / ( /, S2} ... Sp') = Ax f)(81, S2, ... 8P), X being an integer, / (fi i, S2) ... Sp) is called an invariant of the group G. The effect of substituting any other p independent systems of the group G for 8lt S2i ... Sp in an invariant of the group is to reproduce the original function multiplied by a numerical factor which does not vanish. 178. The Null Invariants of a Dual Group. (1) Let S2 and #2 define a dual group, and let 8 = \S1 + /jl82, 8'= \'S1 + fjb'82 A = \// X'fju. Then the expressions wSi. #$2 and X$x. XS2t where x is any point and X is any planar element, are invariants of the group. Call them the Null Invariants. For xS. xS' = AxS,. x82 and XS. XS' = AXS,. XS2. It has already been proved [cf. 172 (4)] that these expressions denote respectively the common null line of the group through the point x, and the common null line of the group in the plane X.
CHAP. III. 177 179] THE HARMONIC INVARIANTS OF A DUAL GROUP. 301 179. The Harmonic Invariants of a Dual Group. (1) Another important invariant of the group is xS . S2 oc82. $1. Call this expression the Harmonic Point Invariant of the group; let it be denoted by H(x). This expression is easily proved to be an invariant by direct substitution. It represents a point. It must be noticed that the intensity to be ascribed to H(x) depends on the special pair of systems (8U 82) which is chosen to define the group. It is obvious that H (\x + /*#') = XH (x) + pH (w'). (2) Similarly if X be any planar element, X81.82 XS2. S, is an invariant of the group. Call this expression the Harmonic Plane Invariant; and let it be denoted by H(X). It represents a planar element. The intensity of H(X) depends on the special pair of systems which define it. Also H{\X + ixX') = \H(X) + pH(X'). (3) If #! and S2 be a pair of reciprocal systems, it is obvious from 167 (5), equation 25, that Hence in this case H(x) = 2#$2. 82 = 2xS2. 8j. Similarly H(X) = 2XS1. S2 = - 2XS2. $. These expressions only hold when Si and S2 are reciprocal. (4) To find the relation between the points x and H(x)y and between the planes Xand H(X). Let the common null line through x meet the director lines of the group in dj and d2\ and let the two director lines be written d1e1 and d2e2. Then Sx and 82y which will be assumed to be reciprocal systems, can be written in the forms [cf. 173 (6)] 8i = dxex -f d2e2, S2 = \ (d^ - d2e2). Also we may write x = ^xdx 4- %2d2. Hence by multiplication xSx= ^dxd2e2-\- \;2d2dxeu wS1. S2 = Xgjdjdfa. dxeY Xg^d^ . d2e2 = X (d1e1d2e2) {^dx 2d2}. Also (SA) = 2 (d d l (SA) = - 2\2 (dntofaj). Therefore H(x) = 2x8,. S2 = V{- (S181) (S2S2)\ (f A - 2d2). But ^idi+f2d2, ^1^1 ^2, di d2 form a harmonic range. Hence H(x) lies on the common null line of the group through x, and is the harmonic conjugate of x with respect to the two points in which the null line meets the director lines. (5) Similarly H(X)= 2XS,. S2= V{-0W0SA)} ( A ~ A); where A and B2 are two planes both containing the common null line in the plane X, and respectively containing the two director lines; and X =
302 INVARIANTS OF GROUPS. [CHAP. III. Hence H(X) contains the common null line of the group which lies in X} and is the harmonic conjugate of X with reference to the two planes containing the null line and the two director lines. (6) Let H{H(x)} be written H2(x\ and let H*(x) denote H{H*(x)}, and so on. Then it has been proved in (4) and (5) that if d1 and d2 lie on the director lines of the group, and x = f A + 2d2, H (x) = V{- (S ) (S2S2)} (f A - f A). It follows that H2 (x) = - (S ) (S2S2) (fA + ? A) = - (8181) (8282) x. Therefore H2(x) = x, and generally HK(x) = xy or =H(x), according as X is an even or an odd integer. Similarly H2 (X) = - (S ) (82S2) X; and hence H* (X) = X. (7) If the group be parabolic [cf. 172 (9)], then H (x) is the null point (common to all the systems) of the plane through x and the single director line. For let D be the director force and 8 any other system of the group, then (DS) = 0. Hence by subsection (3), H (x) = 2xD . 8. Thus H (x) is the null point of the plane xD with respect to S. Accordingly all the points of the type H (x) are concentrated on the director line ; and, if (xyD) = 0, then H (x) = H (y). Similarly H (X) is the null plane of the point DX. 180. Further Properties of Harmonic Invariants. (1) If Sj and 82 are two reciprocal systems of the group, the null plane of x with respect to 8X is the same as the null plane of H(x) with respect to S2. For by 167 (2), since aSi and xS2 are planar elements, H(x)S2 = and H (x) Sj = Similarly the null point of X with respect to Si is the same as the null point of H(X) with respect to 82. For H(X)82 = (2) If S be any system of the dual group, to prove that H(x)S=-H(xS), H(X)S = -H(XS)...............(3). For let 8' be the system reciprocal to 8 belonging to the group. Then we may write H (x) = 2x8.8'. Also from the second of equations (1) in subsection (1), H(x)8=-(SS)xS'. Again by 167 (2) H(xS) = 2x8.8.8' = (SS)xS'=-H(x)8. Similarly H (XS) = - H(X) $.
180* 181] FURTHER PROPERTIES OF HARMONIC INVARIANTS. 303 (3) If the locus of x be the plane X, then the locus of H(x) is the plane H(X). This proposition is obvious from the harmonic relation between x and H(x) and between X and H{X). It can also be proved by means of the important transformation XH(x) = xH(X)..............................(4), where x and X denote respectively any point and any plane. For if Si and 82 be any two reciprocal systems of the group, then remembering that the product of two planar elements and a force, or a system of forces, is a pure regressive product, XH(x) = 2X.(xS1.S2) = 2X.xS1.82 = - 2xSx. XS2 = - 2x (S,. XS2) = 2x (XS1. S2) = xH(X). (4) If ab be a null line of any system 8 of the dual group, then H (a) H(b) is also a null line of 8. For by hypothesis (abS) = 0. And by (2) of this article, H(a)H(b)8=-H(a) H(bS). But by (3) of this article and by 179 (6), H (a) H(bS) = bSH* (a) = bSa = 0. Hence H(a)H(b)S = 0. Since H2(x) = x, this proposition can also be stated thus, if aH(b) be a null line of 8, then bH(a) is a null line of 8. (5) If Sx and S2 be reciprocal systems of the dual group and ab be a null line of $2, then H(a) H(b) is the conjugate of ab with respect to Slm This proposition will be proved [cf. 164 (4)] by proving the important formula H(a)H(b) = -2(S2S2)aS1.bS1.....................(5); where H (x) = 2xSt. 82. For remembering that ($i$2) = 0, and (abS2) = 0, and twice using equations (22) of 167 (3), H(a) H(b) = 4 (aSj) S2. (bSJ S2 = 4 (aS,. bS,. S2) S2-2 (8 ) aS,. bS, = 4 [{(abS,) 5,-1 (SA) ab} S2] S2-2 (S2S2) aS1. bS, = -2(8 )a81.b81. In connection with this proposition and that of subsection (4) the proposition of 166 (2) should be referred to. 181. Formulae connected with Reciprocal Systems. (1) A variety of formulae connected with two reciprocal systems can be deduced from the preceding article. Thus equation (22) of 167 (3) can be written (abS) S = % (SS) ab + aS . bS.
304 INVARIANTS OF GROUPS. [CHAP. III. From this equation and from equation (5) of 180 (5), it immediately follows that, if ft and S2 be reciprocal and ab be a null line of ft, 2 (a6ft) Sl = (ftft) ab - ~ H (a) H (6). .(1). Similarly, 2 (4.8ft) S, = (ftft) AB - -r^r H(A)H(B); where AB is a null line of S2. (2) Also with the same assumptions as in (1), it follows from 180(5) that aH(b) is a null line of ft. Hence by the preceding subsection 2 {aH (6) S2] S2 = (S2S2) aH (6) - ~ H (a) H* (a). But by 179 (6), H- (6) = - (ftft) (S2S2) b; also by an easy transformation Hence 2 (abS,) S2 = aH (b) - bH (a). \ Similarly 2(ABS1)S2 = AH(B) - BH(A).) .................. (3) Also, since (ab82) = 0, (abS2) S2 = 0 = (S2S2) ab + aS2. bS2. 2 Hence ab = - /o a. aS2. bS2. Thus (a6ft) ft = (ftft) ab 4- aft. 6ft = aft . bS1 - J^{ aft. bS2. .,.(3). Similarly, if (ABS2) = 0, (4.8ft) ft = AS,. Bft - - (4) Also, with the same assumptions, equations (26) of 167 (5) become (a6ft) S2 = aS,. bS2 + aS2. 6ft, 182. Systems reciprocal to a Dual Group. (1) Let R be any system reciprocal to a whole dual group. Then R belongs to the reciprocal quadruple group. Also let ft and S2 be two reciprocal systems of the dual group. Then by equation (25) of 167 (5) and remembering that (Eft) = 0 = (RS2), Similarly, H (xR) = H (x) R. (2) We may notice by comparison of this result with 180 (2) that if S be any system of the dual group, But if R be any system of the group reciprocal to the dual group, H(xR) = H(x)R, H(XR) = H(X)R.
182, 183] SYSTEMS RECIPROCAL TO A DUAL GROUP. 305 183. The Pole and Polar Invariants of a Triple Group. (1) Let the triple group 0 be defined by three systems 8lt 82, S3. The same three systems taken in pairs define three dual subgroups. Let these dual subgroups be denoted by glf g2i g3; thus, let the group g1 be defined by S2, S39 the group g2 by S8 Su and the group g3 by 8l9 82. Let the harmonic invariants of the point x or of the plane X with respect to the groups gl9 g2 and g3 be denoted respectively by H1(x), Hl{X)i H2(x), H^X), H3(x), H3(X). (2) The expression /Si , S2 , (S ), (SA), (8,8,), (8J3t), (1). (8A), (828S), (S38S) will be proved to be an invariant of the group, and will be called the Polar Invariant with respect to the group 0. Similarly the expression 81 82 , S3 , (S2S3) , (S3S3) (2), (S2S3) (S3S3) will be proved to be an invariant of the group, and will be called the Pole Invariant with respect to the group G. (3) If Rj be the system of the group G reciprocal to the subgroup gly then by properly choosing the intensity of Rj we may write [cf. 176 (1)] (S2S2), (S2S3) (s3s3) S2 S3 (s2s3) R1= Hence the polar invariant of x with respect to G is H1(x)Rli and the pole invariant of X with respect to G is H1 (X) i^. Let the polar invariant be denoted by P(x) and the pole invariant by P(X). Then P(x) = H1(x)R1, and P(X) = H1(X)R1. (4) Another form for P{x) and P(X) can be found as follows. We have H, (w) 8.2 = {xS,. Ss - xS3. S2} S2 = {(8 ) x - 2xS3. S2] S2 = (82Ss)x82-(82S2)x83. Also H, (x) 8t = {2xS2.83 - (8J3,) x] Ss = (8^) xS2 - (8 ) xSs. Hence from equation (2), P (x) = H, (x) Sy - (8A) xS2 + (S ) xSs.\ Similarly P (X) = H1 (X) S, - (S ) XS2 XS3. ...(3) (5) The invariant property can easily be proved from this latter form. For write x{8l9 S2, ,} for P (x) as defined above, in order to bring out the w. 20
306 INVARIANTS OF GROUPS. [CHAP. III. relations of P (x) to the three systems Slf 82, S3. Then it follows from the form for P (x) given in equations (3) that * {ft, 8 S3}----x {8lt Ss, S2]........................(a), also { ,, , } = ).......................................(b). Furthermore Hx (x) S1 = {2xS2. ft - (ftft) x\ 8, = 2 (ftft) xS2 - 2x8,. ft . ft - (ftft) ft. Hence w {ft, ft, Ss\ = (ftft) xS2 - 2x82. ft . ft - (ftft) aft + (ftft) ft = (ftft) *ft - (ftft) xS, - [2wS2. S, - (ftft) *} 8S = fl"s( ) ft-(ftft) ft +(ftft) ft = *{ft)(S1)(S2}..........................................(c). Lastly * {ft + ft', S2, 8S] = {ft, ^2 ^3J + * {ft', ft, ft)............(d). Now let ft S', 8" be three systems of the group, such that S = Xft + ^/S2 + vft, i8f = Vft + /t'ft + i/ft, S" = X"ft + /'ft + v'%; and let A denote the determinant 2 + Xfx'v". Then from the equations (a), (b), (c), (d), which have just been proved, we deduce at once that x{S,S',S"} = bx{S1,S2,S3}. This proves that P (%) is an invariant of the group. An exactly similar proof shews that P (X) is an invariant of the group. Now that the invariant property is proved we may abandon the notation x{Slf 82}S3} forPO). 184. Conjugate sets of Systems and the Pole and Polar Invariants. (1) Let Rlt R2, R3 be a set of conjugate systems of the group G. Then Also let glt g2 g3 denote the subgroups R2R3 and R3R1} and RYR2 respectively. Hence R1 is reciprocal to the group g1} and R2 to the group g2, and R3 to the group g3. Then P (x) and P (X) take the simple forms 1xR2. R3. Rx and 2XiJ2.iJ3.jRi. This follows at once from the forms for P (oc) and P(X) given in 183 (4), equation (3). (2) It also follows that P(oc) = 2#iJ2. iJ3. RY = 2wR,. iJ2. iJ3 = 2xRs. R,. iJ2 = - 2^iJ3,R2.R,= etc.; with similar transformations for P (X). (3) The equation, aRx = P (6), can be solved for a. For let P(b) = 2bR2.R3.R1. Then multiplying each side of the given equation by R1} aR,. iJi = (iJiiJO a = P(b)R1 = 2bR2. iJ3. iJi. Ri = (iSA) bR2. jR3. Hence a = 2bR2.R3=- 2bR3. i?2.
184,185] CONJUGATE SETS OF SYSTEMS AND POLE AND POLAR INVARIANTS. 307 Also a condition holds. For aR2 = - 2bR3.R2.R2 = -(R2R2)bR . Thus (abR2) = - (baR2) = (R2R2) (bbRs) = 0. Similarly (abR3) = 0. Accordingly aft is a common null line of the subgroup gY. 185. Interpretation of P(x) and P(X). (1) P(x) denotes a planar element, and P (X) denotes a point. To find the plane P (x), write P(x) in the form 2xR2. R3. Rlf which is given in the last article. Let the common null line through x of the subgroup gx intersect the director lines of gx in dx and d/. Then c?2 and d( are on the quadric G [cf. 175 (4)]. Also the four points x, 2xR2.R3, dlt dY' form a harmonic range [cf. 179 (4)]. Hence the point 2xR2. R3 lies on the polar plane of x with respect to this quadric. But P (x) is the null plane of this point with respect to R1; and therefore the plane P (x) passes through the point 2xR2. Rz. Now let Ri, R2, Rs' be another set of conjugate systems of the group G, Then the same proof shews that the plane P (x) passes through the point 2xR2 . R3; and that this point, 2xR2 . R3\ lies in the polar plane of x with respect to the quadric (?. Similarly for a third set of conjugate systems, such as jR/', .fl/', R3". Hence the plane P (x) passes through the three (not collinear) points 2xR2.R3} 2xR2/.R3, 2xR^'.B3". Hence P (x) denotes a planar element of the polar plane of x with respect to the quadric G. Similarly P{X) denotes the pole of the plane X with respect to the quadric G. (2) It follows as a corollary from subsection (1) and from 176 (3) and (4) that we can express the angular points of tetrahedrons self-conjugate with respect to the quadric G, which have one face in a given plane. For let X be the given plane, and R1} R2y R3 a set of conjugate systems of the group G. Then by 176 (3) XR1} XR2, XR3 are three conjugate points in the plane X and by the present article P(X) is the pole of X. Hence these four points are the corners of a self-conjugate tetrahedron with one face in the plane X. By taking different sets of conjugate systems an infinite number of such tetrahedrons may be found. (3) Similarly we can express the four planes which are the faces of a self-conjugate tetrahedron with respect to G, of which one corner is at a given point x. 20 2
308 INVARIANTS OF GROUPS. [CHAP. III. For, by the same reasouing as that just employed, the four planes are xRu xR2i xR3 and P{x). By taking different sets of conjugate systems an infinite number of such tetrahedrons may be found. (4) The interpretations of P (x) and of P (X), which are given in (2) and (3), shew that P2(x) [i.e. P {P (x)}~\ must denote the point x, and that P2 (X) must denote the plane X. This result can also easily be proved by direct transformation. (5) Again it follows from the interpretations of P (x) and P (X) that if y lie on P(x), then x lies on P{y)\ and that if Y contain P(X), then X contains P(Y). This result can also be proved by direct transformation, namely the following equations hold [P(x) y] = [P(y)x\ [P (X) T] = [P (Y) X\ 186. Relations between Conjugate Sets of Systems. (1) It fol- lows from 181 (3), equation (3), that if Rl9 R2, R3 be a conjugate set of systems, and if (abR2) = 0, then (abRJ R, = aR1. bR, - ^^ aR2. bR2. {MM) Now if we take aR1 = P(b)t then by 184(3) the condition (abR2) = 0 is fulfilled; and (aR2) = - (R2R2) bR3. Also (abR,) = - (baRJ = - {bP ( )} = {P (b) b}. Hence finally if b be any point, {P (6) b}R, = P (b). bR, - (R.R,) bR2. bR, ;\ and, since b bears no special relation to Ru by the cyclical interchange of suffixes, a) {P (b) b}R2 = P (b). bR, - {P (b) b}Rs = P (b). bR3 - (R3R3) bR,. bR,., (2) Similarly if B be any plane, {P (B) B}R, = P (B) . BR, - (R.R,) BR2. BR3,\ {P(B)B}R2 = P(B).BR2-(R2R2)BRS.BR1\............(2) {P (B) B}R3 = P (B). BR3 - (R3R3) BR,. BR2] (3) It is to be noticed that P (B\ BR1} BR2) BR3 are the four angular points of a self-conjugate tetrahedron with respect to the quadric G. This tetrahedron has the plane of one face, namely B, arbitrarily chosen, but is otherwise definitely assigned by the conjugate set of systems Rlf R2, R3. Similarly, P (b), bR1} bR2, bR3 are the four faces of a self-conjugate tetra- hedron with respect to the quadric 0. This tetrahedron has one angular point, namely b, arbitrarily chosen, but is otherwise definitely assigned by the set Ru R2, R3.
186] RELATIONS BETWEEN CONJUGATE SETS OF SYSTEMS. 309 (4) Let^, p2, p3, p denote the angular points of a self-conjugate tetra- hedron with respect to the quadric G. Then one reciprocal set of systems with respect to the group 0 can be expressed by PP1 + P1P2P*, where fi1} fi2i /jl3 are given definite numbers. Similarly if P1} P2) P3, P denote planar elements in the faces of a self- conjugate tetrahedron, then one reciprocal set of systems can be expressed by (4) PPS + X.P.P,;) where \lt \ , X3 are given definite numbers. (5) The proposition of the preceding subsection, symbolized in equations (3) and (4), may be enunciated as follows: Corresponding to any given set of conjugate systems of a group G, one and only one tetrahedron self-conjugate with respect to the quadric G can be found with three corners in a given plane, such that its opposite edges taken in pairs are respectively conjugate lines of the three systems of the conjugate set. Also corresponding to any given set of conjugate systems of a group G, one and only one tetrahedron self-conjugate with respect to the quadric G can be found with one corner given, such that its opposite edges taken in pairs are respectively conjugate lines of the three systems of the conjugate set. Such self-conjugate tetrahedrons will be said to be associated with the corresponding conjugate sets of systems, and vice versa. (6) The group Gr reciprocal to the group G is also a triple group, and defines the same quadric as G. Now let p, p1} p.2, p3 be the four angular points of a tetrahedron which is self-conjugate with respect to this quadric. Also let the conjugate set of systems of the group G associated with this tetrahedron be PPi + PyViPz , PP2 + f^p Pi, P Ps + P3P1P2 Then it is obvious that the conjugate set of the reciprocal group (?', associated with this tetrahedron, is PPi ~ frP*P* PPi - M *Pi, PP* ~ V2P1P2 For it follows from mere multiplication that any system of the last set is reciprocal to each system of the first set. Hence the three systems of the last set each belong to the group G'. Furthermore they obviously are reciprocal to each other, and therefore form a conjugate system of the group G\ And lastly, the form in which they are expressed shews them to be the conjugate set of systems associated with the tetrahedron p, pi,p2,ps.
310 INVARIANTS OF GROUPS. [CHAP. III. (7) Similarly an analogous proof shews that if P, P1} P2, P3 be the four faces of a tetrahedron self-conjugate with respect to the quadric of G and G\ and if the associated conjugate set of G be PP. + ^P.Ps, PP2 + fJL2P3P1} PPz then the associated conjugate set of the group Gf is 187. The Conjugate Invariant of a Triple Group. (1) If 8l9 82, S3 be any three systems of the group G} the equation of the quadric G is [cf. 175(3)] (xS1.xS2.S3) = 0. (2) If x and y be any two points and \x + fiy be a point on the quadric lying on the line joining them, then \2 (xSx. xS2. 89) + \/i {(x81. yS2. S3) + (yS,. xS2. S8)} + ^2 (yft. yS2. S3) = 0. Hence the condition that the points x and y should be conjugate is If y be regarded as fixed, this is the equation of the polar plane of y. (3) Let the expression ^ {x8x. yS2. 83 + ySx . xS2. S3] be denoted by G (xy). It will be proved to be an invariant of the group G, and will be called the Conjugate Invariant. The equation G (xx) = 0, is the equation of the quadric G, It follows from symmetry that, G(xy) = G (yx). (4) In order to prove the invariant property let us write G (xy) in the form xy{8u 82) 83}. Then obviously *y {$ + ', 82) 89}=xy{8l9 829 S3] + xy {S/, S2, S3}.........(1). Also (cy{8l98a989}^^x!f{8%98u89} ........................(2), and xff{8l98l989} = 0 .................................(3). Furthermore yS2. #3 = (S2S3) y yS3. S2. Hence xS,. yS2.S3 = x8,.(yS2. ,) = (S2S3)xS .y-xS,. yS3. S2. Similarly yS,. ^2. ^3 = (S2S3) yS,. a? - yS,. ^ S3. S2. Also ajfid. 3/ + ySi. a? = ###! + 3/^/?! = 0. Therefore xy {8l9 8%9 S3} = - xy {8l9 83) S2] = xy {S3} 8l9 S2] ......... (4). Now let S, 8', 8" be any three systems of the group G, such that S = XS1 + julS2 + vS3, S' = VSfj + / + ^3, S7' = V Also let A stand for the determinant X Xfx'v". Then from equations (1), (2), (3), (4) it follows that
187] THE CONJUGATE INVARIANT OF A TRIPLE GROUP. 311 This proves the invariant property of xy [Slt S2, S3}. This expression for the conjugate invariant may now be abandoned in favour of G (ocy), in which the special systems used do not appear. (5) Let R1} R2y R3 be a conjugate set of systems of the group; so that Then (yR,. xR2. R3) = - {yRx. R2. xR3) = (yR2. R . xR3) = - (yR2. R3. xRY) = (xRx. yR2. R3). Hence G (xy) = % {(xR,. yR2. R3) + (yR,. xR2. R3)} = (xR,. yR2. R3) = (xR,. i?3. yR2) = {(xR,. R3) R2.y}. But xR1.R3.R2 = %P(x). Therefore G(xy) = | {P ( ) 3/} = | {P(2/)x}. The equation (? (a?*1) = 0, can be written in the form [P (x) x) = 0. (6) Similarly the condition that the plane X touches the quadric G is The condition that the planes X and Fare conjugate is The expression {(XSX. YS2. S3) + (Y81. XS2. S3)} can be proved to be an invariant of the group and will also be called the conjugate invariant, and denoted by G (XY). Also G(XY) = G(YX). Furthermore if Rlt R2) R3 be a set of conjugate systems of the group, then G(XY) = (XR1.YR2.R3). Also G(XY) = r{P(X) Y] = 1 {P(Y)X}. Therefore the plane-equation of the quadric G is (7) Also from 183 (3), P (*) = JEfj (x) R1, hence G (xy) = i {P (x) y} = \ [H, (x) RlV) = | [H, (as). yB,) ^}, from 182(1) I (^ (wRJ y}. Similarly G (Z Y) = i {^ (X) YE,} = ^{H1( YE,) X) = h{Hi(Y)XRJ = i {ifx(XRJ Y). (8) Corresponding to each director force A of the triple group, there is one parabolic subgroup [cf. 175 (9)]. Let $2 be any system of this subgroup, and let Hs (as) and Hs (X) be the harmonic invariants with respect to this subgroup of a point x and of a plane X. Then we will prove that H3 (x) [cf. 179 (7)] is the point of contact of that tangent line from x to the
312 INVARIANTS OF GROUPS. [CHAP. III. quadric, which intersects D1\ also that H3{X) is the tangent plane con- taining that tangent line to the quadric, which lies in the plane X and intersects the line DY. For let y be another point, and S3 any system of the triple group which does not belong to the given parabolic subgroup. Then, according to sub- section (2), if \x + fxy be a point on the quadric, Now let y = H3 (x) = 2xA. 82. Then by 179 (7), yD1 = 0. Hence (yDL.yS*. 3) = 0; and Also yS2 = 2xD, .82.S2 = xDx(S2S2). Hence (xD,.yS2. S3) = 0. Thus the equation reduces to V = 0. Hence the two points, in which the line xH3 (x) meets the quadric, coincide at the point H3 (x). Similarly for the second part of the theorem which concerns H3 (X). Another mode of stating the propositions of this subsection is that, all quadrics with a given parabolic subgroup touch along the director line of that subgroup. Hence the equation of any one of a group of quadrics which touch along a common generator can be immediately written down. 188. Transformations of G(pp) and G(PP). (1) Let a point p on a plane X be conceived as the null point of X with respect to some system of the group; it is proved in subsection (3) below that, if G (X, X) be not zero, p may be any point on the plane X. Hence p can be written X (Xi-Rj + XA + X3B3); where Rly R2, R3 are three reciprocal systems of the group. Then writing XjXi?! + \2XR2 -f \XR3 for the second p in G(pp), and then using 187 (7) G (pp) = \G (p, XR,) + \G (p, XR2) + \,G (p, XR3) = 1 X.H, (XRJ .pRl + % \,H2 (XR2) .PR2 + i XSH3 (XR3). PR3. But Rx is reciprocal to the dual group glt and therefore by 182 (1) Hx (XR,) = Hx (X) R,; similarly H2(XR2) = H2 (X) R2, H3 (XR,) = #3(X) R3. Hence G(pp) = %\1H1(X)R1.pR1 + \2H2(X)R2.pR2 + \3H3(X)R3.pR3. But [cf. 167, equation (21)] H1(X)R1 .PR1 = B1(X)R1. R,,p = \H1{X)p Similarly H2 (X) R2 .pR2 = % H2 (X)p. (R2R2), and H3 (X) R3 .pR3 = % H3 {X)p. (R3R3). Hence
188] TRANSFORMATIONS OF G(pp) AND G (PP). 313 Again %Hx(X)p = \ H, (X). X (XlR1 + X R2 + \RS). And by 187 (7) } H, (X). XRX = G (XX). Also J^ (X). XR2 = XR2. R3. XR2 = 0; similarly \ H, (X). XR3 = 0. Therefore \ H, (X) p=\G (XX). Similarly Thus finally G (pp) = G (XX) {Xx2 (RXR,) + X22 (i?2 2) + X32 (R3R3)}. (2) Now if G (pp) = 0, p is a point on the section of the quadric G made by the plane X. But (? (j p) = 0 involves either G (XX) = 0, or V (iJA) + v (iW + x32 (R3Rs) = 0. If (? (XX) = 0, the plane touches the quadric and therefore contains one director line of the group and one common null line. The null points of this plane in respect to the various systems of the group must lie on this common null line. If G (XX) be not zero, then V (JKA) + V (R2R2) + V ( 3 3) = 0. But this is the director equation of the group. Hence the director forces of the group are in general the only systems in respect to which the null points of any plane not a tangent plane lie on the quadric. (3) If G(XX) be zero, then by (2) the three points XRlf XR2, XR3 are collinear and lie on the common null line. The point of contact of the plane is P (X)} which is also collinear with the three points. If G(XX) be not zero, the three points XRX, XR2) XR3 form a triangle on the plane. Hence any point on the plane can be represented by X (Xii?i + \2R2 + X3i?3), that is by XS where 8 is any system of the group. Also P (X) does not lie on the plane X. (4) Similarly if P = x (\YRX + \2R2 + \3#3), then G (PP) = G {sex) {V (R,R,) + V (R2R2) + X32 (R3R3)}. The plane P is a tangent plane of the quadric G, if G(PP) = 0. But this equation involves either G (xx) = 0, or v (RiRi) + V (R ) + V (R3R2) = 0. If G (xx) = 0, the point x lies on the quadric and therefore is contained in one director line of the group and one common null line. The null planes of this point with respect to the various systems of the group all touch the quadric (since G(PP) = Q), hence they all contain the common null line through the point. If G(xx) be not zero, then X12(i^1B1) + \^(R2R2) + \32(R3R3) = 0. But this is the director equation of the group. Hence the director forces of the group are in general the only systems of the group with respect to which the null planes of any point not lying on the quadric touch the quadric.
314 INVARIANTS OF GROUPS. [CHAP. III. If G(xx) be zero, the three planes xR1} xR2, xR3 are collinear, and con- tain the common null line through x. The tangent plane at x is P (x) and also contains this null line. If G(xx) be not zero, the three planes xRly xR2i xR3 are not collinear. Hence any plane through x can be represented in the form Also P (x) does not contain the point x. (5) Let p, ply p2 Pi be the corners of a self-conjugate tetrahedron of a quadric; and let a conjugate set of systems of one of the two triple groups defined by the quadric be Rl=PPl + PlPtPs, R,2 = PP2+ ^PzPly Rs=PPs + f^3PlP2' Also let any point x be defined by p + ^px + %2p2 + %3p3. Then xR1 = fi^pp2p3 + frZiPiP-zlh + PP1P2 + PP1P* and xR2 = - fhgpp Hence xR2 .R3 = ( ppip2p9 Therefore finally G (xx) = (pp^p,)2 {/"i/^3 2 + /U-i?i2 + i* + /^3^32). Accordingly the equation of the quadric, G {xx) = 0, can be written in the form (6) Conversely let the equation, be the equation of the quadric; where x is the point p + 1^ + f2p2 + |3p3. Also let one group of the two reciprocal groups, defined by this quadric, be defined by the conjugate set PP\ + lh.p*P* PP2 + P2P3P1, P Then by comparison with (5) we find a a2 a2 a3 Hence Therefore ^1^2/^2= ttia2a3 Finally Pi= Aac^, ji.2= Aaa.2 and /jl3 \olol3 ; where all the upper signs are to be chosen or all the lower signs, and I/A, is put for
188] TRANSFORMATIONS OF G (pp) AND G (PP). 315 Thus one group associated with the quadric is defined by the three systems pp1 + Xaoiipzps, pp2 + XoLOL.psp^ pp3 The reciprocal group is defined by the three systems pp! - Xaa^ps, pp2 - Xolo^p^ , pp3 - If aa OLs be positive, \ is real; if negative, \ is imaginary.
CHAPTER IV. Matrices and Forces. 189. Linear Transformations in Three Dimensions. (1) Let a real linear transformation of elements in a complete region of three dimen- sions be denoted by the matrix / , as in Book IV, Chapter VI. All the matrices considered will be assumed to be of zero nullity [cf. 144 (2)]. (2) The following notation respecting latent and semi-latent regions, which agrees with and extends that in 145 to 150, will be used. The region, such that for each point x in it x = yx, is called the latent region corresponding to the latent root 7 of the matrix. Any subregion of this latent region will be called a latent subregion corresponding to the latent root 7. (3) The region, such that for each point x in it cf x = yx + y, where y is a point in an included latent region, is called a semi-latent region of the first species. Here two cases arise. Firstly, if 7 be a repeated root, then [cf. 149] there may be a semi-latent region of the first species corresponding to the root 7. Similarly [cf. 150], there may be semi-latent subregions of species higher than the first. Secondly, let yx and y2 be two latent roots, of which either or both may be repeated; and let elt e2 be a pair of latent points belonging to the two roots respectively. Then [cf. 146] any point x = f^ + %2e2 is transformed according to the rule, 4 x = Tifrfi + 72^2^2 = Yi^ + = 72# + Si^i- The region defined by the assemblage of independent latent points corresponding to both yx and y2, that is by the points elt e/, e/', etc., e2, e2r, e", etc., is called the semi-latent region of the first species corre- sponding to the two roots conjointly. If neither of the roots be repeated, such a region is necessarily a straight line. (4) A subregion of a semi-latent region of the ath species, which is not contained in a semi-latent region of a lower species, and which is such that
189, 190] LINEAR TRANSFORMATIONS IN THREE DIMENSIONS. 317 any point x in it is transformed into a point in the same subregion, is called a semi-latent subregion of the ath species. The semi-latent regions, or subregions, which are of most importance in the present investigation, are straight lines. A semi-latent straight line is necessarily of the first species ; so that its species need not be mentioned. It necessarily contains at least one latent point: let ^ be a latent point on such a line, and x any point on it. Then the transformation of x takes place according to the law $x y^x + S . If y2 be not the latent root corresponding to elt then another latent point e2 exists on the line, and the line is a semi-latent subregion with respect to the two latent roots y1 and y2 conjointly. If y2 be equal to ylt the latent root corresponding to elt then, either Sj = 0, and all the points on the line are latent, and the line is a latent region (or subregion); or Sa is not zero, and there is only one latent point on the line. Thus, if 72 = 71, the line is either a latent or semi-latent subregion corresponding to the root 7^ (5) A semi-latent plane is at most of the second species. The trans- formation takes place according to the law where x is any point on the plane, and x is some point (depending in general on x) on a semi-latent line lying in the plane. If the plane be semi-latent of the first species, it necessarily contains lines which are semi-latent subregions of the first species. For let x, above, be a latent point, so that ^ x/ = yx/) then the point \x + fix is transformed accord- ing to the law 4 (\x + px') = X (yx + of) + fiyx' = Xyx + (X + fjuy) x'. Hence any point on the line xxf is transformed into another point on the line xx'; and therefore the line is a semi-latent subregion. 190. Enumeration of Types of Latent and Semi-Latent Regions*. (1) Let the four latent roots 7^ 72, 73, 74, of the matrix be distinct. Then [cf. 145 (5), (6), (7)] there are four and only four independent latent points, one corresponding to each root. Let these points be elf e2) e3, e4. The only semi-latent regions of the first species are the six edges of the tetrahedron e^e^; and each edge is semi-latent with respect to two roots * This enumeration has, I find, been made by H. Grassmann (the younger) in a note to 377 390 of the Ausdehnungslehre von 1862 in the new edition, edited by F. Engel (cf. note at end of this chapter). He gives interesting applications to Euclidean Space. He also refers to von Staudt, Beitrage zur Geometrie der Lage, 3rd Ed., 1860, for a similar enumeration made by different methods.
318 MATRICES AND FORCES. [CHAP. IV. conjointly. The only semi-latent regions of the second species are the four faces of the tetrahedron exe$3eA\ and each face is semi-latent with respect to three roots conjointly. All the manifold necessarily belongs to a semi-latent region of the third species; therefore such species need not be further con- sidered. The enumeration will be made without regard to the difference of type which arises according as the roots are real or imaginary. (2) Let there be three distinct latent roots of the matrix. Let these roots be 7^ 73, 74; and let the root y1 be the repeated root. There are necessarily [cf. 148 (2)] three latent points el9 e2, e4, corre- sponding respectively to the three roots. There is a line e^ which is, either (Case I.) a latent region corresponding to the root 7^ or (Case II.) a semi- latent region corresponding to the root ylt Case I. The line e^ is a latent region corresponding to the root ylt e3 is the latent point corresponding to 73, and e4 is the latent point corresponding to 74. The semi-latent regions of the first species are, the plane eYe$3 corresponding to the roots y^ and 73 conjointly, the plane eYe^ corresponding to the roots y1 and 74 conjointly, the line ese4 corresponding to the roots 73 and 74 conjointly. Any line through es, which intersects e^, is a semi-latent subregion corresponding to the roots yx and 73 conjointly; any line through e4, which intersects e^, is a semi-latent subregion corresponding to the two roots y1 and 74 conjointly. The complete manifold forms a semi-latent region of the second species corresponding to all the roots conjointly. All planes through the line e3e4 are semi-latent subregions of the second species corresponding to all the roots conjointly. Case II. The point ex is the sole latent point corresponding to the root 7!, the points e3 and e4 are the latent points corresponding to the roots 73 and 74. The semi-latent regions of the first species are, the line e^ corre- sponding to the root 7^ the line eYe3 corresponding to the roots yY and 73 conjointly, the line exe4 corresponding to the roots yx and 74 con- jointly, the line e3e4 corresponding to the roots 73 and 74 conjointly. The semi-latent regions of the second species are the planes eYe 3i e^e^u exeze4\ the roots to which they correspond need not be mentioned. (3) Let the matrix have a triple latent root yj; and let 74 be the other root. There are necessarily two latent points el and e4 corresponding to these roots respectively. There is [cf. 149, 150] a plane e1e2e3 which is, either (Case I.) a latent region corresponding to the root 7^ or (Case II.) a semi- latent region of the first species corresponding to the root ylt or (Case III.) a semi-latent region of the second species corresponding to the root 7^
190] ENUMERATION OF TYPES OF LATENT AND SEMI-LATENT REGIONS. 319 According to 150 (5), the points elf e2, e3 can always be so chosen that where S2 cannot vanish unless S2 also vanishes. Thus, in Case I. S = 0 = S2; in Case II. 8j = 0, and S2 is not zero; in Case III. neither S1 nor S2 vanishes. Any point # = 2 e is transformed according to the rule t x = (Yi i + si?2 ei + (Yi + ^s) e2 + y^3e3 + y 4e4. Thus in Case I., f)% = yxx -f (74 71) % ...........................(A). In Case II., / # = y,x + S2 3e2 + (74 - 7l) 4e4 .....................(B). In Case III., $x = yxx + + 82fsea + (74 - 7l) ^4..................(C). Case I. The plane 6]e263 is the latent region corresponding to the triple root y1, and e4 is the latent point corresponding to the root 74. Any line in the plane eje2e3 is a latent subregion corresponding to the root 7l. The complete manifold is the semi-latent region of the first species corresponding to the two roots conjointly. Any line through e4 is a semi- latent subregion corresponding to the two roots conjointly. Case II. The latent regions are, the line e^ corresponding to the root 7l, and the point e4 corresponding to the root y4. The semi-latent regions of the first species are, the plane eYe2ez corresponding to the root 7l, and the plane e e^. There is some one point [cf. 149 and 150 (5)] in the line e^ (the point e2, according to the present notation), such that any line through it in the plane eYe2ez is a semi-latent subregion corresponding to the root 7l. Also any line through e4 in the plane e-^e^ is a semi-latent subregion corre- sponding to the roots 7l and 74 conjointly. The semi-latent region of the second species is the complete manifold. Any plane through e2e4 is a semi-latent subregion of the second species; for from equation (B) e2e4 j % = e2e4x. Case III. The only latent points are, the point eY corresponding to the root 7l, and the point e4 corresponding to the root 74. The semi-latent regions of the first species are, the line exe2 corresponding to the root 7l, and the line e1e4 corresponding to the roots 7l and 74 conjointly. The semi-latent regions of the second species are the planes exe^z, and the planes exe2e4. (4) Let there be only one latent root to the matrix, which occurs quadruply. Let 7l be this root, and let ^ be the latent point corresponding to it which necessarily exists [cf. 148 (2)]. Then [cf. 150 (5)] it is always possible to find three other points e2i e3, e4} such that Then any point x = X%e is transformed according to the rule ?2 + S2 3) e2 + (y^3 + S3f4) e3 + 7l 4e4 -f S3 4e3.......................................(D).
320 MATRICES AND FORCES. [CHAP. IV. Five cases now arise according as, either (Case I.) Slt B2 and S3 all vanish; or (Case II.) Sj and S2 vanish, but B3 does not vanish; or (Case III.) S2 vanishes, but S2 and S3 do not vanish ; or (Case IV.) Sly S2, B3 do not vanish ; or (Case V.) S2 vanishes, but Sx and S3 do not vanish. Thus in Case I., cj)X = ry1x.............................................(E). In Case II., $x = y1x+ B3^e3....................................(F). In Case III., j)x = yxx + S2 3e2 -f B3^e3 ........................(G). In Case IV., f*x = 7la? + 8^ + 8 3e2 + 83 e3 ...............(H). In Case V., / # = 7i# + S^ + 83 4e3...........................(I). Case I. Every point is latent; and the operation of the matrix is simply equivalent to a numerical multiplier. This case need not be further con- sidered ; and may generally be neglected in subsequent discussions. Case II. The latent region is the plane eYe2e3. The semi-latent region of the first species is the complete manifold. Every line through the latent point e3 is a semi-latent subregion. All semi-latent planes must pass through e3; and all planes through e3 are semi-latent. Case III. The latent region is the line exe2. The semi-latent region of the first species is the plane e-fifa. Every line through e2 and lying in the plane eYe2e3 is a semi-latent subregion. The semi-latent region of the second species is the complete manifold. Every plane through the line e2e3 is a semi-latent subregion of the second species ; for by equation (G) e2e3(j x = e2e^x. Case IV. The sole latent point is e2. The semi-latent region of the first species is the line exe2. The semi-latent region of the second species is the plane e^e2e3. The semi-latent region of the third species is the complete manifold. Case V. The latent region is the line exe^ The semi-latent region of the first species is the complete manifold. Every point lies on some straight line which is a semi-latent subregion. For consider the condition that the point #(=2 e) may lie on a semi-latent straight line through the latent point a(=a1e1 + a3e3). Now f)x = jjX + S^i + S3 4e3 = 7i# + Xa, by hypothesis. Hence Sj 2 = Xai, S3f4 = \a3. Thus x must lie on the plane of which the equation is This plane passes through a; and all lines lying in it which pass through a are semi-latent. A planar element in this plane is eYe3 (a^^ + ^3^164). (5) Let there be two distinct latent roots, and let each be once repeated. Let 7j and 73 be the two distinct latent roots ; and let ex and e3 be the two latent points, which certainly exist, corresponding to these roots respectively.
190] ENUMERATION OF TYPES OF LATENT AND SEMI-LATENT REGIONS. 321 Then it is always possible to find two points e2 and e4, such that f e2 = 7^2 + 8^!, $e4 = y3e4 + B3e3. Hence any point x = 2 e is transformed according to the rule (f x = (7^ + S ) ex + y 2e2 + (73^ + S3 4) e3 + y3%4e4. Three cases now arise according as, either (Case I.) B1 and S3 both vanish ; or (Case II.) S1 vanishes and S3 does not vanish; or (Case III.) neither Sx nor S3 vanishes. The case, when S3 vanishes and S2 does not vanish, is not a type of case distinct from Case II. Thus in Case I., (f x = 7lx + (73 - 7l) (f3$, + f4e4) = 73# + (7! - 73) ( 01 + 2e2)......(J). In Case II., $x = yxx + {(73 - 7l) 3 + 8 4} e3 + (73 - 7l) |^4e4 = 7^ + (7! - 73) ( ei + ?ae2) + 3^463.........(K). In Case III, /. The latent regions are, the line e 2 corresponding to the root y1} and the line e3e4 corresponding to the root 73. The semi-latent region of the first species is the complete manifold, and it corresponds to the two roots conjointly. Every line intersecting both exe2 and e3e4 is a semi-latent sub- region corresponding to the two roots conjointly. Case II. The latent regions are, the line e^ corresponding to the root y1} and the point e3 corresponding to the root 73. The semi-latent regions of the first species are, the line e3e4 corresponding to the root 73, and the plane exe$3 corresponding to the roots yx and 73 conjointly. Every line through e3 in this plane is a semi-latent subregion corresponding to the two roots conjointly. The semi-latent region of the second species is the complete manifold. Every plane through e3e4 is a semi-latent subregion. Case III. The only two latent points are, ex corresponding to the root y1} and e3 corresponding to the root 73. The semi-latent regions of the first species are, the line eYe2 corresponding to the root 7^ the line e3e4 corresponding to the root 73, the line exe3 corre- sponding to the roots y2 and 73 conjointly. The semi-latent regions of the second species are, the plane eYe^e3 corresponding to the roots 71, 71, 73 con- jointly, and the plane eYe3e4 corresponding to the roots 7^ 73, 73 conjointly. It is to be noticed that, in order to define the roots corresponding to the semi-latent regions of the second species, the repeated roots must be counted twice. The semi-latent region of the third species is the complete manifold. W. 21
322 * MATRICES AND FORCES. [CHAP. IV. 191. Matrices and Forces. (1) Let S denote any system of forces, and f any matrix. Then f 8 is denned in 141 (1) and (3), and denotes another system of forces. If cj S = S, the system 8 is said to be a latent system of the matrix p [cf. 160 (1)]. If every system S, belonging to a group of systems G, is transformed into another system cf)S of the same group, the group G is said to be a semi- latent group of the matrix. If every system of the group is latent, the group is said to be latent. (2) The following properties of the transformation are immediately evident. If (SS') = 0, then ( / / ') = ( f . 88') = 0. Hence if 8 and 8' are reciprocal, f S and $' are also reciprocal; and if S reduce to a single force, f S reduces to a single force. If belong to the group denned by 8l9 S2, ... Spy then / S belongs to the group f 8l9 i S2, ... cj)Sp. (3) Let e^e^t be any fundamental tetrahedron of reference; and let 8 = /SJ"i2^i^2 + ^23^3 + ^31^1 + ^14^1^4 + ^24^4 and f S = 5712/e1e2 + ^13^2^3 + wsi'^i + ^u^i + Then it is obvious [cf. 141] that the coefficients txr12, ..., ^M are transformed into the coefficients 'st12/, ..., vtu' by a linear transformation represented by a matrix of the sixth order. But the most general matrix of the sixth order contains thirty-six constants; whereas the matrix of the fourth order, from which the present transformation is derived, contains only sixteen constants. Accordingly relations must hold between the thirty-six constants reducing the number of independent constants to sixteen. (4) An interpretation of these relations can be found as follows. In general the transformation of the sixth order yields only six latent systems: and in general a matrix of the fourth order has four latent points forming a tetrahedron. The six edges of this tetrahedron are latent forces. Hence in general the six latent systems are six single forces along the edges of a tetrahedron. The expression of these conditions yields twenty independent equations, which reduce the number of independent constants to sixteen. (5) Let the corners of the tetrahedron e e^ be the latent points of the matrix 0. This is always possible, when the roots ylt y2, 73, 74 are unequal. Then the latent roots of the matrix of the sixth order, which transforms the co-ordinates of any system of forces, are given by the sextic (cr - 7X72) (o- - 7374) (o- - 7273) (a - 7^4) (a - yf/J (a - y2y4) = 0.
191, 192] MATRICES AND FORCES. 323 (6) Apart from the special cases when some of the roots of the matrix / are repeated, the only cases, for which latent systems exist other than forces along the six edges of the tetrahedron, arise when two roots of this sextic are equal [cf. 145 (4) and 148]. There are two types of such equality; namely, the type given by 7172 = 7374, and the type given by 7273 = 7371. The second type necessitates y1 = y2; and thus leads back to the special cases when the matrix f possesses repeated latent roots. (7) It is evident from (3) that, if S1, S2}... S6 be any six independent systems, and any system 8 be defined by and f 8 = X/Sj + \2S2 + ...+ X6' 6, then X/, X2', ...X6', can be derived from \lf Xg,... X6 by a linear transformation. 192. Latent Systems and Semi-Latent Groups. (1) A force on any semi-latent line is a latent force. For let e^e2 be the semi-latent line, and ex the latent point on it. Then [cf. 189 (4)] we may assume, / = 72#2 + 8^!, f e1 = 7^. Hence exe2 = (fre^e* = y^n^i^- (2) If one of two conjugate forces of a latent system be latent, the other force is also latent. For let S = \D1 + /u,D2; and assume that j S = S, f Dl = Dj. Then since f S = X^A + fi(f D2 = XA + ftD29 it follows that f)D2 = D2. Also from subsection (1), (f D1 = 77'!^, and f D2 = y"y'"D2; where 7, 7', 7", y'" are latent roots of the matrix, but not necessarily distinct roots. Thus 77r = 7V" This forms another proof of 191 (6). (3) Hence, if a latent system 8 (not a single force) exist and also a semi-latent line which is not a null line of S, then another semi-latent line not intersecting the first must exist, and such that the two lines are conjugates with respect to S. (4) The null plane of a latent point with respect to a latent system is semi-latent. For let e be the latent point, S the latent system, P a planar element in the null plane. Then P = XeS, ) P = \ l eS = \ l e t S = eS = P. Conversely the null point of a semi-latent plane with respect to a latent system is latent. (5) Hence from (2), (3), (4) it is easy to prove that, if a tetrahedron of latent points exist, every latent system must have two semi-latent lines as one pair of conjugates. 21 2
324 MATRICES AND FORCES. [CHAP. IV. (6) If a group is semi-latent, its reciprocal group obviously is also semi-latent. (7) In general a semi-latent group of p 1 dimensions (p ^ 6) contains p latent systems. This follows from 191 (3) and (7). (8) If no special relation holds between the latent roots of the matrix, then [cf. 191 (4)] the only six latent systems are six single forces on the six edges of the tetrahedron formed by the four latent points. It follows that in general a semi-latent group of (p 1) dimensions must have p edges of the fundamental tetrahedron of the matrix as director lines. Thus in general a dual group of the general type can be a semi-latent group, namely, any dual group with two non-intersecting edges of the fundamental tetrahedron as director lines. Also [cf. subsection (6)] in general a quadruple group of the general type can be a semi-latent group, namely, any quadruple group with two non-intersecting edges of the funda- mental tetrahedron as common null lines. But in general a triple group of the general type cannot be semi-latent. For the director lines of a triple group, which are generating lines of the same species of a quadric, do not intersect unless the quadric degenerate into a cone or into two planes. But there are not three non-intersecting edges of a tetrahedron. (9) A matrix can always be constructed with four assigned latent roots, so that any given dual group (or any given quadruple group) is semi-latent. For it is only necessary to choose two of the latent points on one director line (or common null line), and two on the other director line (or common null line). (10) Every semi-latent dual group, which is not parabolic and does not consist of single forces, contains at least two distinct latent systems, which are either the director forces or two reciprocal systems. For, if A and A be the director forces of the group, then A and A remain the director forces of the group after transformation. Hence either j D1 = Dl9 and j D2 = D2; or A = A, and A = A- In the first case the two director forces are the two latent systems. In the second case, let / A = aD2, f D2 = /3A- Let one of the latent systems be S = XA + jjlD2. Then $8 = rS = X A + /* / D2 = Xa A + p/3 A Hence o-X = /u./3, ct/j, Xa. Therefore a2 = a/ Hence =+_ ..
192] LATENT SYSTEMS AND SEMI-LATENT GROUPS. 325 Thus the two latent systems are and and these systems are reciprocal [cf. 173 (6)]. But reciprocal systems of a dual group of the general type are necessarily distinct. (11) The director force of a semi-latent parabolic group is evidently latent. Also [cf. 172 (9)] the null plane of any point on the director line is the same for each system of the group. Hence the null plane of every latent point on the director line is semi-latent. Thus there must be as many semi- latent planes passing through the director line as there are null points on it, for a semi-latent parabolic group to be possible. And conversely there must be as many latent points on the director line as there are semi-latent planes through it. (12) If a semi-latent line exists which does not intersect the director line of a semi-latent parabolic group, then the parabolic group contains at least one latent system in addition to its director force. For let D be its director force and N the semi-latent line. Then [cf. 172 (10)] one and only one system of the group exists, for which N is a null line. This system is D {N8) S (ND) ; where S is any other system of the group. But since the group is semi-latent, and the lines JV and D are semi-latent, this system must be latent. (13) Let D = 0J02, and N=ese4; and let e1 and e3 be the latent points on D and N which certainly exist. Then e^ is a semi-latent line. Now by subsections (3) and (4), either e^ is not a null line of the latent system, and then its conjugate is also a latent line intersecting both D and N in two other null points; or exe2 is a null line of the system. In this last case e2 is the null point with respect to the latent system of the plane e^N, since the two null lines e^ and JV intersect in it: also ex is the null point of the plane ej) for all the systems of the group [cf. 172 (9)], since the null lines and D intersect in it. (14) It follows from (12) that the only possibility, for the existence of a semi-latent dual group with only one latent system, is when all the semi- latent lines intersect one of their number. Then such a semi-latent group is parabolic, and the director force is the single latent system, and is on that one of the semi-latent lines intersected by all the rest. Furthermore, if two non-intersecting semi-latent lines exist, and no latent system exists which is not a single force, then no semi-latent parabolic group with either of these lines as director line is possible. For by (12), such a group must contain a second latent system, and by hypothesis such a system does not exist.
326 MATRICES AND FORCES. [CHAP. IV. (15) Every semi-latent triple group of the general type must contain at least three distinct independent latent systems, unless it contains a semi-latent dual group with only one latent system. This is obvious, remembering that the properties of semi-latent triple groups are particular cases of the properties of the linear transformation of points in a complete region of two dimensions [cf. 145 (4) and 148]. Also by the preceding subsections such a semi-latent dual group is parabolic. Hence, if for any matrix no semi-latent parabolic group with only one latent system exists, every semi-latent triple group must have three distinct latent systems. (16) If a plane region of latent points exists, every director force D of a semi-latent triple group intersects this plane, and therefore has a latent point on it. Thus D and / D are either congruent or intersect. If they intersect, the triple group is not of the general type, since \ D is also a director force. Hence the only possible type of semi-latent triple group is latent. (17) If a line of latent points exists, this line is either a generator of any quadric or intersects it in two points. Now consider the quadric defined by any semi-latent triple group G. The reciprocal group Gf is also semi-latent. Firstly let the latent line intersect the quadric. Then by the same reasoning as in the previous subsection (16), two director forces of G and two of G' must be latent, assuming that G is of the general type. Secondly let the latent line be a director line of G. Then as in (16) all the director forces of G' (null lines of (?) are latent. Thirdly let the latent line be a null line of G'. Then all the director forces of G are latent. 193. Enumeration of Types of Latent Systems and Semi-Latent Groups. (1) Let no two roots of the sextic of 191 (5) be equal. The four latent roots of the matrix / are unequal, and only four latent points 0i #2, e3, e4 exist. Then [cf. 191 (6)] the only latent systems are the six forces on the edges of the tetrahedron e-fi^e^ By 192 (10) the only semi-latent dual groups, not parabolic, have two edges of the tetrahedron as director lines. No semi-latent parabolic group can exist, for by 192 (12) and (14) a latent system of the group must exist which is not a director line; and there is no such system. No semi-latent triple group of the general type can exist. For [cf. 192 (15)] such a group must contain three latent systems, that is, three edges of the tetrahedron e^e.^ as director lines. But three non-intersecting edges cannot be found.
193] ENUMERATION OF LATENT SYSTEMS AND SEMI-LATENT GROUPS. 327 The semi-latent quadruple and quintuple groups can be found by the use of 192 (6). Thus it is not necessary to enumerate them. (2) Let the four roots ylt y2i 73 74 of the matrix be unequal, and [cf. 191 (6)] le* 7i7a=7274- Case I. Let no other roots of the sextic of 191 (5) be equal. Then, as in (1), the four latent points exe2e3e4 form a tetrahedron. The latent systems are the six single forces on the edges of the tetrahedron, and any system of the type Xe 3 + fie2e4. It can be seen by the use of 192 (2) that no other system can be latent. The dual group defined by e1e3 and e2e4 is therefore latent. The dual groups defined by any system of the type Xe1e3 + fxe2e4 together with e^, or e3e4, or exe4, or e3e2 are semi-latent. They are parabolic groups. The only semi-latent dual groups, not parabolic, have two edges of the tetrahedron as director lines. By 192 (14) no semi-latent dual group exists with only one latent system. The semi-latent triple groups of the general type are, all groups of the type defined by Xe^e3-\- pe2e4y e^, e3e4; and all groups of the type defined by \e1e3 + fie2ei, exeAy e2e3. The generality of these types is proved by 175 (9) and (12). By 192 (15) no other semi-iatent triple group (of the general type) exists. Case II. Let ylJz = y2y4} and y^ = y2y3. Then y2 = y22} y32 = y42. Hence, excluding the case of equal roots which is discussed later [in subsection (5)], Yi = -7a 73 =-74- As in Case I. there are only four latent points, which form a tetrahedron. The latent systems are, the six single forces on the edges of the tetrahedron, and [cf. 191 (6)] any system of the type Xexe3 + fjie2e4, and any system of the type XeYe4 4- fie2e3. The semi-latent dual groups are, the semi-latent group defined by eYe2 and e3e4 the latent group defined by exe3 and e2e4, the latent group defined by e1e4 and e2e3, any group defined by systems of the types \e1e^ + fjLe2e4 and Ve^ + fjfe2e3. This last type of semi-latent group is not parabolic, unless one of the four quantities X, //,, X', // vanishes. The semi-latent triple groups of the general type are all groups defined by sets of three systems of the following types : ^2, e3e4i \ele3 #104, e2e3, e^, e2e4, Thus there are four types of semi-latent triple groups for this case. There appears to be a fifth type of group of which a typical specimen is the group defined by Xe^e3 + fie2e4, X'e^ -\- /jue2e3, exe3. But this group cannot
328 MATRICES AND FORCES. [CHAP. IV. be of the general type. For from 175 (9) only one parabolic subgroup of a triple group (of the general type) exists with a given director line of the triple group as its director line. Whereas in the group above mentioned, eYe3 is the director line of two such subgroups. By 192 (15) no other semi-latent triple group of the general type can exist. (3) Let 7i = 72- Then from 190 (2) there are two cases to be con- sidered ; but two extra cases arise, when the relation y? = y374 is satisfied. Thus there are four cases in all; in the first two cases it is assumed that Yi2 4= 7374- Let the notation of 190 (2) be adopted. Case I. [Cf. 190 (2), Case I.] The line e^ is the latent region corre- sponding to the root 7^ Then from 190 (2) the latent systems are, e^, e3e4, any force of the type e3 (X^ + //,e2), any force of the type e4 (X^ + fie2). From 192 (3) no other latent systems exist. Hence no latent systems exist, which are not single forces. The semi-latent dual groups (not consisting entirely of single forces) are the group defined by e^, e3e4\ and the groups of the type defined by e3 (\0! + fjte2), e4 (Vei + fi'e*). No semi-latent parabolic group exists [cf. 192 (14)]. No semi-latent triple group of the general type exists; since three non- intersecting latent forces do not exist, and the only latent systems are single forces. Case II. [Cf. 190 (2), Case II.] The line eYe.2 is the semi-latent region corresponding to the root 7^ The latent systems are the forces exe^ exe3, e1e4) e3e4. It is impossible for any latent system (not a single force) to exist, since e^, e^, e^e4 are not coplanar, and cannot therefore all be null lines. Hence the theorem of 192 (3) applies ; and the truth of the state- ment can easily be seen, since Xe^ + fie3e4 is n t latent. The only semi-latent dual group, not consisting entirely of single forces, is that defined by e^, e3e4. No semi-latent parabolic group can exist [cf. 192 (11) and (14)]. No semi-latent triple group of the general type can exist [cf. 192 (15)]. Case III. (Subcase of Case I.) Let yf = 7374; and let the arrangement of latent and semi-latent points and regions be that of Case I. Then, from Case I. and 192 (3) the only latent systems are, eYe2i e3e4, any force of the type e3 (X^ + fie^), any force of the type eA (X^ + yue^ and any system of the type \eYe L 4- pe3eA. The semi-latent dual groups (not consisting entirely of single forces) are the (latent) group defined by exe2) e3e4; any group of the type defined by e3 (Xex + /j,e2), e4 (X7^ -h pe^); the parabolic groups of the types defined by
198] ENUMERATION OF LATENT SYSTEMS AND SEMI-LATENT GROUPS. 329 + fie3e4 together with either e3 (X'gj + /j,'e2), or e4 (X'tfj + /u/e2). No other semi-latent parabolic groups exist [cf. 192 (14)]. The semi-latent triple groups are groups of the type defined by Xe^z + fie3e4, e3 (X'^ + /e2), e4 (\"e1 + fi"e2). Case IV. (Subcase of Case II.) Let 7^ = 7374; and let the arrangement of latent and semi-latent points and regions be that of Case II. Then the latent systems are the forces eYe2, exez, eYe4i e3e4, and any system of the type X^!^2 + pe3e4. By 192 (3) no other latent systems exist. The semi-latent dual groups, not consisting entirely of single forces, are, the (latent) group defined by exe2y e3e4, all parabolic groups of the types defined by \exe2 4- /jbe3e4 together with e^e3 or exe4. No other semi-latent parabolic groups can exist [cf. 192 (11) and (14)]. No semi-latent triple group of the general type can exist. (4) Let 71 = 72=73. Then from 190 (3) there are three cases to be considered. Case I [Cf. 190 (3), Case I.] The plane exe^ez is the latent region corresponding to the root 7^ The latent systems are, any force in the plane ele2e3) any force through the point e4. No other latent system exists. For all the latent lines in the plane e^e2e3 cannot be null lines, since all the null lines lying in a plane must pass through the null point. Thus some line be lying in the plane exe2e3 may be assumed not to be a null line of any such latent system. Then by 192 (3) the system must be of the form \e4a + fibc, where b and c lie in the plane #1 #2 03 and e4a is any line through e4. Let a be assumed to be the point in which this line meets the plane #$, Hence cf (\e4a + fibc) = 7i74Xe4a + y^fibc. Therefore such systems are not latent. The semi-latent dual groups are all groups of the type defined by e4a and bcy where be lies in the plane exe,2e3. No semi-latent parabolic group can exist [cf. 192 (14)]. No semi-latent triple group of the general type can exist [cf. 192 (16)]. Case II. [Cf. 190 (3), Case II.] The line e 2 is the latent region corresponding to the root 7^ and the plane e1e^e3 is the semi-latent region of the first species corresponding to the first root. Also e2 and e3 are such that (j e3 = y1e3'\-S2e2) where S2 is not zero. The latent forces are, e^, any force through e2 in the plane exe2e3, any force through e4 in the plane e^^^ No other latent systems exist. For a similar proof to that in Case I. shows that a latent system, not a single force, cannot have two non-intersecting semi- latent lines as conjugate lines. Hence such a system must have all the semi- latent lines as null lines. Therefore e2 must be the null point of the plane
330 MATRICES AND FORCES. [CHAP. IV. e es, and e4 the null point of the plane e e*. But since the null line e^ does not go through e4) which is the null point of the plane e e^ this is impossible. The semi-latent dual groups (not entirely consisting of single forces) are all groups defined by a force through e.2 in the plane eYe.2ez and a force through e4 in the plane e^e^. No semi-latent parabolic group exists [cf. 192 (14)]. No semi-latent triple group of the general type exists [cf. 192 (15) and (17)]. Case III [Cf. 190 (3), Case III.] The only latent points are, the point eY corresponding to y1} and the point e4 corresponding to 74. The only latent systems are, the force e^, and the force eYe . There can be no other latent system (not a single force). For by 192 (4) the null point with respect to such a system of the semi-latent plane eYe$z is ei, and therefore the null point of the semi-latent plane e e^ is e4. Hence the line eYe is not a null line of such a system ; and therefore from 192 (3) another semi-latent line, not intersecting e^, is required. But such a line does not exist. There are no semi-latent dual groups, not consisting entirely of single forces. For the only possibility of such a group lies in the possibility of a semi-latent parabolic group with only its director force latent [cf. 192 (14)]. But e^t cannot be the director force of such a group, since by 192 (11) there must be as many semi-latent planes containing e2e4 as there are latent points on eYe . But ex and e4 are both latent points; while e^e2eA is the only semi-latent plane through e^e^ Again e^ cannot be the director force, for by 192 (11) there ought to be as many latent points on it as there are semi-latent planes through it. But there are two semi-latent planes through it, namely e 2ez and e e^ and there is only one latent point on it. There are no semi-latent triple groups of the general type [cf. 192 (15)]. (5) Let there be only one latent root y1 of the matrix. Then [cf. 190 (4)] there are five cases to be considered: but, of these, the first may be dismissed at once. Case II. [Cf. 190 (4), Case II.] The latent region is the plane eYe^ez. The latent systems are of the type \e3a + yJbe, where a is any point and the force be is any force lying in the plane exe$z. If S be any one of these latent systems, f S = y^S. The semi-latent dual groups (not consisting entirely of single forces) are, all groups defined by the forces e3a and be, where a is any point and be is any line lying in the plane e e^ and any parabolic group with a director force of the type e3d, where d lies in the plane eYe$z. All the semi-latent dual groups are thus either latent, or else possess only one latent system.
193] ENUMERATION OF LATENT SYSTEMS AND SEMI-LATENT GROUPS. 331 In order to prove the above statements, first notice that, if fifj and S2 be the latent systems of a semi-latent group with two distinct latent systems, f S1 = y12S1 and f S2 = y12S2. Hence the group is latent, and hence the director forces are latent, if there are two of them. Accordingly the director forces of a non-parabolic semi-latent group are of the type described. Again [cf. 192 (11)] the number of latent points on the director line of a semi-latent parabolic group is equal to the number of semi-latent planes passing through it. Hence a force of the type e3d is the only possible director force of such a group [cf. 190 (4), Case II.]. Now if d be any point on this director line, a system 8 of one of the parabolic groups with this line as director line can be written in the form S = Xe^x + ^dy, where x and y are any points [cf. 172 (9)]. Now the dual group defined by e3d and S is semi-latent. For [cf. 190 (4), equation (F)], j S = \ j es f x + ^ Thus fiS also belongs to the dual group. Furthermore, if 8' be not zero (that is, if dy do not lie in the plane ), esd is the only latent system of the group. But if 8' be zero (that is, if dy do lie in the plane 0i0203), every system is latent and the group is therefore latent. No semi-latent triple group of the general type exists [cf. 192 (16)]. Case III. [Cf. 190 (4), Case III.] The latent region is the line e.e,. The latent systems are all forces in the plane e^e2e3 through the point e2. There are no other latent systems: for all the planes through e2e3 are semi- latent ; and their null points with respect to any latent system must be latent [cf. 192 (4)]. But e2 is the only latent point on all these planes. There are no semi-latent dual non-parabolic groups (not consisting entirely of single forces). For there are evidently no semi-latent dual groups (not single forces) with two latent systems. There are semi-latent parabolic groups of the type defined by e2a and ab + \e2k; where a is any point on the plane exe2e3, but not on eYe2 or e2e3 [cf. 192 (11)], b is any point on the plane 0i0203, k is any point, and X is determined by a certain condition. In order to prove this, note that the form assumed is obvious from 192 (11) and from the consideration, that the null point of the semi-latent plane exe2e3 is latent and lies on the director force e2at and must therefore be e2. Now let a = U + O262 + a3e3t where neither ax nor a3 is zero; let b = P + fi2e2 + /33e3; let k = Kxex + tc2e2 + K3e3 + k^.
