University of Central Florida


Academic Press College Division (Harcourt Brace Jovanovich, Publishers) Orlando San Diego San Francisco New York Toronto MontrealiSydney Tokyo·· Sao Paulo

This textbook is an introductory treatment of parlial dilTerenl i(11equations and boundary value problems. it is



intended for students in mathematics,

and physical sciences who are in either 'their junior or senior year. In some cases it may also serve as an
introductory textbook For beginning

graduate students. necessary 10 study the basic calculus course in ordinary cquariorrs.

The background this book includes sequence and a first differential

The principal objective of the book is 10 pre cnt rnn I of the standard solution techniques used in solving: boundary value problems, using n blend of'theory and basic physical examples. For instance, a simple physical example is often presented first and used for rnoiivation in developing the accompanying theory. 1 have tried to present lhe mathematical theory in a careful, manner without the abstract concepts of functional analysis. Most ofthe major theorems are staled, but some without proof. When a proof is omitted, a reference for the proof is usually cited, While there are numerous. physical examples used thruughnut the [eXI, no special background beyond a basic physics course should be necessary 'to understand them or 10 follow the derivations of [heir governing equations. The material is organized so thai eigenvalue problems and Fourier series are covered before solving partial difIerential equuticns by any general

The method of separation which is the primary 1001 for solving partial diffclrential equations, is one of the few techniques thai works on 0. wide variety of problems. For '[his rea SOli it provides a unifying framework in. which to solve problems involving the heal, wave. and potential equations. III addition to the mel hod separation of variables. the methods of Fourier integrals and integral transrorms arc used where appropriate: the eigenfunclion expansion method is used to solve nenhornegeneous equalions: and some approximation and numericaltechniques tire discu sed.


To aid the student in studying the material in this book, ] have included over 100 worked-cut problems and more than 11O()exercises. The exercise sets contain a blend of routine problems. more difficult problems, and some thai extendthe theory and applications beyondthe exposition. More challenging problems, or those used to extend the theory or applications,are marked with a star (.). Answers are provided for all odd-numbered problems. at the end of the bOOK. .
Tile text is divided intothree parts, with the bulk of Ine material preserued in Part H. Part [ consists 'of Chapters I and 2, which deal only wilh ordinary differential equations. The first chapter provides a comprehensive trearment ooundClry value problems. including (he important study of Sturm-Liouville problems. while the Green's function









for borh inili:ll value ami boundary value problems is presented in Chapter 2. Chapters 3 lhro:ugh 8 make up Part II of the text. The basic concepts of partlal differential equations arc presented in Chapter 3. Chapter 4 begins with Fourier trigonometric series, moves on 10 generalized Fourier series. and ends with Fourier integral represernations. The integral transforms of Laplace :.mdFourier.arelntroduced in Chapter 5, and Chapters 6 through 8 provide a systematic coverage of the heal, wave, and potenualequatiens, respectively. Part III contains Chrlplers II and 10. which deal with special. functions [primarily Legendre polynomials and Bessel functions) and their role in so.lving partial differential equations in coordinate systems other than rectangular. As supplementary material. a

semester or one-term course. To help in ihi s regard, I have marked optional sections wilh an open box {OJ. These sections CIIIl be easily bypassed For a shorter course without any loss of continuity. Naturatty, there arc other seclions and chapters that can be omitted,
depending all the instructor. A stand-

ard one-semester or one-term course can be developed [rom l11aterial in
Chapters J" 4, 6 through 8, ;mo
pUJ'IS of I, 5,9. and 10.. In some cases Chapter I need not be covered except for a short review 011 certain selected topics before Chapter 3.

