INTRODUCTION TO THE MECHANICS OF A CONTINUOUS MEDIUM Lawrence E. Malvern Professor of Mechanics College of Engineering Michigan State University Prentice-Hall, Inc. Englewood Cliffs, New JerseyEnglewood Chis, N J All rights reserved No part of this book may be reproduced in any form or by any means without permission in writing from the publisher. Current printing (last digit): 0987 13-487603-2 Library of Congress Catalog Card Number 69-13712 Printed in the United States of AmericaPreface This book offers a unified presentation of the concepts and general princi- ples common to all branches of solid and fluid mechanics, designed to appeal to the intuition and understanding of advanced undergraduate or first-year postgraduate students in engineering or engineering science. The book arose from the need to provide a general preparation in contin- uum mechanics for students who will pursue further work in specialized fields such as viscous fluids, elasticity, viscoelasticity, and plasticity. Originally the book was introduced for reasons of pedagogical economy—to present the com- mon foundations of these specialized subjects in a unified manner and also to Provide some introduction to each subject for students who will not take courses in all of these areas. This approach develops the foundations more carefully than the traditional separate courses where there is a tendency to hurry on to the applications, and moreover provides a background for later advanced study in modern nonlinear continuum mechanics. The first five chapters devoted to general concepts and principles applicable to all continuous media are followed by a chapter on constitutive equations, the equations defining particular media. The chapter on constitutive theory begins with sections on the specific constitutive equations of linear viscosity, linearized elasticity, linear viscoelasticity, and plasticity, and concludes with two sections on modern constitutive theory. There are also a chapter on fluid mechanics and one on linearized elasticity to serve as examples of how the general principles of the first five chapters are combined with a constitutive equation to formu- late a complete theory. Two appendices on curvilinear tensor components follow, which may be omitted altogether or postponed until after the main exposition is completed. Although the book grew out of lecture notes for a one-quarter course for first-year graduate students taught by the author and several colleagues during the past 12 years, it contains enough material for a two-semester course and is written at a level suitable for advanced undergraduate students, The only vvi Preface prerequisites are the basic mathematics and mechanics equivalent to that usu- ally taught in the first two or three years of an undergraduate engineering program. Chapter 2 reviews vectors and matrices and introduces what tensor methods are needed. Part of this material may be postponed until needed, but it is collected in Chap. 2 for reference. The last 15 to 20 years have seen a great expansion of research and publi- cation in modern continuum mechanics. The most notable developments have been jn the theory of constitutive equations, especially in the formulation of very general principles restricting the possible forms that constitutive equations can take. These new theoretical developments are especially addressed to the formulation of nonlinear constitutive equations, which are only briefly touched upon in this book. But the new developments have also pointed up the limita- tions of some of the widely used linear theories. This does not mean that any of the older linear theories must be discarded, but the new developments pro- vide some guidance to the conditions under which the older theories can be used and the conditions where they are subject to significant error. The last two sections of Chap. 6 survey modern constitutive theory and provide refer- ences to original papers and to more extended treatments of the modern theory than that gi in this introductory text. The booK is a carefully graduated approach to the subject in both content and style. The earlier part of the book is written with a great deal of illustrative detail in the development of the basic concepts of stress and deformation and the mathematical formulation used to represent the concepts. Symbolic forms of the equations, ‘using dyadic notation, are supplemented by expanded Carte- sian component forms, matrix forms, and indicial forms of the same equations to give the student abundant opportunity to master the notations. There are also many simple exercises involving interpretation of the general ideas in con- crete examples. In Chaps. 4 and 5 there is a gradual transition to more reliance on compact notations and a gradual increase in the demands on the reader's ability to comprehend general statements. Until the end of Sec. 4.2, each topic considered is treated fairly completely and (except for the brief section on stress resultants in plate theory) only concepts that will be used repeatedly in the following sections are introduced. Then there begin to appear concepts and formulations whose full implementa- tion is beyond the scope of the book. These include, for example, the relative description of motion, mentioned in Sec. 4.3 and also in some later sections, and the finite rotation and stretch tensors of Sec. 4.6, which are important in some of the modern developments referred to in the last two sections of Chap. 6. The aim in presenting this material is to heighten the reader's awareness that the subject of continuum mechanics is in a state of rapid development, and to encourage his reading of the current literature. The chapters on fluids and on elasticity also refer to published methods and results in addition to those actually presented.Preface ‘vii The sections on the constitutive equations of viscoelasticity and plasticity are introduced by accounts of the observed responses of real materials in order to motivate and also to point up the limitations of the idealized representations that follow. The second section on plasticity includes work-hardening. a part of the theory not in a satisfactory state, but so important in engineering appli- cations that it was believed essential to mention and point out some of the shortcomings of the available formulations. A one-quarter course might well include most of the first five chapters, only part of Chap. 6, and either Chap. 7 on fluids or Chap. 8 on elasticity. Section 3.6 on stress resultants in plates and those parts of Secs. 5.3 and 5.4 treating couple stress can be omitted without destroying the continuity, as also can Secs. 6.5 and 6.6 on plasticity. Section 4.6 can be given only minor emphasis, or omitted altogether if the last two sections of Chap. 6 are not to be covered, The second appendix, presenting only physical components in orthogonal cur- vilinear coordinates might be included if time permits; although not needed in the text, it is useful for applications. A two-term course could include the first appendix on general curvilinear tensor components, useful as a preparation for reading some of the modern lit- erature. There is sufficient textual material in the book for a full year course, but it should probably be supplemented with some challenging applications problems. Most of the exercises in the text are teaching devices to illuminate the theory, rather than applications. The book is a textbook, designed for classroom teaching or self-study, not @ treatise reporting new scientific results. Obviously the author is indebted to hundreds of investigators over a period of more than two centuries as well as to earlier books in the field or in its specialized branches. Some of these inves- tigators and authors are named in the text, but the bibliography at the end of the book includes only the twentieth-century writings cited. Extensive bibli- ographies may be found in the two Encyclopedia of Physics treatises: “The Classical Field Theories,” by C. Truesdell and R. A. Toupin, Vol. I1],’l, pp. 226-793 (1960), and “The Non-Linear Field Theories of Mechanics.” by C. Truesdell and W. Noll, Vol. 111/3 (1965), published by Springer-Verlag, Berlin. These two valuable comprehensive treatises are among the references for collateral reading cited at the end of the introduction. Many of the historical allusions in the text are based on these two sources. The author is indebted to several colleagues at Michigan State University who have used preliminary versions of the book in their classes. These include Dr. C. A. Tatro (now at the Lawrence Radiation Laboratory. Livermore, California)and Professors M. A. Medick, R. W. Little. and K. N. Subramanian. Professors John Foss and Merle Potter read the first version of the material ‘on fivid mechanics. Encouragement and helpful criticism have been provided by these colleagues and also by the dozens of students who have taken the course.vill Preface The author is also indebted to Michigan State University for sabbatical leave during 1966-67 to work on the book and to Prentice-Hall, Inc., for their cooperation and assistance in preparing the final text and illustrations. Finally, thanks are due to the author's wife for inspiration, encouragement and forbearance. Lawrence E, MALVERN East Lansing, Michigan3. 4. Contents Introduction 1 1.1 The Continuous Medium = 1 Vectors and Tensors 7 21 Introduction 7 2.2 Vectors; Vector Addition; Vector and Scalar Components; Indicial Notation; Finite Rotations not Vectors 10 2.3 Scalar Product and Vector Product 17 2.4 Change of Orthonormal Basis (Rotation of Axes); Tensors as Linear Vector Functions; Rectangular Cartesian Tensor Components; Dyadics; Tensor Properties; Review of Ele- mentary Matrix Concepts 25 2.5 Vector and Tensor Calculus; Differentiation; Gradient, Di- vergence and Cus 48 Stress 64 3.1 Body Forces and Surface Forces 64 3.2 Traction or Stress Vector; Stress Components 69 3.3 Principal Axes of Stress and Principal Stresses; Invariants; Spherical and Deviatoric Stress Tensors 85 3.4 Mohr's Circles 95 3.5 Plane Stre fohr’s Circle 102 3.6 Stress Resultants in the Simplified Theory of Bending of Thin Plates 112 Strain and Deformation 120 4.1 Small Strain and Rotation in Two Dimensions 120 4.2 Small Strain and Rotation in Three Dimensions 129 ixContents 43 Kinematics of a Continuous Medium; Material Derivatives 138 44 Rate-of-Deformation Tensor (Stretching); Spin Tensor (Vor- ticity); Natural Strain Increment 145 4.5 Finite Strain and Deformation; Eulerian and Lagrangian Formulations; Geometric Measures of Strain; Relative De- formation Gradient 154 46 Rotation and Stretch Tensors 172 4.7 Compatibility Conditions; Determination of Displacements When Strains are Known = 183 General Principles 5.1 Introduction; Integral Transformations; Flux 197 5.2 Conservation of Mass; The Continuity Equation 205 5.3 Momentum Principles; Equations of Motion and Equilib- tium; Couple Stresses 213 5.4 Energy Balance; First Law of Thermodynamics; Energy Equation 226 5.5 Principle of Virtual Displacements 237 5.6 Entropy and the Second Law of Thermodynamics; the Clausius-Dubem Inequality 248 5.7 The Caloric Equation of State; Gibbs Relation; Thermody- namic Tensions; Thermodynamic Potentials; Dissipation Function 260 Constitutive Equations 6.1 Introduction; Ideal Materials 273 62 Classical Elasticity; Generalized Hooke's Law; Isotropy; Hyperelasticity; The Strain Energy Function or Elastic Po- tential Function; Elastic Symmetry; Thermal Stresses 278 6.3 Fluids; Ideal Frictionless Fluid; Linearly Viscous (New- tonian) Fluid; Stokes Condition of Vanishing Bulk Vis- cosity; Laminar and Turbulent Flow 295 64 Linear Viscoelastic Response 306 6.5 Plasticity I. Plastic Behavior of Metals; Examples of Theo- ties Neglecting Work-Hardening: Levy-Mises Perfectly Plastic; Prandtl-Reuss Elastic, Perfectly Plastic; and Visco- plastic Materials 327 66 Plasticity Il. More Advanced Theories; Yield Conditions; Plastic-Potential Theory; Hardening Assumptions; Older Total-Strain Theory (Deformation Theory) 346 6.1 Theories of Constitutive Equations 1: Principle of Equi- presence; Fundamental Postulates of a Purely Mechanical Theory; Principle of Material Frame-Indifference 378 197 273Contents xi 68 Theories of Constitutive Equations Il: Material Symmetry Restrictions on Constitutive Equations of Simple Materials; Isotropy 406 7. Fluid Mechanics 423 7.1 Field Equations of Newtonian Fluid: Navier-Stokes Equa- tions; Example: Parallel Plane Flow of Incompressible Fluid Between Flat Plates 423 7.2 Perfect Fluid: Euler Equation; Kelvin's Theorem; Bernoulli Equation; Irrotational Flow; Velocity Potential; Acoustic Waves; Gas Dynamics 434 7.3 Potential Flow of Incompressible Perfect Fluid 448 7.4 Similarity of Flow Fields in Experimental Model Analysis; Characteristic Numbers; Dimensional Analysis 462 7.5 Limiting Cases: Creeping-Flow Equation and Boundary- Layer Equations for Plane Flow of Incompressible Viscous Fluid 475 8. Linearized Theory of Elasticity 497 8.1 Field Equations 497 8.2 Plane Elasticity in Rectangular Coordinates 505 83 Cylindrical Coordinate Components; Plane Elasticity in Polar Coordinates 525 84 Three-Dimensional Elasticity; Solution for Displacements; Vector and Scalar Potentials; Wave Equations; Galerkin Vector; Papkovich-Neuber Potentials; Examples, Including Boussinesq Problem $48 Appendix 1. Tensors 569 1.1 Introduction; Vector-Space Axioms; Linear Independence; Basis; Contravariant Components of a Vector; Euclidean Vector Space; Dual Base Vectors; Covariant Components ofa Vector 569 1.2 Change of Basis; Unit-Tensor Components 576 1.3 Dyads and Dyadics; Dyadics as Second-Order Tensors; Determinant Expansions; Vector (cross) Products 588 1.4 Curvilinear Coordinates; Contravariant and Covariant ‘Components Relative to the Natural Basis; The Metric Tensor 596 1.5 Physical Components of Vectors and Tensors 606 1.6 Tensor Calculus; Covariant Derivative and Absolute De-xii Contents ; Christoffel Symbols; Gradient, tivative of a Tensor Fie! Divergence, and Curl; Laplacian 614 1.7 Deformation; Two-Point Tensors; Base Vectors; Metric Tensors; Shifters; Total Covariant Derivative 629 1.8 Summary of General-Tensor Curvilinear-Component Forms of Selected Field Equations of Continuum Me- chanics 634 Appendix Il. Orthogonal Curvilinear Coordinates, Physical Components of Tensors IL 1 Coordinate Definitions; Scale Factors; Physical Compo- nents; Derivatives of Unit Base Vectors and of Dyadics 64l IL. 2 Gradient, Divergence, and Curl in Orthogonal Curvilinear Coordinates 650 IL3 Examples of Field Equations of Continuum Mechanics, ‘Using Physical Components in Orthogonal Curvilinear Coordinates 659 U4 Summary of Differential Formulas in Cylindrical and Spherical Coordinates 667 Bibliography. Twentieth-Century Authors Cited in the Text Author Index Subject Index 641 673 691INTRODUCTION TO THE MECHANICS OF A CONTINUOUS MEDIUMCHAPTER 1 Introduction 1.1 The continuous medium The mechanics of a continuous medium is that branch of mechanics concerned