332 MATRICES AND FORCES. [CHAP. IV. Then, using 190 (4), equation (G), f) (ab -f Xe2k) = 7^ (ab + Xe2k) + S2y1 (a3e2b ft3e2a) + Xy183fc4e2e3. Now (ab + Xe2 ) must belong to the group defined by ab + Xe2 and ae2. Hence Hence S^a^ab XyT?)3ic4e2e3a = 0. (It is easy to deduce from this equation another proof of the limitation of the position of a, namely that it is not to lie on e^e2 or on e2e3.) Again e2ab = e1e2e3 = e2e3a = ae2e3a, say. Then \-**m. These semi-latent parabolic groups have only one latent system. There are no semi-latent triple groups of the general type [cf. 192 (17)]. Case IV. [Cf. 190 (4), Case IV.] The sole latent point is the point ^. The only latent system is the force eYe2. The semi-latent dual groups are parabolic groups, with exe2 as axis, and with e1 as the null point of the semi-latent plane e^e^. Such semi-latent groups have only one latent system, namely the director force e^e2. Also all such groups are not semi-latent; one condition has to be fulfilled, which will be investigated as follows. Let h = K + K2e2 + tc3es + /e4e4, where ka is not zero ; let a = a^i + a2e2, where a2 is not zero; and let b = ^Yex + @2e2 + f33e3f where /33 is not zero. Then a parabolic group of the specified type is defined by exa and ejc + Xab. If k, a and b are given, then X is determined by a condition, which is found as follows. From 190 (4), equation (H), remembering that eYe2 = eYa = e2a, (f (exk + Xab) = 7^ (ejk + Xab) + where f need not be calculated. But / (ejc + Xab) belongs to the parabolic group. Thus 7i83tf4e1e3 + Xy^a^e^b = exe2. Hence y^Kfae^ + Xy1 1a2e1e2b = 0. Also e^jb = $3eYe2e3. Therefore 7iS3 4 + Xy^a2^3 0. But 7x, Slt Os, ^3, k4 are not zero. Hence X = No semi-latent triple group exists. For if such a group existed, e^ must be a director line. But the reciprocal group must also be semi-latent, and therefore it must also contain exe2 as a director line.
193] ENUMERATION OF LATENT SYSTEMS AND SEMI-LATENT GROUPS. 333 Case V. [Cf. 190 (4), Case V.] The latent region is the line e . Those latent systems which are single forces are grouped in planes: thus corre- sponding to the latent point a (= a^ + a3e3), there are an infinite number of latent forces of the type ax; where x is any point on the plane of which the equation is a38i?2 = Aft. Any point x lies on the line of a latent force. For x(f x = x (S^i + S3|4e3). Hence f (x j)x) = y^x (S^^ + S3fte3) = jiX(f x. Also the force e^e3 is latent. The only latent systems, which are not single forces, are formed by any two latent forces as conjugate forces: thus, if x and y be any points, x(f x + ycf y is a latent system. And if S be such system (j S 7l2$. Also it may be noted that e^e3 is a null line of all such systems. There is no a priori impossibility in the existence of latent systems with all the semi-latent lines as null lines. The following investigation shows that such a latent system does not exist. For, since e^e3i e^, e2e3 are to be null lines, systems, with all the semi- latent lines as null lines, must be of the form fts^s + ^14^1^4 + This form is found by assuming then the following equations must hold Also any point a (= a1el + a3#3) on the line exez is the null point of the semi-latent plane ele3(a1Bse2-\-oL^e^, for all values of the ratio of ax to a3. Hence e^ (aA ?2 + a3Si^) . ( 13^3 + fi^i^i + ft^s) = ai^i + s s; that is aAft40i + ^1^3 = * + a3e3. Hence systems with all the semi-latent lines as null lines must be of the type But by operating on such a system with the matrix, it is easy to see that such a system is not latent. For, if S be the system, Any dual group defined by two latent systems is latent, and has therefore (unless it be parabolic) two latent forces as director forces. Any parabolic group with eYe3 as director line is semi-latent. Such a group is either latent, or has only one latent system (the director force). For, let a and b be any two (latent) points on e^, and let x and y be
334 MATRICES AND FORCES. [CHAP. IV. any other two points. Then f a = y1a) j)b = y1b, j x = y1x + x'i f y = y$ + y''; where x and y are points on eYez. Hence, if 8 be the system ax + by, since ax' = 63/' = eYez. Thus S and exez define a semi-latent parabolic group. If 8 be latent (that is, if \ be zero), the group is latent. The semi-latent triple groups (not entirely single forces) are of two kinds, which will be called, type I. and type II. A semi-latent group of type I. is defined by any three non-intersecting latent forces, namely by oo f x9 y j y z$z. Any such group is latent. The groups of type II. are the groups reciprocal to those of type I. Thus gjgg, which is a common null line of every group of type L, is a common director line of every group of type II. Also, since every other semi-latent line intersects eYez, this force is the only latent director force of any group of type II. Again it has been proved that every semi-latent line is a null line of the system B-^e^ + B^e^. Hence this system is reciprocal to every system of every semi-latent group of type I. Accordingly the semi-latent parabolic subgroup defined by e^, B^e^ + Bse2e3 is a subgroup of every reciprocal semi-latent group of type II. The force eYe3 is the only latent system of this common subgroup, since the system B^e^ + B3e2e3 is not latent. Thus a semi-latent group of type II. has only one latent system, namely exez. For, by 175 (9), a director force (eYe3) of a triple group can only be a null line of one dual subgroup of systems; and e es is a null line of every latent system. Semi-latent groups of type II. are defined by exez, Bje^-h B3e2e3, and any system 8. For the system S can be written Hence (f S = y^S + (t*r23S171 -f ^uB^ -f *t2 ) e + ^Ti (B^e^ + S^e^). Thus f S belongs to the triple group defined by S, e^ and B^e^ -f B3e2e3. The other semi-latent parabolic groups cannot belong to any semi-latent triple group. For the reciprocal to such a group must be another semi- latent triple group. But groups of type I. and II. are respectively reciprocal in pairs : so this reciprocal semi-latent group must contain another semi- latent parabolic subgroup. Thus two reciprocal triple groups would each contain the director line e ez in common. This is impossible for triple groups, not of a special kind. (6) Let the roots of the matrix j be equal in pairs ; so that 7l = y2, and ry3 = y4. Then [cf. 190 (5)] there are three cases to be considered, as far as the distribution of latent and semi-latent points and regions is concerned.
193] ENUMERATION OF LATENT SYSTEMS AND SEMI-LATENT GROUPS. 335 But each of these three principal cases gives rise to another case, in which the additional relation 71 = 73 is fulfilled. Thus there are six cases in all. Case I. [Cf. 190 (5), Case L] The latent regions are the lines e^2 and e3e4. The latent systems are the forces e^, e3e4) and any system of the quadruple group which has e^ and e3e4 as common null lines. The semi-latent dual groups are defined by any two of these latent systems; since by 192 (14) no semi-latent parabolic group exists with only one latent system. The semi-latent triple groups are defined by any three of these latent systems. There are only two types of semi-latent triple groups, of a general kind: let them be called type I. and type II. Type I. consists of groups defined by three systems of the form, ab, a'b'', a"b", where a, a', a" lie on eYe2, and b, b'y b" lie on e3e4. Type II. consists of groups defined by three systems of the form exe2i e3e4, \ab + fia'b'. Each group of type II. is reciprocal to a corresponding group in type I. Groups defined by two forces of the form ab and a'b', together with either eYe2 or e3e4, are not of the general kind ; since in them the director force eYe2 (or e3e4) intersects the director forces ab and a'b'. Case II. [Cf. 190 (5), Case II.] The latent regions are the line e^ and the point e3. The latent systems are, the forces eYe^ ese4, and any force of the type e3 (X^ + /xe2). There can be no other latent systems; for the coplanar lines of the type e3 (X^ + fxe2) and e^e2 are not all concurrent and therefore cannot all be null lines. Hence by using 192 (3) the pro- position is easily proved. There is a semi-latent dual group defined by e^ and e3e4. There are semi-latent parabolic groups with any force of the type ez (\e1 + /J e2) as director line [cf. 192 (14)]. Now by 192 (11) the point \ex -f /xe2 is the null point with respect to the group of the plane e^e3; and the point e3 is the null point of the plane (Xe1 + fie2) e3e4; since both exe2 and e3e4 must be common null lines of the group. Hence any other system of such a group must be of the form S = e3 (M + fjie2) + %e2e3 + rj (X^ + fxe^) e4. And by 190 (5), equation (K), it follows that = ryl73# + 77837! (\$i + txe2) e3. Hence any parabolic dual group of the type defined by e3 (X^ + /xe2) and %e2e3 + rj (X^ + fie2) e4 is semi-latent; and all such semi-latent groups contain only one latent system. There can be no semi-latent triple groups of the general type [cf. 192 (17)].
336 MATRICES AND FORCES. [CHAP. IV. Case III. [Cf. 190 (5), Case III.] The only two latent points are eY and e3. The latent systems are e^, e3e4, eYe3y and any system of the type \e3eY + fi { y 3e2e3 73S1e1e4). For [cf. 192 (3) and (4)] the point e1 is the null point of the plane e*fi Zi and the point e3 is the null point of the plane 6364^. Hence any possible latent system 8 (not a single force) must be of the form Now [cf. 190 (5), equation (L)] f S = y^S 4- (7381^23 + 7iS3^14) exez.....................(1). Thus, if S be latent, The semi-latent dual groups (not consisting entirely of single forces) are, the group defined by e^ and e3e4\ any parabolic group [cf. equation (1)] of the type defined by the director force exe3 and a system of the form fi23e2e3 + /^e^; any parabolic group of the type defined by e^ and Xe^ + fi (71836363 73816264) ; any parabolic group of the type defined by e3e4 and \e3ex + jjl {y1h3e2e3 738i6!64). It can easily be proved that no other dual groups are semi-latent, as follows. The only possibility lies in the existence of a parabolic group with a single latent system. But, from 192 (11) and (14), 6^3 is the only possible director force of such a group; and the group must have the form stated above. Further, by equation (1) above, the group stated is actually semi- latent. In searching for triple groups of the general type, it is useful to notice that such a group cannot be defined by a director force D and two systems 8 and S', such that (D8) = 0 = (D8'). For by 175 (9) in a triple group of the general type only one parabolic group with D as director force can exist. The semi-latent triple groups with three latent systems are all groups of the type defined by eYe2, e3e4, \e3eY -f /x (71836263 73SJ6J64). Call such groups the semi-latent groups of type I. No two groups of type I. can be reciprocal to each other, since all such groups have one pair of director forces in common, namely e^ and e3e4. Thus there must be another type of semi-latent groups. Call them the groups of type II. The only semi-latent subgroups, which a group of type II. can contain, are parabolic subgroups of the type defined by 6^3, fi^e^+fji 14e1e4. Now the condition that the system fi^e^ + y^\^e4 may be reciprocal to the system A^! -f /a (71836363 738^164) is 7i^i4 - 7381M23 = 0. Hence all groups of type II. contain the parabolic subgroup defined by 0i03) 7i836263 + 738^64. There is no other latent system which, in conjunction with this subgroup, will define a triple group of the general type. Hence all groups of type II. contain only one latent system, namely 6i63.
193] ENUMERATION OF LATENT SYSTEMS AND SEMI-LATENT GROUPS. 337 Any system 8, which has e^ and e3e4 as null lines, defines in conjunction with the systems exe3, yAe^ + 7381e1e4, a semi-latent triple group of type II. For we may write Then $S = yf/3S + O2A73 + ^ujA + ^Si^s) exe3 + ut24l (yAe^ + yAe^). Hence $S belongs to the required group. There can be no other semi- latent triple groups. For any other semi-latent triple groups mast contain a parabolic subgroup of the type eYe3i ^^2e3-\- yb14eYe4. But two sucn triple groups (of the general type) cannot be reciprocal to each other, since they both contain a common director force exe3. Case IV. (Subcase of Case I.) The latent and semi-latent points and regions are the same as in Case I.: but the additional relation 71 = 73 is satisfied. Thus 7^ = 732. The latent systems are the forces e^, e3e4 any system of the type Xe^-)- jie3e4, and any system of the quadruple group which has eYe2 and e3e4 as common null lines. Any latent system 8 is either such that f S = y,2S = y32S; or such that f S = y^S = - 7^ = - 732 No semi-latent parabolic group exists with only one latent system [cf. 192 (14)] Hence all semi-latent groups have their full number of latent systems. The semi-latent dual groups are defined by any two of these latent systems. The semi-latent triple groups are defined by any three of these latent systems. There are three types of semi-latent triple groups. Type I. and type II. are the same as in Case I.; and their groups are reciprocal in pairs. Type III. consists of groups defined by three latent systems of the form \eYe^ + /Ae3e4, ab, a!b\ The groups of this type are reciprocal in pairs; since the group defined by Xe^ + y 3eXy ab, a'b'', is reciprocal to the group defined by /A03e4, ab', a'b. Case V. (Subcase of Case II.) The latent and semi-latent points and regions are the same as in Case II.: bat the additional relation 7j = 73 is satisfied. The latent systems are, the forces e^, e3e4 any system of the type Xe^z + fjLe3e4, any force of the type ez {Xex + /*e2). There is a latent dual group defined by e^ and e3e4; semi-latent parabolic groups defined by latent systems of the type Xe^ + pe^ and e3 (X'e1 + /i'e2); and semi-latent parabolic groups, with only one latent system, defined by systems of the type e3 (Xex + fie2) and ge2es + tj (\^ + pe2) e4 [cf. Case II.]. There can be no semi-latent triple groups of the general type [cf. 192 (17)]. w. 22
338 MATRICES AND FORCES. [CHAP. IV. Case VI. (Subcase of Case III.) The latent and semi-latent points and regions are the same as in Case III.: but the additional relation 7j = ys is satisfied. The latent systems are the same as in Case III. with the additional set of latent systems of the form Xe1e2 + pe^. The semi-latent dual groups (not consisting entirely of single forces) are, the (latent) group defined by e^ and e3e4; any parabolic group of the type defined by exe3 and ^23^3+^14^64; any parabolic group of the type defined by e^e2 and Xe3eY + /jl {y-$3e2e3 7381e1e4); any parabolic group of the type defined by e3e4 and Xe3eY -f p (y e2e3 7381e1 4); and any group of the type defined by Xe^ + fiese4 and X'e$i + jjl (yi$3e2e3 738101e4). The semi-latent triple groups are simply those of type I. and type II. in Case III. For the only possibility of additional semi-latent triple groups (of a general kind) beyond those of Case III. lies in the semi-latent groups of the type defined by Xe^ 4- fie3e4, X!e3eY -f /jf (yAe2e3 738^ e4), e^e3. But these triple groups are not of the general kind, since the director force eYe3 is the director force of two parabolic subgroups belonging to such groups. 194. Transformation of a Quadric into itself. (1) When a triple group is semi-latent, the matrix must transform the director generators of the associated quadric [cf. 175 (4) and (5)] into director generators of the same quadric. Thus each point on the associated quadric is transformed into a point on the same quadric; and the quadric may be said to be transformed into itself by a direct transformation, the associated triple groups being semi- latent. (2) There is another way in which a matrix may transform a quadric into itself, so that the associated triple groups are not semi-latent. For the generators of one system may be transformed into generators of the other system, and vice versa. Let this be called the skew transformation of a quadric into itself, and let the first method be called the direct trans- formation. (3) If a matrix transforms a quadric into itself, by either direct or skew transformation, then every semi-latent line either has two distinct latent points on it, or touches the quadric. For let any semi-latent line cut the quadric in the two distinct points p and q. Then pp and f q are also on the quadric and on the semi-latent line. Hence either j p =p, f q = q, in which case p and q are two distinct latent points on the line, or cf p = q, f q = p. In the second case f 2p = f q=pf and Now if e be the sole latent point on the line, and 7 be the repeated latent root corresponding to e, then f p = yp + Xe, and (f 2p = y*p -h 2Xye = p. Hence X = 0. Thus p is a latent point. Similarly q is a latent point. Thus
194, 195] TRANSFORMATION OF A QUADRIC INTO ITSELF. 339 the line is a latent region corresponding to the repeated root, and e is not the sole latent point on it. But if the semi-latent line touches the quadric, this reasoning does not apply. The point of contact must be a latent point; and, as far as has been shown, it may be the only latent point on the line. (4) If the semi-latent line does not touch the quadric, and if cf p = q, f q=p, assume that eY and e2 are the two distinct latent points on the line, and that yx and y2 are the corresponding latent roots. Also let p = Xtfx + X2e2. Then f 2p = 712X1e1 + y22\e2 = X^ + \2e2. Hence 7i2 = 722- But if 71 = 72, then f p =p, j q = q, which is contrary to the assump- tion. Hence in the present case ^ = y2) and f p = yx (X^j X2e2) = # Thus (pq, exe2) form a harmonic range. Thus ex and e2 are conjugate points with respect to the quadric. (5) The polar reciprocal of a latent or semi-latent line is itself latent or semi-latent. For since the quadric is unaltered by transformation and the original line retains the same position, its polar reciprocal must also retain the same position. Also if a point be latent, its polar plane must be semi-latent. Hence if the latent point be not on the quadric, at least one other latent point must exist on the semi-latent polar plane. Then the line joining these latent points is semi-latent; and its reciprocally polar line is also semi- latent. 195. Direct Transformation of Quadrics. (1) It follows from the enumeration of 193, that the only cases in which semi-latent triple groups of the general type exist are those cases stated in 193 (2), Cases I. and II.: in 193 (3) Case III.: in 193 (5), Case V.: in 193 (6), Cases I. and III. and IV. and VI. In all these cases the relation, yy = y"y"t holds between the four latent roots of the matrix. In the two cases of 193 (2) the four latent roots are distinct; and there are only four latent points, which form a tetrahedron. In Case III. of 193 (3) two latent roots yx and y2 are equal, and yL2 = 7374: also a latent line exists corresponding to the double root: this is really a subcase of Case I. in 193 (2). In Case V. of 193 (5) all the latent roots are equal, there is a latent line, and an infinite number of semi-latent lines intersecting the latent line. This is really a subcase of Case III. of 193 (6), and will be discussed after that case. In Cases I. and III. of 193 (6), the latent roots are equal in pairs, namely ry1 = y2i ry3 = y^ and either (Case I.) two latent lines exist corresponding respectively to the two distinct roots; or (Case III.) only two latent points exist, one corresponding to each root. Case I. is a subcase of 193 (2). Case IV. of 193 (6) is Case I. with the additional relation 7l = -73. It is partly merely a subcase of (6) Case L: but it also transforms other 22 2
340 MATRICES AND FORCES. [CHAP. IV. quadrics according to the type of (2) Case L: thus it is partly a subcase of (2) Case II. Case VI. of 193 (6) is a subcase of Case III., and in no way differs from it in its properties with regard to the direct transformation of quadrics. (2) The general type of direct transformation of quadrics is given by 193 (2), Case I. Then the associated quadric of any group of the type defined by e-fi^ e2 3, (xe1e3-\- ^e2eA is transformed into itself by direct transfor- mation. The reciprocal group, associated with the same quadric, is e^, e3e4, aeYe3 /3e2e4, and this group is also semi-latent. The semi-latent lines exeAy e2e3 are generators of one system, and the semi-latent lines eYe^ e3e4 are generators of the other system. Hence the quadric has four of its generators semi-latent, two of one system and two of the other. It follows that the semi-latent lines e^e3 and e2e4 are reciprocally polar to each other, so that the polar plane of any point on one contains the other. All quadrics containing these four generators are transformed into them- selves. For they are defined by either group of a pair of reciprocal triple groups of the types mentioned above. (3) In Case II. of 193 (2), the only difference from the general case is that an additional set of quadrics are transformed into themselves; namely the associated quadrics of groups of the type defined by exe2y e3e4, ae^ + fie2e3. The reciprocal group associated with the same quadric is e^, e2eiy ae^ ^e2e:^ and this group is also semi-latent. (4) In Case III. of 193 (3) the only difference from the general case is that any two points aY and a2 on the line eYe2 can be substituted for ex and e2. Then a1e3y a2e4, o^e^ a2e3 are generators of one of the quadrics; and exe2 and e3e4 are reciprocal lines. (5) Case I. of 193 (6) is really only a subcase of that described in the subsection (3). Let c^ and a2 be any two latent points on the line exe*; and let a3 and a4 be any two latent points on the line e3e4. Then the explanation of the previous subsection applies, substituting any tetrahedron such as axa2a3a4 for the tetrahedron eYe2e3e4 in the previous subsection. Thus all quadrics which have the two latent lines eYe2 and e3e4 as generating lines are transformed into themselves by direct transformation. Also all such quadrics have two generators of each system latent, or semi-latent. (6) Case IV. of 193 (6) transforms into themselves the quadrics mentioned above both in (3) and in (5): that is, quadrics with the latent lines eYe2 and e3e4 as generating lines, and quadrics with exe2 and e3e4 as reciprocal lines. The transformation may be represented as follows. Let x be any point; draw through 00 the line xpq intersecting exe2 and e3e4 in p and q. Then x} cpx, p, q form a harmonic range.
195] DIRECT TRANSFORMATION OF QUADR1CS. 341 Let 8 = e^ + e3e4, 8' = e^ e3e4. Then [cf. 179 (4)] we may put f ( (7) In Cases III. and VI. of 193 (6) the semi-latent triple groups belong to two types ; and are reciprocal in pairs, one from each type. A typical specimen of type I. is defined by e^, e3e4) Xe^ + fi(yAe^e3 ^) a typical specimen of type II. is defined by Hence all quadrics, which are transformed into themselves, have two semi-latent generators of one system, namely exe2 and e3e4; and one semi- latent generator of the other system, namely exe3. All the quadrics, so transformed, touch each other along the generator exe3; since [cf. 187 (8)] the parabolic subgroup, defined by exe3 and 7iS3e2e3 + y^e^, is common to them all. The systems of the type Xe3ex -f- jj, (yiS3e2e3 y^e^) have only three null lines which are generators of the quadric associated with the triple group, denned by any one of them together with e^ and e3eA [cf. 175 (13)]. Thus consider the quadric denned by exe2i e3e4, and e3ex + v (yjb3e$3 y^e^). Let it be called the quadric A ; also for brevity write $ = v (yAe2e3 73S1e1e4). Then we have to prove that the system $' = Xe^ + fi (y e^ 73S1 1e4) has only three null lines which are generating lines of the quadric A ; unless Xv = p. For take the quadric A as the self-supplementary quadric, then [cf. 175 (8)] we may assume that Ie^ = exe2i \ e3e = e3e4) | {e + 8} = {e^ 4- S}, |e 3 = - exe3. Hence | S = 2^ 4- 8. Also it is easy to see that {exe3S) = 0. Now vS' = \ve +1*8. Hence v2 (S'S') = p* (SS). Again v \Sf = Xv \e3ex + A4 \8 = (2/x Xv) e3ex + jjlS. Hence v* (8 \8') = fi2 (SS) = i/2(S'S'). Thus (8' \S') = (S'S'). But [cf. 175 (13)], this is the condition that S' may have only three null lines which are generators of the quadric A, assuming that {8' 4= 8'. (8) Case V. of 193 (5) is a subcase of the case discussed in the previous subsection. The semi-latent triple groups belong to two types, such that the groups are reciprocal in pairs, one from each type. A typical group of type I. is defined by three latent forces oc f %, y f y, Z(j z: a typical group of type II. is defined by e^, Si^ + $e2e3, 8\ where S is any system. Thus any quadric transformed into itself has the latent line exe3 as a generator, and has no other latent or semi-latent generator of the same system; also all the generators of the opposite system are semi-latent [cf. 192 (17)]. All quadrics which are transformed into themselves touch along the generator eYeZi since [cf. 187 (8)] they have a common parabolic subgroup with eYez as director force.
342 MATRICES AND FORCES. [CHAP. IV. (9) If the four latent roots 7^ 72,73, 74, satisfying the relation 7173 = 7274, be assigned, then a matrix can in general be constructed, which will transform the quadric into itself by direct transformation, and at the same time make an assigned system latent. For let S be the assigned system. Then [cf. 174 (9)] S has in general one pair of conjugate lines which are polar reciprocal with respect to the given quadric. Let eYe3 and e2eA be these lines, cutting the quadric in the points elt e2, e3) e4. Consider the matrix for which e1} e2, e3y e4 are the latent points corresponding to the latent roots 7^ y2, y3, y4. The given system is obviously latent, since it is of the form ae1e3 + fie2e4. Also all quadrics con- taining the four generators exe4, e2e3y exe2, e3e4 are transformed into themselves, and among them the given quadric. (10) But if the system 8 has only three null lines, which are generators of the quadric to be transformed into itself; then the matrix must be of the type of 193 (6) Case III., or must belong to one of the subcases. Then [cf. subsection (7)] with the notation of 193 (6) Case III., let e1} e2, e3, e4 be so chosen that the three generators, which are null lines are, exe2, eYe3i e3e4. The system S can be written e3eY + ae^ + $eYe4, where a and /3 are known, since S is known. Then 7^ 73, Slt S3 must be so chosen that fi 73S1' Also by an easy extension of subsection (7) [cf. 175 (13) and 187 (8)] the quadric is defined by three systems of the form i ai e3e4) e3eY + \ (ae2e3 + $eYe4). Hence the quadric is transformed into itself, at the same time as 8 is latent, by the operation of the matrix. (11) Thus from (9) and (10), it is always possible to find a matrix which transforms directly a given quadric into itself, and keeps a given system of forces latent. And the matrix is not completely determined by these conditions. 196. Skew Transformation of Quadrics. (1) When a quadric is transformed into itself by a skew transformation, no generator can be semi-latent. (2) If j be a matrix which transforms a certain quadric into itself by a skew transformation, then the matrix j 2 transforms the same quadric into itself by a direct transformation. It is useful to notice that the latent points of j are also latent points of 02, though the converse is not necessarily true. (3) Let the matrix f have four distinct latent roots. Let yu 72, 73, 74 be the distinct latent roots, and ely e2, e3y e4 the corresponding latent points. Then the latent roots of / 2 are 7^, 722, 732, 7/, and elf e2, e3, eA are latent points. Now either 7^, 7/, 732, 7/ are distinct; or, two are equal, 7^ = 7/, so that 7! = y2; or, they are equal in pairs, 7^ = 7/, 732 = 7^.
196] SKEW TRANSFORMATION OF QUADRJCS. 343 Hence f 2 is either of the type of 193 (2), or of 193 (3), Cases I. or IIL, or of 193 (6), Case I. Case II. of 193 (3) cannot occur because [cf. subsection (2)] the line exe2 has two null points, ex and e2, on it. Similarly the other cases of 193 (6) cannot occur: 193 (6) Case IV. is inconsistent with the roots being distinct. But j 2 transforms the quadric into itself by direct transformation. Hence 193 (3), Case I. is impossible ; and 193 (2) and 193 (6), Case I. both make semi-latent lines of j to be generating lines of the quadric, which is impossible by subsection (1) above. If the additional relation, 7^ = y32y42, hold, then / 2 is the type of matrix described in 193 (3) Case III. The latent roots of are connected either by the relations y1 = y2 = V7374 or by 71 = 72 = \/ 7374- With the nota- tion of 195 (4), the points ax and a2 are on a quadric transformed directly into itself by 2; and axe3i a^, a2eAi a^ are generators of this quadric. Hence, if cf transforms this quadric into itself by a skew transformation, cij and a2 cannot be latent [cf. subsection (1)]. Hence, since they lie on a semi-latent line, cf)^ = a.2, ^a2 = alm Hence, by 194 (4), {axa2 exe2) forms a harmonic range. Also Qa^ = yxa2, f a2 = y^. Now for the quadric defined by the group axa2 4- Xe3e4i a2e3, a2e , to be trans- formed by $ by a skew transformation, this group must be transformed by j into the reciprocal group aYa2 Xe3e4f a2e3i a^eA. Now f (aYa2 + \e3e4) = 712o1oa + yjyjte t, atf3 = y!y3a,e3) j a2e4 = y^arf^ Thus it is necessary that y^ = 7374. Hence a matrix with four distinct latent roots, related so that 7i = - 7-2 = transforms into themselves by a skew transformation quadrics, passing through e3 and e4, with e^ and e3e4 as polar reciprocal lines, and with ex and e2 as polar reciprocal points. (4) Let the matrix \ have three distinct roots. Assume el9 e2, e3 e to be such that where Sx may, or may not, be zero. Hence j % = 7^, f 2e2 = y,% + 2^7^, j)% = y3%, f % = Now four cases arise. Case A. Let 7^, y32, y? be distinct. Then the matrix / 2 is of the type described in 193 (3). Hence it cannot transform a quadric by direct transformation into itself; except in Case III. But in 193 (3) Case IIL the lines of the type axe3 a.^, aYe4i a2e2 [cf. 195 (4)] are generators of the transformed quadrics. But these lines are semi-latent lines of (j as well as of f 2: and hence [cf. subsection (1)] this case must be rejected.
344 MATRICES AND FORCES. [CHAP. IV. Case B. Let yi2 = Y32; so that 7! = 73. Then the matrix j 2 is of one of the types (Cases I. and II.) described in 193 (4); either it is Case I. if S2 vanish ; or it is Case II. if 82 do not vanish. In either case f 2 cannot trans- form a quadric into itself by direct transformation. Case C. Let y32 = 7/; and S2 be not zero. Then f 2 is of the type described in 1.93 (6), Cases II. and V. The other cases of 193 (6) cannot occur, since the three latent roots 7,, 73, 74 are by hypothesis distinct; and the points e3 and e4 are both latent points of f 2. Hence / 2 cannot trans- form a quadric into itself by direct transformation. Case D. Let y32 = 7/; and 8X = 0. Then f 2 is of the type described in 193 (6), Cases I. and IV. If / 2 belongs to the type of 193 (6) Case I., then by 195 (5), f 2 transforms into themselves all quadrics with eYe2 and e3e4 as generating lines. But these are semi-latent lines of / . Hence by subsection (1), this case is impossible. But if / 2 belong to the type of 193 (6) Case IV., so that then (j 2 transforms quadrics directly for which eYe2 and e3e4 are polar reciprocal lines. Thus since 732 = 7/ = 7172, we have a subcase of the transformation considered in subsection (3). But it is the alternative case for which (f does not effect a skew transformation. (5) Assume that / has two distinct roots, one root yx occurring triply. Let ely e2) e3} e4 be assumed so that where 8X and 82 may or may not, vanish. Then ^1 = 7^i 4 % = Ji% + 28,7^, $% = y?e3 + 2S27le2 + 8A*, j 2e4 = Let the point e3' = 27^3 82e2. Then elt e2f e3' are such that / %! = 7^, 2e2 = 7!2e2 4- 28x7^, f 2e3 = y?e3 Case A. Let 7/ be not equal to 7^. Then $ 2 must be one of the three types described in 193 (4). But in no one of the three cases of that article does the matrix transform a quadric into itself by direct transformation. Case B. Let 74 = 71. Then (f 2 must be one of the types described in 193 (5). The matrix f 2 is of the type of Case III. of 193 (5), if B, and 82 do not vanish: it is of the type of Case II., if Sj vanishes : it is of the type of Case L, if Sj and 82 both vanish. But in Cases II. and III. no quadric is transformed into itself by direct transformation. In Case I. the matrix f 2 is merely the numerical multiplier 7^. Hence every quadric is transformed into itself, since no point changes its position. Then the matrix j has two latent roots jj and -7^ There is a latent
196] SKEW TRANSFORMATION OF QUADRICS. 345 plane exe corresponding to 7^ and a latent point eiy not on exe2e3, corre- sponding to ylm These are the only latent regions. Hence, by 194 (5), for all quadrics which are transformed into themselves by / , e and exe2e3 must be pole and polar. Also e4 cannot lie on such a quadric, since it does not lie on its polar plane. Now, if p be any latent point on the plane exe2e3, and x = \e4 + pp. then f)x = 7l (fip - \e4). Hence {e4p, x f x} forms a harmonic range. Thus if x be a point on a quadric for which e4 and exe^sz are pole and polar, f x is also on the same quadric. Also the transformation is skew, since by 192 (16) it cannot be direct. This is a subcase of the skew transformation of subsection (3), since Yi = - 74 = (6) Assume that f has only one root. Let ely e2, e3, e4 be assumed so that Then f % = yfa, f % = yfa + Let e3 = 27^3 - 82e2, el = Ly?e Then pel = y1%f + ly^e,, ^ = y*eA' + 27l2836;. Hence ^ 2 is a matrix of one of the types described in 193 (5). The case, when Sj = 82 = 83 = 0, need not be considered : for then f is a mere numerical multiplier. Thus [cf. 195 (1)] the only case of this type in which f 2 transforms a quadric into itself by direct transformation is that of 193, Case V. Then Bx and 83 do not vanish, and 82 = 0. In this case ez' = e3; also [cf. 195 (8)] the latent line exe3 is a generator of all quadrics trans- formed by f)2. But exe3 is also a latent line of . Hence (j cannot transform these quadrics into themselves by a skew transformation. (7) Assume that the latent roots of / are equal in pairs, so that 71 = 72, and 73 = 74. Let elf e2y e3 e be such that (f e1 = yle1} (f e2 = y1e2+S , cf e3 = y3e3t j e4 = y3e4 +83e3. Then (f 2e1 = y12e1 f 2e2 = yi2e2+2yAe1, f % = y32e3, t 2et = y3*e4 + 2y3 3e3. Hence cj 2 belongs to the type described in 193 (6). Of the six cases of this type only Cases I. and III. and IV. and VI. yield quadrics which are transformed into themselves by / * with a direct transformation. In Cases I. and IV., Sj = 0 = S3, and [cf. 195 (5) and (6)] either eYe2 and e3e4 are generating lines of such quadrics, or they are reciprocally polar lines to them. If they are reciprocally polar lines, the four semi-latent lines, joining the two pairs of points in which exe2 and e3e4 meet any such quadric, are generating lines of the quadric. But the latent lines eYe2 and e3ei} and the semi-latent lines joining any point on exe2 to any point on e3e4, are latent and semi-latent
346 MATRICES AND FORGES. [CHAP. IV. lines of cj as well as of / 2. Hence f does not transform any of these quadrics into themselves by a skew transformation. In Cases III. and VI. of 193 (6) neither Sx nor S3 vanishes. All quadrics transformed into themselves by f 2 have [cf. 195 (7)] the three lines exe^ e^, e3e4 as generators, which are semi-latent with respect to as well as with respect to f 2. Thus f does not transform these quadrics into themselves by a skew transformation. (8) Thus there is only one case of skew transformation, namely the case (including its subcase) when 7i = -7a= V7374; and the subcase arises when 7i = 72 = 73 = -74- In the general case the lines eYe2 and eze are polar reciprocal with respect to any quadric so transformed, the points e1 and e2 are polar reciprocal, and the points ez and e4 are on the quadric (except in the subcase, when 73 = 74 and the line eze4 is a latent region). In the subcase the point e4 and the latent plane exe$z are pole and polar with respect to all quadrics so transformed. Note. Homersham Cox, On the Application of Quaternions and Grassmanrfs Algebra to different hinds of Uniform Space, Trans, of (Jamb. Phil. JSoc, 1882, points out the connection between a positional manifold and Descriptive Geometry of any dimensions [cf. Book III.], and applies it to Hamilton's theory of nets. Also he points out the special applicability of Outer Multiplication to Descriptive Geometry [cf. Chapter IV., Book IV.] ; this had already been practically demonstrated by Grassmann in his papers in Crelle's Journal on Cubics. Further [in correction of note, p. 278] he applies the calculus in the manner of this book to deduce some elementary propositions concerning Linear Complexes; he finds the condition for reciprocal systems [cf. 116 (1)], for null lines [cf. 163 (1)], the director equations of dual and triple groups [cf. 172 (1) and 175 (1)], and the condition for a parabolic group [cf. 172 (9)]. He also finds a defining equation of Intensity [cf. note, p. 168], which depends on the distance between points. The bulk of this very suggestive paper is concerned with the Theory of Metrics.
BOOK VI. THEORY OF METRICS.
CHAPTER I. Theory of Distance. 197. Axioms of Distance. (1) In a positional manifold, to which no additional properties have been assigned by definition, no relation between any two points can be stated without reference to other points on the manifold. Thus consider a straight line which is a one-dimensional positional manifold. If ex and e2 represent the reference elements at unit intensity, any point p can be written fi^ + ^2' But is not the ex- pression of a quantitive relation between p and elt For fx depends on f2, and ^/f2 represents a relation of p to the terms eY and e2. But even this does not properly represent a relation of the element represented by p to those represented by eY and e2. For no determinate principles have been assigned by which the terms e1 and e2 should be considered to represent their corresponding elements at unit intensities. Thus the arbitrary assumption as to the intensities is included, when i/ 2 is considered as representing a quantitive relation of p to ex and e2. The only relations between points, which are independent of the intensities, are the anharmonic ratio between four points [cf. 69 (1)], and functions of this anharmonic ratio. (2) A spatial manifold will be defined to be a positional manifold, in which a quantitive relation between any two points is defined to exist. This quantitive relation will be called the distance, and the following axioms will be assumed to hold of it. (3) Axiom I. Any two points in a spatial manifold define a single determinate quantity called their distance, which, when real, may be conceived as measuring the separation or distinction between the points. When the distance vanishes, the points are identical. Axiom II. If p, q, r be three points on a straight line, and q lie between p and r [cf. 90 (3)], then the sum of the distances between p and q and between q and r is equal to the distance between p and r.
350 THEORY OF DISTANCE. [CHAP. I. Axiom III. If a, b, c be any three points in a spatial manifold, and the distances ab and be be finite, then the distance ac is finite. Also if the distance ab be finite and the distance be be infinite, then the distance ac is infinite. Also if the distances ab and be be real, then the distance ac is also real. (4) Let pq be any straight line through p; and assume some rule to exist, by which one of the two intercepts between p and any point q on the line can be considered as the intercept [cf. 90] such that points on it lie between p and q; then points on this line on the same side of p as q are points which either lie between p and q or are such that q lies between them and p. It follows from axiom II. that all points between p and q are at a less distance from p than the distance pq, and that all points on the same side as q, but beyond q, are at a greater distance from p than is q. Also it is evident that there cannot be another point on pq on the same side of p as q and at the same distance as q. For if q' be such a point, then by axiom II. the distance qq must be zero; and hence by axiom I. the points q and qr coincide. (5) Hence the relation of a point q to a point p in a spatial manifold is completely determined by (a) the straight line through p on which q lies, (/S) the determination of the side of p on which q lies, (7) the distance of q from p. Thus any quantitive relation between points on a straight line must be expressible in terms of their distance. 198. Congruent Ranges of Points. (1) Two ranges p, q, r, s, ... and p\ q', rf, s', ... of the same number of points in a spatial manifold are called congruent when the following conditions hold. Let the points py q, r, s, ... and the points p\ q', r', s'} ... be mentioned in order; also let the distance between p and q be equal to that between pf and q\ and the distance between q and r equal to that between q' and r', and so on. (2) It follows from this definition of congruent ranges and from axiom II. of distance, that the distance between any two points on one range is equal to that between the corresponding points on the other range. (3) Also [cf. 197 (5)] any quantitive relation between points in the first range, which can be expressed without reference to other points of the spatial manifold, is equal to the corresponding relation between the corre- sponding points of the second range. Such a relation in a positional manifold is the anharmonic ratio of a range of four points. Hence con- gruent ranges must be homographic [cf. 70]. (4) Also conversely, if on two homographic ranges the distances between three points of one range are respectively equal to the corresponding distances between the three corresponding points on the other range, then the ranges are congruent. For let pqr and p'q'r' be the two groups of three
198, 199] CONGRUENT RANGES OF POINTS. 351 points on the two ranges, and let 5 and s' be any other two corresponding points on the ranges respectively. Then the anharmonic ratio of (pqrs) equals that of (p'q'r's'). But if the range (p'q'r's") be constructed congruent to the range (pqrs), then by the previous part of the proposition the anharmonic ratio of (pqrs) is equal to that of (p'q'r's"). Hence the an- harmonic ratio of (p'q'r's') is equal to that of (p'q'rs"). Hence s and s" coincide. The proposition may be stated thus, if three points of one range are congruent to the three corresponding points of a homographic range, then the ranges are congruent. 199. Cayley's Theory of Distance. (1) Cayley has invented in his 'Sixth Memoir on Quantics'* a generalized expression for the distance between two points of a positional manifold. This work was extended and simplified by Klein-f", who pointed out its connection with Non-Euclidean geometry. (2) Consider in the first place a one-dimensional region. Let a^ and a2 be two arbitrarily assumed points on it. Then any three points xly x2, and x2 of the region can be written X^ -f f*i(h, \ai + /V 2 ^A + ^ci2. The anharmonic ratios /% P13, p\i respectively of the ranges (x^x,, are given by Hence log p12 + log p23 = log pu. Then if 7 be some numerical constant, real or imaginary, we may define [cf. Klein, loc. cit.] ^ log p12 as the distance between any two points xx and x2; where the distance is conceived as a signless quantity, but the ordinary conventions may hold as to the sign of lengths according to the direction of measurement. But the definition and the resulting conventions require further examination according to the different cases, which may arise [cf. subsection (4) below], (3) Let the point-pair al9 Og be called the absolute point-pair. Let these points be either both real, or let their corresponding co-ordinates, referred to any real set of reference elements, be conjugate complex numbers. Then for real points X2 and y^ are both real when ax and a2 are real, and are conjugate imaginaries when a^ and a2 are conjugate imaginary points. Similarly for X2 and /l^, and for X3 and ft3. * Cf. Phil. Trans. 1859, and Collected Papers, Vol. n. No. 158. t ' Ueber die sogenaunte Nicht-Euklidische Geometrie,' Math. Annalen, Bd. iv. 1871.
352 THEORY OF DISTANCE. [CHAP. I. Hence when ax and a2 are real, in order that the distance between real points which both lie on the same intercept between ax and a2 may be real, 7 must be real. When a^ and a2 are conjugate imaginary points, in order that the distance between real points, such as xx and x2, may be real, 7 must be a pure imaginary; for log pl2 is a pure imaginary. Let % be written for 7 in this case. (4) Thus if the absolute point-pair, ax and a2, be real, the distance between any two real points, x1 and x2, lying between them, is denned to be the real positive quantity ^ log p12, where 7 is some real number, and p12 is so chosen as to be greater than unity. Since p12 1, it follows that, with the notation of subsection (2), when ^ ^, and p12 = ^ when -2 ^. fa fl r \fJL fa fa Assuming , then p12 can be described either as the anharmonic ratio of the range (#j#2, a^), or as the anharmonic ratio of the range (x^, a2ax). Thus ax bears the same relation to xly as a2 bears to x2, in this definition of distance. The points ax and x1 will be considered as lying on one side of x2i and the point a2 on the other side [cf. 90]. Let this be called the Hyper- bolic definition of distance. It is to be noticed that, with this definition a pair of points, not in the same intercept between ax and a2, have not a real distance. If the absolute point-pair be two conjugate imaginary points, the distance between two real points xx and x2 is defined to be one of the two values of ^-. log p12 which lies between 0 and 777. The ambiguity as to which value is to be chosen is discussed later in 204, and its determination is possible in two ways. Let this be called the Elliptic definition of distance. (5) Let the limiting case be considered in which the absolute points are coincident at some point u. Let e and u be the two reference points in the one-dimensional manifold. Let aY = axe + /SjW, a2 = cL + ft2u. Then, when ax and a2 ultimately coincide with u, c^//^ and Oj/y j ultimately vanish. Let any other points x and y be written, x = e + %u, y = e + rju. Now putting A for OL1IS2 OL2ftl, Ax = ( - a2f) a, - (ft - (* ) a2i ty = (/32 - olm) a2 - (ft - aw) a2. Hence according to Cayley's definition, the distance between x and y is ( -0,0 (ft-aw)"
200] cayley's theory of distance. 353 Therefore expanding in powers of aj^ and c^/fa and retaining only the lowest powers, the distance becomes Now let 7 increase as aJfti and aJ82 decrease, so that 7 ( 77 tt J remains \P2 pJ finite and equal to 8, say. Then in the limit when ax and a2 coincide with u, the distance between the points e + gu and e+rju is S (?? ) Therefore this definition of distance is a special limiting case of the more general definition first explained. Let it be called the Parabolic definition. 200. Klein's Theorem. (1) It can be shown (cf. Klein, loc. cit.) that this definition of the distance between two points is the only possible definition, which is consistent with the propositions on congruent ranges in 198 (3) and (4). Let p, pi, p2, ... and p1} p2, p3y ... be two congruent ranges. Then by definition the distance ppY = the distance pxp2 = etc. Also by 198 (3) and (4) the ranges are homographic ; therefore [cf. 71 (1)] the first range can be transformed into the second by a linear transformation. Let aY and a2 be the two points on the line which are unaltered by this transformation [cf. 71 (2)], and firstly assume them to be distinct [cf. 71 (4) and (5)]. Then [cf. 71 (8)] Thus the anharmonic ratio (p^p*, axa2) is v ~p. But the distance between pp and pa is (a p) times the distance pYp2, which is any arbitrarily assumed distance X. Accordingly if ax and a2 be Cayley's absolute point-pair and pp = 6ax + / a2, pa ^ax + ^a2, we obtain But this is the definition of distance already given in 199 (2), as far as concerns integral multiples of an arbitrarily assumed length X. But since \ is any length, it may be assumed to be small compared to all lengths which are the subjects of discourse. Thus the definition must hold for all lengths. (2) Secondly, let the two points, ax and a2, unaltered by the linear transformation, be coincident, and write u for either of them [cf. 71 (7)]. Let e be any other reference point on the line ; then if any point p be written in the form e+%11, it is transformed [cf. 71 (7) equation L] into the point e + (f + 8) u, where 8 does not depend on . Thus if the range p Pi,P2-" be transformed into the congruent range p1} p2, p3..., the distance between the points pp and pa is (a - p) times the distance between p and px. But [cf. 71 (9)] p9 = e + (| -h p8) u (= e + vu say) an(^ Pa = e + ( + r8) u (= e + v% say). w. 23
354 THEORY OF DISTANCE. [CHAP. I. Therefore the distance pap? = \ (a p) = ^ (1/ 77), where \ is real. But this is the Parabolic definition of distance of 199 (5). 201. Comparison with the Axioms of Distance. The only difficulty in reconciling the Cay ley-Klein theory of distance with the axioms of 197 (3) arises from axiom I. For in axiom I. the distance is said to relate two points of a spatial manifold, whereas the definition of distance of 199 relates four points of the manifold, namely the two points of which the distance is defined and the two points forming the absolute. But the two points which form the absolute, if real, are at an infinite distance from every point of the spatial manifold. They may be considered as extreme, or limiting points, of the manifold. Thus the distance only relates two points arbitrarily chosen. Again if the absolute point-pair be imaginary, and the distance only relates real arbitrary points, the other points which enter into the definition are special points and are imaginary. 202. Spatial Manifolds of many dimensions. (1) Consider a spatial manifold of v 1 dimensions, where v 2. Assume that Cay ley's definition of distance applies to every straight line in it. Let the whole, or part, of the spatial manifold be such that Cayley's definition of distance, in the same one of its three forms, applies to any two real points in it; so that a real distance exists between them. Then such a manifold, or such a part of a manifold, will be called a Space of v 1 dimensions. If the Space of v 1 dimensions be not the complete spatial manifold, then there must not be a real distance between any point in Space and any point in the remaining part of the spatial manifold. Let the remain- ing real part of the spatial manifold be called Anti-space. Thus a spatial manifold is either such that its complete real portion forms Space; or it is such that its complete real portion is partly Space and partly Anti-space. (2) Consider any triangle abc in the complete spatial manifold, the whole or some part of which forms the space considered. Let b and c be real, and a either real or imaginary. Let the distance between b and c be real and finite. It follows from 197, Axiom III., that if a be one of the points of the absolute point-pair of the line ab, it must also be one of the points of the absolute point-pair of the line ac. Hence all the points which form the absolute point-pairs of all straight lines must form either an entirely imagin- ary surface, or a closed surface; and the part of the spatial manifold, which forms Space, must lie within the surface. For [cf. 82 (1)], when the absolute is real, every straight line through any point in space must cut the absolute in a pair of real points. Then the part of the manifold outside the closed surface is Anti-space.
201 203] SPATIAL MANIFOLDS OF MANY DIMENSIONS. 355 (3) Also every straight line, containing points in the spatial manifold, must cut this surface, whether it be real or imaginary, in one point-pair. The only algebraic surface for which this is possible is a quadric. Let it be assumed in future that the absolute point-pairs form a quadric, which is either entirely imaginary or real and closed. Let this quadric be called the Absolute. (4) When the absolute is imaginary, the spatial manifold is called elliptic*. There are two forms of elliptic geometry; the polar form in which the symbols + x and x represent the same point at opposite intensities [cf. 89 (1)] ; the antipodal form in which -f- x and x represent different points [cf. 89 (2)]. The discrimination between the two forms was first made by Klein. When the absolute is real and closed, the spatial manifold is called hyperbolic. In hyperbolic geometry the symbols + x and x represent the same point at opposite intensities. Parabolic Space is a special limiting form which both Elliptic and Hyperbolic Space can assume, when the absolute degenerates into two coincident planes [cf. 212 below]. (5) Let the distance between the two points a and b be written D (ab) as an abbreviation for ' distance ab.' 203. Division of Space. In the polar form of elliptic geometry a plane does not divide space. For if x and y be any two points and L any plane, the straight line xy cuts the plane L in one point p only. But [cf. 90 (5)] there are two intercepts between x and y. Thus the plane L cuts one of the intercepts and does not cut the other. Hence it is always possible to join any two points by an intercept of a straight line which does not cut a given plane. (2) But two planes do divide space. For it is possible to find two points, such that each of the two intercepts joining them cuts one of the two planes. For any straight line must cut the two planes in two points, say in p and q: on the straight line pq take two points x and y, one on each of the two intercepts joining p and q. Then the planes divide x from y in the way stated. (3) In the antipodal form of elliptic geometry a plane does divide space. For any straight line cuts a plane L in two antipodal points. Then if one intercept between two points x and y contains only one point p on the plane, the other intercept must contain the antipodal point. Thus x and y are divided from each other by the plane. Points x and y, which are not divided * Klein confines the term Elliptic to the Polar form of Elliptic Geometry. The Antipodal form is called by him Spherical Geometry. 23 2
356 THEORY OF DISTANCE. [CHAP. I. from each other by the plane, must be such that one intercept between x and y does not cut the plane and the other intercept contains the two antipodal points of section, namely p. (4) In the hyperbolic geometry a plane does divide space. It might have been wrongly anticipated, since + x represent the same point, that results analogous to those in the polar form hold. But in elliptic geometry space is the whole of the real part of the positional manifold; whereas in hyperbolic geometry space is only the part of the positional manifold within the closed absolute. Now no straight line lies completely within the absolute. Accordingly if one intercept, joining two points in space, itself lie completely in space, the other intercept passes out of space. Hence, ignoring points outside space, points joined by an intercept, lying entirely in space and cut by a plane, are divided from each other by that plane. 204. Elliptic Space. (1) Let the absolute be imaginary, so that the space is elliptic : let it be chosen to be the self-normal quadric [cf. Bk. IV., Ch. in.]. Then its equation can be written (x\x) 0. Also let it always be assumed that, when x represents any real point, (x \ x) is positive. Then from 199 (4) and equation (2) of 123 (9), the distance between any two points xv and x2 is 7l , + (X1\X2) = I log , = y cos- -V{(-^M = 7 where the inverse trigonometrical functions are to denote angles between 0 and 7T. (2) If x and x represent the same point [cf. 89 (1)], the am- biguity of sign must be determined so that + (x11 x2) is positive ; for this choice makes the distance of a point from itself to be zero. Hence, in the polar form of elliptic space, D (#i#2) is n t greater than ^ 7ry. (3) If x and x represent different points [cf. 89 (2)], then the upper sign is to be chosen in determining the ambiguity. Thus if d be the distance between x and y and d! the distance between x and y, d (x\y) , d! ( cos - = -7f- r v'y i Ni , and cos = /f/ V{0k)(2/b)} V{( 7 V{0*k)(2/b)} 7 Hence, in the antipodal form of elliptic space, D (xy) is not greater than iry. 205. Polar Form. (1) It is necessary, for the elucidation of the distance formula of the polar form of elliptic space, to investigate the circum- stances under which (x \z) and (y \z) are of the same and of opposite signs. Let the polar plane of x with respect to the absolute cut xy in x\ and let that of y cut xy in y'. Let the closed [cf. 65 (9)] oval line xyx'y' of the
204, 205] polar form. 357 figure represent the complete straight line xy. Any point z on this line can be written in the form ax + #'. For the sake of simplicity assume that a is positive and does not change, and that alone varies as z shifts its position on the line. Then for one of the two intercepts between x and x, is positive; for the other, is negative; let xyx be the positive intercept. Assume y = ax 4- fix', thus 0 is positive. Also we can write y' = @{x \x)x a{x\x)x thus yf is on the negative intercept. Also let it be noted that with the assumed form of yy (x \y) [= a(x |x)] is positive. Now z = ax + ' = X2 {a2 (x\x) + f0 {xf \a/)} y + \2 (a/3 - af) y'} where \~2 = a2 (x \x) + /32 (V |#') Accordingly (a? \z) is positive at all points of xy. And (y \z) = \*{a2 (* |a) + f/8 (^ ')} (2/ ly). Hence, remembering that as z moves from x' to y' in the direction of the o?(x x) arrow changes gradually from oo to Q , ,. * , we deduce that (y \ z) is p \x \x ) positive when z is on the intercept x'yxy' between x and y', and is negative when z is on the other intercept. (2) Secondly let z be any point not necessarily on the line xy. Now [cf. 72 (5)] z can always be written in the form z1 +p where zY is on the line xy and p is on the subplane which is the intersection of the polars of x and y\ and this representation is possible in one way only. Then (x\z) = (x\z1), and (y \z) = (y \zx). (3) Hence, summing up the results of (1) and (2), we see that if z be separated from x and y by the polar planes of x and y} then (x \z) and (y \z) are necessarily of different signs, provided that (x j y) is positive. But if z be not separated from x and y by the polar planes, then (x \z) and (y \z) are necessarily of the same sign, when (x\y) is positive. Thus if (y \z)} {z\x)t (x \y) are all of the same sign, they are all positive.
358 THEORY OF DISTANCE. [CHAP. I. (4) Let the intercept between x and y on which x and yf do not lie be called the intercept, while that intercept on which x' and y' do lie is called the polar intercept. 206. Length of Intercepts in Polar Form. (1) If x, y, z be three collinear points, it is as yet ambiguous as to which lies between the other two, since the straight line is a closed curve. The definition of distance has however really decided the question, as is shown by the following inves- tigation. (2) Let (x\y) be positive, and firstly let z lie on the intercept [cf. 205 (4)] between x and y. Put z = \x + fjiy', then X, jjl, (x\z) and (y \z) may be assumed to be positive. Hence JHxz) _ (x\z) _ . D(xz) _ I (xz \xz) _ / (xy \xy)_ . 7 " V{(* |*) (* I*)} ' 7 V (x~\x) (z \z) ~ * V (x \x) (z\zY and cos \/ (x\x Thus sin d( ) + d (*y) = *(y k) + ^(* k) / (*y 7 (z\z) V ( k 7 (x\z) (y \z) + (zy \zx) = cos A 7 Hence D (ass) + D (^) = I) (xy). Thus when z lies on the intercept between x and y, as defined in 205 (4), z lies between x and y according to the meaning of 197, axiom II. (3) If z lie on the intercept between y and x, then y lies on the intercept between x and z. Thus from subsection (2), Similarly if z lie on the intercept between x and y\ then # lies on the intercept between y and z, and (4) If z lie between #/ and y\ then / lies on the intercept between x and y, and each of the points x, yy z is separated from remaining two by the pair of polar planes of those two; so that each point lies on the polar intercept of the other two. Assume (x\z) positive and (y \z) negative: also let z = Xx fjiy, where \ and fi are positive [cf. 205 (1)].
206] LENGTH OF INTERCEPTS IN POLAR FORM. 359 mi D(xz) (x z) . D(xz) I (xu\xy) Then cos ^ '- = ,t, \ x / , - , sin v = fiA/rM1 -U) )} 7 rM (x\xj i (v \z) . D (vz) / =------v^LUL_ sin __MU. = X /- f 7 V (; 7 s/{(x\x)(z\z)}} 7 rVOi*0Ok)' D )S Hence cos (y \y)(*I*)" sin Also cos D (y|y) v (^l^) /J^yW)_ + x(#Iy) / Qy (xy\xy) . D(x =sin -A 7 x)(z\z)\ y Hence D (xy) + D (yz) = iry D (xz), Hence no one of the points x, y or z lies between the other two according to the meaning of 197, Axiom II. (o) This difficulty in the reconciliation of the Polar form of Elliptic Geometry to the Axioms of Distance may be obviated as follows. The distance between two points must be specially associated with the intercept between them; since for the intercept only is the axiom II. of 197 true. Let the distance between two points be also called the length of the intercept. Thus the intercept itself is considered as possessing a quantity of length. (6) Again the polar intercept may also be considered as possessing a quantity of length. For, since (x \xf) = 0, D (xx) 7 cos"10 = ^iry. Also and similarly D (yx') = D (xy') = rry-D (xy). Hence D (xfyf) = D (xx) - D (xyr) = D (xy). Thus D(yx') + D(xy') + D(y'x) = iry-D (xy). Hence iry D (xy) may be considered as the length of the polar intercept between x and yy since it is the sum of the lengths of its three parts. Accordingly the whole length of the straight line may be considered to be 777. This also agrees with equation (A) of subsection (4). (7) The paradox of subsection (4) can now be explained. For each of the three points lies on the polar intercept between the other two: and the sum of the distances of any two from the third is in each case equal to the
360 THEORY OF DISTANCE. [CHAP. I. length of the polar intercept. Thus axiom II. of 197 ought to be amended into, the sum of the lengths of the parts which make up either the intercept, or the polar intercept, is equal to the length of the intercept, or of the polar intercept, as the case may be. (8) Also if 7cos"1 ... \ }, . x1 gives the length of the intercept V{( l )(y|y)} (x\y) between x and y. then y cos"1 r, , \ , .-, gives the length of the polar intercept; and vice versa. Let that intercept between x and y of which the length is be called the intercept (x\y), or xy) and let the length of the intercept (x\y) be called xy. This name is useful in the ordinary case in which it is unknown and immaterial whether (x\y) is positive or negative. If (x\y) be positive, the intercept (x\y) is the intercept between x and y according to 205 (4). (9) It is necessary in this connection to distinguish carefully between the points x and y, and the terms x and y by which they are symbolized [cf. 14, Definition]. All congruent terms [cf. 64 (2)] denote the same point (or regional element). Two points x and y divide the complete straight line into two intercepts. The sum of the lengths of the two intercepts is iry. The length of the shortest intercept, which is the distance between the points x and y, is D (xy). The length of the other (polar) intercept is 7ry D(xy). The terms x and y, written in the form (x\y) or xy, define one of these intercepts. If (x\y) be positive, this intercept is the intercept, and is of length xy = D(xy). If (x\y) be negative, this intercept is the polar intercept, and is of length xy 7ry D (xy). Let x" = x y" = y. Then the terms x" and y" denote the same points as x and y. Also (x" \y") = (x\y)\ hence the intercept (x" \y") is the same as the intercept (x\y). But the intercepts (x" \y) and (x\y") which are the same intercept, are always the other intercept to the intercept (x \y) or (x" \y"). Thus, summarizing and repeating the distinctions between D (xy) and xy; D (xy) = D (x"y") = D (x"y) = D (xy"); xy = x"y '; xy"' = x"y; xy + x"y = 7ry; xy = D (xy), if (x \ y) be positive; x"y = D(xy), if (x \y) be negative; Also the length of the intercept (x \y) is written xy. Then xy (as well as (x\y)) may also be taken as this name of the intercept. It is not often of much importance to know whether xy = D (xy) or Try I) (xy).
207] LENGTH OF INTERCEPTS OF POLAR FORM. 361 (10) If z be the point x 4- %y, then when is positive z lies on the intercept (x\y). For cos = ,r.....fr'^tfl*.1 y)}]" Hence as changes gradually from 0 to + oo , cos diminishes gradually 7 (x I y) from 1 to t7/ | x^ , .-, , and this whether (a? y) be positive or negative. \/\(x \x) (u In)\ Thus xz gradually increases from 0 to xy. Similarly at the same time zy gradually decreases from xy to 0. Hence z must lie in the intercept 207. Antipodal form. (1) In the antipodal form of elliptic geometry the intercept between x and y is that intercept which does not contain the Fig. 2. antipodal points x and y; the intercept containing the antipodal points is called [cf. 90 (6)] the antipodal intercept. (2) Now by a proof similar to that in the previous article, if z lie in the intercept between x and y, D (xz) + D (zx) = D (xy). If z lie in the intercept between y and xyD (xy) + D (yz) = D (xz). If z lie in the intercept between x and y,D (yx) + D (xz) = D (yz). (3) If z lie in the intercept between x and y} let z' ( z) be the antipodal point to z. Then zf lies in the intercept between x and y. Hence by subsection (1) But by 204 (3), D (xz) + D (xz') = 7ry = D (yz) + D (yz'). Hence D (xz) + D (zy) = 2ttj - D (xy). Also D (xy) + D (yz) = D (xy) + 7ry-D (yz') = rry + D (xz') = 2-rry - D (xz), and D (yx) + D(xz) = 2iry - D (yz).
362 THEORY OF DISTANCE. [CHAP. I. Thus no one of the three points x, y, z lies between the other two in the sense of axiom II. 197. Accordingly this axiom is not literally satisfied; however the following explanations and additions shew that it is substantially satisfied. (4) Analogously to the similar case of the polar form, let the distance between x and y be called the length of the intercept between x and y. Then the length of the intercept between y and x is Try D(xy), and this is also the length of the intercept between y and x. The length of the intercept between x and y is D (xy). Hence adding the three parts, the length of the antipodal intercept between x and y is 2tty D (xy). Thus the length of the whole straight line is 2iry. (5) The paradox of subsection (3) can now be explained. For each of the three points lies on the antipodal intercept between the other two: and the sum of the distances of any two from the third is in each case equal to the length of the antipodal intercept. Thus axiom II. 197 ought to be amended into, The sum of the lengths of the parts which make up either the intercept or the antipodal intercept is equal to the length of the intercept or of the antipodal intercept, as the case may be. 208. Hyperbolic Space. (1) Secondly let the absolute quadric be real and closed. Then from 199 (4) and equation (3) of 123 (10), the distance D (xy) between any two points x, and y within the quadric is D(Xy) = i7logPl9 = 7cosh"1 /f/ ,(^l/)1 N1 = 7sinh-1 A / \ U) 27 SFi. 1 *J{(x\x)(y\y)} 7 V ( The ambiguity of sign must be determined so that (x\y) is positive. It has been proved in 82 (9) that (x \x) and (y\y) are of the same sign : hence {(x \x)(y\y)} is necessarily positive. (2) The test as to which sign of the ambiguity is to be chosen is derived from the following lemma; which, it is useful to notice, applies to any closed quadric (cc$x2) = 0, and not solely to the absolute in its character of self-supplementary quadric. Let e, x, y, z be four points within the quadric. Then, if (e \x), (e \y), (e\z) are of one sign, also (y \z\ (z \x), (x \ y) are of one sign. For let the line xy cut the polar of e in e\ and let xz cut it in e". Then we may write y = \x -f t)e\ z fix + e". Hence, since (e\e) = 0 = (e\e"), (e\y) = X(e\x) and (e\z) =/J (e \x). Therefore from the hypothesis \ and jjl are positive. Again as rj varies between oo and -f oo, y takes all the positions on the line xe. Also (x \y) = X (x \x) + rj (e' \x). Hence (x \y) is a linear function of the variable 77; and thus as 7/ varies, (x y) can only change sign when it
208, 209] HYPERBOLIC SPACE. 363 vanishes or is infinite. But when (x\y) vanishes, y must lie on the polar plane of x, and this plane is entirely outside the quadric [cf. 82 (6)]; similarly when (x \y) is infinite, rj is infinite and y coincides with e which is outside the quadric since it lies on the polar plane of e. Thus for all points y on that part of the line xe which lies within the quadric, (x \y) has the same sign. Now put rj = 0. Hence (x\y) has the same sign as \{x\x). But \ is positive. Thus (x\y) has the same sign as (x \x). Also (x \x) has the same sign for all points within the quadric, say the positive sign. Hence (x\y) is also positive. Thus the proposition is proved. (3) Let (x'\x) be always assumed to be positive for points within the quadric : also let a point x within the quadric be said to be of standard sign when (e x) is positive, where e is any given point within the quadric chosen as a standard of reference. Then it follows from the above that for all points of standard sign within the quadric, (x | y) is positive. Thus the distance between two points x and y, within the quadric and of standard sign, is D (xy) = i7 log Pv9 = 7 cosh-1 -/U (*-\y) - xl = 7 sinh"1 /)P In future all symbols arbitrarily assumed to represent points within a real closed absolute will be assumed to represent them at standard sign. (4) In hyperbolic space there is only one intercept between two points x and y which lies entirely within the space. Also if z lie within this intercept D() Hence there is no ambiguity as to the application of axiom II. of 197. The distance between x and y will be called the length of the intercept between x and y. The distance of any point from any point on the absolute is infinite. Thus the length of the part of any straight line within the spatial manifold is infinite. 209. The Space Constant. It is formally possible to assume that 7, instead of being an absolute constant, is constant only for each straight line; and accordingly is a function of any quantities which define the special straight line on which xx and x2 lie. Such quantities can necessarily be expressed in terms of the co-ordinates of xx and x2, since these points define the line x^x2. Hence the assumption of 7 as a function of the co-ordinates of the straight line joining the points does not appear necessarily to offend against the axioms of 197. Let the assumption be made that 7 is constant and the same for all lines. Let 7 be called the space-constant.
364 theory of distance. [chap. 1. 210. Law of Intensity in Elliptic and Hyperbolic Geometry. (1) The law of intensity (cf. Bk. III. ch. iv.) is also settled, if the assumption* be made that, when xx and x2 are of the same intensity, w1 + x2 bisects the distance between x1 and x2; where for the polar form of elliptic geometry (^ \x2) is assumed to be positive, and for hyperbolic geometry x1 and x2 are both of standard sign. No special explanation is required for the antipodal form of elliptic geometry, since xx -f x2 is to bisect the distance between xl and x2i and xx x2 is to bisect the distance between xl and x2. Then by 204 and 208, Hence {V(^i |O V( Therefore either (xx |O = (x2 \x2), or (^ |#x) (#2 |#a) (^ \x2)2 = 0. The second alternative is equivalent to (xYx2 x1x2) = 0. This implies that the line xxx2 touches the absolute [cf. 123 (5)]; and this presupposes special positions for x1 and x2. In fact for such a case in elliptic geometry the line xYx2 would then be imaginary ; and in hyperbolic geometry xx and x2 would lie outside the absolute. Hence the alternative, {x1\x^) = {x2\x2)i must be adopted. Accordingly if the point x has a given intensity, (x \x) is independent of the position of x. Thus with a proper choice of constants the intensity of x is \j(x \x); so that (x \x) = 1, when x is at unit intensity. (2) Then, if x1 and x2 be at unit intensity and (#i|#2) De positive (except for antipodal elliptic space), the formulae for the distance between them become, d12 = ^ log p12 = 7 cos"1 (^ \x2) = j sin"1 V(#i# and = 7 cosh"1 (xx \x2) = y sinh"1 V(- according as the space is elliptic space or is hyperbolic space (of any number of dimensions), where in both cases xx and x2 fulfil the condition (3) As an illustration of these formulse consider antipodal elliptic space of two dimensions. Let the absolute be (x \x) = 2 + 22 + 32 = 0. Then the conditions, (x \x) = 1 = (y \y), become * This assumption is made by Homersham Cox, loc. cit.
