COLLEGE OF PETROLEUM & MINERALS LIBRARY Marvin J. Forray Professor of Mathematics C. W. Post College of Long Island University Variational Caleulus in Science and Engineering McGRAW-HILL BOOK COMPANY ) New York San Francisco Toronto London Sydneyyy. MINERALS, GE OF PETROLEY! ms LIBRARY my we: On VARIATIONAL CALCULUS IN SCIENCE AND ENGINEERING Copyright © 1968 by McGraw-Hill, Inc. All Rights Reserved. Printed in the United States of America. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopy- ing, recording, or otherwise, without the prior written permission of the publisher. Library of Congress Catalog Card Number 68-20718 21584 1234567890 MAMM 7543210698Preface The calculus of variations is assuming an increasingly important role in the fields of analysis, physics, and engineering. While the subject originated as a study of certain isolated maximum and minimum problems not treatable by the techniques of elementary calculus, it is at present a powerful method for the solution of problems in dynamics and statics of rigid bodies, general elasticity (including thermal effects), the theory of plates and shells, vibrations, optics, quantum mechanics, optimiza- tion of orbits and controls, ete. Students of mathematics, engineering, and physics receive in many instances only fragmentary instruction in the method and application of the variational calculus. In his quest for additional understanding of this subject and to enhance his ability to read current articles the student usually turns to mathematics books on the variational calculus. While there are some excellent texts in this area, they are difficult to absorb, particularly by the average nonmathematician. This book is written in the hope that it will make the subject more palatable without sacrifice of rigor. Methods will be developed and extensively illustrated with worked examples. Over twenty years of teaching experience on the high school, college, and graduate levels has taught the author that the best, way to make a subject “stick” is to illustrate the method profusely, because—unfortunately—abstract concepts are quickly forgotten or misapplied. 33980vi Preface The first three sections of Chapter 1 are relatively elementary and should serve as a review of the elements of maximum and minimum for functions of one and more than one variable. A number of illustrations should help the reader to renew his acquaintance with these basic calculus concepts, which are significant in the variational calculus also. Sections 1.4 and 1.5 serve as an introduction to extremum problems with con- straints using the multiplier method of Lagrange. This is done at a leisurely pace because it is anticipated that part or perhaps all of this may be new to some readers. Chapter 2 begins by introducing the concept of the functional and in particular what is meant by the first variation of a functional. The variational calculus is then treated formalistically to show the reader the potency of this procedure. In particular, the problems of the cantilever beam and circular plate are exhibited. Many of the concepts which are going to be justified in detail, such as the fundamental lemma of the variational calculus and the commutativity of the operators d/dz and 6, are utilized. The main objective is for us to see the forest before we start clearing the trees. This chapter closes with a few historical remarks and a derivation of some significant classical examples including the isoperimetric, brachistochrone, shortest distance, and minimum sur- face of revolution problems. Chapter 3 is initially devoted to the basic problem of formulating the Euler differential equation for a wide variety of problems. In con- junction with this, the basic lemma of the variational calculus is estab- lished for both one- and two-dimensional problems. Both mathematical and physical examples are given to illustrate the formulation of the Euler equations. The solutions of Euler’s differential equations for various cases are treated in detail. Natiiral boundary conditions followed by a systematic treatment of the variation is given next. Applications are presented in both of these areas. The remainder of the chapter treats the Lagrange multiplier method for functionals. In Chapter 4 we have the application of the variational calculus to discrete mechanics. In particular, starting with Newton’s law of motion, we deduce Hamilton’s principle for a single particle; and for a conserva- tive system, the Lagrangian function is introduced. This is followed by a generalization to n variables, and Lagrange’s equations are deduced. A number of examples and a detailed analysis of free vibrations of a con- servative system about a position of equilibrium are given. This is illustrated in particular by the vibration of a light string with two con- centrated masses attached. Chapter 5 treats the elements of the theory of elasticity. The concepts of displacement, strain, and stress are presented in some detail including equations of equilibrium and compatibility, Also, the transformationsof components of strain and stress tensors about a point are deduced and formulas for principal strains and stresses are obtained. Both Hooke’s law and its generalization are developed and, furthermore, the significance of the elastic constants for isotropic materials is demonstrated. Then the important concept of strain energy is systematically deduced and a proof is given that, in general, the strain energy density is a positive definite form of the strain components. Chapter 6 is, in a sense, a continuation of Chapter 5, because it starts with the proof of the fact that the potential energy of a static elastic continuum must be stationary. This result is then applied to three significant problems, namely, the static response of an elastic string, a beam column, and the deduction of the three-dimensional displacement equations of elasticity. From the concept of stationary potential energy we deduce Hamilton’s principle for a dynamic elastic continuum and illustrate it with the flexural vibrations of elastic strings and beams. At this point, it is expedient to introduce indicial notation and to apply it to a general form of Castigliano’s least-work principle. This theorem, or its corollaries, yields the true stress field from the many states of stress which satisfy the differential equations of equilibrium and traction boundary conditions. They result in the additional requirement that the com- patibility equations must be satisfied by the actual stress field and, furthermore, the complementary energy must be stationary. In the next section a variational theorem due to E. Reissner simultaneously yields relationships between the stress and displacement components, the equilibrium equations, and the boundary conditions for small (or infini- tesimal) displacements. This is followed by a generalization of a second theorem by Castigliano which directly gives the displacements and rotations at arbitrary points of an elastic body. The Maxwell law of reciprocity is deduced for linear elastic bodies and exemplified for a cantilever beam. Chapters 7 and 8 treat some of the direct methods of the calculus of variations. In Chapter 7, the Rayleigh-Ritz method is first outlined and then exemplified by the solution of a particular differential equation and the bending of a clamped rectangular plate. Then the method is applied to simple vibration and stability problems. The torsion of circular cyl ndrical shafts and more general prismatic bodies follow in some detail. This problem is then recast in variational form and exemplified by the torsion of elliptic and rectangular cylinders. In the last chapter, the methods of Galerkin, Kantorovich, and Euler are developed. After the Galerkin method is introduced, its equivalence with the Rayleigh-Ritz method is established for self-adjoint differential equations of order two. Further examples of this equivalence are given for the functionals associated with the bending of plates and membranes.viii Preface This is followed by theCingeniows method of Kantorovich which serves as a generalization of the Rayleigh-Ritz procedure. Also, the method is illustrated for the bending of a clamped rectangular plate. The next section is devoted to the Buler method of finite differences. A detailed application is given to the determination of an extremum associated with a particular functional. In the last section, examples of problems without solutions are demonstrated. Many illustrations of “plausible” extrema problems for which there are no solutions may be given, thereby revealing the significance of questions of existence (and uniqueness). The Appendix is devoted to a more extensive treatment of the summa- tion convention and Cartesian tensors than is presented in the body of the book. In particular, the reader should find this appendix most helpful in his study of Sections 6.3 to 6.6 of Chapter 6. The author is indebted to Professor Robert V. Iosue for critically reading the manuscript and making many useful suggestions for improve- ment of the text. Also, it is a pleasure to acknowledge the very skillful typing assistance of Mrs. Rosalie Ciaravino, Mrs. Doris Vandereedt, and Mrs. Stasia Polster in different phases of the manuscript development. To his wife Muriel, the author is indebted for her deep affection, direct help and persevering efforts in encouraging the completion of this task. Therefore, this book is affectionately dedicated to her. Marvin J. ForayContents Preface ov ONE Maximaand Minima 1 1 Introduction 1 2 Necessary and Sufficient Conditions 2 3 Illustrations 6 4 Mazimum and Minimum Problems with Constraints 10 5 Lagrange Multiplier Method—General Case 14 TWO = Introductory Problems of the Variational Caleulus 18 1 Concept of Functionals 18 2 Stationary Values for Functionals 20 3 Classica! Problems of the Variational Calculus 27 THREE Euler-Lagrange Development with Applications 31 1 The Basic Problem—Euler's Equation 32 2 Euler Equation for More General Cases 35 3. Solution of Euler's Differential Equations 44enrane Contents Natural Boundary Conditions 49 The Variational Notation 58 Lagrange Multipliers for Funetionals 62 Isoperimetric Problems 65 Isoperimetric Problems—Generalization 70 FOUR Hamilton’s Principle and Lagrange’s ansere Equations 75 Hamilton's Principle for a Single Particle 76 Degrees of Freedom—Generalized Coordinates 80 Hamilton's Principle for a System of n Parlicles 81 Lagrange's Equations with Applications 83 Lagrange's Equations for a General System 87 Normal Oscillations 90 FIVE Deformable Bodies—Theory of Elasticity 95 BE CerAaneeNnH SIX NAnR ENE aunewe VEN Rayleigh-i Strain 96 Transformation of Strain Components—Principal Azes 100 Theory of Small Displacements 104 Stress 107 Relation among Traction Vectors at a Point 110 Equilibrium Equations 112 Principal Stress at a Point 114 Hooke's Law 115 Generalized Hooke's Law 117 Significance of Elastic Constants for Isotropic Materials 119 Strain Energy 123 Energy Principles, Methods, and Applications 126 Theorem of Minimum Potential Energy 126 Hamilton's Principle for an Elastic Continuum 134 Indicial Notation—Complementary Energy Density 137 General Form of Castigliano’s Least-work Theorem 144 Reissner's Variational Theorem of Three-dimensional Elasticity 147 A General Form of Castigliano’s Theorem on Deflections 149 Principle of Superposition and Mazwell’s Reciprocity Law 154 itz Method 157 An Outline of the Rayleigh-Ritz Method 158 Rayleigh-Ritz Method Applied to Vibration and Stability Problems 164 Torsion of Circular Cylindrical Shafts 174 Torsion of Prismatic Bodies 174 Transformation to a Dirichlet Problem—The Stress Function 177 Variational Form of Saint-Venant’s Problem, 182EIGHT Methods of Galerkin, Kantorovich, and Euler 189 Galerkin Method 189 Equivalence of Rayleigh-Ritz and Galerkin Methods 193 Method of Kantorovich 199 Euler Method of Finite Differences 204 Final Remarks—Problems without Solutions 204 asewe APPENDIX Summation Convention and Cartesian Tensors 207 1 Summation Convention 207 2 Cartesian Tensors 240 Bibliography 213 Author Index 245 Subject Index 217ONE Maxima and Minima 1 Introduction The calculus of variations is concerned primarily with the determina- tion of maxima and minima of certain expressions involving unknown functions. Certain techniques and concepts of the differential calculus are applicable to the variational calculus. ‘These will be reviewed briefly. One of the basic theorems of the calculus (due to Weierstrass) is that if a function f(z) is defined and continuous at each point of a finite closed interval a <2 0] at that point. If f’’(xo) = 0 and successive differentiation shows that the first derivative which is not zero at z = 2» is odd, no maximum or mini- mum occurs. If, on the other hand, it is even-ordered, a maximum occurs when the value is negative, and a minimum when positive. The problem of determining stationary values for a function of more than one variable is more involved. The theorem of Weierstrass still applies, namely, every function which is continuous in a closed domain D of the variables possesses a largest and a smallest value in the interior or on the boundary of the region. Suppose u = f(x1,22, . . . ,2a) is differ- entiable in a domain D; then a necessary condition for an extremum at a point P in the interior is id ee dagen “379 (2.1) at P; that is, df =0.! The sufficiency conditions are more compli- cated. In the case of a function of two variables, the function f(x,y) yields a maximum at a point P if, in addition to f, =f, = 0 atP, See <0 and fecfy — fay? > 0. It results in a minimum when f.:>0 and fzzfy —fxa? > 0. If, on the other hand, fesfyy — fey? <0, the stationary value is neither a maximum nor a minimum. The case Sexy — fy? = 0 remains undecided; that is, if it occurs, an alternate procedure must be employed. The above conditions have a simple geometrical interpretation. In the case of a function of two variables, the necessary conditions for a stationary value, fz = f, = 0 at (co,yo), imply that the tangent plane to the surface z = f(x,y) at this point is horizontal (i.e., parallel to the zy plane). If the point is an extremum (maximum or minimum), then, 1 The quantity df = > 42, dz; by definition. Note that, for convenience, we will a often use subscripts to denote partial differentiation with respect to that variable. Thus fz = af/ar and fey = a*f/az dy, ete.in its neighborhood, the tangent plane does not intersect the surface. In the case of a saddle point (no maximum or minimum) the plane cuts the surface in a curve which has several branches at the point. ‘The sufficient condition for a maximum or minimum in the case of functions of more than one variable is expressed most conveniently in terms of quadratic form-matrix notation, which we will treat next. DEFINITION: A quadratic form is a homogeneous polynomial of the second degree in several variables. A quadratic form! may be represented in the following form F(eit2, . . 2m) = Guta + Greet, + + + Gintstn + autor, + Arete? + + A Aant2tn t+ Gnitnti + Onatnte + * + Gnntn® where without loss of generality, ai; = aj The matrix ay a2 in an a aay Ae ee (2.2) On Mascaras Onn is called the matrix of the quadratic form. Because of the manner in which it was composed, A is a symmetric matrix; that is, a, = a ‘Thus, every quadratic form is associated in a natural way with a unique sym- metric matrix and, conversely, every symmetric matrix may be associated with a certain quadratic form. A quadratic form may be written in an abbreviated matrix notation as follows: F(@ija, . «+ sn) = B(Gut1 + Grate + + + + + Ginn) + we(derti + Geeta + + + + + Genta) ee es + &n(Qniti + dnote + + + Gnntn) = [tr 2 ++ ta) (Ques + ait, + + dante Gaye, + Arte + + + + + Aonte Gnr%1 + On2t2 + °° * + Anat =f me +++ tel fon ai ais --- Gu] [a @o1 G22 Ges °° * Gon x ni Gn2 Gna ~ °° nn. Tn, F(tyta, .. . te) = 2! Ax (2.3) 1 The letters au, aiz, . . . » nn are taken to represent arbitrary real numbers.4 One where 2’ = [21 22 * ~*~ a] is a row matrix which is the transpose of the column matrix Zn A real quadratic form is called positive definite if, for real values of the variables, its value is always positive, with the obvious exception of y= t= =a, =0. Example 1: 22+ m?