with the stresses in solids, liquids, and gases and the deformation or flow of these materials. The adjective continuous refers to the simplifying concept underlying the analysis: we disregard the molecular structure of matter and picture it as being without gaps or empty spaces. We further suppose that all the mathematical functions entering the theory are continuous functions, except possibly at a finite number of interior surfaces separating regions of continuity. This statement implies that the derivatives of the functions are continuous too, if they enter the theory, since all functions entering the theory are assumed continuous. This hypothetical continuous material we call a con- tinuous medium or continuum, The concept of a continuous medium permits us to define stress at a point, geometric point in space conceived as occupying no volume, by a mathema- tical limit like the definition of the derivative in differential calculus, This ap- proach immediately makes the powerful methods of calculus available for the study of nonuniform distributions of stress, and at the same time provides an easily visualized physical model which agrees with the testimony of everyday observation of matter in the large. Thus this approach allows the mathematical analysis to be guided by intuition. The theories of elasticity, plasticity, and fluid mechanics based on the concept of continuous material lead moreover to quantitative predictions which agree closely with experience over a wide range of conditions. These theories and the further simplified engineering theories of beams and plates and shells, which are also based on the conunuum concept of matter, are adequate for analysis of stress and deformation in most engi- neering problems. There are exceptions of course. lt would be too much to expect that a theory which disregards the molecular nature of matter could account for all 12 Introduction Chap. 1 the observed properties of matter, even in the large. There is, for example, nothing in the theories that accounts for the formation of a fatigue crack in a body after many cycles of reversed loading. Problems in super aerodynamics at altitudes where the air is extremely rarefied may require molecular theories. Even in marginal situations like these, where continuum mechanics alone is inadequate, it is sometimes possible to use the continuum theory in combination with empirical information or with information derived from a physical theory based on the molecular nature of the material. The use of the continuum concept to construct a working theory unifying a large body of observational knowledge and permitting the deduction of useful conclusions obviously has nothing to do with any assumption as to the real nature of matter. The existence of areas in which the theory is not applicable does not destroy its usefulness in other areas. In most of the analyses in continuum mechanics, two further assumptions about the nature of the material are made, namely, that it is homogeneous and isotropic. It should be clearly understood that the three assumptions are com- pletely independent. Continuity. A material is continuous if it completely fills the space that it occupies, leaving no pores or empty spaces, and if furthermore its properties are describable by continuous functions. Homogeneity. A homogeneous material has identical properties at all points, Isotropy. A material is isotropic with respect to certain properties if these Properties are the same in all directions. It is quite possible, for example, to conceive of the atmosphere as a con- tinuous fluid but still suppose that its density decreases with altitude. It is also possible to conceive of a rolled steel plate that is both continuous and homo- geneous but has different tensile strengths or different elastic moduli in tension specimens cut from the plate, one with its axis in the direction of rolling and one with its axis at right angles to the direction of rolling. This kind of an- isotropy can be introduced into continuum mechanics, and has led to useful results, The developments of the concepts of stress and strain in Chaps, 3 and 4 and the general principles of Chap. 5 use only the basic assumption of con- tinuity. The other two simplifying assumptions are not needed until constitu- tive equations (stress-strain relations) or strength properties of a material are considered. It is sometimes said that solutions to engineering problems obtained through the simplifying assumptions of the engineering theory of beams are only ap-See. 1d Introduction 3 proximations to the “exact” solutions of the theory of elasticity. The fore- going paragraphs should have made it clear that the adjective “exact” is not justified. There may be exact mathematical solutions to the equations formulated in elasticity or in other branches of continuum mechanics, but the equations themselves are not exact descriptions of nature. In this respect the difference between the elementary theory and the advanced theory is one of degree rather than of kind. When the elementary theory is formulated consistently and logically, it is just as respectable as the advanced theory from a mathematical or logical point of view. And from a practical point of view it is just as good in those areas where its predictions agree closely enough with experience. The bounds of applicability of these elementary theories are determined by experi- ence, either from experimental verification, or from comparisons with predic- tions of the more advanced continuum theories. The last fifteen years have seen a spectacular growth in general continuum. theory as well as in its specialized branches. The theory divides logically into three parts: (1) general principles, assumptions and consequences applicable to all continuous media, (2) constitutive equations defining the particular idealized material, and (3) specialized theories of each individual idealized material built on the foundations of the general principles and the constitutive equations of that material to the point where boundary-value problems are formulated and solved for application to specific problems. The third part would require a book at least as long as this one for each specialized theory. Moreover, despite some superficial resemblance between the application of Laplace's equa- tion in fluid mechanics and its application in some parts of linearized elasticity, there is very little common ground in the way the complete specialized theories are developed and applied. Hence specialized treatises must be consulted for the individual materials. Two small samples are included in Chap. 7 on fluid mechanics and Chap. 8 on linearized elasticity, to give some introduction to the way these two theories are put together and applied. References to spe- cialized treatises on fluids and elasticity are given in those chapters. Some works ‘on viscoelasticity and plasticity are also cited in the sections dealing with the constitutive equations of viscoelasticity and plasticity in Chap. 6. Until recently continuum mechanics was presented and studied almost exclusively in its separate specialized branches, but during the last ten years a definite trend has developed toward a unified treatment of the first two parts of the theory, the general principles and the constitutive theory. The general principles are best studied within the framework of a general continuum theory. This approach emphasizes their general applicability, which is easily lost from sight when the general principles are interspersed with results applicable only to a specific theory. Some pedagogical economy is also achieved by presenting them only once instead of in two or three courses that might be taken by the same student. But an even more important advantage is4 Introduction Chap. 1 that the general principles can be presented more carefully, avoiding the tend- ency to hurry through them to get to the applications. Most of the material in the first five chapters of this book, covering the general principles, is not new. Important concepts date back to Euler (1770) and much of the formal development was carried out during the early part of the nineteenth century by such pioneers as Cauchy, Navier, and Green, to mention only a few. The foundations of the theory have been subjected 10 ex- tensive recent critical review. C. Truesdell (1952) published an important study, “The Mechanical Foundations of Elasticity and Fluid Mechanics,” reprinted in book form in Continuum Mechanics, Vol. I (1966). Two monu- mental treatises, Classical Field Theories by Truesdell and R. Toupin (1960), and Nonlinear Field Theories of Mechanics by Truesdell and W. Noll (1965), have included comprehensive accounts of the foundations of the theory as well as of the newer developments in constitutive theory, with the most complete historical and bibliographical information on the subject up to that time. Because of the availability of these reference works no attempt will be made in this book to give a complete historical account, especially of the early work. For those early scientists mentioned in this text in connection with develop- ments attributed, to them, complete bibliographical information can be found in the three works cited above, These invaluable reference works and a few recent books at various levels on continuum mechanics are included in the list of references at the end of this chapter. The most noteworthy recent developments in continuum mechanics have been in the theories of constitutive equations. The classical constitutive equa- tions presented in the first six sections of Chap. 6 were developed indepen- dently before the advent of the modern theories of constitutive equations, although a few modifications to them have been made as a result of the modern theories. The modern theories are briefly referred to in Secs, 6.7 and 6.8, but their full development and application, especially to nonlinear constitutive equations, is beyond the scope of this introductory book. The modern theories are in essence theories of theories, setting some limitations on the forms the individual theories can take, once the proposed independent and dependent variables have been chosen. This reduces the amount of experimentation and intuition required to furnish the information needed for completing the def- inition of the individual idealized material. Historically the linear theories and the theory of plasticity developed by a different route, beginning with assump- tions based on intuition and experience and only later being subjected to critical scrutiny to determine whether the form of the theory satisfies the requirements of a theory of theories. Some new constitutive equations will no doubt continue to spring up in this fashion, growing out of the needs of engineers and scientists. But the modern constitutive theories promise valuable aid in formulating new theories, especially of nonlinear material behavior.Sec. 1.4 Introduction 5 Some of the pioneering contributions to the modern theories, especially the work of R. S. Rivlin and his associates and of W. Noll, dating from about 1956, are cited in Secs. 6.7 and 6.8. Many of the early papers and subsequent papers continuing the development of the theory have been reprinted in Continuum Mechanics, Vols. 1, II, 111, 1V (1965-66), a four-volume series reprinting 49 papers published by 18 authors between 1945 and 1961, edited by C. Truesdell, referenced at the end of this chapter. The most complete account of the de- velopment growing out of the work of Noll is in the Nonlinear Field Theories of Mechanics already cited. For more introductory recent accounts see Jaunzemis (1967), Truesdell (1965), Leigh (1968), and Eringen (1967). The general principles of Chap. 5 applicable to all continuous media, and constitutive equations defining the particular material, for example one of those given in Chap. 6, form the field equations of a continuum theory. The me- chanical variables appearing in these equations are the stress and kinematic variables such as strain and rate of deformation. In elementary mechanics of materials, tensile strain is defined as elongation per unit length of the test specimen and is a dimensionless measure of the deformation. Stress is the in- ternal force brought into play by the deformation, per unit cross-sectional area: it is equal to the externally applied force, per unit cross-sectional area. The stress tensor of a three-dimensional combined-stress field is a generalization of the elementary concept of stress, which will be developed in Chap. 3. The small- strain tensor of Secs. 4.1 and 4.2, a generalization of the elementary concept of strain, is the only kinematic variable appearing in the constitutive equation of linearized elasticity. But additional kinds of kinematic variables are needed for more general continua. Some of these additional kinematic concepts, in- cluding rate of deformation and finite strain and rotation tensors, will be developed in Chap. 4. The stress tensor will be treated first, since it is simpler and casier to visualize for most engineers, and it affords an opportunity to develop some skill in the use of tensors before the more complicated kinematic tensor development is encountered. As a preparation for the stress tensor development of Chap. 3 and the kinematic tensor development of Chap. 4, a summary of vectors and tensors is given in Chap. 2, which can serve either as a quick review for the reader already acquainted with tensors or as a self-contained introduction to tensors. Additional material on curvilinear components of tensors is given in Appendices Tand I}, but these are not needed for the main body of the text. Some addi- tional vector and tensor concepts are also presented where they are first needed in Chaps. 