210, 211] LAW OF INTENSITY IN ELLIPTIC AND HYPERBOLIC GEOMETRY. 365 And cos - = (x | y) = f-^ + 2?y2 + f3rjs, sin - = These are the formulae of the ordinary Euclidean geometry of a sphere ; where f2, f2, and Vi V2 V3 are direction cosines. 211. Distances of Planes and of Subregions. (1) As yet only the distance between points has been defined. The same principles can easily be applied to planes. For any planes X and F can be expressed in terms of their polar points with respect to the absolute. Thus X = \x, and Y = | y. Hence, if the absolute be imaginary, (X | X) and (Y\Y) are necessarily of the same sign. If the absolute be real and closed, (X | X) and (F| Y) are of the same sign, when x and y are either both within or both without the absolute. If x lie within the real closed absolute, the plane X contains [cf. 82 (6)] no points lying in space, but only points in anti-space ; but if x lie without the absolute, then [cf. 82 (7)] the plane X contains points in space as well as points in anti-space. Let X and Y be any two planes, and suppose that the plane touches the absolute quadric. Then X//a must be one of the two roots \Jfii and X2//42 of the equation Let Ax and A2 be these two tangent planes; then the anharmonic ratio of the range {XYy A1A2] is Xj/^/Xg/^i, and this ratio is either real or of the form e2i^} where f is real. Let it be called p. Then if p be real, the measure of the separation between X and Y can be defined to be ^ log p; and if p be imaginary, it can be defined to be . log p ; where k and tc are constants. 2/1 There is no reason why either k or k should necessarily be equal to the ' space-constant' 7. But there is no real loss of generality, and there is a gain in the interest of the analogy to ordinary geometry, if k = 7, and k = 1. For it will be found that the hyperbolic measure of separation between planes can then be identified with the distance between two points; and the elliptic measure of separation can be considered as the angle between them, which is of no dimensions in length. (2) Thus, [cf. 124], it follows that the separation between two planes X and Y is that angle between 0 and it given by (XjF) . _x I (XF|XF) Sm when (X|F)2 (X|X)(F|F).
366 THEORY OF DISTANCE. [CHAP. I. And the separation is * 7i u i (X\Y) u i / ~(XY\XY) d = ^ogP = y cosh- ^ ^ (Vm = y smh- J 1(Z,|Z)(r|r)} , when (X|F)2 (XiX)(F|F). It must be noticed that the distinction between these two cases must not be identified simply with that between Elliptic and Hyperbolic Geometry as defined above. The trigonometrical functions must however always be adopted in Elliptic Geometry. This question will be considered in the succeeding chapters as far as it concerns Hyperbolic Geometry. (3) Furthermore the ambiguity of sign is capable of being determined by exactly the same methods as obtained for points. But with respect to planes, in order to obtain an interesting extension of the ideas of ordinary geometry, the ' polar' form is invariably adopted, namely, + X and X are considered as representing the same plane at opposite intensities. (4) If the elliptic measure of distance between planes has to be adopted, the measure of separation of planes is called the angle between them. The ambiguity of sign in the formula for the cosine of the angle leads to the definition that planes make two supplemental angles with each other, 6 and tt 0; and that of the two the acute angle is the measure of the separation of the planes. (5) The law of intensity of planar elements is determined by the same principles as that of points. Let it be assumed that if X and F be planar elements of the same sign and at the same intensity, then X 4- F bisects the distance between X and F. Hence the defining equation of unit intensity can be written, (X \ X) = S, where S is a constant which will be determined later separately for Elliptic and Hyperbolic Geometry according to con- venience. In Elliptic Geometry S is always of the same sign: let it there- fore be chosen to be unity. In Hyperbolic Geometry it is convenient to choose 8 to be positive or negative according as the (real) plane does or does not cut the absolute: let it therefore be chosen to be 1. (6) It is in general impossible to define one single measure of separation between any two subregions Xa and Ya, of a 1 dimensions. But if they are both contained in the same subregion of r dimensions, then considering the latter subregion as the complete region, Xa and Ya have the properties of planes in regard to it. Also the absolute in this complete region may be taken to be the section of the absolute by the region. Hence in this case, cf. 124 (4), the measure of the separation of Xa and Ya (with the conventions, already explained, determining ambiguities) is either ' or 7C0sn v{(
211, 212] DISTANCES OF PLANES AND OF SUBREGIONS. 367 Definition. \ Xff and | Ya are called the absolute polar regions of Xa and Fff. It is obvious that the separation between two regions is equal to that between their absolute polar regions. 212. Parabolic Geometry. (1) If the parabolic definition of distance hold for every straight line, then every straight line must meet the absolute in two coincident points. Hence the absolute must be two coincident planes. It can be seen as follows that the elliptic and hyperbolic definitions for v 1 dimensions both degenerate into the parabolic definition, when the absolute is conceived as transforming itself gradually into two coincident planes. (2) Let the co-ordinate points e1, e2, ... ev be v self-normal points, then the equation of the absolute takes the form, Now conceive the form of the quadric to be gradually modified by a2, ... olv diminishing, till they ultimately vanish, while aY remains finite. Then ultimately the equation of the quadric becomes o^ = 0; that is to say, the quadric becomes two coincident planes, the equation of each plane being x = 0. Also the v 1 co-ordinate points e2, e3, ... ev lie in this plane, and the point e1 without it. Also, cf. 123 (6), {xy \xy) = , (^ - ^Vp)2- Assume that, as the quadric approaches its degenerate form, Ko K3 Kv a2 = , a3 = , ..., a, = , 7 7 7 where the ks are finite and 7.is ultimately infinite. Then ultimately, (xy \ xy) = a * (?*p - fa? + 2 ^ (f ?, - V,Y = ^ (f^ - Similarly (x \x) = a^2, (y \y) = Then if the geometry be elliptic and 7 be the space-constant, d = 7 sin- JMty] i = 7 sin"1 2 Now, since the geometry is elliptic, ax and k2, tc3, ... kv are all of the same sign. Put^ = A . Hence d = 10 **"* If the geometry be hyperbolic, I {x\x)(y y)
368 THEORY OF DISTANCE. [CHAP. I. Now, since the geometry is hyperbolic, the absolute is a real closed quadric; and hence [cf. 82 (5)] ax must have one sign and k2, tf3, ... tcv another sign. Put = j Hence i (3) Thus as a limiting case both of Elliptic and Hyperbolic Geometry, we find a space with the distance between any two elements given by d = 2/32 ^lVp ~ ^Vl^ where the v 1 co-ordinate elements e2, e3, ... ev lie on the absolute plane at an infinite distance. 213. Law of Intensity in Parabolic Geometry. (1) Let e1 be the reference element not in the absolute plane, and let u2, uit...uv be the reference elements in the absolute plane. Let it be assumed, as in 210 (1), that, when x and y are of the same intensity, x + y bisects the distance between x and y. Now let x = + Sf'uy y = rj^ + %VU Also the distance between x and x + y is by 212 (3) Similarly the distance between x-\- y and y is Vi (fi + %) Hence since these distances are equal, ( + ^J = r^y (^ -f t/j), and thence, (2) Hence the intensity of the point a? is a function of only; but by 85 (2) it must be a homogeneous function of the first degree. Thus the intensity of x is \ 2, where \ is some constant; and, if eY be chosen to be at unit intensity, then X=l. Hence the absolute plane is the locus of zero intensity and the law of intensity explained in 87 (4) must hold. And the expression for a point x at unit intensity is e1 + 2fw, where e1 is at unit intensity. Also the distance between the two points el + 2fu and e1 + 2^, both at unit intensity, is 2/3p2 (f t/)2. Furthermore by properly choosing the intensities of u2 w3, ... , this expression for the distance can be reduced to 2(f i?)2. Thus* parabolic * Cf. Riemann, Ueber die Hypothesen, welche der Geometrie zu Grunde liegen, Collected Mathematical Works.
213] LAW OF INTENSITY IN PARABOLIC GEOMETRY. 369 space of v 1 dimensions can be interpreted to be simply an ordinary Euclidean space of that number of dimensions; where e^, eYuZi ...eYuv are v 1 axes at right-angles, and 2, %3...i;v are rectangular Cartesian co- ordinates. The interpretation of (the vectors) n2) u3i... uv will be considered in Book VII. Historical Note. An interesting critical ' Short History of Metageometry' is to be found in Chapter I. of The Foundations of Geometry, by Bertrand A. W. Russell, Cambridge, 1897. Klein also gives an invaluable short history of the subject in his lithographed Vorlesungen iiber Nicht-EuJclidische Geometrie, Gottingen, 1893 ; he makes the important division of the subject into three periods. The following are the creative works of the ideas of the three periods. First Period. Lobatschewsky, Geometrische Untersuchungen zur Theorie der P arallel-linien, Berlin, 1840; translated by Prof. G. B. Halsted, Austin, Texas, 1891. Lobatschewsky's first publication of his discovery was in a discourse at Kasan, 1826 (cf. Halsted's preface); and subsequently in papers (Russian) published at Kasan between 1829 and 1830 (cited by Stackel and Engel, cf. below). John Bolyai, The Science Absolute of Space, 1832 ; translated by Prof. Halsted, 1891 ; also cf. German edition by Frischauf, cited below. The original is written in Latin, and is an appendix to a work on Geometry by his father, Wolfgang Bolyai. Second Period. Rieman, Ueber die Hypothesen, welche der Geometrie zu Grunde liegen, written 1854, Gesammelte Werke ; translated by Clifford, cf. his Collected Mathematical Papers. Helmholtz, Ueber die thatsachlichen Grundlagen der Geometrie, 1866, and Ueber die Thatsachen, die der Geometrie zum Grunde liegen, 1868; both in the Wissenschaftliche Abhandlungen, Vol. 11. Beltrami, Saggio di Interpretazione della Geometria non-Eiwlidea, Giornale di Matematiche, Vol. vi. 1868; translated into French by J. Hoiiel in the Annales Scientifiques de VBcole Normale Superieure, Vol. vi. 1869. Third Period. Cayley, Sixth Memoir upon Quantics, Phil. Trans., 1859; and, Collected Papers, Vol. 11., No. 158. Klein, Ueber die sogenannte Nicht-Euklidische Geometrie, two papers, 1871, 1872, Math. Annalen, Vols. iv., vi. Lindemann, Mechanik bei Projectiven Maasbestimmung, 1873, Math. Annalen, Vol. vu. Lie, Ueber die Grundlagen der Geometrie, Leipziger Berichte, 1890. A bibliography up to 1878 is given by G. B. Halsted, American Journal of Mathematics, Vols. 1., 11. The following very incomplete list of a few out of the large number of writers on the subject may be useful: Flye, Ste Marie, Etudes analytiques sur la theorie des paralleles, Paris, 1871. M. L. Gerard, These, Sur la Geometrie Non-Euclidienne, Paris, 1892. Poincare*, Theorie des Groupes Fuchsiennes, Acta Mathematica, Vol. 1., 1882. Clebsch and Lindemann, Vorlesungen uber Geometrie, Vol. 11. Dritte Abtheilung, Leipzig, 1891. Frischauf, Elemente der Absoluten Geometrie nach Johann Bolyai, Leipzig, 1876. w. 24
370 THEORY OF DISTANCE. [CHAP. I. Killing, Die Nicht-Euklidischen Raumformen in Analytischer Behandlung, Leipzig, 1885. Stackel and Engel, Die Theorie der Parallel-linien von Euklid bis auf Gauss, Leipzig, 1895. This book contains a very useful bibliography of books on the Theory of Parallels fr om the year 1482 to the year 1837. Veronese, cf. loc. cit. p. 161. Burnside, On the Kinematics of Non-Euclidean Space, Proc. of Lond. Math. Soc., 1894. Clifford, Preliminary Sketch of Biquaternions, Proc. of Lond. Math. Soc, 1873, and Collected Mathematical Papers. Newcomb, Elementally Theorems relating to the Geometry of a space of three dimensions and of uniform positive curvature in the fourth dimension, Crelle, Vol. 33, 1877. The philosophical questions suggested by the subject are considered by Russell, Foundations of Geometry (mentioned above); in this work references will be found to the previous philosophical writers on the subject. The first application of an 'extraordinary' algebra to non-Euclidean Geometry was made for Elliptic Space by Clifford, Sketch of Biquaternions, Proc. of London Math. Society, Vol. iv. 1873, also reprinted in his Collected Papers ; this algebra will be con- sidered in Vol. II. of this work. The first applications of Grassmann's Calculus of Extension to Non-Euclidean Geometry were made independently, by Homersham Cox (cf. loc. cit. p. 346), to Hyperbolic and Elliptic Space, and by Buchheim to Elliptic Space; On the Theory of Screws in Elliptic Space, Proc. London Math. Soc, 1884 and 1886, Vols. xv. xvi. xvii. The idea of starting a (pure' Metrical Geometry with a series of definitions referring to a Positional Manifold is obscurely present in Cayley's Sixth Memoir on Quantics; it is explicitly worked out by Homersham Cox {loc. cit.) and by Sir R. S. Ball, On the Theory of Content, Trans, of Roy. Irish Academy, Vol. xxix. 1889. Sir R. S. Ball confines himself to three dimensions, and uses Grassmann's idea of the addition of points, but uses none of Grassmann's formulae for multiplication. But the general idea of a pure science of extension, founded upon conventional definitions, which shall include as a special case the geometry of ordinary experience, is clearly stated in Grassmann's Ausdehnungslehre von 1844 ; and from a point of view other than that of a Positional Manifold it has been carefully elaborated by Veronese {loc. cit.). Homersham Cox constructs a linear algebra [cf. 22] analogous to Clifford's Biquaternions, which applies to Hyperbolic Geometry of two and three and higher dimensions. He also points out the applicability of Grassmann's Inner Multiplication for the expression of the distance formulae both in Elliptic and Hyperbolic Space; and applies it to the metrical theory of systems of forces. His whole paper is most suggestive [cf. notes, p. 346 and at the end of this volume]. Buchheim states the distance formulae for both Elliptic and Hyperbolic Space in the same form as they are given in this chapter, with unimportant variations in notation. He then deduces Clifford's theory of parallel lines; and proceeds to investigate the theory of screws in Elliptic and Hyperbolic Space of three dimensions. In his last paper he obtains an important theorem respecting the motion of a rigid body in Elliptic Space of 2/x 1 dimensions. Many of his results are deduced by the aid of Biquaternions, and of Cayley's Algebra of Matrices. A further account of his important papers is given in the note at the end of the volume.
CHAPTER II. Elliptic Geometry. 214. Introductory. In the following application of the formulae of the Calculus of Extension to the investigation of Elliptic Geometry the polar form will be exclusively considered. Most of the theorems and investi- gations apply, mutatis mutandis, to both forms. But each form requires its own special explanations, which though important geometrically are only remotely possessed of any algebraic interest. So to avoid prolixity one form is adhered to. The space spoken of throughout this chapter will be of v 1 dimensions where v is any number. It is the merit of this Calculus that the general formulae for v 1 dimensions are as simple and short as those for two or for three dimensions. 215. Triangles. (1) Let the terms a, b, c denote three points; there are eight modes of associating the pairs of intercepts [cf. 206 (8)] joining each pair of points; namely, using lengths as named, that by associating be, ca, ab; or Try be, Try ca, Try ab ; or Try be, ca, ab ; or be, Try ca, Try ab ; and so on. (2) Let the angle a between the two intercepts ab and ac be denned to be that angle (out of the two supplementary alternatives) given by [cf. 211 (6)] (ab \ac) COS a = -777 \J{(ab \ab)(ac\ac))' Similarly for the angles j8 and y. Thus the angle between ab and Try ca is found by putting c for c in the above and is cos__lMM ^{(ab\ab)(ac\ac)}' that is 7r a. Let the angles a, ft, y be associated with the intercepts be, ca, ab; and let this system of intercepts and angles be called the triangle abc. 24 2
372 ELLIPTIC GEOMETRY. [CHAP. II. (3) Now (ab \ac) = (a a) (b \c) - (a \b) (a \c). Also [cf. 206 (8)] . ab_ I (ab\ab) . ~ac___ I (ac\ac) Sm 7^V ( \a)(b\bY Sm 7 "~_y a|a)(c|c)' , ab (a 16) ac (a \ c) and cos = ,(/ v ' /7 77V) , cos = -^ ^------ ^ __ i a) (c \ c)}' TT bo ab ac . ab . ac Hence cos = cos cos--h sin sm cos a; 7 7 7 7 7 with similar formulae for j8 and y. (4) When a = 0, then c is collinear with a and 6. Also (ab\ac) is positive: hence we can write either c = %a + b, or c = fja + 6, where is positive. In the first case by 206 (9) c lies in the intercept ab; in the second case, since b = c + a, 6 lies in the intercept ac. A! be ab - ac Also cos = cos------ . __ _ __ 7 7_ _ _ Thus be ab ac in the first case, and be = ac ab in the second case. (5) Let a, b\ c' stand for - a, - 6, c respectively. Then cog^'_ fl'K) = (Me) _co _ _7 ___ Thus 6V = 6c. Similarly c'a' = ca. a'b' = ab. Again it is easy to see from (3) that the angle between ab' and dd is a; and so on. Hence the triangle db'c is the same as the triangle abc, both in its sides and angles and angular points. The two are therefore identical. (6) Consider the triangle a'bc} which by subsection (5) is the same as ab'c'. Its sides are easily seen to be related to those of abc as follows : b'e = be, e'a Try ca, ab' = 7ry ab. Hence by subsection (3) its angles are a, it j8, tt y. Similarly the triangle ab'c, or dbc', has sides Try be, ca, iry ab, and angles tt a, j8, it y. And the triangle abc', or db'c, has sides iry be, Try ca, ab, and angles it - a, it - JS, y. (7) Hence of the eight possible cases mentioned in subsection (1) only four can have angles associated with them in accordance with the convention of subsection (2). Accordingly three points will be said to define four triangles, where a triangle is taken to mean three determinate intercepts and three angles between each pair of intercepts. The triangle defined by the terms a, b, c will be taken to mean the triangle with the intercepts (b\c)t (c\a)f (a\b) as sides, and will be called the triangle abc. The other triangles defined by the points a, b, c are the triangles dbc (or ab'c), ab'e (or dbc), abc' (or db'c).
215] TRIANGLES. 373 There are two main cases to be considered: firstly when one of the four triangles defined by the points a, b, c has all its sides less than ^-rry, that is to say, has the three lengths D(bc), D(ca), D(ab) for its sides [cf. 204 (2)]; secondly, when one at least of the sides of each of the four triangles is greater than %Try. (8) Case I. Let no one of a, b, c be divided from the other two by their polar planes, then [cf. 205 (3)] (b \c), (ca), (a \b) may be assumed to be of the same sign; and this sign must be positive. Hence bc = D (be), ca = D (ca), ab = D (ab). Thus one triangle (the triangle abc) is formed by the intercepts of the lengths D (be), D (ca), D (ab) ; each being less than \iry. Then by subsection (6) the other three triangles formed by the three points are (i) that formed by the intercepts I) (be), Try D (ca), 7ry D(ab), with angles a, it JS, tt y ; (ii) that formed by the intercepts iry D (be), D (ca), iry D (ab), with angles it a, j8, it y; (iii) that formed by the intercepts iry - D (be), try D (ca), D (ab), with angles it a, it j8, y. (9) Each of these last three triangles has two sides greater than Try. Let the triangle with each side less than ^iry be called the principal triangle abc, let the other three be called the secondary triangles. (10) Case II. Assume that a is divided from b and c by the polar planes of b and c. Then [cf. 205 (3)] we may assume (b \c) and (a\b) to be positive, and (a j c) negative. Hence bc = D (be), ca = Try D (ca), ab = D (ab). Also (ab | ac) {= (a | a) (b \ c) (a \ b) (a | c)) is positive ; (be I ba) {= (b | b) (e \ a) - (a \ b) (b \ c)} is negative ; (ca \cb) {= (c \c)(a \b) (b \c) (c \a)} is positive. Thus, considering the triangle abc, the angles a and y are acute, and jS is obtuse; and the obtuse angle is opposite to the side greater than ^wy. The other three triangles, defined by the points a, b, c, are (i) that formed by D (be), D (ca), Try D (ab), with angles a, it jS, tt y. This triangle has one side, namely iry D (ab), greater than \try, and one obtuse angle, 7T y, opposite to it. (ii) The triangle formed by Try D (be), D (ca), D (ab), with angles tt a, tt j8, y. This triangle has one side, namely Try D (be), greater than iry, and one obtuse angle, namely tt a, opposite to it. (iii) The triangle formed by Try D (be), Try D (ca), Try D (ab), with the angles Tr a,J@,Tr y. This triangle has all its sides greater than \Try, and all its angles obtuse. (11) Thus in this case the points a, b, c define three triangles each with one side greater than ^iry, and one triangle with all its sides greater than ^iry. Call this case, the case with no principal triangle. This possibility respecting triangles in elliptic space of the polar form has apparently been overlooked. Let the set, of three triangles, each with one side greater than ^iry, be called the principal set.
374 ELLIPTIC GEOMETRY. [CHAP. II. 216. Further Formula for Triangles. (1) The two typical trans- formations, from which the further formulae connecting the sides and angles are deduced, are (a a) (abc \ abc) = (ab | ab) (ac \ ac) (ab | ac)2...............(i); and (b \c)(abc\abc) = (be \ba)(ca \cb) + (ab \ac)(bc \bc) .........(ii). Both of these formulae can be proved by mere multiplication. Thus for instance [cf. 120] (fee | ba) (ca cb) 4- (ab \ ac) (be ! 6c) = {(b\b)(c\a)-(a\b)(b\c)}{(c\c)(a\b)-{b\c)(c\a)} + {(a\a)(b\c)-(a\b)(c\a)}{(b\b)(c\c)-(b\cy} = (fe c) {2(6 c)(c a)(a b)+(a a)(b\b)(c\c)-(a\a)(b\cf-(b\b)(c\ay ~(c\c)(a\by} = (b\c)(abc \abc). \J\(ab I ab)(ac I ac) (ab ac)2} (2) Since sin a = V{1 - cos2 a[ = -^ it follows from equation (i) of subsection (1) that sin a = Aa 'aH^bcJaM y (ab \ab)(ac\ac)' (6c 16c) Li_ V ab)(ac\ac)} But Hence . bc_ I % V (b\b)(cc)- sin sin 7 sin a 6c . ca . ab sm sin sm 7 7 7 _ /(a \a)(b\6) (c c) (a6c |a6c) ~ V (6c|6c)(ca ca)(a6!a6) ' (3) From equation (ii) of subsection (1) 6c that is, sin j8 sin y cos = cos j8 cos y + cos a ; 6c cos a = cos jS cos y + sin jS sin y cos , with two similar equations. (4) If a, 6, c be at unit intensity then [cf. 120 (1) and 210 (2)] (a6c |a6c) = 1, ab ac cos , cos 7 7 ab - cos-, 1, ac be cos , cos , 7 7 6c" cos 7 This determinant is the square of the well known function, which in Spherical Trigonometry is sometimes called the Staudtian of the triangle.
216, 217] FURTHER FORMULA FOR TRIANGLES. 375 (5) It is evident that the usual formulae of Spherical Trigonometry, for example Napier's Analogies, hold for triangles in Elliptic Geometry. For these formulae are mere algebraic deductions from the fundamental formulae of 215 (3) and of subsections (2) and (3) of this article. (6) Let a circle be defined to be a curve line [cf. 67 (4)] in a two- dimensional subregion, such that each point of it is at the same distance (its radius) from a point (its centre) in the subregion. Then it follows from subsection (2) that the perimeter of a circle of radius p is 27ry sin . For consider the chord pq, subtending an angle a at the centre. Draw cl perpendicular to pq. Then, since by symmetry I is the middle point of pq, . pq .pi . a . cp . a . p sin ^ = sm = sm sin = sin 5 sin -. 2y 7 2 7 27 Therefore when a is made small enough, Yq = *7 sin . Accordingly, assuming that the length of the arc of a curve is to be reckoned as ultimately equal to the chord joining its extremities, the circumference of the circle = 2pq, ultimately, = 7 sin - 2a = 2tt7 sin - . 217. Points inside a Triangle. (1) Consider the triangle abc, that is, the triangle with its sides formed by the intercepts (6,c), (c \a), (a\b). Any point of the form Xa + fib + ve, where X, fi, v are of the same sign, will be said to be inside the triangle. Other points of this form will be said to be outside the triangle. (2) To prove that any straight line, in the two dimensional subregion defined by a, b and c, cuts the sides of the triangle, either two internally and one externally, or all three externally. Write p = Xa + fib + vc; and let px be any line through p and another point x in the two dimensional region. Without loss of generality we may consider that the complete manifold [cf. 103 (3)] is the two-dimensional region defined by a, b, c. Then px ,bc = Xax . be + fibx. be + vex. be = [X (xca) fi (xbc)} b + {X (xab) v (xbe)} e ; px . ea = {fi (xab) v (xca)} c + [p (xbc) X (xca)} a ; px. ab = {v (xbc) X (xab)} a + {v (xca) /jl (xab)} b. Let 01 = fi (xab) v (xca)f 02 = v (xbc) X (xab), 03 = X (xca) fx (xbc). Hence px .be = OJb 02c} px,ca 0xe O^a, px.ab = 02a Ofi.
376 ELLIPTIC GEOMETRY. [CHAP. II. Now px. be is the point of intersection of px and be ; and if 02 and 03 are of the same sign, this point is external to the intercept (6 !c) ; and if 02 and 03 are of opposite sign, the point is within the intercept (b |c). But 0ly 02, 03 are either all three of the same sign, or two are of one sign and the third of the opposite sign. Hence the proposition is evident. (3) Any line in the two dimensional region, which contains a point inside the triangle, cuts two of the sides internally and one externally; also conversely. With the notation of the previous subsection, assume that p lies within the triangle. Then X, fi, v may be assumed to be all positive. Also without any loss of generality, x may be assumed to be on the line be, so that (xbc) = 0. Then 01 = fi(xab) v(xca), 02 = \(xab), 0s = \(xca). Hence, if (xab) and (xca) are of the same sign, 02 and 0S are of opposite signs; also, if (xab) and (xca) are of opposite signs, 6Y is of opposite sign to both 02 and 03. Hence in either case the first part of the proposition is true. To prove the converse, assume that the sides (c\a) and (a\b) are cut internally at the points aa + 7c, OLxa + fij); where a, 7, a1} can be assumed to be all positive. Then any point on the straight line can be written in the form (aa -f 7c) + rj (^a + fab). Hence all points, for which and rj are of the same sign, lie within the triangle. 218. Oval Quadrics. (1) If three points a, 6, c, lie within [cf. 82 (1)] a closed quadric, (a$V)2 = 0, then the quadric cuts all of the sides of one of the triangles defined by the points a, b, c externally. For [cf. 208 (2)] we may assume (a$6][c), (a$c$a), (a$a][ ) to be all positive, when (a$x)2 is positive, x being a point within the quadric. Now with this assumption as to the terms a, b, c, consider the triangle abc. Let any side be cut the quadric in a point ^b + ve. Then fi? ( x bf + 2fjuv (a$6$c) + v2 (a$c)2 = 0. Thus the two roots for jjl : v given by this equation are both negative. Hence any side (b \ c) of the triangle abc is cut by the quadric in two external points. It follows that the sides of any of the remaining three triangles defined by the points a, 6, c are cut two internally and one externally. (2) An oval* quadric is a quadric which cuts externally the sides of any principal [cf. 215 (9)] triangle abc, of which the three angular points lie within it. (3) Let a sphere be defined to be a surface locus contained in the complete manifold [ 67 (1)], such that every point of it lies at a given distance (the radius) from a given point (the centre). * Oval quadrics have not, as far as I am aware, been previously denned. In the special case of Euclidean space of three dimensions, ellipsoids and hyperboloids of two sheets are both closed quadrics; but only ellipsoids are oval quadrics.
218] OVAL QUADRICS. 377 A sphere is a closed quadric. For if e be the centre and p the radius, the equation of the sphere is = cos2 , that is, (e I xf - (e j e) (x ]x) cos2 = 0. Now if ^ be a point at a distance from 6 less than p, then cos2 -; hence (e \y)2 {y\y) (e \e) cos2 - is positive. (y\y)(e\e) 7' Also there must be two real points on any line through y which lie on the surface. For, let any line through y cut the plane \e in e, so that (e \e') = 0. Then any point z on this line can be written y + %e. Hence (e\zf -(z\z)(e\e) cos2 = \(e \yf -(y \y)(e\e)cos2 2 (y \e)(e \e)cos2 f2(e \e) (e \e) cos2 - . 7 7 Hence, since -ue jy)2 (y \y)(e \e)cos2-l is positive, it is always possible to find two real values of for which (e \zf (z z) (e \e) cos2 = 0. Accordingly [cf. 82 (1)] any point at a distance from the centre less than the radius is within the sphere, and for such points (e \y)2 (y\y) (e \e) cos2 - is positive. (4) A sphere of radius less than J777 is an oval quadric. For let e be the centre of the sphere, and x and y two points within it. Then by (3) the two intercepts D (ex), and D (ey) both lie within the sphere and are cut externally by it. Now let the intercepts (e\x) and (e\y) be these intercepts, so that (e\x) and (e\y) are both positive. Then the triangle exy has two sides cut externally by the sphere, and hence by (1) the third side (x \y) is cut externally. ^ xy ex ey . ex . eii ~ But cos = cos cos + sin sin cos 6, 7 7 7 7 7 where 0 is the angle at e of the triangle exy. TT xy ex ey . ex . ey Hence cos cos cos ~ sin sin * 7 7 7 7 7 ex + ey cos------J 7 Hence xy ex + ey ^iry ; since ex and ey are by hypothesis each less than \iry. Thus xy = D (xy). Hence that intercept joining any two points within the sphere, which is cut externally by the sphere, is the shortest intercept. Hence the sphere is an oval quadric.
378 ELLIPTIC GEOMETRY. [CHAP. II. (5) It is also evident by the proof of the preceding subsection that any sphere of radius greater than \iry is not an oval quadric. Hence also it is easy to prove that any oval quadric can be completely contained within some sphere of radius \iry. (6) Furthermore it follows from (1) and (4) that any three points lying within a sphere of radius \iry define a principal triangle. 219. Further Properties of Triangles. (1) Two angles of a principal triangle [cf. 215 (9)] cannot be obtuse. For if possible let a and jS be both obtuse. Then from 215 (3) be ca ab . ab . ca cos = cos cos + sin sin cos a, 7 7 7 7 7 ca be ab . be . ab cos = cos cos----h sin sin cos a. 7 7 7 7 7 TT i ,, be ca ab , ca be ab Hence both cos cos cos and cos-----cos cos are negative, 7 7 7 7 7 7 _ since cos a and cos B are negative. But cos and cos are both H * 7 7 _ be positive by hypothesis and one of them must be the greater, say cos . Then cos-----cos cos has the sign of cos , and is therefore positive. 7 7 7 7 Hence there cannot be two obtuse angles in a principal triangle. It has been proved [cf. 215 (11)] that, if no principal triangle exist, the triangles of the principal set defined by a, 6, c have each only one obtuse angle, while the remaining triangle has three obtuse angles. (2) In any triangle abc [cf. 215 (7)] if j8 and y be both acute or both obtuse, the foot of the perpendicular from a on to be falls within the in- tercept (6 \c) ; otherwise it falls without the intercept (6|c). For let p = \b + fie be the foot of this perpendicular. Then (ap\bc) = \(ab\bc) + fi(ac\bc) = 0. Hence we may write p = (ca \ eb) 6 + (6a j 6c) a Now jS and y are respectively acute or obtuse according as (6a \bc) and (ea \cb) are positive or negative [cf. 215 (2)]. Hence the proposition. (3) If the angles j8 and y be both acute, the triangles abp and acp have JS and y respectively as angles (and not tt jS and ir y\ also the sum of their angles at a is equal to a. This proposition is easily seen to be true. (4) The sum of the three angles of any principal triangle, or of a triangle from a principal set is greater than two right-angles. Firstly, let the angle y be a right-angle, and let a and jS be acute.
219, 220] FURTHER PROPERTIES OF TRIANGLES. 379 Then by one of Napier's Analogies, , be ~ ca . be ~ ca COS ^ ------------ COS i ------------- tan , bc-\- ca , be + ca cos f -------- cos -------- 7 7 _ _ Now since be and ca are each less than ^iry, it follows that /be ~ ca , bc + ca cos i-------- cos i--------. 7 7 Hence a-h j8 ^ - Hence a + J3 -\-y ir. Secondly, let abe be any principal triangle or a triangle from a principal set. Then at least two of its angles are acute, say a and jS. Draw a perpendicular cd from c on to the opposite side. Then d lies between a and 6 on the intercept (a j 6), and abe is divided into two right-angled triangles. Hence obviously from subsection (3) the theorem holds for the triangle abe. 220. Planes one-sided. (1) It has been proved in 203 (1) that a plane does not divide space. An investigation of the meaning to be attached to the idea of the sides of a plane is therefore required. Let two points a and b be said to be on the same side of a plane P, when the intercept D (ab) does not contain the point of intersection of ab and P, that is to say, the point ab. P. Conversely when the intercept D (ab) does contain the point ab. P, let a and b be said to be on opposite sides of the plane. (2) Suppose that a and b are on opposite sides of the plane, but that they each approach the plane along the line ab so that D(ab) diminishes and ultimately vanishes. Then in the limit a and b, though coincident in position, both lie on the plane on opposite sides of it. Thus a plane can be considered to have two sides in the sense, that at each point of the plane there may be considered to be two coincident points on opposite sides of the plane. This idea can obviously be extended to any surface. (3) If a be any point on a plane P, then a and a may be considered as symbolizing the two coincident points on opposite sides of the plane. For let b be any other point not on the plane ; and assume, for example, that (a j b) is positive. Write a' for a. Then if a be considered to be on the same side of the plane as 6, the intercept (a \ b) does not contain a (by the definition of subsection (1)), the intercept {a' \b) does not contain a (since a'b is the length of the long (polar) intercept between a' and 6, namely, Try D (a6)), and the straight line is completed by the indefinitely small intercept D {aa') which passes through the plane.
380 ELLIPTIC GEOMETRY. [CHAP. II. Thus if b be a given representation of the point 6, which is taken as the standard representation, a is on the same side as the point b of any plane on which a lies when (a\b) is positive, and is on the opposite side when (a \b) is negative. It must be carefully noticed that the choice of sides for a and a depends not only on the position of the point 6, but also on the special term b which represents the point. For b represents the same point, and if b be taken as the standard representation, a and a would according to the above definition change sides of any plane on which they lie. (4) Suppose that a sphere of radius \ir^ be described cutting the plane, and that attention be confined to points within this sphere. Then [cf. 218 (6)] any three such points, a, b, c, define a principal triangle: let it be the triangle abc. Now if a and b be on the same side of the plane, then c is on the same side of the plane as a or on the opposite side of it, according as c is on the same side as b or on the opposite side. For the plane cuts the two dimensional region abc in a straight line, and by hypothesis this straight line cuts the intercept (a \b) externally, hence by 217 (2) and (3) it cuts the other two intercepts, (a\c), (b\c), both externally or both internally. Thus when attention is confined to the space within this sphere, the ordinary ideas concerning the two sides of a plane hold good. (5) But if the points a, 6, c do not define a principal triangle, let the triangle abc be one of the principal set. Assume that (a\b), (a\c) are positive and that (be) is negative. Now the straight line, in which any plane cuts the region abc} must cut the sides of the triangle abc either all externally or two internally and one externally. If the line cut all the sides externally, it cuts D (ab)y D (ac) externally and D(bc) internally. Hence a and b are on the same side of the plane, also a and c; but b and c are on opposite sides. If the line cut (a \b), (a\c) internally and (b \c) externally, it cuts D (ab), D (ac), D (be) all internally. Hence any two of the three are on opposite sides of the plane to each other. If the line cut (a \b), (b c) internally and (a \c) externally, it cuts D (ab) internally, and D (be), D (ac) externally. Hence c and b are on the same side of the plane, also c and a; but a and b are on opposite sides. Similarly if the line cut (a c), (b \c) internally and (a b) externally, then c and b are on the same side of the plane, also b and a; but a and c are on opposite sides. Hence, when three points do not form a principal triangle, the ordinary ideas concerning a plane dividing space cannot apply.
220] PLANES ONE-SIDED. 381 (6) It has been defined in (3) that if the point a lie on the plane P and b be another point not on the plane, then the term a symbolizes a point on the same side as the point 6, when (a \b) is positive. Let c be another point on the plane so that the triangle abc is a principal triangle. Then by hypothesis (b\c) is positive, and the term c symbolizes a point on P on the same side as the point 6. Hence, assuming that the theorems of subsections (4) and (5) are to hold when two angular points are on the plane, a and c are on the same side of the plane when (a\c) is positive. (7) If a and c be two points on the plane P and on the same side of it, then the point \a +/ic is defined to describe a straight line without cutting the plane, when any two neighbouring points of the line successively produced by the gradual variation of X and fi are on the same side of the plane. Suppose that X varies from 1 to 0 as ft varies from 0 to 1, then the intercept (a \c) is described without cutting the plane. Also every point on this intercept is on the same side as both a and c. But now starting from c let the moving point describe the other intercept without cutting the plane. Then \ must vary from 0 to 1 while /j, varies from 1 to 0. But the final point reached is a. Thus a moving point, starting from a and traversing a complete straight line drawn on the plane without cutting the plane, ends at a, that is on the opposite side of the plane. Again, if Q be another plane cutting P, and the subplane of inter- section does not cut either P or Q, then when the moving point starting from a has made a complete circuit of a straight line lying in the subplane PQ it is on the opposite side both of P and Q to a. In order to understand the full meaning of this property, consider for example space of three dimensions. Let the two sides of P at a be called the upper and under side, and the two sides of Q at a be called the right and left side. Let a dial with a pointer lie in the plane P at a with face upwards and pointer pointing to the right. Let the dial be carried round the straight line of intersection of the planes so that in neighbour- ing positions both face and pointer respectively look to the same sides of the two planes. Then, when the complete circuit has been made, the dial at a is face downwards and the pointer points to the left. The property of planes proved in this subsection is expressed by saying that planes are one-sided. The discovery of this property of planes in the polar form of elliptic geometry is due to Klein*. (8) The definition in subsection (7) of a straight line drawn on a plane without cutting it can obviously be applied to any curve-line drawn on the plane. Also by the method of (7) it is easy to prove that a point, starting from * Cf. Math. Annul., Vol. VI.
382 ELLIPTIC GEOMETRY. [CHAP. II. a and describing a closed curve on a plane P, returns to a or to a according as the closed curve cuts the subplane of intersection of P and the polar plane of a (that is, the subplane P \a) an even or an odd number of times. 221. Angles between Planes. (1) Since in Elliptic Geometry the absolute is imaginary, the separation [cf. 211 (2)] between planes must necessarily be measured by the trigonometrical formula and not by the hyperbolic formula. The same applies to the separation between any two subregions, when the idea of a measure of separation between them can be applied [cf. 211 (6)]. Let the measure of the separation between planes or between subregions (excluding points) be called the angle between them. Thus the angle between two planes X and Y is one of the two supplementary angles (less than ir). (2) Let XY stand for that one of the two supplementary angles between X and Y which is defined by (3) The points \X(=x) and \Y(=y) are the absolute poles of the planes X and F. The length xy of the intercept (x \y) is given by xy (x I y) ^ TT J{{\)(\)\ y cos = Hence ^ = 7 (4) If Z be a third plane, the angles between the subplanes XY and XZ are the two supplementary angles (less than ir) denned by ^{(XY\XY)(XZ\XZ)}" These angles are the same as those between the lines xy and xz, where z=\Z. 222. Stereometrical Triangles. (1) The angles which the planes A, B, C make with each other, and also the angles which the subplanes BO, CA, AB make with each other can be associated together by definition, so as to form what will be called a stereometrical triangle. Let the stereo- metrical triangle ABC be the association of the three angles BC CA, AB, with the three angles a, J3, y, denned by coca- (AB\AG) /{{AB\AB) (AC \AC)}' with two similar equations for /8 and y.
221 223] STEREOMETRICAL TRIANGLES. 383 (2) Then if a = \Ay b = \B, c = \C, the triangle abc is the ' polar ' triangle of the stereometrical triangle ABC. Also the angles of the triangle abc are respectively equal to a, jB, y\ while the sides of the triangle abc are respectively equal to 7 ( BC), 7 ( GA), y( AB). (3) Accordingly, corresponding to every formula for a triangle defined by three points there exists a formula for a stereometrical triangle defined by three planes. Thus the ordinary formulae of Spherical Trigonometry, in ordinary three dimensional Euclidean space, are shown to hold for the relations between three planes of any number of dimensions in Elliptic Geometry. (4) From 215 (3) it follows that cos BO cos CA cos AB + sin GA sin AB cos a; with two similar formulae. Now if the complete space be three dimensional, the subplanes BC, GA, AB are three straight lines meeting at a point; and thus a, JS, y correspond to the ' sides' of an ordinary three dimensional spherical triangle, while BC, CA, AB, correspond to the angles. Thus according to analogy the above formula ought to be cos BC = cos CA cos AB + sin CA sin AB cos a. This difference of sign is explained by noting that the angles to be associated with the stereometrical triangle ABC were defined by convention in subsection (1); and that if the angles of the triangle ABC had been defined to be ir BC, etc., and it a, etc., the signs of the formulae obtained would have agreed, when the complete region is of three dimen- sions, with those of ordinary Spherical Trigonometry. 223. Perpendiculars. (1) Any two mutually normal [cf. 108 (5)] points x, y are at the same distance from each other. For since (x \ y) = 0, cos = 0, and therefore xy = fary. Such points may also be called quad- rantal. The condition that two lines ab and ac should be at right-angles (or perpendicular) is (ab \ac) = 0. Lines, or other subregions, which are perpendicular must be carefully distinguished from lines, or other subregions, which are mutually normal [cf. 113 (1)]. (2) Let any region L9 of p 1 dimensions be cut by a straight line ab in the point a ; then, if (p 1) independent lines drawn through a in the region L9 be perpendicular to ab, all lines drawn through a in the region Lp are perpendicular to ab. For let aplt ap2,... opp_! be the (/o 1) independent lines. Then by hypothesis (ab ap^ = 0 = (ab ap2) = etc.
384 ELLIPTIC GEOMETRY. [CHAP. II. But \a + 2,/j,p represents any point in Lp. Hence any line through a is (filapl + fi2ap2 + ...) And {ab\(fi1ap1 + p2ap2 + ...)} = /^ {ab\ap1) + fi2(ab \ap2)+ ... =0; which is the required condition of perpendicularity. Then ab will be said to be perpendicular to the region Lp, or at right- angles to it. (3) Any line perpendicular to the region Lp intersects the supple- mentary (or complete normal) region \LP; and conversely, any line inter- secting both Lp and \LP is perpendicular to both. For, with the notation of the previous subsection, let ab be the line; and let b be the point on the line ab normal to a [cf. 113 (5)], then b is normal to every point on Lp. For, if p be such a point, (ab \ap) = 0. Hence (a \a) (b \p) - (a \p) (a\b) = 0. Hence (a \ a) (b \p) = 0. But (a | a) is not in general zero. Hence we must have (b \p) = 0. Hence 6 lies in \LP\ and therefore ab intersects \LP. (4) If Pp and Pa be two regions normal to each other [cf. 113], and if a be any point in Ppy then any line drawn through a in the region Pp is perpendicular to the region aPa. For let a be any other point in Pp, and b be any point in Pa, then by hypothesis, (a b) = (a' | b) = 0. Hence (aa! \ab) = (a\a)(a \b)-(a\b)(a' \a) = 0. (5) Let two planes L and M intersect in the subplane LM, and ax be any point in LM. From ax draw aj, in the plane L perpendicular to the subplane LM, and draw axm in the plane M perpendicular to the subplane LM, then the angle between L and M is equal to that between aj and axm. For in the subplane LM, which is of v 3 dimensions (the space through- out this chapter being of v 1 dimensions), we can find [c 113 (5)] v 3 other points a2, a3, ... av_2, so that aly a2, ... av-2 are mutually normal. Also take I in the line aj to be the point normal to ax. Then by subsection (3) I is normal to a1} a2, ... av-2] and therefore to every point in LM. Similarly in the line a^rrt let m be normal to every point in LM. Then (a21 a2) = (a1 \ a3) = (ap \ aa) = ... = (ap \ a,,_2) = 0, and (a1 \ I) = (a2 I) = ... = (a,_211) = 0, and (on |ra) = (a2 \m) = ... = (av-2 \m) = 0. Also we may write L = (a^ ... av_2l), and M = {a1a2... av-2m). Then from 120 (1) (t |X)=(a1|a1) (a2|a2) ... (l\l), (M M) = (a1 \a,) (a2 |a2) ... (m \m),
224] PERPENDICULARS. 385 Hence if 6 be the angle between L and if, and f between aj and a{m, )} " L){M\M)) *J{(l\l) (m \m)} *J{{aJL \aj) (a.m \aYm)} Thus 0 = 0. Corollary. It is also obvious that 0 = j = . 7 (6) Any line perpendicular to any plane L also passes through its absolute pole [cf. subsection (3)]. Thus if any plane M include one perpendicular to L, then from any point of the subplane LM a perpendicular to L can be drawn lying in M. For, if M includes one perpendicular to i, it includes the pole \L. Then any line joining any point in LM to \L must be perpendicular to L and must lie in M. Also since \L lies in M then \M lies in L. Hence this property is reciprocal. Such planes will be said to be at right angles. It is obvious that, if two planes are at right angles, their poles are quadrantal. (7) If two planes L and U be each cut perpendicularly by a third plane M, it follows at once from the formulae for stereometrical triangles investigated in 222, that the angle between the subplanes LM and L'M is equal to that between the planes L and L'. 224. Shortest Distances from Points to Planes. (1) The shortest distance from a point to a plane is the shortest intercept of the straight line through the point perpendicular to the plane. For let x be the point, p the foot of the perpendicular, and q any other point on the plane. Let the terms x and p be so chosen [cf. 206 (9)] that xp D (xp); so that xp is the shorter of the two intercepts between x and p. Then by 215 (3), cos = cos cos . Hence, if pq be greater than y, xq is also greater than ^ttj. Thus the points x, p, q must define a principal triangle. Let the terms x, p} q be so chosen that xpq is this principal triangle. Then from the above formula, D {xq) D {xp). This length of the perpendicular will be called simply the distance of the point from the plane. (2) It is obvious that the other intercept of the straight line xp is the longest intercept of a straight line drawn from x to the plane. (3) The pole of the plane is easily seen to be the point which is further from the plane than any other point, namely at a distance ^"7. w. 25
386 ELLIPTIC GEOMETRY. [CHAP. II. (4) Let p be the distance of the point x from the plane L. Then \irf p is the distance between x and the point | L. Hence sin - = cos 2 ^ 7 V 7 [ where, as in the other cases, the ambiguity in sign is to be so determined as to make sin - positive. With this understanding we may write , 011^) , {( 10 (| ![ )}* Sm7 {(x\x)(L\L)}*' 225. Common Perpendicular of Planes. (1) The line joining the poles \L and \U of any two planes L and L is obviously [cf. 223 (3)] perpendicular to both planes L and L\ Further, any point on the line \L\L' is normal to any point on the subplane LL'. Let the line \L\L' intersect the planes in I and V. Let a be any point on the subplane LL'. Then it is easy to prove that al and al' are each perpendicular to the subplane LL'. Hence the angle between al and al' is equal to the angle X between the planes. Accordingly in the triangle lal', the two angles at I and V are right- angles, al and al' are each ^7ry, and a is \. Hence IV \y. Fig. 3. It is to be noted that there are two lengths A/y and (ir X) 7 for D (W) ; the shortest of the two is meant according to the usual convention. (2) It is easy to see that D {IV) is greater than the distance of any point x in either plane from the other plane. For let x lie in X, and draw xp perpendicular to 27. Then xp passes through \L'. Also the distance from \L to p equals that from \U to l\ both being 777; but that from \U to I is less than that from \L' to x, since the line from \L' to I is perpendicular to L. Hence D{IV) is greater than D{xp).
226] DISTANCES FROM POINTS TO SUBREGIONS. 387 226. Distances from Points to Subregions*. (1) The least distance of a point a from a line be can be found. For let p be the foot of the perpendicular from a to be, and let b be any other point on be. Then, by the same proof as in 224 (1), the three points a, b, p define a principal triangle. Let this triangle be abp. Then, as in 224 (1), D(ap) D (ab); and hence D (ap) is the least distance which it is required to express. P Fig. 4. But .pa . ab . Q sin ^ = sin sin p, 7 7 where jS is the angle at 6 in the triangle abp, that is, the angle at 6 in the triangle a 6c, if the term c be properly chosen. ab I (ab\ab) But and Hence /{(616)(a6c|a6c)j \f {(ab\ab)(bc\bc)Y . pa I (abc | a6c) m ---- = A / z----j---r- 7 V (a | a) sin ) (6c 16c) Therefore, if F be the force 6c, the distance ( ) of a from the line of F is given by . g_ / (aF\aF) Sm7 V (a a)(F\FY (2) This formula can be extended to give the least distance of any point a from any subregion Pp of p 1 dimensions. For the argument of 224 (1) still holds, and the least distance is evi- dently the length of the perpendicular ap from a to the subregion. One, and only one, such perpendicular line always exists, since [cf. 72 (5) and 223 (3)] it intersects both Pp and | Pp. Let F be a force on any line through p in the region Pp, and let Pp_2 be the subregion in Pp normal to F. Let Pp = PPp_i. ,in^_ /JaF\aF) 7 V (a\a)(F\FY Now since ap is perpendicular to Pp it passes through the normal point to Pp in the region aPp. Let pp be this point. Then pp is normal to every point in Pp. * These formulae, and the deductions from them in subsequent articles, have not been stated before, as far as I am aware. 25 2 Then
388 ELLIPTIC GEOMETRY. [CHAP. II. Let a =pp + \p; where the two equations (pp |PP) = 0, and (pF) = 0, hold. Therefore (aF) = (ppF). Also (aPp \aPp) = (ppFPp_2 \pPFPp,2) = (PpF\PpF) (Pp_2 |PP_2) And Therefore Hence if 8 be the distance of a from the subregion Pp, then . 8 _ / (aPp | aPp) *my~V (a\a)(Pp\Ppy (3) This formula includes all the other formulae for distance from a point. For if Pp denote a point b, then it becomes . 8 / (ablab) sm-= / v, ' ' , 7 V (a a){b\b) which is in accordance with 204 (1). And if Pp denote the plane L, then since (aL) is numerical, hence the formula becomes in accordance with 224 (4). (4) Since ap is perpendicular to Pp, it also intersects |PP and is perpen- dicular to it. Let q be this point of intersection. Then D (pq) = \iry} and D (aq) = tyry ap. Thus if 8' be the distance from a to |PP, 8' . S / (aPp|aPp) cos - = sin - = / v ; . . 7 7 V(a|a)(Pp|Pp) But also by the same formula, replacing Pp by |PP, . S'_ /(ajPp|.a|Pp) Sm7~V (oo)(Pp|Pp)- Hence is obtained the formula This formula is easily obtained by direct transformation by taking [cf. 113 (7)] p mutually normal points in Pp and v p mutually normal points in | Pp as reference points, as in 229 following. 227. Shortest Distances between Subregions. (1) Let Pp and Qa be two non-intersecting subregions of the pth and crth orders respectively, so that p + t v. A line, such that one of its intercepts is a maximum or a minimum distance between them, is perpendicular to both.
227] SHORTEST DISTANCES BETWEEN SUBREGIONS. 389 For let a, b, c be three points, and let D (be) be small. Then a} b and c define a principal triangle : let this triangle be the triangle abc. Then it is easy to prove from the formulae [cf. 215 and 216] that D(ab) ~ D(ac) is a small quantity of the second order compared to D (be) when, and only when, the angle at b is a right angle. The main proposition follows immediately from this lemma. (2) Let p r\ then the polar region, \Qa) of Qai intersects Pp in a subregion of the (p - o-)th order at least. Also any line pq, from a point p in this subregion to any point q in Qa, is perpendicular to Qa, and is of length ^ttj. Accordingly such perpen- diculars from Pp to Qa are of the greatest length possible for the shortest intercept of perpendiculars from Pp to Qa; but they are not necessarily perpendicular to Pp. The polar region, |PP, of Pp does not in general intersect Qa. Hence in general there are no such perpendiculars from Qa to Pp of length \ir y. (3) Let q1} q2, ... qa be a independent points in Qff. Then any point x in Qff can be written 2 #. Hence [cf. 226 (2)] the perpendicular 8 from x to Pp is given by sin = . / 7 V Write X for sin - , and square both sides, and perform the multiplication; then x (P, \pp) {tt(qi\qi) + (q* I ) + ... + 2 (ft |ft) + ...} If S be a maximum or a minimum for variations of x in Qai then X is a maximum or minimum for variations of x, 2, ... a. CT 3x A ax ax Hence ^ = 0 = ^r = ... = k* 9fi a a?,, Thus by differentiation K^Pp |?1PP) - x (Pp I Pp) (fc |?1)} f, + f^Pp !?2Pp) - x*(Pp |PP) (?i I?.)} f. + ... = 0; with o- 1 other similar equations. Thus, by eliminating , fa, ... ^, an equation is found for X2 of the form ai - ^ (ft | ft), 22 - ^ (ft | ft), ., a2a - ^ (ft I ft),
390 ELLIPTIC GEOMETRY. [CHAP. II. ail = -jp^py , au -a, with similar equations defining the other as. (4) Hence, if Pp and Qo- be two subregions of the pth and rth orders respectively (p a), there are in general a common perpendiculars to the two subregions, which are the lines of maximum or minimum lengths joining them. If Pp and Qa had been interchanged in the above reasoning, an equation of the pth degree (p a) would have been found. But this equation would not merely determine the common perpendiculars to Pp and Qa. For, if 8 be the length of the perpendicular from any point in Pp to Q0y then, with the notation of the previous subsection, . 8 d\ 8 d8 X = sin -, ^7 = cos - . -^ 7 9fi 7 79? Hence ^ = 0, when 8 = ^ttj, as well as when 8 is a maximum or a minimum. Thus the infinite number of lines discussed in subsection (2) fulfil the conditions from which this equation of the pth degree is derived. (5) A formula can be found which determines the r feet in Qa of these perpendiculars. For, if q1} q2, ... qa, be these feet, then in the equation for X2 of subsection (3) the a roots must be [cf. 226 (2)] and so on 'ana on* and so on (ft WO{*, W' (ft lft)CPp !A)'ana on* Hence equations must hold of the type {qJPp \ q^Pp) = 0, and of the type (?ilft) = 0. Thus q1} q2, ... q , are the one common set of a polar reciprocal points [cf. 66 (6) and 83 (6)] with respect to the sections by Qa of the two quadrics \x\x) = 0, and (ccPp \xPp) = 0. Thus the a feet in Qa are mutually normal. (6) These common perpendiculars all intersect \PP [cf. 223 (3)]. These a- points of intersection with \PP are also mutually normal. For any line joining Pp and Qff must lie in the region PpQa of the (p -f cr)th order defined by any p + a reference points, p from Pp and cr from Qa. Also the common perpendiculars, being perpendicular to Pp, all intersect the region \PP (of the (v p)th order), and are perpendicular to it. Hence they all intersect the subregion, P^lPp, formed by the intersection of \PP with PpQ r. But this subregion is of the erth order. Then the common perpendiculars of Qa and Pp are also common perpendiculars of Qa and f since PpQa\Pp is part of \PP. But by the previous subsection, if - - p*', be the a feet in PpQ^Pp of these perpendiculars, then
228] SHORTEST DISTANCE BETWEEN SUBREGIONS. 391 Pi\ P2 pJ form a mutually normal set of points. Also they are the one common set of a polar reciprocal points with respect to the sections by PpQo \PP of the two quadrics (w\x)= 0, and (xQff \xQa)= 0. (7) Now, since 2 r v the 2er points ql9 q2y ... qv, px\ p2, ... pj are in general independent. Hence the a- lines Pi'qu p2q2, , cut Pp in a inde- pendent elements plf p2, ... pa, which define a subregion Pa. Then by subsection (5) plf p2, - . p* are mutually normal. But they are also normal to p[, p2y ...pj. Thus [cf. 113] the r lines of the r common perpendiculars are mutually normal, so that any point on one is normal to any point on the other. (8) The theorems of subsections (5) to (7) can be proved otherwise thus, assuming subsection (1) and that one common perpendicular exists between Pp and Qa. For, since this perpendicular (call it i^) intersects Pp at right- angles, then [cf. 223 (3)] any line drawn in Pp through the point of inter- section intersects the region \FX. But p 1 independent such lines can be drawn. Thus \FX intersects Pp in a region of the (p l)th order: similarly it intersects Q^ in a region of the (a l)th order. Let these regions (both contained in \Fj) be called Pp_a and Qa-i. Then by the original assumption Pp_! and Qa-! have a common perpendicular. Call it F2. Then F2 lies in \F1 and is therefore normal to it. Also \F2 intersects Pp_x and Qa^ in two regions of the (p 2)th and (cr 2)th order; and so on. Hence ( r p) by continuing this process, a common perpendiculars can be found, all mutually normal. 228. Spheres. (1) Let b be the centre and p the radius of a sphere ; its equation is (x\x){b\b)cos2P-=(b\xy2. But [cf. 110 (4)], (b |#) = 0, is the equation of the polar plane of b with respect to the absolute. Hence [cf. 78 (2)] a sphere is a surface of the second degree, touching the absolute along the locus of contact of the tangent cone to the absolute with b as vertex. (2) It is obvious that any point on a sphere of radius p and centre 6 is at a distance ^iry p from the polar plane of 6, viz. from \b. But |6 may be any plane since b may be any point. Hence a sphere of radius p is the locus of points at constant distances, ^ttj p, from a plane. A plane can be conceived to be the limiting case of a sphere of radius ^77-7. For if b be the absolute pole of any plane, the equation of the plane is | ) = 0; and this is the degenerate form of the equation of the sphere, when p is put equal to
392 ELLIPTIC GEOMETRY. [CHAP. II. (3) Every line, perpendicular to a plane and passing through the pole of the plane with respect to a sphere, passes through the centre of the sphere. For let b be the centre of the sphere, p its radius and let p be the pole of the plane with respect to the sphere. The equation of the plane can be written (x \p) (b \b) cos2 ^ - (x |b) (p \b) = 0. Hence [cf. 110 (4)] the absolute pole of the plane is But [cf. 223 (3)] the perpendicular lines to the plane pass through the absolute pole. Hence the perpendicular through p to the plane is the line that is, dropping numerical factors, the line pb. Accordingly this line passes through b. Corollary. The perpendicular to a tangent plane of a sphere through its point of contact passes through the centre of the sphere. (4) Let the length of a tangent line from any point x to the sphere, centre 6, radius p, be r. Let the line meet the sphere in p. Then consider- ing the triangle xpb, by the last proposition, the angle at p is a right-angle. Hence the triangle xpb may be assumed to be a principal triangle. Fig. 5. tt xb xp p t p .Hence cos = cos cos - = cos - cos -. 7 7 7 7 7 Therefore the lengths of all tangent lines from x to the sphere are equal. Substituting for cos , we find (x | b)2 = cos2 - cos2 (x I x) (b I b). 7 7 Hence when t is constant the locus of x is a sphere concentric with the original sphere.
228] spheres. 393 In order that the tangent t may be real, we must have ( Hence / )X}J,^ cos2 . (x\x)(b\b) 7 Therefore the point x must be at a distance from b greater than / , that is to say, must be outside the sphere [cf. 218 (3)]. (5) The intersection of any plane with a sphere is another sphere of v 3 dimensions contained in the plane [cf. 67 (1)]. For let L be the plane, and (# |6)2-(#|#)( ! ) cos2 = 0........................(i) be the equation of the sphere. Then any point p on the perpendicular from b to L [cf. subsection (3)] can be written b + \ \L. Hence the distance S from p to any point x is given by J_ (x\pj _ Now let x lie on the locus of intersection of L with the sphere, then (xL) 0, and x satisfies equation (i). s (b\b) cos* Hence C0"2 = ^ Thus the distance of any given point on the line b\L from any point on the locus of intersection of L and a sphere, centre 6, is constant. But this must hold for the point L ,b\L, where the line b |L intersects the plane L. Hence the locus of intersection is a sphere of v 3 dimensions, with the point L. b \ L as centre. The radius Sx of this sphere (of v 3 dimensions) is easily proved by equation (ii) to be given by ~(Lbf nno2 _} ..... T {b\b){L\L) (6) The locus of the intersection of any two spheres is contained in two planes, the radical planes. For let (x \ x) (b \ b) cos2 - (6 | x)2 = 0, and (x \x)(c\c) cos2 - - (c |xf = 0, be the equations of the two spheres.
394 ELLIPTIC GEOMETRY. [CHAP. II. Then two planes containing the locus of intersection are given by (b \xf (e \e) cos2 - - (c \x)2 (b 16) cos2 = 0 ; 7 7 that is, by the two equations, (b \x) V(c ]c) cos - (c \x) \/{b \b) cos - = 0. Let these planes be called the radical planes. (7) These planes are the loci from which equal tangents can be drawn to the spheres. Also from subsection (5), it follows that the locus of inter- section of two spheres consists of two spheres of v 3 dimensions. (8) Spheres cut each other at two angles of intersection, one correspond- ing to each radical plane. For consider the radical plane ............(iii)- )(^)V() 7 7 Then for points on this plane the equations of the two spheres can be written (iv), (b \x) = \/(x |x) (b 16) cos , (c \x) = V(x \x) (c \c) cos - , where the same choice (upper or lower sign) determining the ambiguity is to be made for both equations. Also the angle a , at which the spheres cut each other along this plane, is the angle between the lines bx and ex through a point x on that part of the locus of intersection contained by the plane. TT (xb \xe) Hence cos co = u yv ' y -. *J\(xb \xb)(xe \xc)\ Hence, eliminating x by the use of equations (iii) and (iv), (6 \c) V(6 \b)(c\c)cos^ cos - COS ft) = ' ' *J{(b \b)(c\c)} sin sin - 7 7 S p a cos----cos ~ cos - sin ^ sin - 7 7 where 8 is the length be of that intercept between the points 6 and c defined by (6|c).
228] spheres. 395 Similarly for the other radical plane (b \x) V(c |c) cos - + (c \x) V(6 \b) cos = 0, the angle ' of intersection between the spheres is given by (6 |c) + \/(b\b)(c\c) cos^ cos - ' 7 7 COS ft = ^ ^ i i 8 pa cos - + cos - cos - - i ,.% T................................. sm c sin - 7 7 Hence if 8' be the length of the other intercept between 6 and c, so that 8 -f S' = 7T7, then 8' pa cos-----cos c cos - ry ry ry COS (7T ft)') = . 7 . p . a sin - sin - 7 7 This equation exhibits the identity of type between the formulae of the equations (v) and (vi). Corollary. It may be noted as exemplifying equations (v) and (vi) that, if ) = , then cos - = cos - cos -. 2 7 7 7 Hence cos co = 2 cot - cot - . 7 7 This illustrates the fact that both parts of the intersection are not necessarily simultaneously real. (9) Let it be assumed that (61 c) is positive, so that B = D (be) ^ttj. Also by definition, p and a are both less than ^7ry. Then the two spheres have real or imaginary intersections on the corresponding radical planes, according as cos co and cos / are numerically less or greater than unity. Now cos co is positive ; and cos cof 1, .P D(bc) p t . p . a if cos - - + cos - cos - sm - sm - ; 7 7 7 7 7 .Li a. ' r (^C) 7r7 ~ P "~ a that is, if cos - cos c-----; 7 7 that is, if D(bc) + p + r 7ry ...........................(vii). Also if cos )' be less than unity, cos co is necessarily numerically less than unity. Hence if co be real, co is also real; and both parts of the intersection are real.
396 ELLIPTIC GEOMETRY. [CHAP. II. Thus the condition that the intersections of the spheres with both radical planes may be real is D (be) +p + r 7ry. (10) If D (6c) + p + cr 7ry, then one of the intersections is imaginary. The condition that the other may be real is D (be) p r cos - - cos - cos . p . J sin - sin - u D (be) p + Hence cos - cosE 7 7 p ~ a cos~----. 7 Therefore the condition that one intersection (at least) may be real is p~a D(be) p + r........................(viii). (11) It follows from the inequality (vii) of subsection (9) that two oval spheres cannot have two real intersections. For D (be) |7ry, and by 218 (4) for oval spheres p and a are both less than J7ry. Hence D (be) + p + a 7ry. (12) Let the sphere, centre c, reduce to the plane Z, so that we may write | c = L, and a = \irj. Hence the angle to at which the plane cuts the sphere, centre 6, is given by sin - cos co = If the plane cut the sphere at right-angles, (bL) = 0. Hence the plane contains the centre of the sphere. (13) If the plane touches the sphere, o= 0. Hence the plane-equation [cf. 78 (8)] of the sphere, of centre b and of radius p, is 7 If the sphere be defined as the locus of equal distance a from the plane B, then remembering that B = \b, and that p + r = ^7ry, this equation be- comes (B\B)(L\L)cos2-=(B\L)\ 7 Similarly the point equation of the sphere takes the two forms (x \x)(b\b)cos2P- = (b\x)2, and (x \x)(B\ B) sin2 - = (xB)\ 7
229] PARALLEL SUBREGIONS. 397 229. Parallel Subregions. (1) Let Pp be a subregional element of the pth order, then the locus of points x} which are at a given distance 8 from the subregion Pp, by 226 (2) is determined by the equation, This is the equation of a quadric surface. In the special cases in which Pp is either a point or a plane, the surface reduces to a sphere. (2) If Pp be neither a point nor a plane, real generating regions exist on this surface. For let e1} e2,... ep be p mutually normal points in Pp, each at unit normal intensity; and let ep+1, ... ev be v p mutually normal points in |PP, each at unit normal intensity. Then elf e2i... ep, ep+1 ...ev form a set of v mutually normal points at unit normal intensity. Let e e2... ephe written for Pp, then (Pp |PP) = 1. Also let x be written %%e + 2 ye. Then (x \x) = f + if. i p+i i p+i Also (xPp\xPp)= 2 7}k2(eKe1...ep\eke1...ep)+ 2 2 A=p+1 A + l where \ and fi are assumed to be unequal in the double summation. But (e* \ea) = 1, for all values of a ; and (ea \er) = 0, for all unequal values of a and t. Hence {eKex ...ep le^ ... ep) = 1, and (e^ ...ep\e^ex... ep) = 0. Thus (P|P) ^ p+i Hence the equation of the surface takes the form sin2 - i^ - cos2 - 2 r} * = 0. 7 i Yp+i If p \v, then [cf. (80) (1) and (5)] noting fche formation of conjugate sets of points from reciprocally polar sets, and remembering Sylvester's theorem [cf. 82 (6)], it is evident that real generating regions defined by p points exist on the surface, that is, regions of p 1 dimensions. If p \v, then real generating regions defined by v p points exist on the surface, that is, regions of v p 1 dimensions. If p = \v (y even), then real generating regions defined by p points exist on the surface, that is, regions of p 1 dimensions. (3) Let these real generating regions be said to be parallel to Pp. Thus a region parallel to Pp is by definition such that the distances from all points in it to Pp are equal, and has been proved to be of the type Qp or Qv-P, according as p or v p is least. Also from 226 (4) a surface of equal distance from Pp is also a surface of equal distance from |PP. Thus all regions parallel to Pp are also parallel to |PP, and conversely.