+ 2+ -- + +? is a positive definite form. The term “positive definite” is also extended to symmetric matrices. Areal symmetric matriz [A] = [ay] is called positive definite if the quadratic form 2 Y aerers 1 it F(a1,t2,03 - - » stn) = ime is positive definite. Thus, for example, the unit matrix is positive definite, since the quadratic form corresponding to it is positive definite. The following theorem (due to Sylvester) is significant for determining necessary and sufficient conditions for maxima and minima. THEOREM: A quadratic form )' )) airs; is positive definite if and Aah only if all the determinants (principal first minors) ay a2 Ar = |an| 42 = aos An = are positive. Similarly, a quadratic form F is negative definite if —F is positive definite, while a matrix [A] is negative definite if [— A] is positive definite. If a form is neither positive or negative definite, then it is said to be indefinite. Example 2: The quadratic form P= ay? + 2x, + 5x3? — Qayte + 4aits — deersis easily shown to be positive definite. For this example, eae eee 2 faJ =|-1 2 -2 2 -2 5 so that 4: = [1] = 1 aa-|_} 1 A= |-1 2 Since all the principal first minors are positive, it follows that F is positive definite. In fact, the quadratic form can be written equivalently as = (ey — a2 + 2s)? + a? + as? which, being the sum of squares, is zero if and only if tm —t%+27=0 w%=0 and 2s=0 which implies that 21 = 22 = x3 = 0. On the other hand, the quadratic form P = 2x,’ + 2ae? + 2xs? — 4a, — Avors + ear 1 -2 1| (x1 [z. we wal] -2 2 —2| 4m 1 -2 2) lea, is readily shown to be indefinite. The principal minors are given by A= [1] =1 1 -2 oe et 4s=|-2 2 -2 1-2 2 Two particular values of F are F(1,0,0) = 1 and F(1,1,1) = —1, thereby verifying that the form is actually indefinite. Let f(e1,2n3, . . . ;2n) be three times continuously differentiable in the neighborhood of a stationary point P: 2 = 2° @ = 1, 2,.... , That is, f2, = 7°, = =f’, =0. Consider the second total differ-6 One ential of f at the point 2°; then df? = )' ) f0.,, dx dz; This is a afr quadratic form in the variables dz, . . . ,dtn. The quantity d’f? may be (1) Positive definite (2) Negative definite (3) Indefinite It can be shown by Taylor’s theorem that, corresponding to (1), f has a minimum, and to (2), a maximum, whereas condition (3) yields neither a maximum nor a minimum provided that Fin. Seer °° * Son D=lfie fen *** Sem| 9 3. Illustrations Example 1: Find the absolute maximum and minimum for f@@)=x'-2 -2<2<2 Solution: f'(x) = 4° — 5x4 For critical values, f’(x) = 0 must necessarily hold. Therefore, 2(4 — 52) =0 z = 0 (3 times) and cat are the critical values. S'(@) = 12a? — 202% F'(8) = 12(8)? — 20(8)* = ($)2(12 — 16) < 0 Thus x = yields a local or relative maximum, f($) = 44/5%. Also, J) = 0, so that higher derivatives must be calculated and evaluated atz = 0, "(@) = Az — 602% 70) = 0 f"(@) = 24-1202 f”"(0) = 24 Since the first derivative which does not vanish is even-ordered and its value is positive, we have a relative minimum at z = 0, f(0) = 0. Aref($) = 44/5* and f(0) = 0 the absolute maximum and minimum for f(x) in —2 0. Comparison of (@j) and (i) results in u |" = «| 6 Example 3: Find the dimensions of an open rectangular box with a capacity of 4 cu ft which shall have the smallest possible surface area. Solution: Let x and y be the horizontal edges and z the depth, all positive from physical considerations. The surface area is A = ay + nz + 2ye The volume is ayz = 4, so that z = 4/zy. Therefore, . ee Aamtits For critical values 8 ~ a Ar=8 One so that r=y=2 and A=12 ‘The corresponding depth, z = 1, is half the edge. ‘The second derivatives at the point (2,2) are 16 16 An => Ay = qt? daa so that at (2,2) AgAy — Ax’ = Since Az: > 0, we have arelative minimum. This may be shown to be an absolute minimum. Example 4: Test for relative maxima and minima: flaye) = 22 + y® + 382? — ay + 2a2 + ye Solution: So = 2 —y +2 fy=%yr-aote fe = 62+ 2 +y For critical values, f2 = fy = fe = 0, which yields 2 = y = 2 = 0. For the determination of whether we have a maximum or minimum, the principal minor test will be used. fos =2>0° fox fv else eee _ tee ‘| “| | ~870 Jez fav Sue 22 Juz Sv a7 oe ed 0 fee fav oie Thus I(x,yz) = F,0,0) = 0 in the neighborhood of the point; that is, (0,0,0) yields a relative minimum." Example 5: Snell’s Law of Refraction. A very significant applica- tion of maxima and minima is to the derivation of Snell’s law of refraction. It is based upon a minimum principle of Fermat which states that the time 0, y > 0). The area of the rectangle is f(x,y) = zy, and f is to be maximized subject to the restriction that the perimeter is L. Thus the constraintis g(a,y) = 2x + 2y-L=0 at (a,b) (4.1) (4.2)12 One Substitution of the expressions for f(z,y) and g(z,y) into (4.2) yields y+2r=0 @ z+2=0 Gi) Also, Qe +24 —L=0 Gi) ‘The simultaneous solution of (i) to (lii) results in zea yad=% md y=-2 Therefore the square (a = b) makes the area stationary for a given perim- eter. Furthermore, it is easily shown that the square actually maximizes the area, It is also of interest that this example has a “dual” result; namely, that of all rectangles of given area the square has the least perimeter. ‘The above argument may fail to be valid if the curve g(x,y) = 0 has a singular point (say, a cusp) at ¢ = a, y = 6. By a singular point we mean a point for which both partial derivatives are zero. For example, the origin (0,0) is a singular point of the curve y? — 2* = 0 which actually has a cusp there. In any event, our intuitive investigation suggests the following rule (which will be established rigorously next): In order that a function f(2,y) attain an extremum at =a, y =b subject to the restriction g(x,y) = 0 where gs, = gy],, = 0 does not occur, it is necessary that a constant of proportionality \ exists such that Fe(a,b) + dgo(a,b) = 0 Fu(a,b) + rg,(a,b) = 0 are satisfied together with the equation g(x,y) = 0. Theabove rule is known as Lagrange’s method of undetermined multipliers, and the factor is known as Lagrange’s multiplier. As expected, the analytic proof of the validity of the multiplier rule is established by reducing the constraint problem to the known problem with “free” extreme values. Let it be assumed that at the extreme point (ab) at least one of the partial derivatives, say g,(a,b), does not equal zero. Then, from the implicit function theorem of the calculus {assuming that 9(z,y) has continuous first partial derivatives] it is known that there exists a neighborhood of (a,b) for which the equation g =0 determines y uniquely as a continuously differentiable function of z, say y = ¥(z). Substitution of y = ¥(z) into f(2,y) results in the function f(x,¥(z)), which must have a free, i.e., unconstrained, stationary value at the point x = a.Therefore, '@) =f2+ iW @) =0 (4.3a) must hold atx =a, Also, from g(x,y) = g(z,¥(2)) = 0 we have go: + gW'(z) = 0 (4.36) Multiplication of (4.3b) by \ = —f,/g, and addition of this to (4.3a) yields fet dge + (fy + Agu)/(z) = 0 (4.4) But by our choice of 2, fy t dy =0 (4.5) and this reduces (4.4) to foto: =0 (4.6) Equations (4.5) and (4.6) are precisely those of the method of undeter- mined multipliers, and therefore the procedure is established. The multiplier method can be summarized by the rule: V Form F(z,y,d) = f(v,y) + dg(@,y) and treat the variables z, y, and ) as if they are independent of each other. Then make F stationary by setting the partial derivatives with respect to each of the variables equal to zero, i.e, oF ag Oth tM fit This is identical with (4.5), (4.6), and the constraint condition. Finally, before generalizing our results, we shall indicate the application of differentials to this problem. In order that the function f(z,y) may have an extreme value at the point z = a, y = b subject to the subsidiary equation g(x,y) = 0, it is necessary that the differential df shall vanish at that point; ie, ee a7 Also from g(z,y) = 0, Y/Y 9x(ajb) dz + 9,(a,b) dy = 0 48) Multiplication of (4.8) by \ and addition to (4.7) yields [f-(a,b) + dge(a,b)] dz + [f,(a,b) + dg,(a,0)] dy = 014 One Choose ) so that Sy(a,b) + g,(ab) = 0 assuming g,(a,b) + 0 Then since dz # 0 in general, Sela,b) + dgz(a,b) = 0 and Eqs. (4.5) and (4.6) have been deduced again. 5 Lagrange Multiplier Method—General Case More generally, consider the problem of extremizing a function of n variables F(wiyte, . . - sn) (5.1) subject to constraints gr(ti,22, .. . tn) = 0 9r(ti,G2, . . . stn) (5.2) of the n quai terms of the remaining n — k variables. Substitu- tion into (5.1) yields a function of n — k independent variables. This, of course, may then be treated as an ordinary maximum and minimum problem by using the techniques previously developed. Alternatively, the following symmetric procedure may be employed. Since f is to be made stationary, it follows that g= VS Da, = (5.3) a However, the n variables are not independent, but instead are subject to the constraints (5.2), so that only n — k variables are independent. The differentials dz; must satisfy k conditions of constraint, that is, 995 ae = ee 3 Ben 0 j=1,2,...,k (5.4) Multiply (5.4) by A; and sum with respect to j from j = 1 to j =k. Combination with (5.3) yields d ¢ + i #) dx; = 0 (5.5) which is true for arbitrary \1, Xs, . . . , Xe and arbitrary values of n — kof the differentials. In expanded form we have af i Ogata z the tg to +g ms) + “(zt +S Se ts #) des foe + (Zang aye “+n 32) dn =0 (5.6) Choose the )’s so that k of the n bracketed expressions are zero. It then follows that the remaining n — k bracketed expressions must be zero, since the n — k remaining differentials are arbitrary. Thus for a relative maximum or minimum, we have oa e — gia 0 G=1,2...50 Co) +e. t=O §1,2,...,k as necessary conditions. This system (5.7) is n + k equations to be solved for the same number of unknowns 21, 2; . . « ;tnjMyAz . » 0 y Me These equations (5.7) are also obtained by forming F(@it2, . «+ s®nyAayd2, - «+ ade) k = ferry, ... tn) + Y Agilent, - - stn) j= and setting the partial derivatives with respect to the n +k variables 2%... Tn My bay». . » Xe Equal to zero: oe Ox; (58) Fo 5=1,2...,k ay 12, , This again is the Lagrangian rule. Example 1: Minimize x? + 2:? + - - + + 24? subject to the condi- tion ayer + dete + °° + ante = 1, where clearly a:* + a2? +--+ a,? is positive. Solution: Form F(a1,22, « « - jn) Satta tos tat + Mati + ante + + ~ + ante — 1) oF Ge, 7 2 tm = 0 SF = 20, + dar = 0 Ooze oF x 2m, + da = 0 at,16 One and yxy + ay, + ++ + + Ont, = 1 Therefore a= —da/2 j=1,2, mn Thus, ¥ d ms or the parameter ) is given by ra--2 and j= —% . i The corresponding value of a1? + 22 + +++ + aq? is Note that the dual problem is to maximize the expression )\ aja; if mA > a; =1, and it is recommended that the reader solve this as an a exercise. Example 2: Find the greatest and least values of z on the ellipse formed by the intersection of the plane « + y +z = 1 and the ellipsoid 16x? + 4y? + 2 = 16. Solution: Form* F(xyyyzsAa,a) = 2 + aw + y + 2 — 1) + Aa(16z? + 4y? + 2? — 16) Set the partial derivatives equal to zero: F,=1+%+ 2d =0 @ Fy = + 8y =0 7) F, =i + 82dr = 0 Gai) and atytz-1=0 Gv) 16x? + dy? + 2? — 16 =0 ) 1 Since f(2,y,2) = z is to be extremized subject to the two constraints gi(z,y,2) = aty+z2—1=Oand gx(z,yz) = 162? + dy? + 2? — 16 = 0.Instead of solving for the five unknowns, we note that (ii) and (iii) are consistent if y = 4z[1 = d2 = 0 is excluded, since (i) is violated]. Sub- stitution of y = 4z into (iv) yields x = (1 — 2)/5. Then (v) becomes (in terms of 2) 16G —2)*, (416) —2)? | sige ee or, simplifying, this reduces to Ql? — 322 — 64 = 0 The maximum and minimum values of z are therefore § and —§, respectively.TWO Introductory Problems of the Variational Calculus It is the initial objective of this chapter to introduce the concept of the functional and then to extend the ideas of stationary values from functions to functionals. The notion of the vanishing of the first variation of a functional according to the theorem of minimum potential energy will then lead to the Euler equations and natural boundary conditions. ‘This will be done without stressing rigor in order to give the reader a feeling for the scope of the variational method. (These ideas will be presented more precisely in Chap. 3.) Examples of the static deflection of a beam and a circular plate are developed to show how the differential equations of equilibrium and boundary conditions may be derived. This is done without recourse to free-body diagrams, which can often yield incorrect field equations and boundary conditions in, for example, the case of structural elements such as plates and shells. Finally, the chapter will close with a brief discussion of some of the classical problems of the variational calculus. 1 Concept of Functionals Chapter 1 dealt with the subject of stationary values and more par- ticularly with extrema, i.e., maximum and minimum for functions of oneor more variables. Similarly, the calculus of variations originated from the endeavor to determine extrema or stationary values for functionals. A functional is defined as a quantity or function which depends upon the entire course or path of one or more functions rather than on a number of discrete variables. The domain of a functional is a set or collection of admissible functions which belongs to a function space or class rather than to a region in coordinate space. Several examples should help to make this concept clear. Example 1: The length L of a curve y = y(z) defined over the closed interval zo < x < 2, or more compactly [zo,2)], is given by the integral L@) = [2 VIF UP a & a (1.1) Thus the value of L depends on the path or form of the function y(c), called the argument function. In order that the integral (1.1) exist, it is sufficient that y(v) be taken as an arbitrary continuous function with a piecewise continuous derivative. This defines the space or domain of admissible functions from which the argument functions may be selected. Thus symbolically L = L(y) expresses this dependence of L on y. Example 2: As a second example, consider the functional $9) = [> “Fr @ at (1.2) where s is a parameter independent of ¢ and thus is constant during the integration. The “result” function f(s) depends upon the “object” function F(#). We express this functional correspondence in the form J(s) = L{FO} (1.3) and f(s) is called the Laplace transform of the function F(t). The admis- sible functions F({) are taken to be piecewise continuous in any finite interval and to be of exponential order as i—> © ; that is, a positive number M and a real number « exist such that |F(t)| < Me for all ¢ sufficiently large. These conditions are sufficient but not necessary to guarantee that f(s) exists for s > a; that is, the integral (1.2) converges. Example 3: Functionals may also depend upon the behavior of more than one function. In this example, we restrict ourselves to two admis- sible functions y(x) and z(z), which are defined by the conditions 1. y(e) and z(2) are continuously differentiable on the closed segment [20,21] 1 See, for example, R. V. Churchill, “Modern Operational Mathematics in Engineer- ing,” Ist ed., p. 5, McGraw-Hill Book Company, New York, 1944.20 Two 2. y(x) and 2(z) have specified values at xo and 21; that is y(t) = yo (ms) = oe) = 2% ea) = a4) A functional of y and z is T(ya) = [2 Feyany's’) de (1.5) where F is a continuous funetion of its five arguments x, y, 2, y',2". Our considerations can be easily extended to accommodate n functions of m variables involving, if m > 1, multiple integrals. For example, Chap. 3, Sec. 2, pages 37 to 41, discusses the case of a single function of two variables (n = 1, m = 2) in connection with making a double integral stationary, where the class of admissible functions F is more severely restricted than the integrand in (1.5). 2 Stationary Values for Functionals The application of minimum-energy principles involves the definition of stationary values for functions, or more generally, functionals, A stationary condition is necessary in order that an extremum (maximum or minimum) exist. In the case of a function of n independent: variables S(e1,22, . . . ,tn), the conditions for a stationary value at a point P are ef) _ of eae Oxy |p ~ Axe |p cone ae ey) as discussed in Chap. 1. In the case of functionals the problem is more complex. For example, suppose that we are given the functional Vw) = [2 Feyy'y") ae (2.2) where F is a given explicit function of a, y, y’, y’’ and xo, a1 are fixed numbers (Fig. 2.1). We seek the function y = y(x) (assuming its exist- y Fig. 2.1 Variation of extremizing function with fixed ends. yx) + By(x) —— Ora yx) EXTREMIZING FUNCTION (x05 Yo)ence) which makes V stationary where y and y’ are prescribed at z = xo and z =. Let us examine the effect on V of a variation in y by a small amount éy, where dy = dy’ = 0 at z — xo and x = a (Fig. 2.1). If V is an extremized functional, then v= cle y+ a + 3 oy" | ae = (2.3) The left-hand side is called the first variation of the integral V. In order to eliminate the variations éy/ and éy”’ from Eq. (2.3), the second and third terms are integrated by parts: okie: OF. Js pu d (aF fo Sy BY de = [# af aia #) by de 2 OP an ay a fF ay (4 FY) gy] 5p "= [Soo -[e ar) of. co 2 da? (OF ds (spr) ee so that aye ie [# d oF @ oF jy aay tae xl by dz ard oF ar +[(2-z%) af + [Bw f-0 eo If the boundary conditions (by = dy’ = 0 at 2 =z and 2 = m) are imposed, then nfor d oF, dt aF ic av = [e-2% Ew vee -0 (2.6) ‘The above integral must vanish for all admissible 6y, which requires that the expression in the bracket of Eq. (2.6) be zero, i.e., OF d (aF d@? (oF oy ~ & (i) + (Sp) -° at This follows from the fundamental lemma of the variational calculus discussed in Chap. 3. The differential equation (2.7) is known as the Euler differential equation corresponding to the functional defined by (2.2). This, together with the boundary conditions, determines y(z). Suppose that the requirements that y and y’ be prescribed at the ends are relaxed. The competing curves would then appear as in Fig. 2.2. For 8V to be zero, it is sufficient that (2.7) be satisfied in addition to OF d oF oF ay aay 9 Md By = 9 Gea at x =a and x=. These equations (2.8) are referred to as the “natural boundary conditions,” since they occur naturally from the22 Two Fig. 2.2 Variation of extremizing function with free ends. -y(x) + By(x) coe xex, y(x) a f EXTREMIZING FUNCTION Ke Ko variational process. Their physical significance will be discussed below in the examples of a cantilever beam and a circular plate. In general, the quantity 6V = 0 if at zo and 2; either oF d oF eee OF ay ~° (@_ y is specified or =0 (2.9) (ii) is prescribed or and the differential equation (2.7) is satisfied. THEOREM: We now state the following important theorems for elastic bodies: the theorem of minimum potential energy and its converse. Of all compatible displacements for a stable structure satisfying given displacement boundary conditions, those which satisfy the equilibrium equations make the potential energy V a minimum. (The potential energy of a structure is the strain energy plus gains in potential energy of the external forces.) The converse theorem, which is the one used most often, is as follows: Of all the displacements satisfying the boundary conditions, those which make the potential energy a minimum satisfy the equilibrium equations. Since the potential energy is a minimum for the true equilibrium state, it must also be stationary; ie., sv =0 (2.10) Example 1: A uniform clastic cantilever beam (Fig. 2.3) with bending rigidity EI and length L is uniformly loaded. Derive the differential equation and boundary conditions by a variational procedure. Deter- mine the end deflection y(L).2 A . EL is Solution: "The strain energy is =" |," y’” dz (due to beam bending). L The gain in potential energy of the external load q is — f qy dx. There- fore the total potential energy of the system is which implies that [EL , v=f [Fy a |de Comparison with the general form (2.2) yields EL FaSu?