3,5, and 7, and the reader to whom the tensor methods are new will find that the repeated use of them required throughout the book and the Many cross references to Chap. 2 will greatly strengthen his understanding beyond that acquired in a first reading of Chap. 2.6 Introduction Chap. 1 Selected references for collateral reading COLLECTIONS OF ORIGINAL PAPERS AND TREATISES Truesdell, C., ed., Continuum Mechanics (reprints). New York: Gordon and Breach, vy. 1 (1966), v. II, IIT, LV (1965). Truesdell, C. and R. A. Toupin, “The Classical Field Theories,” : Encyclopedia of Physics, v. W1/1, ed. S. Fliigge. Berlin: Springer-Verlag, 1960. Truesdell, C. and W. Noll, “The Non-Linear Field Theories of Mechanics,” in Ency- clopedia of Physics, v. 11/3, ed. S. Fliigge. Berlin: Springer-Verlag, 1965. IntRopucTory TexTsooxs Not INCLUDING MODERN CONSTITUTIVE THEORIES Aris, R., Vectors, Tensors, and the Basic Equations of Fluid Mechanics. Englewood Cliffs, N. J.: Prentice-Hall, Inc., 1962. Frederick, D. and T.S. Chang, Continuum Mechanics. Boston: Allyn and Bacon, Inc., 1965. Fung, ¥.C., Foundations of Solid Mechanics, Englewood Cliffs, N.J.: Prentice- Hall, Inc., 1965." Long, R.R., Mechanics of Solids and Fluids. Englewood Cliffs, N.J.: Prentice- Hall, Inc., 1961. Prager, W., Introduction to Mechanics of Continua. New York: Ginn and Com- pany, 1961. Scipio, L. A., Principles of Continua with Applications. New York: John Wiley & Sons, Inc., 1967. TEXTBOOKS INCLUDING MODERN CONSTITUTIVE THEORIES Eringen, A. C., Mechanics of Continua. New York: John Wiley & Sons, Inc., 1967. Eringen, A. C., Nonlinear Theory of Continuous Media. New York: McGraw-Hill Book Company, 1962. Jaunzemis, W., Continuum Mechanics. New York: The Macmillan Company, 1967. Leigh, D. C., Nonlinear Continuum Mechanics. New York: McGraw-Hill Book Company, 1968. Truesdell, C., The Elements of Continuum Mechanics. New York: Springer-Verlag, 1965.CHAPTER 2 Vectors and Tensors 2.1 Introduction Physical laws, if they really describe the physical world, should be independent of the position and orientation of the observer. That is, if two scientists using different coordinate systems observe the same physical event, it should be pos- sible to state a physical law governing the event in such a way that if the law is true for one observer, it is also true for the other. For this reason, the equa- tions of physical laws are vector equations or tensor equations, since vectors and tensors transform from one coordinate system to another in such a way that if a vector or tensor equation holds in one coordinate system, it holds in any other coordinate system not moving relative to the first one, i.e., in any other coordinate system in the same reference frame. Invariance of the form of a pirysical law referred to two frames of reference in accelerated motion rela- tive to each other is more difficult and requires the apparatus of general rela~ tivity theory, tensors in four-dimensional space-time. For simplicity, we limit ourselves in this book to tensors in three-dimensional Euclidean space and for most of the discussion use only rectangular Cartesian coordinates for the com- ponents of tensors and vectors. The way in which the physical laws defining material response transform from one frame of reference to another frame moving relative to the first will be explained in Sec. 6.7 by means of the pos- tulated principle of material frame indifference. Most readers will have used vectors before; however, we review in Secs. 2.2 to 2.5 and Sec. 5.1 some of the ideas of vector algebra and calculus. The con- cept of a tensor asa linear vector operator will be presented in this introduction and further developed in Sec. 2.4, where some properties of tensors are sum- marized and Cartesian tensor components are defined. The Cartesian com- ponent formulation is sufficient for the development of the theory of continuum mechanics and will be used throughout the main body of the text. For the solution of specific problems, orthogonal curvilinear coordinates suitable to the ans8 Vectors and Tensors Chap. 2 geometry of the problem may lead to simplification of the analysis. Appendix TJ presents the physical components of vectors and tensors in orthogonal cur- vilineat coordinates and summarizes the equations of continuum mechanics in this form. It is recommended that on a first reading the material of this appendix be omitted or postponed unti] the continuum theory has been com- pleted. Appendix I presents the more complicated formulation of general tensor components referred to an arbitrary curvilinear coordinate system; this general formulation is not actually necessary for the development of continuum mechanics, and is included only because many modern writers on the subject have chosen to use this language. The advanced student can profit from some acquaintance with the language of general tensors in order to read the literature, but it is recommended that App. I be omitted in an introductory course. Tensors as linear vector operators. In three-dimensional space. a vector may be visualized as an arrow that has length equal to the magnitude of the vector and that is pointing in the direction of the vector; it thus is seen to have an existence quite independent of any coordinate system in which it might be observed. Although a tensor cannot be visualized simply as an arrow, it also has an existence independent of the coordinate system. Tensors frequently arise as physical entities whose components are the coefficients of a linear relationship between vectors. For example, in Chap. 3 we will find that for every direc- tion at a point, specified by a vector f, there may be associated a second vector t, the traction vector (force per unit area) acting at the point across the surface whose unit normal is A. At a given point, t is thus a vector function of fi; symbolically t = t(f). The functional dependence in this case is an example of a linear vector function. (A vector function F is said to be a linear vector function if F(an + by) = 2F(u) + bF(v) for every choice of the vectors u, ¥, and real numbers a, b.) Since the argument fi of the function and the func- tional value t associated with it are both vectors, and therefore have an exis- tence independent of any coordinate system, the functional relationship must then be independent of any coordinate system. Actually, in practice, we observe the two vectors in a particular coordinate system and express the func tional relationship by means of coefficients referred to that coordinate system. These coefficients will be the components of the tensor that relates one vector to the other. And we will derive tensor transformation formulas giving the tensor components in any other coordinate system, such that when these com- ponents are used as coefficients of the vector function, the function yields the same vector as before, but the vector is represented by its components in the new coordinate system. In any one rectangular coordinate system, the linear vector functional rela- tionship may be represented as a matrix equation involving the square matrixSec. 2.1 Vectors and Tensors 9 of tensor components either premultiplied by the argument vector expressed as a row matrix or postmultiplied by the argument vector expressed as a column matrix. (See Part 4 of Sec. 2.4, for a review of elementary matrix concepts.) This is a very important kind of relationship in all branches of physics. Some examples are: 1. Tis stress tensor (square matrix). nis unit vector normal of surface (row matrix). tis traction vector on surface (row matrix): t=nT, 2. Tis inertia tensor of a rigid body (square matrix). is angular velocity vector of body (column matrix). A is angular momentum vector of body (column matrix): h=Io. 3. E_ is small strain tensor (square matrix). dX is infinitesimal initial relative position vector of adjacent material particles (column matrix). du is relative displacement vector produced by straining (column matrix): du = Edx. 4. K_ is thermal conductivity of anisotropic medium (square matrix). q_ is heat flux vector (column matrix). ‘V8 is gradient of the temperature (column matrix with components 26 20 26). ax’ By 32” q=~—KV6. 5. 3 is electrical conductivity of anisotropic medium (square matrix). e iselectric field intensity vector (column matrix). i is current density vector (column matrix): i=De. These are just a few of the many physical laws in which a tensor is used as the function associating one vector with another. Remember that the relation- ship between vectors must be independent of the coordinate system, although we usually observe it in some one system; and remember that it is precisely this independence of the coordinate system that motivates us to study tensors at all. Before beginning our study of tensors we first review (in Secs. 2.2 and 2.3) some of the notions of elementary vector algebra.10 Vectors and Tensors Chap. 2 2.2 Vectors;+ vector addition; vector and scalar components; indicial notation; summation convention; finite rotations are not vectors In a three-dimensional Euclidean space, vector quantities (or briefly, rectors) are entities possessing both magnitude and direction and obeying certain laws; a vector is conventionally represented by an arrow that is pointing in the direction associated with the vector and that has a length proportional to the magnitude of the vector. The simplest example is a directed line joining two points. Other familiar examples are force, velocity, and acceleration. A scalar quantity, on the other hand, does not have any direction associated with it, although some scalar quantities may have positive and negative values, e.g., Fahrenheit temperature. In printed text, a vector is usually denoted by a boldface letter. The magnitude of the vector is then usually denoted by the same letter in lightface italic type. In typewritten or handwritten text, the wavy underscore used by copyreaders to designate boldface type is often used for both vectors and tensors, or the vector may be identified by an arrow above the fetter, or by underlining. The magnitude of vector a is denoted by a (the same letter in italic instead of boldface and not underlined), or by [al, and is always a positive number or zero. In script @ = Jal. Two sectors are equal if they have the same direction and the same mag- nitude. Thus a vector is unchanged if it is moved parallel to itself. In some applications, however, the point or the line of action of a vector quantity is important. For example, a force is a vector quantity with a definite point of action; it is a fixed or hound vector. In rigid-body dynamics, only the line of action of the force is important, not the point; force is then a sliding vector. When we say that two forces are equal, however, we mean only that they have the same magnitude and direction; this does not imply that they are mechani- cally equivalent. Vector addition. Any two vector quantities of the same kind (e.g., two forces or two velocities) may be represented as two vectors a and h so placed that the initial point of b coincides with the terminal point of a, as in Fig. 2.1. The swum of a and b is then defined as the vector ¢ extending from the ini- tial point of a to the terminal point of b: +b. (2.2.1) Fig. 2.1 Addition of vectors. Only those definitions and formulas used in the text are summarized here; they are so important in mechanics that they should be mastered by every serious student of the subject. For additional details, or for an axiomatic development of the theory, see books ‘on vector analysis and vector spaces.See. 2.2 Vectors and Tensors W The actual addition of two vectors is most conveniently performed by using their rectangular components in some coordinate system as in Eq. (8). but the definition and properties of vector addition do not depend on the introduction of a coordinate system. From the definition, it follows that b+a=a-+t as is indicated by the dashed arrows in Fig. 2.1. Thus vector addition obeys the commutative law. It also follows from the definition that the associative law (a+b) +ce=a+ (b+ c) is satisfied. The common value of the two expressions is written as the sum 2 +b + of the three vectors Ifa vector is reversed in direction with no change in magnitude, the re- sulting vector is called the negative of the original vector. For example, BA = — AB in Fig. 2.1. To subtract b from a, add the negative of b to a: a—b=a+(—b). (2.2.2) A vector a may be multiplied by a scalar c to yield a new vector, ca or ac, of magnitude |ca|. If c is positive, ca has the same direction as a; if ¢ is nega- tive, ca has the direction of —a. If c is zero, the product-is the zero vector with zero magnitude and undefined direction. Thus (0)a = 0. Sometimes 0 is written as 0, not in boldface and not underlined in script. A vector of unit length is called a unit vector. Any vector may be written as the product of the scalar magnitude of the given vector by a dimensionless unit vector in the given direction. Thus a = ai, where i is a dimensionless unit vector in the direction of a. A unit vector may be signified by the caret (*) placed over its letter symbol. The caret is often omitted, when the context makes it clear that the vector is a unit vector. Vector components and scalar components. Two or more vectors whose sum is a given vector are called rector components of the given vector. It is often convenient to resolve a given vector into three components parallel to three noncoplanar base vectors; the resolution is unique for a given vector and given base vectors. The set of base vectors is called a basis. If the basis con- sists of mutually orthogonal unit vectors, it is called an orthonormal basis. In this book, the base vectors are usually specified as unit vectors in the positive directions of a set of rectangular Cartesian coordinate axes. The components are then called the rectangular vector components of the given vector in that coordinate system. The name rectangular components is also used for the numerical coefficients of the base vectors. that is, for a,,.a, and a. in Eq. (4) below.12 Vectors and Tensors Chap. 2 Fig. 2.2. Rectangular components and unit vectors. In Fig. 2.2, for example, PA, AD, and DE are the rectangular vector com- ponents of the vector PE, in the directions of the unit base vectors i, j,k. We denote the vector by a, its magnitude by a, and the angles it makes with the positive coordinate directions by @,,@,,@,. Then the three rectangular com- ponents of the vector, that is, the numerical coefficients of the base vectors, are @, = acosa,, a, = acosa,, a, = @Cos a, (2.2.3) and the rectangular vector components are a,i, a,j, @,k so that a=a,i+aj+a.k. (2.2.4) Note that a, is a number (positive, negative, or zero), while a.i is a vector, 4 directed physical quantity parallel to i. The rectangular components are the orthogonal projections of the given vector onto the three coordinate axes. The word component is also often used as a synonym for “projection on an arbitrary direction,” For example, a unit vector fi may be given (not necessarily parallel to a coordinate axis); “the component a, of » in the direction of a” then simply means the orthogonal projection of a onto a line having the direc: tion of f, namely a, = @ cos 6, (2.2.5) where @ is the angle between the directions of a and i. This equation applies whether or not a and f are drawn from the same point, but the angle @ is more easily visualized if one of the vectors is moved parallel to itself to the same initial point as the other.Sec. 22 Vectors and Tensors 13 The three cosines, cos az, cos @y, cos @,, are the direction cosines of the vector. By repeated application of the theorem of Pythagoras we see that the magnitude of the vector is given in terms of the scalar components by at = ab +a} + al. (2.2.6) Substitution of Eqs. (3) into this relationship yields cos? a, + cos? a, + costa, = 1, (2.2.7) which expresses the fact that the sum of the squares of the direction cosines of any direction is unity. The addition of two or more vectors is most simply performed in terms of their rectangular components. For example, if c=a+b and d=a—b then the components of c and d are given by es=a,tb, d,=a,—b, ty=ay+b, dy=a,—b, (2.2.8) ¢:=a,+b, d, = a, — b,. Equations (8) are special cases of the relationship Cy = an + by, 2.2.9) which is valid for the components in an arbitrary direction fi (projections), as is evident from Fig. 2.3, where c = a + b and where a, = OA,b, = AB and c= OB. Fig. 2.3 Projection OB of sum ¢ equals sum of projections.14 Vectors and Tensors Chap. 2 Indicial notation. Instead of denoting the rectangular Cartesian coordi- nates of a point by (x, y, 2), we may call them (x;, X2, x) as is indicated by the labeling of the coordinate axes in Fig. 2.4. eee K Fig. 2.4 Indicial notation: x,-, xr, xyaxes. Vector a. The components of a vector a are then denoted by a,,/ = 1,2,3. The three components are (a, ds, 23). The symbol a, is used for a typical one of the components or for the set of three components. It is also sometimes used instead of a to represent the vector itself; the context prevents serious mis- understanding of this usege. Then if @ is a unit vector in the direction of the vector a, its components are Ny = COS Ay, where the @, are the angles between a and the x,-axes. Hence the components of the vector a are given by a = an, (2.2.10) Equation (10) represents a set of three equations, obtained by giving i, succes- sively, the values 1, 2,3. In fact Eq. (10) is a compact form of Eq. (3). Simi- larly the three equations (8) for the components of ¢ can be written compactly as ey ay + by Summation convention. The indicial notation permits us to write many of the formulas of continuum mechanics in a much shorter form than would otherwise be possible. This brevity makes the formulas easier 10 remember and also easier to understand, once the notation is learned. Many of our formulasSec. 