398 ELLIPTIC GEOMETRY. [CHAP. II. (4) Let the region Qa be parallel to Pp; where a is equal to the least of the two p and v p. Let q be any point in Qai and let qp be the perpen- dicular from Qa to Pp. Then gp is also the perpendicular from p to Qff. For if not, let pq' be this perpendicular. Then D (pq) D(pq). Also from q\ let ^'p' be drawn perpendicular to Pp. Then either^' coincides withp, or D(p'q') D(pq'). Hence in any case D (p'q') 1) (pq). But, since the region Qa is parallel to Pp, D (p'q') = D (pq). Thus pg must be a common perpendicular of Pp and Qa. Thus, if for example p be less than v pf so that r = p, then Pp is parallel to Qp. Thus Pp and Qp are mutually parallel to each other. But the same proof does not shew that \PP is parallel to Qp. For, by the preceding subsection, no region parallel to Qp can be of an order greater than the pth ; and by hypothesis v p is greater than p. Also, if S be the distance of Pp from Qp, the entire region parallel to Qp at a distance ^7ry S from Q9 must be contained in |PP. Hence a subregion P'p of |PP of the pth order is parallel to Qp. Accordingly a distinction must be drawn between the fact that one region is parallel to another region, and the fact that two regions are mutuall}7 parallel. Thus with the above notation, Qp is parallel both to Pp and to |PP; also (p v p) Pp and Qp are mutually parallel; but !PP and Qp are not mutually parallel, though P/ (a subregion of | Pp) and Qp are mutually parallel. The feet of the perpendiculars from all points in Qp to I Pp must lie in this subregion Pp'. This agrees with 227 (7) : the perpendiculars found by the method of that article must be all equal. (5) This theory of parallel regions is an extension* of Clifford's-)- theory of parallel lines in Elliptic Space of three dimensions. Consider a straight line L in a space of v 1 dimensions, (v ^ 4). Then the regions parallel to L are also straight lines, whatever be the dimensions of the space, provided that they are equal to or greater than 3. If the space be of three dimensions, then only two parallels to L can be drawn through any given point %, being the two generating lines of the quadric surface through x of equal distance from L. But if the space be of more than three dimensions, an indefinite number of parallels to L can be drawn through any given point. Also the tangent plane at x to the surface of equal distance from L which passes through x cuts this surface (a quadric) in another quadric of one lower dimension. Hence [cf. 80 (8) and (12)] this quadric is a conical quadric formed by the parallels through x. Thus in a region of v 1 dimensions the parallels through x to a line L form a conical quadric of v 3 dimensions with x as vertex. * Hitherto unnoticed as far as I am aware. t Cf. Preliminary Sketch of Biquaternions, Proc. of Lond. Math. Soc, vol. 4, 1873, reprinted in his Collected Papers.
CHAPTER III. Extensive Manifolds and Elliptic Geometry. 230. Intensities of Forces. (1) In considering the special metrical properties of extensive manifolds we shall confine ourself to three dimensions. The only regional elements in this case are planar elements and forces. The intensity of a planar element X is now taken to be [cf. 211 (5)] (X \X). The intensity of a force F has now to be determined. (2) Let* it be denned that the intensity of the force xy is some function of the distance xy multiplied by the product of the intensities of x and y. Thus assume that the intensity of xy is *J{(x \x) (y \y)] f (xy) ; where the function j (xy) has now to be determined. Now let x and y be at unit intensity, and let a be any number, then xy x{y + ax). Hence {(y + ax) \(y +ax)} = (y\y)+2a (x\y) + a* (x\x" = 1 + 2a cos ^ + a2. 7 Accordingly the intensity of xy = the intensity of x (y + olx) ^J j [x (y + cue)}. Therefore, / (xy) = (1+ a2 + 2a cos -^) f {x (y + ax)}. . xy sin But sin Hence __ . xy . x (y + i sin sin-^--------- 7 7 * This reasoning is very analogous to some reasoning in Homersham Cox's paper, cf. loc. dt. p. 370.
400 EXTENSIVE MANIFOLDS AND ELLIPTIC GEOMETRY. [CHAP. III. But x (y + ax) can be made to be any length ( ^7) by choosing a suitable value for a. Hence xy sin = a constant = 1, say. Therefore whatever points x and y are, the intensity of xy is K^k)(yly)}*sin . (3) Hence the intensity of the force F is (F\F)K Thus if the force F be written PQ, where P and Q are planar elements, the intensity of F is 231. Relations between two forces. Let .F and F' be any two forces. In general there are only two lines intersecting the four lines F, F', \F, \F'. These two lines [cf. 223 (3)] are two common perpendiculars to F and F' [cf. 227 (7) and (8)]. Let one perpendicular intersect F and Ff in a and b respectively, and let the other intersect F and F' in c and d. Let ab = 8, and cd = 8'. Then one Fig. 6. of the two is the shortest distance from one line to the other, and the other is the longest perpendicular distance. Also since ab intersects F, F\ \F, \F', \ab intersects the same four lines. Hence cd is the line \ab, and ab is the line \cd. Thus db = ^7 = ac. Hence (a \c) = 0 = (b \d) = (a \d) = (b |c). (2) To prove that* (FFr) .8.8' (F\Ff) 8 8r (a I Let Then Hence But Similarly, \ac, F' = \'bd. , cos 0 = 8 8' (a 16) (eld) C0S7cos7-{(a|a)(6|6)(c|C)(d|d)}i- |oc) = V {(a \a)(c \c) - (a (J" | J") = \'2(616)(d \d). * Cf. Homersham Cox, loc. cit. X2 (a |a)(c |
231, 232] RELATIONS BETWEEN TWO FORCES. 401 Also (F|F) = XX' (ac \bd) = X\'(a \b)(c d). Therefore cos - cos - = ^ (3) Again, from (1) we may write cd = fi \ab. Hence (cd | cd) = /*2 (a 1 ab). Therefore [cf. 204 (1)] 7 Also (a cd) = /x (aft |a6) = fi (a \ a) (b \b) sin2 - = sin - sin - {(a \a)(b\b) (c\c) (d\d)}K But (FF') = \\'(acbd). Hence assuming that the ambiguity of sign is so determined as to make both sides positive, (FFf) . 8 . 8' {{F\F)(F'\F')}*-*mv*my' (4) If the lines F and F' intersect, either 8 or 8' vanishes, say 8' = 0. Then (FF) = 0, and {(F ,-jy^yjjj = cos ?. This agrees with 211 (6). 232. Axes of a System of Forces. (1) A system of forces (S) can in general [cf. 174 (9) and 175 (14)] be reduced in one way and in one way only to the form*, S = a1a2 + e \aYa2. Then the lines of the forces a^2 and e [a^ are called the axes of the system (sometimes, the central axes), and the ratio of their intensities, namely e for -), is called the parameter of the system. Then (88) = 2e {axa2 \ a^), and (8 \8) = (1 + e2) (a,a2 \ (2) Let 8 denote the system F + e\F, and 8' the system F' + rj\F'. Also with the notation of 231 let 8 and 8' be the perpendicular distances between the lines F and F\ reckoned algebraically as to sign. * Cf. Homersham Cox, loc. cit., p. 370. W, 26
402 EXTENSIVE MANIFOLDS AND ELLIPTIC GEOMETRY. [CHAP. III. Then {SS') = {I+ V){FF')+{ + V){F\F') = {(F\F) (Ff \F*)}* J(l + eV) sin 8 sin - + (e + v) cos- cos -1; and (S\S') = ( + v)(FF')+(l+eV)F\F') = {{F F) (Ff \F')) j(e + v) sin S sin - + (1 + eV) cos - cos -I. (3) The simultaneous equations (#$') = 0, (8\8') = 0, in general secure that the axes of S and S' intersect at right angles. For from (2) unless either e or rj be + 1, 8 S' A . S . S' cos - cos - = 0 = sin - sin - . 7 7 7 7 Thus S = 0, 8' = ^Try; or vice versa. Hence F and F' intersect at right angles. Therefore [cf. 223 (3)] ^intersects \F' as well as F'\ and F' intersects ! F as well as F. Also, since {FF') = 0, (] F \ F') = | {FF') = 0. Thus | F and .F' intersect. Also these various pairs of lines [cf. 223 (3)] intersect at right angles. (4) Every dual group contains one pair of systems, and in general only one pair, such that their axes intersect at right angles. Let Si and S2 define the dual group, and let S = XA + \ , S' = m Then, (/S5f/) = 0, becomes V, (8181) + {Xlfi2 4- X^O (S ) + X^ ( Sf2^2) = 0. Similarly {S\S') = 0, becomes . ft) = o. Hence eliminating ^ : /x2, the pair of systems are given by the quadratic for Xj: X2, V {(SA) (5,1-S.) - (8,8,) (8, [S,)} - V {(8 ) (8, \8J - (8,8,) (8, \S2)} + X,X2 {(-8,5.) (8, \8,) - (8 ) (S, 18,)} = 0. Let this pair of systems be called the central pair of the dual group, and let the points at which their axes intersect be called the centres of the group. There are [cf. subsection (3)] four centres to a dual group, forming a complete normal system of points. If the group be not parabolic [cf. 172 (9)], the two director forces D1 and D2 may be written for ft and S2 in the above equation. The equation then becomes (AA) {V (A | A) - V (A! A)} = 0.
232] AXES OF A SYSTEM OF FORCES. 403 Hence (considering only real groups) there are always two distinct roots to this equation. But, if D1 and D2 be both self-normal, this equation is an identity. For this exceptional case, cf. 235 following. If the group be parabolic. Let S be any system, and replace S2 by the single director force D. Then (DS^) = 0, and the equation for Xj : X2 becomes Thus the central pair of the group are D and {D D)S1 (S \D) D. (5) To find* the locus of the axes of the systems of a dual group. Let the four centres be e, elf e2, e3, forming a complete normal system of unit points; also let eel and \eel be the axes of one central system, and ee2 and I ee2 of the other. Let S1 = ee1-\- 1\ee1, S2 ee2-\-e2\ee2y denote this central pair of systems of the group; and let e2 and e2 be called the principal parameters of the dual group. Any other system S' of the group can be written = e (X + \e2) -f | e (X^ ) Consider the system 8" = (e + fes) Oi^i + A^a) + e \(e It is a system of which a central axis intersects the line ees at right angles [cf. 223 (3)] in the point e 4- 3; also its parameter is e. But we may assume e3e1 = \ee2, 2e3 = \eelt \e^ex = ee2 \e2ea = ee1. Hence 8" = e {(^ - /t2fe) ex + |e {(^6 - /^f But in general we can make S" and S' identical by putting ^i-^2f = ^i.................................(1), fJL2 + fM^=X2.................................(2), ^i-rf=eAi ..............................(3), ji2 + fiit; = 62x2 ..............................(4). Hence, by elimination, we find 0.........(5). This is a quadratic to find e; the two roots are reciprocals, namely -, corresponding to the two axes of any system. Again, let any point x on either axis of the system $' be Cf. Homersham Cox, loc. cit. 26 2
404 EXTENSIVE MANIFOLDS AND ELLIPTIC GEOMETRY. [CHAP. III. Then, assuming for example that x is on the axis (e -+ ^e3) (fju^i + A^); D)7 comparing with the original form of S", Also by elimination between equations (1), (2), (3), (4), (6) ( . - o ftft (p+e.1)=(i - w ( 2+ ) f f,...............(7). This surface is the analogue in Elliptic space of the cylindroid. It is the locus of all the axes of systems of the dual group. All the central axes intersect at right angles the lines ee3 and \eez which are called the axes of the dual group. (6) Equation (5) of the previous subsection can also be found thus. Assume that ab is the axis of the system $'. Thus 8' = e (Xjd + X^) + | e (X^^ 4- X2^a) = a b + e | ab. Then (8'8f) = 2 ( lV + e2V) (ee1 \ee,) = 2e (ab \ab); and (S' 18') = {(1 + !2) V + (1 + *22) V} (ee, \ ee,) = (1 + e2) (at | a6). Thus finally 2e 2 ( lV + 6aV) This is equation (5) of the previous subsection. 233. Non-Axal Systems of Forces. (1) If a system of forces, S, be such that (88) = (S \8), then [cf. 174 (8) and 175 (13)] 8 has not a pair of axes [cf. 232 (1)]; provided that S be not self-supplementary [cf. 235, following], in which case it has an infinite number of pairs of axes. Such systems may be called non-axal. It will now be proved that all non-axal systems are imaginary. (2) No real hyperbolic dual group can contain a real non-axal system. For let F and Ff be the real director forces of this group, and let the non- axal system be \F + \'F\ Then by subsection (1) X2 (F\F) + 2XX' {(F|F') + (FF')} + X'2(F' |F') = 0. Hence from 231 (2), if 8 and S7 are the lengths of the two common perpendiculars to F and F', this equation becomes X2 (F\F) + 2XV {(F\F) (F' \F')f cos S-^ + V2 (Ff \F') = 0. But the roots of this equation are necessarily imaginary. Hence the four non-axal systems, which belong to any real hyperbolic group, are necessarily imaginary.
233, 234] PARALLEL LINES. 405 (3) But any real system must belong to some real hyperbolic groups. For [cf. 162 (2)] the conjugate with respect to the system of any real line, not a null line, is a real line. Now the dual group with these two lines as director lines is a real hyperbolic group, and contains the real system. (4) It therefore follows from (2) and (3) that all non-axal systems are imaginary. Hence any real system S, for which (SS) = (S\8)y is self-supplementary. 234. Parallel Lines. (1) An interesting case arises [cf. 231] with regard to lines with a special relation discovered by Clifford*, and called by him the parallelism of lines [cf. 229]. It is to be noted that the parallel lines of Hyperbolic Space [cf. Ch. iv. of this Book] do not exist (as real lines) in Elliptic Space, and conversely these parallel lines of Elliptic Space do not exist in Hyperbolic Space. In general only two lines intersect the four lines F, F', \F, \F. But if these four lines are generators of a quadric, then an infinite number of lines namely, the generators of the opposite system intersect them. The two lines F and F' have then the peculiarity that an infinite number of common perpendiculars can be drawn. F and Ff will then be proved to be mutually parallel according to the definitions of 229 (3) and (4). (2) Since the four lines are generators of the same quadric a relation [cf. 175 (4) and (5)] of the formf, \F + fi \F+ X'F' + / jjP' = 0 must exist. Taking its supplement, it must be identical with \ \F+ jiF+ V |F + fi F' = 0. TT X fM \ /JL Hence - = !=- = =-,. fJU \ fJL \ Accordingly X = + p, \' = jjl. Firstly, let X = ft, V = ///. The condition becomes \(F+\F) + \' (F + \F)=0. Let the relation of F and F' be called ' right parallelism.' Secondly, let X = - fju, \' = - yl. The condition becomes \(F-\F) + \'(F'-\F')=0. Let the relation of F and F' be called ' left parallelism.' * Cf. loc. cit., p. 370. t This form of the relation between parallel lines was first given by Buchheim, Proc. London Math. Society, loc. cit.
406 EXTENSIVE MANIFOLDS AND ELLIPTIC GEOMETRY. [CHAP. III. (3) Consider the equation Multiply it successively by F and F', then X (F \F) + V {(FFf) + (F \F')} = 0, and X {(FFf) + (F \ F')) + X' (F' Ff) = 0. Hence by eliminating X: X', (F F){F'\F)={{FF') + (F F')\\ Therefore - - ** - ~ +____^^ - + 1 ineretore ^F\F) (F \Fy + V{(^|F) OF' \F')} ~ " Similarly from the equation \(F-\F) + \'(F'-\F') = 0, we deduce (F\F)_ Wl*")} *Ji(F\F)(F' F But it has been proved in 231, using its notation, that (F \F) Si 89 (FF) . Si . o2 ^ = cos - cos - , ,TrpTP\/iP'\~wi = sm " sm Hence for right-parallelism, assuming that and are both acute angles but not necessarily both positive (with the usual conventions as to signs of lengths), cos S2 cos S2 + sin Sx sin B2 = 1 ; therefore Sj = S2. For left-parallelism, cos S2 cos S.2 sin 8X sin S2 = 1; therefore Sx = 82. But S2 and S2 taken positively are the greatest and least perpendicular distances from one line to the other. Hence the lines are parallel according to 229. (4) Thus through any point b, a right-parallel line and a left-parallel line to any line F may be obtained by the following construction. Draw ba perpendicular to F, and let the least of the two distances of b from F be 8, which is ba. Find the polar line of ab, which must intersect F at right-angles in some point c. On this line take d and df on opposite sides of c at distance S from it. Then bd and bdr are respectively the right and left-parallel to F through b. It is to be noted that \F is both a right and a left-parallel to F; and that a line parallel to F is parallel to \F. (5) Since two parallel lines are generators of the same quadric [cf. sub- section (2)], they are not coplanar.
234, 235] vector systems. 407 235. Vector Systems*. (1) Any system (S) of the type F \F is called a vector system. Such a system has an infinite number of pairs of axes, consisting of all lines parallel to F taken in pairs. For let F' be any right or left-parallel to F. Then a relation exists of the form, F \F=\(F' \F'). Accordingly S = \(F' \F'). Let a system of the form F+\Fbe called a right-vector system. If R be a right-vector system, R = \R, and (R R) = (RR): either of these equa- tions is a sufficient test, if the system is known to be real [cf. 233 (4)]. Let a system of the form F \F be called a left-vector system. If L be a left-vector system, L = |Z, and (L\L) = (LL) : either of these equations is a sufficient test, if the system is known to be real. Vector systems are the self-supplementary systems of 174 (2). (2) The sum of two right-vector systems is a right-vector system, and the sum of two left-vector systems is a left-vector system ; but the sum of a right-vector system and of a left-vector system is not a vector system f. For let R = axa2 + ja^, and R' = bxh2 4- \bj)2, be two right-vector systems. Then R + R' = {axa2 + A) + I ( i 2 + A). Now let axa.2 + bj)2 = ca + e | dc2. Then R + R' = (1 + e) (ca + i ca). Accordingly R + Rf is a right-vector system. A similar proof shows that the sum of two left-vector systems is a left- vector system. It is also obvious that another statement of the same proposition is that the dual group denned by two vector systems of the same name (right or left) contains only vector systems of that name. (3) But if R is a right-vector system and L is a left-vector system, then R + L is not a vector system. For if it were, R + L = \(R + L) = R - L. Hence R = L. But a system cannot be both a right and a left vector system ; since for such a system, |#=$ = S, which is impossible. Any system J S can be written in the form R + L. For and f ( + | S) is a right-vector system, and \ (S - \S) is a left-vector system. This reduction is unique. For if S = R + L = R' + L\ then R-R' = L'-L. Hence a right-vector system would be equal to a left-vector system, which is impossible. * This use of the word * vector' seems to me to be very unfortunate. But an analogous use is too well established in connection with the kinematics of Elliptic space to be altered now. The theory of systems of forces is very analogous, as will be proved later, to the theory of motors and vectors investigated by Clifford ; cf. loc. cit., p. 370. + Cf. Sir R. S. Ball, "On the Theory of Content," Royal Irish Academy, Transactions, 1889. + Cf. Clifford, loc. cit., p. 370.
408 EXTENSIVE MANIFOLDS AND ELLIPTIC GEOMETRY. [CHAP. III. (4) Any right-vector system and any left-vector system are reciprocal*. For let R = axa2 + \aYa2, and L = bj)2 \bj)2, then (RL) = 0. 236. Vector Systems and Parallel Lines. (1) Let ex and e2 be two unit quadrantal points : then the vector systems e$2 \exe2 are called unit vector systems. If R = exe2 + \exe2y then (RR) = 2 (e |exe2) = 2 (e, \e,) (e2 \e2) = 2 = (R\R). Also if L = exe2 \exe2y then (LL) = 2 = (L \L). (2) A vector system (* possesses an infinite number of axes, it being possible to draw an axis of the system through any point; and this set of axes forms a set of parallels, right or left according as the vector system is right or left. For let x be any other unit point and let p be another unit point on the right parallel to exe2 and quadrantal to x. Then by 234 (2), R= e1e2+\e1e2 = X(xp+\xp). Hence xp and \xp are also axes of R. Also (R \R) = 2 = 2\2. Hence \= + l. Thus a unit vector system, when expressed in terms of one pair of axes, is a unit vector system when expressed in terms of any other pair of axes. (3) A simple expression! for a line drawn through a given point right or left-parallel to a given line can be found. For let aYa2 be the given line and x the given point. Consider the right-vector system R = a1a2 + \a1a2. Let xp be the required right-parallel to aYa2 drawn through x. Then R = \(xp + \xp). Hence xR = Xx\xp = \(x \p) \x X(x\x)\p; therefore x \xR X (x \x) xp = xp. Hence x \xR is a force on the right-parallel to axa2 drawn through x, where R = a^a2 + \ajCi2. Similarly if L = a1a2 |axa2i then x\xL is the left-parallel to aYa2 drawn through x. (4) It follows that, if any two lines are each parallels of the same name (right or left) to a third line, they are parallels of that name to each other. Let all the lines parallel (of the same name) to a given line be called a parallel set of lines. (5) Any pair of conjugates of a vector system is a pair of parallel lines of an opposite denomination (right or left) to that of the system J. * Cf. Sir R. S. Ball, Transactions R.I.A. t Cf. Clifford, loc. cit., p. 370. X Not previously published, as far as I am aware.
236, 237] vector systems and parallel lines. 409 For let R = \F+fiF'. Then R = \F+ pF' = \R = \\F + fi\F'. Hence X (F- \F) + p (Ff - \F') = 0. Thus by 234 (2) F and F' are left-parallels, while R is a right-vector system. It is to be noted that by 234 (4) any pair of axes of a vector system, since they are reciprocally polar lines, are both right and left-parallels. (6) Any* right-parallel set of lines and any left-parallel set of lines have one and only one pair of reciprocally polar lines in common [cf. 234 (4)]. For let R and L be the associated vector systems of the two sets of parallels. Then they are necessarily reciprocal; also they have one and only one pair of common conjugates. These common conjugates are the lines F and F\ where F = R J(LL) + L V{- (RR)}; F' = R *J(LL) -L /{- (RE)}. Hence F'=\F. Also R = (F+\F), L = (F-\F). Thus F and F belong to the right-parallel set of R and to the left- parallel set of L. (7) The common conjugates of two vector systems of the same denomi- nation are a pair of imaginary generating lines of the absolute. This follows from 235 (2). 237. Further Properties of Parallel Linesf. (1) If any straight line meet two parallel straight lines, it makes each exterior angle equal to the interior and opposite angle, or in other words the two interior angles equal to two right angles. For let xp and yq be two parallel lines (say right-parallel); and let xy be any line meeting them. Then posy, and qyx are the two interior angles. Fig. 7. Let x, y, p, q be all unit points, and let (x \p) = 0 = (y q). Then, by 236 (2), it may be assumed that ocp+\xp = yq + \yq..............................(1). * Not previously published, as far as I am aware. t Cf. Clifford, toe. cit.y p. 370.
410 EXTENSIVE MANIFOLDS AND ELLIPTIC GEOMETRY. [CHAP. III. ^^ (xy xp) (xy \xp) cos yxp = v B v ---------- v ^ f F/ Also Similarly cos Qq =____(^IM___= _ (^fog) y J{(xy \xy) (yq yq)} x/{xy\xy}' But from equation (1), multiplying by xy, we find (xy xp) = (xy yq). Hence cos yxp + cos xyq = 0, and yxp 4- xyq = ir. A similar proof applies to left-parallel lines. (2) Conversely, if two straight lines be such that every line intersecting both makes the two interior angles equal to two right angles, then the lines are parallel. For let any line xy cut the lines xp, yq; so that yxp = xyq . Fig. 8. Draw xl and ym perpendicular to yq and xp respectively. Then from 216, considering the triangles xyl and xym, ym . xl . xy sm sm sin : sm xyq sin = L = sm - = 7r 7 sin yxp Hence xl = y?n. Therefore the lines xp and yq are parallel. (3) Parallelograms can be proved to exist in Elliptic Space: but they are not plane figures [cf. 234 (5)]. For let ab and ac be any two lines intersecting at a. Then the right- JcT /c Fig. 9.
237] FURTHER PROPERTIES OF PARALLEL LINES. 411 parallel through b to ac is, by 236 (3), F=b b(ac+\ac). Similarly the left-parallel through c to ab is F' = c ' c (ab | ab). To prove that these lines intersect, we have to prove that {FF') 0. But it is easy to prove by multiplication and reduction that, {FF') = b {b (ac + |ac)} .c\{c(ab- \ab)} = 0. Therefore the two parallels through b and c intersect in some point d. Therefore the opposite sides of the figure abdc are parallel, one pair being right-parallels and the other pair being left-parallels. Also if the angle cab be 6, then abd = it 6, bdc = 0, dca = ir 6. Further it is easy to prove that ab = cd, and bd = ac. Thus the opposite sides are equal. Hence if ac and bd be any two parallels and ac bd, then ab and cd are parallels of opposite name (right or left) to ac and bd; and also ab = cd. (4) Let ab, ab' be one pair of parallels, and let ac, a'c be another pair of the same name as the first pair: also let ab = a'b'y ac = ac , then be and b'c are parallels of the same name, and be = b'c'. For join aa\ bb\ cc'. Then by (3) aa! and bb' are equal and parallels, of the opposite name to ab and a'V; also aa' and cc' are equal and parallels of the opposite name to ac and a'c. Hence [cf. 236 (4)] bb' and cc' are equal and parallels of the opposite name to ab and ab'. Hence be and be' are equal and parallels of the same name to ab and ab'. Fig. 10. It is further obvious that the angle cab is equal to the angle c'a'V. Hence if from any point a! two parallels of the same name are drawn to any two lines ab, ac, the two pairs of intersecting lines contain the same angle.
412 EXTENSIVE MANIFOLDS AND ELLIPTIC GEOMETRY. [CHAP. III. 238. Planes and Parallel Lines*. (1) One line, and only one line, belonging to a given parallel set of lines, lies in a given plane. For let P be the given plane and e^ a line of the given parallel set. Now if F be one of the set lying in P,\F is also one of the set and passes through the point |P; and conversely. But one and only one of the set can be drawn through |P, hence one and only one of the set lies in P. If p stand for \P, then by 236 (3) the right-parallel \F, which passes through P, is p pR, where R stands for eYe.2 + exe2. Hence Thus the single right-parallel in the plane P is the line P\PR. Similarly for left-parallels. (2) To any point p in a given plane P there corresponds one and only one point q in any other given plane Q, such that if any line through p be drawn in the plane P, the right-parallel line through q lies in the plane Q (or in other words the right-parallel in the plane Q passes through q). For draw any two lines pp', pp" in the plane P. Let their right parallels in the plane Q be qq, qq" intersecting in q. Then q is the required point. For take pp' = qq, pp" = qq". Then pp' + \ pp = qq' + qq', pp" + pp" = qq" + | qq". Any other line through p and in the plane P can be written pp' -f Xpp". But from the above equations, pp + Xpp" + | (pp' + Xpp") = qq' + Xqq" + \(qq9 + Xqq"). Hence the line qq + Xqq', which passes through q and lies in the plane Q, is the right-parallel to the line pp' + Xpp". * These properties have not been stated before, as far as I am aware.
238] PLANES AND PARALLEL LINES. 413 Similarly a unique point qx in the plane Q corresponds to the point p in the plane P with similar properties for left-parallels. (3) With the construction of the preceding proposition (where pp = qq', pip" = qq"), it follows from 237 (4) that p'p" is the right-parallel to q'q". Hence the points pf and q' in the planes P and Q correspond. Thus, given two corresponding points p and q, it is easy to find the point on one plane corre- sponding to any point on the other. For consider the point pf on P. Join pp and draw qq parallel to pp' and of the same length. Then q corresponds to p'. (4) The common perpendicular of two planes P and Q, namely \PQ, cuts the planes in two points p and q which are corresponding points both for Fig. 12. right and left-parallels. For in the plane P draw any line pin cutting the line PQ in m. Take two points I and r on PQ such that hn rmr=pq. Join qr and ql. Then [cf. 234 (4)] one of them (say qr) is a right-parallel to pm and the other ql is a left-parallel. Accordingly, knowing that p and q are corresponding points, it is possible by (3) to construct the points qx and q2 on Q corresponding to any point pf on P for right and left-parallels respectively.
CHAPTER IV. Hyperbolic Geometry. 239. Space and Anti-space. (1) In hyperbolic geometry [cf. 208] the absolute O|#)=0, is a real closed quadric. If e be any point within such a quadric, then [cf. 82 (6)] its polar plane does not cut the quadric in real points and the polar plane lies entirely without the quadric. Hence if e, eu e2, ... ev-x form a normal system, and if e lie within the quadric, then the remaining points elf e2 ... ev^ lie without it. Similarly [cf. 82 (7)] if E, Eu E2 ... EV_Y form a normal system of planes, and if E does not cut the quadric in real points, then E1} E2i ... Ev_x must all cut the quadric in real points and include points within the quadric. (2) Let that part of the complete spatial manifold of v 1 dimensions which is enclosed within the absolute be called Space [cf. 202], Let the part without the absolute be called Anti-space, or Ideal Space. Let a point within space be called spatial, a point in anti-space anti-spatial. (3) A subregion may lie completely in anti-space, as far as its real elements are concerned, but cannot lie completely in space. Let a subregion which comprises spatial elements be called spatial, and a subregion which does not comprise spatial elements anti-spatial. (4) Then a normal system of real elements e, elf ... ev-x consists of one spatial element e, and of v 1 anti-spatial elements. Let e be called the origin of this system. A normal system of planes E, Eu ... Ev^ consists of one anti-spatial plane E, and of v 1 spatial planes. If a plane P be spatial, its absolute pole \P is anti-spatial; if the plane be anti-spatial, its absolute pole is spatial. If an element p be spatial, its absolute polar \p is anti-spatial; if p be anti-spatial, \p is spatial. If any subregion Pp, of p 1 dimensions, be spatial, the subregion \PP is anti-spatial; if Pp be anti-spatial, \P? is spatial.
239, 240] INTENSITIES OF POINTS AND PLANES. 415 240. Intensities of Points and Planes. (1) Let the absolute be referred to the v normal elements e% e1} ... ev-1} of which e is spatial. Let a, ia1} iot2,...iaiv-i be the normal intensities [cf. 110] of these elements; and let iv~x A stand for iv~l aa^... _!. Then A is also real, where a, aD ... av^ are real. Also let i1"1 A (ee1... ev-i) = 1. Then |g = ^ ^2... ^_1? |g1= _g2-^ ..- V-i, e2 =------- - ^1^3 ^i/-i; and so on. Hence if x = e + f ^ + ... -f %v-xev- , Thus [cf. 82 (9)] if ^ be spatial and its co-ordinates real, (x\x) is positive. This supposition will be adhered to. (2) Any real plane is given by L = \exe2... ev-Y X^ee^... ev-Y + X^ee^ ... 0,-1 + etc.; where the ratios \ : \x: A,2: etc. are real. But if so (=2 e) be a real point with its co-ordinates real, we may suppose x to be the pole of L, and write L = \x = i*-1 A (|2 e^a... ev^ ~^2ee2... ev^ + etc. Hence \ = {"-i_f? \l=iv-i^} etc Therefore if v be even, \, Xly etc., are pure imaginaries, so that their ratios are real. A plane will be considered to be in its standard form, when expressed in the form L = iv~l\exe2... ev^ i^^ee^... ev^ + etc., where \, Xly etc. are real. Then we can write L = \x, where the coefficients of x are real. Thus a real plane is, if v be even, intensively imaginary [cf. 88 (3)]; while a real point, spatial or anti-spatial, is always intensively real. (3) If x be spatial, its intensity is unity when (x \x) = 1, and is real when (x\x) is positive. If x be anti-spatial, the intensity of x will be defined to be real when (x \x) is negative, and to be unity when (x\x) = l. Thus in both cases the intensity is real when the coefficients are real,
416 HYPERBOLIC GEOMETRY. [CHAP. IV. The intensity [cf. 211 (5)] of a plane L will be denned to be unity, when it is in the standard form \x, where x is at unit intensity. Thus for anti-spatial planes at unit intensity, \x is spatial, and For spatial planes at unit intensity (L\L)=-1. (4) Thus for a spatial point 2 e at unit intensity a2 a* '" Vi ' For a spatial plane %iv~Y \E at unit intensity, i2 V + a22 V -f ... + o^A^-i2 - *2X2 = A2. (5) But if the reference elements e, eli ... ev-x be at unit intensities, spatial and anti-spatial, then a = ax = ... = a,^ = 1. Hence [cf. subsection (1)] the point-equation of the absolute is the plane-equation of the absolute is V + V+... + X2^-X2 = 0. The intensity of a spatial point 2 e is ( 2 f^ f22 ... f^-i)*; the intensity of a spatial plane Si*""1 X# is (V + X22 + ... + X2,_i - X2)i Also iv~l (eex... ev-j) = 1; and | e = iv~xexe^... ev^, \ex= iv~Y ee2... ev^, \e2 = iy~xeexe^... ev-x, | ez ^~^eexeA... ^_i, and so on. This supposition will be adhered to, unless it is otherwise stated. 241. Distances of Points. (1) It will be seen that in the case, in which the line joining two points in anti-space does not cut the absolute in real points, the usual hyperbolic formula does not give a real distance between them. In this case it is convenient to use the Elliptic measure of distance. Thus any two points in anti-space (as well as any two points in space) are separated by a real distance, Elliptic or Hyperbolic. But a point in space cannot have a real distance of either type from a point in anti-space. (2) Firstly, let two points x and y both be spatial, and of standard sign [cf. 208 (3)] ; then xy is given by and xy, thus determined, is real. Also, since there can be no distinction between D (xy) and xy, the latter symbol will always be used for the distance.
241, 242] DISTANCES OF POINTS. 417 (3) Secondly, let x and y both be anti-spatial, but let the line xy be spatial. Then if ax and a2 be the points where the line meets the quadric, x and y lie together on the same intercept between ax and a2. Also (x\x) and (y \y) are of the same negative sign. Hence the hyperbolic functions give a real distance. Thus xy is determined as a real quantity by the formula where the ambiguity ^f sign must be so determined that the right-hand mde is positive. (4) Thirdly, let x be spatial and y be anti-spatial. Then both the formulae cosh xy (x | y) cos 7 /{(x\x)(y\y)} must make xy imaginary, since (x \x) and (y \y) are of opposite signs. (5) Fourthly, let x and y both be anti-spatial, and let the line xy be anti-spatial. Then the two elements a^ and a2 in which xy meets the absolute are imaginary. Hence the elliptic law applies. Let the distance between x and y, determined by this law, be called the angular distance between the points, and denoted by Z xy. Then cos Z xy = .,. ^'^ _ ; where the conventions of 206 apply: so that, if x stand for x, (x\y) cos Zxy= , ; ,*; . = - cos Z xy. Hence Z x'y + /_xy ir. (6) Thus, to conclude, if the line xy be spatial, and x and y be both spatial or both anti-spatial, then xy is real. If the line xy be anti-spatial, then z xy is real. If x be spatial and y anti-spatial, both xy and Z xy are imaginary. 242. Distances of Planes. (1) Consider the formulae for the separa- tion between two planes P and Q. Firstly, let both planes be spatial, and let the subplane PQ be spatial. Then \P and \Q are both anti-spatial, and P \Q is anti-spatial. Hence Z \P \Q is real, and is given by COSZipiq=____(mo)(im 1 |y y/{(} Hence the separation between P and Q is real, when determined by the elliptic formula. Let it be called the angle between P and Q, and denoted by zPQ. w. 27
418 HYPERBOLIC GEOMETRY. [CHAP. IV. Then cos Z PQ = It is to be noticed that there are two angles Z PQ and ir Z PQ, corre- sponding to the ambiguity of sign on the right-hand side. (2) Secondly, let both planes be spatial, and let the subplane PQ be anti-spatial. Then \P and \Q are both anti-spatial, and P\Q is spatial. Hence \P \Q is real, and is determined by J{(\P\\P){\Q\\Q)}' Hence the separation between P and Q is to be measured by the hyper- bolic formula, and will be called the distance between the planes, and denoted by PQ. _ Then where as usual the terms P and Q are so chosen that (P | Q) is positive. (3) Thirdly, let P be spatial and Q be anti-spatial. Then (P|P) and (Q \Q) are of opposite signs. Hence both Z PQ and PQ are imaginary. (4) Fourthly, let the planes P and Q both be anti-spatial. Then \P and \Q and \P \Q are spatial, and \P \Q is real. Hence cosh - IPI' * ____ 7 VKIPilPXIQI^Q)} (P\Q) ' /{(? \P)(Q\Q)V Hence PQ is real and Z PQ is imaginary. Also cosh~7 =7T(P\P)(Q\Q)}] where the terms P and Q are so chosen that (P | Q) is positive. 243. Spatial and Anti-spatial Lines. (1) If the elliptic measure for separation holds, then [cf. 204 and 211] and if the hyperbolic measure holds, then [cf. 208 and 211] smh ^ = / -(xy\xy) y V \(x\x)(y\y)}
243, 244] SPATIAL AND ANTI-SPATIAL LINES. 419 (2) Thus if xy be spatial, (xy xy) is negative. For if x and y be either both spatial or both anti-spatial, the proposition follows from the expression for sioh . But if x be spatial and y anti-spatial, then (xy \xy) = (x\x) (y \y)- (x \y)\ Now (x I x) is positive and (y \ y) negative; hence again the proposition follows. But if xy be anti-spatial, then, from the expression for sin Z xy, (wy\xy) is positive. (3) Furthermore if x be anti-spatial and y be any point on the cone, (xy \xy) = 0, which envelopes the quadric, then xy = 0 and Zxy = 0. Hence any two points on a tangent line to the quadric are at zero distance from each other. (4) Again, by similar reasoning, if the intersection of two spatial planes P and Q be spatial, (PQ \PQ) is positive. If the intersection of two spatial planes be anti-spatial, (PQ \PQ) is negative. If P be spatial and Q anti- spatial, (PQ \PQ) is negative. Hence if PQ be spatial, (PQ \PQ) is positive : if PQ be anti-spatial, (PQIPQ) is negative. 244. Distances of Subregions. (1) If two subregions Pp and Qp, each of p 1 dimensions, are contained in the same subregion (L) of p dimensions, then [cf. 211 (6)] a single measure of the separation of Pp and Qp can be assigned. (2) Let the section of the absolute by L be real; and firstly let the intersection of Pp and Qp be spatial Then Z PPQP is real, and ( PpIG,) Secondly, let the intersection of Pp and Qp be anti-spatial, but Pp and be both spatial. Then PPQP is real, and ___ Thirdly, let Pp be spatial and Qp be anti-spatial. Then PPQP and Z PPQP are both imaginary. Fourthly, let Pp and Qp be both anti-spatial. Then PPQP is real, and cosh 7 (3) Let the section of the absolute by L be imaginary. Then Pp and Qp are anti-spatial, and we have a fifth case when Z PPQP is real, and given by the formula of the first case. 27 2
420 HYPERBOLIC GEOMETRY. [CHAP. IV. 245. Geometrical Signification. Geometrical meanings can be assigned to the co-ordinates of any spatial point x, at unit intensity and of standard sign, referred to a normal system e, ely ... ev-. at unit in- tensities, of which e is the spatial origin. Let x = %e + % + ... + ,_ie,-i, where (x \x) = 1 = f2 - \ - ... - ,_!2. Then cosh = (e \x) = . Let the angles, xeel = Xly xee2 = X2, etc. (eeY; ex) Then cos Xx = \J{(ex\ex)(ee1 ee-)} . , ex' smh 7 Hence fl = sinh cos \, 2 = sinh cos A2 etc. 7 7 Also the angles X1? \2, ... X,,^ are connected by, 2cos2X = l. Similar geometrical interpretations hold for Elliptic Geometry. 246. Poles and Polars. (1) It will be noticed that the only case, when there is no real measure of separation between two points x and y, is when x is spatial and y is anti-spatial. In this case the point of intersec- tion of xy and the polar of y is spatial. For this point is xy\y = (x y)y- (y ,y) x = y\ say. Then by simple multiplication we find (3/ \y) = (y \yf (x\x)- O \yf (y y)- But (x x) is positive and (y {y) is negative. Hence (y'' \y') is positive, and therefore y is spatial. Also the term y' is of standard sign. For [cf. 208 (3)] x is by hypothesis of standard sign, and (x \y') = (x \yy - (x \ x) (y \y) = - (xy \xy). But xy is spatial; hence, by 243 (2), (xy\xy) is negative, and (x y') is positive. Similarly the point of intersection of xy and the polar of x is x' = yx and x is anti-spatial, since \x is anti-spatial. (2) Now (y \y') = (y \y) {(x \x) (y \y) - (x \yf) = (y \y) (xy\xy), and (y \x) = (x\yf-(x x) (y \y) = - (xy \xy). x ; Hence co.h ^ - 7 ""
245, 246] poles and polars. 421 Also since {y\y)y as well as {xy\xy\ is negative, y'x is real as given by this formula. Similarly sinh V = V{cosh* V - ll *J{-{x\x){y\y)\' Also since x and y are both anti-spatial, and x'y is spatial, then x'y is real. And by a similar proof cosh = J^^VL cosh ^. 7 V 0 !^)(2/|y) 7 Hence x'y = y'#. (3) Let x and y be both anti-spatial; and let xy be spatial. Then xy is real. Also x' = yx x} and y' xy\y are both spatial points. For y W)=(y \yf (x\x)-(x \y)2 (y y)=(y \y)(xy \xy)- Now (y \y) and (xy\xy) are, by hypothesis, both negative. Hence (y'\y) is positive, and y is a spatial point. Similarly x is a spatial point. Also (x \y') = (x\yf -(x x)(y\y)(x\y) = - (x \y) {xy \xy). Hence if the terms x and y be so chosen as to sign that (x \y) is positive, (x \y') is also positive. Now, since x and y1 are both spatial, x'y' is real; and 7 VKI Now, since {xy \ xy) is negative, T, . ^V' (^ I y) i Hence cosh = ; 'y = cosh 7 V{(^]f)(2/Iy)} 7 Therefore #Y = ^. (4) Exactly in the same way let the plane P be spatial and Q anti- spatial, then \P is anti-spatial and |Q is spatial. Let the plane through PQ and \P be called P' and that through PQ and \Q be called Q'. Then Q' is obviously spatial, and P' can be proved to be anti-spatial. Then P and Q' are two spatial planes with an anti-spatial intersection. Hence cosh^= P LQ1_ But
422 HYPERBOLIC GEOMETRY. [CHAP. IV. Therefore (P |Q') = (P \Qf -(Q'Q)(P\P) = - (PQ \PQ), and (Q'\Q')=(Q\Q)(PQ\PQ). ,PQ I (PQ\PQ) UPV Hence cosh = ^(PTPR^ Q) = cosh -*. PQ (P\0) P'Q Similarly smh - = ^^ ,ff(QjQ)} = sinh -* . Hence PQf = PrQ. 247. Points on the Absolute. (1) A point m on the absolute is at an infinite distance from any other point. For (u \u) = 0, and hence . ux (x I ti) cosh = -7,7 | w | x) = oo . (2) To find a point u in which any spatial line xy cuts the absolute, put u =x + Xy. Hence \2(y j^/) + 2\0 !y)+ (^ ja?) = 0. Now let p stand for the distance xy, and let x and y be spatial points at unit intensity and of standard sign. Hence X2 + 2\ cosh + 1 = (X + e*) (X -f 6 *) = 0. Accordingly the two points, in which the line xy cuts the absolute, are _p p ^ ~ e y V an(i oo ey y. (3) In the same way if the line xy be spatial, but x and y be both anti- spatial at unit anti-spatial intensity, and (x\y) be positive, then the points, _p ?_ in which xy cuts the absolute, are x + e y y, x -f ey y. (4) Similarly let P, Q be two spatial planes, and PQ be anti-spatial: also let P and Q be at unit spatial intensity and let p be the distance between them (hyperbolic measure). Then the planes through PQ touching the absolute are P + e y Q, and P + ey Q. Also if P and Q be both anti-spatial at unit anti-spatial intensity, the _p p^ tangent planes are P e y Q, and P ey Q. 248. Triangles. (1) Consider a triangle a c, in which the measures for the separation of the angular points are all real. Then the cases which arise are (1) a, b, c all spatial; (2) a, 6, c all anti-spatial, and 6c, ca, ab all spatial; (3) a, 6, c all anti-spatial, and be, ca, ab all anti-spatial; (4) a, 6, c all anti-spatial, and be, ca, ab two spatial and one anti-spatial; (5) a, 6, c all anti-spatial, and be, ca, ab being one spatial and two anti-spatial.
247, 248] triangles. 423 (2) Case /. a, 6, c all spatial. Let the triangle abc in this case be called a spatial triangle. Let the angle between ab and ac be a, that between ba and 6c be yff, and that between ca and cb be y. To discriminate between a and ?r a, let a be that angle which vanishes when b coincides with c ; and similarly for jS and y. Thus, b and c being of standard sign according to the usual convention, (ab \ac) cos a = =-, \/{(ab \ab) (ac \ac)}' And (a6 ; ac) = (a \ a) (6 ; c) (a 16) (a | c); i . ,ab / (ab\ab) . , ac / (ac\ac) also sinh = A / ^ ' /. , smh = A / , , ' . \ . 7 'y (a I a) (6 16) 7 V (a|a)(c|c) cosh-----cosh cosh Hence cos a = . , ab . , ac sinh smh TTi- 11 1 be , ab , ac . , ab . , ac r1 mally, cosh = cosh cosh-----sinh smh cos a. 7 7 7 7 7 (3) Also [cf. 216 (1)] sin a = ; ac)} _ \/{(a I a) (a6c ,a6c)} ~ \/{(a6|a6)(ac|ac)}' And sinh 6c _ / (6c 16c) inh = . / 7 V' Therefore sin a (c c)(abc \abc) . . 6c V -(6c 6c)(ca ca)(a6ja6) smh 7 sm /j sin 7 7 (4) It is easily proved, exactly as in Elliptic Geometry [cf. 216 (6)], that the perimeter of a spatial circle, with a spatial centre and of radius p, is 27T7 sinh -. And that the length of an arc subtending an angle a at the centre is ct7 sinh - . 7 (5) Case II. The angular points a, 6, c are anti-spatial, and the sides 6c, cay ab are spatial. Let the triangle a6c in this case be called a semi- spatial triangle.
424 HYPERBOLIC GEOMETRY. [CHAP. IV. The distances between the sides 6c, ca, ab must now be measured by the hyperbolic measure. Thus let a, jS, y be assumed to be lengths and not angles. Also adopt the conventions of Case I. Then by a similar proof , be , ab t ac . . ab . , ac , a cosh = cosh cosh------sinh smh cosh - . 7 7 7 7 7 7 sinh - sinh sinh Also ------L = . , 6c . , ca . , a sinh smh smh 7 7 7 _ I (a | a) (6 ! 6) (c | c) (a6c ; a6c) !6c)(ca |ca)(a6:a6) (6) Case III. The angular points a, 6, c all anti-spatial, and 6c, cay ab also anti-spatial. This case gives simply the ordinary formulae of Elliptic Geometry [cf. 215]. (7) Case IV. The angular points a, 6, c are anti-spatial, the two sides a6, ac are spatial, and the third side 6c is anti-spatial. The sides a6 and ac have a real measure of separation a, reckoned according to the hyperbolic formula, but the sides 6a and 6c, and the sides ca and c6 have no real measure of separation. Tjr , , a6 ,ac . , ab . , ac , a Hence cos Z 6c = cosh cosh-----smh smh cosh - . 7 7 7 7 7 (8) Case V. The angular points a, 6, c are anti-spatial, and a6, ac are anti-spatial and 6c is spatial. Then a is real, and J3 and y are imaginary; and a is measured by the elliptic formula. And 6c cosh = cos Z a6 cos Zac + sin Z a6 sin Z ac cos a. 7 There are no corresponding formulae to be obtained by cyclic interchange; since JS and y are imaginary. (9) The theory, given in 217, of points inside a triangle holds without change for Hyperbolic Geometry. 249. Properties of Angles of a Spatial Triangle. (1) Two angles of a spatial triangle cannot be obtuse. For if a and JS be both obtuse, cos a and cos JS are both negative. Hence from 248 (2) , a6 , ac , 6c , , 6c , a6 , ac cosh cosh cosh , and cosh cosh cosh . _ 7 7 7 7 7 7 But cosh is necessarily greater than unity; hence these two inequalities are inconsistent.
249, 250] PROPERTIES OF ANGLES OF A SPATIAL TRIANGLE. 425 (2) It follows from 247 (2) that, when b and c are spatial points of standard sign, all points of the form \b + /ac, where X/fi is positive, lie on the intercept between b and c; since the two points, in which xy cuts the absolute, both lie on that intercept for which \//jl is negative. Hence it maybe proved, exactly in the same way as in 219 (2) dealing with Elliptic Geometry, that if in any triangle abc jS and y be both acute, the foot of the perpendicular from a on to be falls within the intercept be. (3) The sum of the angles of any spatial triangle is less than two right- angles. Firstly, let the angle y be a right-angle. Then as in Elliptic Geometry, [cf. 219 (4)]. _ _ , 1 be ~ ca cosh ^--------- tan i(a + /B) =----- 7 cot \y l6 , 1 be ~ ca cosh H--------- , 1 be + ca cosh---------- 2 7 XT i 1 be ~ ca . lbc + ca JNow cosh---------- cosh^---------. 2 7 2 7 Hence a + /S ^ . Hence a + jS + y 7r. Secondly, the theorem can be extended to any triangle by the reasoning of 219 (4). 250. Stereometrical Triangles. (1) It is obvious by the theory of duality that a complete set of formulae for stereometrical triangles [cf. 222 (1)] can be set down, and that these can be ranged under eight cases just as in the case of ordinary triangles. It will be sufficient to obtain the results for the two most important cases. (2) Firstly, let the planes A, B, G be spatial, and let the subplanes BC, CA, AB be also spatial. For shortness put BC = Al9 CA = Blt AC= CY. Let zBC=a, ZCA=J , zAB = y. Also ZB , Zfti, zi^are real. Then if \A = a, \B=b, \C = c, the triangle abc is anti-spatial, and be, ca, ab are anti-spatial. Hence from 248 (6) cos Zbc = cos Z ab cos Z ac + sin Z ab sin Z ac cos Z (ab) (ac).
426 HYPERBOLIC GEOMETRY. [CHAP. IV. But Zbc = ZBG= a, Zca= zCA=j8, Zab = ZAB = y. Also Z(ab)(ac) = zB1C1. Hence, cos a = cos j8 cos y + sin JS sin y cos Z B^. (3) When the complete region is of two dimensions, this does not agree with the ordinary formula in Euclidean space for spherical trigonometry; and, as in the analogous case of Elliptic Geometry, the discrepancy is removed by replacing the angles by their supplements. When the complete region is of three or more dimensions, we deduce, as in the case of Elliptic Geometry, that the ' Spherical Trigonometry' of Hyperbolic Space is the same as that of ordinary Euclidean Space. This theorem is due to J. Bolyai, as far as space of three dimensions is concerned : it is here extended to planes of any number of dimensions. (4) Secondly, let the planes A, B, C be spatial, but let the subplanes Aly Bly Cx be anti-spatial. Then the triangle abc has its three angular points anti-spatial, but its three sides be, ca, ab spatial. Hence from 248(5) _ __ ______ , be , ab , ac . , ab . ac , (ab) (ac) cosh = cosh cosh------smn sin cosh ----- 7 7 7 7 7 7 Hence cosh - = cosh cosh sinh sinh cosh - ; 7 7 7 7 7 7 with two similar formulae. 251. Perpendiculars. (1) The theory of normal points and of per- pendiculars in Hyperbolic Geometry is much the same as in Elliptic Geometry (cf. 223). The proofs of corresponding propositions will be omitted. Any two mutually normal points satisfy, (x \y) = 0. If xy be spatial, then one point must be spatial and the other anti-spatial [cf. 239 (4)]. In this case no real measure of distance exists between x and y. If xy be anti- spatial, then the elliptic measure holds, and Zxy = \ir. The condition that two lines ab, ac should be at right-angles (or perpen- dicular) is (ab \ac) = 0. If a be spatial, the measure of distance between the lines is elliptic, and the angle between them is a right-angle. If a be anti-spatial, and both lines be anti-spatial, the measure of distance is elliptic and the angle is a right-angle. But if a be anti-spatial, ab be spatial, and ac be anti-spatial, there is no real measure of distance between the lines. It is impossible for two lines to be at right-angles when a is anti-spatial, and ab, ac both spatial. (2) If a line ab cut any region Lp, of p 1 dimensions in the point a, and if p 1 independent lines drawn from a in Lp are perpendicular to ab, then all lines drawn from a in Lp are perpendicular to ab. The line ab is then said to be perpendicular to the region Lp.
251, 252] PERPENDICULARS. 427 (3) Any line perpendicular to the region Lp intersects the supplementary (or complete normal) region \LP; and conversely, any line intersecting both Lp and Lp is perpendicular to both. (4) If Pp and Pa be two regions normal to each other, and if a be any point in Pp? then any line drawn through a in the region Pp is perpendicular to the region aPa, (5) Let two planes L and M intersect in the subplane LM, and let ax be any point in LM. From ax draw aYl in the plane L perpendicular to the subplane LM, and draw a^n in the plane M perpendicular to LM, then the separation between L and M is equal to that between ajl, and ajn. For as in the Elliptic Geometry, (L\M) (ajajn) V{(L\L)(M\M)} ~ ^{(aj laj) (ajn \aimj\' Hence if /.LM be real, then / LM = / (aj) (a^n) : and if LM be real, then LM (aj (6) Any line, perpendicular to any plane Lt also passes through its absolute pole. If any plane M include one perpendicular to L, then from any point of the subplane LM a perpendicular to L can be drawn lying in M. Also if | L lies in M, then | M lies in L ; hence this property is reciprocal. If two planes are at right-angles, their poles are mutually normal. (7) Also if two planes L and Lr be each cut perpendicularly by a third plane My then the measure of separation between L and L' is the same as that between LM and LM. 252. The Feet of Perpendiculars. (1) Let p be the foot of the perpendicular xp, drawn from any spatial point x to a spatial plane L. Then p is spatial. For p = x\L .L] also put l=\L. Then it can easily be proved that (p\p) = (L\L)(xl\xl). Now since xl is spatial, (xl\xl) is negative [cf. 243 (2)]; and (L\L) is negative, since the plane L is spatial. Hence (p \p) is positive, and p is spatial. This can be extended to any spatial subregion Pp by noticing that Pp has the property of a plane with respect to the region xPp. (2) If the plane L and the point x be both anti-spatial, then the perpendicular from x to is spatial, since it passes through the spatial point \L. (3) The line joining the poles \L and \L' of two planes L and Lr is evidently the only common perpendicular to the two planes L and L\ It is anti-spatial, if LL be spatial: it is spatial, if LU be anti-spatial.
428 HYPERBOLIC GEOMETRY. [CHAP. IV. For let l=\L and V = \L. Then if LL be spatial, by 243 (4) {LL \LL) is positive. But (IV\U') = {LL LL). Hence IV is anti-spatial by 243 (2). If LL be anti-spatial, by 243 (4) {LL' \LL) is negative, and hence {IV \IV) is negative. Therefore IV is spatial. (4) If \LL = (\L\L) be spatial and L be spatial, then \LL intersects L in a spatial point. For let d be this point. Then d = L LL =( \L) \L - (L \L) \L. Hence (d \ d) = {L\ L) {LL \ LL). But {L \L) is negative, and {LL LL) is positive. Therefore d is spatial. The theorem also follows immediately from subsections (1) and (3). 253. Distance between Planes. (1) To prove that the distance (hyperbolic or elliptic) between two planes is equal to the distance between the feet of their common perpendicular line. For let L and L be the two planes ; and let d = L LL, d' = L \LL. Then d = {L\L) \L - {L Z) \L\ d' = {L \L) \L-{L \L) \L. Hence {d d') = {L jLJ ~{L\L) {L \L) {L \L) = -{L\L){LL\LL). Hence if LL be anti-spatial and {L \L) be positive, [cf. 252 (3) and (4) and 242 (2)] -{L\L){LL\LL) -------- ^dd cosh = C0Sft _ Hence dd' is the distance which has been denned as the measure of separation between the planes. (2) Secondly if LL be spatial, d and d' are anti-spatial and on an anti- spatial line [cf. 252 (3) and (4) and 242 (1)]. Then cos Z dd' = /UT\ T[ \, . = cos Z LL. y\\L \L) {L \L )\ Then Z dd' is the angle which has been denned as the measure of separa- tion between the planes. (3) Also [cf. 211 (6)], when the distance formula can be applied to two subregions Pp and Qp, each of p 1 dimensions, these subregions are both contained in the same region of p dimensions; and therefore they have the properties of planes. Hence they possess a single common perpendicular; and, when Pp and Qp are spatial and their common subregion anti-spatial, the length of this (spatial) perpendicular is the measure of separation between
253, 254] DISTANCE BETWEEN PLANES. 429 the subregions ; also when the common subregion is spatial, the angular length of this (anti-spatial) perpendicular is the angle between the sub- regions. 254. Shortest Distances. (1) The least distance from a spatial point x to a spatial plane L is the perpendicular distance xp, where p is the foot of the perpendicular. For let q be any other spatial point on the plane L. Then since [cf. 251 (2)] the angle between px and pq is a right-angle, cosh = cosh cosh . 7 7 7 Hence xq xp. This length of the perpendicular will be called the distance of x from the plane L. (2) To find this distance xp, write I for L, then p = xl\l = (x l)l-{l\l)x. Hence (p\p) = (L\L)(xl\xl), and {xp xp) = (x \lf (xl \xl) = (xL)2 (xl \xl). Thus sinh This formula gives the distance from a spatial point x to a spatial plane Z. (3) The greatest hyperbolic distance from an anti-spatial point x to an anti-spatial plane L is the perpendicular distance xp, where p is the foot of the perpendicular from x to L. Let q be any other point on L such that xq is spatial: also xp is spatial from 252 (2). But pq is anti-spatial. Hence cosh = cos pq cosh - . 7 _ _ 7 Hence xq xp. (4) It follows from (1) of this article and from 253 that the length of the common perpendicular is the least distance between the spatial points of spatial planes with an anti-spatial intersection and that this least distance is what has been defined as the distance between the planes. The same holds for any two subregions of the same dimensions with a single measure of distance between them. (5) A formula, analogous to the formula for Elliptic Geometry, in 226 (1) and (2), can be found for the perpendicular distance of any point a from any subregion Pp, of p - 1 dimensions. 7 " V 0* k) (P |P) ~ V - (*!*) ( !i) " V t- (x x) (L \L)}"
430 HYPERBOLIC GEOMETRY. [CHAP. IV. Case I. If a and Pp be both spatial, then (aPp\aPp) sinh I-(aPP\ V (a\a)(P Case II. If a and Pp be both anti-spatial, and aPp be spatial sinh - = Case III. If a and Pp be both anti-spatial, and aPp be also anti-spatial aPp) . ~ / (aPp a sin 8 = A / . s^d V (a a)(P (6) To prove these formulae first consider the distance of a from the straight line F. Let b and c be two spatial points on F and let ap be the perpendicular from a on F [fig. 1]. Then F=bc=pb. Also by hypothesis (pa \be) = 0 = (pa pb), since jd6 = be. Hence by formula (i) of 216 (1) (p \p) (pab \pab) = (pb \pb) (pa \pa) - (pa \pb)2 = (pb pb) (pa \pa). (pa |pa) _ (pab \pab) _ (abc |abc) (p \p)(a\a)~ (pb\pb)(a\a) ~ (be \bc) (a\a)' Now, if the hyperbolic formula hold, Hence and if the elliptic formula hold * / (pa\pa) j (aF\aF) SU1 V(p\p)(a\a)-\/ (F\F){a\a,y In Case II the hyperbolic formula holds; since F is anti-spatial, and therefore the point, in which \F meets the two dimensional region aF, must be spatial; re- membering that the section of the absolute by aF is real. But ap passes through this point. These formulae may be extended to the general case of subregions of p 1 dimensions by exactly the same reasoning as that used for the analogous theorem of Elliptic Geometry in 226 (2). 255. Shortest Distances between Subregions. (1) Let Pp and Qo be two non-intersecting sub-regions of the pth and xth orders respectively, so that p + cr v. A series of propositions concerning lines of maximum and minimum distance between Pp and Qa can be proved analogous to those for Elliptic Geometry in 227. Let it be assumed throughout this article that
255] SHORTEST DISTANCES BETWEEN SUBREGIONS. 431 p a. Four different cases arise according as Pp and Qa are respectively spatial or anti-spatial. We will only consider here the single case in which Pp and Q9 are both spatial. (2) It can be proved, as in 227 (1), that a line (spatial or anti-spatial) of maximum or minimum distance (hyperbolic or angular) between them is perpendicular to both. (3) The polar regions \PP and \Qa are both anti-spatial, and of the (v p)th and (y cr)th orders respectively. In general the region | Qv intersects Pp in an anti-spatial subregion of the (p r)th order at least. The regions \PP and Qa do not in general intersect. (4) Let qlf q2, ... q9 be a independent points in Qa. Then any point x in Qa can be written 2f#. Also write X2 If x be spatial, and xp be the perpendicular to P from w, then [cf. 252 (1)] p is spatial. Hence by 254 (5), Case I, sinh2 = X2. If xp be anti- 7 spatial, so that the elliptic measure of distance holds, sin2 Z xp = X2. Hence in all cases of lines of maximum or minimum length between Pp and Q9, a conditions of the type, = 0, hold; where , f2, ... %a are successively put for Thus by the same reasoning as in 227 (3) a determinantal equation of the 6th degree is found for X2 of the form *22 ^2 ( ?2122), ", a-2 r X2 {q2 \q2), with similar equations defining the other a's. (6) Hence there are in general a common perpendiculars to the two subregions Pp and Qa,{p r)- If Pp and Qa had been interchanged in the above reasoning, so that x is a point in Pp, and
432 HYPERBOLIC GEOMETRY. [CHAP. IV. then an equation of the pth degree for A,2 would have been found. But by the formula (i) of 226 (4) (xQ*\xQ.) + (*\Q*\-x\Q,) = (x\x)(Q*\Q*) ...............(i). Now by subsection (3) above x may be supposed to lie in the region Pp \Q r which is a subregion of Pp. Thus for all points x in this subregion, x | Qv = 0; and equation (i) becomes Now differentiate , f2, with respect to any variable 0, and put x for 2 t| p. Then equation (i) becomes, after differentiation, But {x'\Qa\.x\Qa) = 0. Hence (x'Q ,\xQ r) = (x'\x)(Qa\Qir).....................(iii); which holds for any point x in the subregion P^Q,,, which has made any infinitesimal variation to the position x + x'B0 in the region PP. Thus differentiating X2, and using equations (ii) and (iii), dX? {x'Qa \ bQ.) (x tx) - (x' ,x) (xQ, \xQ,) dd {x\x?(Q.\q.) Thus the infinite number of lines drawn from any point in Pp \ Qa to Qa, which are not necessarily perpendicular to Pp, fulfil the conditions from which the equation of the pth degree is derived. The analysis of this subsection could have been used for the corresponding subsection in Elliptic Geometry [cf. 227 (4)]. (7) It follows by the method of 227 (5) that the a- feet in Qa of these a perpendiculars are the one common set of a polar reciprocal points with respect to the sections by Qa of the two quadrics (x \x) = 0, and (xPp \xPp) = 0. (8) It follows by the method of 227 (5) that the a common perpen- diculars all intersect | Pp; and that the r points of intersection with j Pp are mutually normal. (9) It follows by the method of 227 (7) that the a lines of the perpendiculars are mutually normal; and that therefore they intersect Pp in t mutually normal points, which define a subregion Pa of the crth order. (10) Also these theorems can be proved by the method of 227 (8). (11) One, and only one, of the a perpendiculars is spatial. For consider a spatial point p in Pp and a spatial point q in Qa. Then the distance pq is real and finite; it varies continuously as p and q vary their positions continuously on Pp and Q^; and it approaches infinity as a limit, when p or q or both approach the absolute.
256] SHORTEST DISTANCES BETWEEN SUBREGIONS. 433 Hence there must be at least one position of pq, for which pq has a minimum value. Thus there is at least one spatial common perpendicular to Pp and Qa. Let F be a force on the line of this perpendicular: then by (9) the remaining a 1 perpendiculars must lie in | F. Now | F is anti-spatial [cf. 239 (4)]. Hence the other (tr 1) perpendiculars are anti-spatial lines, and their lengths must be measured in angular measure. 256. Rectangular Rectilinear Figures*. (1) Let attention be confined to rectilinear figures lying in a two-dimensional subregion. Then the straight lines of the figures have the properties of planes in this contain- ing region. Let the two-dimensional region cut the absolute in a real section. Let all the rectilinear figures have all their corners spatial, unless otherwise stated. (2) A rectangular quadrilateral (a rectangle) cannot exist. For in such a figure two opposite sides would have two common perpendiculars, contrary to 253 (3). (3) Two alternate sides of any rectangular figure intersect on the (anti-spatial) pole of the included side. Thus [see fig. 2] let A, B, and 0 be three consecutive sides of a rectangular figure, so that the angles at the intersections of A and B, and of B and 0, are right-angles. Let the closed conic in the figure be the section of the absolute. These results have not been given before, as far as I am aware. W. 28
434 HYPERBOLIC GEOMETRY. [CHAP. IV. Then [cf. 251 (3)] A and C must intersect in b\ the pole of B with respect to the absolute. Hence, corresponding to the rectangular spatial figure formed by the lines A, B, C, ..., there is the figure of which the anti-spatial corners d, b\ c, ..., are the poles of the lines of the original figure. Let this be called the reciprocal figure. Then in the reciprocal figure each corner, such as b'', is normal to the two adjacent corners, such as d and c': so that (b' \a) = 0 = ( ' |c'). (4) Let the point of intersection of A and B be a, and the point of intersection of B and G be b. Then ab is the side of the given figure corre- sponding to b'. Also by 246 (3), dd = ab. (5) A rectangular pentagon can be described as follows [cf. fig. 3]: take any two mutually normal anti-spatial points, d and d!; and let a be any third Fig. 3. anti-spatial point, such that dd and dd! are spatial. Let the line A be the polar of d, G of c, D of df; let B be the line dd and E the line dd\ Then the lines A, B, G, D, E, taken in this order, form a rectangular pentagon with its corners spatial. Let A and C intersect in b\ and A and D in e': then db'c'd'e is the reciprocal pentagon. (6) The following formula holds for the rectangular pentagon, giving the length of any side ae in terms of the two adjacent sides ab and de: , ae ,, ab ,. de ... cosh = cotn coth ...........................(i). 7 7 7 In order to prove this formula, assume that the two-dimensional region is the complete region with respect to which supplements are taken; so that we may write b'=\dc\ e=\dd'.
256] RECTANGULAR RECTILINEAR FIGURES. 435 Then, by subsection (4), i a# , be cosh2 = cosh2 = A 7 7 (I (a'c'\ddj _ (a!\cy(a'\d!)* (a'c' | a'd) {a'd' \ a'd') (a'c' | a'c') (a'd! \ a'd') ' since But and Hence COth2 = COth2 = ; / V//x , 7 7 -(ac\ac) 7 7 (ad ad) cosh2 = coth2 coth2 . 7 7 7 (7) The reciprocal figure of a rectangular hexagon can be decomposed into a pair of triangles conjugate with respect to the absolute. For, in figure 4, let the conic be the absolute: take any three anti-spatial points a', c\ e\ so that the three sides of the triangle doe' are spatial. Let A, Gy E be the polars of these points; and let B be the line dc\ D the line ce, F the line e'a'. Then A, B, C, D, E, F, taken in this order, form a rectangular hexagon. Let A and B intersect in a\ B and 0 in b and so on. Thus abcdef is the rectangular hexagon, and a'b'c'd'e'f is its reciprocal figure. 28 2
436 HYPERBOLIC GEOMETRY. [CHAP. IV. (8) The formulae connecting the sides of a rectangular hexagon are simply the formulas of 248 (5) for a semi-spatial triangle. For consider the semi-spatial triangle a'c'e': let a, y, e be the measures of the separation between its sides. Then [cf. 248 (5)], , c'e' , a'c' , a'e' . , a'c' . , a'e' , a cosh = cosh cosh------smh smh cosh ; 7 7 7 7 7 7 sinh sinh -- sinh and / I f f , r . i c e . i a e -. a c smh smh---- sinh 7 7 7 But, by subsection (4), cd = c'e', e/ = e'a', ab = a'c'; and by 253 (3), a = af, y = bc, e = de. Hence the formulae connecting the sides of the hexagon are , cd , ab , ef . , ab . , ef , af ,.., cosh = cosh cosh smh smh cosh ...........(n); 7 7 7 7 7 7 . , ab smh sinh cd and 1 \AiKj . -i /(X smh smh J 7 7 . t be smh 257. Parallel Lines. (1) Two spatial straight lines in a subregion of two dimensions may intersect spatially, or non-spatially, or on the absolute. In the first case let them be called secant*, in the second case non- secant, in the third case parallel. These parallel lines are not the analogues of parallel lines in Elliptic Geometry, cf. 234. (2) If the straight lines be secant, then by starting from a spatial point on either line the point of intersection can be reached after traversing a finite distance. This case is illustrated in figure 5. Fig. 5. * Cf. Lobatschewsky and J. Bolyai (loc. cit.).