—q (2.11) To find the Euler equation (2.7) for this specific F we calculate oF oF oF 7 gy Tt Gy tO gy Bly (2.12) Substitution into (2.7) yields the differential equation of equilibrium Ely" —q=0 (2.18) The boundary conditions at the clamped end are y(0) = y’(0) =0 (deflection and rotation prevented). The unsupported ends are free to deflect and rotate, so that the boundary conditions (2.8) apply. Substitution of (2.12) into (2.8) yields Ely" =0 (2.14a) and Ely” =0 (2 .14b) Equation (2.14a) states that the shear is zero, while (2.146) implies that the bending moment is zero at the free end. These are the natural boundary conditions. Fig. 2.3 Cantilever beam under uniform load.24 Two The solution of (2.13) may be effected by successive integration: wn qh a EI wn He aoe Oe ay +e, ¢, = arbitrary const 0 thus ¢, = — 7 BI wn Vg y Br e-) na (e i uw =ay(F- Leta) ee = arbitrary const L y"|,_, =0 therefore cs = 5 or ma ge 2 u! = gay @ — We + 1) Integrating again, Buty’ |,_, = 0 and soe: =0. Finally, gh (ee tae if an(i ooo where y i = 0 and consequently ci = 0. Thus the deflection of the beam is given by a rere v= (S- $+ 2.15) In particular, the end deflection is = 7 (M_l IN _ dt Yea ~ DET G 3 +o) 7 eT (2.16) Example 2: A uniform elastic circular plate of radius b with a concen- tric unsupported hole of radius ais shown in Fig. 2.4. The plate is loaded axisymmetrically with a distributed normal force per unit surface area q = q(r) at the top and is reacted at its outer edge as shown in the figure. ‘The differential equation and boundary conditions are required.ix iG pelt SULLA ke Q : Fig. 2.4 Section of oe plate with concentric hole loaded, and reacted axisymmetrically. Solution: The potential energy of the system (loads + reactions) may be shown to be 2 V=aaD ds [rent +E+ anv ar — on f? qrw dr + 2n{rMw, — Qu)» (2.17) where D = plate flexural rigidity w = normal deflection (positive downward) wy = dw/dr Wr = dw/t ‘dr? M = radial bending moment per unit of circumferential length Q = radial shear per unit of circumferential length (The positive direction is as shown in Fig. 2.4.) = Poisson’s ratio For stationary potential energy, 8V = 0 must hold. Noting that 20 Wer = 2 and that 54 = 4 8V = 2D i [2rd + 22 += w,* d a 6, 2 by, rf (2u, in) | — nr in gr bw dr + 2n(rM bw, — 7Q 600)» = 2D { rh ‘ [re Btbpp + (rte + 9) bu, + rw, 7 ar 6 gr bw _ 7Q bw — PoP a+ (Gow D )} Integration by parts yields 3 o- [ ew. + 1.) boy + (vm + 2) ww]. i ifs [z (rim + un) buy + 4 § (mm +2 on ar - [oe +5 bu, 28) =0 lo Bs 1 This permutation of the order of operations is justified in Chap. u °26 Two or integrating the term containing 6w, by parts, and collecting terms, we have ie [# (rtm + van) — 4 (rn + 9) - al bw dr + {(r%0 + ny) B06 + [~- $F Grog + ve) | on! +(5 M sy, — a ow) =0 (218) The differential equation is found by setting equal to zero the coefficient of 6w in the integral with respect to r.!_ Thus @ d w) gr _ oe Since the terms in » cancel, we have ' fm~E(2)-§-« or, expanding ee TWeere + Were — zo and y: > yo by asmooth curve y = u(z) for which the time taken by a particle sliding without friction from A to B along the curve under gravity is as short as possible (Fig. 2.5). Let vo = initial velocity; then from energy considera- tions, the velocity at any point (z,y) is determined by v? — vo? = gy — yo) or v? = Qgy + vo? — 2gyo Let a be defined by 2gec = 2Wgyo — vo®Therefore v? = 2g(y — a), and taking the nonnegative square root of both sides yields a v= FG = VM Vy ‘The time taken in the fall is Pa eds c EVO nD WS 2 ae Ignoring the unimportant factor +/2g, the problem of the brachistochrone may be stated as follows: Among all continuously differentiable functions y(z) satisfying y(t0) = yo and y(z1) = y1, find the one which minimizes r= [7 ew VIF OY Vy —a (3.3) The solution to this problem was independently published by the two Bernoulli brothers, James (1654-1705) and John (1667-1748). The techniques at arriving at the solution were quite different. It is inter- esting that a rivalry had arisen between these two men, and this gave an unusual impetus and zest to the beginnings of the variational calculus. Solutions to this problem were also obtained by Newton (1642-1727), Leibniz (1646-1716), and L’ Hospital (1661-1704). More important, the two papers of James Bernoulli in 1697 and 1701 were the starting point for the researches of Euler (1707-1783), who was a pupil of John Bernoulli. Buler was responsible for the first important result in the modern theory of the calculus of variations. Finally, we would like to mention briefly several other problems. 1. The simplest of all is that of finding the shortest smooth plane curve joining two points A(zoyo) and B(z1,y:). ‘The length is given by T= [?ViF Gide (3.4) and J is to be minimized. Our intuition suggests that the answer is a straight line connecting the two points. This is proved mathematically in Chap. 8, Secs. 1 and 3. In three dimensions, the corresponding problem is to determine two funetions y = y(2), 2 = 2(x) such that the integral JE VIF OFF CP ae (3.5) has the least possible value, where y(z) and 2(z) are prescribed at the end points of the interval. Still more generally, we have the problem of finding the geodesics on a given surface G(z,y,z) = 0, that is, of joining two points on the surface30 Two Axo, ¥0) Bix yi) Fig. 2.6 Minimum surface of revo- lution. with coordinates (xo,yo,20) and (z,yx,21) by the shortest curve possible lying in the surface. This problem is not simple. Analytically, we seek among all triads of functions 2(1), y(0), 2(t) which satisfy identically the equation of the surface G(x,y2) = 0 and for which x(t), y(t), 2(4i) ( = 0, 1) are prescribed, to find that triad for which the integral [" V+ 9 + dt is least. 2. The problem of minimum surface of revolution: Two points A (x0,y0) and B(t:,y:), where 21 > zo, yo > 0, y1 > 0, are to be joined by a curve y = u(z) lying above the z axis in such a way that the area of the surface of revolution formed by rotating this curve about the z axis is least (Fig. 2.6). The functional to be minimized is T= 28 ["yVIF Gear cD) This and some related questions have been approached both analytically and experimentally using soap solutions.’ The problems introduced in this section are more extensively treated in Chap. 3. 1 The reader is referred to a very informative discussion of the soap film experiments in R. Courant and H. Robbins, “What Is Mathematics?,” 4th ed., pp. 385-397, Oxford University Press, New York, 1947.THREE Euler-Lagrange Development with Applications This chapter begins by returning to the basic problem of formulating Euler's equation for a wide class of problems where the integrand in the functional is of the form F(z,y,y’). The problem of the determination of the shortest distance between two points is solved from both a variational and a nonvariational viewpoint. The theoretical development is then extended to more general integrands involving (1) more than one depend- ent variable, (2) higher derivatives, and (3) several independent variables. This is followed by a section on natural boundary conditions with applica- tions to the problems of determining the minimum distance between parallel lines and the minimum surface of revolution. Still more general movable boundary problems with associated trans- versality conditions are considered somewhat briefly. The brachisto- chrone problem is then presented next and application to isochronous clocks is considered in some detail. A systematic treatment of variational notation follows; it shows the analogy between stationary problems for functions and functionals. This notation is then applied to the small deflections of a rotating shaft. The last three sections of this chapter are devoted to Lagrange multipliers for functionals subject to the side or constraint conditions in function and integral forms. Applications to geodesic and isoperimetric problems are given to illustrate the method. 3132 Three 1. The Basic Problem—Euler’s Equation Our object is to determine necessary conditions that a function y = u(z) makes the integral = [2 Feuy! de aa) stationary where y(to) = yo and y(v:) = y: are preseribed. The function F(z,y,y’) is assumed to possess continuous second derivatives with respect to its three arguments. Also, y’’ is taken to be continuous in the interval concerned. To this end we construct y(z) = u(z) + en(x),! where (2) is contin- uously differentiable and (0) = n(2:) = 0. The quantity ¢ is a param- eter which is independent of 2. Substitution into (1.1) yields T= 10 = [2F@utenu +e) a (1.2) In order that J be stationary at e = 0 it is necessary that a F\ = 0 (13) But al a oF oF mae (acta Fay * ow Fay *) i and al a [-[EerSeleene a where F = F(z,u,w’). A a integration by parts of the second term results in 2 OF oF ype (a aF on Bul? = Gy? of a £ a & a) a =- [*,(2 2 = Ox du! since 7 is zero at each end of the interval. Therefore dl n(@F _ d aF 2 mina (3-die) c. However, 7 is arbitrary outside its differentiability requirements and boundary conditions, Then, as shown in the proof of the fundamental 1 Note that in this case dy = en(z). Weshall be concerned with variations in which both Sy and 6y' are small. These are referred to as weak variations.lemma of the variational calculus, (proved below) it follows that au ~ daw ~° (8) This is Fuler’s equation, and it is an ordinary differential equation of the second order. It must be satisfied subject to the boundary conditions u(t) = yo —U(as) = yn (1.6) Proof of the Fundamental Lemma of the Variational Calculus LEMMA: If a function ¢(x) which is continuous in [0,7] satisfies the relation [E 1@o@ ax = 0 (7) where n(z) is any function such that n(z0) = (zi) = 0 (1.8) and 1/(z) is continuous in [x0,21], then ¢(z) =0 ins SrSu (1.9) PROOF: Suppose on the contrary that $(z) #0 at x = £, where zo <&0 in (ff:). Therefore hypothesis (1.7) is violated and we have a contradiction. A similar contradiction is reached if $(8) <0, which proves the lemma. In some applications the basic lemma is required in a more restrictive form. If, for example, the integral34 Three (1.7) is to vanish for all twice differentiable 7 for which (1.8) holds, then by using the previous argument with n(x) = (x — &)%(% —2)* in fo 0. Then since »/a? + y? > 2, it followsthat [i vEFRaz fPea= frac =a Therefore, L > a and the equality holds if and only if y = 0 or the curve is the straight line. ‘This completes the proof and the example. If the quantity d/dz F,, (in the general case) is determined by applica- tion of the chain rule of differentiation, then (1.5) becomes in expanded form (after multiplying through by —1) wl Pu + U'Pan + Paz — Fu = 0 qa.) where subscript denotes partial differentiation (e.g., Fvu = 0°F/du' du). Every solution of (1.5) or (1.11) makes the given functional stationary. The stationary character is necessary for the occurrence of a maximum or minimum. However, as in the case of ordinary maxima and minima, it is not sufficient to ensure the occurrence of either of these possibilities. The determination of sufficient conditions is considerably more difficult than the corresponding problem of the differential calculus. ‘The formal systematic treatment of the sufficiency conditions as developed by Jacobi, Weierstrass, and Hilbert, for example, is beyond the scope of the present volume. The reader interested in this phase should consult references by Akhiezer, Bliss, Bolza, Courant and Hilbert, Elsgole, Fox, Gelfand and Fomin, and Pars. 2 Euler Equation for More General Cases Integrals with More Than One Dependent Variable. The problem of determining the stationary values of an integral can be ‘extended to the case of n dependent functions y; = y:(z),i = 1,2, ... ,m. Let F(yrys, -- - Ym¥o¥» +» + sYn) be a function of the 2n +1 arguments z, yi, Y2,-.., y, which has continuous second deriva- tives in the interval concerned. If we substitute specific functions us = fila) = 1,2, . . . ,n) with continuous second derivatives into T= [2 Fyn... evita) de (21) then the result is a number. In general, T= Tys,Y2, + + + Yn) The variational problem now requires that we determine among all these systems of continuously differentiable functions y:(x) (i = 1,2, . . . ,n) prescribed at 2p and 23, the functions y; = u,(z) which make J stationary. In many cases, the stationary value will be an absolute maximum or a minimum.36 Three We consider the family of functions yz) = wr) ten) 1=1,2,...,0 (2.2) where m(z), m2(z), . . - , m(z) are n arbitrarily chosen functions which satisfy (xo) = 0, ni(t1) = 0 and furthermore that n;(z) possess continuous first derivatives in [20,21]. Substitution of (2.2) into (2.1) results in 10) = [2 Futon... tebe ui pen, oo. su pent) de A necessary condition for an extremum ys = 1(z) is that leo = 9 But we ean choose, for example, mz = 95 = -* > = 1 =O and 1 #0, so that. 160) = 2 FG, Ws $ ens, as us, «yy WL Henly hy 5 sh) de Then, — as we did for the one-variable case of See. 1, there results oF d a. fl (je hiu) ne =0 which implies that oF d oF @ oe OF = Feats, «6. Unie, . 6. UA) More generally, if u; alone is varied, then there results oF _ d oF : oo eS #=1,2,. n (2.3) This is a system of n differential equations of the second order for the n functions u(x), ..., Un(z). Each function u(x) must satisfy the boundary conditions which are imposed on the corresponding functions y(z)(@ = 1,2, . . . ,n). Integrals Involving Higher Derivatives. The above development may be extended to the case where the integrand assumes the form Faeyy'y”, sy™). To be specific, attention will be confined to the casen = 2. Therefore, we are interested in determining the requirements on y(z) to make the integral Ty) = [2 Fey y") de (2.4) stationary subject to the conditions that y and y’ are prescribed at each end. Assume y = u(x) to be the function which makes J an extremum.Form the family of functions y(x) = u(x) + en(x), where ¢ is a constant and 7 is an arbitrary function with continuous derivatives up to the fourth order which together with its first derivative vanishes at the end points of the interval.' Substitution into (2.4) yields 1 = [2 FG, ut ew! + of, ul + a1”) de (2.5) Hall Again 9 = 0 is required. We employ differentiation under the integral sign using the chain rule of differentiation and then integrate by parts to eliminate 1” and 7! using the boundary conditions on 7. ‘This yields foo[F Biot Bho ae = 0 26) ze “Ga dete: : where F = F(2,uju',u!’). Since n is arbitrary outside its differentiahility requirements and boundary conditions, we deduce the Euler differential equation corresponding to (2.4), : ret ret Sie =0 (2.7) This is an ordinary differential equation of the fourth order. It must be solved subject to the boundary conditions u and wu’ specified at x = zo and z = 21. Still more generally, if F(x,y,y/,y", - . . , y%) is the integrand and y, y', y", . +4 yt) are prescribed at the ends of the interval,’ the corresponding Euler equation is da @ a oe in a i.) = Pe — Pe + Per — Pert + (-I FF = 0 (2.8) where F = F(z,uw'u, ... ,u®). This is a differential equation of order 2n which must be satisfied subject to the 2n boundary conditions. Integrals Involving Several Independent Variables. The above method for determining the necessary conditions for an extreme value can be applied equally well when the integral is a multiple integral. For example, consider the instance of a double integral. Let D be a given region bounded by a sectionally smooth curve C in the zy plane. The integral to be treated is I= J F(x,y,2,t02y0y) dx dy (2.9) 1 Euler's equations (2.7) and (2.8) still remain valid when the boundary conditions are different from these in this section (see Sec. 4 of this chapter, on natural boundary conditions).38 Three where F is continuous and possesses two continuous derivatives with respect to its five arguments and w(z,y) is prescribed on C. To effect the extremization of (2.9) we introduce a one-parameter family of comparison functions w(ay) = u(ay) + en(z,y) (2.10) where u(z,y) is the function which makes (2.9) stationary and n(z,y) has continuous derivatives up to the second order and is zero on the bound- ary C. The integral then becomes a function of e: 10 = [f F@yw,t0.s,) de dy (2.11) 4 which for stationary values must satisfy J’(0) = 0. But ye oF oF 10 = [f [Font feet $n ety J : and r@= [ [ier inet oF 4, ac ay = 0 where nz = 41/dz, 1, = 4n/dy. To eliminate the terms in 7: and 7, we integrate by parts—the former with respect to x and the last term with respect toy. Therefore,! od OF o OF If o[ a sein ~ ayan |e” OF dy aF dz], _ + fo [ Ee |e =o (2.12) where de = & ds, dy = ds, and dsis the differential arc length C. But 1 = 0 on C, so that the second integral is zero and then there results OF a OF 0 OF [fol ~ ia ou, ~ Hie] = If we invoke an extension of the basic lemma of the variational calculus (proved below), then the Euler partial differential equation of the second 1 The integral if means that C is followed in the conventionally positive sense of description. This implies that the interior of the region D bounded by the closed curve C'remains on the left as C’is traversed. More simply, (2.12) follows by applica- tion of Green’s theorem in the plane.See os =0 (2.18) is obtained. Proof of the Extension of the Basic Lemma. Let D be a domain of the zy plane for which Jf emGtey) ax dy = 0 (214) y for all continuously differentiable (x,y) that are zero on the boundary C. Furthermore, G(z,y) is continuous for all and yin D. We will establish that G = 0 in D. Suppose on the contrary that G # Oin D; then there exists a rectangular subdomain D, of D for which G # 0. Let this rectangular subdomain be i <2 < & m <1 m and assume (without loss in generality) that G@ > Oinits interior. Choose 9(2,y) = (x — £:)?