22 Vectors and Tensors 15 require a summation over all values of the index. For example, the magnitude a of a vector a, is given by Eq. (6) as (2.2.11) a Lael We adopt the summation convention which states that in Cartesian coordinates whenever the same letter subscript occurs twice in a term, that subscript is to be given all possible values and the results added together. The symbol 5 may then be omitted because the repeated subscript indicates the summation. For exampie, in three dimensions, ay = ai + a + ai, De = Oy + Ose 7 Assy Gybsy = aids + Geabss + Gabry When general tensor components in curvilinear coordinates are used, summa- tion is only implied by repeated indices if one of the repeated indices appears as a superscript and one as a subscript; ¢.2., by + ath, + abby. by = Equation (11) can now be written as a= aa, (2.2.12) This is a single equation, while Eq. (10) is a set of three equations, because Eq. (10) has a free subscript, ic., a letter subscript not repeated twice in the same term. Note that c, = a; + ; is also a set of three equations because the ‘same subscript does not occur twice in one term. The notation ¢,; = a1; + bry Tepresents nine equations, one for each of the nine possible combinations of nu- merical values 1, 2,3 that the free subscripts i and j can have. Finite rotations are not vectors. Not ail entities representable as directed lines are vectors; to qualify as vectors they must satisfy the laws of vectors. In particular, they must combine according to the laws of vector addition. For example, it is one of the basic postulates of Newtonian mechanics that two forces applied at the same point may be combined according to these laws, i.c., the vector sum applied at the point produces the same effect as the two forces would produce. And this postulate is crucial in establishing that forces are vector quantities. Finite rigid-body rotations are not vector quantities, since they do not combine vectorially, as the following counterexample shows, de- spite the fact that any one rigid-body rotation about a fixed axis may be repre- sented as an arrow laid off to scale along the axis of rotation.16 Vectors and Tensors Chap. 2 = zk Fig. 2.5 Example showing that finite rotations are ‘not vectors. In Fig. 2.5, the rigid body is taken to be the rectangle initially lying in the xp-plane with diagonal OA». It is given the following sequence of 90-degree rotations, each about a fixed axis: 1. About x-axis to position OA,, represented as an arrow (z/2)i in the x- direction, 2. About y-axis to position O.4,, represented by (x/2)j, 3. About z-axis to position OAs, represented by (2/2)k Tf finite rotations were vector quantities, then the single rotation about a fixed axis, defined by the vector sum (2/2)(i + j + k) of the three individual rota- tions, would take the rectangle from the initial position OA, to the final posi- tion OAs. But this is not the case, since evidently the single rotation which takes OA, to OA; is a 90-degree rotation about the y-axis, represented as (x/2)j, which is not the vector sum of the three individual rotations. This shows that finite rotations are not vectors. If a vector quantity is conceived of as a physical entity (e.g., force or ve- locity), it is clear that it has an existence independent of the coordinate system. If another rectangular Cartesian system is used, the components of the vector in the second system are different from those in the first, but it is still the same, vector. lt must therefore be possible to express the vector’s components in the new coordinate system in terms of those in the old one; the formulas for such a change of axes will be given in Sec. 2.4. Thus, if the components of a vector are given in one coordinate system, the vector is completely specified. Before considering change of axes in Sec. 2.4, we will review the notions of scalar product and vector product: two different ways of multiplying two vectors.Sec. 23 Vectors and Tensors 17 2.3 Scalar product and vector product In Sec. 2.2, multiplication of a vector by a scalar was defined to yield a vector. We now define two different kinds of multiplication of one vector by another. In one case, the product is not a vector but a scalar; it is called the scaler product or dot product. The second case yields a vector; it is called the vector product of cross product, A third kind of multiplication yielding a tensor is defined in Eq. (2.4.21). The scalar product is defined as the product of the two magnimdes times the cosine of the angle between the vectors: arb = abcosé. (2.3.1) It follows immediately from the definition that if m and 7 are scalars (numbers), then (ma)+(nb) = mn(a+b). (2.3.2) It is also evident from the definition that the scalar product is commutative, acb = bea. (2.3.3) ‘We may interpret ab cos @ either as b times the projection of a onto the direc- tion of b or as a times the projection of b onto a. ‘The scalar product is also distributive, i.c., a+(b + c) = (ab) + (ac), (23.4) as the following argument shows. Since the projection of b + ¢ onto a equals the sum of the projections of b and ¢ onto a, according to Eq. (2.2.9); and since a+(b + c) may, by the scalar product definition, be interpreted as the product of a by the projection of b ++ ¢ onto a, Eq. (4) takes the form abe + Ca) = abs + ala, which is known to be true by the distributive property of the arithmetic prod- uct of numbers. Here b, and c, are the components (projections) of b and ¢ in the direction of a. The distributive property of the scalar product permits us to derive a formula for the scalar product in terms of the rectangular com- Ponents by using the unit base vectors i, j, k. From the defining equation (1), it follows that the scalar product of any two different unit base vectors is zero, since cos 90° = 0, while the scalar Product of any unit vector by itself equals unity.18 Vectors and Tensors Chap. 2 i= jjakk=I bjs pk =ki=0. (235) Then the scalar product acb = (a,1-+ a,j + ask)-(bei + byj + bk) yields a,b, + ayby + a,b or (2.3.6) in idicial notation. Remember this very important formula. It furnishes the method by which we actually evaluate scalar products. But do not forget Eq. (1), which gives physi- cal significance to what we accomplish by using Eq. (6). One of the most common applications of the scalar product is the projection of a given vector onto a line having a given direction. If a is the given vector and fis a unit vector in the given direction, then the projection a, is given by ay = ah = ayy ayny ban, OF ay = am 3.7) The three components (11,, My, 12) of the unit vector fi are the direction cosines of the given direction. Another important geometrical application is the determination of the angle between two vectors given in terms of their components. From Eq. (1), we obtain (2.3.8) ‘The numerator of this formula is given by Fq. (6), and the two magnitudes in the denominator may be calculated in terms of the components by Eq. (2.2.6). It follows from Eq. (8), that a necessary and sufficient condition for two nonzero vectors to be perpendicular is 0. (2.3.9) Another important application of the scalar product is the calculation of the work done by a force. If force f acts on a point of application which moves through the infinitesimal displacement dr, then the work dIV done is AW =: fede ~ fos 0 ds, (2.3.10)Sec. 2.3 Vectors and Tensors 19 where @ is the angle between f and dr, and where fand ds are the magnitudes of f and dr, respectively. Since, in general, the force may vary both in mag- nitude and direction during any finite motion, it will be necessary to express the force asa function of position and evaluate the total work as a line integral along the path of the motion, When f is constant and the motion is along a straight line, this results in a total work IV given by W=f-Ar (2.3.11) where Ar is the finite displacement. We note again that the work IV isa scalar quantity, obtained by multiplying two vectors, g the scalar product. We now consider another way of multiplying two vectors; this time the product itself be a vector. The second kind of product of one vector multiplied by another is the vector product ot cross product: c=axb, (2.3.12) defined as a vector ¢ perpendicular to both a and b in the sense that makes a, b, ca right-handed system. Magnitude c is given by = absin 6. (2.3.13) (This always yields a nonnegative value for c because the angle is between zero and 180°.) Because of the sense prescription, vector multiplication is not commutative. We have instead bx a= —(a x b), (2.3.14) as illustrated in Fig. 2.6. The magnitudes are the same; in either case the magnitude c is numeri- cally equal to the area absin 6 of the parallelo- gram formed with the two given vectors as sides; but the senses are opposite. : For example, the unit vectors i, j, k of a right- Yoc= bxa handed system form the following products: Fig. 2.6 Vector products. ixi xj=kxk=0 (the zero vector), ixjak, Jxk= kxia} (2.3.15) jxt=-k, kx] . ix k=]. The algebraic signs for the products (15) can be remembered by the scheme shown in the figure on page 20. If the two factors occur in the same sequence20 Vectors and Tensors Chap. 2 as i, j,k on the circle, the product is positive. If the order is reversed, the product is negative. (All the signs 7 # are reversed for a left-handed system.) The vector product is also distributive, but we omit the proof.t Thus ax(b+e)=(ax b+ (axe). (2.3.16) 7 The distributive character of the product permits us to evaluate it in terms of the components of the given vectors. This is done by writing each vector in the form given in Eq. (2.2.4), multiplying the two vectors together, and simplifying by means of Eqs. (15). The result is, for a right- handed system, a X b= (a,b, — a,by)i + (a,bz — azb,)j + (a2by — aybz)k, (2.3.17) which may be remembered conveniently as the formal expansion of the deter- minant ij ok axb=|a, a, a, (2.3.18) jb, byob in terms of the minors of the elements in the first row. For a left-handed coordinate system a minus sign is prefixed to the formulas. The cross product would be an example of an axial vector if we did not employ this con- vention of prefixing a minus sign to Eqs. (17) and (18) in a left-handed system. With this convention, the cross product behaves as a polar rector or true vector under transformation of coordinates (see Sec. 2.4). When the vectors have several zero components, it is often more convenient to apply Eq. (15) directly instead of using Eqs. (17) or (18). For example, 2X (3} — 2k) = 61 x J— AE x k= 6k + 4}. Here and in the derivation of Eq. (17), we have tacitly assumed in addition to the distributive property that (ma) x (nb) = rin(a x b) (2.3.19) for arbitrary scalars m and n, a consequence immediately evident from the definition of the vector product. However, the vector product is not associative aX (bX c)# (a xb) xc. See, for example, Nathaniel Coburn, Vector Analysis (New York: The Macmillan Company, 1955), p. 14.Sec. 2.3 Vectors and Tensors 2 Hence parentheses must be used to indicate the order of multiplication in the vector triple product. With the scalar triple product a x bec, parentheses are not needed, since this product has meaning only if the cross product is evaluated first. However, the dot and cross may be interchanged without changing the (scalar) value of the product. You can easily verify this by writing each vector in the standard component form of Eq. (2.2.4) and carrying out the multiplications for a right- handed coordinate system: a, ay a, axbe=abxec=|b, by by}. (2.3.20) Cr Cy Ce In fact if a, b, ¢ are oriented so that they form a right-handed system (so that this product is positive), the scalar triple product of Eq. (20) gives the volume of the parallelepiped formed with a, b and ¢ as three edges. Applications of the vector product include the calculation of the moment of a force, the determination of the displacements produced by a small rotation, or the particle velocities induced by rotation. One important geometric appli- cation is the determination of a direction perpendicular to two given directions. A unit vector ft perpendicular to both a and b is obtained as A= 2xb. (2.3.21) The rectangular components thus obtained wilt be the direction cosines of the common perpendicular, This is especially useful in relating new coordinate axes to the old ones, An example will be given in Chap. 3; when we are finding the principal axes of stress, we will obtain the third axis as the common perpendic- ular to the first two. In order to write the vector product in indicial notation, we introduce the permutation symbol enn, defined to have the value 0, --1, or —1 as follows: 0 when any two indices are equal; +1 when m,n,r are I, 2,3 or an even nar permutation of 1, 2, 3; (2.3.22) when m,n, 7 are an odd permutation of 1, 2,3. Thus, for example, €1n9 = Cant = Can = in = Cn = Orn = —E eu = ems = ers = 0, ele,22 Vectors and Tensors Chap. 2 The sign convention can be remembered by marking the numbers I, 2,3 on a circle, Then any even permutation will be in cyclic order, ic., it will go around the circle in the same sense as 1, 2, 3, while an odd permutation will go in the reverse direction or acyclic order, Then the three components c, are given in a right-handed coordinate system by C=AXD Cy = Cyardyby = Cryer, (2.3.23) as you can verify by writing out the sums represented in Eq. (23) and compar- ing the results with Eq. (17). Note that because there is one free subscript p, there are three Eqs. (23), and each equation contains nine terms: one for each possible choice of the two summation indices q and r. But when these nine terms are written out, all but two will have a repeated index on the permuta- tion symbol, which means that there ure only two nonzero terms in each nine-term sum—one requiring a plus sign and one a minus sign, The vector equation (17) may be written for a right-handed system as aX b= Cyerdeb,t, (vector sum on p). (2.3.24) For actual numerical evaluations, the determinant scheme of Eq. (18) or direct use of the unit vector products, as in the example after Eq. (18), is more con- venient than the indicial Eqs. (23) or (24), but the indicial expressions are useful in theoretical derivations because they can be substituted conveniently into other equations. The Kronecker delta 8,q is another symbol useful to simplify and shorten equations in index notation. It is defined by oa ee (2.3.25) la By 3. ronecker dela Ba [oer where pand g can each have any positive integer value appearing in the dis- cussion, that is 1, 2, or 3 for the usual case of three dimensions, Asan example of the use of the Kronecker delta, we note that the statement that 4,, 1,, 1, form an orthonormal set, ie. a set of three mutually orthogonal vectors each of unit magnitude, can be expressed by the equation iri, = By (2.3.26) ing out the summations Eq. (25) we can ea implied by the repeated indices, and using y establish the following useful identities. By wri the definition ofSec. 