257] PARALLEL LINES. 437 If the straight lines be non-secant, then the point of intersection has neither a real linear nor a real angular distance from a spatial point on either of the lines. This case is illustrated in figure 6. Fig. 6. If the straight lines be parallel then the point of intersection is at an Fig. 7. infinite distance from any spatial point on either of the lines. This case is illustrated in figure 7. (3) Let the two straight lines ac be intersect at a point c on the absolute. Then (ac \bc) = (a\b)(c\c)- (a \c) (b c) = - (a c) (b c). Hence if 6 be the acute angle between ac and be, on* 0 = sj{(ac\ac) (bc\bc)} /{(a Hence 0 = 0. Therefore the angle, which two parallel lines make with each other, is zero.
438 HYPERBOLIC GEOMETRY. [CHAP. IV. (4) Any spatial straight line meets the absolute in two points ax and a2. Hence through any spatial point p two straight lines can be drawn in the plane parallel to the given straight line, namely the line pax and the line pa2. From p draw the perpendicular pd on to the line a^. The length between p and d is pd. Let the angle between dp and pax or pa2 be called the 'angle of parallelism;' there is only one angle of parallelism, since it follows from the subsequent analysis that these angles are equal. Then the angle of parallelism is a function of pd only. For in the right-angled triangle pda1} we have Fig. 8. But cos Z ! = sin Z.p cosh . cosZ a1 = 1. sin Z p = sech . 7 Hence This relation can also be written in either of the forms, cot /.p = sinh , and where e is here the base of Napierian logarithms. Let the angle of parallelism corresponding to a perpendicular distance, from the given straight line be denoted by n (f )*. Then the formula above becomes cot .-A*-' = eY. It is to be noticed that when f = 0, n(f) = ^7r; as f increases, II (f) diminishes; and when f is infinite, II (f) is zero. (5) It is possible to draw a straight line parallel to two secant straight lines. For let the two straight lines intersect at an angle a; draw cp bisecting the angle a; and produce it to p so that ^ = 7 log cot |. The perpendicular to cp through p is parallel Fig. 9. to both the secant lines. * Cf. Lobatschewsky, loc. cit.
258] PARALLEL PLANES. 439 258. Parallel Planes. (1) Let the complete region be of three dimensions, then the planes are ordinary two-dimensional planes, and the subplanes are lines. Let two planes L and U intersect in a spatial line LL', and let this line cut the absolute in the two points aY and a2. Then through any two points p and p in the planes L and L' respectively two pairs of parallel lines can be drawn, namely pa1, p'a1, and pa2}pa2. Thus all the lines through ax in the two planes L and L' form one series of parallel lines distributed between the two planes; and all the lines through a2 form another series. And both series are parallel to the line of intersection of the planes. (2) If the line LL' touches the quadric, the points a1 and a2 coincide, and the two series of parallel lines coincide. The planes may then be called parallel. The condition that LL' may touch the absolute quadric is (LL'\LL') = 0. Hence sin Z LL' = 0, and Z LL' = 0. Therefore parallel planes are in- clined to each other at a zero angle. (3) The planes through any point p which are parallel to a given plane L envelope a cone, which has p for vertex and the section of the absolute by L for its section in the plane L. Let pd be drawn perpendicular to Ly and let ab be any tangent line to the absolute lying in the plane L and touching it at a. Then the plane pab is one of the parallel planes through p. Let L' be Fig. 10. written for the plane pab. Now draw pn perpendicular to L'. Then pa, pd and pn are co-planar. For pd and pn pass through \L and \L\ But the line \L\L' is the normal subregion to the line ab. Hence \LL' passes through a,
440 HYPERBOLIC GEOMETRY. [CHAP. IV. since ab touches the absolute quadric. Hence the three lines pn, pd, pa are co-planar. Now the two lines da and pa are parallel. Hence the angle Z apd is n (pd). But the angle Z apn = \tt. Hence Z dpn = tt n (pd). Thus if through any point p, distance from any plane L} all the parallel planes to L are drawn, the normals to these planes form a cone of which all the generators make an angle \tt n ( ) with the perpendicular from p to L.
CHAPTER V. Hyperbolic Geometry (continued). 259. The Sphere. (1) The equation of a sphere of radius p and of spatial centre b is (x \x)(b\b) cosh2 = (b\x)2. Every point of this sphere is spatial. For (b \x)2 is positive and (b\b) is positive ; hence (x\x) is positive [cf. 240 (1)]. (2) The equation e2{x\x)(bb) = {b\x)2 represents an anti-spatial locus, when b is anti-spatial. For then (b\b) is negative, and hence (x\x) is negative [cf. 240 (3)]. (3) The equation -e2(x\x){b\b) = (b\xf represents an anti-spatial locus, when b is spatial. For then (b \b) is positive, and hence (x \x) is negative. But the equation represents a purely spatial locus, wheu b is anti-spatial. For then (b\b) is negative, and hence (x \x) is positive. Let e be written sinh - ; then [cf. 254 (2)] a is the distance of the spatial point x from the spatial plane \b. (4) Thus, if the equation v(x\x)(b\b) = (b\xf be considered as the general form of equation of a sphere, there are two types of real spatial spheres; namely the type with spatial centre b, of which the equation is (x \x)(b\b) cosh2 ^ = (b\x)2; and the type with anti-spatial centre b, of which the equation is -(x\x) (b \b) sinh2 - = (6 \x)\
442 HYPERBOLIC GEOMETRY. [CHAP. V. (5) This latter surface is the locus of points at a given distance r from the spatial plane B(= \b); and the equation can be written [cf. 254 (2)]. ~{x\x) (B \B) sinh2 - = {xB)\ Let the spheres of this second type be also called Surfaces of Equal Distance; and let the plane B be called the Central Plane. (6) The spatial sphere is a closed surface. For firstly, let the centre be spatial, and let it be taken as the origin, e, of a normal system of reference points with the notation of 240. Then (a.|fl.) = ?_g_..._|g, | Hence the equation of the sphere is, 7 2' that is, ^^^ Therefore the sphere is a closed surface [cf. 82 (5)]. (7) Secondly, let the centre be the anti-spatial point, e1} of this normal system of reference points. Then (ei|x) = ^1(ei|eI) = -|L2; sinh2 - and - 0i 10i) sinh2 - =-------21. 7 i Hence K~K~ - "^- = cosech2 - % a2 ct!2 aVi 7 i Therefore the equation of the sphere becomes q a2 a,2 a22 "" a2^ * This is the equation of a closed surface [cf. 82 (5)]. (8) It is to be noticed that this last closed surface touches the absolute along the real locus of v 3 dimensions given by the equations (9) If the centre of a sphere be spatial, it lies within the surface. For let b be the centre of (a?. \x)(b\ b) cosh2 9- - (b | xf = 0.
259] THE SPHERE. 443 Then if x be any point, the point b + \x lies on the surface, when (b 16)2 sinh2 + 2X (6 \x) (b \b) sinh2 + X2 {(x \x) (b \b) cosh2 - (61^)2} = 0. The roots of this quadratic for X are real, if (b | xj (b 16)2 sinh2 e - (6 16)2 {(a? | a?) (6 16) cosh2 - (6 | xf) is positive ; that is, if (6 \x)2cosh2- (x\x) (b \b) cosh2 - is positive ; that is, if (bx \bx) is positive. But [cf. 243 (2)] since the line bx is spatial, this condition is fulfilled. Hence [cf. 82 (1)] b lies within the sphere. Also if b be substituted in the expression (x | x) (b | b) cosh2 -(b | x)\ there results (6 j 6)2 sinh2 - , which is positive. Hence [cf. 82 (9)] any point x, which makes this expression positive, lies within the surface. (10) But if the centre b be anti-spatial, then the equation of the sphere is - 01x) (b |b) sinh2 - - (b \xf = 0. Then if x be any point, the point b + \x lies on the surface, when (|)(|){( 7 7 The roots of this equation for X are real, if (x \b)2 (b \b)2 cosh2 - - (6 \b)2 {(x \x) (b \b) sinh2 - + (b \xf\ is positive ; that is, if (w \b)2 (x \x) (b \b) is positive ; that is, if (xb \xb) is positive. But xb may or may not be spatial; and therefore (xb \xb) may be positive or negative. Hence the anti-spatial centre of a sphere lies without the surface. But it is interesting to note that any spatial line, drawn through the (anti-spatial) centre, cuts the sphere in real points. Also substituting b in the expression we obtain (61 b)2 cosh2 - , which is negative. Hence any point x, which makes this expression positive, lies within the surface [cf. 82 (9)]. Thus any
444 HYPERBOLIC GEOMETRY. [CHAP. V. spatial point on the central plane | b lies within the surface. For, if x be such a point, (x \b) = 0, and (x\x) (b\b) sinh2 - is positive. (11) It can be proved, exactly as in the case of Elliptic Geometry [cf. 228 (3)], that the line, perpendicular to any plane and passing through its pole with respect to a sphere, passes through the centre of the sphere. Hence it follows as a corollary, that the perpendicular to a tangent plane of a sphere through its point of contact passes through the centre of the sphere. 260. Intersection of Spheres. (1) The locus of the intersection of two spheres lies on the two planes {b\x)_ (c\x) A A These planes are respectively the absolute polar planes of the points b o But if b and c are both spatial and of standard sign, the point -5- + t* Pi H2 is spatial, and its polar plane is anti-spatial. This plane can only meet the spheres, which are entirely spatial, in imaginary points. If b and c are anti- b _ c spatial, one of the two points ^ + -5: must be anti-spatial, and hence its Pi P2 polar plane spatial. The other point may or may not be anti-spatial. These radical planes are perpendicular to the line be, since be passes through their poles. (2) It can be proved, as in the Elliptic Geometry [cf. 228 (4)] that the lengths of all tangent lines from any given point to a sphere with a spatial centre are equal. Also if p be the radius, and b the centre of the sphere, and t the length of the tangent line from x, then , r (x\b) cosh- = ' ' 7 7 Also, by an easy modification of the proof and by reference to 246 (2), we find when the centre b is anti-spatial, and the distance from the plane |6 is r, cosni=
260] INTERSECTION OF SPHERES. 445 (3) The locus of points, from which equal tangents are drawn to two spheres with spatial centres b and c, and with radii px and p2, is given by (*\b) (x\c) V(|) V(|) This locus is the spatial radical plane. The cases when 6 or c, or both, are anti-spatial can easily be discussed. (4) The theorems of 228 (5) also hold, with necessary alterations. (5) The angles of intersection of two spheres can be investigated by the method of 228 (8). Let the two spheres be e(b\b)(x\x) = and v (c \c) (x \x) = (c \x)2; where e stands for cosh2 , if b be spatial, and for - sinh2 1, if b be anti- 7 7 spatial [cf. 259 (4)] ; and rj stands for cosh2 ?-, or for sinh2 , according as c is spatial, or anti-spatial. Then it can easily be proved, as in the analogous theorem of Elliptic Geometry, that the angles of intersection, co and (c)', of the two spheres, which correspond to the two radical planes, are given by + /{eV(b\b)(c\c)}-(b\c) and cos '- Also let it be assumed that in all cases (b\c) is positive. Four separate cases now arise. (6) Firstly, let b and c be both spatial. cosh cosh cosh Then cos a, =------1--------1----------2 , sinh ^ sinh ^ 7 7 cosh cosh cosh and cos o' = 7 V 7 sinh^1 sinh ^ 7 7 Since coth coth is necessarily greater than unity, / is always imaginary.
446 HYPERBOLIC GEOMETRY. [CHAP. V. The spheres have one real intersection, if 1 cos co 1; that is, if pi ~ p2 bc p1 + p2. (7) Secondly, let b and c be both anti-spatial; and let be be anti-spatial. Let b = \B, and c = \C\ so that B and G are the central planes of the spheres. Then, since be is anti-spatial, BG is spatial. sinh sinh cos BG Now cos (c) = , cosh cosh 7 7 sinh sinh cos BG and cos ' = ^ ' cosh cosh 7 7 Then co and a/ are both necessarily real. (8) Thirdly, let b and c be both anti-spatial; and let be be spatial. Then BG is anti-spatial. sinh sinn-----cosh Then cos co =-------2--------2------------21 y cosh cosh 7 7 sinh ^ sinh cosh 7 7 7 cos (c) = /---------' ' . cosh cosh 7 7 The angle w is real, if The angle w is real, if The first condition secures a real intersection on one radical plane ; the second condition secures a real intersection on both radical planes. (9) Fourthly, let b be spatial, and c be anti-spatial. Let 8 be the dis- tance from b to the central plane G. Then [cf. 254 (2)], . , 8 ^ (bG) sinh - = + T v 7 y cosh sinh sinh - Hence, cos co = '; sinh cosh 7 7
261] INTERSECTION OF SPHERES. 447 cosh ^ sinh sinh - and cos co' = 7 7 7 ^ sinh ^ cosh 7 7 Then co is real, if cr2 px S a2 + pY; and / is real, if p1 r2. This condition for co' includes the conditions for co, since 8 has been assumed to be positive. (10) Now a spatial plane is a particular case of a sphere with an anti- spatial centre c, when cr 0 ; the plane is then \c. Hence from subsection (9) the plane L cuts the sphere, with spatial centre 6, at an angle co given by x. Pi ~ (* L) sinh cos co = -j. , ,, ' .... . And from subsection (8), the plane L cuts the sphere, with anti-spatial centre 6, at an angle co given by cosh cos co Hence putting co 0, the plane-equation of a sphere is when the centre is spatial; and is when the centre is anti-spatial. 261. Limit-Surfaces. (1) If b be on the absolute, the surface denoted by 2(tf|#) = ( |#)2 is called a limit-surface. It must be conceived as a sphere of infinite radius. Since the centre is on the absolute, by 259 (11) all the perpendiculars from points on to their polars with respect to the surface are parallel lines. (2) Let a distance S be measured from every point w on the above limit- surface along the normal xb, either towards or away from b. Let y be the point reached. Then x and X can be eliminated from = y, cosh2-= ^ffV 7 O* The result, remembering that (b\b) = 0, is easily seen to be je* exp (^)\y \y) - (b |y)j je* exp (- ^) (y \y) - (b |y) j = 0.
448 HYPERBOLIC GEOMETRY. [CHAP. V. The surface obtained by measuring towards b is therefore The surface obtained by measuring from b is Both these surfaces are again limit-surfaces with b as centre. (3) Now assume that the spatial origin e, of a normal system of unit reference points e, elf ... ev-ly is on the surface. Let ee1 pass through the centre b. Then b is of the form e elf say e + ei. The equation of the surface becomes But since e is on the surface, we can put x= e in this equation. Hence e2 = 1. The equation now is (x\x)={x\{e + e,)}\ This form, by its freedom from arbitrary constants, shows that all limit- surfaces are merely repetitions of the same surface differently placed. 262. Great Circles on Spheres. (1) Let any two-dimensional region, through the centre of a sphere and cutting the sphere in real points, be said to cut the sphere in a great circle. Accordingly a great circle is in general defined by two points on a sphere, since these two points and the centre of the sphere (if not collinear) are sufficient to define the two- dimensional region. The radius of the circle is the radius of the sphere, and the centre of the circle is the centre of the sphere. If the centre be anti- spatial, the circle is the surface of equal distance in the two-dimensional region from the line of intersection of the two-dimensional region with the polar plane of the centre. The two-dimensional region, since it contains the centre, is perpendicular to the polar plane of the centre, that is, to the central plane of the sphere. (2) If the centre b be spatial, and two points pq on the surface define a great circle, then the length of the arc pq of the great circle [cf. 248 (4)] is ay sinh , where p is the radius of the circle, and a is the acute angle between pb and qb. (3) Let the centre be anti-spatial. Consider any two points p and p' on a surface of equal distance a from any given plane. Let the two perpen- diculars from p and pf meet the given plane in q and q'. Then the length of
262] GREAT CIRCLES ON SPHERES. 449 the arc pp', traced on the great circle joining^ and p\ can be found in terms of q and q\ For putting qq =8, it is easy to prove that cosh ^ L sinh2 - + cosh2 - cosh - . 7 7 7 7 V P' P Fig. 1. Hence when pp' and S are small, Therefore But ultimately, pp' = arc pp'. Therefore pp = S cosh - . 7 qq 7 But if p" be any point on the arc pp' prolonged to a finite distance, and if p"q" be drawn perpendicular to the plane, it is obvious that arc pp" arc pp , a S, = -IT = cosh - . qq qq 7 (4) Let d be the centre of a spatial sphere of radius p, and let a, b, c be three points on the sphere. Let the acute angle between db and dc be a!, that between dc and da be ', that between da and db be y ; let the angle between the two-dimensional regions dab and dac be a, that between dab and c$c be /3, that between c?6c and dca be y. Then the three two-dimen- sional regions can be conceived as planes in a three-dimensional region. Hence by 250 (2), cos a' = cos /3; cos y + sin ft' sin y cos a. Now a, j9, y are the angles of the curvilinear triangle formed by the great circles joining a, b, c. Also if be, ca, ab stand for the lengths of the arcs of great circles, by (2) of the present article, a' = cosech ff = cosech - , y = cosech - . 7 y 7 7 y n' w. 29
450 HYPERBOLIC GEOMETRY. [CHAP. V. ca ab ca 7 sinh 7 sinh - 7 sinh - 1 7 7 7 with similar equations. Thus the relations between the lengths of the arcs, forming a triangle of great circles on a sphere of spatial centre, and the angles between them are the same as the relations between the sides and angles of a triangle in an Elliptic Space, of which the space constant is 7 sinh -. Thus an Elliptic Space of v 2 dimensions can always be conceived as a sphere of radius p with spatial centre in Hyperbolic Space of v 1 dimensions, the great circles being the straight lines of the Elliptic Space, 7 being the space constant of the Hyperbolic Space, and 7 sinh - that of the Elliptic Space. (5) Let a sphere with anti-spatial centre be a surface of equal distance a from a spatial plane; and let a, 6, c be three points on the sphere, and d, b', c be the feet of the perpendiculars from a, b, c on to the plane of equi-distance. Let a, b, c be joined by great circles of lengths be, ca, ab. Let a, J3, y be the angles of the curvilinear triangle abc; they are also the angles of the triangle db'c\ since the two-dimensional regions containing the great circles are perpendicular to the plane of equal distance. Then [cf. 248 (2)] , b'c , c'd , db' . , daf . , db' cosh = cosh cosh - sinh - sinh cos a. 7 7 7 7 7 _ __ c But by subsection (3) of the present article, b'c' = ^^ , with similar cosh- y equations. Hence i be , ca , ab . , ca . , ab 7 cosh - 7 cosh 7 cosh 7 cosh 7 cosh 7 7 7 7 7 Thus the relations between the lengths of the arcs, forming a triangle of great circles on a sphere of equal distance cf from a spatial plane, and the angles between them are the same as the relations between the sides and angles of a triangle in a Hyperbolic Space, of which the space constant is 7 cosh - . Thus, since 7 cosh is always greater than 7, a Hyperbolic Space of v 2 dimensions can always be conceived as a spherical locus with anti- spatial centre in a Hyperbolic Space of v 1 dimensions and of smaller space constant.
263] GREAT CIRCLES ON SPHERES. 451 (6) The relations between the sides and angles of a curvilinear triangle formed by great circles on a Limit-surface can be found either from (4) or (5) by making p or a ultimately infinite. Then with the notation of (4) or (5) 6c2 = ca2 + ab2 2ca. ab cos a. Hence triangles formed by great circles on Limit-surfaces have the same geometry as triangles in ordinary Euclidean Space; for instance, the sum of the angles of any such triangle must equal two right-angles. Thus a Euclidean Space of v 2 dimensions can be conceived as a Limit-surface in a Hyperbolic Space of v 1 dimensions*. 263. Surfaces of Equal Distance from Subregions. (1) Let Pp be a spatial subregion of p 1 dimensions, and let Pp be the regional element of the pth order which represents it. Then locus of points x at the given distance S from this subregion is by 254 (5), (x \x) (Pp |PP) sinh2 - + (xPp\xPp) = 0. (2) Now take as reference elements v normal points, of which the spatial origin e and p 1 other points ely e2, ... ep^ lie in Pp, and the remaining v p elements lie in \PP. Also let e be at unit spatial intensity, and ex, e2, ... ev-i at unit anti-spatial intensity. Let Pp = ee1 ... ep-x. Then (Pp|Pp)=(-l)p- Let x = f e + fYex + ... +; ^-i^-! + t)pep + ... + v^-i^-i- Then (^PP|^PP) = (-1 S7?2, (^|^) = ^_^-...-^_i Hence the equation of the surface of equal distance from Pp becomes that is, 7 This is a closed surface with no real generating regions. Hence the parallel regions of Elliptic Space [cf. 229] have no existence in Hyperbolic Space. * The idea of a space of one type as a locus in a space of another type, and of dimensions higher by one, is due partly to J. Bolyai, and partly to Beltrami. Bolyai points out that the relations between lines formed by great circles on a two-dimensional limit-surface are the same as those of straight lines in a Euclidean plane of two dimensions. Beltrami proves, by the use of the pseudosphere, that a Hyperbolic space of any number of dimensions can be considered as a locus in Euclidean space of higher dimensions. There is an error, popular even among mathematicians misled by a useful technical phraseology, that Euclidean space is in a special sense flat, and that this flatness is exemplified by the possibility of an Euclidean space containing surfaces with the properties of Hyperbolic and Elliptic spaces. But the text shows that this relation of Hyperbolic to Euclidean space can be inverted. Thus no theory of the flatness of Euclidean space can be founded on it. 29 2
452 HYPERBOLIC GEOMETRY. [CHAP. V. 264. Intensities of Forces. (1) Consider an extensive manifold of three dimensions. The only regional elements are planar elements and forces. A spatial planar element X is at unit intensity when (X]X) = 1, and an anti-spatial planar element X is at unit intensity when (X \X) = 1 [cf. 240 (3)]. (2) In order to determine the intensity of a force xy, let it be defined that the intensity of xy is some function of the distance xy, or Z xy if the measure of distance be elliptic, multiplied by the product of the intensities of x and y. Then by the same reasoning as in 230 (2) for Elliptic Space it can be proved that: (a) if x and y be both spatial, the intensity of xy is a/{(# \oo)(y \y)} sinh , that is { xy\xy}h; where it is to be noticed that, by 243 (2), (xy \xy) is negative, when xy is spatial: (/S) if x and y be both anti- spatial and xy be spatial, the same law of intensity holds as in (a): (7) if xy be anti-spatial, the intensity of xy is \/{{x\x)(y\y)} sinxy, that is (xy\xy)^. Hence the intensity of a spatial force F is { F\F}%, that of an anti-spatial force F is {F\F}K (3) If P and Q be two planes the standard form of a real force of the type xy is iPQ. If the force be anti-spatial, its intensity is if the force be spatial, its intensity is y/{(P \P) (Q \Q)} sin PQ. 265. Relations between two Spatial Forces. (1) In general [cf. 231 (1) and 255 (6)] there are only* two lines intersecting the four lines F' Fig. 2. F, F', \F, \F\ Let these be the lines ab and cd. Then ab and cd are perpendicular to both F and F''; also each is the polar line of the other. One of the two must be spatial and the other anti-spatial. Assume ab spatial. Let ab = S, and Zcd=0. Then 8 is the shortest distance between the lines, and 0 will be called the angle between the lines. * For the discussion of an exceptional case for imaginary lines see the corresponding discus- sion for Elliptic Space, cf. 234, in which case the lines are real.
264, 265] RELATIONS BETWEEN TWO SPATIAL FORCES. 453 Let F=ac, F' = bd. Then S (a'6) (rf|c) also (i^|i^) = (ac| d) = (a |6)(c|d), since (a \d) = 0 = (b \c). Also (JfT|JPT) = (ac|ac) = (a|a) (c|c), since (a|c) = 0. And (i*7' |^) = (b\b)(d\d), since (6 | d) = 0. ( |a) (616) (c |c) (d \d)} ~ (2) Again, let a' be the point normal to a on the line ab, and c' be the point normal to c on the line cd. Also let (a |a') = (a \a), where a is assumed to be spatial, and (c' \c) =(c \c). , a cosh - + a' sinh - , , . Then -______2!__________2 and d _ c cos (9 + c sm 9 inen V (b |6) ~ V( | ) ' V (d |d)" V- (c |c) ' Hence / i. j\ * (aa'cc') sin ^ sinh - i (acbd) v 7 7 7{(a~\a)(b\b)(c\c)(d\dj\ = -(a\a){b\b) = + sin 8 sinh -; 7 since [cf. 240 (5)] i(aa'cc) = (a\a)(b\b). (3) If the forces intersect, the point of intersection is either spatial or anti-spatial or on the absolute. If the point of intersection be spatial, then 8 = 0. Hence (FF') = 0; and (F\F') If the point of intersection be anti-spatial, then 0=0. Hence (FF') = 0 ; and W\F') ,8 ?\ / EV i ei/m = COSh - . If the point of intersection be on the absolute, so that the lines are parallel, then 8 = 0 = 6. Hence (FF') = 0, and (F\FJ = (F\F)(F'\F'). (4) Let two forces F and F' have a spatial intersection, and let their intensities be p and p'. Let the single force F + F' be of intensity x. Let 6 be the angle between F and F'. Then T* = -(F + Ff)\(F+F') = p* + p'* 2PP'cos0. The upper sign of the ambiguity must be chosen so that, when 0 = 0,
454 HYPERBOLIC GEOMETRY. [CHAP. V. If F and F' be spatial but have an anti-spatial point of intersection, let S be the shortest distance between F and F'. Then as before, if F + F' be spatial, o-s =/02 + p'* + 2pp cosh-. But it is possible that, though F and F' are spatial, F + F' may be anti- spatial, the intersection of F and F' being anti-spatial. In such a case a2 = 2pp cosh --P2-p'2. 266. Central axis of a System of Forces. (1) It has been proved in 175 (14), also cf. 232, that any system of forces has in general one and only one pair of conjugate lines, which are reciprocally polar with respect to a given quadric. Now let a system S have the two conjugate lines a^a2, 3 4 which are reciprocally polar with respect to the absolute. One of the two must be spatial, the other anti-spatial. Let a2a2 be spatial. Then S= \axa2 + fia3a4; and this form of reduction is unique. The line a^a2 will be called the central axis of the system. Let the points alf a2, a3, a4 be so chosen at unit intensity, that 8 = a^a + a^; then aYa2 ( = 8) and Z aBa4 (= a) will be called the parameters of the system. A system, Sy referred to its central axis may also be written in the form a1a2-\-ie\a1a2y where e is real. Then (88) = 2ie (a,a2 {cw*), and (8 \S) = (1 - e2) (axa2 \aYa2). (2) Let 8 denote the system F -f ie \ F, and 8' the system F' + irj \F\ Also with the notation of 265, let S be the shortest distance between the lines and 8 the angle between the two. Then (88T) = (1 - eV) (FFf) + % ( + v) (F [ F') = {(F \F) (Ff \F')f | i (1 - 67;) sin 6 sinh - + i(e + v) cos 0 cosh -I. And (S \S') = i(e + v) (FF') + (l - eV)(F\Ff) = {(F\F) (F' \F')f |+ (c + V) sm 6 sinh - + (1 - V) cos 0 cosh - (3) The simultaneous equations (SSf) = 0, (8\ S') = 0, secure that the axes of S and S' intersect at right angles. For from (2), unless e or 77 be i, which is the case of an imaginary system analogous to the real vector system of Elliptic Geometry [cf. 235], (88') = 0 and (8 \Sr) = 0 entail cos 0 cosh - = 0, sin 0 sinh - = 0 ; that is 8 = 0, 0 = . 7 7 ^
266, 267] CENTRAL AXIS OF A SYSTEM OF FORCES. 455 (4) Every dual group contains one pair of systems, and only one pair, such that their axes intersect at right angles. The proof is exactly the same as for the analogous theorem of Elliptic Geometry, cf. 232 (4). Let this pair of systems be called the central systems of the group, and let the point in which their central axes intersect be the centre of the group. (5) Dual groups with real director lines can be discriminated into three types according as, either (a) both director lines are spatial, or (/8) both director lines are anti-spatial, or (7) one director line is spatial and one is anti-spatial. (6) To find the locus of the central axes of a dual group, let e be the centre, eeY and ee2 the axes of the central system, and ee3 a line perpendicular to the lines ee1} ee2. Also let eeYe$z be a normal system at unit intensities. Let $j = eex + % .Y \eel9 S2 = ee2 + ie2 \ee2 be the central systems of the group. Now [cf. 240 (5)] we may assune eBe1 = -i \ee2, e2e.3 = -i\ee1} \e^x = -iee2, \e2e3 = -ieex. Any other system S' of the group can be written S' = Xj^! -1- Xg j = e (X^i -f X2e2) + i \. e (X^^ + X2e2e2). Then, as in 232 (5), this system can be identified with the system (e + fo) (fi1e1 + fi2e2) + ie\.(e + 3) (/^ + fi ). Hence all the central axes of systems of the group intersect ees at right- angles. Let ee3 be called the axis of the group. The equation to find e is ( 2 - 1) {e^2 + 622V} - e {( l - 1) X,2 + (e22 - 1) X22} = 0. The locus of any point 2 e on an axis of any system of the group is (* ~ *) ftfi (P " ) = (1 + exeO ff, (ft + m 267. Non-Axal Systems of Forces. (1) A system of forces, not self-supplementary, and such that (SS) = (S 8)% is called a non-axal system [cf. 233]. (2) All such systems are imaginary. For if S be real, then [cf. 240 (5)] it is easily proved that (SS) is a pure imaginary, and that (S 18) is real. (3) Hence, from this article and from 266 (3), any system S, such that (SS) = + (S IS), whether it be self-supplementary or not, is imaginary. (4) Accordingly^ the theorems of 266 hold for all real systems of forces.
CHAPTER VI. Kinematics in Three Dimensions. 268. Congruent Transformations*. (1) A congruent transformation is a linear transformation, such that (a) the internal measure relations of any figure are unaltered by the transformation; e.g. if abc is transformed into a'b'c', then ab = a'b', and the angle between ab and ac is equal to that between a'V and ac, and similarly for the other sides and angles: (ft) the transformation can be conceived as the result of another congruent transformation p times repeated, where p is any integer: (7) real points are transformed into real points: and (S) the intensities of points are un- changed by transformation. It follows from (a) that points on the absolute must be transformed into points on the absolute. Hence congruent transformations must transform the absolute into itself. It follows from (a), (ft) and (7) that spatial points must be transformed into spatial points. For from (a) [cf. 241 (1)], either all spatial points are transformed into spatial points, or all spatial points into anti-spatial points. Also from (ft), since the integer p may be taken indefinitely large, a finite transformation can be considered as the result of p repetitions of an infini- tesimal transformation. But an infinitesimal transformation must transform spatial points into spatial points. Hence the same holds for a finite trans- formation. (2) Let the discussion be now confined to regions of three dimensions. To prove that a congruent transformation must transform the absolute by a direct transformation (cf. 194). For by (ft) of (1) any congruent transformation can be conceived as the result of p repetitions of another congruent transformation. But an even number of applications of either a direct or a skew transformation of a quadric produces a direct transformation of the quadric. Hence every congruent transformation is a direct transformation of the absolute. * The theory of congruent transformations is due to Klein, cf. loc. cit. p. 369 ; Buchheim has applied Grassman's algebra to this subject, cf. loc. cit. p. 370.
268] CONGRUENT TRANSFORMATIONS. 457 (3) By 195 (2) to (6) among the latent points of a direct transforma- tion of the general type there are the points of intersection of two conjugate polar lines with the quadric. Now in Elliptic Space the absolute is imaginary; and therefore the co-ordinates of the latent points on either one of the polar lines, referred to real reference points, form pairs of conjugate imaginaries. In Hyperbolic Space one polar line must be anti-spatial and one is spatial: the co-ordinates of the latent points on the anti-spatial polar line, referred to real reference points, are pairs of conjugate imaginaries: the latent points on the spatial polar line are two real points, and their co-ordinates are real. Now either in Elliptic or in Hyperbolic Space let aly a2, a3, a4 be the four above-mentioned latent points of a congruent transformation, and let aYa2 and z3a4 be conjugate polar lines. Also let a3 and aA be imaginary points, then their co-ordinates are pairs of conjugate imaginaries. Hence if a3 and a4 be taken as reference points, the co-ordinates, tj3 and 774, of a real point y (= rj3a3 + ^4a4) are conjugate imaginaries. (4) Let the latent roots of the congruent matrix be a1} a2} a3, ar4. Then y is changed into 7}3a3a3 + rj^a^, and by (7) of (1), ^3a3a3H- ?;4a4 z4 is a real point. Hence a3 and a4 must be conjugate imaginaries. Similarly if aY and a2 are imaginary points as in Elliptic Space, ax and a2 are conjugate imaginaries; but if aa and a2 are real points as in Hyperbolic Space, ax and 0^ must be real. Hence in Elliptic Space the latent roots are two pairs of conjugate imaginaries, in Hyperbolic Space one pair are real and one pair are conjugate imaginaries. (5) In Hyperbolic Space both the real latent roots c^ and a2 must be positive. For the given congruent matrix may be conceived, according to (/3) of subsection (1), as the result of another congruent matrix twice applied. Let /31? /32, /33, /34 be the latent roots of this matrix. Then /3j2 = a,, and /322 = a2. But and /32 are real by the same proof as that for aY and a2; hence ax and a2 are positive. Therefore the real roots of a congruent matrix in Hyperbolic Space are positive. (6) By 195, aid.} = a3or4. Hence in Elliptic Space we may put, And in Hyperbolic Space we may put, S ax = \ey, a2 = Xe *, as = \eia, a4 = Xe~ia. Also by (8) of (1) the intensity of any point is unaltered by trans- formation. Now the intensity of 7]^ + ^2^2 is {viVz (ai K)}2 and the intensity of the transformed point is
458 KINEMATICS IN THREE DIMENSIONS. [CHAP. VI. Hence \ = 1. Thus the latent roots in Elliptic Space take the form and in Hyperbolic Space t _s_ (7) The special type of direct transformation, with only three semi- latent lines [cf. 195 (7)], cannot apply to Elliptic or Hyperbolic Space, so as to give a real congruent transformation. For, with the notation of 195 (7), the points elt e2, e3, e4 and the planes eYe2e3 and exe3e4 are imaginary, both in Elliptic and Hyperbolic Space. But an imaginary plane always contains one straight line of real points. Hence the semi-latent plane exe2e3 contains one real line. But, since real points are transformed into real points, this line must be semi-latent. Also the semi-latent lines exe2 and exe3i which lie in this plane must be imaginary, since they are generators. Hence a third semi-latent line must lie in this plane; and this is impossible in this type of transformation. (8) The theory of congruent transformations in Hyperbolic Space will first be discussed, cf. 269 to 280, and then that of congruent transforma- tions in Elliptic Space, cf. 281 to 286. 269. Elementary Formulae. (1) Let alt a2, a3, a4 be the latent points of a congruent transformation in Hyperbolic Space. Let axa2 and a3a4 be conjugate polar lines; and let them be called the axes of the transformation. Let axa2 be spatial, and be called the spatial axis, or more shortly the axis; then a3a4 is anti-spatial, and may be called the anti-spatial axis. Thus ax and a2 are real; and a3 and a4 are imaginary. Then it at once follows that, (a ax) = 0 = (a2 |a2) = (a3 \a3) = (a4 \a4) = (a^ \a3) = (ax \a4) = (a2\a3) = (a2 \a4). (2) Let e, e1} e2i e3 be a normal system of elements at unit intensity, of which e is the spatial origin. Let eel be the line axa2, and e2e3 the line a3a4. Then by 247, we may write in in in in a__ /y I gy /-f __ /y ___ /y /~f _ sy 0 i ^4 gy /-w __ /} 4 yy I jy 4 S) I t/\ ~T" t Uj2 Xj\ C/, U/3 C/ t?2 T~ 3 l"4 t/ t?2 "t" t? f?3 Hence (aj \a2) = (eY \e^ (e |e) = 2 ; and (a3 \a4) = (e2 \e2) + (e3 \ e3) = - 2. Also from 240 (5) (eexe2e3) = i. Hence (a^a^) = 4i {eexe2e3} = 4 = (ax | a2) (a3 \ a4).
269, 270] ELEMENTARY FORMULAE. 459 (3) Again, by substituting for c^, (h, s5 4 in the expression for any real point x (= ^ax + 2a2 + 3a3 + 4a4), we find * = (fi - ) + (fi + f.)* + (f rf" + f46*) e. + (frf7 + f 4 f *7)*,. . But since x is real, the coefficients of e, ely e2, e3 are real. Hence 3 and 4 are conjugate imaginaries. Let 3 = peid, 4 = pe~^. Then x = (ft - ) e + ( + )* + 2p cos (d- j) e, + 2p cos (o +1) e3 = ve + V\e\ + ^2^2 + ^3^3 (say). Any spatial point x must satisfy the condition that (x \x) be positive. But (x | a?) = 2 9 (ax | a2) + 2f,f4 (a, |a4) Hence for a spatial point 2 is negative, and ^^2 p2. (4) The congruent matrix transforms # into x\ where = (fi^ - f.e"r) 6 + (f^ + ^2e ^) e, 2p cos ^ + a - j) ^ + 2p cos + ih )\( ih = (77 cosh - + ^i sinh -)e-\-(rj sinh - + Vi cosh -) ^i \ 7 7^ V 7 7/ cos a + ^3 sin a) e2 + (773 cos a ?72 sin a) e3. 270. Simple Geometrical Properties. (1) Consider any point x - -~ (=1^4-^2) on the axis a^. It is transformed into of = g^a,! + %# ya2, which is again on the axis. Furthermore, s_ zii? = l log K, lffl2} = I log 1^- = I log e = 8. Thus all points on the axis are transferred through the same distance 8. (2) Again, consider any plane P through the axis aYa2 and any point x (= 3a3 + 4 z4) on the anti-spatial axis a3a4. Let the plane axa2az be called AZy the plane a^a^a^ be called A4. Then u43 and A are the two planes through oa which touch the absolute in imaginary points az and a4. Also p = 3A3 + 4-44; and P is transformed into the plane P' = %ze -Az -f f46~lfltJ.4. Furthermore, Z Pi" = 1 log {PF, ASA ) = g log |^| = Hence every plane through the axis is rotated through the same angle a.
460 KINEMATICS IN THREE DIMENSIONS. [CHAP. VI. (3) The distance, ot, of any spatial point x from the axis is given by . , ^ _ J (xaYa21 xa^2) \2 _ %p 7 ~~ I" O \x) (^i 21ofo)) ~~ (x |xf' where a; is written in the form g^ + %2a2 + peiea3 + pe^a^. But the trans- formed point x' is ^aY-\-^2e ?a2 + jo^(e+a)a3-f pe~l^e+a)a^ and it is obvious that its distance from the axis is the same as that of x. Hence the distance of a point from the axis is unaltered by the congruent transformation. Also it is easy to prove that ------; r r ----------- -,----;----r------- y \{x\x) ) \ k{x\x) J (4) To find the distance of the transformed point x' from the plane through x perpendicular to the axis axa2 of the transformation. This plane is represented by xa3a4. Now xa3a4 = 2ixe2e3] but 2ixe2e3 is in the standard form of 240 (2); hence xa3aA is in the standard form. Now if f be the distance of x' from this plane, it is easily seen after some reduction that i K , (xxa3a^) i r 18 sinn - = -77 TfTTF----\-------v = cosn sinn - 7 vl (^ x)(xa3ai\xa3a4)i 7 7 Hence f is independent of a, and depends only on S and on the distance of x from aY(ju^ Also when S = 0, f=0; and when ^ = 0, f becomes B in accordance with subsection (1). (5) It is easily proved that __, , . /x 1I2 cosh - + p2 cos a . . xx (x\x) b b 7 . w , o . . r cosh = 7-----^rr": 7vr=--------j^^-----^2-------=cosh2- cosh- + sinn2- cos a. 7 sj{(x\x)(xf\af)} Z1 + P2 7 7 7 271. Translations and Rotations. (1) Let the quantities 8 and a be called the parameters of the transformation. If a = 0, the congruent transformation is called a translation through the distance S with aYa2 as axis. The effect of the translation on a point x is that the transformed point xY is at the same distance as x from a^, is in the plane xa^a^, and is at a distance 7 sinh"1 j cosh sinh -J- from the plane xa3a4, where tn- is the distance of x from axa2y or in other words the plane x^a^ is at a distance S from the plane xa3a4, a^2 being their common perpendicular. (2) If 8 = 0, the congruent transformation is called a rotation through the angle a with a^a2 as axis. The effect of the rotation on the point x is that the transformed point x2 is at the same distance zj from axa2i and is in the plane xa3aiy and that the plane x2axa2 makes an angle a with the plane xaxa2.
271] TRANSLATIONS AND ROTATIONS. 461 (3) It is obvious that the general congruent transformation, axis a^, parameters (8, a), in its effect on any point x is identical with the effect first of the translation, axis a^, parameters (8, 0), bringing x to xlf and then of the rotation, axis aYa2i parameters (0, a), bringing x1 to x'\ or first of the rotation bringing x to x2, and then of the translation bringing x2 to x. It is to be noticed that congruent transformations with the same axis are convertible as to order, but that when the axes of the transformations are different the order of operation affects the result. It will be convenient for the future to use the letter K for the matrix representing a congruent transformation, so that Kx is the transformed position of x. (4) A further peculiarity of translations and rotations is established by seeking the condition that a real plane, other than one of the faces of the tetrahedron a^a^a^ may remain unchanged in position, i.e. be semi-latent. Let x be any point in such a plane. Then x, Kx, K2x, K*x must all lie in the plane. Therefore (xKxK2xK3x) = 0. But x = f^ + faGfe + pePa* + per^a^, Kx = a3 Therefore 25 35 23 35 K3x = fay a, (xKxK*xK*x) y a2 + pel ^+Sa) a, + pe 1, 1 , 1 , e+la e~m 25 35 25 35 e+s e~sia Now gi = 0, or f2 = 0, or p = 0, each makes the plane to be one of the three real faces of the tetrahedron {a^a^a^). Hence the determinant must vanish, if another real plane is semi-latent. But this condition can be written, 5 _5 6 5 _5 6 {ey - e y) (ey - ep) (ey - er4*) (e y - eu) (e~y - er*") (e^ - e~ia) = 0. The only solutions, making 8 and a real, of this equation are 8 0 or a = 0. Thus rotations and translations are the only congruent transformations by which planes do not change their positions. For rotations, planes per- pendicular to the axis are rotated so as to remain coincident with them- selves; for translations, planes containing the axis are translated so as to remain coincident with themselves. In other words, the only possible motions
462 KINEMATICS IN THREE DIMENSIONS. [CHAP. VI. of a plane, which remains coincident with itself, are either rotations about a perpendicular axis, or translations along an axis lying in it. Similarly it is obvious that rotations and translations are the only con- gruent transformations by which points other than the four corners of the self-corresponding tetrahedron do not alter their positions. For rotations these points are points on the axis, for translations they are points on the anti-spatial axis. (5) The property, which discriminates a translation from a rotation, is that, as a translation is continually repeated, the distance between the original and final positions of any spatial point grows continually greater. For let the translation, axis a^, parameters (S, 0), be repeated v times, and let Then Kvx = ^ a* + %2ev a2 -f peiea3 + pe~i9a4. Hence (x \Kvx) = 4 2 2 cosh------4p2. Therefore xKvx = y cosh" (x \x) Hence xKvx grows continually greater as v is increased. But in the case of a rotation, if the parameter a bear a commensurable ratio to four right angles, then after a certain number of repetitions every point coincides with its original position. 272. Locus of Points of Equal Displacement. (1) The locus of points, for which the distance of displacement [cf. 270 (5)] in the general transformation, parameters (S, a), is equal to a given length a, is the quadric surface 2 2 cosh - + 2p2 cos a = cosh - {2^2 + 2o2j; 7 7 that is (# |#) (cosh----cos a j = I cosh----cos a J (x\a1)(x \a2). Also by referring to 195 (2) we find that all quadrics of the form, remain coincident with themselves after the congruent transformation, axis They are quadrics which touch the absolute at the ends of the axis But by comparison with the quadric, which is the locus of points transferred through a given distance r, we see that the system of quadrics found by varying a is the same* as that found by varying //,. * Cf. Sir R. S. Ball, " On the Theory of Content," Transactions R. I. A., loc. cit. p. 370.
272,273] locus of points of equal displacement. 463 In fact we have cosh----cos a ) i___' [cosh - cos a) (2) The sections of these quadrics by planes perpendicular to axa2 are circles. For let pazaA be any such plane, and let x = p 4- ?3 x3+ 4a4. Then (oo\a1)=^(p\a1), (oc\a2)=t;(p\a2)', and (xa3a4b) = ^(pa^aj)), where b is any fixed point. Hence (w \ a,) (x\a,) = Therefore the section of the quadric, (x\x) + n2(x a1)(x\a2) = 0, by the plane pasa4 is the intersection of the plane with the sphere (3) The centre of this sphere is the point |a3a46, that is the point where meets \b. Hence the centre of the circle, which is the intersection of with the quadric [cf. 260 (4)], is the point where axa2 meets It is otherwise evident from 270 (5), that all points on such a circle must receive equal displacements, and that the distances of their displaced positions from the plane of this circle must be the same for each point. Similarly the curves of intersection of the system of quadrics with planes through axa2, that is of the form paxa2, are lines of equal distance from a^, that is are circles with their anti-spatial centres on a3a4. 273. Equivalent Sets of Congruent Transformations. (1) Any congruent transformation (K) may be replaced by a combination of two transformations consisting first of a translation with its axis through any arbitrarily chosen spatial point b, and then of a rotation with its axis through Kb, where the given transformation changes 6 into Kb. First apply a translation, axis bKb and parameter bKb\ which is always a possible transformation. This translation converts b into Kb. The second trans- formation, which brings all the points into their final positions, must leave Kb unchanged. Hence the transformation must be a rotation with its axis through Kb. (2) Applying the principles of 2 75 below with regard to small displacements, it will be easy to see that any small transformation is equi- valent to the combination of a rotation and a translation with their axes through any arbitrarily assigned point b.
464 KINEMATICS IN THREE DIMENSIONS. [CHAP. VI. 274. Commutative Law. The operations K and | performed succes- sively on any point x are commutative, that is \Kx = K\x. For let et elt e2, e3 be a unit normal system, e being the spatial origin. Let the transformation K have eex as axis, and 8, a as parameters. Then if x = %e + f^ + f;2e2 + f3e3, iT# = (g cosh - + fi sinh - J e + f sinh - + fi cosh - J e1 + (f2 cos a + 3 sin a) e2 + (f3 cos a - 2 sin a) e3. Also [cf. 240 (5)] \e = ieYe2e3, \eY = iee2e3, |e2 = and Ke = cosh - e + sinh - elt Ke1 = sinh - e + cosh - e2, 7 7 7 7 -Ke2 = cos a . e2 sin a . e3, jfiT^3 = sin a . e2 4- cos a . e3. Hence ^^ = KeKe1 = eelf and iT^s = e2e3. Thus | a? = - i (fa -f ^e) e2e3 - i (% - 2e3) ee1; and K \ x = i (^Ke1 + ^iTe) ^2e3 i ( 3Ke2 %JKe3) eex. Now %Ke1 + iKe= [^cosh- + ^sinh-J e1 H- [ sinh- H-^cosh -J e, and ^3Ke2 - %2Ke3 = (f8 sin a 4- f2 cos a) es + (|3 cos a - 2 sin a) e2. Hence by substitution and comparison \Kx = K\x. 275. Small Displacements. (1) Two finite congruent transforma- tions, when successively applied, produce in general different results according to the order of operation. If however each transformation be small, and squares of small quantities be neglected, the order of operation is indifferent. This is a general theorem, which holds for any linear transformation what- ever. Thus let e^e^ be any four reference points, and let x be transformed into Kx by the transformation / = (1 + au) + ajafa + 13 3 + a14f4, ' = Oaifi + (1 + Ofaa) 2 + O^s + Om?4, with two similar equations; where all letters of the form app or apff are small, and their squares and products are neglected. Again, let Kx be transformed into K'Kx by the transformation V = (i + Ai) '+/Saf,' + 1 + *', with three similar equations for f2", 3/r, f4"; where all the letters of the form /3PP or /3p r are small and their squares and products are neglected. Substituting for /, 2', etc. in the equations for /', 2", etc. we find l" = (1 + U + Al) f 1 + ( with three similar equations.
274 276] SMALL DISPLACEMENTS. 465 It is obvious from the form of these equations for the co-ordinates of KfKxy that K'Kx = KK'x. (2) Let a point x with reference to the normal system e, elt e2i e3 (spatial origin e) be written in the form rje + 77^ 4- 77^2 + ^3^3. Apply a congruent transformation, axis eelf parameters 8 and a. Let x become Kx. Then, as in 269 (4), Kx = f 77 cosh - + 77! sinh - ] e -f (w, cosh - + 77 sinh - ) ex V 7 7/ V 7 7/ + (773 cos a + 773 sin a) e2 + (773 cos a rj2 sin a) e{. Assuming that 8 and a are small, this equation becomes / S \ / S V 7 / V 7 (3) Accordingly if in any order small translations 8lf 82, 83 and small rotations alf Oa, a3 be applied with axes eeu ee2, ee3 respectively, and as the total result x becomes Kx, then Jvw [ 77 -\- 771 h 772 H 773) e 4" I Vi H~ V ~^~ 3V2 V 7 7 7 / \ 7 -2 )e ( - XU ryV l 3 7 2 V 8 7 2 ' ' / 276. Small Translations and Rotations. (1) The result (K) of the three small translations of the preceding article by themselves is a 8 8 8 translation, havinef as axis the line joining e and ex + e + 3 e3} and as 7 7 " 7 Si , So S3 parameters V[Si2 +S22 +S32} and 0. For let . 7 7 7 Then Ke = e + d, and ^Td = ------\------ e + d. Hence every point on ed is transferred to another point on ed; and any plane of the form edx is semi-latent. Therefore the resulting displacement is a translation with ed as axis. Let 8 be the parameter. Then 8 8 - = sinh - = ^{ (eKe \eKe)\ = V (ed\ed) 7 7 (2) The result (jST) of the three rotations by themselves is a rotation, having as axis the line joining e and a^ + a2e2 + a3^, and as parameters 0 and For let a = 01 + a2e2 + a3e3. w. 30
466 KINEMATICS IN THREE DIMENSIONS. [CHAP. VI. Then Ke = e, and Ka a. Hence every point on ea is unaltered by the resulting transformation. Therefore the transformation is a rotation with ea as axis. To find its parameter, calculate the angle a between the planes ieexa and iK {ee^a) ; this is the required parameter. Let A = iee a. Then KA = iKex. Kea = i (e1 a3e2 + a2e3) ea. Also (A \A) = -(af + af) = (KA \KA). Hence AKA = [el (ex - a3e2 + a2e3) ea} ea = i (a^ + a32) ea= i(A\A) ea. (3) By properly choosing Slf S2, Ss the line ed, which is the axis of the translation, can be made to be any line through e, while at the same time the parameter V^i2 + S22 + $2} can be made to assume any small value. Similarly the axis of rotation ea can be made to be any line through e by properly choosing alt a^ 3, while at the same time the parameter y/la^-f a22 + a32} can be made to assume any small value. Hence it follows from 273 (2) that the combination of the three translations along eely ee2) ee3 and the three rotations round the same axes may be made equivalent to any small congruent trans- formation whatever. 277. Associated System of Forces. (1) Let 8 denote the system of forces 6 7 7 7 17 7 7 Then it is immediately evident by performing the operations indicated [cf. 99 (7) and 240 (5)] that* (2) Similarly if P be any plane, then KP =P + i\.P\8. For let P = \p. Then Kp=p + i \pS. By taking the supplement of both sides of this equation Thus the system \S bears the same relation to the transformation of planes that the system S bears to the transformation of points. (3) It follows that, corresponding to every theorem referring to systems of forces, there exists a theorem referring to small congruent transformations. The system S will be called the associated system of the transformation. Also since 8 completely defines the transformation, it will be adopted as its name. Thus we shall speak of the transformation S. * This formula has not been given before, as far as I am aware.
277, 278] PROPERTIES DEDUCED FROM THE ASSOCIATED SYSTEM. 467 278. Properties deduced from the Associated System. (1) If the X associated system be (with the previous notation) e2es + ezex + eYe2i the transformation is a translation along the line e (S 4- S^e2 + S3ez) of para- meter V{Si2 + 822 + S32}. Hence if the associated system be the single force F, where .Pis anti-spatial, the transformation is a translation of axis \F and of parameter .y/{F\F}. If the associated system be the single force (a1ee1 + a^ee2 + asee3), the trans- formation is a rotation round this line of parameter V{ai2 + 22 + 32}- Hence if the associated system be the single force Fy where F is spatial, the trans- formation is a rotation round F of parameter */{ F\F}. (2) Hence the condition that the transformation S be a translation or a rotation is (SS) = 0. The additional condition, that it be a translation, is that (S\S) be positive, and the additional condition, that it be a rotation, is that (S\8) be negative. (3) The system S can be reduced [cf. 160 and 162] to two forces F + F\ in such a way that either one force is in a given line, or one force passes through a given point and the other force lies in a given plane. If both F and F' are spatial, the transformation has been reduced to two rotations round the two lines. If one (or both) of the forces be anti-spatial, instead of a rotation round the line of that force a translation along its polar line must be substituted. Since the given point may be assumed to be spatial, and the pole of the given plane may be assumed to be a given spatial point, it follows that it is always possible to reduce a small congruent transformation to a rotation round an axis through one given spatial point and a translation along an axis through another given spatial point. The two given points may be chosen to coincide. (4) The axis of the transformation is the axis of the system S. Let F be a force on this axis, then If F = Xeel + fiee2 + vee3 + we + pe + aexe2\ then the condition, X'ST + fip + V J = 0, must be fulfilled. Also S = (0\+ / ot) eej + (dfi + f p) ee2 + (0v + j r) ee3 + ( zr - f \) e^s + (Op - $ H ) e3el + (0a - f v) e . Hence with the previous notation, 0\ + f 'SF = a1, 0/jl + j p== a2, r 7 7 7 30 2
468 KINEMATICS IN THREE DIMENSIONS. [CHAP. VI. 0OLX (j 0-------- j)OLi Therefore X = ^---- % , w = -% 77-, with four similar equations; where the ratio of 0 to f is given by and a and 8 are put for VK2 + s3 + 32} and V{^i2 + g22 + S 2}. The equation of condition gives two ratios for 0: j , and hence two lines are indicated as the axis. One of these is F and the other is \F; the spatial line is the axis. The equation of condition for 0 : f can be given in another form. For S|5) = -aM.^, and (88) = - 2i \"A + ^ Therefore 20 f (8\8) + ( f 2 - 6'2) i (SS) = 0. (5) The character of a small congruent transformation will be conceived as completely determined by its axis and the ratio of 0 to f ; the remaining constant simply determines its intensity. In other words, the two transfor- mations, of which the associated systems [cf. 160 (1)] are 8 and \8, are of the same character with their intensities in the ratio 1 : X. The performance of a given transformation X times increases the intensity in the ratio X: 1. For let S be the associated system. Then Kx = x + i \x89 K*x = Kx + i \KxS=Kx + % \xS = x+2i \a;8, neglecting squares of small quantities. Hence Kkx = x + Xi\ xS. 279. Work. (1) Let any two spatial points a and b on a line ab be transformed into Ka and Kb, let the angle between aKa and ab be \fr and that between bKb and ab be ^, then . ,^ffa , . .bKb smh------cos-ur = smh-----cos v 7 ^ 7 X ____ or aKa cos y(r = bKb cos %, since the transformation is small. To prove this proposition, notice that . h l-(aKa\aKd) _ *J{-(aKa\aKa)} _ I - (a ~V (a\a y ~V (a\a)(Ka\Ka)~ (a\a) j * (aKa I ab) and cos 0 = u L^-s. ^ ; 7, , , . s/{(aKa \aKa) (ab \ab)\ u , aKa A (aKa lab) Hence sinh-----cos 0 = xv f ' . 7 (a a) V{-(oft a6)
279] work. 469 Therefore we have to prove that --(aKa \ab) = -(bKb\ab) (a | a) Jb\b) ' Let the associated system of the transformation be 8, where S = X(ab + cd). Then Ka = a + i \aS = a + Xi \acd, and aKa = iXa | acd. Hence (aKa \ab) = (ab \aKa) = iX (ab . acd \a) = iX [ab {(a \a) cd -f (d |a) ac + (c | a) daj] = i\ (a | a) (abed) = i (a |a) (abS). mi r (aKa\ab) ., , ,Y, (bKb\ab) Therefore v , \ J = ^ O tf) = Wrjr-^ (a \a) We notice that if 8 be the associated system of the transformation ir t tTtti i(abS) aKa cos vr = 6iL0 cos v = . / 7 , -, .T . ^ s/{ (ab \ab)\ (2) Definition of Work. Let any point a on the line of a force of intensity p be transformed by a small congruent transformation to Ka, so that aKa makes an angle -\/r with the force, then p. aKa. cos sfr is said to be the work done by the force during the transformation. It follows from the previous proposition that the work done by the force is the same for all points on its line. (3) Let b be another point on the force (F) so that F = ab. Then Hence by (1) the work done by F during the transformation S is (4) Let the work done by any system of forces S' be defined to be the sum of the works done by the separate forces of the system. Thus let Then the work done by 8' during the transformation 8 is - i (FS) - i (F'S) - i (F"S) - etc., but this is -i(SS'). We notice (a) that the work done by a system of forces during a small congruent transformation is the same however the system be resolved into component forces; and (/3) that the work done by the system S' during the transformation XS is equal to the work done by the system $ during the transformation XS'] where X is small, but the intensities of 8 and Sr are not necessarily small. If two systems be reciprocal, that is if (SS;) = 0, then no work is done by either one during the transformation symbolized by the other.
470 KINEMATICS IN THREE DIMENSIONS. [CHAP. VI. 280. Characteristic Lines. (1) Let the line joining any point with its transformed position, after a small congruent transformation, be called a characteristic line of the point: and let the line of intersection of a plane with its transformed position be called a characteristic line of the plane. Thus if x be changed to Kxy the line xKx is the characteristic line of the point x; and if the plane P be changed to KP, the line PKP is the characteristic line of the plane P. (2) Let L be any line, and KL its transformed position, and let L intersect KL, then L is the characteristic line of some point and also of some plane. For let L and KL intersect in Ka, and consider the points a and Ka. Since Ka lies on KL, then a lies on L, hence L = aKa. Thus L is the characteristic line of the point a. Also consider the plane P = K~Aa. a. Ka, then KP = aKaK2a. Hence aKa (i.e. L) is the characteristic line of the plane P. (3) If S be the associated system, the characteristic line of any point x is x \xS, and the characteristic line of the plane P is P . P \S. (4) The locus of points x on the characteristic lines, which pass through a given point a, is given by (axKax) = 0. This is a quadric cone through the point a. The equation can also be written (aKa. xKx) = 0. Hence the characteristic lines of the points x are those characteristic lines, which intersect the characteristic line aKa. The equation can also be written (a \aS .x \xS) = 0; that is [cf. 167 (3)], (axS) (ax 18) - \ (SS) (ax \ ax) = 0. (5) Similarly if AP be a characteristic line lying in the plane A, then the planes P envelope the conic (APKAP) = 0, which lies in the plane A ; that is to say, the characteristic lines in the plane A envelope a conic. The equation can also be written (AKA . PKP) = 0. Hence the planes P are such that their characteristic lines intersect the characteristic line AKA. The plane-equation of the conic can also be written (APS) (AP \S) - %(SS) (AP AP) = 0. 281. Elliptic Space. (1) The Kinematics of Elliptic Space can be developed in almost identically the same manner as that of Hyperbolic Space [cf. 268], only with a greater simplicity. The absolute quadric being now imaginary, the four corners of the self- corresponding tetrahedron in any congruent transformation must also be imaginary.
280, 281] ELLIPTIC SPACE. 471 Let au a2, a3, a4 be these four corners. Let the lines axa2 and a3a4 be real conjugate polar lines. Then ax and a2 are conjugate imaginaries, and so are a3 and a4. Thus let ex and e2 be two real quadrantal points on axa2, and let e3 and e4 be two real quadrantal points on a3a4. Then iff in in- _iv in in in in aY e *ei e4e2, a2-=e* e1 e 4 e2 a3 = e * e3 e* e4, a4 = e4e3 e 4 e4. The transformation changes ax into e~iaaly a2 into eiaa2, a3 into a4 into elPa4. Hence any point x = 77^ 4 v*^ 4 i)3e3 4 ?74e4 is changed into ifa? = (77! cos a 4 772 sin a) ex 4- (772 cos a r}i sin a) e2 4 (rj3 cos /3 4 *74 sin/8) 63 4 (^4 cos /3 773 sin^S) e4. Thus (JK'a; I iT^) = 77^ 4 77./ 4 ?732 4 ^7/ = (x x). And (x I if a?) = (77^ 4 77./) cos a 4 (77^ 4 77/) cos fi. (2) Thus any point 77^ + 772e2 on eYe2 is transferred through a distance 7a, any point on e3e4 through a distance y/3. Similarly any plane through exe2 is transferred through an angle ft, any plane through e3e4 is transferred through an angle a. (3) The distance of Kx from e^ is S, where i \eYe2Kx) . S _ / (exe2Kx \eYe.2 sin But (e Kx le Kx) (r}3 cos /3 4 Vi ^n fif 4- (^/4 cos /3 773 sin /3)2 = 7732 4 ^42. Hence sin 8 sin - = A / ' 7 V Thus the distance of ifo; from ^^ is the same as that of x from e^. Similarly for the distances from e3e4. (4) Let a and be called the parameters of the transformation. The transformation will be described as the transformation, axis exe2, parameters a, /3, or as the transformation, axis e3e4 parameters /3, a. The transformation, axis e^, parameters a, 0, will be called the transla- tion, axis 6^2, parameter a; or else, the rotation, axis e3e4, parameter a. (5) Any congruent transformation may be conceived as the combination of a rotation round and a translation along the same axis; or as the com- bination of two rotations round two reciprocally polar lines; or as the combination of two translations with two reciprocally polar lines as axes. The distinction between translations and rotations, which exists in Hyperbolic Space, does not exist in Elliptic Space.
472 KINEMATICS IN THREE DIMENSIONS. [CHAP. VI. 282. Surfaces of Equal Displacement. (1) The locus of points, for which the distance of displacement in the transformation axis ^3, parameters a, /?, is y rt is the quadric surface (x\Kx) Hence employing the notation of the previous articles, (V + V22) cos a 4- (vs2 + V2) cos 0 = cos a (rjf + v*3 + Vz + V^)- Therefore (V -f iy22) (cos a cos r) + (t^2 + rjf) (cos /3 cos a) = 0. (2) The system of quadric surfaces found by varying a is also easily shown to be the system which remains coincident with itself during the transformation. Its sections by planes through exe2 or e3e4 (i.e. by planes perpendicular to e3e4 or exe^) are circles. 283. Vector Transformations. (1) An interesting special case discovered by Clifford* arises in Elliptic Geometry which does not occur in Hyperbolic Geometry. Let a congruent transformation, of which the parameters are numerically equal, be called a vector transformation. Thus, with the above notation, = + . Hence [cf. 282 (1)] any point x is transferred through a distance 7a. Accordingly in a vector transformation all points are transferred through the same distance, and similarly all planes are rotated round the same angle. (2) Again, the line xKx is parallel to the axis exe2. For, taking a = /3, xKx = (rji2 -f rj2~) sin a. eYe. (i]s2 -f 7/42) sin a . e3e4 + (^4 - V3V2) sin a. ^63 - (rjtfz + r)2r)4) sin a . e^ + (v-tVi + ^1^3) sin a. e2e^ + (^^ rj2r)3) sin a . e 2e4 = - (Vi2 + V22) sin a . e^2 - (i/8a + rj^) sin a. | e^2 + (^1^4 - V2V2) sin a {e - |^8) + (772^4 + ViVs) sin a (g2e3 - \e2e3). Hence ^iTa? + \xKx = - (V -f- v 22 + ^32 + ^42) sin a Similarly if a = y3, #if# |^j5T^ = (# I a?) sin a {exe2 Thus [cf. 234 (2)] if a, a be the parameters of the vector transformation, all points are transferred along right-parallels to exe2\ and, if a, a be the parameters, all points are transferred along left-parallels. Let these transformations be called right-vector transformations and left- vector transformations respectively. It follows therefore that any one of the lines parallel to the axis of a vector transformation may with equal right be itself conceived as the axis. * Cf. Clifford, Collected Papers, Preliminary Sketch of Biquateimions, loc. cit. p. 370.
282 285] ASSOCIATED VECTOR SYSTEMS OF FORCES. 473 284. Associated Vector Systems of Forces. Let jR be the unit right-vector system of forces e^-f \e^e.2 [cf. 236 (1)], then the right-vector transformation, axis e^ and parameter a, can be represented by Kx = x cos a sin a \xR. This representation, unlike the preceding formula of 277 and the subse- quent formula of 286, is not confined to the case when a is small. Let R be called the associated unit system of the right-vector transformation, axis e^. Similarly if L be the unit left-vector system exe2 \exe2i then the left-vector transformation, axis e 2 and parameter /3, can be represented by K'x = x cos /3 -f sin /3 j xL. Let L be called the associated unit system of the left-vector transformation, axis #! ,. 285. Successive Vector Transformations. (1) If a right-vector transformation and a left-vector transformation be successively applied, the result is independent of the order of application *. For let K and K' be the matrices denoting the vector transformations of the last article only with different axes, namely e^ and e/e/. Then KK'x = K'x cos a - sin a (K'x . R) = x cos a cos fi + cos a sin ft j xL sin a cos ft \ xR sin a sin fi |( \xL . R). Also K'Kx = Kx cos fi + sin fi\(Kx. L) = x cos a cos # sin a cos/31#22 + cos a sin /3 \xL sin a sin /3\(\xR . i). But |( \xL .R) = \\xL.\R = -ocL.R] and similarly |( \xR .L) = xR.L. But, by 167 equation (25), since (RL) = 0 [cf. 235 (4)], it follows that (xL . R) + (xR. X) = 0. Hence finally, KK 9x = ^'ifa?. (2) Again, let i and R! be both of the same name, say both unit right- vector systems of the forms e^ + U^, tfiV + ki'^'; and let K and K' be the corresponding matrices. Then K'Kx = x cos a cos /3 cos a sin /3 \xR' cos /3 sin a \xR + sin a sin /3 !(\xR'. R). Now KI^U' ,R) = -xR'.\R = -xR'.R = - (RR')x + ## . #'. Hence K'Kx = # cos a cos cos a sin /3 j#ft' cos sin a #i2 (RR') sin a sin /3. # + sin a sin /3. -.R. R\ * Cf. Sir R. S. Ball, loc. cit. p. 370.