(@ — &2)*(y — m)*(y — m2)? in D, and » = 0 over the rest of D. Then [J reneew ae dy = [ n(ayy) Oey) de dy > 0 D ! contrary to hypothesis (2.14). Thus the assumption that G > 0 in Dy has resulted in a contradiction. Analogously, we can show that G <0 in a subdomain also leads to a contradiction, which implies that G = 0 is the only other possibility. Example: Poisson and Laplace Equations—Static Membrane Deflec- tion. The equations of Poisson and Laplace occur very frequently in mathematical physies. Let us consider a problem which yields them as Euler equations corresponding to making a particular functional stationary. Suppose that we wish to minimize the integral ie {f fue? + uy? + 2f(z,y)ul de dy (2.15) where f(z,y) is a prescribed continuous function and u(x,y) possesses continuous first and second partial derivatives in D. Furthermore, u is prescribed on the sectionally smooth boundary C of the two-dimensional domain D, that is, u=g(s) on€ (2.16) ‘The parameter s is the are length along C measured from a fixed (but arbitrarily selected) point on C. Also, wis assumed to possess continuous first partial derivatives on C.40 Three For stationary J, it is necessary that (2.13) holds; i.e., Use + Uy =f (2.17) which is Poisson’s equation. In particular if f = 0, (2.17) reduces to Laplace's equation Use + Uy = 0 (2.18) For this example, we shall now show that the solution of (2.17) makes the functional (2.15) an absolute minimum with respect to all functions with continuous second partial derivatives which satisfy the boundary restraints (2.16). Let (x,y) be any function of the set, and we write U(z,y) = u@y) + @y) where u(z,y) is the solution of (2.17) subject to (2.16). Since a(z,y) also satisfies (2.16), the function n(z,y) = OonC. Define AI = 1(@) — I(u) fue + ma)* + (uy + my) + Fu + aM] de dy 5 4 - { [ut + ry? + 2ful de dy = 2 ff tue + van Sade dy + ff tne + 98 ae dy From Green’s theorem (or simply integration by parts) [f (oon + wan ae dy = fon oe ag [J (1 + 1g) de dy where d/dn stands for the derivative in the outward normal direction. But then we have a= 2 fongyas + [f 2 ~ (ee + uel de dy + ff Int + mA de dy The first and second integrals are zero, since 7 = 0 on C and (2.17) holds. Therefore, at = ff int + nA1dz dy > 0 b as the integrand is nonnegative. Moreover, AI = 0 if and only if nz = ny = 0 in D, which implies that 7 is independent of both x and y or merely a constant. However, » = 0 on C, so that 7 = 0 in D and on C.‘Thus AI > 0 if @ # u, which signifies that the function u minimizes the functional I. This completes the proof. Let us treat briefly a simple mechanical illustration which gives rise to the integral (2.15). Consider the case of the static equilibrium of a stretched homogeneous elastic membrane. By definition, a membrane is an clastic skin which is incapable of resisting bending but does resist stretching. In fact, the elastic membrane may be considered as the two- dimensional analogue of the elastic string, which is treated in detail in Chap. 6, Sec. 1, Example 1. It is assumed that the elastic membrane is stretehed over a certain simply connected planar domain D bounded by a rectifiable contour C. If u(z,y) denotes the displacement normal to the plane, then the potential energy (for small deflections) may be shown to be V=4 if] [Suet + uy?) — 2qul de dy (2.19) The quantity S is the tension per unit of length. It is constant if it has this constant value at all the points of boundary C. This means that along any arbitrary cut through the membrane, the tension will be trans- ferred from one side to the other without change of magnitude. Also q = 4(z,y) is the external load per unit of projected area acting in the direction of positive u. Comparison of (2.19) with (2.15) shows that the Euler equation corresponding to (2.19) is Use + Uy = — 3 (2.20) and it is this Poisson differential equation of equilibrium which must be solved subject to u=0 on (2.21) Solutions of the boundary-value problem defined by (2.20) and (2.21) for particular domains such as rectangles or circles are readily obtained by the method of separation of variables (in rectangular and polar coordinates, respectively). The reader may refer to many sources for detailed solutions; see, for example, R. Courant and D. Hilbert, pages 297 to 307. In this excellent reference book the free and forced vibra- tions of such membranes are treated in an interesting and comprehensive fashion, Additional Examples Example 1: Two Dependent Variables. Suppose that T= J Cys — 2u? + (i)? — WD) de is to be made stationary.42 Three Then F(x,11,u2,uj,u3) = 2uyu, — Qu? + (ui)? — (uh)®, so that oF du, 7 2s — Ae oe Fa 720 The first Euler equation is [from (2.3)] ard oF San 7 de dat 7 22 ~ An — 2a’ = or uy — Quy — ui’ = 0 @ Similarly, oF uz oF aus = Qu = —2uh so that the second Euler equation is OF d oF Su, de du, or um tu =0 Co) Equations (i) and (ii) form a system of linear equations which is readily solved. Differentiation of (i) twice and combination with (ji) and thereby eliminating w, yields the ordinary differential equation of the fourth order with constant coefficients uy” + Quy’ + wr = 0 ii) Let uw: = &* be substituted into (iii), which yields the characteristic algebraic equation O+)2=0 (iv) with corresponding solutions } = +i, +i. This means that us = (ext + ¢2) cose + (osx + cx) sine ) is the complete solution of (iii) where c1, cx, cy and cy are arbitrary (real) constants of integration. Then substitution of (v) into (i) gives the expression us = ¢x(—2sin x + x cos x) + cy cost + (2 cosa -+2sinz) +esinz (vi)A simple check shows that (ji) is identically satisfied. The constants ¢1, C2, C3, Cs must be determined from the boundary conditions at each end involving the quantities uw: and us. Example 2: Integral Involving Higher Derivative. Find the Euler differential equation associated with JZ tou — W? + 6@)] ae where ¢(z) is an arbitrary continuous function of x. Solution: The integrand is Fayy' iy") = ly? — (y)? + o(@) where y’ is not explicitly present. Therefore F(x,u,w ul’) = 16u? — (u’)? + $(2) Fy = 32u Fu =O Fun = —2u"” and the Euler equation (2.7) is a a Fy- Fw + ale =0 this yields the fourth-order ordinary differential equation 32u — uv” = 0 or w” — 16u =0 @ The solution of (i) is readily found by substitution of u =e into (i). This yields the characteristic algebraic equation M—16=0 with roots \ = +2, +27. Therefore, u = ce + cxe-** + cz cos 2c + sin 2x (ii) where c1, ¢2, ¢s, and cs are arbitrary constants, is the complete solution. Again, these constants are to be determined from the boundary restrainis. Example 3: Find the Euler partial differential equation for the functional T= fff Peeyeni0, 0,0 ttee Dry es ay ele) de dy dz (2.92) + where V is a definite region in xyz space. The functional (2.22) arises in problems of static equilibrium of elastic solids and dynamics of “two-dimensional” bodies such as plates, shells,ccd Three and membranes. In the former instance we have three spatial coordi- nates, while in the latter, time replaces the third spatial coordinate. We define the class of admissible functions w(z,y,z) as follows: 1. w and all its partial derivatives of order four are taken to be continuous. + 2. w must satisfy on the surface S bounding V the linear forced con- ditions of the form aw + 8 dw/dn = y, where a, B, y are prescribed functions of the coordinates z, y, z. Furthermore, F is assumed to possess continuous third partial deriva- tives with respect to all of its arguments for all x, y, z in V. To effect the extremization (again we assume its existence) let u(x,y,z) make J stationary and write w(z,y,z) = u(z,y,z) + en(a,y,z), where 7 and its first partial derivatives may be taken to be zero on S, since our imme- diate interest is im the formulation of the Euler-Lagrange equation corresponding to (2.22). Consequently, if we form J(e) and then a Flico = 0: there results [ff toPe + nee, + mPa, + Po, + tesPese + tivP og + MP on t e t+ aya, + taFy,, + t2F u,,] dx dy de = 0 From the divergence theorem (or integration by parts), application of the boundary conditions on 7, and the basic lemma of the variational calculus, we conclude that a a 3 a a F.- -& fr) = a ey) -Z 2 +& 7 + i Poa) 2) + poe Pu,) + 525 Fun) = 0 (2.23) = 5 z az in On Equation (2.23) is the Euler partial differential equation. This must be solved subject to the boundary restraints. Furthermore, we may derive “natural boundary conditions” by not imposing all forced restraints on w and u at the boundary. These conditions will occur by determining that certain integrands of surface integrals must be zero. Natural bound- ary conditions will be discussed in See. 4 of this chapter. 3 Solution of Euler’s Differential Equations The Euler differential equation (1.5) of Sec. 1 is a second-order differ- ential equation which in general cannot be explicitly integrated. How- ever, in special cases the equation can be solved in quadratures.Dependent Variable u Absent—F = F(z,u'). For this problem, F,, = 0, so that Euler’s differential equation becomes J x) =0 (a) which integrates to Fy = ¢ = const (3.2) Thus the problem of the determination of the extremization function is reduced to a first-order differential equation. If Eq. (3.2) can be solved for u’, such that w! = Gee) then ua [Peeodth (3.3) where k is another arbitrary constant of integration. In this instance, the Euler equation is solved by quadrature. ‘The constants c and & are determined from the conditions at the end points zo and 2 of the interval. Suppose further that F = F(w), that is, both the independent and dependent variables are absent. ‘Then oF = 2 — sanction of w = (4) which implies that u’ = k, where k is a constant which is a function of ¢, that is, k = k(c). Consequently, the extremizing functions corre- sponding to F = F(w’) are necessarily linear functions of x or straight lines. Therefore, for the shortest distance in plane problem, treated in Sec. 1, F = »/1 + (w)? and the solution is a straight line connecting the two given points. Example 1: It is required to extremize the functional 1) = [>MEOe 2>0 Solution: ‘The integrand F(w’z) = ~/1 + (w/z does not explici depend onu. According to (3.2), the first integral of the Euler equation is Pox = ¢ = const (3.6) rV1 Fwy Equation (3.6) is solved most simply by setting u’ = tant, —1/2 1 — wu? sin’ a 2 > 1 —sin Q = C8 > Therefore the actual period T' is bounded by the estimates cL an VL/g Qn VE = aru =0. But aH _ jn [30.mu 4 aoa" Og dy OG | By’ dG mS [ut av] ® ag ie ; ={2 [Sea+ 28 ai]ae =0 j=1,2,...,N41 Integration by parts and utilization of 1j(¢0) = n(x) = 0 yields nn [ae ae = fon[%- & (3) |# =o f=1,2,...,N+1 @9) where the 7; are arbitrary differentiable functions for which 1j(to) = 9;(@1) = 0 Thus ae a9) (3) - 0 (8.10) which is a differential equation of order two. Integration of (8.10) will give rise to two arbitrary constants of integration. Thus we have N + 2 unknown quantities \1, X2, .. . , Av and the two constants of integration. These may be determined from the N + 2 equations (8.2) and (8.3). More generally, we now seck to extremize the integral firey... 26m... ad oof 6.11) with respect to continuously differentiable functions x(!), y(), . . . » 2(0)72 Three subject to the N constraint conditions in integral form Ky= [i Giaw-..att... ad 7 =1,2.-.50 (8.12) The required equations to be solved are a¢ _ d (a¢\ _ ad (2) ma a¢ _ 4 (dd) _ 9 dy = dt \ay. (8.13) oo a (eye a di @) ae where n o=F+ YG ron and F, Gi, . . . , Gy are defined by (8.11) and (8.12). The equations (8.13) must be solved subject to the boundary and constraint con- ditions (8.12). Example: Find the closed plane curve of given length which encloses a maximum area. Solution: The curves which compete are taken to be of the form z = x(t) bo nates ars — Fe br dt = 0 mH Integration by parts of the first term and use of the constraint condition br; = 0 at t = t and tz yields filer+ y s+ ars] dt = 0 (3.4) a where T=} S mi? (3.5) & is the kinetic energy of the system of particles. Equation (3.4) is an expression of Hamilton’s principle for a discrete system of particles in general form (for a holonomic system). Various important specializations can be made. First, the force acting on the ith particle can be divided into external or applied forces F* and internal or constraint forces Fy, so that F, = Ft + Fe (3.6) In many practical instances the work done by the constraint forces is zero, i.e, >; Fy: or, = 0 (3.7) 1 This is true, for example, for (1) the internal forces in a rigid body, (2) the reaction on a body rolling (without slipping) on a fixed body, and (8) the tensions in an inextensible string of an Atwood machine. For this situation, Hamilton's principle reduces to fi ler+ 5 Fe ori] dt = 0 (3.8) 4 Finally, if the applied forces are derivable from a potential function V, we have aV aV av Fe=— @ x, #) (3.9) where zi, ti2, aia are the rectangular coordinates of the ith particle.Alternatively, (3.10) and therefore for a conservative system, a fs a [" or — avy) a= sf" —v)a=5 ("La =0 (11) In words, Hamilton's principle states that the motion is such that the integral of the difference between the kinetic and potential energies of the system is stationary for the true path when the initial and final configurations are specified. 4 Lagrange’s Equations with Applications The Lagrange equations may be deduced immediately from Hamilton’s principle for a conservative holonomic system. From (3.3) .yn The kinetic energy of the system is T= 4) mite = Tass, «+ + dordudes « « « sdnt Ft Similarly, from (3.3), V = V(rurs, . . . sta) = Vgugn - - - gost) so that the Lagrangian L=T—V = L(gygs, - - - sdorduda, » - + Gost) From condition (3.11) and the basic Euler formula there results the equa- tions of motion doaL_ aL tae sp (4.1) or equivalently Golaor ev. ie HG 3 7 age TTP (4.2) ‘These are the Lagrange equations for a conservative system. The advantage of the Lagrangian equations is that the equations governing the motion of the system (with ideal constraints) can be written down by84 Four k ma ™ www Eo Tl Fig. 4.2, Basic two-mass-spring sys- tem performing translational vibra- tions. tule, thereby enabling the practitioner to find them with a minimum of mental effort. On the negative side, this procedure does not promote insight into the physical processes that are taking place. In short, the use of Lagrange’s equations is analogous to the function served by integral tables to the user of mathematics. While these tables do not give insight into the integration process, they expedite the ensuing calculations and considerably reduce the chance of error. The application of faulty (but reasonable appearing) free-body diagrams for complex systems such as plates and shells leads to the formulation of erroneous differential equa- tions, thereby emphasizing the need for the Lagrange method. Three examples will be demonstrated to illustrate the formulation of the Lagrange’s equations for oscillating systems. Example 1: Linear Oscillations of Two Coupled Masses. Consider the particular case of small free vibrations of the system shown in Fig. 4.2. ‘Two masses m; and m: can slide without friction along the horizontal zaxis. The connecting springs have spring constants given by ky and he. We seek the equations of motion for the two-mass system. Let x1 and zz denote the respective displacements of the two masses from their equilibrium positions. The kinetic energy of the system is T = 30m? + mets?) while the potential energy (which in this case is the strain energy) is given by V = Heys? + Ha(er — 21)? Lagrange’s equations are (L = T — V) d (aL dt \ oR, so that we have —kyer + heo(a, — 21) —kn(e2 — a1) midy mtr (4.3)vertical oscillations where the displacements are taken from the static equilibrium position. The above system is completely analogous to the torsional oscillations of two circular disks suspended from two coaxial circular shafts as shown in Fig. 4.3. The equations to be solved are analogously T Gi, Gals at be Th =~“ a+ F — 0) iL | ; Gal: The = —"F7 — 0) 1 T where 63, #2 = twists of the disks i ir Gi, Gz = shear moduli of the shafts [ ; J;, J2 = axial polar moments of inertia | of shafts cross sections wast wd.