2.3 Vectors and Tensors 23 @) Ban = 3 (0) 8nn Bnn = 3 (©) Um Ban = Hn) Tran Ban = Tax oe The e-B identity of Eq, (28) below can also be verified by writing out the expression, although this requires a more lengthy enumeration of the indepen- dent choices of the free indices. entity cs en By (2.3.28) This identity relating the three-index symbols ¢,,,, of Eq. (22) to the Kronecker delta of Eq. (25) is useful for proving many vector identities. For example, the following identity for triple vector product a X (bX ¢) = (arc)b — (ab)e (2.3.29) may be proved as follows. Let vy; denote the rectangular Cartesian components of the triple product on the left, Then tH = €114(Cirsb-e,) by Eq. (23) Fer Cirayb,C, = (By8ir — By5..)ajb¢. by Eq. (28). ty = (By yC)8ird, — (Byes beBises = (a,¢5)be — (ayby)ce by Eqs. (27c) and (27d). But this is the component form of the right side of Eq. (29), which establishes the required identity. A similar derivation shows that (a Xb) X € = (aeyb — (bee)a, (2.3.30) which is a completely different vector from a x (bX ¢). The vector product therefore is not associative, so that the parentheses must not be omitted on the left sides of Eqs. (29) and (30). We have emphasized the physical or geometric character of a vector, which exists independently of any coordinate system. However, most of our actual computations are made with rectangular components, which require the intro- duction of a coordinate system. We now turn our attention to the relationship between the components of a vector or a tensor in one rectangular coordinate system and the components in any other rectangular coordinate system.24 Vectors and Tensors Chap. 2 EXERCISES 1, Express in terms of I, j, k a unit vector parallel to a= 41 — 2) + 4k. 2, A vector of magnitude 100 is directed along the line from point A(10, ~5, 0) to point B(9, 0, 24). Express the vector in terms of i, J, k. 3, Find: wey, u X v, and v Xu for (a) u= 61 — 4)— 6k, v= 4i—2)— 8k (b)u=3+ J—2k, v= —i+2)— 3k. Find the projection of ¢ = 181 — 27) + 81k onto d, if @) d= 1-24 2k, (b) d= 1-4 2h — 2k. . Show that the vector Al + Bj + Ck is normat to the plane whose equation is Ax + By + Cz =D. Hint: Show that it is perpendicular to the vector joining any two points (x1, ¥1 21) and (xs, ¥2, 22) lying in the plane, Find the component of the force f = 6001 -- 1200) — 900k pounds in the ditec- tion of the normal to the plane 2x — 2y + 2 = 3, choosing the positive normal to be on the opposite side of the plane from the origin, . Find the equation of the locus of all points (x.,', 2) such that a vector from point (2, =1, 4) to (x, », 2) Is perpendicular to the vector from (2, 1, 4) to (3, 3, 2). » a 2 Find the angle between the two vectors from the origin to the two points A(A, 3, 2) and B(—2, 9. Find the area of the parallelogram formed on the two vectors of Ex. 8. 10. Find the unit vector perpendicular to both vectors of Ex. 8. 11, Find a unit vector parallel to the line x — z = y = 2z. Hint: The line is the in- tersection of the two planes x — y — z = O and y — 2z=0. 12, New &, ‘axes are chosen so that the new X-axis makes equal angles with the positive x- and y-axes and the negative z-axis, while the new j-axis lies in the x2~ plane and bisects the angle between the old x- and z-axes. Show that these new 5- and j-axes ate perpendicular, and find the direction cosines of the new 3- axis for a right-handed system. |. Write out the following expressions: @u=Tan @ Ty =264, (emen FE + pir = pa, (©) W= Tye = 14, Verify the identities of Eqs. (2.3.27). 15, Verify Eq. (2.3.20). 16. Verify that Eq. (2.3.23) gives the same results as Eq. (2.3.18), when Eq. (2.3.18) is written with fy, fy, fy for J, J, k and ay for ae, etc.Sec. 2.4 Vectors and Tensors 25 17. Prove the identity of Eq. (2.3.30). 18. Prove the e-8 identity e1y €1rs = 85,51, — 84554, of Eq. (2.3.28). Hint: Two of the four free indices jkrs must be equal. Show first that if » then both sides vanish identically. [For example, eres) = 0, and Bay Buy. — SayeSuyr = 0, where the parentheses around k are used to indicate that there is no sum on k.] Then show that even with j/k and r % 5, both sides vanish when j =r unless also & = 5, and then prove that exer etirney = Soren Bexn — SinwySieny fore # k. What cases remain to consider? 19. If vis any vector and fi is any unit vector, show that v= (vA) fA x (¥ xO). This gives a resolution of v into vector components parallel and perpendicular to A. 20. Show that three concurrent vectors a, b, ¢ are coplanar if eyardybecr 2.4 Change of orthonormal basis (rotation of axes); tensors as linear vector functions; rectany 7 conlponents; dyadics; tensor properties; ary matrix concepts Part 1. Change of Orthonormal Basis A change of rectangular coordinate systems can be made by a translation of the axes without rotation followed by a rigid-body rotation of the axes keeping the origin fixed (and possibly followed by a reflection in a coordinate plane if we allow transformations between right-handed and left-handed systems). The Cartesian components of a vector are not changed by translation of axes, since the projections of a vector onto parallel axes are the same in both magnitude and sense. Hence we need to consider only transformations with fixed origin. Rotation of axes. If new right-handed rectangular Cartesian ¥,, 1, ¥s-axes are chosen with the same origin as the original right-handed x,, xs, Xs-aXxes, as in Fig. 2.7, the direction of each new axis can be specified by giving its direc- tion cosines referred to the old axes, denotedt aj = cos (%,x-). The subscript refers to a new (barred) axis ¥,, while the superscript refers to the original (unbarred) axis x. The direction cosines defining the change of axes are con- veniently tabulated as in Table 2-1. The components of the new unit vector Some authors use ays, and some use air, for the direction cosine. We denote it by af to emphasize that it is not a Cartesian tensor component.26 Vectors and Tensors Chap. 2 Fig. 2.7 Rotation of axes. 1, for example, referred to the old axes, are found in the column below the symbol {,. ‘TABLE 2-1 Direction Costnes FOR ROTATION OF AXES foe if ol al “ha at by ay ay ay Each new unit vector J, is thus given as a linear combination of the old unit vectors i, with coefficients listed in the column. Inversely, each old unit vec- tor is a linear combination of the new unit vectors with coefficients given in a row. Thust {,= ati, and i, = atl, (2.4.1) (vector sum on s in the first case and on r in the second). Until you learn that the subscript refers to the barred system, you can write the cosines as a}. Because of the orthogonality of the unit vectors we must have il, =0 for pq, while {,-1, = 1. This is conveniently expressed by introducing the Kronecker delta 8,9, defined tin the curvilinearcomponent notation of App. I, the second Eq, (2.4.1) would be written as {f= a#I’,so that s would appear as a superscript on both sides of the equa- tion. But in Cartesian components the free indices need not appear on the same level, and we repiace {# by I.Sec. 2.4 Vectors and Tensors 27 to be zero when p and q differ and to be unity when p = g. See Eq. (2.3.25). 7 2.4.2) if pq. oe wt Then the fact that each basis (set of base vectors) is orthonormal, i.e., com- posed of mutually orthogonal unit vectors can be expressed by the orthogonality conditions and in Ban (2.4.3a) or aha = 8s, and aPa? = Sun, (2.4.36) where Eqs. (3b) follow from Eqs. (3a) by use of Eqs. (1), [Note that the left side of each of the Eqs. (3b) is a sum of three terms because of the repeated index s or r.] An additional identity involving the direction cosines follows from the fact that the determinant det |a7| with elements a¥ in the mth row and nth column is equal to the scalar triple product i, X iz+4s of the old unit vectors in terms of their components referred to the new axes, according to Eq. (2.3.20). But this triple product equals unity, the volume of the unit cube whose edges are the three unit vectors 4, I, 15, provided the old axes are right- handed. The triple product equals —1 if the old axes are left-handed. Hence for rotation: det fan] = 1; (2.4.48) while for rotation and refiection: det ja] = —1. (2.4.4b) A linear change of basis of the form of Eqs. (1) is called an orthogonal transformation if the coefficients a{ satisfy Eqs. (3b). Every orthogonal trans- formation satisfies one of the two Eqs. (4). If it satisfies Eq. (4a) [positive de- terminant], it is called a proper orthogonal transformation; if the determinant is negative, the transformation is improper. Hence rotation of axes is repre- sented by a proper orthogonal transformation, while reflection is represented by an improper orthogonal transformation. For most of our purposes we need only consider proper orthogonal transformations, but occasionally an improper transformation is useful. [See, for example, Eqs. (6.2.22).] Polar vectors and axial vectors, transformation of Cartesian components. When we consider how the rectangular Cartesian components of a vector transform under the change of basis given above, and admit the possibility of28 Vectors and Tensors Chap. 2 improper orthogonal transformations, it is necessary to distinguish two kinds of vectors called polar vectors and axial vectors. These vectors are distinguished by the way their components change under an improper orthogonal trans- formation. The axial vector transformation formula differs in having an extra minus sign prefixed if the transformation is improper. No distinction is needed if only proper orthogonal transformations are used. An example of an axial vector would be furnished by the cross product, if we used the definition of Eqs. (2.3.17) and (2.3.23) without prefixing a minus sign in a left-handed system. But with the convention we adopted in Sec. 2.3, the components of the cross product will transform as a polar vector. The polar vector is the typical vector or true vector. Its component trans- formation formula follows from the requirement that the representations of the vector in the two systems have exactly the same form whether or not the transformation is proper. Thus the vector equality i, (2.4.5) ol, = must hold between the two representations, Substituting in Eq. (5) the second Eg. (1) and collecting terms, we obtain G, — vat)h, = 0. Since all the components of the zero vector 0 must be zero, the expression in parentheses must be zero for each r, whence we obtain the three equations Polar Vector v 6 = atv, or inversely v, = até. (2.4.6a) The inverse equations follow in a similar fashion from substituting the first Eg. (1) on the left side of Eq. (5). We sce that the transformation equations for the vector components (6a) have exactly the same form as the transforma- tion equations for the base vectors (1). Despite the similarity in form, the two transformation equations [(1) and (6a)] are quite different, since Eqs. (6a) give the components of the same vector y in two different systems, while the vector equations (1) pertain to two different sets of vectors, one set for each system. Equations (6a) may be written in the following matrix form (see Part 4 of this section for a review of matrix notation): t=ATr v= Ab, (2.4.6b) where A is the matrix shown in the array of Table 2-1, with element a} in row sand column r, A” is the transpose of A, and § and v are column matricesSec. 