474 KINEMATICS IN THREE DIMENSIONS. [CHAP. VI. Similarly KK'x = x cos a cos ft cos a sin ft \ xR' sin a cos ft xR sin a sin ft. xR. R'. Hence K'Kx is not equal to KK'x, Thus two vector transformations of the same name (left or right) applied successively produce different results according to their order*. (3) The resultant transformation, which is equivalent to two successive vector transformations of the same name, is itself a vector transformationf of that name. Let R and Rr be the two unit associated right-vector systems with any axes, and let a and ft be the respective parameters. Then it is proved in the preceding subsection that KK'x = x cos a cos ft cos a sin ft \xR' sin a cos ft1xR sin a sin ft. xR. R'. Let Kxx be written for KK'x. Also let x be a point at unit intensity, so that (x \x) = l. Then (x \ KYx) = (x | x) cos a cos ft sin a sin $ [x | (xR . jB')}. Now x\(xR.R') = x.(\xR.\R') = x.(R\x.R') = x.R\x. R'. Again R = e1e2+ \e1e., = xp + \xp, where p is the unit point on the right- parallel to exe2 through x and normal to x, so that (x \p) = 0. Similarly R' = eie2' + \e-[el xp' -f- \xp\ where p is a similar point such that xp' is a right-parallel to e/e/. Then i2 !a? = xp \x = (a? U')^ (x \p) x = p. Hence x \ (xR. R) = ay (ap' + | xp') = (ap ] xp) = (RR). Thus (x | if^) = cos a cos /3 J (jRi?') sin a sin ft. Accordingly, if y r be the distance through which x is transferred, remembering that (^a? |^a?) = (x \x) = 1, we find cos a = cos a cos ft \ (RR') sin a sin ft. Therefore the resultant distance of displacement of x to Kxx is the same for all positions of x. Therefore [cf. 283 (1)] the transformation is a vector transformation of which the parameter is r. The proof is exactly similar if the two component transformations are left-vector transformations. (4) It now remains to be proved that the resultant transformation is of the same name (left or right) as the component transformations. The method of proof will in fact prove the first part of the proposition also. It is easy to prove that, with the notation of the previous sub-section, KK'x = cos a. x cos a sin ft .p' sin a cos ft. p + sin a sin ft \ (xp'p), and KKx = cos a . x cos a sin ft .pf sin a cos ft. p + sin a sin ft \ (xpp'). * Sir R. S. Ball, loc. cit. t Cf. Sir R. S. Ball, loc. cit.
285] SUCCESSIVE VECTOR TRANSFORMATIONS. 475 Then, if K, stand for KK\ we find xKxx = sin a cos ft . xp sin ft cos a. xp + sin a sin #. \p'p, since # Kaj/p) = (# |p) |#p' + (a? \p') \px + (# |#) \pp = |p'p. Now let y be any other point, and let q and g' stand in the same relation to y as do p and pf respectively to x. Then yK-iy = sin a cos ft . yq sin /8 cos a. i/^' 4- sin a sin $. | q'q. But p + l !P = yg + |y?, p' + |a?p/ = y?/ + |y?/, and by 236 (2) pp' + \pp' = g^ + [ggf. Hence ^if^ + \xKxx = t/^y 4- \yKYy. Therefore yK^y and ^A^^ are right-parallels, which was to be proved. (5) These theorems, due to Sir Robert Ball, have been proved analytically in order to illustrate the algebraic transformations. They can however be more easily proved geometrically. For consider two vector transformations of opposite names applied successively. Let the right-vector trans- formation transfer a to b, and the left- vector transformation transfer b to c [cf. fig. 1]. Complete the parallelogram abed [cf. 237 (3)]. Then the left-vector transformation transfers a to d, and the right-vector transformation transfers d Fig. l. to c. Hence in whatever order they are applied the same ultimate result is reached. But [cf. fig. 2] if any other point a' is transferred by the same Fig. 2.
476 KINEMATICS IN THREE DIMENSIONS. [CHAP. VI. combination of transformations successively to b' and to c', ac is not parallel to ac, since ab and ab' are not parallels of the same name as be and b'c [cf. 237 (4)]. Hence the combination is not itself a vector transformation. Consider again two vector transformations of the same name, say both right-vector transformations, call them Tx and T2, Let Tx transfer a to b, a to V; let T2 transfer b to c and b' to c' [cf. fig. 2]. Then ab=a'bf and bc=b'c\ and the pairs of lines are parallels of the same name. Hence by 237 (4), a'c' and ac are equal and parallels of that name. Hence the resultant of the combination is itself a vector transformation of the same name. But a parallelogram cannot be formed in which the opposite sides are parallels of the same name. Hence the combination of Tx first, T2 second, gives a different result from that of T2 first, Tx second. 286. Small Displacements. (1) The theory of small displacements in Elliptic Space is the same as that in Hyperbolic Space [cf. 275 et Let S = a^exe2 4- a^e^ + a^e^ + aue. + a13e4e, 4- where the coefficients a12, etc., are small. Then Kx = x + | xS gives Kx = ( - otuf2 - a13 3 - a14 4) + ( + aMfi + Oafs + aw^4) e3 + (f4 + a14?i 4- a42f2 + au^3) e4. This is the most general type of small congruent transformation. (2) Also K]x = \Kx. Let P be any plane, then \KP = K\P = \P-\(\P.S). Hence KP = - \K \P = P + ||(|P. S) = P- |(P|S). Thus | S bears the same formal relation to the transformation of a plane P, as S does to that of a point p. (3) It follows, as in Hyperbolic Space, that corresponding to every theorem referring to a system of forces there exists a theorem referring to small congruent transformations. The system 8 will be called the associated system, and S will be used as the name of the transformation. (4) Thus if 8 = F+ F\ the transformation 8 is equivalent to two rota- tions round F and F\ of parameters *J(F\F) and %J{F'\Fr) respectively. Every transformation possesses two axes, which are found as in Hyperbolic Space. (5) Also [cf. 160 (2)] if b be any point, S = pb + P\b. Thus the transformation is equivalent to a rotation, parameter ^(pb \pb), round the axis pby and a rotation, parameter y/(P \b |. P \b)f round the axis P \b : this last rotation can also be described as a translation, parameter *J(P \b \.P \b), along the axis b P.
286] SMALL DISPLACEMENTS. 477 (6) Let any two points a and b on a line ab be transferred to Ka and Kb; let the angle between aKa and ab be 0, and that between bKb and ab be ^ , then . aKa . bKb , sin -----. cos 0 = sin-----. cos p, 7 7 or, since the transformation is small, aKa. cos 6 = bKb . cos . This theorem is proved (as in Hyperbolic Space) if we prove that (aKa\ab)_(bKb\ab) (a\a) (b\b) ' Now let the associated system 8 be written in the form X (ab -f cd). Then Ka = a + \a8=a + \\ acd, and aKa = \a\acd = X(a\d)\ac + \(a \c) \da -f X(a\a) \cd. Hence (aKa | ab) = (ab \ aKa) = X (a | a) (abed). mi r (aKa\ab) ^ . T ,. (bKb\ab) Therefore -7 r \ = ^ (abed) = v /7 .' 7 . (a I a) v 7 (616) (7) The definition of work is the same as in Hyperbolic Space. The work done by a force is (by the last proposition) the same for all points on its line. If \8 be the associated system of a transformation, where X is small and S is not necessarily small, and S' be any system of forces, then the work done by 8' during the transformation \S is \(SS'). This is equal to the work done by S during the transformation \8'. The proof of this theorem is exactly as in the case of Hyperbolic Space. If two systems S, S' are reciprocal, so that (SS') = 0, then the work done by 8 (or S') during the transformation \S' (or \S) is zero.
CHAPTER VII. Curves and Surfaces"'. 287. Curve Lines. (1) Let the Space be Elliptic (polar) and of three dimensions, though the resulting formulae will in general hold for Hyperbolic Space. Let any point x be represented, as usual, by (^ + 2e2 + $ 3+ fA, where the co-ordinate points ely e2, eSf e4 form a unit normal system. Now let the co-ordinates of x, namely f1? f2, %* De functions of some variable t. Then, as r varies, x traces out a curve line. When t becomes t + Sr, where St is indefinitely small, let x become Then obviously x + 262 + f3e3 + 404. Let xx stand for x+x8t. Let xy x etc. be derived in regular sequence by the same process as x is derived from x. (2) Now as x changes its position to xi9 it might change its intensity as expressed in the above notation. Let it be assumed that the intensity of x remains always at unit intensity. Hence (x\x) = l, (x\x) = 0, (x \x) +(x \x) = 0, and so on, by successive differentiations. Hence x is a point on the polar plane of x. Also the same variation of r to t -f St, which changes x to xlt will change xl to #2, where x2 = xx 4- x^r = x + 2xBt -f x (St)2, and will change x2 to x3, where x2 = x + 3i?ST + Sx (8t)2 + a? (St)3, and so on. (3) Let 8 r denote the length of the arc xxlf then . Sa 8a I (, sm = = a / / 7 7 V (* = V{(# |#) (# !^)2} St = *J{x \x) St. mi_ p ^ " // i \ FlQ. 1. 1 here fore -r- = 7 v(^ ^)- " This application of the Calculus of Extension has not been made before, as far as I am aware.
287, 288] curve lines. 479 (4) Again, let 8e denote the angle of contingence of the curve at x corresponding to the small arc 8a. Then 8e denotes the angle between xxx and xxx2. Hence by considering the triangle xxxx2y we find [cf. 216 (2)] / = A / 7 V (X Se = sin Se - A , , , w , . Now xxYx2 = xxx (St)3 ; and (x^21xYx2) = (xx1 \xxj) = (xx \xx)(St)2, ultimately. XT ~ J(xxx \xxx) ~ J(xxx \xxx) c Hence Se = \ . ,' or = ^.-'.y - dr. (## |##) (x \x) Also we may notice [cf. subsection (2)] that (xxx xxx) = 1, 0, (x\x) 0, ( *!*), ( *|*) (x \x\ (x x)} (x \x) = (x \x)(x \x) - (x \x)2 - (x [xf = (xx \xx) (x \xf. tt * \/{(xx \xx) (x \xf\ ^ Hence Se = ^ rVS (5) The tangent line at x is the line xxly that is, the line xx. The normal plane at x is the plane through x perpendicular to xxY. This plane is the plane x \xxlt that is, x \xx. Now x \xx = (x \x) \x (x \x) \x = \x. Hence \x is the normal plane of the curve at x. The normal plane at x^ is therefore \x + x8t. Monge's ' polar liney of the curve at x is the line of intersection of the two planes, that is, the line \xx. (6) The osculating plane of the curve at x is xxxx.ly that is, the plane xxx. The neighbouring osculating plane at xY is xxx + xxxBr. The angle between these two planes, namely, 80, is the angle of torsion corresponding to the arc 8cr. If the first plane be P and the second plane be Q, then PQ = xxx. xxx8t = (xxxx) xx8t. Hence ~n . /, / (PQ|PQ) (xxxx) *J(xx \xx) ~ (xxxx) \f(x \x) V (P | P) (Q | Q) (^^iT \xxx) (xxx \xxx) 288. Curvature and Torsion. (1) Let the 'measure of curvature' or the ' curvature' of the curve be defined to be the rate of increase of e per unit length of a, and let it be denoted by - .
480 CURVES AND SURFACES. [CHAP. VII. Let the ' torsion' of the curve be defined to be the rate of increase of 9 per unit length of a, and let it be denoted by . m, 1 de _ 1 *J(xxx \xxx) __ 1 *J{(xx \xx) (x \x)3} p da y (x \xy y (x \xy 1 _ dd _ 1 (xxxx) Hence k da y (xxx \xxx) ' 1 1 (xxxx) The condition, that a curve is plane, is - = 0; that is, (xxxx) = 0. tc (2) Newton's geometrical formula for the curvature still holds. For let 8(7 be the distance between the points x and x + 8x, and let Sv be the perpendicular from x -f S# on to the tangent aac. Then [cf. 226 (1)] ^ = J{xfX \Xffi. L 3 v /J 7 V (xx \xx) But 8x = x mi r 1 Kb X \XXX) ' , /(^^ |flK ) / v Therefore oz = iy A / , . -^ . (ot)2 = I7 A / - 7. , .. -- . (ot)2. 2/V (a 6|aa ) V (x\x) v 7 w Sz/ __ 1 1 I (xxx \xxx) 1 Mence (8^=27V (i|^)3 =V (3) The normal planes envelope the * polar developable/ of which the edge of regression is the locus of intersection of three neighbouring normal planes. The polar line \xx is a tangent to this edge of regression at the point corresponding to x. The three normal planes at x, xly x2 are \x, \x + \x.hT, \x + 2\x.St + \x.(St)2. Hence the point on the edge of regression of the polar developable which corresponds to x is \xxx. Let the distance, plf of this point from x be called the radius of spherical curvature. Then cos ^ = 7 \J(xxx \xx"x) " (4) The polar line \xx meets the osculating plane in a point called the 1 centre of curvature/ This point is xxx xx. The distance of this point from x is called the radius of curvature which is not in general equal to the inverse of the measure of curvature except in parabolic geometry.
289] CURVATURE AND TORSION. 481 In order to find the radius of curvature, which will be called p2, we notice that \xx is perpendicular to xxx. For the line through the point \xx"x perpendicular to xxx is the line \xxx. \xxx = \(xx"x. xxx) (xxxx) \xx. Hence neglecting the numerical factor, this is the line \xx. Therefore p2 is really the distance of x from the polar line \xx. But the distance of x from any line F is Now x \xx = (x \x) \x (x \x) \x = (x \x) \x. And (\xx. \\xx) = (xx \xx). . p2 /(x\xf(x\x) (x\xf Hence sin = A /v . .. -~- = .; -., . 7 V \xx xw) \/(xx \xx) Therefore ^ = cosec2 - ] = cot2 ^ . P2 72 7 Hence P = J t'an (5) The principal normal is the normal line in the osculating plane, that is the line, xxx \x = (x \x) xx -f (x \x) xx + (x \x) xx = (x \x)xx -f (^ \x) xx. The binormal is the normal perpendicular to the osculating plane, that is the line x xxx. It will be useful later to notice that the intensity of xxx x is \]\{x \x) (xxx \xxxj\. For if P and Q be two planes, then (PQ\PQ) = (P\P)(Q \Q) - (P\Q)2; and by writing xxx for P and |i? for Q the result follows. 289. Planar Formulae. (1) The complete duality both of Elliptic and of Hyperbolic Space allows formulae similar to the above to be deduced from the plane equation of a curve. Let P denote the plane Vi^3^4 ¥i¥j + ¥i^4- v^e^e^. Let v1} v2, vZy vA be functions of one variable r. Then the plane P, as it moves, envelopes a developable surface, of which the curve under consideration is the edge of regression. Let P1 = P + P r, where P v^e^ + etc., and let P be derived from P as P is from Py and so on. Let P2 = P + 2P . Sr + P (St)2, P3 = etc. Let (P |P) = 1, so that (P|P) = O, (P|P) + (P|P) = O. (2) The point on the curve corresponding to P is the point PPjP2, that is the point PPP. The tangent line in P is the line PP. w, 81
482 CURVES AND SURFACES. [CHAP. VII. The angle of torsion $0 is given by (3) The angle of contingence Se is the angle between the lilies PPi and 2. Hence by considering the stereometrical triangle P, P1? P2 and deducing the formula from that for a point-triangle by the theory of duality, we find Se-sinSe- /- (W.IW - sin - ^ {pp-\ppj{pP (p ]P) (4) The length of the arc So- is obviously found from the analogous formula to that which gives the angle of torsion in the point-equation. Thus Sa _(PPPFU(PjP) 7" (PPP\PPP) St- da y(PPPP) ati~(PPP\PPPy do- y(PPPP)(P\Pf Hence p = ^ = (PPP\PPP)S ' do- y(pppp) (ppp\pppy (5) The point \P is a point in the plane P, such that any line in P through \P is perpendicular to the tangent line PP. For the point, where |PP meets P, is the point P|PP= (P |P) |P-(P|P)|P = -(P |P) |P, and this is the point \P, neglecting the numerical factor. Thus \P is the point on the principal normal distant ^7ry from the point PPP. The normal plane is the plane PPP |PP, this can easily be shown to be the plane (P |P) P + (P\PyP-(P \P) P. 290. Velocity and Acceleration. (1) Some of the main propositions respecting the Kinematics of a point in Elliptic or in Hyperbolic Space can now be easily deduced. It must be remembered, that in such spaces the idea of direction, as abstracted from the idea of the fully determined position of a line, does not exist. In Euclidean Space all parallel lines are said to have the same direction. Again, in considering the small displacements which a moving point has received in a small time Sr, we must remember that
290] VELOCITY AND ACCELERATION. 483 the propositions of Euclidean Space are applicable to infinitely small figures in Elliptic or Hyperbolic Space. As in the case of curved lines the reasoning will explicitly be confined to Elliptic Space. (2) Let the variable r of the preceding sections be now considered to be the time. Then the point x in the time St has moved to the position a?-f xBt. The line joining these points is the line of the velocity, and the length of the arc traversed is its magnitude. Hence the linear element xx represents the line of the velocity, and its intensity represents the magnitude of the velocity. Hence xx may be said completely to represent the velocity. Hence at a time t + 8t the linear element (x + x8t) (x + xSt), that is xx + xx. St, represents the velocity. Therefore, remembering that the propositions of Euclidean Space can be applied to the infinitely small figures which are being considered, the linear element xx completely represents the acceleration. (3) It is also obvious, from the applicability of the ideas of Euclidean Space to small figures, that two component velocities vx and v2 along lines at an angle co are equivalent to one resultant velocity of magnitude V(vi2 + V2 + 2^2 cos w), making an angle 6 with the line of v1; where Vi _ V2 _ V sin ( 6) sin 6 sin ay' The same theorem holds for accelerations. Thus if xx and xx1 denote two component velocities of the point x, then xx-\-xx represents the resultant velocity [cf. 265 (4)]; and if xx and xx' represent two component accelerations, then xx+xx represents the resultant acceleration. (4) The magnitude of the velocity is , or y s/(x\%). It will now be shown that the acceleration is equivalent to two components, one along the (x \x) tangent of magnitude a, or 7 ).. , . , and the other along the principal normal of magnitude . For a unit linear element along the tangent is . ; also [cf. 288 (5)] i\J\X\X) Tver I or a unit linear element along the principal normal is + /f/. .w .'. rrrri f r ~ *J{(x\x)(ocxx\xxx)) ... . , . (x \x) xx (x \x) xx which can also be written + . ' . , .. . ...,' ...xl . *J{{x \x) (xxx \xxx)\ .. _ (x\x) xx /Uxxx I xxx)) (x I x) xx (x I x) xx Also 7ra_7 _|-- . _^_^ + 7y |___^| ______ XX 2 XXX Id) p \/{(x\x)(xxx\xxx)}' 31 2
484 CURVES AND SURFACES. [CHAP. VII. Hence the acceleration is equivalent to two components as specified above. (5) Let the point x be called the velocity-point of x, and x the accelera- tion-point of x. The velocity of x is directed along the line from x towards x and the acceleration towards x. Furthermore, if v be the magnitude of the velocity and a of the acceleration, then, v = y \/(x\x); and a = 7 *J(xx \xx) = 7 aJ{(x \x) (x \x)2} = 7 *J{(x\x) (x \x)2}. Hence 72a2 + v4 = 7* (x \ x). 291. The Circle. (1) To illustrate the formula of 287, 288 con- sider the circle, radius a, in Hyperbolic Space [cf. 216 (6) and 259]. Firstly let the centre be the spatial point e. Let e, ely e2, e3 be a system of normal points at unit intensities, spatial and anti-spatial respectively, and let eexe2 be the plane of the circle. Then if x be any point on the circle [cf. 216 (6)], and f be the parameter r of 287(1), x = e cosh - + eY sinh - cos b + e2 sinh - sin 6, 7 7 7 x = sinh - { e1 sin j + e2 cos 0}, x = sirm - { ex cos f e2 sin f } = e cosh x. Therefore - = 7 V~ (^ I ) = 7 sinh - : d p 7 and cr = (j ysinh ; and the length of the whole circumference is 2^7sinh- . Also xxx = sinh - (e1 cos f -\- e2 sin f ) . sinh - (- e1 sin f -f e.2 cos f ). e cosh - 7 7 7 = eeje2. sinh2 - cosh - . 7 7 Hence ,^ = T( ..' . J = cosh - . aq (x \x) 7 The normal plane is \x = i sinh - {ee2e3 sin f -f ee^ cos ^ 1. All the normal planes pass through the line ee3, drawn through the centre perpendicular to the plane of the circle. The measure of curvature is given by 1 1 sjixxx I xxx) 1 a - _ ^_----- = _ coth . p 7 (-j\xy 7 7
291] THE CIRCLE. 485 Hence 0 = 7 tanh -. 7 When the radius is infinite, the measure of the curvature becomes -, and 7 this is its least possible value; when a is zero, the measure becomes infinite. The binormals obviously all intersect in e3, and form a cone with an anti- spatial vertex. (2) Secondly, let the circle have an anti-spatial centre, e2, and let it lie in the plane eeYe^\ where e, ely e2, es form an unit normal system of points with e as spatial origin. Then [cf. 259 (5)] the circle is a line of equal distance from the line ee1. Let y9 be the distance of a point x on the circle from this line. Then sinh - = / - --- \ / , x = *J(xeel \xee^), when x, e, ex are at unit intensities, spatial and anti-spatial. Hence if x = %e -f % + #2, we have ^2 i2 ^22 = 1, and 2 = sinh . Hence we may put f = cosh cosh f), g! = cosh sinh d . 7 7 A? A? /3 Thus ^ = e cosh - cosh f + e1 cosh - sinh f + e2 sinh - , 7 7 7 /^? ^ = cosh - \e sinh 6 + el cosh f ], R x = cosh \e cosh f -f- ex sinh f } ry l ' ; = x e2 sinh - . 7 The normal plane is given by R \x = i cosh - { e e-s sinh $ e^3 cos 7 7 The principal normal to the curve is the intersection of this plane with ; and it is perpendicular to the axis, since \x contains the polar line of the axis, viz. e2e3. This normal meets the axis ee1 in the point ee1 \x, that is in the point ex sinh + e cosh . Call this the point y, then y = e sinh f + ex cosh j . Let 8 r' denote the distance between y and y + yh f .
486 CURVES AND SURFACES. [CHAP. VII. Then ^ = 7 V" 0* \x) = 7 cosh -, Hence [cf. 262 (3)] ^7 = cosh , r = * ' cosh . Again ### = ee^o cosh2 - sinh . 7 7 , cosh2 - sinh - ~ de 7 7 . , p Hence 7 . = n = smh - , ^ cosh2^ * 7 and - = - tanh . p 7 7 When /3 is infinite, p = 7; and when /3 is zero, the curvature is zero. In this latter case the curve is identical with the straight line which is the axis. (3) Thirdly, let the centre of the circle be on the absolute, so that the curve is a limit-line. Let e be a spatial point on the curve, and let eeY be the normal at e, and ee^ be the plane of the curve ; also let e, elt e2, e3 be a normal system of unit reference elements. Then from 261 (3), the limit- line becomes the section of the limit-surface by the plane ee e2. Thus, if x be of unit intensity, it can be exhibited in the form Then, if 6 be the parameter t of 287 (1), x = de + 0ex + e2i x = e + ex. Hence 3Z = 7 V~ (# I ^) = 7 cr = $7- Thus we can write 1 where a is the length of the arc from e to w. Then writing x' and x" for x and a? with this new variable ay x' = e + -. e1 + - ^2, #" = (e + ej, and ic^V' = - 7 7 7 T T w de 1 1 Hence = - = -. da p 7
292] THE CIRCLE. 487 This agrees with the deduction above, that a circle of infinite radius is of curvature -. 7 Again let y be a point on a second limit-line with the same centre as the first one, and at a constant normal distance 8 from it; so that, if x and y be corresponding points on the first and second limit-lines respectively, the line xy is normal to both curves, and ayy is equal to 8. Then if y be of unit intensity, it is easily proved that J x + sinh - (e 4- eY). Hence if cr be an arc of the x curve and a an arc of the y curve, and if a be the independent variable, Hence -7 = 7 J (y \y) = 7exp y (x \x ) - Hence a = exp ( J. cr. This result is proved by both Lobatschewsky and J. Bolyai. 292. Motion of a Rigid Body. (1) Let S stand for the system and let SBr be the associated system of the transformation which would dis- place the body from its position at time r to its position at time t + Bt. Let any point of the rigid body be x at time t and x + xSr at time t 4- St. Then x + St = x + |#S . St. Hence = +1 xS. Let S be called the associated system of the motion of the body. (2) The theorem of 286 (6) can be stated in the form: the resolved parts of the velocities of two points of a rigid body along the line joining these points are equal. Thus if x, y be the two points, and xx and yy make angles 6 and 0 with xyy then xx cos 0 = yy cos / . (3) The velocity of any point x of the moving body is xx = xS. Hence the velocity of each point is perpendicular to its null plane with respect to S.
488 CURVES AND SURFACES. [CHAP. VII. (4) Again if 8 change to # + 8 . 8r at the time t + 8t, then x = \xS+ \x8= \xS + \x\8 = \x8 + x8\8. (5) It is obvious that all the theorems which have been enunciated with respect to small congruent transformations hold with verbal alterations for the continuous motion of a rigid body. Thus if 8 be a vector system of the form axa2 | x2a3 then \S = S. Also suppose that 8 is constant with respect to the time. Then Thus xx is always zero and no point of the moving body has any acceleration. Thus each point of the body is moving uniformly in a straight line. This is a vector motion of the body. 293. Gauss' Curvilinear Co-ordinates. (1) Let x be any unit point on a surface in Elliptic Space. Then the co-ordinates of x, referred to any four reference elements, may be conceived as definite functions of two independent variables, 0 and / . And the two equations, 0 = constant, l = constant, represent two families of curves traced on the surface. (2) Suppose that the unit point x -f 8x corresponds to the values 6 + 80 and f + 8 f of the variables. Then we may write 8x in the form 8x = x,80 + x28 f + | {xn (80)2 + 2x12808 f + x22 (S )2} + etc. Since the point remains a unit point, we see by making 80 and 8 f infinitely small, and by remembering that the ratio of 80 to 8 f is arbitrary, (x x^) = 0 = (x \ x2). Hence xl and x2 are in the polar plane of x. In order to exhibit the meanings of xly x2j xu, etc., let e^e^ be a set of four unit quadrantal points; and let x = S e. Then xx = X Jk e, x2 = 2 ^- e d2E xu = X^g-2e} and so on ; where the condition, f{- + 22 + 32 + f42 = 1 is fulfilled. It will be an obviously convenient notation to write ^ for xx, and so ou on, when occasion requires it. (3) By differentiating the equations, (x\x1) = 0 = (x\x2), with respect to 0 and / successively, we obtain Oi |#i) = -(x ;#n), (X2 \X2) = - (X |afc), (X1 \X2) = - (X \X12). (4) The distance 8a between x and x + 8x is given by (S\B) ( |)(8 9)3 2 (awj | = sin2 = sin 2 (^ |
293, 294] gauss* curvilinear co-ordinates. 489 (5) The tangent line to the curve joining the points x and x -f 8x is x8x, that is x (x^O + x28$). Hence xxx and xx2 are two tangent lines to the surface at x, and therefore the plane xx^x2 is the tangent plane at x. The normal at x is the line x \xxYx2. But x | xxYx2 = (x | x2) | xx-l + (x\x1)\ X + (x I x) I a?^ = I xxx2. Hence the line \xxx2 is the normal at x. 294. Curvature of Surfaces. (1) Let 8v be the perpendicular from the point x -f 8x on to the tangent plane at x. Then [cf. 224 (4)] 8v {xxYx28x) 7 ^{ ) (SO)2 + 2 (xXyVzXy) 808 f) Let - be the measure of the curvature of the normal section at x through the tangent line xhx. Then . {(x, | 2 (x, \ = (f _ 2 Sv (2) Now seek for the maximum and minimum values of p, when the ratio of 80 to 8 f is varied. Let p1 and p2 be the maximum and minimum values found, and let SOJSfa and 802/8 f 2 be the corresponding ratios of Then Pi/yVI^A |^a} and ^/yV^aI^i^} are the roots of the quadratic for f: - (^ |#2)f = 0. n) 2 ( 72 {(^i _1 1 _ (#i! #1) {xxxX zz) + (#2 ^2 (3) The expression for can be put in terms of (x11^ ^ijo2 and of their differential coefficients with respect to 0 and / . For (xXiX n) (xx^x y?) = (xaw n IxXjX^^) = | 1 , {x\a$ , (x\x2) , {x\x (Xu a?2), (a?2 |ar2), \X2),
490 CURVES AND SURFACES. Hence, since (x\x1) = 0 = (x \x2)t [CHAP. VII. Similarly 0 ! Y) (x21 x2) - (^ I x2)2} {(xu I a^) - (a;! xn) (x | {(x, |a - (x \x12f}. 7) 7) Now write ^ (^ |^) in the form {x1 \x^9 and ^r (^1 |^i) in the form (xx l^, with a similar notation for the dififerential coeflBcients of similar quantities. Then by successive differentiations of the equations (x\x1) = 0 and (x \oo2) = 0, we find (x \xu) = - (^ 1^), (#|#22) = --(tf2|#2), (a? |^12) = -( ! | 2); Also Hence #22) - (x12 \x12) = (x1 \x2)12 - J (a?2 \x2)n - k) Thus is expressed in the required form. P1P2 (4) Hence follows the extension to Elliptic and Hyperbolic Space of Gauss' theorem with respect to the applicability to each other of two small elements of surface. It is evidently a necessary condition, that should Pip* be the same for each element. 295. Lines of Curvature. (1) By the usual methods of the ele- mentary Differential Calculus it is easily shown that the ratios 80^8^ and a which give the directions of the lines of curvature (defined as lines
295] LINES OF CURVATURE. 491 of maximum or of minimum curvature) through #, are the roots of the follow- ing quadratic for 80/80: {Oi \x2)(xx1x2x11)-(x1 \xl\xx1x2xl^)(80y-\-{(x2\x2)(xx1x2x11)--(x1 \x1)(xx1x2x22))808$ + {(#2 k2) (xx^ u) - Ox I x2) (xXiX n)} mf = 0......(i). This equation can be put into another form. For [cf. 293 (3)] x2) (xX zXn) - (xx l^j) (xx^X^CY2) = (x \x xn )*y v Of* I ) i /Y /V* /y /y* /y \ i /v 1 by 103 (3), equation (4). But the product {(xxxx2. xux12) \ x] is pure [cf. 101], and therefore associative. Hence {(xxxx2. xux12) \x} = (xxYx2 \x. xux12) = (xYx2 xux12), since xxxx2 \x = (x \x) x1x2-f (xx \x) x^ + (x2 \x) xxx = xYx2i by 293 (2). Similarly (x2\x2)(xx1x2xn) (x1 and (x21 x2) (xx1x2x12) (c Accordingly equation (i) takes the form (XiX nXw) 802 + (^1^2^11^22) 808(j) -f (x1x2x12x22) 8 f 2 = 0............(ii). It easily follows from equation (i) that (2) Let x + 8x and x + 8'x be any two neighbouring points to x on the surface, where Then the angle yjr between the two tangent lines x8x and x8'x is given by . (x8x \x8'x) V {{xox I xox) (xo x I xo x) j , (x, 1^) 808r0 + (xx \x2)(808'cj where 80- and 8V are the arcs between x and x + 8x, x and x + 8'x. Corollary. The lines of curvature cut each other at right angles. (3) Since (x \ 8x) = 0 = (x \ 8'x)t where 8x and 8'x are infinitely small, ______(8x\8'x) , / (s^'# 18x8'x) Hence smyfr = / (?^ IS . ' ' x). T V {(8x\8x)(8x\8x)} Therefore So-Sa' sin yfr = rf*J{8x8'x \ 8x8fx} = 72 (808'$ - 8'08$) V{( i \*i) (^ 1^)}. sin a); where co is the angle at x between the curves 0 = constant, / = constant.
492 CURVES AND SURFACES. [CHAP. VII. (4) If the curves, 6 = constant, f = constant, be lines of curvature at all points, then the equation for the lines of curvature must reduce to B0S(j = 0. Hence from subsection (1), equation (i), and these equations must hold for all values of 6 and f . (5) Let be the measure of curvature of the normal section through Pi xx1} and of that through xx2\ where the 0 and f curves are lines of P2 curvature. The radius of curvature of any normal section [cf. 294 (1)] is given by {(#! \x1) {buy + (#2 I #2) \P p)\ = ---------\ V) H----------K^y) P Pi P* The angle yfr, which the tangent line xSx makes with the tangent line xxl, is given by cos sir = -777- TT 1 cos2 ylr sin2 ylr Hence - =------T -\-------- . P Pi P-2 This is Euler's Theorem. (6) The condition for the 6 and curves being lines of curvature may be put into a simpler form than that in subsection (4). For we have (x \ xx) = 0 = (x | x2) = {xx \ x2) = (xx rfc^. Hence [cf. 293 (3)] (x \xu) = 0. Hence since (x x^) = 0 = (x \ x2) = (x | x12), either the three equations are not independent and x12 can be written in the form or x is of the form v lavtv^. Taking the latter alternative, and substituting in the equation (xx1x^c1^ = 0, we find (Aye^ ^1^2^12) = 0. But the condition (P|P) = 0 cannot be satisfied in Elliptic Space by a real plane area. Hence it implies, if the plane area is known to be real, P = 0. Thus we are brought back to the first alternative, namely x^.jc^ = 0.
296, 297] lines of curvature. 493 If the space be Hyperbolic, the condition, (P | P) = 0, implies that all the points on the plane P be anti-spatial, except its point of contact with the absolute. Hence if x be a spatial point, (xP) cannot vanish. So again the first alternative is the only one satisfying the conditions. 296. Meunier's Theorem. The measure of curvature of the curve, = constant, iP is given by = constant, is found from the formula of 288 (1). Writing for it, The measure of curvature of the normal section through xx1 is given by (XXiXtyX'u) ( Hence = p V {(0*0111 The osculating plane of the curve is xxYxu. Let % be the angle between this plane and the normal section, which is the plane x1 \xxx2. Then cos % = + Now x11 xxx2 = (xx \ x2) xx {xy I xY) I x2; hence {xY \xYx2 \. And ( Thus cos X = xxlxu)(xlx2 iXj p' Therefore p cos % = p. 297. Normals. (1) The normal at the point x is N = \xYx2i the normal at the point x + hx is JV7 = | #^#2 -f J S^a^ + ! x^x2 + I S^S^,. Hence (NNf) = | (x1x^x18x2) (x1x28x1Sx2). Now S^! = xnB0 + x12h f y Sx2 = Therefore Therefore in general normals at neighbouring points of the surface do not intersect. But [cf. 295(1) equation (ii)] normals at neighbouring points on a line of curvature do intersect,
494 CURVES AND SURFACES. [CHAP. VII. (2) If the 0 and / curves are lines of curvature, then by 295 (6) Hence (#i4Wi2) = 0 Thus (NN') = fax Hence neighbouring normals on the curve 0 = constant, or on the curve f = constant, intersect; that is to say, neighbouring normals on a line of curvature intersect. 298. Curvilinear Co-ordinates. (1) Let x be conceived as a function of three variables, 0, , yfr. Then the equations 0 = constant, = constant, and y(r constant, determine three families of surfaces. On the surface, 0 = constant, x is a function of the variables f and yfr; on the surface, f =constant, it is a function of i/r and 0; on the surface, yfr = constant, a function of 0 and j . Let 7rr, = xli - = # , ^-T=x3. with a corresponding notation for the cu o p oy x higher differential coefficients. (2) Now suppose that the three families of surfaces intersect orthogonally wherever they meet. Then {x1 \ x2) = 0 = (x21 x3) = (x3 \ x^). Hence (x13 \x2) + (xj {x^) = 0, (x12 \x3) 4- (x2 \x31) = 0, (x^ |a?2) 4- (x3 \xl2) = 0. Therefore (x11x^) =0 = (x2\x31) = (x3 \x12). Also [cf. 293 (2)] (x |^) = (x \x2) = (x \x3) = 0. Hence since (^ \x) = 0 = (#3 oc2) = (^ |^3) = (^1 |a?23) it follows that But the equations (x2\x3) = 0 = (xx^x^) are the conditions [cf. 295 (4)] that the / and yfr curves should be lines of curvature on the surface, 0 = con- stant. Thus the lines of intersection are lines of curvature on each surface. This is Dupin's Theorem. 299. Limit-Surfaces. As a simple illustration of some of the above formulae, adapted to Hyperbolic Space, consider the limit-surface 2(x\x) = (x\b)\ It has been proved [cf. 261 (3)] that if the spatial origin e be taken on the surface, and if the line ee1 be taken to be through the point b on the absolute, then the equation takes the form (cc\x) = {x\{e + ei)y. Now if a? be at unit intensity, we may write
298, 299] LIMIT-SURFACES. Then ^2 I 9o-2 72 72*1 "V2J 1 2 a r8 72^ 7 3' Let So- be the element of arc between the points a? and ir + Sa;, then rf * K) So-2 x2)S ^ - 495 = T2 ' Hence So-2 = S 722 + S r32. Accordingly the metrical properties of the surface must be the same as those of a Euclidean plane. The same result had been arrived at before [cf. 262(6)] when it was proved, that the sum of the angles of a triangle formed by great circles on a sphere of infinite radius is equal to two right-angles. The curvature (- J of any normal section [cf. 294(1)] is given by _ V{ i21 i2} { (i I i) 22 2 (#! | x2) 8 r2S r3 (x2 \ x2) 8 r82} __ ~ (xx^^) r2 + 2 {xxxx^ BaSa- + (xxx^^ r2 Hence every normal section is a limit-line, a result otherwise evident.
CHAPTER VIII. Transition to Parabolic Geometry. 300. Parabolic Geometry. (1) The interest of Parabolic Geometry centres in the fact that it includes the three dimensional space of ordinary experience. Any generalization of our space conceptions, which does not at the same time generalize them into the more perfect forms of Hyperbolic or Elliptic Geometry, is of comparatively slight interest. We will therefore confine our investigations of Parabolic Geometry to space of three dimensions, in other words, to ordinary Euclidean space. (2) The absolute quadric as represented by the point-equation has degenerated into the two coincident planes [cf. 212] The intensity of any point #(=X e) must therefore [cf. 213] be con- ceived to be the square root of the left-hand side of this equation, that is, ai?i + ... 4- a4(-4. The absolute plane itself being the locus of zero intensity. (3) It is proved in 87 that, if the intensities of the unit reference points be properly chosen, the equation of the absolute plane becomes and the intensity of any point 2fe is 2 . The intensities of all points in this plane are zero. Hence, if a and b be any two points at unit intensity, the point a b, which is at zero intensity, lies in the absolute plane. (4) If three of the reference points, namely, uly u2, u3, be taken to be in the absolute plane, and e be any other reference point, then any point x is denoted by ]-e 4- 1^1 + 2w2 + %fih; and its intensity is . Thus the expression e + 2fw is the typical form for all points at unit intensity. 301. Plane Equation of the Absolute. (1) In order to discuss completely the formulae for the measurement of distances and angles, it is requisite to write down the most general plane-equation of the absolute, which is consistent with the point-equation reducing to two coincident planes. This question was discussed in 84 (4).
301] PLANE EQUATION OF THE ABSOLUTE. 497 (2) Let any planar element be denoted by P = Xu^Us T^eu^ + X2eu1iis X^eu^, Then it has been proved in 84 (4) that the plane-equation an V + ... + 2a12\1X2 + ... = 0, where the terms involving X are omitted, necessarily implies the point equation, r=o, where any point x is written ge + %YuY + %2u2 -f f3w3. The fully determined absolute quadric may therefore be considered as a conic section lying in the absolute plane. The points on the absolute are the points of the plane, the planes enveloping the absolute are the planes touching the conic section. The absolute plane is also called the plane at infinity; and the conic section denoted by the plane-equation of the absolute may be called the absolute conic lying in the plane at infinity. (3) Let this conic section be assumed to be imaginary, so that the elliptic measure of separation holds for planes [cf. 211 (2)]. (4) It may be as well at this point to note that the operation of taking the supplement with respect to the absolute becomes entirely nugatory. The operation therefore symbolized by | will in Parabolic Geometry represent as at its first introduction in 99 the fact that the reference points (what- ever four points they may be) are replaced according to the following scheme, ex by e2ese4) e2 by - e e^ e3 by exe2eAi e4 by - eYe2ez. This operation of taking the supplement, as thus defined, will (as previously) be useful in exhibiting the duality of the formula, when it exists. Its utility for metrical relations will be considered later [cf. Book VII., Chapter il] (5) It has been proved in 212 that in either Elliptic or Hyperbolic space if we start with an absolute of the form, ? + n l2 + 22^2 + 33 2 = 0, and make it gradually degenerate to 2 = 0, at the same time increasing the space-constant, then the distance between any two points x and y, where x is fe + ^ -f f2a2 + 3^3 and y is rje + rj^ + tj2u2 + v^h takes the form V^Y + K2 It will be observed that the assumption of the initial form of the absolute, from which the degeneration takes place, is equivalent to the assumption that eitj, eu2, eus are mutually at right-angles. The most general assumption for the plane-equation of the absolute is then [cf. 84 (4)] (p*$xy=a%2+ 2x22 + w = 0. w. 32
498 TRANSITION TO PARABOLIC GEOMETRY. [CHAP. VIII. And if 6 be the angle between the two planes and then [cf. 211] (6) But the K's are not independent of the /8's as these two detached forms of statement may suggest. In order to perceive the connection it is better to conduct the gradual degeneration of the absolute as follows. Let the plane-equation of the absolute be /^X2 + A2 V + %* + A2 V = 0, where $ will ultimately be made to vanish. Then the point-equation of the absolute is Hence by reference to 212 we see that Therefore if K be some finite constant, K2 = Accordingly, when ft is made to vanish, the distance between two points x and y takes the form If x and 2/ be two points of unit intensity, they are of the form e + Sf?/ and e -f S^w, and their distance is K V{/8aW (ft " ^)2 + AW (ft - V*)2 + AW (ft - %)2}. 302. Intensities. (1) The intensities of the points which lie on the plane at infinity, which is the degenerate form represented by the point- equation of the absolute, are all zero according to the general law of inten- sity. It was explained in the chapter on Intensity [cf. 86 (2)] that some special law of intensity, applying to these points on the locus of zero intensity, must be introduced. (2) Consider two points x and x' on the line eux. Let x = e + e + \f1w1, so that x and x' are at unit intensity. The distance ex is i the distance ex' is ITAA^ft. Hence eaf
302] INTENSITIES. 499 The difference of the two points x and e each at unit intensity is a point at infinity, in fact x e = ^uY; similarly xf e = X^u^ Hence x e and x' e denote the same point at infinity, but at intensities (according to some new law) which are proportional to the distances ex and ex*. (3) Let the intensity of a point at infinity be so defined that, if a and b be any two unit points at unit distance, the point a b is at unit intensity, positive or negative. Also let the three points ul9 w2, u3i used above, be at unit intensity. Then any point a = e + ux is a unit point on eax at unit distance from e. But its distance from e is if/32y83. Hence K$$3 = 1. Similarly for points on eii2 and eu3. 1 = )92=/33 = /9, say; andiT/32=l. (4) Hence with these definitions, the plane-equation of the absolute is, The angle 6 between the two planes \2 eu2ii$ -f \ eiiiU3 \3 is given by ~ vkv+v+v) (V2+Vs+V2) j' The distance between any two unit points e -f 2fi and e + S97?/ is The intensity of the first of the planar elements given above is The intensity of the point on the absolute plane, X1Mi + X2w2 + X3'W3, is the distance between the points e and e -f 2\w, that is, V{^i2 + V + V}- (5) The transition* from Hyperbolic or from Elliptic Geometry to that of ordinary Space has now been fully investigated. The logical results of the definitions, which have finally been attained, will be investigated in the next book. * Since Euclidean space is the limit both of Elliptic and of Hyperbolic space with infinitely large space-constants, it follows that the properties of figures in Elliptic or Hyperbolic Space, contained within a sphere of radius small compared to the space-constant, become ultimately those of figures in Euclidean space. Hence the experience of our senses, which can never attain to measurements of absolute accuracy, although competent to determine that the space-constant of the space of ordinary experience is greater than some large value, yet cannot, from the nature of the case, prove that this space is absolutely Euclidean. 32 2
500 TRANSITION TO PARABOLIC GEOMETRY. [CHAP. VIII. 303. Congruent Transformations. (1) It will however be instruc- tive to work out the properties of Congruent Transformations for Parabolic Geometry in the same way as that in which they were discussed in the preceding chapter for Elliptic and Hyperbolic Geometry. (2) The special properties of a congruent transformation are, as stated in 268 (1), (a) the internal measure relations of any figure are unaltered by the transformation: (/3) the transformation can be conceived as the result of another congruent transformation p times repeated, where p is any integer: (7) real points are transformed into real points: (S) the intensities of points are unaltered by transformation. (3) It follows from (a) firstly that the plane at infinity is unaltered by the transformation; and secondly, that the degenerate quadric represented by the plane-equation of the absolute, which is a conic in the plane at infinity, is transformed into itself. (4) Thus the plane at infinity is one semi-latent plane of a congruent transformation. It is proved in the next subsection that there must be at least three distinct latent points on this plane. Now, by reference to 190, it can be verified that semi-latent planes, with at least three distinct latent points on them, only exist in the cases enumerated in 190 (1), in 190 (2), in 190 (3) Cases I. and II., in 190 (4) Cases I. and II., and in 190 (5) Cases I. and II. But in each of these cases a semi-latent (or latent) line exists, which does not lie in the semi-latent plane containing the three distinct latent points. Now by Klein's Theorem [cf. 200] the points on the absolute on this line are the latent points of the congruent transforma- tion. But these points on the absolute are the two coincident points in which the line meets the plane at infinity. Hence the line is in general a semi-latent line with only one latent point on it, namely, the point at infinity. (5) Now consider three unit points (ult v2, u3) on the plane at infinity, so that the three lines drawn to them from a unit point e, not on this plane, are at right-angles to each other. Then any point on the absolute is fritti + ft^+fs^sj and any plane is Xti1u2u3 \1eu.2u3-\-\2eu1us \3ei(1u2. The plane-equation of the degenerate absolute conic is Xx2 + A^2 + \32 = 0. Hence, confining attention to points and lines on the absolute, any line on the absolute is XjU2u3 4- A^t^H- X^u^: the line-equation of the absolute conic is V + X22 + \32 = 0 ; and its point-equation is %f + f 22 + 32 = 0. Now it is easily proved* that a linear transformation in two dimensions, which transforms a conic into itself, must be such that two of its latent * Cf. Klein, Joe. cit. p. 3C9.
303] CONGRUENT TRANSFORMATIONS. 501 points are on the conic, and a third is the pole of the line joining the other two. Assume ux to be the latent point not on the absolute conic: then the polar of v^ is the line u2u3. Let this line cut the absolute in the points v and v'. Then uly v vf are the latent points of the transformation. Since the conic is imaginary, the points v and v' are conjugate imaginary points; and hence it is easily proved, that the three points vt v and uY are necessarily distinct. The equation of the line n2us is f1 = 0; hence v and v are given by this equation and by f22 + f32 = 0. Thus we may write iir in ht itr v=e*u.2+e ^u3, v' e ^w2 + e4w3. (6) Let the latent roots of the matrix be a, fty ft' corresponding to tiiy v v\ Then ^8 and ft' must be conjugate imaginaries; accordingly put Again, considering the complete three dimensional transformation, the semi-latent line, not lying in the plane at infinity, corresponds to two equal roots. This repeated root must be real: hence the line also must be real, and cut the plane at infinity in a real latent point. Thus ux is the point in which the semi-latent line cuts the plane at infinity. Now if p be any point on this semi-latent line, and j be the matrix representing the complete three dimensional transformation, f p= ap + Xiii. But from assumption (8) of subsection (2), a = 1. Again*, in order that the conic may be transformed into itself, a? = ft ft'. Hence fto2= 1, and therefore fto= 1. Thus finally the latent roots of the transformation are 1, 1, * and e~i . (7) Now let u2 and uz be transformed into u2' and u3\ Then e*u2'+e *ih'=ei8v = ey */^ + e"n "4^, e *u2 +e4w3' = e-tSv =e x 4/ u2 + e v 4/ws. Hence u2 = u2 cos 8 + us sin 8, u3' = uA cos 8 - u2 sin 8. Also let e be any unit point on the semi-latent line cutting the absolute in ih. Then [cf. 200 (2)] any point e + f^i on this line is transformed to # + ( + 7)^. Thus all points on this line are displaced through the same distance 7. Let this line be called the axis of the transformation. Any point e + %%u becomes ?i + 7) ui + ( cos S - f3sin 8) ^2 + (f3 cos 8 + f2 sin 8) w3. * Cf. Klein, loc. dt. p. 369.
502 TRANSITION TO PARABOLIC GEOMETRY [CHAP. VIII. (8) If 8 = 0, the transformation is called a translation. The axis of a translation is indeterminate, since any line parallel to eux possesses the same properties with regard to it as eu^ If 7 = 0, the transformation is a rotation. Every point on the axis euY of the rotation is a latent point. If any point at a finite distance is unchanged by a congruent transforma- tion, then the axis must pass through that point, and 7 = 0. Hence the transformation is a rotation.
BOOK VII. APPLICATION OF THE CALCULUS OF EXTENSION TO GEOMETRY.
CHAPTER I. Vectors. 304. Introductory. (1) The analytical formulae applicable to Euclidean space relations were arrived at, under the name of Parabolic Geometry, as a special limiting case of a generalized theory of distance. We will now start afresh, and, apart from any generalized theory, will con- sider the applicability of the Calculus of Extension to the investigation of Euclidean Geometry of three dimensions. Neither will it be endeavoured to assume a minimum of axioms and definitions in Geometry, and thence to build up the whole science by the aid of the Calculus. Such a scientific point of view was adopted in the investigation of the generalized metrical theory of the previous book. At present the propositions of elementary analytical Geometry will be assumed as known, and the suitability of the Calculus for geometrical investigation demonstrated by their aid. It may be further noticed that the propositions, which fall under the head of what is ordinarily called Projective Geometry, have been sufficiently exemplified in Book III., so that now metrical propositions will be chiefly attended to. Fig. 1. (2) Let the points e1} e2, es, e4 form a tetrahedron; and let x be any other point. Let the co-ordinates of x in tetrahedral co-ordinates be fi, s referred to the fundamental tetrahedron e^e.^; so that, for
506 VECTORS. [CHAP. i. instance, is the ratio of the volume of the tetrahedron we2ese4i to that of the fundamental tetrahedron. Similarly for 2, f3, and f4. Also is positive when x is on the same side of the plane e2e3e4 as the point ely and is negative when x is on the other side, with similar conventions for the signs of the other co-ordinates. With these conventions the co-ordinates of x always satisfy the equation, 6+ + ?. + ?4=l. (3) Now let elt e2, e3, e4 also stand for four reference elements of the first order [cf. 20 and 94] in the calculus, and let x denote the element ^l + ^+^s + l^- And let x be at unit intensity [cf. 87], when i + ft + + f4 = 15 and be at intensity \, when ft + 2 -f f3 + 4 = X. (4) Then, when x is at unit intensity, the co-ordinates ft, ft, ft, ft of subsections (2) and (3) can be identified. For [cf. 64 and 65] if x and y be any two points with tetrahedral co-ordinates ft, ft, ft, ft and r)!, r)2, Vs V* respectively, then the point z which divides the line xy in the ratio X: fi, so that ~zx is to z~y as X is to /a, has as its co-ordinates Thus if a? and y also stand for unit elements in the calculus, the point z stands for the element (fix -f X#)/(X + A*), and is also at unit intensity as thus represented. Thus conversely (fix + Xy) can be made to represent any point on the straight line xy, by a proper choice of X/fi. (5) For instance let x, y z denote the three angular points of a triangle at unit intensity. The middle points of the sides, also at unit intensity, are i (V + z) \ (z + x\ i (x + y)- ^ny Points, not necessarily at unit intensity, on the three medians respectively are i\(y+ *) + /**, iVOer + aO + ^y, \" (x + y) + tf'z. It is obvious therefore that the three medians meet in the point (x + y -h s), which, as thus represented, is at intensity 3. 305. Points at Infinity. (1) The point fix Xy, assuming X and p to be positive, divides the line xy externally in the ratio X to fi. In particular, the point x y divides xy externally in a ratio of equality, and is therefore the point on xy at an infinite distance. It is to be noticed that the intensity of x y is necessarily zero, and therefore that the intensity of X (x y) is also zero. Thus the plane at infinity is the locus of points at zero intensity [cf. 86 (1)]. (2) A special law of intensity must therefore be assumed to hold for the points on the plane at infinity [cf. 86 (2)]. Thus if xt y, z be three collinear points at unit intensity, y x and z x both denote the same point at infinity, but not at the same intensity according to this special law.
305, 306] POINTS AT INFINITY. 507 Suppose for instance that z divides the distance between x and y in the ratio X to fi, so that z = (fix + Xy)/(X + /i). Then s - x = ^ (y - x). Hence the intensity of z x is to that of y x in the ratio of the distance x~z to that of the distance xy [cf. 302 (2)]. (3) Any law of intensity may be assumed to hold in the plane at infinity, which preserves this property [cf. 85 (2)]. But great simplicity is gained by defining the distance xy as the intensity of the element y x. (4) Let the line ab be parallel to the line cd and of the same length. Fig. 2. Let the points a, b, c, d be at unit intensity. Then the elements b a and d c are the same point at infinity at the same intensity. Hence b a = d c. Therefore a c=b d, and the symbols express the fact, that ac and bd are equal and parallel. Also a + d = b + c ; which expresses the fact, that ad and be bisect each other. 306. Vectors. (1) Let a point at infinity be called a vector line, or shortly, a vector. A vector may be conceived as a directed length associated with any one of the series of parallel lines in its direction. Thus if u denote the vector parallel to ab and cd, and of length equal to ab or cd reckoned from a to or from c to d, then u = b a = d c. (2) The conception of vectors is rendered clearer by the introduction of the idea of steps, which is explained in 18. Thus the addition of it to a is the step by which we pass from a to 6, for a + u = 6; and the intensity of u measures the length of the step. Since also c + u = d, we must reckon, in accordance with this definition, all parallel steps in the same sense and of the same length as equivalent [cf. 3]. Again if v denote the vector, or step, from d to 6, then v = b d. So u + v = d c+6 d=6 c. Thus the sum of two steps is found by the parallelogram law.
508 VECTORS. [CHAP. I. (3) The fundamental tetrahedron may be chosen to have for its corners any unit point e and three independent vectors ulf u2, u3, each of unit length. Any point x is then symbolized by e + iWi-f ^^a + fs^, and the intensity of x is [cf. 87 (4)]. Thus if x be at unit intensity, it is written 0 + M1 + %2u2 + ?3^s- Thus the lines eulf eu2, euz are three Cartesian axes, and f2, f2, 3 are the Cartesian co-ordinates of the point. For let ely e2} es be three unit points on the lines eulf eu2y eu3 respectively, and each at unit distance from e. Then ux = e1 e, u2 = e2~- e, u3 = e3 e. Also let x=];e + i;ie1'\-l;2e2 + 363, where f, , f2, 3 are tetrahedral co-ordinates of x. Then But is the ratio of the tetrahedron exe2e3 to the tetrahedron ee je^ that is, the ratio of that Cartesian co-ordinate of x, measured on euly to a unit length. Similarly for 2 and f3. Hence , 2, f3 may be considered as the Cartesian co-ordinates of x, referred to the axes eu1} eu2, eu3. (4) Any vector can be written in the form A vector of the form \u + /jlv must denote a vector parallel to the series of planes which are parallel to the pair of vectors u and v. 307. Linear Elements. (1) A linear element, or the product of two points, must be conceived as a magnitude associated with a definite line. Thus, if a and b be two points, the linear element ah is a magnitude asso- ciated with the definite line ah. Suppose that c is another point on ah such that the length from a to c is X times the length from a to b. Then ac = Xab, cb = (1 \) ah. Therefore c=(l \) a + \b, and c is at unit intensity, if a and b are also supposed to be at unit intensity. But ac = Xab. Hence the intensity of ac is \ times that of ah, when the length ac is \ times the length ah. (2) We may therefore define the intensity of the product of two unit points as the length of the line joining them. If the two points a and b are at intensities a and y8, the intensity of ah is a/3 times the length ah. (3) The vector b a (a and b being at unit intensities) and the product ab should be carefully compared. The intensity of each is defined as the length ab} but they are magni- tudes of different kinds. For b a is a directed length associated with any one of the infinite set of straight lines parallel to ab, and is an extensive magnitude of the first order, being really a point at infinity. While ab
307, 308] LINEAR ELEMENTS. 509 must be conceived as a directed length associated with the one definite line ab, and is an extensive magnitude of the second order. (4) Also ab can be written in the form a (b a), since aa = 0. Hence the linear element ab may be conceived as the vector b a, fixed down or anchored to a particular line; and the unit point a, as a factor, may be conceived as not affecting the intensity, but as representing the operation of fixing the vector. (5) Also if c be any other unit point on the line ab, then (a - c) (b - a) = 0; since a c and b a represent the same point at infinity at different intensities. Hence a (b a) = [c + (a c)} (b a) = c(b a). Thus any other unit point in the line ab may be substituted as a factor in place of a. (6) Hence if a, b, c be three collinear unit points, ab 4- be = ac. For by (5) b (c b) = a (c b); and hence ab + bc = a(b a) + b(c b) = a(b a) + a(c b) = a(c a) = ac. Fig. 3. If a, by c be not collinear, then b 4- c ab + ac = 2 . a - = ad, where d is the opposite corner of the parallelogram found by completing the parallelogram ab, ac. 308. Vector Areas. (1) A product of two vectors will be called a vector area. If uv be any vector area, then only the intensity is altered when any two vectors parallel to the system of parallel planes defined by u and v are substituted for u and v. For let u^ = \xu 4- ^v, u2 = X^u + ^v; then u z = (X^a X^) uv. Hence u^ denotes the same vector area as uv only at different intensity.
510 VECTORS. [CHAP. I. (2) From any point e draw two lines ep and eq representing in magnitude and direction the vectors u and v respectively, and complete the parallelogram eprq. Also draw epx and ep2 to represent the vectors u^ and u2 respectively, and complete the parallelogram epip3p2. Then, conceiving eu and ev as two Cartesian axes and assuming that the vectors u and v are of equal length, the co-ordinates of px and p2 are \l9 fa and Xo, /jl2 respectively. Hence the area of the parallelogram epip3p2 is to that of the parallelogram eprq as (\ifi2 Xg/ii) is to unity. Therefore the intensities of uv and uYu2 are in the ratio of the areas of the parallelograms formed by uv and uni2. (3) But the intensities of vector areas are necessarily zero according to the general definition [cf. 307 (2)] of the intensity of a linear element. For if au be a linear element where u is a vector of length S and a is a point at intensity a, then the intensity of au is aS. But when a becomes a vector, a is zero. Therefore the intensity of a product of two vectors is necessarily zero. Accordingly a special definition must be adopted for the intensity of vector areas; and the above investigation shews that we may consistently adopt the definition that the intensity is the area of the parallelogram formed by completing the parallelogram eu, ev. The intensity of uv will be considered by convention as positive when by traversing the perimeter of the parallelogram so as first to move in the direction of u and then of v, the direction of motion is clockwise relatively to the enclosed area. Then in the above figure for uv, we start from e and traverse ep which represents u and then pr which represents v, and the motion is anti-clockwise so that the area is negative.
309] VECTOR AREAS. 511 (4) A vector area will be conceived as possessing an aspect or direction, namely the aspect of the system of parallel planes which are parallel to the two vectors. A line parallel to this system of planes will be called parallel to the plane of the area, or parallel to its aspect. 309. Vector Areas as Carriers. (1) The addition of a vector area to any linear element, which is parallel to its aspect, simply transfers the linear element to a parallel line without altering its intensity. Fig. 5. For ab = a (b a) = [c + (a c)} (6 a) = c (b a) -f (a - c) (b - a). Now let d c = b a. Then c (d c) + (a c)(d c) = a(b a). Thus the addition to cd of the vector area (a c)(d c) transfers it to ab, which is an equal and parallel linear element. (2) It is also to be noticed that, if cd is conceived as continuously moved into its new position by being kept parallel to itself with its ends on ca and dbt then it sweeps through the area abdc, which is the area of the parallelogram representing the intensity of the vector area. (3) Let x and y denote any two unit points e + Ziiii + f2**2 + 3w3, and e + Vi*h Then by multiplication This is the form which any linear element must assume. Any vector area takes the less general form
512 VECTORS. [CHAP. I. 310. Planar Elements. (1) A planar element, or the product of three points, must be conceived as a magnitude associated with a definite plane. Thus if abc be a planar element formed by the product of the three points a, by and c, then abc is a magnitude associated with the definite plane abc. (2) Let v^ and u2 be two unit vectors parallel to this plane but not parallel to one another, and let e be any other point in it. Then we may write a = e + a^j + a2w2, b = e + ftt^ + fi2u2 c = e + yxuY 4- y2u2. Hence abc = \l 1 1 eu^iu. 2 02 72 Also eu^ = e(e + u^) (e 4- u2). Therefore the intensity of the planar element abc is to that of the planar element e (e + i^) (e 4- u2) in the ratio of the area of the triangle formed by a, 6, c to that of the triangle formed by e, e 4- uly e -\-u2. (3) We may therefore consistently define the intensity of the planar element abc as twice the area of the triangle abc. Also the convention will be made that the intensity is positive when the order of letters in abc directs that the perimeter of the triangle be traversed in a clockwise direction. If a be at intensity a, b at intensity y8, c at intensity 7, then the intensity abc is 2a/3y times the area of the triangle abc. (4) In comparing a vector area with a planar element it must be noticed that a vector area is conceived as an area associated with any one of a series of parallel planes, while a planar element is conceived as an area associated with a definite plane.
310 312] PLANAR ELEMENTS. 513 The planar element abc, where a, 6, c are unit points, can be written in the form a(b a){c a). Then (b a)(c a) is a vector area of which the area representing the intensity is the same in magnitude and sign as the area representing the intensity of the planar element abc. Accordingly if U represent a vector area, and a be any unit point, then the planar element all may be conceived as the tying down of the vector area to the particular plane of the parallel system which passes through a. Also this operation of fixing the vector area makes no change in the intensity. 311. Vector Volumes. (1) A product of three vectors will be called a vector volume. The intensity of a vector volume is necessarily zero according to the general definition of the intensity of a planar element. For if U be any vector area of area 8 and a any point of intensity a, then the intensity of all is a8. Accordingly, when a becomes a vector and a is therefore zero, the intensity of the planar element vanishes. A special definition of the intensity of a vector volume must therefore be adopted. (2) We may first notice that all vector volumes are simply numerical multiples of any assigned vector volume. For let ult u2t u3 be any three non-coplanar vectors. Then since there can only be three independent vectors, any other vectors u, v, w can be written respectively in the forms Then by multiplication Thus any vector volume is a numerical multiple of uxu#iz. (3) Also let two parallelepipeds be formed with lines representing respectively ul3 u2, us and u, v, w as conterminous edges. Then the intensities of u^Ui and uvw are in the ratio of the volumes of these parallelepipeds. Thus we may consistently define the intensity of a vector volume as the volume of the corresponding parallelepiped. 312. Vector Volumes as Carriers. (1) The addition of any vector volume to a planar element transfers the planar element to a parallel plane without altering its intensity. w. 33
514 VECTORS. [CHAP. 1. For consider any planar element abc and any vector volume V. Then we may write abc = a (b a) (c a)y and F= u (b a) (c a), where u is some vector. Hence abc + V = where a! a is the vector u, and a'b\ a'd are equal and parallel to ah and ac respec- tively. Fig. 7. (2) Also it is obvious that if abc moves continuously into its new position remaining parallel to itself with its corners on aa\ bb', cc' respectively, it sweeps out a volume equal to half the volume of V. 313. Product of Four Points. (1) Since the complete region is of three dimensions, the product of four points is a mere numerical quantity. Let elt e2i e3t e4 be any four unit points forming a tetrahedron, and let a, by c, d be any four other unit points, also expressible in the forms tae, S0e, Xye, 2 . Then (abed) = a2, a2, a3, a4 ?i, , A. A 7i 72 73 74 B4 Accordingly (from a well-known proposition respecting tetrahedral co- ordinates) the numbers expressed by the two products (abed) and are in the ratios of the tetrahedrons abed and (2) Let the product of four points, such as abed, be defined to be equal to the volume of the parallelepiped, which has the three lines ab} ac, ad as conterminous edges. Also {abed} = [a(b a) (c a)(d a)}. Hence {abed} = (aV), where V is a vector volume of volume equal to the volume (abed). 314. Point and Vector Factors. (1) It has now been proved that every non-vector product of an order higher than the first may be conceived as consisting of two parts, the point factor, which will be conceived as of unit intensity, and the vector factor. Also the intensity of the product, which is either a length, or an area, or a volume, is also the intensity of the vector factor.
313 315] POINT AND VECTOR FACTORS. 515 (2) Thus any linear element can be written in the form au, where u is a vector line and a is a unit point; any planar element in the form aM, where M is a vector area; any numerical product of four points in the form (aV), where V is a vector volume. (3) Also since a is a unit point and not a vector, it follows that ait = 0 involves ti = 0, and aM= 0 involves M = 0, and aV= 0 involves V 0. Thus au = au', involves u u'; and aM = aMf, involves M = M'; and aV = aV, involves V=V. (4) Again, if a and b be two unit points in the line au, then au = bu. If a and b be two unit points in the plane aM, then aM = bM. If a and b be any two unit points whatever, then (aV) = (bV). 315. Interpretation of Formula. (1) It will serve as an illustra- tion of the above discussion to observe the geometrical meanings of the leading formulae of the Calculus of Extension in this application of it. In the first place, let the complete region be a plane so that the multi- plication of two lines is regressive [cf. 100]. Let p, q, r, s be four points, and let t be the point of intersection of the two lines pq and rs. Fig. 8. Then t=pq.rs = (pqs) r - (pqr) s = (prs) q - (qrs)p. Hence t divides rs in the ratio of the area of the triangle rpq to that of the triangle spq; and the section is external, if the order of the letters in rpq and spq makes the circuit of the triangles in the same direction; and it is internal, if the circuits are made in opposite directions. Similarly t divides pq in the ratio of the area prs to qrs. (2) In the second place, let three dimensional space form the complete region. Then the products of a line and a plane, and also of two planes, are regressive. Let p, q, r, s, t be any five points, and let st meet the plane pqr in x. Then x = pqr . st = (pqrt) s - (pqrs) t = (pqst) r + (rpst) q + ( qrst)p. 33 2
516 VECTORS. [CHAP. 1. Hence x divides st in the ratio of the volumes of the tetrahedrons pqrs and pqrt; and the section is external, if the products (pqrs) and (pqrt) are of the same sign, that is, if s and t are on the same side of the plane pqr: otherwise the section is internal. Also, the last form for x states that the areal co-ordinates of x referred to the triangle pqr are in the ratio of the volumes of the tetrahedrons qrst, rpst, pqst. (3) We may also notice here that according to these formulae any five points in space are connected by the equation (qrst) p (prst) q + (pqst) r (pqrt) s + (pqrs) t = 0. The formulae for the line of intersection of two planes abc and def re abc. def= (abcf) de + (abce)fd -f (abed) ef = (adef) be + (bdef) ca + (cdef) ah. The geometrical meanings of these formulae are obvious, though they would be rather lengthy to describe. 316. Vector Formulae. (1) Some of these formulae take a special form, if four vectors ult u2 vlf v2 be substituted for four of the points. The special peculiarities arise from the fact that the product of four vectors is necessarily zero; and that if V be any vector volume, and a and b any two unit points, then (aV) = (bV). (2) The formula for five points becomes (au2vxv2) ! (auYvYv2) u2 4- (au1u2v2) ^ (an^ii^) v2 = 0. (3) Again, au^ . i\v2 = (au1u2v2) v1 (au^vj v2 = (an^v^ u2 (ai^v^) u^ Also ai\v2. uxu2 = (avxv2u2) uY (av^u^) v2 (av^u^ v2 (av2u1u2) v1. Hence auxu2. v^v2 -f avYv2. uxu2 = 0. Or, if M and M' be any two vector areas, aM.M'+aM'.M=0. Again, it is obvious that, if a and b be any two unit points, 317. Operation of Taking the Vector. (1) Let a unit vector volume be denoted by the symbol U, and let the sense of U be such that (all) = 1, where a is any point of unit intensity. Also if u be any vector volume, and if M be any vector area, then (uW) = 0, and (MM) = 0. (2) Now, if F be any linear element, it can be written in the form au and FX\ = au. U = (all) u = u.