* Fig. 4.3 Torsional oscil- 32° 32 lation of circular shafts. Iy, Lz = lengths of the shafts J;, Iz = axial moments of inertia of disks and quantities G,J:/L; are the “spring constants” of the shafts. Example 2: Find the differential equations governing the free vibrations of the system in Fig. 4.4a (about its static equilibrium position) 77 Fig. .4 Vibrations about static equilibrium. eee fee Bie a 1 : na I 2q2-4) Se we ()86 Four assuming that the springs are massless and that the bar to which masses m are attached is rigid and massless. Solution: Let q, and q: be the displacement of the masses (Fig. 4.4b). The system has two degrees of freedom. We choose q: and gz as the generalized coordinates. ‘The rigidity of the bar requires that the dis- placement of the other spring is 292 — qi. The kinetic energy of the system is T = ¥(mix? + més’) and the potential energy of the system (strain energy in springs) is kg? ik = +5 Cn — a)? Lagrange’s equations become mgr + 2k — m2) = 0 m2 + 2k(2q. — qi) = 0 so that the mode shapes and natural frequencies may be readily obtained. Example 3: As a final example, we consider the determination of the equations of motion of a spherical pendulum (Fig. 4.5). ‘The position of the particle of mass m is completely defined by the spherical coordinates, namely, the colatitude @ and the longitude @. Thus we choose these angles as generalized coordinates: q: = @ and q: = ¢. Then the kinetic Fig. 4.5 Spherical pendulum.and potential energies are 2 T= oe (6 + 6% sin? 8) V = mgacos@ — (VJpazn = 0 is chosen as datum) and differentiation yields —= 2 2 2 gi: 0 ma? ma? ¢? sin @ cos 0 oF = mga sin 6 a6 or : or av pelle ey ee a 36 ma? ¢ sin?? 6 0 6 0 Therefore (4.2) yield the following equations of motion for a spherical pendulum: ma*§ — ma? ¢? sin 6 cos @ = mga sin 6 d j ay (ia? ¢ sin? 8) = 0 The particular cases of a simple pendulum (¢ = 0) and the conical pendulum (6 = constant) may be readily treated. 8 Lagrange’s Equations for a General System We seek the Lagrange’s equations for a general nonconservative holonomic system of p degrees of freedom. The inerement of work done 6W by the resultant force field F; during a virtual displacement dr; of the ith particle is oW = Y) Fie ors (6.1) ron But from r; = ri(qi,g2 . . . ,Qp,t) it follows that > ors a= Y ag (5.2) hh qe Substitution of (5.2) into (5.1) yields 2s = peed aw = > Fis 5 ba (53) ee ia (Result of sum is independent of ordering.) Define a new quantity Q,88 Four called the generalized force component by the formula 2 aW = Y) Qu du (6.4) mo so that os a= Sr. ar (5.5) The generalized force Q; has the dimension of a force when q is a length and the dimension of a moment when q: is an angular coordinate. In general, the dimension of Q, is determined by the rule that the product Qiqi has the dimension of work. It is not necessary to use (5.5) to caleu- late the generalized forces, since they can be found directly from (5.1). Consider, for example, the spherical pendulum. The only external forces are the weight —mgk (where k is a unit vector in the positive z direction) and the tension in the string. Then the virtual work corre- sponding to a virtual “displacement” 66, 6¢ is 8W = —mgk - [a 80 ey + asin 8 6¢ es] where ey and eg are unit vectors in the direction of increasing @ and 9, respectively (the string tension does no work). Thus if @, ® are the generalized forces in the @ and ¢ directions, 8W = © 0 + © 54 = mgasin 656 so that the generalized forces are @ = mga sin 6 o=0 66) Let us return to the problem of formulating the equations of Lagrange for a general (conservative or nonconservative) holonomic system of n particles with p degrees of freedom. Substitution of (5.4) and (5.1) into (8.4) yields fort 3 Q dqx) dt = 0 (0) or if we perform the variation in the usual way (integrate by parts and use the fact that gi is zero at time t = t; and ¢ = fy), there results adoT ik 3 (B- a8 +e) naoHowever, the quantities 64 can be varied independently, so that det at Hag ~ au 7 k=w1,2...,7 (5.8) If the system is conservative, av a= - 5, (6.9) and (5.8) reduces to (4.2). If some I < p of the generalized forces are conservative while the remaining are not, then (5.8) may be replaced by d eT _ aT , aV He Gu tag 9 RELA. d a0 daT_ aT _ hale 6.10) Hin ~ ag 7 =1+1,1+2,...,p It is important to emphasize that (1) the generalized coordinates can be chosen in a number of ways, but the form of the Lagrange equations remains the same independently of the choice and (2) workless constraints are automatically eliminated, since they do not contribute to the expres- sion for 6W. Consider next the particular case when the kinetic energy is a function of the generalized coordinates q: and the generalized velocities 4 and not explicitly of the time. This will occur when the displacement vectors are functions of the generalized coordinates only, i.e. 12 Also, the potential energy V is assumed to depend only on the generalized coordinates and not explicitly on the time. Thus the integrand 7 — V in Hamilton’s integral does not depend explicitly on the independent variable ¢. It is therefore permissible to use the first integral of Euler’s equation: M92)» + Gp) 2 T-V— > & rT = const (6.11) mh which may be rewritten in the form - —TH+ V4) Got =H = const (6.12) he ee But 7 is a homogeneous quadratic function of the form a) Dy disdids (5.13) jar i90 Four where by = by (6.14) Then Cer yoet) recy an = Del he + 3g, % But ag _ {1 ifh=j Se (0 fk 5 since the generalized velocity components are independent. Con- sequently, 2 2 = d, bade +) bess ry or from (5.14), ee oa? > bade (5.15) so that 2 ae . oT a ” 55 7 2 2 Dade = 27 (5.16) Substitution into (5.12) yields T+ V =H = const (5.17) Therefore the principle of conservation of energy has been deduced, namely, that during the motion the sum of the kinetic energy and potential energy does not vary with the time. 6 Normal Oscillations This section will give a more detailed analysis of the free vibrations of a conservative system about a position of stable equilibrium. Consider a dynamical system with a finite number of degrees of freedom. It can be shown that the most general motion of such a system which is disturbed from a position of stable equilibrium may be represented as a superposition of independent periodic motions known as principal or normal modes. The number of such special modes of motion is equal to the number ofdegrees of freedom. These particular motions are characterized by the following properties 1. The motion of every particle of the system is simple harmonic. 2. The period and phase of the simple harmonic motion are the same for all the particles. 3. The ratio of the amplitudes of the displacements of the normal modes is determined by solving a homogeneous linear system of equations. 4, The normal periods (and frequencies) are found by solving a deter- minantal equation or by iterative matrix techniques. To illustrate the ideas, let us consider the vibrations of a light elastic string of length L; + L: + La stretched between the two points A and B (Fig. 4.6). Let two particles of mass m, and m, be located at the distances L, and L; from the ends. The particles are assumed to be supported on a smooth horizontal plane, or, equivalently, gravity can be neglected. Initially the particles are at rest and the tension & in the string is constant throughout. The particles are given a small displacement perpendicular to the rest position and then released. It is required to find the resulting oscillations. ‘The configuration at any time t (and somewhat exaggerated) is shown in Fig. 4.6. ‘The changes in length of the three string elements are 2 Als = Lise 1) = 48 ga, 2,8 Since the potential energy V is the tension k multiplied by the increase in length and oe 4e8Te md eB we have —klg?, Ga)? , a Vea3 [% i iy The kinetic energy is T = }(mgi? + mags®), and the Lagrange equa- Fig. 4.6 Vibrations of a light elastic string with two attached masses.92 Four tions (4.2) are mas tk (P+ OH 7) =0 (1) mage +k (52 a +E e =0 In order to simplify the formulas (without masking the concepts) we proceed to the special case m, = m: = m; Ly = Lz = Li = L. Lagrange’s equations become mas + Ms a Ha. — %) 9 (6.2) mi + _ a) , ka _ 9 On examining these equations, it is noted that, when q: = 2 or q1 = —q», the two equations become identical, namely mg + fq =0 (6.3) when qi = gz, and mg + & q=0 (6.4) ifq = —g. These equations are expressions of harmonic motion and may be interpreted as follows: 1. If the two masses are started with equal amplitudes on the same side (q = q), the motion will be harmonic with frequency w: = /k/mL. The center string will then remain horizontal, and the masses, being unable to exert a vertical force, cannot influence each other (Fig. 4.7). 2. If, on the other hand, the two masses are started with equal and opposite amplitudes g: = —q2, the motion is again harmonic with fre- quency «2 = ~/3k/mL but the two masses will remain out of phase with each other (Fig. 4.7). Note: These are the only two starting conditions for which the motion is harmonic, as can be shown by examining the original equations. Let us return to the simultaneous equations (6.2) and a more direct procedure for determining the normal modes of vibration. Introduce the quantity k Meo (6.5)Fig. 4.7 Normal modes of vibrations. so that equations (6.2) become hi + Wo — Mg: = 0 di — Xn + 2N%Q2 = 0 6.6) Let us try solutions of the form qi = @ cos (nt + e) 2 = B cos (nt + 6) 6.7) where a, 8, n, and e are constants, Substitution of (6.7) into (6.6) yields the homogeneous linear system: a(—n? + 2n) — Br? = 0 aan =o (6.8) ‘This system possesses a nontrivial solution if and only if the determinant of the coefficients nt — 2° oN PNG n® — 2n? or, by expansion, nt — 4dr? + 34 = 0 (6.9) which has the solutions mt= mt = 3 (6.10) For each 7%, the ralio of 6 to a is readily found. For n* = ni? = 24, B/a = 1; and for n? = nz? = 30%, 6/a = —1. Thus qi = = Cos (At + 41) q: = a cos (At + e1) (6.1) represents one set of normal modes. Similarly, a 2 08 (4/3 At + €2) 6.12) q2 = —a2 cos (4/3 M+ «)94 Four is a second set of normal modes. The complete or general solution is obtained by superposition, namely qi = c4 Cos (ME + €1) + a cos (4/3 Mt + €) gz = a1 cos (At + €1) — a2 cos (4/3 Mt + &) This is the most general solution, because it contains four constants a1, a2, 4, €2 which may be adapted to satisfy arbitrary initial conditions corre- sponding to given positions and velocities of the two particles at t = 0. (6.13)FIVE Deformable Bodies— Theory of Elasticity The first portion of this chapter will be devoted to the geometrical concept of the relative deformations of a deformable body. The material of the body under consideration is taken to be a continuum, i.e., the atomistic structure of matter is disregarded. For example, when we introduce the mass density of a body “at a point P,” this means the mass per unit of volume at the point. In other words, if P is surrounded by a portion of the body of mass Am and Volume AV, then the density p of the body at P is defined by p = iim. SP? where it is assumed that the limit exists and is a continuous funetion of position. ‘Therefore the body is replaced by a continuous mathematical model whose geometrical points are identified with the corresponding material points of the body. It is important to note that the range of applicability of the concepts in the sections on strain and the following Secs. 4 to 7 on stress applies equally well to elastic! or even more general inelastic solid bodies. Furthermore, these formulas also apply to nonsolid bodies such as liquids and gases. Therefore, these concepts may be applied, for example, to problems in elasticity, plasticity, aerodynamics, and hydrodynamics. 1A deformed body is called elastic if it possesses the property that it regains its original unloaded shape when the forces causing its deformation are removed. 9596 Five Starting with Sec. 8, our considerations will be restricted to linear elastic bodies, i.e., those for which the stress and strain components are linearly related. Examples of materials which essentially exhibit this linear relation in considerable load-deflection ranges are high-strength aluminum alloy, low-carbon steel, and glass. The logical extension of the basic Hooke’s law, termed the generalized Hooke’s law, is treated in Sec. 9. It is used in all developments of the linear theory of elasticity and is applicable to the most general case of anisotropic elastic bodies. It involves 36 constants, but with the assumption of a strain energy density function the number of elastic constants reduces to 21. Furthermore, by elastic symmetric properties the number of elastic constants may be further reduced, and we have various crystal classes depending upon the assumed levels of symmetry. Finally, when the elastic properties of the body are the same in all directions, the body is elastically isotropic and the number of fundamental elastic constants reduces to 2. The physical significance of the isotropic elastic constants is treated in Sec. 10. Sec- tion 11 is concerned with the intimate connection between the strain energy density function and stress and strain components. The expres- sion for the strain energy is derived for isotropic structural elements. 1 Strain Consider an arbitrary point in a three-dimensional elastic body B (Fig. 5.1) which before deformation has rectangular coordinates (z,y,2). Let: (u,0,10) denote the displacement field of the point at any time t so that u = u(a,y,2), » = v(e,y.2), and w = w(x,y,z). Therefore, after deforma- tion the rectangular coordinates of the point are given by vw =2+u(zyz) y =y + vey.) (Lt) z+ w(zy,2) Q ds Plictu,yty,z+w) Possy2) 8 Fig. 5.1 Vector displacement field for two neighboring points P and Q.Let (ds)* denote the square of the distance between two neighboring points before deformation and (ds’)? denote the corresponding distance after deformation. From (1.1), dx! = de + du = dx + tz da + uy dy + us dz dy! = dy + dv = dy + vez +0, dy +0. dz (1.2) de! = dz + dw = dz + wz de + w, dy + w, de so that (ds')? = (da!)® + (dy')? + (de)? = [dx + uz dz + uy dy + u, dz}? + [dy + v2 da + vy dy + v, dz]? + [dz + ws dx + w, dy + w, dz]? or (ds’)* = (dx)? + (dy)? + (dz)* + Yece da? + Ley dy? + ers de? + ery die dy + eye dy de + ees de dx + ese dx dz + Leys dy dz + 2exy dz dy (1.3) where Us + $(us® + v2? + wz") vy +H (uy? + vy? + w,?) ws + 3(us* + v2? + w,") (1.4) ye = B[Uy + Ve + Ustly + vary + Wetdy] $l. + wy + Uytls + oye + wywr] Gz = ez = Ewe + Ue + Uets + 002 + wwe) We introduce the quantity ¢ defined by _ ds’ —ds -2 = (1.5) and called the strain in the line element originally of length ds. As repre- sented by (1.5), ¢ (which is greater than —1) is the change in distance between the points P and Q (due to the deformation field) divided by the distance between them before deformation. A simple calculation shows that — (as? = (ds)? dst et+ie= (1.6) so that if the initial direction cosines of the line element are (lym,n), then. e+ de? = eral? + end + eadln + eal + eqn? + exmn + earl + eynm + en? (17) Thus from (1.7), in order to calculate e it is sufficient to know the six98 Five coefficients €22, €yy, €s; xy) €ys) €z, Which are expressed in terms of the displacement components according to (1.4). In particular, for a rigid- body displacement, the six quantities are zero. Note that subscripts (x,y,z) connected with (u,v,w) denote partial derivatives (for example, v2 = dv/an). The symmetric matrix (czy = €yz, €ye = €sy Gz = €xz) fez fay Exe ei fy fe Gz Gy Gee is called the strain tensor, since its components obey the tensor law of transformation of a Cartesian tensor of rank 2, as shown in the next section. The components és, éy, and ¢, are readily interpreted physically. If we set 1 = 1, m = 0, n = 0 in (1.7) (corresponding to the case of a linear element which before deformation is parallel to the « axis and is of length ds = dz) then Cz + de2? = tox (1.8) If (1.8) is solved for ez, and noting that e, must be greater than —1, we have uniquely é: = V1 + 22 —1 (1.9) For most engineering applications, e.g. where aluminum, structural steel, etc. are employed, the components of strain are very small in comparison with unity (¢.g., in many applications |e,| < 0.001), so that fez = Ce (1.10) Thus, the component e,, of the strain tensor represents the extension or change in length per unit length of an element originally parallel to the z axis. Similarly, the relative extension of a fiber or element initially parallel to the y and 2 axes may be expressed analogously as oo. (1.11) 6 = V1 Ft 26 — 1 Next we shall deduce an interpretation of the strain components éy €m and éz. ‘To this end, let 0 and 0, be two neighboring points at a small distance r apart. If J, m, n denote the direction cosines of the ele- ment 00; and (6z,éy,6z) the rectangular coordinates of 0; with respect to O, then bx = rl by = rm (1.12) éz = rnIf (w,w) are the z, y, and z components of displacement of point 0, the displacement of point 01 is (by Taylor's expansion) ty = Ut Us bx + tty by + Ue Be v1 =v + ve bx + dy by + ve bz (1.18) wy = w+ wz bx + w, by + w, be where terms of second and higher order in 6z, dy, and 6z are neglected because the quantities 6x, éy, and éz are numerically very small in com- parison with unity. Owing to the deformation, let 0 move to O while Ox is displaced to O;. If r(1 + e) is the distance between the two points after deformation and if (l1,m1,n1) are the direction cosines of segment joining 0; and Oy, then, using (1.13), y= etme _ At us) te + wy by + te be Oe Eile Oe 7 + €) 7d +6) _ dy tur —o _ vede t+ (1 +0) by + & ere a ee aa _ be + wi —w_ vedo + wy dy ++) & cen at) Substitution of (1.14) into (1.12) yields (+e) = (+ us)l + uym + un (1 + e)mi = vel + (1 + 0,)m + vn (1.15) (1 + e)m: = wil + wym + (1 + w,)n Similarly, if we consider a second element of length 7 through the same point with direction cosines (I/,m’,n’), there results (+e) = (+ wa)! + uym! + un (Lt e’)mi, = vd! + (1 + 0,)m! + on’ (1.16) (tent = wal! + wym! + (1 t+ w,)n’ If each equation in (1.15) is multiplied by the corresponding equation in (1.16) and addition is performed, then, utilizing the definitions (1.4), (1+ e)(L +e!) + mami, + many) = Ul + mm! + mn’ + Qlezall! + eymm! + enn! + xy(Im! + I'm) + eye(mn! + m’n) + ee(nl’ + In’)| (1.17) In particular, if the two elements of lengths r and r’ are orthogonal, then W' + mm! + nn’ = 0, and denoting @ as the angle between the elements after deformation (so that cos @ = lil; + mym; + ninj), we have G+ 6)0 +e’) cos 6 _ 2 eeall! + eymm! + enn! + en(lm’ + Um) + ee(mn! + min) + ee(nl! + n'l) (1.18)100 Five Specializing still further, we choose l=1,m=0,n=0 and Jl =0,m' =1,n'=0 so that the originally orthogonal segments are parallel to the x and y directions, respectively. Substitution into (1.18) gives A+ OG +e) e080 19) The angle ¥ = 5 — 0 is known as the shearing strain. If we note that the three quantities — 0, ¢, and e’ are generally very small in magnitude 2 in comparison with 1, then Y= eay (1.20) Again, the other quantities ¢,, and «2 may be interpreted analogously as approximately one-half the shearing strain of segments originally parallel to the y and z axes and the z and z axes, respectively. 