2.4 Vectors and Tensors 29 with elements @, and v,. Note that in the matrix formulation the matrix A is the matrix used for the inverse transformation from 6 to v. For the ortho- gonal transformation, the inverse A~' of the matrix A is equal to its transpose ‘A’, In the indicial form, the second Eq. (6a) represents multiplication by the matrix A, since the summation is on the column index r, while in the first Eq. (6a) the summation is on the row index s. An axial vector differs from a polar vector in that for an improper orthog- onal transformation of an axial vector a minus sign must be introduced in the defining Eq. (5), which becomes é,1, = —v,i,. Since the minus sign is not needed for a proper orthogonal transformation, both cases may be included by prefixing (det |a™|) in place of the minus sign. Thus axial vector y: Gi, = (det |az|)o, whence as in the derivation of Eqs. (6) we obtain the component transformations Axial Vector ¥ 6 = (det |az|are, —-v, = (det lar |)ate,; (2.4.7a) or in matrix form 6 = (det ja"|)A7v v = (det |at|)A6. (2.4.76) We have already noted that one example of an axial vector would be the vector product of two polar vectors if we did not prefix a minus sign to Eqs. (2.3.17) and (2.3.23) in a left-handed system. Another example would be the curl of a vector field, to be discussed in Sec. 2.5, if we did not employ the same convention about the minus sign as in the cross-product formulas. With our conventions, all the vectors we encounter will transform as polar vectors. But the reader should be warned that some writers do not use the minus sign in the cross-product formulas; then the cross product is an axial vector. Coordinate transformation formulas for the change of axes under rotation and reflection are merely special cases of Eqs. (6) in which the vector v is the position vector r with components , or x, in the two systems. If a translation is added, the most general transformation of rectangular Cartesian coordinates has the form %, = ay(x, — 6,) or inversely x, = b, + ati, (2.4.8) where the a! satisfy Eqs. (3b) and either (4a) or (4b). The transformation property of rectangular components under rotation of axes is sometimes made a part of the formal definition of a vector quantity as follows. A physical entity, defined by three components in every system of rectangular Cartesian coordinates, is a vector quantity if30 Vectors and Tensors Chap. 2 1, the three components in any one system are related to the three com- ponents in any other (meaning every other) system by Eqs. (6), and if 2. two such physical entities of the same type, say a and b with compo- nents a, and 5, in one coordinate system, together produce the same effect as would be produced by the single one, c with components c, = a, + 6, in the same system. In short, to be a vector quantity, the physical entity must transform like a vector and compound like a vector. We remark that it is possible to invent physical entities with three components not satisfying these two conditions. We might take the triplet (p, v, 1) representing the pressure, specific volume, and temperature at a point in a gas. These values do not transform as the components of a vector under rotation of axes; each one maintains its value, ie., each is a scalar point function under rotation of axes. We could always construct a vector artificially by specifying its components in one system of axes and prescribing that its components in any other system should be given by the vector transformation formulas, When we are dealing with physical entities, however, we do not have this much freedom of definition, as is shown by the absurd example (p, v, 1). Part 2. Second-Order Tensors as Linear Vector Functions (Transformations) A vector function is a correspondence which associates with each argument vector v in a certain set of vectors (the domain of definition of the function) another vector, which is the value of the function. The function is also called a transformation, since it transforms one set of vectors (the domain) into another set of vectors. A vector function F is a linear vector function if it satisfies the two requirements that for any vectors u and y in the domain and any real number c, F(u + v) = Fu) + F(v) and F(cu) = cF(u), (2.4.9) For the purposes of this book we define a second-order tensor (also called a second-rank tensor) to be a linear vector function. A vector is a first-order tensor, while a scalar is a tensor of order zero. Higher-order tensors will also be defined later, but the second-order tensors are the ones of primary interest in this book. For brevity, we will often use the word tensor to mean second- order tensor. Since the two sets of vectors have an existence quite independent of any coordinate system in which they might happen to be observed, the ten- sor correspondence or pairing of each antecedent element in the domain with its image element in the second set must also have an existence independent of any coordinate system. In order to make explicit the way in which any partic-See. 24 Vectors and Tensors 3M ular tensor operates on the argument vector, however, we will introduce a coordinate system. For simplicity, we limit ourselves for the time being to rectangular Cartesian coordinate systems, but the tensor itself is not limited to any coordinate system. We introduce a notation for the second-order tensor itself (as distinct from any possible component representation). A boldface capital letter symbol will be used in printed text for a second-order tensor, while a boldface lower-case letter will identify a vector. Some exceptions to this distinction between capitals and lower-case symbols may be made with commonly used tensors or vectors whose physical nature will help us to re- member whether they are vectors or second-order tensors. In typed or hand- written text, the second-order tensor will be identified by wavy underscores, eg, TorE. In the following, the word “tensor” will mean “second-order tensor” unless otherwise stated, Linear transformations of rectangular Cartesian vector components, tensor components. The second Eg. (9) requires that the zero vector be transformed into the zero vector (put ¢ = 0 to see this). In any one rectangular Cartesian system, the most general linear transformation of the components v, of the argument vector v into the components u, of the image vector u, which always transforms the zero vector into the zero vector, has the form % =Tyv, with matrixt notation u = Tv (2.4.10a) or the form, for S = TT and T = S*, 4 = Sy, with matrix notation uv = oS. (2.4.10b) If Eqs. (10b) always give the same output vector components as are given by (10a) for an arbitrary input vector v, then the two forms are not fundamentally different, since they differ only in how the same nine coefficients are labeled: ‘Sy, in (b) being equal to 7,, in (a), ie., in the matrix notation the matrix Sin the second equation is the transpose of the one which should be used in the first equation. Each of the two tensors T and S with rectangular Cartesian component matrices T and S = T” is said to be the tensor transpose of the other, denoted S=T and T=S". (2.4.10c) When we do not plan to write both forms for the same tensor, we let T denote the matrix of components in the form which we actually use. The nine coef- ficients are the rectangular Cartesian components of the tensor T. They are denoted by the same letter printed in ordinary type instead of boldface (without tA brief review of some elementary matrix properties is given in Part 4 of this section.32 Vectors and Tensors Chap. 2 the wavy underscore in handwritten text) and with two subscripts attached, eg., Ti Tr, etc. Some exceptions will be made to the rule that components are denoted with the same letter as the tensor. For example, the strain tensor is often denoted by E with components e,;. The linear transformation of Eq. (10) is the representation of the tensor T (linear vector function) in one system of Cartesian components. This transformation of the components of one vec- tor into the components of another vector referred to the same system should not be confused with the change of components of one vector under rotation of axes, given by Eq. (6a). Symmetric tensors and skew or antisymmetric tensors. A tensor is said to be symmetric, if its matrix of rectangular Cartesian components is symmetric, ie, if Ty, = Ty). If the matrix of a tensor is symmetric in one Cartesian coor- dinate system, it is symmetric in all such systems; thus symmetry is really a tensor property. A skew or antisymmetric tensor has T,, = —T,, in any Cartesian system, which implies that the diagonal elements, 7, Tr, T'ss are all zero. We will now derive the rule for change of components of a tensor under rotation of axes. Rotation of axes, change of tensor components. The rule for changing second-order tensor components under rotation of axes follows from the rule for changing vector components when the transformation (10) is required to apply in each of the two systems. For example, Eq. (62) yields a = alu, aiT,% by Eq. (108) =alT,,a$6, by Eq. (6a). Also ii, = T,,8, by Eq. (10a) in the barred system. By equating these two different expressions for i, and by collecting terms on one side of the equation, we conclude that (Tip — ala$T,)8, = 0 for every possible choice of the argument vector vy. Hence the quantity in parentheses on the left must vanish, whence Tip = alagT,, and inversely T,, = alatT,,, (2.4.11) or in matrix notation To=ATA T= ATA, (2.4.12) where for the proper orthogonal transformation A~! = A’, The inverse relationSec. 24 Vectors and Tensors 33 can be derived in a way similar to the derivation of the first Eq. (11). Com- paring these formulas for the change of components of a second-order tensor with Eqs. (6a) for a polar vector (first-order tensor), we see that they follow a similar pattern. There is on the right side one free index on each direction cosine, corresponding to the free indices on the left, and each direction cosine also has a summation index corresponding to one on a tensor component on the right. Each of the nine first equations (11) has a sum of nine terms on the right, corresponding to the nine possible values of the pair of indices jg. For example, Fay = alalTy, + abalTy + alah Ty, + dlaiTy + aah + atai Ts (2.4.13) + ataiTy, + alaiTs: + atasT as. ‘We remark that exactly the same change-of-component formulas apply whether the components are taken in the form of Eq. (10a) or of Eq. (10b), provided the same choice is made in both coordinate systems. These change-of-component formulas are sometimes made the basis for the definition of a “Cartesian tensor.” Briefly, a second-order Cartesian tensor would be characterized as an entity with nine components in any Cartesian coordinate system, such that the components in any one system are related to the components in any other system by Eqs. (11) and (12). Higher-order ten- sors can be similarly defined, for example, with three subscripts T,,., and with change-of-axis formulas similar to Eq. (11) but with three direction cosines in each term. We will be concerned primarily with vectors and second-order tensors. For our purposes it is more convenient to define the second-order tensor as @ linear vector function and to deduce that Eqs. (11) and (12) relate its components in any two rectangular Cartesian systems. We postpone until the appendix any consideration of how the tensor would be represented by components in a curvilinear coordinate system, but our basic definition will still apply, since it is not dependent on any coordinate system. Dyadic (single-dot) product notation for a second-order tensor. Many of the general equations containing tensors have a simpler appearance when writ- ten in terms of the symbolic representation T instead of in terms of the com- ponents, We use the Gibbs dyadic notation (see Part 3 of this section) to give a symbolic representation independent of any choice of coordinate system for the linear vector operation represented in rectangular Cartesians by Eqs. (10). Dyadic Notation for Tensors as Linear Vector Operators u=Tey usvS — whereS = 7 (2.4.14a) (2.4.14b)34 Vectors and Tensors Chap. 2 The symbolic representation T+v is not merely another notation for the matrix product Tv of Eq. (10a). T+v symbolizes an operation pairing one physical or geometric vector with another one, while Ty is the mathematical product of appropriate matrices of components of T and of v in one coordinate system. In general curvilinear tensor components some care is necessary in evaluating T+v as a matrix product [see App. I, Eq. (1.2.14)]. For example, the matrix prod- uct of two matrices of covariant components is not in general equal to the matrix of covariant components of u. The covariant component matrix of o can be obtained as the matrix product of the covariant components of T by the contravariant components of y instead of by the covariant components of vy. In rectangular coordinates there is no difference between the different kinds of components, and the matrix forms of Eqs. (10) can always be used to rep- resent in any one coordinate system the operation symbolized by Eqs. (14). Some authorst use a symbolic operational product notation without the dot. For example the relationship we have expressed by Eq. (14a) would be written =Ty, (2.4.15) an operation completely independent of any component representation. Some properties of second-order tensors. The sum S = T + U of two tensors is the single tensor which operates on any vector v to yield the sum of the two vectors obtained by operating with T and U separately; in rectangular Cartesian components, S;; = T,, + U,,. The zero tensor 0 assigns the zero vector as image to every argument vector y; its components are all zero in every coordinate system. Multiplication by a scalar c is defined, for example, by the requirement that (cT)-v = c(T+y) for every vector vy. The rectangular Cartesian components of cT = Te are cT,,=T jc. Tensor addition and multiplication by a scalar obey the following four addition axioms and four scalar-multiple axioms, characteristic of a generalized vector space; hence the set of all tensors is a vector space (see Sec. I.1). ADDITION AXIOMS @) T+U=U4+T commutative (ob) T+(U+V)=(T+U)+V associative © T+0=T (4) For each T, there exists another, called —T, such that T+(-T)=0. (2.4.16) See, for example, Truesdell and Noll (1965) and Jaunzemis (1967).Sec.24 Vectors and Tensors 35 SCALAR-MULTIPLE AXIOMS (Note aT = Ta) (a) a(bT) = (ab)T @) IT=T (©) (a+ dT =aT +06 @) a&T + U)=aT+aU (2.4.17) The scalar product of two tensors is a scalar, denoted T : U, which can be calculated in terms of components of the two tensors in any one rectangular Cartesian system, by T:U=TyUy (2.4.18a) ie. by the sum of the nine products of components. This formula applies only to rectangular Cartesian components, but the result, a scalar, is independent of any coordinate system. Some authors define the scalar product as T,,U,, in- stead of T,,U,, and some use the notation T++U instead of T:U. In this book, T: U will always mean 7,,U,, in rectangular Cartesians, while THU = TyUy (2.4.18b) will represent another possible scalar product, different from T : U unless one of the two tensors T or U is symmetric. The scalar products satisfy the fol- lowing four axioms, which with (16) and (17) characterize an inner-product space. See also Eq. (25b) below, where the scalar products are expressed as traces of the product of two tensors. Inner-PRODUCT AXIOMS @) T:U=U:T (b) T3U+V)=T:U4¢T:V © &T:U)=(@T):U =T: (aU) @) T:T>0 unless T=0 (2.4.19) The identity tensor, also called the unit tensor and denoted by 1, is de- fined by the requirement that levovl=y (2.4.20) for all vectors y. The rectangular Cartesian components of 1 are 8,,; its matrix of rectangular Cartesian components is the identity matrix with zero in each off-diagonal position and one in each diagonal position.36 Vectors and Tensors Chap. 2 The tensor product or open product of two vectors, denoted ab, is a tensor called a dyad, defined by the requirement that (ab)-7 = a(b-v) (24.218) for all vectors vy. That is, if T = ab, then T-v = a(b-y) for all vectors v; in rectangular Cartesian components Ty = aby. (2.4.21b) Some writers use the notation 2 © b instead of ab. The product of two second-order tensors, denoted T+ U in this book, means the composition of the two operations T and U, with U performed first, de- fined by the requirement that (T-U)-y = T-(U-y) (2.4.22a) for all vectors v. If TU (2.4,22b) then Pi, Ur or in matrix notation, where P, 7, and U are the matrices of components in any one rectangular Cartesian system. Some authors omit the dot and write the tensor product as - P=TU. (2.4.22c) The product satisfies the linearity rules and is therefore also a second-order tensor. It also satisfies the following five axioms, which together with (16) and (17) characterize the set of all second-order tensors as an algebra. Similar def- initions and axioms apply to ve(U-T) = (v-U)-T. TENSOR ALGEBRA AXIOMS (a) (T+U)*R = T-(U-R) (b) T(R+U)=TeR4T-U © (R+U)-TH=R-T+U-T (2.4.23) @) ATU) = (aT)-U = T-(aU) © PT=Te1=T The product is not commutative; in general U-T # T-U. Instead, Eq. (49) below, when applied to the matrices of rectangular Cartesian components, shows thatSec. 