Laplace transforms and their inv rses is provided in Appendlx A. with a sirnilar table of Fourier transfornts in Appendix. B, Because of the frequent need thrnughout the text to discus - the solutions or linear systems of equations. n short review of Cramer's rule arul Its role in developing the theory of simultaneous equations is presented in Appendix C.
Since 197-1. a twc-course sequence in bound.II~.\ alue problems has been offered ill the Uni\'el'''ll}' ofCcnlral

or some

of the most common

Florida to a mix of mathematics: eng •. ncering, nnd physics students. The lecture notes developed for that equeuce of courses have finally le~110 this textbouk. However, [IHWC lJ led tl) arrange the material S(1 (hal it is equally \~ell-suited for a more rradinonal one-

[ owe many thanks to all the students who sat through my course over 'the last ten years while Ithi,s text was being written, Their patience, understanding. and helpful Sll!;P::Sli(lI1sarc \'cry much appreCIUlo:!u.1 wishnlso [0 thank my colleagues ond rric-m.h. Frank L Sail' mann rind Rubert C. i:lfigham. \\ ho made numerous suggestions while lC:lChing from early drufrs of (he 11I;II1Uscript, The dirticull task of Checking <1m.! correcting the answers was done by Lisa M. Potchen. I arn extremely grateful (0 Philip Crooke. Vanderbih UniYcrsily; Edward ['\'1. Landesman, University of Califurnia. Sama Cruz: Rolarul DiFran 0, University of rhe Pacific. Herman Gollwitzer, Drexel U[li"eJ'~J(y; and Monty J. Strauss. Texas Tech University; who served as
reviewers and provided many helpful


'Finally, I wish


my appreciation to We~ky La" tun, j\..!ath~nHllli:S Editor. and the entire preducuon staff of Acndernic Press for their .:ITOrlS in producing Ihis lext






_-_ .._-.----_._-

._-_. __



_ ..

_._---_._. __ ._-










Copyright \0 1986 bY,A<:.1dcmic All rights reserved ..". t. No part or [his Dul'lIcnlli~n led incl in any J) and publisher.



I ...

Press, Inc.


.reproduced or transmitor mechanical. information storage nr.rTn"c<i~.~ in writing from the

- ---.--.------.

-.----- -- - --- _-





Chapter 1 Boundary Value and Eigenvalue Problems 3
1.1 Introduction 4 1.2 Steady-State Heat Conduction: An Example 6 1.3 Boundary Value Problems 9 1.3.1 Homogeneous DEs 10 1.3.2 Nonhomogeneous DEs 14 104 Eigenvalue Problems 20 1.5 Physical Applications 31 1.5.1 Deflections of an Elastic String 31 1.5.2 Rotating String 35 1.5.3 Displacement Curve of an Elastic Beam 36 1.504 Buckling of a Long Column 40 1.6 Sturm-Liouville Theory 46 1.6.1 Properties of a Symmetric Operator 50 1.6.2 Regular Sturm-Liouville Systems 54 1.6.3 Periodic Sturm-Liouville Systems 55 1.604 Singular Sturm-Liouville Systems 57 1.7 Approximation Methods: Weighted Residuals 62 I.7.1 Nonhomogeneous Equations 63 1.7.2 Eigenvalue Problems 70




Chapter 2

The Method of Green's Functions


Introduction 77 2.1.1 A Physical Example 78 2.2 Initial Value Problems 79 2.2.1 One-Sided Green's Function 81 2.3 Impulse Functions 88 02.3.1 Physical Interpretation and a More General Green's Function 91 . 2.4 Boundary Value Problems 94 2.4. L . Green's Function 95 . 2.4.2 'Physical Interpretation of the Green 's Function 104 .Bibliography 110 .



Chapter 3 . An Introduction to Partial Differential Equations I 15


Introduction 116 3.1.1 Properly-Posed Problems 117 3.2 Classification of Second-Order PDEs in Two Variables 119 3.2.1 Constant-Coefficient Equations 120 3.2.2 . D' Alernbert's Solution of the Wave Equation 122 3.3 Equations of Mathematical Physics 127 3.3.1 Separation of Variables 128 03.3.2 Solutions by ODE Methods 131

Introduction 134 Fourier Series of Periodic Functions 135 4.2.1 Cosine and Sine Series 140 4.3 Convergence of the Series 144 4.3.1 Uniform Convergence 150 04.3.2 Operations on Fourier Series 152 04.4 A Proof of Pointwise Convergence 159 4.4.1 Bessel's Inequality and Riemann's Theorem 161 4.4.2 Convergence at a Point of Continuity 162 4.5 General Periods and Intervals 164 4.5.1 Nonperiodic Functions Over Finite Intervals 166 4.5.2 Half-Range Expansions 167 4.6 Generalized Fourier Series 171 4.6.1 Nonhomogeneous Boundary Value Problems 175 04.6.2 Bilinear Formula for the Green's Function 181 4.7 Fourier Integral Representations 185 4.7.1 Cosine and Sine Integral Representations 189