2 Transformation of Strain Components—Principal Axes Next we seek the transformation of components of strain at a point from one Cartesian coordinate system to another. Let (l,m,n) and (Um! ,n!) be the direction cosines of an element with respect to the two systems, namely, (z,y,z) and (z’,y’,2’). Also, we define the direction cosines by the following table: 2 y z a 4 my, m y ly ms Mm 2 L m na 2 where, for example, 1s = cos (z,2'). From (1.7), since ¢ + 5 is inde- pendent of the particular Cartesian system, eed? + eqn? + ean? + eal + 2eumn + eanl = eva (U!)? + eval’)? + ere(n!)? + eryl’m! + 2eyym'n! + ern! (2.1) But L= bl + bam! + lin’ m = ml! + mam! + msn! (2.2) n= nal! + nym! +n! Substituting (2.2) into the left member of (2.1) and noting that I’, m’, and‘n’ are arbitrary within the constraint condition (’)? + (m’)? + (n’)? = 1, there results eet = Ly%€ee + matey + tatece + Qlimr€sy + 2mamreye + Balers eyy = Unters + matey + natece + Lamatay + Qmanaeye + 2aleee eve = [sez + matey + mae + Qamseey + mame: + WMalseee cory = Lalates + mitmatyy + Matacer + (Lama + Lams)ery (23) + (mana + mumi)eys + (mile + Melrose ates + MeMaeyy + Nase + (lems + lam2)exy + (mans + mana)eys + (Mala + Nsla)ere eer = Uslites + mamrey + Natiers + (Lym + lims)exy + (mani + mins)eye + (mali + mils)ece It may be recognized that the strain components are Cartesian tensors of rank two because of the form of the transformation. This becomes more apparent when indicial notation is used. From the expressions for the transformation of the strain components it is a simple matter to determine the direction for which the direct com- ponents of strain are extremized. These are the directions of principal strains. From (1.9) it is clear that maximizing ¢. is equivalent to maximizing the strain or relative deformation e.. Let 2’ denote the required direction which extremizes evs as given by the first formula of (2.3). The direction cosines lh, m1, m1 are connected by the equation L? +m? +n2—1=0 (2.4) Thus the Lagrange method of undetermined multipliers may be used. To this end we form the function eye A = exp — ely? + mye +1? — 1) (2.5) where e is the Lagrangian constant multiplier, and then set oH _ oH _ ay Lh Performance of this differentiation results in (cz — Oli + exymi + exxtrr = 0 eed + (Ew — 6a + thy = 0 (2.7) €zal1 + €y2m1 + (Ce — €)m1 = 0 The system (2.7) has the trivial solution], = m, = ni = 0, which must be rejected because it contradicts the constraint condition (2.4). For non- trivial solutions, it is necessary and sufficient that the determinant of its 1 Detailed treatment of indicial notation and the elements of Cartesian tensors is given in Appendix A. For further applications of indicial notation, the reader is referred to Chap. 6, starting with Sec. 3.102 Five coefficients be zero: tes € bay fy fw 0 (2.8) fae ye Expansion of the determinant yields the following cubic equation for the determination of the principal strains é-he+he-Ih=0 (2.9) where Ty = eee + Gy + Ge Ta = tasty + €yyts + €ss€ez — Exy? = ey? — Ges? (2.10) fez xy xs Ts= ley Gv & cs Gye as Equation (2.9) is a cubic equation and therefore must possess at least one real root. Furthermore, because of the symmetry of the determinant (2.7), it is known that all roots must be real. Denote these roots as , €2, and ¢. Since the values of the principal strains must be independent of the particular Cartesian coordinate system, the coefficients I;, Is, and Is which enter into (2.9) must be invariant quantities, ie, have the same values for any rotated coordinate system. The quantities 1, Is, and I; are designated as the first, second, and third invariants of the strain tensor. Associated with each principal strain there is a definite direction found from the solution of system (2.7) (which determines the ratios of the direc- tion cosines Js, m1, m1) and the constraint condition ly? + m1? + mi? = 1. It will now be established that the three principal directions are mutually orthogonal. Let ¢1, ¢2, es be the principal strains and (11,m1,n1), (Inman), and (ls,ms,ns) the direction cosines of their respective directions. Then from (2.7), corresponding to ¢ and e2: ely = xaly + cay + eta (2.11) ey = eal + ey + et (2.110) ext = zal + eet + eat (2.110) and eale = eeala + enya + €xxtta (2.12a) exttty = exyla + equa + eta (2.126) eat, = ela + ea + esta (2.12¢) If both sides of (2.11) are multiplied by Js, (2.118) by ma, and (2.11¢)by nz and the resulting equations are summed, we have eallale + mame + mina] = eselile + eyymime + examine + exy(male + mals) + eye(nm2 + nym) + ee(lime + lam) (2.182) Similarly, multiplication of (2.12a) by l, (2.12b) by ms, end (2.12c) by m and adding yields eq{lile + myms + nine] = exalila + eyymyme + een in2 + exy(tmale + mals) + eye(name + mam) + ece(lime + lena) (2.180) Subtracting (2.13b) from (2.13a), there results (4: — €)[lile + mym2 + nine] = 0 (2.14) In general «: * es, which implies that le + mymz + rum = 0 (2.15) and proves the condition of orthogonality. (If 1 = ¢2, then all directions in the plane of the two principal axes can be shown to be principal directions and orthogonality may still be satisfied by selecting directions which are perpendicular.) Ina similar manner, lils + mms + mits = 0 is readily established, which proves that the directions of the three prin- cipal strains are mutually perpendicular to each other. It can also be shown that if the principal strains are such that © > @ > e, then «: is the absolute maximum direct strain and ¢; the minimum, while ¢ is neither a maximum or minimum. Suppose now that the 2’, y’, and 2’ axes correspond to principal direc- tions, so that eve, ¢vy, and éve are principal direct strains. We shall now establish that the corresponding shearing strains ey, «', e’ ate 7eT0. From (2.3) we have eey = lallress + Mr€sy + Mees] + mallreay + Mey + Maeve] + malliere + mitye + Miés] (2.16) Replace the expressions in brackets by their equivalent expressions according to (2.7): Lafels] + mafems] + nafena} e{lils + mim2 + mina} = 0 ery due to the orthogonality condition for principal strains. In a completely analogous fashion we have eys = ¢ = 0. This means that there is no change in the right angles between the principal 2’, y’, and z’ axes as a result of the deformation. Consider an element of a strained medium which before deformation occupies a volume V. Let the final deformed volume be Vp. The volu-104 Five metric strain (or cubical dilatation or simply the dilatation) is defined as _Vo-V a- 2 (2.17) In particular, let the volume element V be in the form of an infinitesimal rectangular parallelepiped with its edges in the direction of the principal axes of strain. Upon deformation, this element becomes again a rec- tangular parallelepiped. Let ¢1, ex, es denote the strains of the elements in the principal directions. Then Vo = (1 + es) + e)(1 + es) 0 so that from (2.17) the dilatation is given by A= e1 + €2 + €3 + e1€2 + €2€3 + e301 + e1€0€3 (2.18) 3 Theory of Small Displacements The formulas developed in the first two sections apply to so-called states of finitestrain. They (or special modifications of them) are required to accommodate large displacements of thin plates, shells, and columns for both nonbuckling and buckling problems. However, many problems in the theory of elasticity and plasticity involve very small (or infini- tesimal) deformations, so that squares or products of quantities are neg- lected in, comparison with the first powers. ‘Thus, for example, the strain- displacement relations (1.4) are replaced by the linear approximations Gee = Ur Gy = Wy Gee = We fyz = E(u, + 02) 40. + wy) fez = Css = (Ws + Us) (3.1) Also, no distinction is made between the e’s and the e quantities. Therefore, for example, Eq. (1.8) is replaced by e: = és. This means that the term e,*/2 is neglected in comparison with e:. Similarly, the dilatation A as given by Eq. (2.18) for finite strains reduces to Atatatasateta e+ 2447, G2) Therefore the dilatation is simply the divergence of the displacement field. Now, given a displacement field with differentiable (and therefore continuous) components u(2,y,2), v(z,y,2), w(2,y,2) it is a simple matter to calculate the corresponding strain components by substitution into Eq. (8.1). If, however, we try to solve the converse problem of first selectingthe six components of strain in an arbitrary fashion, it soon becomes apparent that in general an associated displacement field cannot be found. It can be shown that for simply connected bodies a single-valued displace- ment field which satisfies (3.1) exists if and only if the strain components satisfy the following diJerential equations of compatibility. Sey , O%er _ 2é%exy oy? da dy Or, Dey _ 26%eys ay? | dz? ~ dy dz dere y Pew _ 26%ere et * Ox? ~ Oe ac (3.3) Bere | Oye _ ere , ry " dy dz" Ox? ~ dz dy * dz dx Dey 1 Ieee _ 9% Weve dzdx' dy? dydz dx dy op Mery Pee 5 See a2 dean" Oy a It may be recalled that a region of space Ris said to be simply connected, or, more briefly, simple, if and only if every closed curve lying in R can be shrunk continuously to a point without passing outside the boundaries of the region. Thus the region between two concentric spheres is simple, while the interior of a torus (ie., a doughnut) is multiply connected. Similarly, the interior of a rectangle is a planar simply connected region, while the annular region between two concentric circles is multiply connected (more specifically, doubly connected). It should be noted that additional conditions to (3.3) must be satisfied for multiply connected bodies. Example 1: An elastic solid of arbitrary shape is subjected to a temperature distribution of the form T = T(z,y,2), where aye is a Cartesian system of coordinates. If each element in the body is free to expand (i.e., the body is not stressed), the strain components are fez = Gy = Gee = aT (3.4) fy = G2 = Gz = 0 where « is the linear coefficient of thermal expansion and is assumed to be constant. Determine the necessary form of the temperature distribution. Solution: If (3.4) is substituted into (3.3), there résults Tz + Ty =0 Ty + Tee Ti t+ Tiz =0 Tye = Tez = Ty =0 (3.5)106 Five which implies that Pex = Ty = Te = Ty = Tye = Ter = 0 Since all the second partial derivatives with respect to the space variables are zero, the temperature field is at most a linear function of z, y, and z, that is, T = A + Be + Cy + Dz, where A, B, C, D are constants. It may be shown conversely that if the temperature distribution in a simply or multiply connected body is linear in rectangular z, y, 2 coordinates and there are no surface forces, body forces, or surfaces of displacement dis- continuity, then there will be no stresses throughout the body. Example 2: A procedure for determining the displacement field is to first determine the strain field and then integrate the strain-displacement relations. The rigid-body displacements are usually disregarded in these calculations. Therefore, it is of interest to determine the form of the rigid-body displacements corresponding to the linear strain-displacement relations (3.1). The rigidity requirements imply that all components of strain are zero. From ez = ¢y = é = 0, it follows that u, v, w are independent of =, y, z, respectively. Therefore u = uoly,z) v = v0(Z,2) w = wolz,y) If these functional forms are substituted into the expressions for the shear strains, we have AuUo , Wo _ Oo | IW _ wo Spe a eae du oz If the first equation is differentiated partially with respect to y and it is noted that vo does not depend on y, then a : a =0 @ Similarly if the third equation is differentiated with respect to z, there results au, i a 79 wi) ‘The most general solution of (i) and (ii) is the bilinear form tug = Aye + ay + bz + Cy Similarly, v9 = Bex + dx +ez+Cr wo = Cay + gx + hy + Cswhere all coefficients are arbitrary constants thus far. Substitution into dup , a Gy t Ge 7 0 yields (Az +a) + Bze+d=0 or A +B=Oanda+d=0. Similarly, B+C=0 and e+h=0 C+A=0 and g+b=0 so that Therefore, U = Up = we — wy + Cr v = 9 = wt — oy + Cr w= Wo = wry — wot + Cs If we let CH=CitGjt+Ck r=aityjtck © = ai + wx + wk and S = ui + oj + uk there results the vector relation S=C+exr (3.6) Thus the displacement vector S consists of a translation C of the body as a whole and a rotation. Of course, this agrees with the fact that the most general displacement of a rigid body is of this form. 4 Stress The concept of stress arises from an analysis of the internal forces which act in a body. Suppose that we have a three-dimensional solid body subjected to the action of surface and body forces. Select. a point 0 in its interior and pass a plane through O (Fig. 5.2). Consider the forces exerted by one part of the body B: on the remainder Bz with the plane as part of its boundary. Surround the point O with a closed carve in the plane of area AA, The forces acting on AA may be replaced by a resultant force AP at O and « couple AC. Let n be a unit outward normal to AA.108 Five PLANE SECTION THROUGH POINT O EXTERNAL FORCES ON BODY: Fig. 5.2 External and internal forces acting on and within a body, respectively. By definition, the average force per unit area or average traction at O on an area with normal n is then AP/AA and the average moment is AC/AA. Let AA tend to zero, so that O remains inside AA and all its linear dimensions tend to zero. Assume that the limit exists independ- ently of the manner in which AA approaches zero. The traction vector or stress vector is defined by _ AP Tew jim ad oY Furthermore, it is usually assumed that |AC| is of the same order of magni- tude as |AP| multiplied by a representative moment arm whose length lies within the shrinking area AA. Thus while AP/AA does not in general become zero in the limit as AA approaches zero, it will be assumed that Ac aan BA 9 (2) In effect, we are not accommodating point moments, which could occur when an external magnetic field acts on a magnetized medium. There are several facts concerning the stress vector T, that are impor- tant to note, namely, 1. T,, in general, is not parallel to the normal n. 2. T, is not only a function of position but also depends upon the orientation of the planar element (or equivalently, the normal n). 3. If T, refers to the action of By on Bz, then we denote T_, as theaction of B:on By. But, since action and reaction are equal and opposite (Newton's third law), T..=—-T, (4.3) In particular, for the tractions T., T,, T; on elements perpendicular to the z, y, and z directions, respectively, we have (4.4) Since the traction T,, is a vector, it may be resolved into two com- ponents, one perpendicular to the planar clement and the other in the plane. We may write Tr = ont + ons (4.5) where n and s are unit vectors perpendicular to and in the planar element, respectively. The quantity cnn is called the normal or direct stress com- ponent and is positive when tensile and negative if compressive (see Fig. 5.3). Also, ons is referred to as the component of shear, since it tends to cause relative sliding of one element along a contiguous element. Alternatively, T, may be resolved into its Cartesian components. If i, j, k denote three mutually orthogonal unit vectors in the 2, y, 2 direc- tions, respectively, then let = Ons + Omyj + onk 46) and, in particular, if m = i, j, and k in turn, there results Gack + onj + oak Oval + onj + ok (4.7) Te = Gest + nj + ook 2 A normal stress component is tensile if it is directed away from the surface upon which it acts. If it is directed toward the surface, it is compressive. Sanh Ty Fig. 5.3 Resolution of traction vec- Tm8 tor into orthogonal normal and shear components. AA110 Five 5.4 Cartesian components of traction vectors acting on a rec- tangular parallelepiped showing the positive directions. The nine Cartesian scalar components gzz, gz, - . - , dz: are shown in Fig. 5.4. There ozs, oy, and gz, are the normal or direct-stress com- ponents and oxy, ozs, Gye) Ty; Jzz; Jxy are the shearing components. The meaning of the subscripts zz, zy, ... , zz is that the first subscript indicates the coordinate axis normal to the element of area upon which the traction acts and the second subscript indicates the direction of the component of the traction vector. Therefore, for example, o., represents the shearing component in the z direction of the traction vector T, acting on a planar element z = constant. Also, a tensile normal component implies that it is positive and a compressive normal component implies that it is negative. 