24 Vectors and Tensors 37 BeA = [AT-BT}r (2.4.24) where the superscript T denotes the transpose as in Eqs. (10). In exceptional cases, when T-U = U-T, we say that T and U commute. For example, the unit tensor 1 commutes with any other tensor. The trace, denoted trT, of a second-order tensor T is a scalar invariant function of the tensor, having the same numerical value in all coordinates systems. In rectangular Cartesians (see Sec. 3.3) uT=Tn. (2.4.25) Thus tr T is equal to the trace of the matrix of rectangular Cartesian com- ponents, the sum of the elements on the main diagonal. The scalar products of Eqs. (18) can be represented as traces of tensor products. It is left es an exercise for the student to show that As+B = tr(A-B) A:B = tr(A-B’) = tr(A7-B) tr(A-B) = tr(B-A) tr (A+B-C) = tr (B+C+A) = tr (C+A-B) (2.4.25b) The last line demonstrates the cyclic property of the trace of the continued tensor product of any number of factors; the trace is unaltered in value by any cyclic rearrangement of the factors. A test for tensor character. An entity T defined by nine components in any coordinate system is a tensor, if and only if T:(av) is a scalar invariant for arbitrary vectors u and y, where the symbol “:” denotes the scalar product of Eq. (18a), while uv denotes the tensor product of Eq. (21). This means that in rectangular Cartesian components Tesi, = Toate (2.4.26) in every pair of Cartesian systems, for every choice of the vectors a and v. If the components being tested satisfy the conditions for symmetry of a tensor in all coordinate systems, it is sufficient to show that T: (vv) is a scalar invariant for arbitrary v and symmetric T, (2.4.27) which means that F,0,6, =Tyevpr, and Ty =Ty Ty = Te (2.4.28) in every pair of Cartesian systems for arbitrary choice of y.38 Vectors and Tensors Chap. 2 Part 3. Algebra of Dyads and Polyads; Dyadies In Eq. (21), an operational definition was given for a dyad or open product of two vectors. Higher-order polyads can be formed by open products of more than two vectors, .g., triads abe or tetrads abed. Without assigning any phys- ical significance to these polyads for the moment, we postulate that open multiplication obeys all the rules of ordinary algebra except that open multi- plication is not commutative. ab ¥ ba. (2.4.29a) We define the conjugate dyad (ab), to ab by (ab). = ba. (2.4.29) A linear combination of dyads with real coefficients is called a dyadic, e.g., mab + ned. Contraction: single-dot product, double-dot (scalar) product of dyads. If in a polyad we replace one of the open products by a scalar product of the two vectors, we obtain a polyad of order two less than that of the original polyad. This process is called contraction. For example, with the tetrad abed, contraction can be done in three ways with adjacent vectors to yield three dif- ferent dyadics: a-bed = (a+b)cd ab-cd = (b-c)ad (2.4.30) abe+d = (c-d)ab, where the dot product in parentheses is a scalar factor multiplying the dyad. The second of these contractions defines the single-dot product of the dyad ab by the dyad ed. Note that ab-ed % cd-ab (2.4.31) s0 that (unlike the dot product of two vectors) the single-dot product of two dyads is not commutative. The scalar product of two dyads (or donble-dot product) denoted ab: cd is defined as the scalar obtained by multiplying together the two scalar products a-cand bed, Note that the first vector of the first dyad multiplies the first vector of the second dyad, and the second vector of the first dyad multiplies the second vector of the second dyad. The scalar product is comnmtative. ab: ed = (are)(bed) = (c+a)(d-b) = ed : ab. (2.4.32)Sec, 2.4 Vectors and Tensors 39 Some authors denote the scalar product of Eq. (32) by ab ++ cd instead of ab: cd. Also, some authors define the scalar product differently, multiplying the two outside vectors together and the two inside vectors together. In this book, the double-dot notation with the two dots on the same line will denote the product obtained by multiplying the two outside vectors together and the two inside vectors together. ab ++ cd = (a-d)(bee), (2.4.33) Dyadics as second-order tensors. Since the single-dot product of a dyad ab by @ vector y yields another vector, as in Eq. (21), and since this operation is a linear vector function, a dyad operates as a second-order tensor. Is the converse true? Can every second-order tensor be represented as a dyad? The answer is no, but every second-order tensor can be represented as a dyadic, a linear combination of the dyads formed with the base vectors. Thus, in rec- tangular Cartesians, T= Tabb + Tabs + Tits + Trbh + Tabl + Tabls + Tybt) + Tatsis + Trisis or T=T,,1,4,. (2.4.34) ‘The algebra of dyads and polyads, as outlined in Eqs. (29) through (33) can then be shown to imply the results for tensors given earlier, when the tensors are written in the form of Eq. (34), Moreover, if the change of basis Eqs. (1) are substituted into Eq. (34), and the tensor is required to have the same form in the new coordinate system, i.e., Trl, =T = Tall, (2.4.35) it can be shown that this implies that the tensor component transformation formulas of Eq. (I!) must hold between the coefficients 7,, and 7,,. Thus the coefficients of the dyadic representation of Eq. (34) ate simply the rectan- gular Cartesian components of the tensor. Section 1.3 considers dyadic representation with an arbitrary basis b, instead of the orthonormat basis },. The dyadic representation is especiatly convenient for introducing derivatives of tensor fields, In Cartesian coordinates, since i, is independent of position, 2T/2x, = (3T;,/2x,)i,1,, but in curvilinear coor- dinates the unit base vector orientations vary with position. Thus in curvilinear coordinates with unit base vectors @, it is necessary to differentiate each term of the dyadic representation T=7"8,8, (2.4.36)40 Vectors and Tensors Chap. 2 as a product of three functions. For orthogonal curvilinear coordinates this is discussed in App. II, which avoids the general tensor complications of App. I. (It is not necessary to read either appendix before proceeding with the text.) Part 4. Review of Elementary Matrix Concepts Definition of a matrix. A matrix is a rectangular array of elements. We will consider here only matrices whose elements are real numbers. Ay Aw Aw A= (2.4.37, Fe Ay a] ( ; Ay, is the element in row ¢ and column j. (We will sometimes denote the element Aj.) Two matrices are equal only if their elements are equal term by term. Un- like a determinant, a matrix does not have a numerical value. Also it may not have the same number of rows and columns. An m x n matrix has m rows and n columns. The example above is a 2 x 3 matrix. Other common notations are || lor ( ) instead of [ ]. Addition of matrices, Matrices of the same order add term by term. Aut Bu Aw+ Be A+ By A+B= 2.4.38) 7 [i + By Ant Ba An + | : ’ Multiplication by a numerical factor. Multiply every term. kAy kKAy kA kA= 2.4.39) ‘An kAn | ; y Multiplication of two matrices. Matrices must be conformable or else they cannot be multiplied. Matrices A and B are conformable and can be multiplied in the order AB if the first’ matrix has the same number of columns as the second matrix has rows. For example, iccaae : : ah and ap are conformable in the order AB, but not in the order BA. The product of the 2 x 3 matrix A by the 3 x 1 matrix Bisa 2 x 1 matrix C.Sec. 2.4 Vectors and Teasors 41 2-1 3° | an i ae [o + (14) + ree _ [il =c (4)(6) + (0\(4) + 208)! ~ Lad The element in the first row and first column of C is obtained as the inner product (like a vector scalar product) of the first row of A by the first col- umn of B, as indicated above. The element in the second row and first column of the product matrix C is similarly obtained from the inner product of the second row by the first column, Formulas for the elements of the product matrix. Consider the product AB, schematically shown as AB = [m x nj[n x s}=[m x sJ=C. The two matrices are conformable, since the first matrix has the same number of columns as the second matrix has rows, namely n. The element C,, in the ith row and jth column of the product is Cy = 4uBy (summed). (2.4.40) If s = m, it is possible to multiply in the opposite order, and BA = [Bu Ags) (summed). (2.4.41) In general BA + AB. Square matrices and linear transformations. A square matrix has the same number of rows as columns. The system of equations Mux + Mix + Misxs = i Max, + Maxi + Maxi =r (2.4.42) Myx, + Max, 4 Masxa = Ys may be abbreviated as Mx=y, (2.4.43) where M is the square matrix of coefficients and x and y represent the two column matrices xelml y= (2.4.44)42 Vectors and Tensors Chap. 2 Equations (42) and (43) are said to transform the column matrix x into the column matrix y, This isan example of a linear transformation. Solving (43) consists of finding the column matrix x transformed into a given column matrix y by the given square matrix M, The example illustrated is for a 3 x 3 system, but the ideas apply for any n x n system. A diagonal matrix is a square matrix with nonzero elements only on the main diagonal Mj,, Mz, Mss,.... If Mis diagonal, the system (42) may be solved immediately, one equation at a time. When M is not diagonal, we may try to diagonalize it by performing on it a succession of operations equivalent to the kind of operations performed in solving (42) by elimination. An identity matrix, denoted by I, is a diagonal matrix with each diagonal element equal to unity, The elements of the identity matrix are 8,,. An iden- tity matrix transforms a column vector into itself (the identity transformation). Ix = x, (2.4.45) ilarly for any row matrix x, i.c., any matrix with only one row, xt = x. (2.4.46) The inverse of a square matrix M (if it has an inverse) is denoted by M=* and defined by MM"'= M'M =I. (2.4.47) Multiplying (43) by M7? (from the left), we get M-'Mx = M-ty or Ik = M"y whence x=Mty, (2.4.48) Thus, finding the inverse of M is equivalent to solving the system of equations, which can be done provided the determinant is nonzero. A square matrix with a nonzero determinant is called nonsingular. Thus, every nonsingular square matrix has an inverse. A formula for the inverse is given in Eq. (51). The transpose M? of a square matrix M is the matrix obtained by inter- changing rows and columns. For example, My My Ms M'=|My Mi Mss}. Mia Mi MasSee. 2.4 Vectors and Tensors 43 A symmetric matrix is a square matrix with equal elements in the positions symmetrically placed with respect to the main diagonal, i.e, with M,,= My so that M” = M. A skew or antisymmetric matrix has My = —M,, so that M? = —M, which implies that the diagonal elements are all zero. For square matrices A and B A = (AT BY, (2.4.49) The proof of Eq. (49) is left as an exercise. Formula for the inyerse. For a nonsingular matrix M, if D=deM is the determinant of the matrix, the solution of Eq. (42) by Cramer's rule gives, for example, yi Ma Mis ie Wie Ma M, =4ln Ma Mal = (M" "| - aie Hag y [Ma ‘| Mee Mi Mss Mix Mul Ma Mas} or - ee + Dy, = Diy where D™® is the cofactor (signed minor subdeterminant) of the element Mun in the matrix M. Thus, the solution of Eqs. (42) is = a Wn (2.4.50a) or, in matrix form, (2.4,50b) Hence the inverse matrix M~! of Eq, (48) is given by a to T : pe Mt = [ D with element D (2.4.51) in row n and column m, where D™* is the cofactor of Mna. This method of finding the inverse is prohibitively complicated for a large matrix, For numer- ical methods of inverting a large matrix, see a text on numerical analysis.44 Vectors and Tensors Chap. 2 Similar matrices. If M and N are two n x 1 matrices, then N is said to be similar to M if there exists a nonsingular matrix A such that N= A"'MA (2.4.52) Then M is also similar to N, since M = ANA" =(A")'N(A'). Orthogonal matrix. A matrix @ is said to be orthogonal if QFQ= QO" =I whence Q" =". (2.4.53) Equations (3b) show that the matrix A of direction cosines for change of rec- tangular Cartesian axes is orthogonal. Hence the matrices 7 and T of rec- tangular Cartesian components of a tensor T are similar matrices under the class of orthogonal transformation matrices, according to Eqs. (12) and (52). Eigenvectors and elgenvalues of a matrix. The transformation y = Mx of Eqs. (42) and (43) may be interpreted as associating to each point P(x1, xs, Xs) another point Q()1, Ys, ys). Since the transformation is homogeneous, it as- sociates to any other point (rx,,rx2,rxs) on the line OP another point (ry ryn rys) on the line OQ, and we may consider the transformation to be @ transformation of the line OP into the line OQ. We may ask whether there are any lines not changed by this transformation, i.e., any lines transformed into themselves so that (x,,x;,%s) transforms into (Ax, xs, A%3) for some scalar X. Any line transformed into itself is called an eigenvector of the matrix. (Other names for this are proper vector, characteristic vector, or latent vector.) If there is such an eigenvector x, then y = Xx and Mx = dx = Alx or (M—A)x =0, (2.4.54) and the eigenvector is given by the solution of the set of homogeneous alge- braic equations (54). A nontrivial solution will exist if and only if the deter- minant vanishes: IM — a] =0. (2.4.55) When the 3 x 3 determinant is expanded, Eq. (55) is a polynomial equation, a cubic in three dimensions, whose roots 4,4, are called the eigenvalues (proper numbers, characteristic values, or latent roots) of the matrix. When »Sec. 2.4 Vectors and Tensors 45 is put equal to one of these cigenvalues, Eq. (54) can be solved for the x, which are determined only up to a scalar multiple. They are determined up toa factor of +1, if we require in addition that the x, be the components of a unit vector. The whole discussion can be generalized to n dimensions. A real symmetric matrix has only real eigenvalues. For, if there were 2 complex root & of the polynomial equation (54) (with real coefficients), then Mx = dx, whence MX = XX (where overbars denote complex conjugates). Multiplying the first of these equations by 37 and the second by x”, we have S°Mx=d8"x and xTMF = aks. (2.4.56) But 37x = x78 = x18) + %y-+ +++ + x48; and, since M is symmetric x" MZ = 3° Mx, Hence, subtracting the two Eqs. (56) we find 0. (A — )3 Since x is nontrivial #7x + 0, and it follows that N=d (2.4.57) so that X must be real. It is also known that when no two of the eigenvalues are equal, the eigenvectors are uniquely determined up to a scalar multiple of each vector, so that the three lines in space (one associated with each cigen- value) are uniquely determined. Furthermore, if any two eigenvalues are distinct, say 4, % A», then the two associated eigenvectors must be orthogonal. This will be demonstrated in Chap. 3 for the particular example where the eigenvectors are the principal axes of the stress tensor. If all three eigenvalues are equal, then