316, 317] OPERATION OF TAKING THE VECTOR. 517 Similarly if P denote any planar element, it can be written in the form aM, and P. U = aM. U = (aU) . M = M. Hence the operation of multiplying U on to any non-vector element of any order yields the vector factor of that element. This operation will therefore be called the operation of taking the vector. (3) We may notice that, if this operation be applied to a vector, the result is zero; and if to a point, the result is the intensity of the point with its proper sign. (4) Thus if any force be ab where a and b are unit points, by taking the vector we have by the ordinary rule of multiplication ab. U = (all)b - (bU) a=b-a. Again, if any plane area be abc where a, 6, and c are unit points, taking the vector we have abc.U = (all) be + (bVL) ca + (cU) ab = bc + ca + ab, which is therefore the required vector factor. (5) In considering the effect of this operation on regressive products, it is well to notice that, if p be any point, the product (pU) can be conceived both as progressive and as regressive. Therefore the multiplication of U on to any pure regressive product still leaves a pure regressive product, which is therefore associative. (6) The regressive product an. bM is a point, so taking its vector must yield the intensity of the point. Also the product au . bM. U is a pure regressive product and is therefore associative. Hence au .bM.VL = au(bM. II) = (auM). Therefore (auM), which also equals (buM\ is the intensity of the point. (7) Again, aM. bM' is a linear element, and its vector factor is given by aM.bM'.\X = aM. (bM' .U) = aM.M' = - bMf. M. Also, since the result is a vector, it is evident that any unit point c can be substituted for a or b in these two formulae for the vector factor. These results should be compared with the last formulae in 316 (3). (8) Finally aM. bM'. cM" is a point. To find its intensity take the vector, then aM. bM'.cM" .XX = aM.bM'.(cM" . U) = aM. bM'. M". (9) As an illustration of these formulae, let us find the vector factor of abc . def. Then by subsection (4) of this article abc . def. U = abc . (def. U) = abc. (ef+fd + de) = def. (be + ca+ ab).
518 VECTORS. [CHAP. I. Also abc. (ef+fd 4- de) = (abed) (/- e) + (abce) (d -/) + (abef) (e - d). (10) Again, let a be any unit point, and F be any linear element. Then F can be written be, where b and c are unit points. Hence aF. it = abc. U = (he + ca + a ) But c-6 = P.U. Therefore aP. U = F- a . FU. (11) Let P be any linear element. Then the linear element through any point d parallel to F is d. FU. Thus if F be in the form ab} where a and 6 are unit points, the parallel line through d is d (b a). Let P be any planar element. Then the plane through any point d parallel to P is d. Pit. Thus if P be in the form abc, where a, b, and c are unit points, the parallel plane through d is d (be + ca -f- at). 318. Theory of Forces. (1) The theory of forces or linear elements, as discussed in Book V., holds in the Euclidean Space now under discussion. But some further propositions involving vectors must be added. (2) In 160 (2) it is proved that, if a be any given point, and A any given planar element, then any system of forces 8 can be written in the form where p and P are respectively a point and planar element depending on the system 8. Now let A denote a vector volume; then AP denotes some vector area, call it M. Also ap can be written in the form an, where u is a vector. Thus S = au + M. (3) The vector u is independent of the point a. For taking the vector of both sides 8VL=au.VL + MVL = au.tt = u, Hence, since u can be written in the form 8Ut it is independent of any special method of writing 8. (4) Let SIX be called the ' principal vector* of 8. It is the sum of the vector parts of those separate forces which can be conceived as forming S. Let the vector area M be called the vector moment of the system round the point a, or the couple of S with respect to a. Let a be called the base point to which the system is reduced. (5) Also M depends on the position of a. For a8 = aM. Hence M = aS. U, and therefore M is the vector factor of the planar element aS, which is the planar-element representing the null plane with respect to a.
318] THEORY OF FORCES. 519 The same results respecting M and u follow directly from 317 (10). For by adding the results of applying the formula of that subsection to each component force of 8, we at once obtain Let a' be any other unit point, and let M' be the vector moment of 8 with respect to it. Then S = an + M = {a + (a - a')} u + M = a'u + {(a- a') u + M}. Hence M' = (a - a!) u + M. (6) Also (S8) = 2 (auM); and since uM is a vector volume, (auM) = (auM) = {a!uM'\ And since aM = a$, ( $) = 2 (auS), where a is the principal vector of S. (7) Again evidently (aa'M) = (aa'M') = (aa'fif). And (auS) = ( S) = (a'uS). Therefore {(a-a')uS} = 0, where a and a' are any two unit points. This is only an expression of the fact that u8 is a vector volume, where u is the principal vector of 8. In fact from 167 (2) we have Thus the plane at infinity is the null plane of the principal vector. (8) To find the locus of base points with the same vector moment M. Let a be one such point and x any other such point. Then by hypothesis S S Hence x(8 M) = 0. But 8 M is the linear element a. 8U. There- fore the equation, x(S M) = 0,denotes that x lies on a straight line parallel to 8VL (9) Let Mo be any given vector area, then if 0M0 be the vector moment of S (round an appropriate base point) which is parallel to Mo, uS= 0uMQ, where u is the principal vector of S. Hence if a be any unit point Therefore 0 =\ {S8) 2 (a,uMQ)' Thus the locus of a point x such that the vector moment of S with respect to it is parallel to Mo or in other words, the point of which the null plane with respect to S is parallel to Mo is given by V 2 2 (auM0)
520 VECTORS. [CHAP. I. 1 S a But it was proved in 162 (2) that the conjugate of any line ab is ab. Hence the conjugate of any vector area Mo is a straight line parallel to the principal vector, and this line is also the locus of points corresponding to which the vector moments are parallel to Mo. 319. Graphic Statics. (1) It will illustrate the methods of the Calculus of Extension as applied to Euclidean Space, if we investigate at this point the ordinary graphic construction for finding the resultant of any number of forces lying in one plane. (2) Let the given system, S, of coplanar forces be also denoted by ZiWi + a2u2 + ... + avuv\ where a^, a2u2, etc. are given forces (cf. fig. 9). We require to construct their resultant. ap Fig. 9. Let v be any arbitrarily assumed vector in the plane. Then Thus - v), Up = (V - th ... - Wp_i) + (ui + W2 + + Up - V), S = axv + (a2 - a2) (v - Wj) + ... + Op - ap_j) (v - Uj ...-wp_1) + ... + (a, - _! (v -Ui...- w_!) .(1). (2).
319] GRAPHIC STATICS. 521 (3) The equations (1) giving the vector parts of the forces are equiva- lent to starting from any point b0 (cf. fig. 10) and drawing boc to represent v and W^ to represent ux. Then bYc represents d h^. Also from bx draw bj)2 Fig. 10. to represent u2y then hjo represents v u^ Uo, and so on. This is the ordinary Graphic construction for the force polygon with any pole c. (4) To simplify the expression for S notice that alf a2, ... av may be any points on the lines of the forces ; hence we may assume (cf. figs. 9 and 10) that a2 a2 is drawn parallel to v ult a3 a2 parallel to v ux ii2, and so on. Then 8 = ct^v + av (ux + u2 + ... + uv v). Thus the resultant force passes through the point d which is the point of intersection of aYv and av (^ + u2 + ... + uv v) and is the force (5) This gives the ordinary construction for a funicular polygon, thus: start from any point aOi draw a^ parallel to v, axa2 parallel to v nly and so on, finally avav+1 parallel to v ux u2... uv. Then the resultant passes through the point of intersection of a^ and avav+1. (6) Suppose that two different funicular polygons are drawn, namely ...a, and ao'ai'... a/, corresponding to the arbitrary assumptions of two
522 VECTORS. [CHAP. I different vectors v and v' respectively with which to commence the construc- tions (cf. fig. 11). We will prove the well-known theorem that the points of intersection of corresponding sides are collinear. For apap is parallel to up, hence apup = ap'up. Again, apap-Y is parallel to v uY ... wp_i, hence ap (v ux ... Up-!) = df^! (v uY ... Wp-j). Similarly, ap (vr v^ ... u^) = ap-1 (v uY ... up_x). Therefore axv a^v = ax(v i^) a/(v Ui) = a2(v u^) a2r (vf iii) = a2 (v uY u2) a2r (v ux u2) a3 (v uY u2) a3' (v ux u2) = etc. Let cip-iftp and a''P-Yap intersect in dp_Y, then axv - a^v =do(v v'), a2 (v u^) a2 (v u^) = dx (v - v)s and so on. Hence dQ (v v') = d1(v v)=... = dr (v v'). Fig. 11. Thus the points d0, c?i, etc. all lie on a straight line parallel to v v'. Also if the force polygon bobj)2 etc. be identical in the two cases (cf. fig. 11), and if boc represents v and boc' represents v\ then cc' is parallel to v v'. Note. Grassmann considers vectors in the Ausdehnungslehre von 1862; but not in connection with points. The formulae and ideas of the present and the next chapter are, I believe, in this respect new. The two operations of 'Taking the Vector' and of * Taking the Flux' [cf. 325] are, I believe, new operations which have never been defined before. Since this note was in print I have seen the work of M. Burali-Forti, mentioned in the Note on Grassmann at the end of this volume.
CHAPTER II. Vectors {continued). 320. Supplements. (1) The theory of supplements and of inner multiplication has important relations to vector properties. Take any self-normal quadric [cf. 110, 111], real or imaginary, and let el7 e2, e3, e4 be four real unit points forming a self-conjugate tetrahedron with respect to this quadric. Let elt e2, e3, e4 be respectively the normal intensities [cf. 109 (3)] of these reference points, where e^ e2, e3, e4 are any real or pure imaginary quantities. Then (e1|61) = lM2, (e2\e2)=l/ 2*, {e,\ez) = lje,\ (e4|e4) = l/tf. Also if x be any point 2 e, then Suppose that y, which is 2?/e, is on the polar plane of x with respect to the quadric, then Also all the points normal to x with respect to the quadric must lie on this plane. (2) Hence the pole of the plane at infinity is the only point which has three vectors, not coplanar, normal to it. This is the point a 2 c 2 C2 2 c 2 i ^2 I n 2 | ^.2* tj fc2 *-3 ^4 fcl " ^ ^2 ~T~ t3 T" t4 This point is the centre of the quadric. Let it be denoted by e, where e is at unit intensity, and let the normal intensity of e be e. Then we may write (612 + 22 + 32 + 642)e = e12^1- 1 Therefore e2 = e^ + e22 + e32 + e42.
524 VECTORS. [CHAP. II. (3) JBut elt e2, e3 e4 are any four points forming a normal system with respect to the quadric. Hence the last equation proves that, when the quadric has not its centre at infinity (in which case e^2 + e22 4- e32 -f e42 = 0), the sum of the squares of the normal intensities of any normal system of points is constant and is equal to the square of the normal intensity of the centre. 321. Rectangular Normal Systems. (1) Now let e be the centre of the self-normal quadric, and let u1} u2, u3 be three unit vectors forming with e a normal system with respect to the quadric. We may assume without loss of generality that the normal intensity of e is unity. Since uu u-2, u3 lie in the locus of zero intensity their normal intensities according to the general definition of intensity holding for all points, are zero. But [cf. 109 (3) and 110 (1)] let the normal lengths (or intensities) of ^, u2, u3 be *!, eta, ot3, according to the special definition of intensity for vectors [cf. 305 (8)]. Then (e\e) = 1, (% |^) = , (u2\u2) = , (u3 \u3) = . i O2 a3 (2) Also any point x at unit intensity is of the form e + 2%u. Hence fc 2 fc 2 t-2 (^) = l+il+ L+*t. a,3 a* 32 Thus the self-normal quadric is, 2 fc2 fc2 r a/ 32 and is accordingly purely imaginary, unless one or more of a1} a2, a3 are pure imaginaries. (3) It is obvious from the equation of the self-normal quadric that, if ulyu2,2*3 be any three mutually normal vectors, then eu1} eu2i eu3 are three conjugate diameters of the quadric. In general one set and only one set of such conjugate diameters are mutually at right angles. But if the quadric be a sphere with a real or imaginary radius, then all such sets are at right angles. In such a case let the normal systems be called rectangular normal systems. The centre (e) of the self-normal sphere will be called the Origin. 322. Imaginary Self-Normal Sphere. (1) Firstly, let the sphere be imaginary with radius V( !) Then a2 = o^ = a3 = 1. Hence if u1} u2J u3 be any set of unit vectors at right angles, then (ih luj) = 1 = (u.2!u2) = (u3 \u3); and (u21 u3) = 0 = (u31 ux) = Also if e be the centre of the sphere, (e\u1) = 0 = (e\ut) and (e\e) = l.
321-323] IMAGINARY SELF-NORMAL SPHERE. 525 Again, |e = w^2w3, jt^ = eu2us, \ua = eusulf \u3 = -eit^. And | eux = u2u3, \ u2u3 = eu^, and | eu2 = u^, | tt^ = eu2, and 1^3 = 1^2, |^1w2 = ^3. Also [u^u^is = e, \eu2u3 = i 2, \eu3ux = u2i \euiU2 = us, (2) Let y be any vector (^ -f 2w2 + f^, then (v V) = ^2 + 22 + f32. Hence (v |v) is the square of the length of v. Again, let v and vf be two vectors f^ + %2u2 + %zuz and ^uY + f/i/2 + f3'tt3, of lengths /3 and /o' respectively. Then ( I *') = ftf/ + ^/ + W = PP cos 0, where 6 is the angle between the two vectors. mi n (v v') i /i / (vi/\w') Thus cos6 = /f/ , w , tsti and sin 0 = A / / , ' ./.. . V{(viv)(v \v)} V (v!v)(v \v) (3) Again, let ilf be any vector area ^iU2u3 4- |2^3^i + ^i^h, and let //, be its area. Then (M\M) = tf + + 2 = ^2. Also let M' be another vector area in a plane making an angle 6 with that of Af; let if' be written %iU2ii3 4- 2'%wi + ^zuiu2 and let its intensity be//. Then (Jf | M') = ' + ftfa' + f8f/ = w' cos 0. (4) The inner squares and products of points and linear elements in general have no important signification. It will be observed that these formulae for the product of two vectors or of two vector areas are entirely independent of the centre of the self-normal quadric. The expressions (M \M) and (u\u) will be often shortened into M'2 and u2, on the understanding that the normal system is rectangular and the radius of the self-normal sphere is \/( 1). 323. Real Self-Normal Sphere. (1) Secondly, let the self-normal sphere be real with radius unity. Then with the notation of 321 ax = a2 = a3 = V(- 1 (2) Hence if ulf u2t u3 be any set of unit vectors mutually at right-angles, then (Wl ih) = - 1 = (u.2 \n2) = 31^ ; and (u21UJ = 0 = (u31 ux) = (u, i u2). Also, if e be the centre of the sphere, (e\u1) = 0 = (e\u2) and (e\e) = l.
526 VECTORS. [CHAP. II. Thus if v and v' be any two vectors 2|fy and %%'u, of lengths p and p, and making an angle 0 with each other, (v \v') = - ( / + #+u:)=- ' cos e. And in fact a set of formulae can be deduced analogous to those which obtain in the first case, when the sphere is imaginary with radius V(- !) (3) But the constant introduction of the negative sign is very incon- venient, we will therefore for the future, unless it is otherwise expressly stated, assume that rectangular normal systems are employed and that the self-normal sphere is imaginary as in 322. 324. Geometrical Formulae. (1) If u and v are vectors the square of the length of u + v is given by (u + vf = (u + v) | (u + v) = n2 + v2 + 2 (u \v) = u2 + v2 + 2 *J(u2v2) cos 6, where 0 is the angle between u and v. (2) To express the area of the triangle abc. Assume that a, b and c are at unit intensities. Then taking the vector abc. U = be + ca + ab = (a b) (a c). Hence, if A be the required area, A2 = } (abc.U)2 = J {(a -6)(a- c)\(a-b) (a- c)} If a, 13, 7 be the lengths of the sides, and a, jSy y the corresponding angles, (a Vf = y2, (a c)2 = 2, (a 6) | (a c) = /3y cos a. Thus A2 (3) The angle, 0, between two linear elements C2 and O2 is given [cf. 322 (2)] by The angle, #, between two planar elements P and Q is given [cf. 322 (3)] by (4) The length of the perpendicular from any point a on to the plane of the planar element P is obviously [cf. 311 (3) and 322 (3)] (5) To find the shortest distance between two lines Gx and C2. Let aY and a2 be any two unit points on C^ and G2 respectively. Then Ox = aY. CXU, and O2 = a2. O2U. Also the required perpendicular (ot) is equal to the perpendicular from a2 on to the plane ax. C^U. O2U.
324, 325] GEOMETRICAL FORMULAE. 527 _(a2a1.C1U.C2l\)_ (6) If 7! and y2 be the lengths of the linear elements Gx and C2, 6 the angle between them, and w the length of the common perpendicular, then [cf. 322 (2)] Hence (GYC2) = y^1^ sin 0. We may notice here that sind/= J l-cos20 = 325. Taking the Flux. (1) It is often necessary to express the vector-line normal to a vector-area, or the vector-area perpendicular to a vector-line. This is accomplished by a combination of operations, which we will call the operation of taking the ' Flux/ Let v be any vector and let e be the origin. Then | ev = j | eux + 21eu2 + 31 euz = %iii2u3 + ^hux + hu2. Accordingly \ev is the vector-area perpendicular to the vector v. Also the length of v is \l{%-? -f 22 + ?32} times the unit of length, and the area of | ev is V{ i2+ 2 + ^32} times the unit of area. The vector-area \ev will be called the flux of the vector v. Again, let M be the vector-area Then | eilf = x | ew2w3 + f21 euz'Ux 4- ^3 | MiW2 = ^i + Hence the vector-line |eilf, which has been defined as the flux of M, is perpendicular to the plane of M. The operation of taking the flux can of course be applied to non-vector elements, but the results are of no interest as permanent formulae and are easily worked out afresh when required. (2) It will be noticed that the result of the operation on vector elements is really entirely independent of the position of e, the centre of the self- normal sphere. Also furthermore the operation is capable of an alternative form. For |w = |e|v = U|v = | .U; and \eM = \e\M = XX\M= \M. U. Hence, except in respect to sign, the operation is identical with that of taking the vector of the supplement of the vector-line or of the vector-area. (3) For these two reasons it is desirable to adopt a single symbol for the combination of operations denoted by \e. Let \ev and \eM be written %v and %M respectively.
528 VECTORS. [CHAP. II. We notice that ggv = v and gglf = M. Also the operation is distributive as regards addition ; but it is not distributive as regards multiplication. 326. Flux Multiplication. (1) The operation of multiplying the flux of a vector-line or vector-area into a vector-line or vector-area will be called Flux Multiplication. Its formulae are almost identical with those of Inner Multiplication [cf. 119]. They are all independent of the position of the origin: this fact will be obvious, if it be remembered that (eVL) = (e'tt), where e and e' are any two unit points. (2) Firstly, it is obvious that (*%v) = ( 2 + + f, ) It = (v \v) U, (MdM) = ( * + + ) \X = (M\M)M. Similarly (vgt/) = (v \ if) U = (t/gv), and (M%Mf) = (M \ Mf) U = (M'%M). (3) Again, v$v2v3 = v1(\v2.\va.\l) = (vx \ v,) U | v2 + (^ | v2) \vsAl And ^sS i = 0 = i/g^ ..............................(ii). Equation (i) may be compared with From equation (ii), it follows as a particular case that ^1.3^2 = 0. (4) From equation (i), replacing v1 by %uu\ we have But [cf. 99 (4)] glT |v = l^ift;) = (eMv). Hence gww'. gtw' = (euu'tf) gv - (eimV) gt/.....................(iii). As a particular case of equation (iii) we deduce gtw\gtw" = (etraV/)8tf...........................(iv). (5) Let a be any unit point, then a%u = - a(\u. VI) = - (all) \u + (a | w) II = - \u + (a |w) U. Hence gw. gv = [\u (a\u)U} gv = \u. %v= \u . \v. Vl = \wv.VL = %uv......(v). Also if b be another unit point, and M and M' be vector-areas, aM. b%M' = (aifgJf0 6 - (aM ) gif' = (if \W) b - (aMb) %M'......(vi). And as particular cases aM. b%M = M26 - (aJf6) %M\ {M\M')a ).....................(vn)*
326 328] GEOMETRICAL FORMULA. 529 327. Geometrical Formula. (1) To find the foot of the perpen- dicular from the vertex d of the tetrahedron abed on to the opposite face. The required point [cf. 317 (4) and 325 (1)] is abe.dft (be + ca + ab), and by the transformation of equation (vii) this can be written (be + ca + ab)2 d - (abed) % (be + ca + ab). (2) To find the line perpendicular to two given lines. Let G1 and G2 be linear elements on the two given lines. Then CjU . G2ll is a vector-area perpendicular to the required line. Hence % (CjU. C2\\) is a vector parallel to the required line. But the required line intersects both the lines Cx and G2. Hence it lies in both the planes G1%(G1U. G'2U) and O2g ( 72U . G2U). The required line is therefore C1%(C1VL.C2VI).C2%(C1VI.C2VI). (3) Now if the two lines Ox and G2 are given respectively in the forms ! ! and a2u2, this expression for the line of the common perpendicular can be transformed by equation (i). For C$ (G^X. G2U) axu$uYu2 = ax. u$u{ii2 = fa \u2) afiih - ( ! K) a$u2, and similarly (72g (02U. GJ\) = (^ | u2) a$u2 - (u2 \ u2) a,^. Hence the required line is now [cf. 102 (7)] in the form 1^) (n2 \u2) a$U2 328. The Central Axis. (1) It follows as a particular case of 318 (9) that any system S can be written in the form an -f- ?3- ;w; where a is any point on a certain line parallel to u, called the Central Axis, and w is called the pitch. Also [cf. 318 (9)] (2) It is obvious that S can be written, without the use of u, in the form a.S\X + tsrg (SU). Also other expressions for isj are found from (818) = ^2 Hence (3) We are now prepared to discuss the signification of the addition of any vector-area M to any force system S. W. 34
530 VECTORS. [CHAP. II. It has been proved [cf. 309] that the addition of M to any linear element C1 in its plane is equivalent to the transference of Cu kept parallel to itself, to a new position, so that in the transference G1 sweeps out an area \l(Mf with the aspect of M. Let S be written in the form au + vt%u. Then if M and %u have the vector w in common, M can be written in the form Xgw + wu. Thus 8 4- M = au + vr%u + \$u + wu Hence the component of M which is parallel to the central axis (namely, wu) transfers the central axis according to the ordinary rule for transferring a linear element parallel to the plane of the vector; and the component of M which is perpendicular to the central axis (namely, \%u) simply alters the pitch. It. will be noticed that the principal vector of S is unaltered. 329. Planes containing the Central Axis. (1) The plane through the central axis of 8 and parallel to any vector v is where ct is the pitch of 8. For, writing S in the form au 4- rg^, the required plane is auv = (8 vrftu) v = Sv zrv^u = Sv w (v \u) U. If v be perpendicular to $U, the plane becomes Sv. (2) Hence the plane containing the central axis of 8, and perpendicular to the given vector area Mt is For by subsection (1) the plane is SgJf - r (gJf . ISM) XI Now (gjf iSU) =(\eM. \SU) = (eM. SVL) = (eMVL. S) = (MS); since the three terms eM, S and U, form a pure regressive product [cf. 101 (3)]. 330. Dual Groups of Systems of Forces. (1) The metrical pro- perties of dual groups and therefore also of quadruple groups [cf. 169 and 170] can now be discussed. Let Si and S2 be any two systems defining a dual group. Let Sx be written in the form a{u + ^gt^, and 82 in the form a2v2 + vr2 $V'2' Then (S^) = (a^v^^) + vr1 (a^igO + ^2 (oatfegvi) = (axvxa2v2) + fa + tja) (t^ \v2). Accordingly [cf. 324 (6)] if 0 be the angle between aYvx and a2v2f and 8 the shortest distance between them, } {d sin 9 + (Wl 4- w2) cos 5}.
329 332] DUAL GROUPS OF SYSTEMS OF FORCES. 531 (2) Let -Tf- o\t\9 /c*Ttv be called the virtual coefficient of the two systems jSx and $2 This virtual coefficient can accordingly also be written in the form {d sin 0 + (' -1 4- tnv) cos 0}. The idea of the virtual coefficient, and this latter form of it are due to Sir R S. Ball in his Theory of Screws. (3) Since the condition that the two systems be reciprocal can be written d sin 6 4- (^i + sr2) cos 0=0, it follows that if the central axes of two reciprocal systems intersect, either they are at right angles, or the sum of the pitches is zero. 331. Invariants of a Dual Group. (1) A vector-area parallel to the two central axes is ^U . S2Xl. This vector-area is an invariant [cf. 177] of the system. For let Then 8H . flf'U = (X^2 - X^) 81H . 8JU. Hence all the central axes of the dual group are perpendicular to the same vector g (fl^U. #2U). (2) The plane through the central axis of Slt and parallel to the normal to the vector-area 81VL. 82U is, by 329 (2), 8$ (Silt. S2U). Hence the line of the shortest distance between the central axes of Si and S2 is But this linear element is an invariant of the group. For substituting 8 and Sf for Si and 82, we find by the use of the previous subsection Hence all the central axes of the systems of a dual group intersect the same line at right angles. Let this line be called the axis of the group. 332. Secondary Axes of a Dual Group. (1) In any dual group there are in general two, and only two, reciprocal systems with their central axes at right angles. For let 8 and 8' be two such systems. Then (88') = 0, gives Xi/*i (SA) + X^2 (8A) + (XiA*. + VpOOSA) = 0; and ( SfU|fif/U) = O, gives 4- (x^ 4- \2/O (^u | s2u) = 0. 34 2
532 VECTORS. [CHAP. II. From these two equations by eliminating fii/fa, we deduce V K = o. Let a//3 and a'/fi' be the two roots of this quadratic for Xx/Xg. Then oSx + ^flf, and aS1 + ^/S2 are two reciprocal systems with their central axes at right angles, and there are only two such systems belonging to the group Si82* (2) It follows from 330 (3) that the central axes of these two systems must intersect. Hence they intersect at a point on the axis of the group. Let this point be called the ' Centre of the Group/ Also let the two central axes of these systems be called the ' Secondary Axes of the Group'; and let the systems Sx and 82 be called the ' Principal Systems' of the group. Let the plane through the centre perpendicular to the axis be called the ' Diametral Plane of the Group.' 333. The Cylindroid. (1) Let the assemblage of central axes of a dual group, each axis being conceived as associated with its pitch, be termed a Cylindroid. (Cf. Sir R. S. Ball's Theory of Screws.) Take the centre of the group as the origin e, let eux and eu2 be the two secondary axes of the group and eu3 the axis of the group. Then eu^, eu2, eu3 are a system of rectangular axes. Assume that ult u2y u3 are unit vectors. Let istj and vr2 be the pitches associated respectively with eux and eu2y then we may write S g S2 = eu2 + /Gr23^2 (2) Then any other system S of the group, with its principal vector of unit length and making an angle 6 with ult is Sj cos 0 + 82sin 0. Hence S e^ cos 0 + u2sin 0) + iff1 cos 0. u2u3 + sr2sin 0. u3nx (e + s^s) (^i cos 0 + u2 sin 0) + ('OTi cos 0 + 3 sin 0) u2u3 + (vr2 sin 0 g3 cos 0) u3iij; where 3 is a quantity which can be expressed in terms of ml9 ot2, and 0. Now assume that the central axis of 8 cuts eu3 in the point e+ g3u3, and that -or is the pitch of S. Then (^ cos 0 -f 3 sin 0) ^3 + (^2 sin 0 f3 cos 0) u3n^ = tzrg: (iii cos 0 -h w2 sin 0) = sr cos 0. ^^ 4- 'cr sin 0. i^t/!. Hence vrx cos ^ + f 3 sin 0 = t r cos 0, -cj2 sin 0 !j3 cos 0 ts sin 0,
333 335] THE CYLlNDROlD. 533 Thus ot = vrx cos20 + -bt2 sina0, and 3 = (*sr2 ^ sin 0 cos 0. These equations completely define the cylindroid. (3) If x = e + i^i + fa M2 + ?3W3 be any point on one of the central axes, the equation of the surface, on which it lies, is obviously (4) The director forces of the group are the two systems of zero pitch. Hence the angles their lines make with eux are given by -ctj cos2 6 + ot2 sin2 0 = 0. It follows that the directrices of the group are real if ot^ be negative, and are imaginary if ot^ be positive. The distances from the centre of the points of intersection of the directrices with the axis are easily seen to be + \/{ ^i 3"2). 334. The Harmonic Invariants. (1) The harmonic iuvariant H(x) [cf. 179 (3)] of any point x can be written 2xSx . S2; where Sx and $2 are the principal systems of the group. Hence, with the notation of 333 (1), by an easy reduction H(x) = 2 (xeuiu2) e + 2 r2 (xeu^) i^ 2'gt1 (xeu.jis) u2 + 257^ {xun^u^ ud. (2) Thus if x lie on the plane eu^u^ H (x) is a vector. If x be the centre of the group, the harmonic invariant is the vector u6 parallel to the axis of the group. Also the harmonic invariant of us is the centre of the group. (3) Also by a similar proof the harmonic plane-invariant of euxu2 is the plane at infinity; and therefore conversely the harmonic plane-invariant of the plane at infinity is the diametral plane of the group. Hence H (U) = \eu{a2. This can be easily verified by direct transformation. For H (U) = 2HSj. S2 = 2ux (eu2 + ar 335. Triple Groups. (1) Any triple group defines a quadric surface, called the director quadric. It was proved in 186 (4) that if p, p1} p2, p% be the vertices of a self-conjugate tetrahedron of this quadric, then a set of three reciprocal systems are given by (i), where filt /lu2, fh are three numbers, real or imaginary, depending on the given group.
534 vectors. [chap. ll- Also one of the vertices of the tetrahedron may be arbitrarily chosen. Let the point p be taken to be the centre of the quadric, then ply p2} p3 are on the plane at infinity, that is to say, they are vectors. Hence if the centre of the quadric be called c, and v1} v2, v3 be unit vectors in the directions of three conjugate diameters, a set of reciprocal systems can be written, Sj = CV1 S2 = Thus if Slf S2, S3 be any three reciprocal systems of the group, $iU, S2U, S3VL are the directions of a set of three conjugate diameters of the quadric. (2) If u1} a2i u3 be three unit vectors in the directions of the principal axes of the quadric, then the three corresponding systems, . 2j = CUX + VTyVvUz CUi + WiS^i,] 22 = cm2 + W2W3W1 = cw8 + vr$u2,1.................. (iii), S3 = cu3 + iz3ihu2 cu3 + i*t$u3 , j will be called the principal systems of the group. 336. The Pole and Polar Invariants. (1) The polar invariant of the point x is denoted in 183 (3) by P (a?) and is the polar plane of x with respect to the quadric [cf. 185 (1)]. Similarly the pole invariant of the plane X is P (X) and is the pole of the plane X. Thus the centre of the quadric is the point P (ll). (2) Now let Sly S2, 83 denote a set of reciprocal systems. Then P(U) = 2U^ .S2.S3 = 2 U. 82.S3. But 2 SjU. S2 is the diametral plane of the dual group S [cf. 334 (3)]. Hence the centre is the null point of this plane with respect to 83. Accordingly the diametral plane of any dual subgroup contained in the triple group passes through the centre of the quadric. Also the centre of the quadric is the null point of such a diametral plane with respect to the system of the triple group reciprocal to the corresponding dual subgroup. (3) It may be noticed that, if Hx (x)} H2 (x), H3 (x) denote the harmonic invariants of x with respect to the dual groups S2S3 S38ly S respectively, then [cf. 184 (2)] Xc = P (U) = J?! (^U) = H2 (S2U) = H3 (83U). These are four alternative methods of expressing the centre of the quadric. (4) Since the diametral planes of the three dual groups are c.SJX.S3U, c.S3U. SjU, c.81U.S2U, it follows that they intersect in pairs in the edges c . 8$, c . 82VL} c. 3U.
336 338] THE POLE AND POLAR INVARIANTS. 535 (5) The diametral plane, which bisects chords parallel to any vector v, is obviously P (v). Thus the diametral plane, of which the conjugate chords are parallel to any line L, is P (ill). The diametral plane parallel to any vector area M is P (U) M; the vector parallel to its conjugate chords is P {P(U) M). 337. Equation of the Associated Quadric. (1) The condition, that any point p is on the quadric, can by 187 (6) be written in the form Now let p1 be the length of the semi-diameter c . ^U. Then the point p = c + piVj is on the quadric. Hence by 335, equations (ii), p81. S2 p81. aS2 . 3 And pjSd ,S2.83.p = - Hence o - - i* ** - Hence Pl - ^s- 4 (c . with similar expressions for p2 and p3. (2) Let Si, 22, 23 be the principal systems of the group, and ult u.2y u3 unit vectors parallel to the principal vectors of 2i, 22, 23. Let it be assumed that Si = CUX 23 = CMs 4" T8 M3 = CU3 + Let be the point c + ^ih -r tj2u2 + 3w3; then fx, f2J are the rectangular co-ordinates of x referred to the three axes cult cu2, cu3. Also P(a?) = 2a?21.21.2. = 2 { where c is the centre of the rectangular normal system. The equation of the quadric is \xP (x)} = 0, that is 338. Normals. (1) The vector parallel to the normal at a point a? on the quadric is g [P (x) Uj; and this can be transformed into \P(x- c), where a; is a unit point, and c is the centre of the quadric. The first expression requires no proof, if it be remembered that P (x) is a planar element in the tangent plane at x.
536 VECTORS. [CHAP. II. (2) To prove the second expression, we have Now, c. P 0) U = (cU) P (x) - {cP (x)} U = P (x) -{xP (c)} U. But P (c) = XU. Hence {xP (c)} = X. Hence g [P (x) U] = \P(x) - X |U = \{P (x) -P(c)} = \P(x - c). Thus the plane P (x c) is the diametral plane perpendicular to the normal. 339. Small Displacements of a Rigid Body. (1) If any rigid body be successively displaced according to the specifications of two small congruent transformations [cf. 268 (1) and 303], it is obviously immaterial which of the two transformations is applied first, so long as small quantities of the second order are neglected. JMow let e be any origin and eu1} eu2) eu3 any set of rectangular axes, uly u2, u3 being of unit length. Assume the three small translations defined by the vectors Xa^, Xa2w2, Xa3^3, and the three small rotations with axes eu1} eu,2, eu3 through angles \S1} XS2, XS8 successively applied in any order; where X is a small fraction whose square may be neglected, and i a2, *3, i, S3, 3 are not necessarily small. (2) Then any point x e + Xgu becomes Kx, where Kx = e + ( + Xotx + XS2^3 XS3 2) uY + (f 2 + Xa2 + \hj;i XS^) u2 Hence the combined effect is equivalent to the combination of the small translation X (a1nl + c^ + 0Lsu3)f and of the small rotation round an axis e (BjUj + B2u2 + S3u3) through an angle X ^/(Sf + 8.22 + S32). But by properly choosing a1? a2, a5 and Sl7 82, S3 these can be made to be any small translation and any small rotation with its axis through e. Accordingly the above linear transformation is equivalent to the most general form of small congruent transformation. (3) Let S denote the system of linear elements a^Us + cizUsUj 4- a3uxu2 + h^eu^ + S2ew 2 + 83eu3. Then xS = (o^ 4- S2|3 - S3f2) eu2u3 + (a2 + S3^ - S^3) eu^ + ( s + ^2 - S2 ) eu t + (aj + a2 + a3) w^./^. Hence Kx = x + Xg ( ?. U). Let /S be termed the system of linear elements associated with the trans- formation, or more shortly the associated system. And conversely XS will be used as the name of the transformation. (4) If two small congruent transformations X^ and Xg j be successively applied, then neglecting small quantities of the second order the combined effect is that of the single transformation X^ + X^.
339, 340] SMALL DISPLACEMENTS OF A RIGID BODY. 537 For K,x = x + X$ (Sxx. U), neglecting It is now obvious that every theorem respecting systems of linear elements possesses its analogue respecting small congruent transformations. (5) When 8 is a single linear element through any point e, the trans- formation \S is a rotation with its axis through e. When S is a vector area, the transformation \S is a translation in a direction perpendicular to the vector area. The transformation \S can be decomposed into two rotations round any two conjugate lines of S. (6) Let S be written in the form av + rgv, where v is a unit vector, then Kx = x + Xg (avx . U) + X rg {gv . x . U}. Now g {%v . x . U} = g {gi;} = v. Hence iT# = # + Xg (ava? . U) + X-cjv. Now Xav denotes a rotation through an angle X round the axis av; and Xts-gv denotes a translation parallel to v through a distance Xtc. Thus the axis of the transformation is the central axis of the associated force system, and the pitch of the force system is the ratio of the distance of the translation to the angle of the rotation. Let this pitch also be called the pitch of the transformation. (7) Whatever point x may be, we may write S in the form xv + M. Hence Kx = x + Xg/¥. In other words, \Sx. U defines the translation re- quired to bring x into its final position. 340. Work. (1) If the force F pass through the point x, and x be displaced to + Xv, where v is a vector and X is small, then the work done by F is X {F%v). This is obviously in accordance with the common definition of work. The work can also be written in the equivalent forms, X (FU \ v) and \(v\FVL). (2) Let F=xu, and let \v be the displacement of x produced by the congruent transformation XS. Then v = %(8x. U). Hence ga = 8x. It And the work done by F is X {xu .(8x.Vl)} = -\{u.x (Sx. U)} = - \ {u xS}=\ (FS). It then follows, that the work done by a force F during the small con- gruent transformation \8 is the same at whatever point of its line of action F be supposed to be applied.
538 VECTORS. [CHAP. II. (3) Let Fly F2, etc. be any number of forces acting on the rigid body during its transformation. Then the sum of the work done by them is where S' is the force system Fx + F2 + .... Hence the work done by the force system 8' during the small congruent transformation X8 is equal to that done by the force system 8 during the small congruent transformation \8'. Also if S and S' be reciprocal, the work done is in both cases zero.
CHAPTER III. Curves and Surfaces. 341. Curves. (1) Let the point x (= e + Sf-it) be conceived to be in motion, so that , 2, 3 are continuous functions of the single variable t, which is the time. Then x -f #St = e + 2|?w + St2|m, is the position of the point at the time r + St. Hence the vector x which is 2 i*, represents the velocity of x in magnitude and direction. Similarly the vector x, or Xffw, represents the acceleration of x in magni- tude and direction. Let x, "x\ etc. be formed according to the same law. (2) Let a be the length of the arc traversed during the time t, and T-\-h r that traversed during the time t + St. Then ^t = V^2' (3) The tangent line to the path of x at the time t is xx, that at the time t + St is xx + #$St. Let Se be the angle of contingence between these tangent lines, then by 322 (2) __ *J{xx \xx) (4) Hence, if p be the principal radius of curvature of the path, 1 de V(^ |^) Therefore p* {(*)? (A)4 (if *
540 CURVES AND SURFACES. [CHAP. III. 342. Osculating Plane and Normals. (1) At the end of a second interval St, x has moved to x + 2xSr + 2x (St)2. Therefore the osculating plane is xxx. The vector factor of this product is xx. Hence the direction of the binormal is that of the vector $xx. The binormal is x%xx\ (2) The neighbouring osculating plane is x {xx + xxSt). The vector factor of the neighbouring binormal is (## + xxhr). Let 8\ be the angle of torsion. Then [cf. 322 (2)] f = A L v /J dr V since giu?. 3 (## + xx'Sr) = i$ . S##St. Now by 326 (4) equation (iv), $xx. g^ = (exxx) $x. TT d\ /(exxx)'2 (x)2 (exxx) v Hence -7- = ./- ttt^~ - = - yv~ Cv i \f KtA/JL/f \JuUDf Now let be the measure of the torsion. K Then -= **? /C (## XX) 1 Thus (3) The normal plane at x is x^x. The principal normal at x is the line N, where N = xxx. # = (^Si) xx (a?a?gi;) ^i; = (i?)2 ^ (i? j^) xx. The vector parallel to the principal normal is therefore (x)2x (x \x)x. 343. Acceleration. In order to resolve the acceleration along the tangent and the principal normal, notice that ., _(x\x) . (x)2 x (x jx) x Now let t and n be unit vectors along the tangent and principal normal respectively. Then x = (i?2)* t, (xf x - (x x) x = n. P Hence x = \,\2L t + ^~ n. Therefore the acceleration is equivalent to a component ^ ' alOng the (i;)2 tangent, and along the normal.
342 345] SIMPLIFIED FORMULAE. 541 344. Simplified Formulae. (1) In order to develope more fully the theory of curves, let us make use of the simplification introduced by the supposition that the curve is traversed with uniform velocity by the moving point. We may then take a as the independent variable, and use dashes to denote differential coefficients with respect to t. (2) Collecting our formulae, we have in this case O')2 = 1, and therefore (x' \af') = 0, (x'J + {of \af") = 0. p p K The tangent line is xx ; the normal plane is x%x ; the osculating plane is xx'x"; the binormal is x%xx"; the principal normal is xx". (3) To find Monge's polar line of the point x of the curve, notice that the normal plane at the point x + x' r is (x + x'$ r) {%x' -f %af'. So-}, that is, x$x' + (x%x" + U) S r, since (x'%x') = U. Hence the polar line is x$x'. (x$x" + U), that is, x$x . x%x" -f %x'. It is obvious that x%x . x$x" is some line through x. Assume that it is xv. Then v = xv. U = x%x . a$x". U = x%x . %x" = %x'x'\ by 326 (5). Hence the polar line of x is x%x'x" + %x. (4) The centre of curvature is the point where this line meets the osculating plane, that is the point (x%x'x" + %x). xx'x" = O'V)2 x + %x'. xx x". Now %di.xafx" = - \x .U. xx'x" = |^. x'x" = -(^ \x)x1 + (^ !^) x" = ^'. a?" Hence the centre of principal curvature is the point x -f -j--, ,/t-2 , that is the point x + pV'. 345. Spherical Curvature. (1) The centre of spherical curvature is the point where three neighbouring normal planes intersect, that is the point x%af . (a%x" 4- U). x%x" = (x%afx" -h %xf). x%af'\ by use of the transformation of 344 (3). Now by the rule of the middle factor [cf. 102 (7)] x%x'x" . x%x'" = (xftx'x" . %x'") x = iaftaT^afx") x. Also %x"'%afx" = | {ex"'. exx") = {ex'x"x'") \ e = (exx"x'/f) U. Hence a%afx". xftx'" = {ex'x"x'") x. Also by 326 (5) %x'. x%x"f = g#"V = - g^V". Hence the centre of spherical curvature is %x'x'" x - t , , , , that is x g*k%x'x'". (exx x )
542 CURVES AND SURFACES. [CHAP. III. (2) Therefore, if p1 be the radius of spherical curvature, p* = pW {afaf'J = pV {{af'J - ' \x"J) = pV { '")2 - O" K')2} = A2 O'")2 - *2- (3) Now O"')2 can be found by squaring the determinant {ex'x"x"f). Thus [cf. 342 (2)1 -L = (eafaf'af" \ex'x"x'") o, o, Now - = (^71 x") = -(af\ x'"). And - ~ ^ = 2 (#" | ^'). J_ _ ^!12 _ I ^ V _ I Hence Therefore p22 = / This is the well-known formula for p1. 346. Locus of Centre of Curvature. (1) It is easy to see that the inner products of the various differential coefficients of xy such as x', x\ etc., can be expressed in terms of p and k and their differential coefficients with respect to a. Let pl be written for A /\p2 + k2 (- } [. V T \d rJ) Then we have by successive differentiation O')2 = 1, {x\x") = 0, (x'J + ( ' l^') = 0, 3 (a?" |O + (x'\xw) = 0, 3 (^'")2 + 4 " |^iv) + (x \xv) = 0, and so on. Also (V')2 = = (x'\xf"). Hence (x" \x'") =----- ~- = - (^' |^iv). Again (*'")2 = ^ + = ^ sa7- And ( ' |O = X Also ^(-wl-0-(0' + ^K)-^^ Thus (x" xiv) is expressed in the required form, and so on.
346,347] locus of centre of curvature. 543 (2) Let y be the centre of principal curvature of the curve at the point x Let a be the arc of the locus of y measured from some fixed point. Then when a is the independent variable, and y is used for ~, y = x + pV, y = x' + 2p - x" + pV". Hence f^-Y = {yf = 1 + 4p* (^Y(a")2 + 2p* (x \af") Therefore =- = . (3) Also if Se' be the angle of contingence corresponding to 8a', then -du t y = ( L-f-2 1 ^ I + Thus by mere multiplication -^-, and hence -^ ,, is expressed in terms of p and k and of their differential coefficients with respect to r. 347. Gauss' Curvilinear Co-ordinates. (1) Let x be any unit point on a given surface. Then the co-ordinates of x referred to any four reference elements may be conceived as definite functions of two independent variables 0 and j . Then the two equations, 0 = constant, and / = constant, represent two families of curves traced on the surface. (2) Suppose that the unit point x + 8x corresponds to the values 0 + 80 and f + 8(f of the variables. Then 8x is the vector representing the line joining the point (0, f ) with the point (0 + 80, f + 8 f ) of the surface. Also 8x can be written in the form OX = {X^U + X2O p) + ^ \XnOv | + ZX12OUO(p + X22O(p^*) + ... , where xl3 x2 xU) x12, x^ are vectors. Hence, if e be the origin, (e k) = 0 = (e \x2) = (e \xlx) = (e \x12) = (e 1^). (3) In order to exhibit the meanings of these vectors, let e be any origin and euu eu2, eu3 rectangular unit axes. Then x is e 4- ^u, and x1 is 2 rfj 1*, a?2 is S ^ i*, a?u is S ^ u, and so on. d" v p vtf
544 CURVES AND SURFACES. [CHAP. III. dx It will at times be an obviously convenient notation to write ^ for x1, dx r d2x r , ^7 for Xo, ^zz for ocn, and so on. 0 p Gv (4) The distance 8a between x and x + 8x is given by (8a)2 = (8x)2 = (x1 \xY) (80)2 + 2 (xY \x2) 808 f + (#2 \x2) (8 f )2. (5) The tangent line of the curve joining the points is x (xx80 + x28cf ). Now let ^ + o\# be a neighbouring point on the curve cf = constant, and x -f 2# on ^ne curve ^ = constant. Then x H- o\# = ^ + ^S^ + 2" ^n (^^)2? ^ + ^2^ = ^ + x28 f + 2" ^g^r* Hence two tangent lines to the curves and 0 respectively are xx1 and ##2. Accordingly the tangent plane at x is xxxx2. The normal at x is , (6) The angle o between the tangents at x to the 6 curve and the (f curve is given by cos w = /f/ ; ' ---------rr, sm = V {(^ | #0 (#21 x2)} (7) Let / be the perpendicular from x + e on to the tangent plane. Inen let. 5 oz4 (4) 01/ = vXu) (80)2 + 2 (ea dVjs) 868 f 2 sKpCyK^ XXX2) 348. Curvature. (1) Let p be the radius of curvature of the normal 1 /^ \2 section through x and x + 8x. Then p = - ^-^-. = V(A j^) {(^ k) () + 2 (^ \x2) 808 f + (a?81 (ex Xn) (80)2 + 2 (e^^) g^S ^ + (^^^^22) (8 f )2 (2) Now seek for the maximum and minimum values of p when the ratio of 80 to 8 f is varied. Let pY and /o2 be the maximum and minimum values found, and let 801 to 8^x and 802 to 8 f 2 be the corresponding ratios of 80 to S . Then p1/^/(x1x2\x1x2) and po\\](xxx2 xYx2) are the roots of the following quadratic for f: ? - (^ |^2)}2 = 0. Hence 1 1 I P1P2 K^i I#1) (#21 ^2) - Oi (x2 \x2) (exxoc n) 2
348, 349] CURVATURE. 545 (3) The expression for----can be put in terms of fa |^), (x2 \x2\ fa \x2), Plp2 and of their differential coefficients with respect to 0 and f . For, since by 347 (2) 0k)= 0 = ... = and (^n k), 2= | fa [Xj), (^ Hence fa\xa), fa |, fa\x2), (x2 fa\xn), fa\xn), 0 , fa\x3), fa\xu) fa\x2), (x2\x2), fa\x12) + {fa \xj fa \x2) - fa \x2f) {(xu |a?a) - fa2 !# )}. Now let (^ Then (XY \X2)12 = (Xn ^ , etc. stand for ^ (^ i = 2 (^ |xn), fa |.^)2 = 2 i = 2 (a?2 |a?u), fa \x2% = 2 2) 4- (^1 |^i2), (a?! I#2)2 = fa* \x2) + fa , etc- 2 (x1 \xm)t (x2 \x2)n = 2 (# 2) + (^112 I X2) + fa \xm). 2 (x2 \xU2), Hence (^n 1^) - (x12 \x12) = fa \x2)12 - ^ fa {x^ - \ fa \x2)u. Thus----can be expressed in the required manner. 349. Lines of Curvature. (1) Also if and stand for p1j J(x1x2\xlx2) and p.2/^/(x1x2\x1x2) respectively. Then the ratios S^ to S / 2 and S02 to S 2 are given by - Ox I a?,) and - fa \x^' - fa \x2) (2) By the aid of the quadratic for and 2 which have been found, it can easily be seen that the lines of curvature are given by xx) (ex1x^Bl2)\ (S0)2 - fa \xx) (exjx n)} 808$ - (a?! \x2) (exjx^)} (8 f )2 = 0. W, 35
546 CURVES AND SURFACES. [CHAP. III. And therefore it follows that (x, O 80,802 + (xl \x2) {80x8 f 2 + 80^} + (x2\x2) %8 2 = 0. (3) Let x + 8x and x+ 8'x be two neighbouring points to x, where 8x = xx80 + #2S -f ..., and 8'x = xx8'0 + x28' j) + .... Then the angle yfr between the two tangent lines x8x and x8'x is given by (8x\8'x) C0S^ (x, x,) 808'0 + p! \x2) (808' j + 8 r8'a (808' f - Hence So-SVcos ^r = (^ IarO S^^^ + (xx \x2) (808 4- S'08^) + (^21 2) 8 f 8' f , and [cf. 347 (6)] 8 r8' r sin -^ = (808' j - 8'08 f ) sin o V1O1 ki) (^2 | 2)}. Corollary. The tangents to the lines of curvature at x are at right angles. (4) The conditions that the 0 curves and the j curves should be lines of curvature at each point are, from subsection (2) and the corollary of subsection (3), that the equations, (x1 \x2) = 0 = (exyXtf^, should obtain at each point. 350. Dupin's Theorem. (1) Let x be conceived as a function of three variables 0, p, yfr. Then the equations 0 = constant, } = constant, and yfr = constant, determine three families of surfaces. On the surface, 0 = constant, x is a function of the variables f and yfr; on the surface, / = constant, a function of yf/ and 0 ; on the surface, yfr = constant, of 0 and f . Let = #i3 ^-7 = xi,i ^-r-=^, with a corresponding notation for the da o p ' ayfr ' higher differential coefficients. (2) Now suppose that the three families of surfaces intersect orthogonally. Then OilO = 0, (#2|#3) = 0, (^3|^) = 0 ..................(1). Hence by differentiation (#13 ! ^2) + (^1 !^23) = 0, (^2! x3) + (x21 xn) = 0, (#23 xY) + (xs 1 xl2) = 0 Hence (#2 i^) = 0 =(x2 \x31)=(xs \xl2) ........................(2). The condition that the lines of intersection of 0 and f with the surface yfr should be lines of curvature is (ex^x^c^) 0. But from equations (1) x1x2 = ^xs, and hence we may write xYx2 = And therefore from equations (2) = X (e%x$, x12) = X(x^\ xXo) = 0.
350 352] dupin's theorem. 547 Hence the lines of intersection are lines of curvature of the surfaces on which they lie. 351. Euler's Theorem. (1) Let the curves 0 = constant and f = constant be lines of curvature, so that Let pj be the radius of curvature of the normal section through xx1 and p2 of that through xx2 The radius of curvature of any normal section is given by {(, ^(y (2 \x2 (2) The angle - fry which the tangent line xSx makes with the tangent line xxu is given by +( #o \x2) . _ ______ TT 1 cos2 ilr sin2 Hence - =------r H--------T . P Pi P2 352. Meunier's Theorem. (1) The principal radius of curvature of the curve, f = constant, is The radius of curvature of the normal section through xxl is = Hence lP - / (2) The osculating plane of the curve / is xx1wUt the normal section is xx-^x1x2. If % be the angle between these planes, it is given by But Therefore {oo^x^c^ \ x1xu) = (^ | ^ And Hence cos v = Note. I do not think that any of the formulae or proofs of the present chapter have been given before in terms of the Calculus of Extension, 35-2
CHAPTER IV. Pure Vector Formulae. 353. Introductory. (1) A simple and useful form of the Calculus of Extension for application to physical problems is arrived at by dropping altogether the representation of the point as the primary element, and only retaining vectors. The relations between vectors of unit length will give the expressions in terms of the Calculus of Extension for the formulae of Spherical Trigonometry. Also many formulae of Mathematical Physics can be immediately translated into this notation. These vectors may [cf. 210 (3)] also be conceived as the elements of a two-dimensional region, and their metrical relations are those of two-dimensional Elliptic Geometry. (2) Let i,j, k represent any three unit vectors at right angles. Then any other vector x takes the form fi + rjj 4- f A\ (3) We will recapitulate the forms which the formulae assume in this case. It will be obvious that, as stated above, they form a special case of Elliptic Geometry. i = jk, \j=zki, | k = ij. \jk = i = \\i, \ki j = H^*, \ij = k = \\k. (j\lc)=O=(k\i) = (i\j). (4) The multiplication formulae are the ordinary formulae for a two- dimensional region: we mention them all for the sake of convenience of reference. Let x, y be any two vectors; and X, Y any two vector areas. Then Xy=-yxy XY = - YX. Also xy represents a vector area, and XY a vector: the vector area xy is parallel to both vectors x and y, and the vector XY is parallel to the intersection of the vector areas X and F,
353 355] introductory. 549 Again, (*|y) = (y|tf), (X\Y) = (Y\X)', and the result is in each case a purely numerical quantity. 354. Lengths and Areas. (1) The length of the vector x is V(* k). The angle 6 between two vectors x and x' is given by COS a (x\xf) . I {xx'\xxf) (2) Any vector area X takes the form l*jk + ^^* + fy. The magnitude of the area is *J(X \X). The angle 6 between two vector areas X and X' is given by T) . / Z'|Z')}' sm " V cos " vt(x (3) Also |X denotes a vector line of length \](X \X), and \x denotes a vector area of magnitude ^(x \x). It will be useful at times to employ the term ' flux' to denote a vector area. (4) Let and rj be the lengths of the vectors x and y, and let 0 be the angle between them. Then the magnitude of the vector area xy is V{ y I xy} = V{(# | a?) (3/12/) - O 12/)2} = sin 6. Again, let f and 77 be the magnitudes of the vector areas X and F, and 0 the angle between their planes. Then the length of the vector XY is 355. Formula (1) The extended rule of the middle factor [cf. 103] gives the following formulae: X .xy = {Xy)x-(Xx)y...........................(i); x.XY=(xY)X-(xX)Y........................(ii). The second can also be deduced from the first by taking supplements. (2) The same rule also gives the following formulae for inner multipli- cation : xy\z = (x\2)y-(y\z)x....................................(iii); z\xy = z.\x\y = (z\y)\x-(z\x)\y .....................(iv). (3) Also from 105, xy.XY=(xX)(yY)-(xY)(yX).....................(v). And writing \u for X and \v for F, we deduce (xy\uv) = (x\u)(y\v)-(x\v)(y\u) ..................(vi). Similarly, the supplemental formula = (X\U)(7\r)-(X\V)(Y\U)............(vii).
550 PURE VECTOR FORMULA. [CHAP. IV. (4) Two particular castes of the formulae (vi) and (vii) have been already used above, namely, (xy\xy) = {x\x)(y\y)-{x\yf.....................(viii); {XY\XY) = (X\X){Y\Y)-(X\YY..................(ix). It will be convenient to write any expression of the form (x\x) in the form O)2 or x% and (X\X) in the form (X)2 or X2. Thus (xy)1 stands for (xy \xy). 356. The Origin. By conceiving the vectors drawn from any arbitrary origin 0, any vector x may be considered as representing a point. Thus it is the point P such that the line from 0 to P can be taken to represent the vector x in magnitude and direction. This origin however is not symbolized in the present application of the Calculus. 357. New Convention. (1) Before proceeding with the development of this Calculus it will be advisable explicitly to abandon, for this chapter only, the convention [cf. 61 (1)] which has hitherto been rigorously adhered to, that letters of the Italic alphabet represent algebraic extraordinaries and letters of the Greek alphabet numerical quantities of ordinary algebra. As a matter of practical use and not merely of theoretical capabilities it would be found so necessary by any investigator in mathematical physics to continually form the Cartesian equivalents of his equations if only for comparison with other investigations that the capabilities of the Greek alphabet for the representation of numerical quantities would not be found sufficient. The German alphabet is found by most people difficult to write and to read. But let the following convention, which is a modified form of one adopted by Oliver* Heaviside, be adopted. (2) Let all letters of the Greek alphabet denote numerical quantities. Let all letters, capital and small, of the Latin alphabet without subscripts denote respectively vector areas and vectors; except that in formulae con- cerned with Kinematics or with Mathematical Physics t always denotes the time. Let i,j, k denote invariably three rectangular unit vectors. If x denote any vector, let x1} x.2, x3 be numerical quantities denoting the magnitudes of the resolved parts of x in the directions i, j, k respectively: so that x = x{i -J- x2j 4- xjc. Let x0 be a numerical quantity denoting the magnitude of x. Thus * Cf. 'On the Forces, Stresses, and Fluxes of Energy iu the Electromagnetic Field,' Phil. Trans. 1892.
356 359] NEW CONVENTION. 551 If X denote any Hux, let Xly X2, X3 be numerical quantities denoting the magnitudes of the resolved parts of X along the unit fluxes jk, hi, ij: so that X = Xxjk + XJci -h X3ij. Let Xo denote the magnitude of the flux, so that XQ = + V(Z)2 = {Xf + X? f (3) This notation avoids a too rapid consumption of the letters of the alphabet, and shews at a glance the relationships of the various symbols employed. We note as obvious truths; it X= \j\ then X0 = x0, X1 = xly X.1 = x.1) Xs = x3. Also we note that if w= X, then x X = X, and the same results follow. 358. System of Forges. (1) Let forces represented by the vectors fy f\ ..., act at points denoted by the vectors x, x\ ..., drawn from any assigned origin. Then any force f at x is equivalent to f at the origin and a vector area xf representing the moment of f at x about the origin. (2) Hence the system is equivalent to 2/ at origin and the vector area Xxf, representing a couple ; which may be called the vector moment of the system about the origin. If L be this 'vector area/ Ll9 L2, Ls are the three moments of the system about axes through the origin parallel to i, j, k. 359. Kinematics. (1) Let any point in space be determined [cf. 350 (1)] by the three generalized co-ordinates (0, j , yfr). It will be called the point (0, , tjr). If the point be referred to three rectangular axes, the rectangular co-ordinates will be written xly x2, xS) and the point will be represented by the vector x. If 0, , yfr be conceived as the co-ordinates of a moving particle, they are functions of the time. Let u be the vector which denotes the velocity of the particle at each instant; then corresponding to each position (0, f , yfr) there is a definite velocity u. Hence u must be conceived as a function of 0, f yyfr: that is to say, if i, j, k be any three fixed rectangular vectors, and u = u1i+ u2j + ujc, then i^, u^, us are functions of 0, f } yfr. Since 0, , yfr are functions of the time, u can also be conceived as a function of the time. (2) Let u be the velocity of the point at the time t} and u + uht at the time t + St. Then when St is made infinitely small, u is the acceleration.
552 PURE VECTOR FORMULA. [CHAP. IV. Also evidently, du (3) The aspect of the osculating plane of the curve traced by the point is represented by the vector-area uu. The binormal is represented by the vector \uu. The normal plane is represented by the vector-area j u. The principal normal is represented by the unit vector n, where uu __ (u \u)it (u \ii)u */(uu | u)2 *J{(u I u) (uu I uu)} ' The distance traversed in the short time St is *J(u)2. St (4) The angle Se between the directions of motion at the times t and t + St is given by uu) 6e = sin Se = ./ v , ,, . ot. The radius of curvature is } . , ^ . V(ww|wtf) (5) Thus u = ,, XJ This represents the ordinary normal and tangential resolution of the acceleration. 360. A Continuously Distributed Substance. (1) Many branches of Mathematical Physics depend upon the investigation of the kinematical properties of substances (ordinary matter or some other medium) distributed continuously throughout all, or some portion of, space. The continuously distributed substance will possess various properties dependent on its motion and on other intrinsic properties. Let any quantity associated with a particle of matter, which does not require a direction for its specification, be termed scalar, according to Hamilton's nomenclature. (2) Then scalar quantities, such as the density, and vector quantities, such as the acceleration, are associated at each point with the existence of the continuously distributed matter. These quantities, scalar or vector, may be associated either with the varying elements of matter occupying given points of space, or with the given elements of matter occupying varying points in space. (3) If the quantities be thus associated in the first way with the given points of space, then the co-ordinates say 0} f , yjr of any point are not to
360] A CONTINUOUSLY DISTRIBUTED SUBSTANCE. 553 be considered as functions of the time. Let % be any scalar function of the matter at the point (0, , i/r) at any time t, then at the subsequent instant t + St a fresh element of matter occupies the position (0, f , y(r). Let its corresponding scalar function at the time t + Bt be x + - St. Thus x is dt conceived as expressed in the form ^(0, , yfr, t), where 0 / , \fr are not functions of t The operators d0' 3 / ' df dt applied to x have therefore the relative properties of operators denoting partial differentiation. Call ^ the stationary differential operator with respect to the time. (4) Similarly if u be any vector function of the matter at the given point {0, f , yfr) at any time t, then at the subsequent instant t + St the correspond- ing vector function of the new element which occupies the position {0, f , yjr) can be written u 4- -^- St. Also it is obvious that ot dn _ dux . du2 . du3 j dt~~dt% + ^t3 + Tt ' Let ^- and ^ be abbreviated into n and x', or mto Mt %t- ot ot (5) We shall assume, except where the limitation is expressly stated, that the scalar and vector functions spoken of are continuous functions of the variables: and that if x ^e any scalar and u any vector, %, uly u2, u3 have finite and continuous partial and stationary differential coefficients with respect to 0, , i|r, t (6) If the quantities be associated with the given particles of matter, let the co-ordinates 0, f yfr mark the position of any given particle at the time t. Then at the subsequent time t 4- S , the co-ordinates of that particle have become 0 -f 0 t, f 4- f , ^ + yjrSt. Also if % be any scalar function of that particle at the time t} the same function of the same particle at the time t + St will be denoted by x + X^ or by % + -^f St. The function % can be dt conceived now as a function of 0, , yjr, t and written 0, ^, ty are functions of the time. Thus Similarly if ^ be any vector associated with the particle, at the time t+ St the same vector function associated with the same particle is u+iiSt orw+-j- St.
554 PUKE VECTOR FORMULA. [CHAP. IV. m, i , du diii . , du2 . du 7 Thus obviously _=_ + _ ; + _i; where *~* + '8* + *a* + *a*' with two similar equations. Hence du du aB d* = 8* + ^ Call -j- the mobile differential operator with respect to the time. 361. Hamilton's Differential Operator. (1) Let the position of any point be denoted by the vector x which is represented by the line drawn to it from any arbitrarily assumed origin. Then xlt x2, xz are the rectangular co-ordinates of the point referred to axes parallel to itj, k; and xlt x2, x$ may be conceived as taking the place of the unspecified co-ordinates 6y j yfr of the previous investigations. (2) Let x be any scalar function of position at a given instant. Then i^+j~t+k^ obviously represents the rate of change of % at the point x OX} CX2 uX$ iii the direction of the normal to the surface ^ = constant, which passes through x. It follows that the function i ^ + j ^ + k ~- is independent of 0 3a?x J 3#2 H$ the directions of the vectors i, j, k so long as they are a rectangular set. 7) 7\ r) Let the symbol operator, i- I- j h k ^ , be written V, and called OXi * 0X2 OXg Hamilton's* Differential Operator, or more shortly, the Hamiltonian. Its properties were first fully investigated by Prof. Tait** for the very similar case of quaternions. (3) The Hamiltonian may accordingly be conceived as operating on a vector by means of the conventions Vu = V^!. i + Vu2 .j + Vw3. k, (V \u) = (Vth \i) + (Vu, \j) + (Vuz I ft). (4) Hence Vu = ]k U- -^} + ki (_i-^-i +17 _-r-_J . 1 V9#2 3^3/ \dx3 dxj J \3aa dxj I V'M- is called the Curl of the vector u, Vu is the Curl-flux of the vector it. /-\ ai /r7I N duri du2 du3 (o) Also (V|w) = ^-+_ + _. (V I u) is called the Divergence of the vector u and is a scalar quantity. * Cf. Hamilton's Lectures on Quaternions, Lecture vu. 620. ** Cf. his Elementary Treatise on Quaternions, 1st Edition, 1867, 3rd Edition, 1890.
361,362] Hamilton's differential operator. 555 (6) It is obvious with this symbolic use of V that it can be treated as a vector as far as formal algebraical transformations are concerned, so long as in the product it is kept to the left of the quantity which it operates on, and so long as those quantities to its right on which it does not operate are noted. (7) Thus in accordance with the rest of our algebraical notation we may write V V^ in the form V2^, where [cf. 357 (2)] It is obvious that V^ = 0; Wu = 0. Again, V |V . u becomes V2u, which is VLX . i + V2u2 .j + V2m3 . k. (8) An important example of the possibility of formal algebraic transformations of expressions involving V is as follows : If a, b, c be any vectors, ab \c = (a \c) b (b \c)a = \(o \ab) = \(c \ba). Hence (b\c)a = b (c \a) j (c16a). Now putting V for both 6 and c and u for a. we rind v2u,=v (v | u) - j (V | Vuy. 362. Conventions and Formula. (1) The symbol V is to be assumed as operating on all the subsequent vectors in a product in which it stands, in the absence of some special mark attached to a vector. If a vector such as v is not operated on by a preceding V, let it be written with a bar on the top, thus v. For instance, Vuv implies that V operates both on u and v; but Vuv implies that V operates on u and not on v, and Viiv implies that V operates on v and not on u. Similarly V (u \ v) implies that V operates on v and not on u. (2) The advantage of affixing a sign to a vector not operated on by V is that, as far as the vanishing of a product is concerned owing to the formal laws of multiplication [cf. 93 (4)] a vector behaves differently according as it is or is not operated on by V. For instance if u, v, are any two vectors, uvu = vuu = 0. This is true by reason of the formal laws of multiplication. Now substitute the symbolic vector V for v, then tiVu = Vuu, and this is not zero. Thus it is convenient, as far as formal multiplication is concerned, to reckon u and u as different vectors. It sometimes conduces to clearness in tracing the algebraic transformations to preserve the bar over a vector even when it is placed in front of V ; thus Vme = uVu. In such cases the bar may obviously be placed or dropped without express mention.