5 Relation among Traction Vectors at a Point In the preceding section it is shown that the traction vector varies not only with position but also with orientation of the element of area AA. We now seek a useful relation among the traction vectors at a point from the condition of force equilibrium. Let OEDG be an elemental tetrahedron and h the distance from 0 to triangle DEG (Fig. 5.5). Also, let AA, AAs, Ay, AA, represents the areas of the triangles DEG, OEG, ODG, and ODE, respectively. The unit normal n from the origin to the plane triangular element is assumed to be oriented with direction cosines J, m, n referred to the Cartesian system so that n= Li + mj + nk (5.1)and AA, =1AA AAy=mAA AA, = AA (5.2) There are in general two kinds of forces acting on the tetrahedron: 1. Surface forces acting on its boundary. The resultants are given by AAP,, AA,P_., Ady P_,, 44, P_, on the respective faces DEG, OBG, ODG, and ODE. 2. Body forces distributed over the volume of the body such as gravi- tational forces. Let F denote the average body force per unit of volume (i.e., average body force density). The vector equation of force equilibrium is therefore P_.AA, + P_, AA, + P_,AA, + P,AA + 3h AA F =0 (5.3) Division by AA and application of (5.2) and (4.4) yields IP. + mP, + nP, = P, — $hF (6.4) If we let h— 0, then P, > T., P, > T,, P. > T:, Ps Ts, the tractions at O (assuming these tractions are continuous functions of position). In the limit there results T, = IT. + mT, + nT, (5.5) If we then substitute (5.5) into (4.6) and (4.7) and equate coefficients of i, j, k, One = laze + Moye + Nose Ony = lasy + mo yy + Roy (5.6) One = laze + Moye + Noe DAP Fig. 5.5 Surface forces acting on Aa,P, elemental tetrahedron of body.uz Five Finally, from (4.5), (5.5) and (4.7) oon = n+ T, = n+ (IT, + mT, + nT.) = UGzel + 24m + o2n) + m(gyal + oyym + oyn) + (rel + 0.4m + oun) or Onn = Poze + Moy + Woe + lm(Gxy + oy2) + mn(oys + Gey) + MUGre + oz2) (5.7) Also, let the unit vector s in (4.5) be given by s=li+mj+n'k (6.8) From (4.5), (5.5) and (4.7) we have s-T, = s- (IT. + mT, + nT.) Is+ (Gzsi + onj + onk) $ms+ (6 + n§ + ok) $.ns+ (Casi + onj + onsk) Finally, from (5.8) there results Tne one = Woe + mm'oyy + nN’ + Im'oe, t+ Umoy, + mn!oy: + non (5.9) + alors + n'los Equations (5.7) and (5.9) give the normal and shear components of traction in terms of the nine Cartesian components of stress. 6 Equilibrium Equations Consider a body which is in a deformed state of equilibrium due to the action of both body and surface forces (Fig. 5.6). For equilibrium, it is necessary that the vector resultant of all the forces and moments about an arbitrary origin be zero. Let T, denote the traction at the surface S, F the body force per unit of volume (assumed to be continuous over V), s TqdS Fig. 5.6 Body and surface forces acting on deformed body.and r the position vector from the arbitrary origin O to an element of volume dV and surface dS. Then for force and moment equilibrium f,T»a8 + [Fav =0 6.1) and far xTadS + [rx Fav =0 (6.2) We shall now show that (6.1) yields the stress equations of equilibrium and (6.2) will lead to the equality of cross shears, that is on = oan Substitution of (5.5) into (6.1) results in f, (Ts + mT, +n, as + [Fav =0 (6.3) and if the divergence theorem of the vector calculus is applied (assuming that the traction T,, possesses continuous first derivatives), then eT at, cat, _ (E+ +B +r)av-0 64) Equation (6.4) must hold for all volumes, i.e., for any proper subvolume of the body. Therefore, since the integrand is a continuous function of a ot: 4 Oy at + Ee aT. a which is a vector equation of equilibrium. Finally, substitution of (6.5) into (4.7) and equating coefficients of i, j, and k to zero gives the corresponding scalar form dors +F=0 (6.5) oye dee ae + Ses Met = 0 Sout 4 Sn 4, Son ta +F,=0 Co) i i: se : ie a where F = iF. +3f, + KP, 6.2) Next, consider the consequence of the vanishing of the resultant moment produced by the body and surface forces. From (6.2) and (5.5), Jar x (Ts + mT, +n) aS + [pr x Fav =0 (68) Application of Gauss’ pa theorem to the first integral implies that he @xT) +5 OxT) +2 @xT) +rxFlav =0 (6.9)14 Five If we carry out the differentiation of (6.9), there results Jprx «(E49 eons one + 2 av “i xT + ExT, +ExT,)av =o (6.10) However, the integrand of the first vote is zero in accordance with (6.5), and furthermore, since (6.10) must hold for arbitrary volume V, the second a is zero, that is, ext +S ene kenoe (6.11) or or But 5. =i, 5, hs 5s oe = k, and consequently (6.11) reduces to ixTe+)xTtkxT.=0 (6.12) Insertion of (6.12) into (4.7) and equating the coefficients of i, j, and k to zero yields the equality of cross shears, that is Oxy = Fyz Oye = eye = ee (6.13) Therefore, the state of stress in a body is determined by the six com- ponents of stress o2z, Cy) Gs2) Tay = Sys) Sys = Oxy Ore = Tze. More generally, it is a simple matter to show that Ons = Ten (6.14) All that is required is the interchange of the roles of n and s in addition to (m,n) and (U'ym’,n’) in (5.9). Then apply (5.9) and (6.13) to yield (6.14). 7 Principal Stress at a Point In Secs. 5 and 6 the relationships between the stress components at a point were obtained. In particular, the normal component of stress nn as given by (5.7) was expressed in terms of the Cartesian components and the direction cosines (/,m,n) of the normal with respect to the zyz system of coordinates. It is seen that with oz) = oyz, oy = Ory, and Gzz = xz the relation (5.7) is completely analogous to the corresponding formulas for the transformation of strain such as given by the first of Eqs. (2.3). Thus (omitting the unnecessary details) it follows that there must be three mutually orthogonal directions for which the normal components of stress are stationary. ‘These directions are known as the principal directions, and the corresponding stresses are termed the prin- cipal stresses. The normal stress along two of the axes will actually be the largest and smallest normal stresses, while along the third axis wewill have neither a maximum or minimum. Let «, o2, os denote the principal normal stresses at the point. The planes normal to the prin- cipal directions are called the principal planes of stress. The principal stresses o are determined from the determinantal equation [analogous to (2.8)] Gx Oxy on oy Owe Oe | =0 (7.1) Ox on on — 8 Expansion of the determinant results in the cubic equation o — Jy? + Ja — Js = 0 (7.2) where Ji, J2, and Js are stress invariants defined by Ji = G22 + Oy + on Se = Czy + Oyee + on22 — Tny' Tze Try Tze Js=|ou ow ow Oss Tye Oe 2 2 = Ons" (7.3) Oy Furthermore, the three roots of (7.2) must be real. Finally, it is very easy to establish that the shear components of stress are zero on the principal planes, i.e., the traction vector on a principal plane is per- pendicular to the plane. For from (5.5) and (4.7), Ts = om is equivalent to the system of equations (a = li + mj + nk) Uox2 — 0) + mow + Now = 0 losy + My — 0) + oye loss + Moy + N(Ge2 — 0) = 0 (7.4) This is the system of equations which, for nontrivial [m,n (recall 1? + m? + n? = 1), leads to the determinantal equation (7.1) in exactly the same manner as (2.7) implied (2.8). Thus the proof is complete. Note that the stress invariants are related to the principal stresses by the relations Jv=atortos Jy = a102 + 0105 + om (7.5) Js = 10203 8 Hooke’s Law The relationships among the strain components in Secs. 1 to 3 and then the stress components in Secs. 4 to 7 are applicable to liquids and gases as well as solids provided that the bodies may be represented as a con-116 Five tinuum. The elastic properties of materials have thus far not entered into the relations deduced up to this point. However, the interrelation between the components of stress and strain is based primarily on experimen- tal data varying from material to ma- terial. This and ensuing sections of this chapter are concerned with Linear elastic solid bodies. The general meaning of the word “elastic” is that there is an absence of any permanent deformation when a specimen is loaded and then unloaded. The term “linearly elastic” implies that the relationship between “load” and “deflection” is a linear one. For example, the load may actually bea Fig. 5.7 Shaft twisted by twist- twisting moment while the deflection is ing moment or torque. the corresponding rotation. Thus if a steel shaft is subjected to a twisting couple as shown in Fig. 5.7, a typical response curve of moment: vs. rotation will be as shown in Fig. 5.8. It is noted that the twisting moment M is directly proportional to the angle of twist 0 until the pro- portional limit is reached. This defines the linear elastic response range. FRACTURE TWISTING MOMENT M INELASTIC NONLINEAR RESPONSE PROPORTIONAL LIMIT LINEAR ELASTIC RESPONSE ° ANGLE OF TWIST 6 Fig. 5.8 Representative moment—angular displacement curve.The greatest moment that can be applied without producing a per- manent angle of twist upon unloading is called the elastic limit. For steel the elastic limit very nearly coincides with the proportional limit, and the distinction between the two may be disregarded. The range of moment beyond the elastic limit defines the inelastic nonlinear response. Note that a point in the loading process is reached after which the loading actually reduces in the neighborhood of material fracture. Similarly, consider a slender steel bar of initial cross-sectional area A subjested to an increasing tensile load P (Fig. 5.9). If the load is assumed to be distributed uniformly over the cross section and if small changes are neglected in the cross-sectional dimensions, then the longitudinal stress is given by oz: = P/A, while the longitudinal strain is ee = AL/L (the change in length per unit of length in the rod), Hooke in 1678 postulated that the axial stress 0. is proportional to the strain; that is Ow = Bess (8.1) where E is the constant of proportionality and is known as Young’s modulus. Equation (8.1) is known as Hooke’s law. It is essentially valid for many materials until the proportional limit is reached. (Some materials exhibit only an approximately linear stress-strain curve.) 9 Generalized Hooke’s Law In our discussion of a rod under a simple uniaxial tension, we neglected the change in the cross-sectional dimensions. When a rubber band is stretched, it is easily verified that the dimensions of the cross section are decreasing. In other words, an axial stress causes strains not only in the same direction but also in the transverse direction. A logical gener- alization of Hooke’s law (due to Cauchy) is to assume that at each point of the medium the stress components are linear functions of all the strain components and conversely. The resulting law is called the generalized Hooke’s law and is expressed mathematically (for an isothermal state) as G22 = Caress + Crreyy + Crates + 2Crueys + 2Crseee + 2Creeay Caress + Cortyy + Crates + 2C240y2 + 2Cas€es + 2Coceey (9.1) Fig. 5.9 Slender rod under axial tension.118 Five The 36 coefficients Cu, Cis, . . . , Cesin (9.1) in general may be functions of position. If, however, they are constants which are independent of position, the medium is defined to be elastically homogeneous. We shall from this point restrict our considerations to elastically homogeneous systems and denote the 36 constants Cu, Cuz, . . . » Ces as the elastic constants of the material. Also, a body is said to be isotropic if its elastic properties are the same in all directions. What this amounts to is that, for an isotropic body, the stress-strain relations referred to any orthogonal Oz'y'z' coordinates must have the same coefficients as in (9.1). In other words, for arbitrary orientation of the z’y'z’ Cartesian coordinate system with respect to the zyz system the following will be the stress-strain law: ate! = Creare + Cireyry + Crseerer H+ Waser + Wrsesrer + WCreesry yy = Crrezrar + Cortyys + Crsterer + WCreeyer + Woaseere + Woreesry (9.2) ary = Corearr + Cortyy + Costes H+ 2Coseyrer + Wester + Woeeery where, it must be emphasized, the coefficients in (9.2) must be identical with the coefficients in (9.1). It may be proved that for an isotropic body the 36 constants are not independent of each other. In fact, the number of elastic constants reduces to only 2. This may be established by using the invariance of the stress-strain relations with respect to various covering operations such as (1) a rotation about a line or axis and (2) a reflection in a plane. ‘The detailed derivation of the reduction from the 36 constants to 2 constants will not be given in this book. Instead, the following facts will be noted: 1. It can be shown that with the assumption of the existence of a strain energy function that Cy = Cx, ij, and i,j=1,2,...,6. This reduces the number of elastic constants to 21 in the generalized Hooke’s law and applies to the most general case of an anisotropic elastie body. Crystals possessing no symmetry at all define the triclinic system. 2. If the medium is symmetric with respect to a plane, then the con- dition of the invariance of C;; results in a reduction to 13 constants. ‘This is called the monoclinic class. 3. For a material with three mutually orthogonal planes of elastic symmetry, the number of elastic constants is reduced to 9. Such 1 For detailed discussions the reader is referred to (1) I. S. Sokolnikoff, “Mathe- matical Theory of Elasticity,” 2d ed., pp. 56-70, McGraw-Hill Book Company, New York, 1956 and (2) A. E. H. Love, “A Treatise on the Mathematical Theory of Elasticity,” 4th ed., chap. 6, Dover Publications, Inc., New York, 1944.materials as wood possess these properties and are said to be orthotropic. 4. Further symmetry requirements of independence of the stress-strain laws with respect to one pair of the mutually orthogonal reference axes result in the hexagonal erystal class with 6 elastic constants (for example, quartz). If, in addition, the same holds true for a second pair of axes, we have the so-called face-centered or body-centered cubic class exemplified by crystals of the majority of pure metals—iron, copper, nickel, and aluminum, for example. It is readily shown that the number of elastic constants reduces to 3 for this class. While it is true that for a crystalline body the elastic properties depend on orientation of axes, metals on a macroscopic scale may be considered as isotropic. That is because many metals have a sufficiently small and random-oriented crystal grain structure. For an isotropic body the number of elastic constants reduces to 2 and the stress-strain law may be expressed as Gx2 =O + Qwese — oy = OT Bey oe = NOT Byles Oye = Queye — Ore = Were ay = Aplery (9.3) where 8 = exe + éy + er is called the volumetric strain or dilatation and the constants \ and u are called Lamé constants. Conversely, (9.3) may be solved for the strain components in terms of the stress components: — Ot Wore _ x “WGN 2a) — Bal -F Bay Cow + en) CeO ee HBX + 2y) ~ Bu(BN -F Bp) Me 2) = Ot Won 2 » “* ~ WGNF Bu) ~ TaGr Fo—) Cs + em) ou ne wi ee ae 10 Significance of Elastic Constants for Isotropic Materials Very important physical insight may be gained by considering the response of elastic isotropic materials to particular loadings which result in states of pure shear, uniaxial tension, and hydrostatic compression, respectively. Pure Shear. We may characterize this state by the requirements x = % = T = constant throughout the body while all the rest of the stress components are zero. From (9.4) it follows that correspondingly there is only one nonzero strain component éy = é2 = T/2u. Thus in120 Five accordance with Sec. 1, we may interpret » = 7'/2ez, as the shear stress 7’ divided by the decrease in angle between linear elements originally parallel to the z and y axes (Fig. 5.10). The quantity wis called the shear ‘modulus and in the usual stress analysis is denoted by the letter G, that is, aS eB (10.1) and is a constant which assumes different values for different materials. Uniaxial Stress Field. Next let us return to the slender cylindrical bar of Fig. 5.9 subjected to the action of longitudinal forces applied at the ends. If the load is taken to be uniformly distributed over the cross section, then the stress field will be o,, = constant while the other com- ponents are zero. Then from (9.4), the direct strains are Obes = (Bd + 2p) = Mas fa = © Dux + 2u) (10.2) But for the uniaxial stress distribution, we define Young’s modulus EB by the formula E = ¢z:/éz. Thus from (10.2), y= He = 2) moe (10.3) This relates i, #, and Z, but \ has no physical significance. Elimination of ozz from Eqs. (10.2) results in Deze aL) = Te oo SS 2H) Fig. 5.10 Pure shear—rectangular element ABCD before deformation becomes a parallelogram AB'C'D T after deformation with the samewhere » is defined by x Secret 10.4 FH) oe ‘The constant », called Poisson’s ratio, is the negative of the ratio of lateral strain to