every direction is an eigenvector; in this case the eigenvectors are not unique, but it is still possible to choose a set of three mutually orthog- onal eigenvectors. If two are equal, say Ax = Ay # Ay, then the eigenvector associated A, is unique (up to a scalar multiple) and any direction orthog- onal to it is an eigenvector, so that the other two are not uniquely determined. In any case, it is possible to choose a set of mutually orthogonal eigenvectors. If the three orthogonal eigenvectors are normalized to be unit vectors, then the square matrix L whose columns are the column matrices of the orthonormal eigenvectors will be an orthogonal matrix, such that D=L'ML (2.4.58) is a diagonal matrix whose elements are the eigenvalues. Thus every real sym- metric matrix M is similar to a diagonal matrix whose elements are the eigen- values, and the transformation matrix L required for the similarity relation is an orthogonal matrix whose columns are a set of mutually orthonormal eigenvectors of M.46 Vectors and Tensors Chap. 2 This completes our introduction to tensor algebra. Some additional tensor properties having to do with principal axes of tensors and the algebraic invariants of a tensor will be presented in Sec. 3.3 in connection with the development of the stress tensor. Although developed specifically for the stress tensor, these properties are also relevant to other tensors. Also presented in Sec, 3.3 will be the separation of a tensor into the sum of a spherical and a deviatoric tensor. And in Sec, 4.6 the concept of the inverse of a tensor and of positive, negative, and fractional powers of a tensor will be presented. The differential calculus of vectors and tensors, which will be needed in our study of strain and deformation, is considered in Sec. 2.5. EXERCISES 1, New right-handed coordinate axes are chosen with new origin at (4,1, —2) and with Fy = (2h - le + i:)/3 and Ty = (51 — ts)/W2- (a) Express [, in terms of the ip. (b) If t = 101, + 10h, — 20K, express { in terms of the new basis, {c) Express the old coordinates x1, x2, xs in terms of ¥1 51, 25. . Complete the following table of direction cosines for rotation of axes. If S2 = Sy =f, = 1000 1b, find the components of f in the barred system. g 3 x 0 i i: 1 0 . Verify that Eq. (2.4.66) gives the same results as Eq. (6a). Carry out the details of deriving the inverse transformation of the second Eq. (2.4.6a). Derive the first Eq, (2.4.72). If (ab)-v means a(b+¥), show that its rectangular Cartesian components are the same as those of T+¥, where T is the tensor with components aib,. . Write out the nine-term sum of the first Eq. (2.4.11) for Ts1, for Tas. AY ae 23 . Rotated Zm-axes are chosen making angles with the x,-axes as shown in the following table. g 5 z x 90° 45° 135°Sec. 2.4 Vectors and Tensors 47 (a) Verify that the new axes are a right-handed orthogonal system, and display the matrix A of direction cosines. (b) If tensor T has the components shown in the matrix below, referred to the raxes, compute the matrix of fm-components of T. 2-2 0 2 6 ccc: ). Derive the second Eq. (2.4.11). 10, Show that the first Eq. (2.4.12) is equivalent to the first Eq. (11). 11, Show that the two scalar products T:U and T ++ U, defined in Eqs. (2.4.18), are equal if U is eymmetric, 12, Show that T: W = 0, if T is symmetric and W is skew. 13, Show by use of Eqs. (2.4.6a) that the test of Eq. (26) implies the first transforma- tion of Eq. (11). 14, Show by use of Eqs. (2.4.6a) that the test of Eq. (28) implies the first Eq. (11). 15, Show that if Eq. (2.4.35) holds for arbitrary rotations of axes, then the Tp, are elated to the Trs by Eq. (11). 16. Find the coefficients af for the rotation of axes, if the new %,-axes are obtained by a counterclockwise rotation through angle @ about the old xs-axis. 17. Find the rotation matrix A for the following coordinate changes: (a) rotate 180° about x-axis, (b) rotate 45° about x,-axis, (c) rotate 90° about £1-axis and then. rotate 90° about the position assumed by the x2-axis after the first rotation. 18. Determine the product matrices AB and BA, if A and B are the matrices shown. ' 000 A=|0 00 B=|0 00 10 0, 19. Multiply 8, if 12 fo 1 45[ | B=|3 0 24-1 2-1 Can you multiply as BA? 20, Multiply AB and BC, ma -12 6 4 A=([} —-b 4) B=/-12. 90 c=|-3|. 6 0 6. +48 Vectors and Tensors Chap. 2 21, Verify that, if 321 ot A=|4 65 then At=| ys —$ 48: oo 4 oie te 22, Show that if u,v, w are orthonormal eigenvectors of matrix M, corresponding to eigenvalues 21, 21, As, respectively, and L is the square matrix whose columns are u,v, w, then L'(ML) is a diagonal matrix, as asserted in Eq. (2.4.58). 23, and 24, Determine the eigenvalues and eigenvectors of the matrices shown below, and verify that the orthogonal matrix L of Eq. (2. 4. 58) diagonalizes the given matrix in each case. uw4o0 70 -2 23. 430 24, 05 0 oe ee 25. Verify Eq. (2.4.49). 26. Verify the first two Eqs. (2.4.25b) for A ++ Band A:B. 27. Verify the last two lines of Eqs. (2. 4. 25b). 2.5 Vector and tensor calculus; differentiation; gradient, divergence, and curlt The preceding sections have dealt with the algebra of vectors at a point. The physical entities defined as vectors, in general, vary from point to point and with time. The vector quantity is then a vector field or vector function, Study of the rate of change of such a vector function leads naturally to the study of derivatives with respect to space coordinates. A brief summary of vector dif- ferential calculus is given in this section; for more details and proofs see a textbook on vector analysis. Some integral theorems will be given in Sec. 5.1, since these are useful in deriving the general principles of continuum mechanics introduced there, We introduce the total time derivative of a vector function of ¢ by an example, the differentiation of the position vector of a moying particle. If p(e) is the position vector at time # of a particle (material point), then its velocity y is given by dp/dt, which is defined in the same way as the de- tivative of a scalar function, namely This section may be postponed until after Chap. 3, since it will not be needed until the development of the concepts of strain and deformation in Chap. 4.See. 2.5 Vectors and Tensors 49 dp — tim AP (2.5.1) It ato If the particle moves on a smooth path, as in Fig. 2.8, then Wal as (2.8.2) Fig, 2.8 Differentiation of position vector p. and in the limit, assuming that |Ap|/As approaches unity and that the direc- tion of the secant PQ approaches that of the tangent to the path at P dp ds 4 (2.5.3) where @, is the unit vector tangent to the path, For the evaluation of the derivative, if we express the Cartesian coordinates of the particle in a “fixed” system of reference as functions of t, p= x()i + (Oj + 20k, (2.5.4) we obtain dp dt dx, dy) dzy_ y Git alta = Simba (2.5.5a) since i, j,k do not change with time. Inversely the time integral is50 Vectors and Tensors Chap. 2 ‘ ‘ 4 p—pe=fivde =if} tadt + If) oat + kf dt. (2.5.5b) The time derivative of any other vector a is evaluated in the same way, dia _ dag 4 day y 4 dos y a dey 5 Mie 4 Mek = daly (25.6) and a similar interpretation to that of Fig. 2.8 can be made if we plot succes- sive values of a in a space with coordinates (a,, ay, 4,). It is easy to show that the derivatives of a product of a scalar by a vector, of the dot product of two vectors, of the cross product of two vectors, and of the tensor products all obey the usual calculus rule for the derivative of a product. Proofs will not be given here; the essential ingredients for the proofs are the distributive character of the products considered and certain limit theorems—such as that the limit of a product is the product of the limits when both limits exist. Under the assumption that the vector functions considered have continuous derivatives, the whole algebra of derivatives of scalar functions carries over to the derivatives of vector functions. In applying the product tule to the cross product of two vectors, however, it it necessary to keep the two factors always in the same order, since the cross product is not commuta- tive, Thus dx yaux ts Bx y, 25.7) and it would be incorrect to write the last term as v x(du/dt) in analogy with the way the scalar formula is often written. If A and B in Fig, 2.9 are two points on a rigid body with instantaneous angular velocity w, the relative velocity va), of B relative to A is given entirely by the rotation, Thus, the magnitude Waal = 42 cB] = 4 legs sin 6, (2.5.8) and the direction of vg. is tangent to the circle of radius CB in a plane perpendicular to the axis of instantaneous relative rotation drawn through A. Hence Vaya =O X Faye. (2.59) If any other vector of constant magnitude, say a, is fixed to a rigid body, then it follows also that a =axa, (2.5.10)Sec. 2.5 Vectors and Tensors SL Fig. 2.9 Relative velocity vaya for two points on a rigid body. @ result which can be used to express vector derivatives in curvilinear coordi- nates, or with respect to moving axes. In cylindrical coordinates (r, 8, 2) (Fig. 2.10), where x=reos6, y=rsin8, the position vector p is given by pare, + 28, (2.5.11) where @, and @, are unit vectors in the direction of increasing r and z, respec- tively. Note that @, = k is constant, but that in a general motion @, changes with time as does the third unit vector, @, in the transverse direction of in- creasing 6. If we differentiate Eq. (11), we obtain v= 2= ta Bey He, (2.5.12) The triad of unit vectors @,, &,, @, may be considered as a rigid body moving with the particle; since @, always intersects the z-axis at a right angle, the52 Vectors and Tensors Chap. 2 Fig. 2.10 Vector p in cylindrical coordinates r, 6, 2. angular velocity @ of this little rigid body is (d@/dt)@,, and the rates of change of the unit vectors are given by Eq. (10) as a, 4 oe FAUX eHGuxba dt a TUX eH Gaxea= (2.5.13) since =@,x@, while @ x & = —@,. Hence _dr a a = Fe tr Pet Fe, (2.5.14) This last equation may be differentiated again to yield the acceleration, using once more Eqs. (13) for the derivatives of the unit vectors. For further ex- amples of this kind see a textbook on vector dynamics. We turn our attention now to differentiation with respect to space coordinates.Sec. 2.5 Vectors and Tensors 53 Scalar point function. A scalar point function has a single component in any coordinate system, which has the same numerical value in all coordinate systems. Thus FG, 5,2) = Fy), (2.5.15) where, by definition, FR, I, DS FUR, Fs Ds WEI, Ds 28 FD). (2.5.16) It is common practice to write Eq. (15) as FUR, Fs 2) = Fs ys 2), but it must be remembered that F docs not represent the same function of the barred coordinates as it docs of the unbarred coordinates. For example, if F(x, y, 2) = 25(x* — y*) + 2°, and if the transformation is x = (4% — 39)/5, y = (38 + 49/5, and z= Z, then F(x, J, 2) = 7(8% — f) — 4B 8p 4+ 27, an apparently quite different function from F(x, y, z) although it has the same value at every point. The gradient of a scalar F is defined to be the vector, denoted VF or grad F, such that for any unit vector 0, the directional derivative dF/ds in the direction of @ is given by the scalar product dF _ op, F = vr, (2.5.17) where = a=. Note that this definition makes no reference to any coordinate system. The gradient is thus a rector invariant independent of the coordinate system. To find its components in any rectangular Cartesian coordinate system we note that (2.5.18) while, by the chain rule of calculus, ae Olax (2.5.19) ds ~ Ox, ds *54 Vectors and Tensors Chap. 2 If (VF), denotes the x,-component of VF, then Eq. (17) yields F = (WF), Comparing this with Eq. (19), we see that [ery - x] tr =o for arbitrary choice of the dx,/ds (arbitrary direction 0). Hence (WF) = x or 5. grad F = VF = zh a In indiciat notation partial derivatives are frequently denoted by a subscript preceded by a comma: 8F/@x, = Fy. Another notation places the operator 2, ahead of the function. Thus OFaF.= x= and VF =hOF =F, (25.206) A geometric interpretation of the gradient follows from Eq. (17) if we take 0 = tas a tangent vector to the level surface F(x, y, 2) = constant. Then dF/ds = 0, and Eq. (17) implies that (VF)-t = 0. (2.5.21) Since according to Eq. (21), the gradient vector is perpendicular to every tangent vector at the point, the gradient vector must be normal to the level surface through the point. Moreover the rate of change of the function F with respect to distance in the direction fi of increasing F normal to the surface F = constant, namely, (VF)-+A, is equal to |VF], since VF is parallel to the normal and since (VF)-f is positive because f was taken as the direction of increasing F. The unit normal vector fi to the level surface, taken in the direc- tion of increasing F, is therefore given by dF f= ea (2.5.22) Other important applications of the gradient in mechanics relate the force vector to the gradient of the potential of a conservative force field and the velocity vector to the gradient of a velocity potential in irrotational fluid flow.Sec.2.5 Vectors and Tensors 55 The symbolic vector operator del, denoted by V, tA a i v= a tla tka =in = bm (2.5.23) may be used formally to obtain Eq. (20) fee (pe Oyo, OF aF end FHV = (IP +p +kg)F= i ties eeea(25 24) The operator operates on the quantity written to the right of it. When the quantity to the right is a scalar function F, the result is grad F. When the quantity to the right is a vector function, it is necessary to specify how the vector operator operates on it. Two common applications are V+y and vx y, the first yielding a scalar, the second a vector. These will be intro- duced here briefly and formally; applications and physical interpretations are given later. Vv without dot or cross, is a tensor; see Eq. (31). Divergence V-v. When the vector operator V operates in a manner anal- ogous to scalar multiplication, the result is a scalar point function, divy, called the divergence of the vector field y. In rectangular Cartesians, divy = Vor = (2 es ig a 2)-Gv. + Ju, + ke.) or (2.5.25a) divy = Ses Bea y Be = Mey ey. A physical interpretation of div v is given in Sec, 5.2, where it is shown that if v is the velocity field in a flow of a continuous medium, then divy = —(1/p)(dp/dt), where p is density, Thus the divergence of the velocity vector measures the rate of flow of material away from the particle and is equal to the unit rate of decrease of density p in the neighborhood of the particle. The Laplacian Operator V* = V+V gives a scalar point function when it operates on a twice-differentiable scalar field (oes Or ror VFS VIVE = Fa = oat ort oF (2.5.25b) in rectangular Cartesian coordinates. For orthogonal curvilinear coordinates, see Eq. (1.2.14). For a general-tensor component form, see Eq. (1.6.28). For the Laplacian V*y of a vector v, see Eq. (2.5.35). Curl V x vy. When the vector operator V operates in a manner analopous to vector multiplication, the result is a vector, curl v, called the curl «' the vector field v (sometimes called the rotation and denoted Rot v).