556 PURE VECTOR FORMULAE. [CHAP. IV. (3) The following are important examples : V (u\v) = V (u\v) + V (u\v),\ . hence V(u\u) = %Vu* j........................^ ;* Vuv Vuv + Vuv = vVu uVv..................... (ii). Also [cf. 361 (8)] u\Vu = - | (Vu u) = | V (u j u) - (u \V) \u = i \Vu? -(u \V)\u. But \(u \Vu) = \u . Vu = Vw- |u. Therefore Vw | u = (w | V) u - % Vu2,) or {u\V)u = Vu\u + \Vu? j........................^m^ (4) If the operation V is repeated in a product, a little care must be exercised so that the use of the bar may be unambiguous. For instance the following transformation exemplifies this remark. We wish to operate on Vu\u with V. The new operation of course operates both on the u and the u of the existing expression; since the bar merely refers to the existing operation V. Write Vu \u \u . Vu. Then V (\u . Vu) = V (\u . Vu) + V O . Vu), where obviously the newly placed bars refer to the V outside the bracket. (5) But the introduction of a new symbol, such as \v = Vu is often the simplest solution of the difficulty. For instance Vu\u becomes \vu. An important example is arrived at by operating with V\ on Vu\u. Write \v for Vu. Then V |. | vu = Vvu = uVv vVu, by equation (ii). But vV u = (v | v) = v2 = (Vu)2, and uVv = u . V \Vu = (u\V) (V \u) - u V'ht. Hence Vvu = (u\V)(V \u) - u\V2u-(Vu)2..................(iv), where v = | Vu. (6) It will be convenient to adhere to the further convention that V immediately preceding a scalar such as f does not operate on a subsequent vector unless some stop is placed between the V and the scalar. Thus V(j u has the same meaning as V j u, but V . f u implies that V operates on j u. This convention is useful in dealing with such expressions as V V^: it avoids the clumsy form V^Vifr. It is however often better to place bars where there is a risk of mistake, so as not unduly to burden the memory with conventions. (7) The preceding transformations have brought into prominence the symbolic operator (^|V). It is a scalar operator, and in the Cartesian notation [cf. S 357 (2)1 is Wj^ Vu2^ hw3^-=woT- , where da is an L c /J dxx dx2 dxs do- element of arc at the point x in the direction of u.
363] CONVENTIONS AND FORMULAE. 557 It follows that, if u be the velocity of the matter at the point x, d 3 3 3 3 3 , / irrx 363. Polar Co-ordinates. (1) The analytical transformations of V into polar and cylindrical co-ordinates can be easily established. Let P be the point x, and 0 the origin: let the position of P be defined Fig. 1. by the length p of OP, the angle 6 between OP and the direction of k, the angle f between the plane through OP and h and the plane hi. It may be noted that by the convention of 357 (2) p has also been denoted by x0. (2) Let r be the unit vector in the direction of OP, thus r = - = . p x0 Let v be the unit vector perpendicular to the plane through OP and Z\ positive in the direction of j increasing. Let u be the unit vector perpendicular to OP in the above plane and positive in the direction of 6 increasing. Thus (r \r) = 1 = (u\u) = (v \v); and (r \u) = 0 = (r\v) = (u\v); and u = | vr, v = | ru, r = j uv. (3) Also r = i cos f sin 0 + j sin sin 6 -f cos 0. Hence v = rk = (i sin j cos ) sin 0; therefore v jcos / * si*1 And i* = |w = i cos f cos 0 + j sin f cos 0 k sin ft
558 PURE VECTOR FORMULA. [CHAP. IV. (4) Again, V = + dp pdO p sin ( M -^a) + ( a,. 1 = ( Fj + fo -) = -; poo/ \ \p sm uo p/ \ \py \ \p' p (V\v)=(v du l psmO /u\r\ ( vcos0\ cot 6 Hence ? sin* (5) Again, let p be any vector function of the position of P, and let where r, v, w are the three rectangular unit vectors as defined above which correspond to the position of P, and mlt ? 72, zt3, are scalar functions of p, 0, f . Then (V 'p) = (r \ VOTl) + (w | V r2) + (v | Visr3) + r,(V |r) + ^2(v !^) + ots(V |v) 3^3 2-5TJ ^2 cot 6 d pdO 2-5TJ dp pdO psinddcfy p p (6) Again, let ? be the curl of p and let q = ^r + /c2?^ + ^v. Then \q = Vp = V^ir + VtEr2u + Vcr3v + OTjVr + w2V?t + s73Vv. Now Vr = w-h-v- = 0: P P _, ?' v cos 0 r? V 7^ = u - + ^ = h = ; p psmv p __ ( j sin d 4- i cos ^ ) (r sin ^ + w cos 6) cot 0 1 V 7; = y v-^-------- .---------------^ = V-----------------:-----pp-------/ =---------UV-------VV. p sin v p sin 6 p p TT , f 1 dsr2 , 1 d-GT3 tir^ cot 0~\ Hence \q = uv \-----: ^ ^r+-^-i :----- ^ j_ p sm ^ o / /) ^ p J f 1 d'UT1 d-STz ZJ3~] [~ 1 dtVj 3-572 CTol + vr , a ^7 - ^------+ ru - - W + T~ + |j sm 6 o p dp p ] [_ p dv dp p} gicot p sin dnr3 'dp 3ZB2 1 StlT! BT2 dp p dd p 364. Cylindrical Co-ordinates. (1) Employing cylindrical co- ordinates, let # denote the length ON in the annexed figure, and a denote
364] CYLINDRICAL CO-ORDINATES. 559 the length NP. Also let v denote the same vector as in 363, and to denote a unit vector parallel to NP. (2) Then v =j cos i sin , w = icos f +j sin0 ; and w = I vk, v = \ Jciv, k = | wv. Fig. 2. = V "a = 0, V w = V V o" _1 cr d2 92 d2 1 a ~da~2 + a2d f 2 dx.? cr w ivv V 0 ~~ a a (7 Also And Hence Again, (3) Let any vector p be written vy^u + written kxw + #2u + /c3/;. Then 13 = cr 8 f ^, and let its curl q be 9*3"! 1 9t*T2 B^ ^l da a d(f doo3 a -f
560 PURE VECTOR FORMULA. [CHAP. IV. Hence tc3= ^--------"5X + ~ Off T O(p T _ 1 9-573 9C72 j dcf 9#3 __ dzTj 9-573 doc?, d r 365. Orthogonal Curvilinear Co-ordinates*. (1) The formulae may be generalized thus: let I, m, n be three unit rectangular vectors associated with any point P, such that the system of vectors suffers a continuous change in direction as the point P moves in a continuous line to any other point P'. Let 8 r1} 8cr2i 8 r3 be elements of arc traversed by the point as it moves through small distances from P in the directions I, m, n respectively. (2) Let P be determined by three curvilinear co-ordinates 0l9 02t 0S) such that during the small motion 8al} 01 becomes 01 + S0lf and 02 and 0* are unaltered ; with two other similar specifications. Also assume that j S#! ^ 802 * 803 h\ h2 ns where hlt h2, h* are functions of 0ly 02) 03. Then[ef. 361(2)] V^+m^ + n^.. (3) Thus if f be any scalar function of 0l9 02 0^, Again, let p be any vector, and let p = tb^I + -572ra + fsr^n. Then Similarly Vp = +nl w\ zar3V?i...............(iii). * As far as I am aware, the methods of transformation of the present and the two preceding articles have not been employed before. The methods are the analogue in this Algebra of Webb's method of Vector-Differentiation, published in the Messenger of Mathematics, 1882, and fully explained and applied to this case in Love's Treatise on the Mathematical Theory of Elasticity, 119,
365] ORTHOGONAL CURVILINEAR CO-ORDINATES. 561 Thus when {V \l), (V |m), (V \n), VI, Vra, Vw have been obtained, the formulas for transformation are complete. (4) Now 0! = constant, 02 = constant, 03 = constant, form three sets of mutually orthogonal surfaces. Hence I = \mn = ~ V01 = ^ \V02V0s. ill fl2lt 3 Hence (V\l) = ^rV. V0,V03 + V02V03V ttl ^rV V0V0 + V0V0V L Now V. V02V03 = VV02. V 93 - V(92. VV03 = 0. Therefore Similarly Hence from equation (ii) (V W = ^A^.jA g- + ^ g- + ^ g- (5) Again, VI = V . 1V0, = V 1. V0, = ;-2 Vtf.V/j, = J /?j //j /i^ / _h2 j dhi h3 j dhY dh2, Aj dh3 7 A2 dh3 n - r I ^^ n^ Hence from equation (iii) .... _ hz dh^ hx dh2j 3 7 2 3 Similarly Vm = y^ ~ ran - r ~k I , ^^ = tt -IF n^ ~ r 5n mn- h3 ou2 hs dh2 h2 dh3) , GTj V to 3 Accordingly if q be the curl of p, that is |Vjt ; and if q be written then #! = Ms ( kq- i-----fiTT tx1 d \ w. 36
562 PURE VECTOR FORMULAE. [CHAP. IV. These formulae of course include as special cases the preceding formulae for polar and cylindrical co-ordinates [cf. 363 and 364]. 366. Volume, Surface, and Line Integrals. (1) Let dr stand for an element of volume at the point x. Let dS be a vector-area representing in magnitude and direction an element of surface at the point x. Then dS =jkd81 4- kidS2 + ijdS3; where dS1 = dx2dx3 dS2 = dx3dx1, dSs = dxxdx2. Let \dS denote the normal, positive when drawn outwards. Let dx be a vector line denoting in magnitude and direction an element of arc at x. Then dx = idxY +jdx2 -f kdx3, and dx0 is often denoted by da. (2) Then the well-known theorem connecting the volume and surface integrals of any continuous function of position within a closed surface is (3) Green's Theorem can be written JYj(V / |V^r) dr = jj( This can obviously be deduced from subsection (2) by writing / V /r for u, then (V \u) = (V \ j Vyfr)=(V(j \Vf) + ( V^). (4) Stokes' Theorem connecting line integrals and surface integrals is expressed by fj(Vu\dS)=f(u\dx), where the line integral is taken completely round any closed circuit, and the surface integral is taken over any surface with its edge coincident with the surface. 367. The Equations of Hydrodynamics. (1) Let the vector u denote the velocity of a frictionless fluid at any point represented by the vector x drawn from an arbitrarily chosen origin. Let p be the density at that point, and tzr the pressure. Let the vector f denote the external force per unit mass at x. Also let the vector q denote the curl of u, so that q = j Vw. The vector q defines the vortex motion at each point of the fluid: portions of the fluid, for which q = 0 at each point, are moving irrotationally. Then the fundamental equation of motion is
366 368] THE EQUATIONS OF HYDRODYNAMICS. 563 (2) Assume in the first place that the fluid is homogeneous and in- compressible : also that / is derivable from a force potential \|r, so that Equation (i) becomes -- This can be transformed [cf. 362 (7)] into (^ ) ........................(ii). But by equation (iii) of 362 (3) Hence ^+ \qu = - V ( - + ^ + |w2).....................(iii). at \p J (3) The equation of continuity becomes (V|^)=0....................................(iv). (4) These equations are independent of any special co-ordinate system. Thus let 6l9 02, 0z denote any set of orthogonal curvilinear co-ordinates, so that 0X = constant, 02 = constant, 03 = constant, denote three systems of mutually orthogonal surfaces. Let I, m, n and hly h2 h5 have the meanings assigned to them in 365. Let u = vji 4- v2m + v3n, q = kxI + k2vi + tezn. Then kx = hji3 f ?r^- y3 ^r ^ J , with two similar equations for ko \ov2 hz dV3 hj and k3. Now Z, m, n are independent of t. Hence equation (iii) splits up into three equations of the type hxd f'ur 1 And equation (iv) becomes ^J^L J^ iLj^ ........................(vi). ^_+ + =o ........................ d01 hji% d02 h2fi1 d03 h1h2 These are the general equations of motion of a homogeneous incom- pressible fluid referred to any orthogonal curvilinear co-ordinates. They include as special cases the equations referred to polar or to cylindrical co-ordinates. 368. Moving Origin. (1) Equation (iii) of the preceding article may be extended to the case of a moving origin. Suppose that the origin moves with velocity v, then v may be a function of t, but of course is not a function of position. 36 2
564 PURE VECTOR FORMULAE. [CHAP. IV. The point, which at time t is defined by the vector x, at the time t + St is defined by the vector x vSt $ Let ^r denote the stationary differential operator with respect to the ot time relatively to the moving origin, so that kt gives the rate of change at a moving point which is defined by the constant vector x drawn from the moving origin. Then !-! + !*). Hence equation (iii) of 367 becomes ^-(v|V)tt+|jtt = -v(- + ^.+4t*)...............(vii). (2) Equation (ii) of 367 becomes Now let u' = u v. Then u' is the velocity of the fluid at any point relatively to the origin. Substitute u + v for u and remember that (u |V) v 0 = (v | V) y, since v is not a function of x. Also if v be the acceleration of the origin, Su 8u' . st = Jt+v- Hence the equation of motion becomes ^ + * + (*'|V)*' = -?(-+*)..................(viii). The equation of continuity is (V | uf) = 0. (3) The curl of v! is the same as that of u, since \Vu=\V (u +v) = \Vu\ Hence equation (viii) can be transformed into Furthermore, since if is not a function of position, v = V (v \x). Hence finally (4) Therefore a uniform motion of the origin does not aflfect the form of the hydrodynamical equation, when the velocity is reckoned relatively to the origin.
369, 370] moving origin. 565 An acceleration of the origin adds a term to the force potential. The vortices are the same whatever motion be assigned to the origin. Therefore by suitable modifications of yfr, equations (ii) and (iii) of 367 may be looked on as the typical hydrodynamical equations, whether the origin be at rest or be moving in any way. 369. Transformations of Hydrodynamical Equations. (1) Opera- ting on (iii) of 367 with V But V^= ?3. ot ot Also V \qu = (V \u) \q + (V \u) \q - (V |g) ju - (V |g) |w = (u |V) | j - (g |V)|w; since (V |m) = 0 = (V \q). Hence by taking the supplement This can also be written (2) Again, operate on (iii) with V|. Now V|| = |( Hence Vqu = - V2 ( - + yfr + %u*)........................(xi). Also by 362 equation (iv) Vqu = (u | V) (V I u) - uV2u - (Vuf = - uVhi - (Vt^)2. 370. Vector Potential of Velocity. (1) Assume that there are no boundaries to the fluid which extends to infinity in all directions; also that the vortices [cf. 367 (1)] only extend to a finite distance from the origin. Now q = | Vw. Hence [cf. 355 (2) equation (iv)] Vq = V |Vm=|V(V \u)- |V2m=- \Vhi. Therefore by the ordinary theory of the potential, since by assumption q = 0 at all points beyond a certain finite distance from the origin, , 1 fff-ZL^;
566 PURE VECTOR FORMULA. [CHAP. IV. where q represents the curl of the velocity at the point x, and V stands for i j: , +j ~ , + k ^- , , and dr is an element of volume at x. (2) The integration may be assumed to be confined within any surface large enough to contain all the vortices and such that none of them lie on the surface. Integrate by parts, and remember that by the assumption q is zero at all points of the surface. Then [cf. 362 (1)] j u. = ^ ^(x-xj ' Now V -_* =_V- Hence |^ = i-///v q' dr' = ~ V [[ f .. q' ,drf......(xii). 4ttJJJ ^{x xf 4fir JJJ sj{x-xf v J Hence -7- III r. - 7-r dr is a vector such that u is its curl. Let this ^TrJJJ sf(x xf vector be denoted by p, then u = \ Vp. (3) Also by integrating by parts, it is easily seen that (V .p) = I- 4nr At since (V|5)=0. The vector p is called the vector potential of the velocity. (4) The same suppositions as to the absence of boundaries and as to vorticity enable us similarly to solve equation (xi). For by the ordinary theory of the potential equation (xi) can be trans- formed into 2+ constant = -i (If -7/V'gV/v dr\ - P Now integrating by parts exactly as above, Hence if L denote the flux 7 I 11 ,qU /x-dr} then 4nrJJJ \/{x xy
371] VECTOR POTENTIAL OF VELOCITY. 567 371. Curl Filaments of Constant Strength. (1) Let v be any vector wThich at each point is definitely associated with the fluid at that point: the magnitude and direction of v may depend, wholly or in part, on the velocity of the fluid and on its differential coefficients, and it may depend partly on other properties of the fluid not here specified. Let it be assumed that the components of v, namely vl9 v2y v3y and their differential coefficients are single-valued and continuous functions. Let r denote the curl of v, so that r = \Vv. Lines formed by continually moving in the direction of the r of the point are called the curl lines of the vector v. Since r fulfils the solenoidal condition, namely (V,r) = O, such lines must either be closed or must begin and end on a boundary. (2) A curl filament is formed by drawing the curl lines through every point of any small circuit in the fluid. If dS be the vector area at any point of a curl filament, then rdS is called the strength of the filament. It follows from the solenoidal condition by a well-known proposition that the strength of a given curl filament is the same at all points of it. Let any finite circuit be filled in with any surface, then by 366 (4), jj(rdS)=J(v\dx); where the line integral is taken round the circuit. (3) Let us now find the condition that the sum of the strengths of the filaments, which pass through any circuit consisting of given particles of the fluid, may be independent of the time. Also for the sake of brevity assume that the region of space considered is not multiply connected. The condition is /(, 1*0-0. This becomes l( Tt\dw) + j(v\dn) = 0' Now du = (dx\V)u. Therefore (v \du) = (dx IV) (v u) = V(v \u) \dx. Hence j(^\dx)+j (v\du) = 0, becomes J|J~ + V (v |W)| | ete] = 0. Now if ^ be any scalar function of x and t which together with its differential coefficient is continuous and single-valued,
568 PURE VECTOR FORMULA. [CHAP. IV. where the integration is completely round the circuit. Hence if yfr be some such scalar function, we deduce W Also sr + "lv * Now (u \V)v = V (v \u)+Vv \u = V (v | xZ) i7 .Vv = V (v\u) \ .u \Vv = V (v\u) - |m\ Hence ^- | ur -f V (y I u) = -=- Viir. ot Now put % = yfr + (# | m). Then g + | = _Vx..............................(xiv). (5) To eliminate ^ we operate with V, then Now V I /r = (V !r) \u (V |u) \r. But (v = 0 = (V|tt)|. Hence V | wr = (r | V) \u - (^ | V) \r. Therefore the equation becomes after taking supplements, r/r This is T7=(r|V)j(.................................(xv). (6) This condition should be compared with equation (x). It follows from the comparison that the strengths of all vortex filaments are constant. In other words, that if equation (xiv) be conceived as an equation to find the unknown vector r, then q is one solution for r. But q is not necessarily the most general solution. Thus there are other curl filaments in the fluid with the same property of constancy. (7) But equations (xiv) and (xv) are more general than these enunciations would suggest. For in the derivation of (xiv) neither the equation of con- tinuity for an incompressible substance nor the kinetic equation of fluid motion were used. It follows that if the motion of any continuous substance be assumed given, so that u is a given function of x and t, then any vector v, as defined in subsection (1), with its curl r which satisfies equation (xiv) is such that the curl filaments are of constant strength. (8) Equation (xv) involves the equation (V |^) = 0. Hence this equation holds for any incompressible substance moving in any continuous manner.
372] CURL FILAMENTS OF CONSTANT STRENGTH. 569 An extended form of (xv) can be deduced by writing (V \u) = 6, where 6 is a known function of x and t, since u is such a function. Hence V |w = (r|V)|tt (w|V)|r- 0\ | (|)| (|)| \r. Therefore ^ + (u | V) r = (r | V) u - 0r ; that is, j-=(r|V)tt-0r..............................(xvi). 372. Carried Functions. (1) Let 0 be a scalar function of x and t such that for all values of t any surface j = y, where 7 is any particular constant, represents the same sheet of particles of the substance. Then the function j will be called a carried function*. (2) The analytical condition which f must satisfy is or as it may be written, $-* ) ................................................ (xvii). (3) Now let f) and i/r be any two carried functions. Then by equation (xvii) j (V Vtfr) = - V (V^ I ll) Vi/r - by 355 (2) equation (iii). (4) Also if / , yfr, % be any three carried functions by the extended rule of the middle factor, where the product of three vectors is treated as an extensive magnitude formed by progressive multiplication. Hence * These functions for a perfect fluid have been investigated by Clebsch in Crelle, Bd. lvi. 1860, and by M. J. M. Hill, in the Transactions of the Cambridge Philosophical Society, Vol. xiv. 1883.
570 PURE VECTOR FORMULA. [CHAP. IV. This result is obtained by Hill, without the use of the Calculus of Extension, in the paper cited. The brevity of the necessary analysis by this method is to be noted. (5) Putting S for the determinant (V^V-^V^), equation (xix) can be written and it follows at once that *? 0 g 0*)8, and so on. Hence if none of the series 6, 6, 6, and so on, are infinite, then all the successive mobile differential coefficients of S with regard to the time are zero when 8 is zero. Hence if S is zero at each point at any one instant, it remains zero at all subsequent times. 373. Clebsch's Transformations. (1) The curl filaments, defined by 371, equation (xvi), move with the substance with unaltered strength. Let two systems of surfaces be drawn at any instant on which the curl lines lie. Then if these surfaces be defined at any instant by the carried functions j and yjr, the intersections of the two systems at any subsequent instant will define the curl lines. Therefore, remembering that V and Vyfr are vectors at each point respectively perpendicular to the surfaces f = constant, and ty constant, passing through that point, we may write r = | V j Vyfr = -cr |V ViJr, where st is some function of x and t dr (2) But from equation (xvi), = (r V) u Or. Cut tvt dr Now -j- = ta dt Also (r | V) u Br Tn {(V j V^V) z^ | V0V-\/r (V |w)} l )} By equating these results we obtain zzr | V^ V^|r = 0. But by hypothesis the vector |V / V-^r is not null. Therefore iir = Hence is a carried function of the substance.
373] clebsch's transformations. 571 Let cr be replaced by the carried function ^. Thus r = x |V V f, and I r = (3) Now the solenoidal condition (V | r) = 0, gives V. %V / V-i/r = 0, that is V^V^Vi/r = 0, since V. V Vi/r = W . V^/r - V / . Wi/r = 0. But V^y f Vyfr is the well-known Jacobian whose vanishing is the con- dition that % is a function of 0 and yjr, where t is regarded as a constant. Hence %=/( ^ 0- But since f , i/r, ^ are carried functions, -^ = ^ = 0; where ~~ means that c/ and i/r are regarded as constant. Hence % is a function of / and yfr only, where t is regarded as a variable. Thus % =/( / , ^). (4) It is now easy to prove that the most general form for these curl filaments, which satisfy equation (xvi), is .................................(xx). For let t*r be a carried function of the form ,/( / , yjr). Then ^ and are carried functions of the same form. Then we have proved that the most general form for r is \V f Vyfr. dtp But ^^ Hence VvrVyfr = ^ o p Thus the most general form can be converted into |V-5rV |r, which is the form stated in equation (xx). (5) Now V f Vy{r = V . j Vyfr = - V . yfrVcj). Hence from the preceding subsections of this article the most general form of the solution of equation (xvi) for the vector v, of which r is the curl, is given by VJ V(xxi), where $ and yjr are carried functions, and ur is any continuous function of x and t. (6) We can also solve for the function x which appears in equation (xiv) in terms of f , yfr and sr. It is to be noted that the % of equation (xiv) is not a carried function.
572 PURE VECTOR FORMULAE. [CHAP. IV. Now by equation (xx) since [cf. 272 (2)] Also % = lt K W) + VotJ = V^ + 0V^ + V where f * is written for ~ , and so on. ot Hence equation (xiv) can be written Therefore v {% + 0^ + t] = - Now only the differential coefficients of % appear in equation (xiv), so we may with perfect generality write X = - t ^t-*rt..............................(xxii). Equations (xxi) and (xxii) are the extension of Clebsch's transformations for the velocity of a perfect fluid. 374. Flow of a Vector. (1) The flow of a vector v along any unclosed curve will be defined to be the integral f(v \dx), where the lower limit is the starting point of the line, curved or straight, and the upper limit is the end point. (2) If the vector v be such that its curl filaments are of constant strength, then its flow between any two points P and Q along a defined line takes by equation (xxi) the form I f dyfr + -cjq -srp. [Q (3) The part I f dy{r is such that it is independent of the time if the J p same line of particles be always considered. But it does in general depend on the special line of particles chosen, and is not completely defined by the terminal particles. The part 'utq -crp is completely defined by the terminal particles, but varies with the time. (4) Suppose that the mobile differential coefficient of the flow of any vector v along any line of particles in the substance is always equal to the
374] FLOW OF A VECTOR. 573 flow of some vector p along the same line of particles, then p will be called the motive vector of the flow of v. 7 / /. The definition of p is therefore -ji\(y \dx) = \{p \doc). By attending to the derivation of equation (xiv) it is easy to see by the use of the same analysis as there employed that p = ~- + | m + Vm........................(xxiii). where sr is some single-valued scalar function of x and t, r is the curl of v, u is the velocity of the substance at the point x. This equation should be compared with the equations of Electromotive Force in Oerk Maxwell's Electricity and Magnetism, Vol. il, Article 598. Note. The present chapter is written to shew that formulae and methods which have been developed by Hamilton and Tait for Quaternions are equally applicable to the Calculus of Extension. The pure vector formulae have some affinity to those of the very interesting algebra developed by Prof. J. W. Gibbs, of Yale, U.S.A., and called by him Vector Analysis. Unfortunately the pamphlet called, ' Elements of Vector Analysis,' New Haven, 1881 4, in which he developed the algebra, is not published, and therefore is not generally accessible to students. The algebra is explained and used by Oliver Heaviside, loe. cit. p. 550 ; it will be noticed in its place among the Linear Algebras. Note on Grassmann. H. Grassmann's Ausdehnungslehre von 1844 was republished by him in 1878 (Otto Wigand, Leipzig). A note by the publisher in this edition states that the author died while the work, was passing through the press. A complete edition of Grassmann's Mathematical and Physical Works (he also wrote important papers on Comparative Philology) with admirable notes is now being published under the auspices of the Royal Saxon Academy of Sciences, edited by F. Engel (Leipzig, Teubner) Band I. Theil I. 1894, Band I. Theil n. 1896 ; the remaining parts are not yet published (December 1897). I have not been able to make any substantial use of this admirable edition: the present work has been many years in composition and already nearly two years in the press; and the parts most closely connected with Grassmann's own work were, for the most part, the first written. It must be distinctly understood that the present work does not pretend to exhaust the suggestions in Grassmann's two versions of the * Ausdehnungslehre': I only deal with those parts, which I have been able to develope and to bring under one dominant idea. Thus Grassmann's important contribution to the theory of Pfaff's Equation by the use of the Calculus of Extension, given in the Ausdehnungslehre of 1862, is not touched upon here. It is explained in Forsyth's work, Theory of inferential Equations, Part I. Chapter v. The following list of the mathematical papers of Grassmann is taken from the Royal Society Catalogue of Scientific Papers.
574 NOTE ON GRASSMANN. Theorie der Centralen, Crelle xxiv. 1842 ; and xxv. 1843. Ueber die Wissenschaft der extensiven Grosse oder die Ausdehnungslehre, Grunert, Archiv vi. 1845. Neue Theorie der Electrodynamik, Poggend. Annal. lxiv. 1845. Grundziige zu einer rein geometrischen Theorie der Curven, mit Anwendung einer rein geometrischen Analyse, Crelle xxxi. 1846. Geometrische Analyse gekniipft an die von Leibnitz erfundene geometrische Characte- ristik, Leipzig, Jablon. Preisschr. (No. 1) 1847. Ueber die Erzeugung der Curven dritter Ordnung durch gerade Linien, und iiber geometrische Definitionen dieser Curven, Crelle xxxvi. 1848. Der allgemeine Satz iiber die Erzeugung aller algebraischer Curven durch Bewegung gerader Linien, Crelle xlii. 1851. Die hohere Projectivitat und Perspectivitat in der Ebene; dargestellt durch geo- metrische Analyse, Crelle xlii. 1851. Die hohere Projectivitat in der Ebene, dargestellt durch Functionsverkniipfungen, Crelle xlii. 1851. Erzeugung der Curven vierter Ordnung durch Bewegung gerader Linien, Crelle xliv. 1852. Zur Theorie der Farbenmischung, Poggend. Annal. lxxxix. 1853; and Phil. Mag. xii. 1854. Allgemeiner Satz iiber die lineale Erzeugung aller algebraischer Oberflachen, Crelle xlix. 1855. Grundsatze der stereometrischen Multiplication, Crelle xlix. 1855. Ueber die verschiedenen Arten der linealen Erzeugung algebraischer Oberflachen, Crelle xlix. 1855. Die stereometrische Gleichung zweiten Grades, und die dadurch dargestellten Ober- flachen, Crelle xlix. 1855. Die stereometrischen Gleichungen dritten Grades, und die dadurch erzeugten Ober- flachen, Crelle xlix. 1855. Sur les difterents genres de multiplication, Crelle xlix. 1855. Die lineale Erzeugung von Curven dritter Ordnung, Crelle lii. 1856. Ueber eine neue Eigenschaft der Steiner'schen Gegenpunkte des Pascal'schen Sechs- ecks, Crelle lviii. 1861. Bildung rationaler Dreiecke. Angenaherte Construction von 7r, Archiv Math. Phys. xlix. 1869. Losung der Gleichung oft + y3 + z3 -f u3 = 0 in ganzen Zahlen, Archiv Math. Phys. xlix. 1869. Elementare Auflosung der allgemeinen Gleichung vierten Grades, Archiv Math. Phys. li. 1870. Zur Theorie der Curven dritter Ordnung, Gottingen Nachrichten, 1872. Ueber zusammengehorige Pole und ihre Darstellung durch Producte, Gottingen Nachrichteriy 1872. Die neuere Algebra und die Ausdehnungslehre, Math. Annal. vn. 1874. Zur Elektrodynamik, Crelle lxxxiii. 1877. Die Mechanik nach den Principien der Ausdehnungslehre, Math. Annal. xn. 1877. Der Ort der Hamilton'schen Quaternionen in der Ausdehnungslehre, Math. Annal. xii. 1877. Verwendung der Ausdehnungslehre fur die allgemeine Theorie der Polaren und den Zusammenhang algebraischer Gebilde (posthumous), Crelle lxxxiv. 1878. An obituary notice will be found in the Zeitschrift Math. Phys. Vol. xxiii. 1878, by Prof. F, Junghans of Stettin.
NOTE ON GRASSMANN. 575 The works on the Calculus of Extension by other authors deal chiefly with the application of the Calculus to Euclidean Space of three dimensions, to the Theory of Determinants, and to the Theory of Invariants and Covariants in ordinary Algebra. Thus they hardly cover the same ground as the parts of the present work, dealing with Grassmann's Calculus, except so far as all are immediately, or almost immediately, derived from Grassmann's own work. Some important and interesting works have been written, among them are: Abriss des geometrischen Kalkuls, by F. Kraft, Leipzig 1893 (Teubner). Die Ausdehnungslehre oder die Wissenschaft von den extensiven Grbssen in strenger Formel- Entwiciclung, by Robert Grassmann, Stettin 1891. System der Raumlehre, by V. Schlegel, Part I. 1872, Part II. 1875, Leipzig (Teubner). Calcolo Geometrico, by G. Peano, 1888, Turin (Fratelli Bocca). Introduction a la Geometrie Differentielle, suivant la Me'thode de H. Grassmann, by C. Burali-Forti, 1897, Paris (Gauthier-Villars et fils). The Directional Calculus, by E. W. Hyde, 1890, Boston (Ginn and Co.). I did not see the above-mentioned work by C. Burali-Forti till the whole of the present volume was in print. It deals with the theory of Vectors and of Curves and Surfaces in Euclidean Space, in a similar way to that in which they are here dealt with in Chapters I., in., and iv. 359 of Book vn. The operation of taking the Vector is explained and denned. The formulae of multiplication in so far as they involve supple- ments are however pure vector formulae: some interesting investigations are given which I should like to have included: the application to Gauss' method of curvilinear co-ordinates is also pointed out. Buchheim's and Homersham Cox's important papers have already been mentioned [cf. notes pp. 248, 370]. I find that Buchheim has already proved [cf. Proc. Lond. Math. Soc. Vol. xviii.] the properties of skew matrices of 155: also the extension of the idea of Supplements in Chapter in. Book iv. is to some extent the same as his idea of taking the polar [cf. Proc. of Lond. Math. Soc. Vol. xvi.]. I had not noticed this, when writing the above chapter. He does not use the idea of * normal intensity'; accordingly his point of view is rather different. He does not bring out the fundamental identity of his process of taking the polar with Grassmann's process of taking the supplement. Homersham Cox has also written a paper*, Application of Grassmann's Ausdehnungs- lehre to Properties of Circles, Quarterly Journal of Mathematics, October, 1890. There are two papers by E. Lasker, An Essay on the Geometrical Calculus, Proc. of the London Math. Soc. Vol. xxvin. 1896 and 1897. The paper applies the Calculus to Euclidean space of n dimensions and to point-groups in such a space. It contains results which I should like to have used, if I had seen it in time. Helmholtz uses Grassmann's Calculus, as far as concerns addition, in his Handbuch der physiologischen Optik, 20, pp. 327 to 330 (2nd Edition). * In this paper by a slip of the pen the words 'Outer' and 'Inner' as applied to multiplication are interchanged.
IndexINDEX. The references are to pages. Absolute in connection with Congruent Trans- formations, 456 sqq., 500 sqq.; Conic Section, defined, 497; Plane, 367, 496 sqq.; Point-pair, defined, 351; Points on, in Hyperbolic Geometry, 422; Polar Regions, 367, 384, 420; Quadric, defined, 355. Absorption, Law of, 36 sqq. Acceleration in Euclidean Space, 540 sqq.; in Non-Euclidean Space, 482 sqq. Addition in connection with Classification of Algebras, 29 sqq.; and Multiplication, 25 sqq.; and Positional Manifolds, 120 sqq.; in Algebra of Symbolic Logic, 35 sqq.; of Vectors, 507 sqq.; Principles of, in Universal Algebra, de- fined, 19 sqq.; Relations, 123. Algebra, Linear, mentioned, 32, 172; and Algebra of Symbolic Logic, 35; defined, 28; Universal, mentioned, 11, 35 ; defined, 18. Algebras, Classification of, 29 sqq.; Linear Associative, 30; Species of, defined, 27; Numerical Genus of, 29, 119. Angle of Contingence, in Euclidean Space, 539 and 552; in Non-Euclidean Space, 479; of Parallelism, in Hyperbolic Space, 438; of Torsion, in Euclidean Space, 540; in Non-Euclidean Space, 479. Angular Distance between Points in Anti -space, 417. Anharmonic Ratio in Positional Manifolds, defined, 132; of Systems of Forces, 290. W. Antipodal Elements, defined, 166; in Elliptic Space, 361; Form of Elliptic Geometry, defined, 355; Intercept, defined, 168; length of, 362. Anti-space, considered, 414 sqq.; defined, 354. Anti-spatial Elements, 414 sqq. Arbitrary Regions in Algebra of Symbolic Logic, 55 sqq. Arithmetic and Algebra, 11. Associated Quadric of Triple Group, see Group; System of Forces, see System of Forces. Associative Law and Algebra of Symbolic Logic, 37; and Combinatorial Multiplication, 174; and Matrices, 251; and Multiplication, 27; and Pure and Mixed Products, 185 ; and Steps, 25; defined, 21. Ausdehnungslehre, 13, 19, 32, 115, 131, 168, 171, 172, 180, 198, 210, 219, 229, 248, 262, 278, 317, 522, 573. Axis, Central, in Elliptic Geometry, 401; in Euclidean Geometry, 529; in Hyperbolic Geometry, 454; of a Congruent Transformation in Elliptic Space, 471; in Euclidean Space, 501; in Hyperbolic Space, 458 ; of a Dual Group, see Group. Ball, Sir R. S., 281, 370, 406, 462, 473, 475, 531, 532. Base Point, 518. Beltrami, 369, 451. Binomial Expressions in Symbolic Logic, 45 sqq. 37
578 INDEX. Biquaternions, 370, 398. Bolyai, JM 369, 426, 436, 451, 487; Wolfgang, 369. Boole, 4, 10, 35, 46, 63, 96, 111, 115, 116. Bradley, 6, 10. Brianchon's Theorem, 231. Buchheim, 248, 253, 254, 278, 370, 405, 575. Burali-Forti, 522, 575. Burnside, 370. Calculus, General Nature of, denned and dis- cussed, 4 sqq.; Differential, distinguished from Uni- versal Algebra, 18 sqq.; of Extension, Algebraic Species of, 28,31; and Descriptive Geometry, 132, 214 sqq.; and Theory of Duality, 146, 196, 212, 481; investigated, 169 sqq. Cantor, G., 16. Carried Functions, 569 sqq. Cayley, 119, 131, 135, 161, 248, 249, 351, 352, 353, 354, 369. Central Axis, see Axis. Central Plane of Sphere in Hyperbolic Geo- metry, 442. Centre of Dual Group, see Group. Characteristic Lines of Congruent Transforma- tion, 470. Characteristics of a Scheme, 9, 14; of a Manifold, 13. Chasles, 246. Circle, denned, and Perimeter of, in Elliptic Geometry, 375; Great, on Sphere in Hyperbolic Geo- metry, 448; in Hyperbolic Geometry, 484. Classification (operation in Symbolic Logic), 41. Clebsch, 369, 569, 570; and Lindemann, 278, 280, 294. Clifford, 13, 369, 370, 398, 406, 407, 409, 472. Combination, General Definition of, 8. Common Null Line of a Group, 286. Commutative Law and Algebra of Symbolic Logic, 37; and Congruent Transforma- tions, 464; and Multiplication, 27; defined, 21. Complete Manifold, defined, 15. Complex Intensity, 120. Complexes, Conjunctive and Disjunctive, de- fined and discussed, 107 sqq.; Linear, 278 sqq. Compound Extensive Magnitudes, defined, 177 sqq. Congruence of Terms, defined, 122. Congruences of Lines and Dual Groups, 288. Congruent Ranges, defined, 350; Klein's Theorem concerning, 353 ; Transformations in Euclidean Geometry, 500 sqq., 536 sqq.; in Non-Euclidean Geometry, 456 sqq.; and Work, 469, 477, 537 ; Associated Systems of Forces of, 466, 476, 536; Parameters of, 460, 471. Conies, Descriptive Geometry of, 229 sqq. Conjugate Co-ordinates, defined and discussed, 148sqq.; Lines, 277; Sets of Systems of Forces, 298, 306, 308. Conjunctive Complex, 107 sqq. Construction, Grassmann's, 219 sqq.; in a Positional Manifold, defined, 214; Linear, of Cubics, 233 sqq.; von Staudt's, 215 sqq. Content, Theory of, 370, 406, 462. Contingence, Angle of, see Angle. Co-ordinate Elements, defined, 125; Region, defined, 126. Co-ordinates, Curvilinear, in Euclidean Space, 543; in Non-Euclidean Space, 488, 494; Conjugate, defined, 148. Cox, Homersham, 346, 370, 399, 400, 401, 575. Cubics, Linear Construction of, 233 sqq. Curl, defined, 554; Flux, defined, 554; Lines and Filaments, defined, 567. Curvature of Curves and Surfaces in Euclidean Space, 544 sqq,; in Non-Euclidean Space, 479 sqq, Curve-Locus in Space of v Dimensions, defined, 130; Quadriquadric, 144, 151. Curvilinear Locus in Space of v Dimensions, defined, 130. Cylindroid in Euclidean Space, 532 ; in Non-Euclidean Space, 403, 455. De Morgan, 32, 100, 122, 131. Dependence in a Positional Manifold, defined, 123.
INDEX. 579 Derivation, General Definition of, 8. Determinants and Combinatorial Multiplica- tion, 180; associated with Matrices, 252. Determining Property of a Scheme, 8. Developable Surface in Non-Euclidean Geome- try, 481. Development in the Algebra of Symbolic Logic, 45 sqq. Diametral Plane of a Dual Group in Euclidean Space, 532. Dimensions of a Manifold, defined, 17. Direct Transformation of a Quadric, 338 sqq.; of the Absolute, 456 sqq. Director Force, Line, or Equation, of a Group, 286 sqq. Discourse, Universe of, 100. Discriminants of Equations, defined and dis- cussed, 51 sqq.; of Subsumptions, defined, 59. Disjunctive Complex, 107 sqq. Displacements of Eigid Bodies in Euclidean Space, 500 sqq., 536 ; in Non-Euclidean Space, 456 sqq.; Small, 464, 476; Surfaces of Equal, 462, 472; Vector, 472 sqq. Distance, General Theory of, 349 sqq.; in Elliptic Geometry, Shortest, 385, 387 sqq.; in Hyperbolic Geometry, 416 sqq.; Angular between Points, 417; between Planes, 428; Shortest, 429 sqq.; in Non-Euclidean Geometry between Sub-regions, 365 sqq.; Definition of, 352. Distributive Law in Algebra of Symbolic Logic, 37, 84, 174. Divergence, defined, 554. Division in Algebra of Symbolic Logic, 80 sqq.; of Space in Non-Euclidean Geometry, 355, 379. Dual Group, see Group. Duality, Theory of, 147, 196, 481. Dupin's Theorem in Euclidean Space, 546; in Non-Euclidean Space, 494. Elements and Terms, 21; Antipodal, 166, 361; Co-ordinate, 125; Intensively Imaginary, 166; Real, 166; Linear, defined, 177; Elements, Linear, in Euclidean Space, 508 sqq.; in Non-Euclidean Space, 399 sqq., 452 sqq.; in Positional Manifold, and Me- chanical Forces, 273; Null, in Symbolic Logic, 35, 37, 38; and Propositional Interpretation, 109, 111; in Universal Algebra, 24 sqq., 28; Planar, defined, 177; of a Manifold, defined, 13; Secondary Properties of, 14; Regional, defined, 177; Self-Normal, 204 sqq.; Spatial and Antispatial, in Hyperbolic Space, 414 sqq.; Supplementary, in Symbolic Logic, 36 sqq. Elimination and Syllogisms, 103 sqq.; from Existential Expressions, 89 sqq.; in Symbolic Logic, defined and discuss- ed, 47 sqq.; Formula for, 55. Elliptic Definition of Distance, 352; Space, Formulae for, 356 sqq.; Kinematics of, 470 sqq.; Parallel Lines in, 404, 407 sqq.; Subregions in, 397 sqq.; Vector Systems of Forces in, 06 sqq.; Transformations in, 472 sqq.; Spatial Manifold, defined, 355. Engel, F., 317, 369, 370, 573. Equal Displacement, Surface of, 462, 472. Equations, Identical of Matrices, 256, 261; in Extensive and Positional Manifolds, viz. Defining, defined, 162; Director, of Groups, 286; of Condition, defined, 172; of Subregions, 195; Plane and Point, 147; Reciprocal, defined, 147; in Algebra of Symbolic Logic, viz. and Universal Propositions, 105; Auxiliary, 71, 78; Limiting and Unlimiting, defined, 59; Negative and Positive Constituents of, defined, 50; Simultaneous, 51 sqq.; Discriminants of, defined, 52; Resultant of, defined, 52; with many Unknowns, 52; Discriminants of, 53; Resultants of, 53;
580 INDEX. Equations, with many Unknowns, Solution, 65 sqq.; Johnson's Method, 73 sqq.; Skew-Symmetrical, 71 sqq.; Symmetrical, 67 sqq., 73 sqq., 75 sqq.; with one Unknown, 49 sqq.; Discriminants of, 51; Resultant of, 51; Solution of, 55; Standard Form of, 49. Equivalence, defined, 5; Definition concerning, in Universal Al- gebra, 18; in Symbolic Logic, interpreted, 38; for Propositions, 108; Proof of, 36. Euler's Theorem, in Euclidean Geometry, 547; in Non-Euclidean Geometry, 492. Existential Expressions, 83 sqq.; and Propositional Interpretation, 111 sqq.; Resultant of, defined, 90; Solution of, 91. Expressions, Field of, defined, 60. Extension, Calculus of, investigated, 169 sqq.; mentioned, 28, 31, 132, 146; of Field, maximum and minimum, de- fined, 61. Extensive Magnitudes, defined, 176 sqq.; Manifolds and Non-Euclidean Geometry, 399 sqq., 452 sqq.; defined, 177; of three dimensions, 273 sqq. Extraordinaries, defined, 119. Field, Limited, defined, 61; of an Expression, defined, 60; of an Unknown, defined, 60. Flow of a Vector, defined, 572. Flux, defined for Point and Vector Formulae, 527; defined for Pure Vector Formulae, 549; Multiplication, 528 sqq.; Operation of Taking the, 522, 527. Flye, Ste Marie, 369. Force, defined, 177; Compared with Mechanical Force, 273 ; Single, Condition for, 277. Forces, Director, of a Group, 286; Groups of Systems of, see Groups; Intensity of, in Euclidean Space, 525; in Non-Euclidean Space, 399, 452; Investigated, 273 sqq.; Spatial, investigated, 452 sqq.; Forces, Systems of, see Systems. Forsyth, 17, 573. Franklin, Mrs, 98, 116. Frischauf, 369. Functions, Theory of, 11. Gauss, 488, 490, 543. Generating Regions of Quadrics, 147 sqq., 152 sqq.; in Non-Euclidean Geometry, 397, 451. Generators, Positive and Negative Systems of, 207. Geometry and Algebra, 11; and Extensive Manifolds, 273; Descriptive, and Calculus of Extension, 214; of many Dimensions, 131; Elliptic, Polar and Antipodal Forms, defined, 355; Hyperbolic, Investigated, 414 sqq.; Line-, 278; Non-Euclidean, and Cayley's Theory of Distance, 351; and Lindemann's Theory of Forces, 281; Historical Note upon, 369 sqq.; of a Sphere, Euclidean, 365; Parabolic, 496 sqq.; as a limiting Form, 367; Spherical, 355. Gerard, 369. Gibbs, 573. Graphic Statics, 520. Grassmann, H., 13, 19, 28, 31, 32, 115, 122, 131, 132, 146, 168, 171, 172, 201, 210, 219, 229, 233, 235, 246, 248, 249, 262, 278, 370, 522. Grassmann, H., Note upon, 573 sqq.; H. (The Younger), 317; R., 575. Gravelius, H., 281. Green's Theorem, 562. Groups of Systems, 284 sqq.; Common Null Lines of, 286 ; Director Forces of, 286; Dual and Quadruple, 287 sqq.; and Congruences, 288; Central Systems of, 402, 455; Centres of, 402; Diametral Plane of, 532 ; Elliptic and Hyperbolic, defined, 292; Parabolic, 289, 296 sqq.;
INDEX. 581 Groups, Dual, in Elliptic Space, 402 sqq.; in Euclidean Space, 531 sqq.; in Hyperbolic Space, 455; Principal and Secondary Axes of, 532; Self-Supplementary, 292, 296 sqq.; Invariants of, see Invariants; Quintuple, 286 sqq.; Eeciprocal, 285 sqq.; Semi-latent and Latent, denned, 322 ; Types of, 326 sqq.; Triple, 295 sqq.; Associated Quadric of, 295, 314, 535; Conjugate Sets of Systems in, 298, 306 sqq.; in Euclidean Space, 533 sqq. Halsted, 369. Hamilton, W. E., 32, 115, 131, 552, 573. Hamiltonian, defined, 554. Hamilton's Differential Operator, 554. Hankel, 32. Harmonic Invariant, see Invariant. Heaviside, Oliver, 550, 573. Helmholtz, 168, 369, 575. Hill, M. J. M.,569, 570. Homography of Eanges, defined, 133. Houel, 369. Hyde, E. W., 575. Hydrodynamics, 562 sqq. Hyperbolic Definition of Distance, 352; Dual Group, see Group; Geometry, Formulae for, 362 sqq.; investigated, 414 sqq.; Spatial Manifold, defined, 355. Ideal Space, 414. Identical Equation of a Matrix, 256, 261. Incident Eegions in Algebra of Symbolic Logic, 42; in Positional Manifolds, 125. Independence of Elements in Positional Mani- folds, defined, 122. Inference and a Calculus, 10. Infinity, Plane at, 497 ; Points at, 506. Inner Multiplication, 207 sqq., 528. Integrals, Volume, Surface, and Line, 562. Intensity, 119 sqq., 162 sqq.; and Secondary Properties, 15 ; Complex, 120; in Non-Euclidean Geometry, 364, 366; in Hyperbolic Geometry, of Points and Planes, 415 sqq.; Intensity in Parabolic Geometry, 368, 498; Locus of Zero, defined, 163; Negative, 120; Normal, defined, 200; of Forces in Non-Euclidean Geometry, 399, 452; Opposite, defined, 166. Intensively Imaginary, or Eeal, Elements, 166, 415. Intercept, 167 sqq., 358 sqq.; Antipodal, defined, 168; Length of, defined, 359, 363; Polar, defined, 358; The, defined, 358. Interpretation, Propositional, in Symbolic Logic, 107. Intersection of Manifolds, defined, 15. Invariant Equations of Condition, 172. Invariants of Groups of Systems, 300 sqq., 531 sqq.; Conjugate, of Triple Groups, 310 sqq.; Harmonic, of Dual Groups, 301, 533; Null, of Dual Groups, 300 ; Pole and Polar, of Triple Groups, 305, 534; of Groups in Euclidean Space, 531 sqq. Involution, Lines in, 280; of Systems of Forces, 291; Foci of, 291. Jevons, 38, 39, 115. Johnson, W. E., 28, 44, 48, 67, 73, 88, 116, 183. Junghans, 574. Killing, 370. Kinematics in Euclidean Geometry, 536 sqq.f 551 sqq.; in Non-Euclidean Geometry, 456 sqq. Klein, 127, 135, 278, 351, 353, 354, 369, 381, 456, 500, 501. Koenigs, 278. Kraft, F., 575. Lachlan, 138. Ladd, Miss Christine, 98, 116. Lasker, E., 575. Latent and Semi-Latent Eegions, Types of, in Three Dimensions, 317 sqq.; Groups and Systems, Types of, 326 sqq.; Group, defined, 322; Point, defined and discussed, 254 sqq.; Eegions, ,, ,, 248, 256 sqq.;
582 INDEX. Latent Regions, corresponding to roots con- jointly, 316; Root, 254; Repeated, 257 ; System, defined and discussed, 322 sqq. Law, Associative, and Combinatorial Multipli- cation, 174; and Matrices, 251; and Symbolic Logic, 37; and Universal Algebra, 25, 27; Commutative, and Congruent Transforma- tions, 464 ; and Symbolic Logic, 37 ; and Universal Algebra, 27 ; Distributive, and Combinatorial Multipli- cation, 174; and Existential Expressions, 84 ; and Symbolic Logic, 37; and Universal Algebra, 26; of Absorption, 37; of Simplicity, 39; Partial Suspension of, 88 ; of Unity, 38; Partial Suspension of, 88. Laws of Thought, 110. Letters, Greek, Roman, and Capital, Con- ventions concerning, 86, 119, 177, 550; Regional, 87; Umbral, 86. Leibnitz's Theorem, 273. Lie, 369. Limit Line, 495; Surface, 447 sqq., 486, 494. Limited Field, defined, 61. Limiting Equation, defined, 59. Lindemann, 281, 369. Line-Geometry, 278. Lines, Null, 278, 286; Parallel, in Elliptic Space, 404 sqq.; in Hyperbolic Space, 436 sqq.; Secant and Non-Secant, 436 ; Spatial and Anti- Spatial, 418; Straight, defined, 130. Linear Complexes, 278; and Quintuple Groups, 286; Theorems concerning, 292 ; Element, defined, 177; compared to Mechanical Force, 273. Lobatschewsky, 369, 436, 438, 487. Locus, Containing, defined, 131; Curvilinear, defined, 130; defined, 128; Flat, defined, 129; of Zero Intensity, defined, 163 ; Locus, Surface and Curve, defined, 130. Logic, Application of Algebra to, 99; Generalization of Formal, 105; Symbolic, Algebra of, mentioned, 22, 29; Formal Laws of Algebra of, 35 sqq.; Interpretations of Algebra of, 38, 99, 107. Lotze, 6, 116. Love, 560. M'Coll, 112, 116. Magnitude, Extensive, defined, 176 sqq. Manifolds, 13 sqq.; Algebraic, 22, 26 sqq.; Orders of, 27, 171, 175; Self-Multiplicative, 27; Complete Algebraic System of, 27 ; Derived, 175; Extensive, and Elliptic Geometry, 399 sqq.; and Hyperbolic Geometry, 452 sqq.; mentioned, 31; of Three Dimensions, 273 sqq.; Positional, investigated, 117 sqq.; mentioned, 30; Spatial, defined, 349, 355; Special, defined, 16 sqq. Matrices, 248 sqq., 316 sqq.y 456 sqq., 500 sqq.; and Forces, 316 sqq.; Congruent, 457 sqq., 500, 536 ; Denominators and Numerators of, 249 ; Null Spaces of, 252; Nullity of, 253; Skew, 248, 267; symbolized, 280; Spaces (or Regions) preserved by, 253 ; Sums and Products of, 250 ; Symmetrical, 248, 262; Vacuity of, 261; Vacuous Regions of, 262. Maxwell, 573. Metageometry, 369. Metrics, Theory of, 273, 347 sqq. Meunier's Theorem, in Euclidean Space, 547; in Non-Euclidean Space, 493. Middle Factor, Extended Rule of the, 188; Rule of the, 185; for Inner Multiplication, 208. Mitchell, Dr, 116. Mixed Product, defined, 184. Mobile Differential Operator, defined, 554. Mobius, 131. Mode of a Property, 8, 13. Moment of a System of Forces, 278.
INDEX. 583 Monge, 479, 541. Motion, Associated System with, 487. Miiller, 192. Multiplication, Combinatorial, investigated, 171 sqq.; Formulae for, in Three Di- mensions, 274; Flux, 528; in Symbolic Logic, interpreted, 38,108; Inner and Outer, 207 ; in Euclidean Space, 523 sqq.; mentioned, 18; Principles of, denned and discussed, 25 sqq.; Progressive and Regressive, 181 sqq. Multiplicative Combination, defined, 175. Napier's Analogies and Non-Euclidean Geo- metry, 375, 425. Negation, Primitive, 113. Negative Intensity, 120; System of Generators, 207. Newcomb, 370. Non-Euclidean Geometry, Historical Note upon, 369 sqq.; investigated, 347 sqq. Non-secant Lines, 436. Normal Intensity, defined, 200; Points, defined, 199 sqq.; Regions, defined, 203; and Points in Euclidean Geometry, 523 sqq.; in Non-Euclidean Geometry, 383 sqq.y 426 sqq.; Systems of Points, defined, 200; Rectangular, 524. Nugatory Forms of Propositions, 100. Null Element and Self-Condemned Propositions, 111; in Symbolic Logic, 35, 37 sqq.; interpreted, 38, 109; in Universal Algebra, 24 sqq., 28; Invariants of a Dual Group, 300; Lines, Planes, and Points, 278; Points, Latency of, 323 sqq.; Space of Matrix, 253; Term, 24. Nullity of Matrix, defined, 253. Numbers, Alternate, 180. One-sided Planes, 379. Operation, General Definition of, 7 sqq.; of Taking the Flux, 522, 527 ; Vector, 516, 522. Order and the Operation of Addition, 19 sqq.; of Manifolds, 27, 171, 175 ; of Tortuosity, 131. Origin, defined, for Euclidean Geometry, 524; for Hyperbolic Geometry, 414; with Pure Vector Formulae, 550. Outer Multiplication, 207. Oval Quadrics, defined, 376; Spheres, intersection of, 396. Over-Strong Premises, 104. Parabolic Definition of Distance, 353 sqq.; Geometry, 355, 496 sqq.; as a Limiting Form, 367; Group, 289, 296; Semi-Latent, and Latent, 322 sqq.; Linear Transformation, 135; Self-Supplementary Group, General Type of, 296; Subgroup of Triple Group, 296, 311 sqq. Parallel Lines in Elliptic Geometry, 404 sqq.; in Hyperbolic Geometry, 436 sqq.; Planes in Hyperbolic Geometry, 439; Regions in Elliptic Geometry, 397 sqq. Parallelism, Angle of, 438 ; Right and Left, 405. Parallelogram of Forces, 273 ; in Elliptic Space, 410. Parameters of a Congruent Transformation, 460, 471; of a System of Forces, 401, 454. Partition of a Manifold, 15. Pascal's Theorem, 231, 237. Peano, G., 575. Peirce, B., 172; C. S., 3, 10, 37, 42, 115. Perpendiculars in Non-Euclidean Geometry, 383 sqq., 426 sqq. Perspective, 139 sqq. Pfaff's Equation, 573. Planar Elements, defined, 177; Intensities of, in Non-Euclidean Geo- metry, 366, 415. Planes, Angles between, in Non-Euclidean Geometry, 365, 382, 417 sqq.; Central, of Spheres, 442 ; Defined, 130; Diametral, of Groups, 532 ; Parallel, in Hyperbolic Space, 439; One-sided, 379 sqq. Pliicker, 213, 278. Polar Form of Elliptic Geometry, defined, 355; Intercept, defined, 358;
584 INDEX. Polar Invariant, see Invariant; Reciprocally, defined, 145; Regions, Absolute, 367, 384, 420; Self-, defined, 145. Pole Invariant, see Invariant. Pole and Polar, 145; in Hyperbolic Geometry, 420. PoincarS, 369. Point and Vector Factors, 514 ; Inside a Triangle, defined, 375 ; Latent, 254, 257, 317. Points, Normal, 199 sqq.; Normal or quadrantal, in Non-Euclidean Geometry, 383, 414, 426. Positional Manifolds, defined, 30; investigated, 117 sqq. Positive System of Generators, 207. Premises, Over-Strong, 104. Primitive Predication and Negation, 112 sqq. Principal Systems of a Dual Group, defined, 532; Triangles, 373 sqq., 380 sqq.; Vector of a System of Forces, 518. Products, Order of, defined, 175; Pure and mixed, 184. Progressive Multiplication, 181. Projection, 224 sqq. Propositional Interpretation of Symbolic Logic, 107 sqq. Propositions, Equivalence of, defined, 108; Nugatory Forms of, 100; Reciprocal, 38; Self-condemned, defined, 111; Simple, 107; Symbolic Forms of, 99 sqq., Ill sqq. Pure Products, defined, 184. Quadrantal Points, 383. Quadrics, 144 sqq.; Absolute, 355; and Inner Multiplication, 210; Associated with Triple Groups, see Group; Conical, defined, 150; investigated, 155; Closed, 153, 355, 376; Line-Equation of, generalized, 213 ; Oval, defined, 376; Self-normal, defined, 199; Extended Definition of, 201; Transformation of, Direct and Skew, 338 sqq. Quaternions, mentioned, 32, 115, 131, 554, 573. Quotient, name for Matrix, 249. Ratio, Anharmonic, defined, 132; of Systems of Forces, 291. Reciprocal Groups, 285; Propositions, 38; Systems of Forces, 281, 303; and Work, 469, 477, 538. Reciprocity between Addition and Multiplication in Symbolic Logic, 37. Rectangular Normal Systems in Euclidean Space, 524 ; Rectilinear Figures in Hyperbolic Space, 433. Reference Figures, 138. Regional Element, in Calculus of Extension, 177; Letters, in Symbolic Logic, 87. Regions and Symbolic Logic, 38; Incident, 42; Complete, defined, 123; Containing, 126; Co-ordinate, 126; Generating, 147; and Non-Euclidean Geometry, 397, 451; Latent and Semi-Latent, 248, 256 sqq.; corresponding to Roots conjointly, 316; Types of, in Three Dimensions, 317 sqq.; Mutually Normal, 203; Non-vertical, 159; Null, of Matrices, 252 ; Parallel, 397, 451; Preserved by Matrices, 253 ; Semi-Latent, see Regions, Latent; Supplementary, 126; S37mbolism for, in Calculus of Exten- sion, 177; Vacuous, of Matrices, 262. Regressive Multiplication, 181 sqq. Relation, General Definition of, 8. Resultants, defined, 51 sqq.; Existential, 90 sqq.; of Subsumptions, 59. Riemann, 13, 17, 368, 369. Rigid Body, Motion of, in Non-Euclidean Geometry, 487. Rotation in Euclidean Geometry, 502; in Non-Euclidean Geometry, 460, 471. Russell, B. A. W., 369, 370. Salmon, 151. Scalar, 552.
INDEX. 585 Schemes, Algebraic, 22, 26, 35 ; Substitutive, 8 sqq., 14. Schlegel, V., 575. Schroder, 37, 42, 51, 62, 66, 67, 74,82,115, 116. Scott, K. F., 180. Secant Lines, 436. Secondary Properties of Elements, 14,120 sqq.; Triangles, 373. Self-condemned Propositions, defined, 111. Self-Conjugate Tetrahedrons and Conjugate Sets of Systems, 308. Self-normal Elements, 199, 204; Quadric, 199, 201, 204 ; Sphere, 524. Semi-Latent, see Latent. Shadows, 87 sqq.; Weak Forms of, 84. Sign, Standard, 363, 416. Signs and Symbolism, 3 sqq. Simple Extensive Magnitude, defined, 177. Simplicity, Law of, 39; Partial Suspension of, 88. Skew Transformation of a Quadric, 338, 342 Space and Symbolic Logic, 30 sqq., 38 sqq.; Constant, 363; defined, 354; Descriptive Properties of, 119 sqq.; Division of, 355, 379; Elliptic, Hyperbolic, and Parabolic, 355 sqq.; Euclidean, 505 sqq.; Flatness of, 451; Ideal, 414; Non-Euclidean, Formulae for, 356 sqq., 362 sqq.; Null, of a Matrix, 252; Preserved by a Matrix, 253 ; Vacuous, of a Matrix, 262. Spaces as Spherical Loci, 450 sqq. Spatial Elements, 414 sqq.; Interpretation of Universal Algebra, 31; Manifolds, defined, 354 sqq. Species of Algebras, 27. Sphere, defined, 376; investigated, 391 sqq., 441 sqq.; Self-normal, 524. Spherical Geometry, 355. Stackel, 369, 370. Standard Form of Planes, 415 ; Sign, see Sign. Stationary Differential Operator, 553. Staudt, von, 215, 317. Staudtian, Formulae for, 374. W, Steps, 25, 507. Stereometrical Triangles, 382, 425. Stokes' Theorem, 562. Stout, 3, 4. Sturm, Dr Rudolf, 278. Subgroup, Parabolic, of Triple Group, 296, 311. Submanifold, defined, 15. Subplane, defined, 130. Subregions, defined and investigated, 123, 125 ; Distances of, 365; Parallel, 397, 451. Substitutive Schemes, 8; Signs, 3 sqq. Subsumptions, defined and investigated, 42 sqq., 59. Subtraction and Symbolic Logic, 80 ; and Universal Algebra, 22 sqq., 28. Supplementary Terms, interpreted, 39, 110. Supplements and Reciprocal Polars, 201; defined, 181 sqq.; Different kinds of, 202 ; Extended Definition of, 201; in Euclidean Space, 523 sqq.; in Symbolic Logic, 36 sqq.; investigated, 199 sqq. Surface Locus, defined, 130. Surfaces in Euclidean Space, 539 sqq.; in Non-Euclidean Space, 488 sqq.; of Equal Distance from Planes, 391, 442; from Subregions, 397, 451. Syllogism, 101; Generalization of, 105; Symbolic Equivalents of, 103. Sylvester, 154, 253, 254. Symbolic Logic, Formal Laws of Algebra of, 35 sqq.; interpreted, 38, 99 sqq., 107 sqq.; mentioned, 22, 29. Synthesis, General Definition of, 8. Systems of Forces, 273 sqq.; and Quadrics, 294 sqq., 298, 338, 535; associated with Congruent Transform- ation, 466, 473, 476, 536; Axes of, 401, 454 ; Groups of, see Groups ; in Euclidean Space, 518 sqq.; in Non-Euclidean Space, 401 sqq., 454 sqq.; in Notation of Pure Vector Formulae, 551; Latent, defined, 322; Non-Axal, 404, 455; Parameters of, 401, 454 ; Reciprocal, 281; 38
586 INDEX. Systems, Reciprocal, and Work, 469, 477, 538; Vector, 406 sqq., 454 ; and Vector Transformations, 473 ; Unit, 407. Tait, 554, 573. Taylor, H. M., 215. Terms, defined, 20; and Points, distinguished, 360; Constituent, defined, 47 ; Congruence of, defined, 122 ; defining Triangle, 372 ; Equivalence of, interpreted, 38 ; Null, 24; Supplementary, interpreted, 39. Thought, Laws of, 110. Tortuosity, Order of, 131. Transformation, Congruent, 456, 500, 537 ; and Characteristic Lines, 470; and Work, 469, 477, 537 ; Associated Systems of Forces of, 466, 473, 476, 536; Axes of, 458, 471; Parameters of, 460, 471; Elliptic, Hyperbolic, Parabolic, 134; Linear, 133, 227, 248, 316; of a Quadric, Direct and Skew, 338, 456 sqq.; Vector, 472 sqq. Translation in Euclidean Space, 502 ; in Non-Euclidean Space, 460, 471. Triangles, 371 sqq., 378, 422 sqq.; defined by Terms, 372; Principal, 373, 380; Set of, 373; Secondary, 373; Spatial and Semi-Spatial, 423 ; Stereometrical, 425, 482. Trigonometry, Spherical, 375, 383, 426. Umbral Letters, 86 sqq. Uninterpretable Expressions in a Calculus, 10. Units, Independent, defined, 122. Unity, Law of, 38 ; Partial Suspension of, 88. Universal Algebra, 11, 18 ; Propositions and Equations, 105. Universe and Primitive Predication, 113 ; defined, 36 ; interpreted, 39, 109; Properties of, 37; of Discourse, 100. Unlimiting Equations, 59 sqq. Vacuity of a Matrix, 261. Vacuous Region of a Matrix, 262. Vector in Elliptic Space, viz. Vector System of Forces, 406 sqq., 454, 473 ; Associated with Motion, 473; Right and Left, 406 sqq.; Unit, defined, 407; Transformations, 472. Vector in Euclidean Space, viz. Vector Analysis, 573 ; Areas, 509 sqq.; Differentiation, 560; Factor, 514; Flow of a, 572 ; Formulae, Pure, 548 sqq.; Moment, 518, 551; Operation of Taking the, 516 sqq., 522 ; Potential, 565; Principal, of System of Forces, 518 ; Volumes, 513 sqq. Velocity in Non-Euclidean Space, 482. Venn, 115, 116. Veronese, 139, 147, 152, 161, 370. Vortex Motion, 562. Weak forms of Shadow Letters, 84. Weakened Conclusion, 104. Webb, 560. Work in Euclidean Space, 537 ; in Non-Euclidean Space, 468, 477. CAMBRIDGE : PRINTED BY J. AND C. F. CLAY, AT THE UNIVERSITY PRESS.