longitudinal strain when subjected to uniaxial longitudinal stress. Hydrostatic Compression. In the case of » uniform hydrostatic pressure of amount P, we have =-P Oy = Os = oy = 0 Osx = Oy = Then from the stress-strain relations eee Gee ioe oe eae ce Skat or solving for P, P= ere | where K is defined by Ke= (10.5) The constant K, called the bulk modulus of elasticity (or modulus of volume expansion), is the ratio of applied hydrostatic pressure to the change in volume per unit of volume. By means of (10.1), (10.3), (10.4), and (10.5) it is seen that any one of the elastic constants E, », @, K, d, « may be expressed in terms of any other two constants, i.e., only two constants are independent. If quantities E and y are selected as the two basic elastic constants, Eqs. (9.4) and (9.3) assume the more usual form es = Fae — 1 + 60)) en = Flew — 1a + on] fe = flow — (ou + on)] ys ui (10.6) l+p fee = ee122 Five or, conversely, E ee = ET Ml — eee + olen + —. wise [0 = rey + r( eee + €ee)] E 2 = BID Ih = vee + v(ces + ew)] ey v)(1 — 2v) (10.7) ee ee eri E oe re ace oe = TG From their physical interpretations, the elastic constants u or @, B, and K must be positive quantities. But from w= 3K(1 — 2»)|2(1 +») it follows that the values of Poisson’s ratio are bracketed by —1 < » <}. The following table gives average values of H and u obtained experi- mentally for some materials in this linear clastic range. Note that the Walues of ¢ arc deduced from y= z —1, although they may also be determined experimentally. ‘The values of Z and u are given in millions of pounds per square inch, and y is nondimensional. Material Aluminum alloy 0.313 Carbon steels 0.283 Cast iron 0.269 Glass 0.250 Magnesium 0.354 Malleable iron 0.269 ‘Titanium A very important consequence of the stress-strain law (10.6) or its equiv- alent (10.7) is that, for isotropic elastic materials, the principal azes of stress and strain coincide. This follows from the simple argument that if the 2, y, and z axes are chosen along the principal axes of stress, then Ory = Oye = G2 = 0. But from (10.6), this implies that ex, = ¢y2 = é2 = 0. Therefore, the z, y, and z axes are also principal axes of strain. This coincidence of the principal axes of stress and strain for isotropic bodies may be employed to derive (in reverse) the stress-strain relations using theconcept of superposition. If we imagine a rectangular parallelepiped element of the body subject to the simultaneous action of normal stresses in the three coordinate directions, then from the definition of Z and », = G->(B+9) Similar results hold for ¢,, and és, showing agreement with (10.6). This approach is not as satisfying from a theoretical viewpoint. However, it does appeal strongly to our physical intuition. 11 Strain Energy Consider a body acted upon by surface tractions T, and body forces F per unit volume. If the change in the body heat is considered negligible, and assuming that the body is in a statically deformed state, then the work done by the external forces must equal the strain energy stored in the material. Let U denote the strain energy of the material and Us denote the strain energy per unit of volume, so that u=f, Ueav (uy) where the integration is performed over the volume of the body. Suppose that the displacement field u undergoes a change from u to u + du (a virtual displacement compatible with the constraints). The corresponding work done by the surface tractions is [, T,+ du dS, where the integration is taken over the surface of the body. Similarly, the work done by the body force F per unit of volume is i: F- éudv. Thus, the balance of energy requires that J, Ueav = [,T.- duds +f, F-ouav = [, (Ts + mT, + nT) suds + f,P-suaVv (11.2) where (l,m,n) are the direction cosines of the unit outward normal n to the surface. Application of the divergence theorem to the first integral on the right side of (11.2) yields a a J, avvav = f,[Z ow + Z cry ou) a +56 av + [,F-suav124 Five which upon expanding becomes aT , aT, , aT J, eUeav = i (E+ Fe SE yw). amar yu du du +f, [te 040 oe Ela (11.3) But oT. Ox oT, me oy +F=0 from (6.5), so that |, sUeav = ffm +138 ET, 62 ]av (14) From (4.7) and the expressions (3.1) for the strains in terms of displace- ments we have fi, 8U0aV = f, loxsbeee + Obey + ober + Qondery + 2oysbeys + e262] dV (11.5) If it is assumed that the strain energy density is a function of the strain components, that is, Uo = Uol€szs€vvy6s2s€zys€ves€zz) then Uo 4. 4 Ue eee aus jf, oUeav = i [se Bee + Ge Sew + + Fee] av (11.6) The right sides of (11.5) and (11.6) may be equated. This equality must hold for arbitrary volumes and, furthermore, des, dé, . . - , dex are independent variables to within equilibrium and compatibility require- ments. Therefore, we deduce the equalities , = We _aUy _ ao at es aasae ess ges aU _ Uo aUo a. Cele open ee Ogee odes Finally, the significant expression for the strain energy density Uo can be obtained by expressing it as a homogeneous quadratic function of the components of strain (as a trial function). Then, utilization of (11.7)and the linear stress-strain law for isotropic bodies (9.3) yields Uo = ATH esata? He) + Menten + tant + ete) A Qu (exy? + eve? + €rs*) (11.8) which may be rewritten in the form = xe Hh Mlean* + yy? + eee” + lay? + 2eys® + ees”) (11.9) Since 4 > 0, that is, » > 0 for any physical material, it follows that the strain energy density Up is a positive definite form in the strain components. Substituting Eqs. (10.6) into (11.9) and using the expressions for \ and yz in terms of H and », the function Uy is expressible in terms of the stress components as Uo = iy least + om? + 0n8 — 29(Cyytee + orsbee + ons0) $20 + Gon? + on2 + on)] (11.10) and this is also a positive definite form in the stress components, since vcd Also, Uo may be expressed in terms of both stresses and strains as Uo = Bosstes + ytiy + Gestes + 2ozyery + yeeys + rates] (11.11) Finally, if the expression (11.10) is differentiated partially with respect to each of the stress components and use is made of the linear stress- strain relations (10.6), we have sth ay | _ ats "Gee lon | A ae aU> aU. Us ae elgg ee oaze | oe a ageSix Energy Principles, Methods, and Applications 1 Theorem of Minimum Potential Energy In the preceding chapter we employed the principle [expressed by (11.2)] that the virtual work performed by the body and surface forces during an arbitrary variation in displacements is equal to the increment of strain energy. Since the volume V is fixed and the surface tractions T, and body forces F do not vary with arbitrary virtual displacements (com- patible with constraints of system), (11.2) of Chap. 5 may be rewritten in the form a[f, vear - [,T.-uas - f,F-uav] =0 (a. ‘The expression in brackets is defined to be the potential energy V of an elastic continuum, that is v= f, Usa — [,T.-uas — f,F-ua? (1.2) and (1.1) becomes simply sv =0 (1.3)This states that the potential energy of an elastic system in equilibrium must be stationary with respect to all displacements satisfying the given boundary conditions. In fact, more strongly, not only must the potential energy be stationary but it can be shown that it actually assumes a minimum value when the displacement field is that of the stable equilib- rium state. Conversely, it can be established that if the potential energy of an elastic system in equilibrium is an absolute minimum for a displace- ment field u which also satisfies the boundary conditions, then the equa- tions of equilibrium must also be satisfied by the field. It is this converse form that is most prevalent in applications. To illustrate the application of the theorem we shall consider two examples. Example 1: Deflections of an Elastic String. The principle of minimum potential energy will now be applied to determine the ordinary differential equation for the small deflections of a flexible string. A string (Fig. 6.1) is shown loaded by a distributed vertical force of intensity q(x). The string is of length L and is under a large tensile load S$ when q(2) is applied. We neglect the small change in the tension due to the distributed force q(z). It is further assumed that the string is perfectly flexible, so that the tensile force is directed along the tangent. ‘The strain energy in the string is composed of two parts: (1) U’ due to the strain from the original tensile force and (2) U’" due to the deflection of the string resulting from the transverse loading. The total strain energy is given by U = U’ + U”. Let us find an expression for U”. ‘The increase in strain energy due to the deflection is obtained by multi- plying S by the increase in length due to the deflection. If y denotes the string deflection, ur = 8 fo (as— ae) = 8 [2 [ahi + (#) - Jae Small deflections are taken to mean that (dy/dz)*<< 1. Then according Fig. 6.1 Vertically loaded string. (a) String under tension S with no (o) vertical load. (b) Same string with vertical load added.128 Six to the binomial expansion, ae a rr S pz (ay i U! = 8 f7[1+3 (2) + ee dz (i) and the total strain energy is given by _ oS pe (au? i u=uU+5f (ya (i) Substitution of (ii) into (1.1) for the ease of no body forces results in cee ae L ue ae ale +5), (2) ax — [ aye] = 0 (iii) where — [(" ay dx is the potential energy of the distributed load q(a). But U’ is constant for all variations in displacement, so that 5U’ = 0. Thus (iii) reduces to [5 (Be) a - faves] -0 (v) Therefore the Euler equation corresponding to the integrand S (ay)? 3 (Hy - 9 FY + g(a) =0 (4) is which is the differential equation of the vertically loaded string. Its solution may be obtained by repeated integration where constants of integration are determined from the boundary restraints (0) = y(L) = 0. For example, in the case of a uniform load q = go = constant, y= gee — 2) (1.8) which is the well-known parabolic deflection shape. Example 2: Beam Column. The beam column represents a general- ization of the ordinary beam problem in that it is a beam subjected to both axial and transverse forces. We shall be concerned with the deriva- tion of the differential equation of equilibrium and boundary conditions using variational notation (Chap. 3, Sec. 5) in conjunction with the principle of minimum potential energy. Also, a solution will be developed for the deflections from which the bending moment and stress may be readily calculated. Figure 6.2 shows the problem to be analyzed. The applied loadings are compressive axial end forces P, distributed transverse loading q(z),and two concentrated end moments My and M,—all taken to be acting in a principal plane of bending. Forces and moments are taken to be positive as displayed, and the beam ends are simply supported for bending. ‘The strain energy of bending is U=sf Bry’) ae @ while the potential energy of the distributed load q(2) is Vex — fl wae Gi) Similarly, the potential energy of the moment Mo is Vu, = —Moly')2-0 Gi) while the corresponding quantity at the other end is Vu, = Mily')enn (iv) (Note that the rotation in the direction of Mz atz = Lis (—y’)ea1-) For the compressive end loadings, the potential energy is — P times the axial shortening. The latter quantity is found from the condition that, during bending, the axial beam length does not change. This is a char- acteristic of a ‘small-deflection” analysis. If A denotes the displacement of the end x = L in the direction of the load P, then [PO vIF Uae = is required. But V1 + (y’)? = 1 + 3(y’), so that fru +40 a = 7 or Lats [wars Fig. 6.2 Loads applied to beam column.130 Six Finally, since A is negligible in comparison with L, asa fl? wre sd [wae w) Therefore, Ve= 2 [i wae (vi) The principle of stationary potential energy requires that 6(U + Va + Mo+ Mi + Vr) =0 (vii) or from (i) to (iv) and (vi) oe : {ft [ere — Fw — a] de + Mada — Maly} = 0 (vii) If the variation is performed, we have 1 fh [EIy" by” — Py! by’ — q by] dx + M1 (3y')2nr, — Mo(5y')zno = 0 (ix) Application of integration by parts twice yields Jo Blu" by" de = (Bly ay’ — f= Ely") by de 4 1 = tery’ ov — [2 ty") ov | ae + [C5 Ev) yar a d miyt = Wly" iy! + ff F5 ly") by de @) while i : d Jo Py oy ax = Pry’ by - [2 ey) ay ae aa, : = — [PE Pv) iy ae xi) since 6y = Oateach end. Substitution of (x) and (xi) into (ix) results in LT da? = d - kl fen + Zen ~ a] ee + (Mz + Ely'!) by! — (Mo + Ely}’) by’ = 0 (xii)Therefore, the required differential equation is Oe qa Ely") + (Py) —9=0 as) and the boundary conditions are (Mi + Ely) =0 (My + BIW = 0 eo and yruy=0 (1.8) If q = Mo = Mz = 0 (that is, if there are no transverse loading or end moments), then we have the eigenvalue problem for pin-ended columns 2 # ly") + Py’ =0 (1.9) subject to the boundary restraints Yo =O (y)zan = 0 (1.10) and y0)=0 y(L) =0 (ay) If, in particular, the bending rigidity ET is a constant, then the problem is further reduced to the Euler buckling problem. Also, if P = 0 (no axial end loads) the differential equation (1.6) becomes the beam equation, £ Gy) -¢=0 (1.12) Solution of the Boundary-value Problem. We now return to the problem of solving the boundary-value problem (1.6) to (1.8) for the case when the bending rigidity and the applied transverse loading q are constant. Equation (1.6) then becomes y+ hey! = (1.13) where P woe (1.14) Let y/’ = u, so that (1.13) becomes ul + ku = H (1.18) ‘The solution of (1.15) is composed of any particular solution added to the132 Six complete solution of the homogeneous equation w” + ku = 0. Therefore, rae ae q u = A’ cos ke + BY sin kx + Ses (1.16) where A’ and B’ are arbitrary constants. Note that since cos ke and sin kx are linearly independent functions," the solution is complete. Two successive integrations of u yield = i ge? y = A cos ke + Bein ke + sia; + Ce +D (1.17) where A = —A'/k? and B = —B’/k? are new arbitrary constants, while C and D are constants of integration. These quantities are now to be determined from the boundary conditions. The deflection is zero at x = 0, (yo =0=A+D (a) Also [y’"Je0 = — Me, But y” = —KA cos kx — WB sin ke + pha therefore = ge Me fh eee aeRT, or A-ae (é + Me) (b) From (a) and (b) substituted into (1.17), ’- TH (™ + 4) (cos ke — 1) + Bsinkr + 52 4 Ce (1.18) and y= ~ By (Mo + fh) eos he ~ Bit sin be + phy BI B PBI ‘The satisfaction of the conditions [yJeur = 0 and [y"lez = — M¥ yiela B and C, so that after some straightforward simplification y= per (Mery B) (eos ke — 0 io ae | - [a -M+% by, - »]z (1.19) The stresses may now be calculated in a routine manner. ‘The formal definition of linear independence of n functions is given in Chap. 7, See. 1.Final Comments 1. For completeness, it may be remarked that beam columns supported differently from the foregoing development can be accommodated with equal facility. If, for example, the member is built in, the boundary condition is on the slopes rather than the moments. 2. The above development is not valid if the transverse loads are con- centrated at points along the beam length. This is due to the finite jumps in the shearing force and therefore the third derivatives at such points. The differential equation (1.6) is correct in between the loads, and this problem can be treated by integrating in steps. Example 3: Three-dimensional Displacement Equations of Elas- ticity. An isotropic body obeys Hooke’s law relating stress and strain components. It is in equilibrium under the action of body forees per unit of volume (F.,F,,F.) and surface tractions (oa2,¢n,0x2). It is required to find the small displacement differential equations of equilibrium for (w,»,w) by application of the principle of stationary potential energy. Before we proceed to the solution it should be remarked that the following technique is utilized frequently for derivation of equations in structural mechanics. The potential energy V from its definition and (11.8) of Chap. 5 is given by Ve I if [G + ) (eae? + ean? + east) + Nest + frites + Gentes) + 2ulea? +e + ee) — Pau + Fw + Fw] de dy dz — ff nat + ony0 + onetd) dS (1.20) is Next, let us vary only wu, that is, 6» = 6 = 0. Thus Us bere = Suz = (5U)z vy bey = ov, =0 fe = Ws bes = bw, = ty +02 buy _ (5u) oe ea, = She = Oude Us + Wy Ge bee 2 wee BEM gg, = ue The surface tractions enter only into the boundary conditions and do not affect the equations of equilibrium. Therefore, the surface integral may be disregarded for our purposes. Setting the variation of V equal to zero134 Siz results in If f {G + +) 2us( bux + M(vy + 2)(Bu)s} + wl(ty + v2) (5u)y + (te ++ we) (3u),] — Pe au de dy az + boundary terms = 0 Integration of terms involving (éu)., (6u)y, and (5u), by parts and changing signs yields Jf] (0 + 2eduae + Nye + tee) + lt + Om) 7 + pluses + w22] + Fe} du dx dy dz + boundary conditions = 0 Since u is arbitrary, the differential equation is found by setting the expression in brackets equal to zero. Rearrangement of terms yields aVut +n) 2+ 7 =0 where : 2 du, ov , dw A = dilatation = ie + ay + 7 and VU = ez + Uyy + Use ‘The other two equations of equilibrium may be determined by cyclical permutation of the variables (u,v,w) and (z,y,z). Thus we have, by eliminating \ from (10.4) of Chap. 5, the coupled system of equations Oh oye (1 = 29) Vu + 5 + a F,=0 oA , 1-2: (= 29) ve + FF + FF, =0 (1.21) oA, 1-2 Ge 2?) Vea iz F,=0 2 Hamilton’s Principle for an Elastic Continuum In Chap. 4, Sec. 3, Hamilton’s principle for a discrete system was derived for a conservative force field. It is natural to anticipate that this concept may be extended to a continuum. Our objective is to establish this result and then to show some examples illustrating its use in problems of motion.