Chapter 4
4.1 4.2

Fourier Series and Fourier Integrals



Chapter 5 Integral Transforms of Laplace and Fourier '193' Introduction 193 Laplace Transform 194 ' 5.2.1 Evaluating Transforms 197 5.3 Operational Properties and Inverse Transforms 204 5.3.1 Inverse Laplace Transforms 206 5.3.2 Convolution Theorem 209 5.4 Initial Value Problems 214 05.4.1 One-Sided Green's Function 216 5.5 Fourier Transform Pairs .220 5.5.1 Fourier Cosine and Sine Transforms 221 5.1 5.2



... :-.•.. t.. ~ ••..•.









5.5.2 Evaluating Transforms 223 5.5.3 Operational Properties 225 5.504 Relation Between Fourier and Laplace Transforms 5.6 Boundary Value Problems 233 05.6.1 Green's Function 236


. 228

Chapter 6 The Heat Equation 239
Introduction 239 6.1.1 One-Dimensional Model 240 6.2 Heat Flow in a Rod with No Sources 242 6.2.1 Rod with Ends Held at Zero Temperature 243 6.2.2 Rod with Ends Held at Constant Temperature 247 6.2.3 Rod with Ends Impervious to Heat 251 6.2.4 Convective Heat Transfer at One Endpoint 253 6.2.5 Summary and Discussion 256 6.3 Nonhomogeneous Problems 261 6.3.1 Nonhomogeneous Terms Independent of Time 261 6.3.2 Eigenfunction Expansion Method 263 06.3.3 Time-Varying End Conditions 266 604 Infinite Rods and Fourier Integrals 270 6.4.1 A Semi-Infinite Rod 270 604.2 An Infinite Rod 273 6.5 Method of Integral Transforms 275 6.5.1 Laplace Transform 276 6.5.2 Temperatures in a Rod of Finite Length 279 6.5.3 Fourier Transforms 280 6.504 Nonhomogeneous Equations 283 6.6 Numerical Method: Finite Differences 289 6.6.1 Finite-Difference Operators 290 6.6.2 Steady-State Problems 292 6.6.3 Heat Equation 295 6.1

o o

Chapter 7 The Wave Equation 303
7.1 7.2 7.3 Introduction 303 7.1.1 Equation of Motion for a Vibrating String 304 Free Motions of a Vibrating String 305 7.2:1 Validity of the Formal Solution 310 Forced Motions of a Vibrating String 316 7.3.1 Sinusoidal Forcing Functions 316 7.3.2 Eigenfunction Expansion Method 320 7.3.3 Nonhomogeneous Boundary Conditions 324 Infinite Domains and Fourier Integrals 326 704.1 A Semi-Infinite String 327 Method of Integral Transforms 328 7.5.1 Laplace Transform." 328 7.5.2 Fourier Transforms 330 Numerical Method: Finite Differences 334·

07.4 07.5





Chapter 8 The Potential Equation 338
Introduction 338 Dirichlet and Neumann Problems 339 8.2.1 Rectangular Domains 340 8.3 Circular Domains: Polar Coordinates 346 8.3.1 Dirichlet Problem for a Disk 347 8.3.2 Neumann Problem for a Disk 352 8.3.3 Potential Problems in a Circular Annulus 352 8.4 Properties of Harmonic Functions 357 804.1 Uniqueness of the Dirichlet Problem 358 804.2 Maximum-Minimum Principle and Stability 359 8.4.3 Uniqueness of the Neumann Problem 360 8.5 Unbounded Domains 364 8.5.1 Method of Fourier Integrals 364 8.5.2 Poisson Integral Formula for the Half-Plane 366 8.5.3 Fourier Transform Method 367 08.6 Approximation Methods 371 8.6.1 Weighted Residuals 371 8.6.2 Finite Differences 374 Bibliography 379 8.1 8.2


Chapter 9 Special Functions 383
9.1 9.2 Introduction 383 Legendre's Equation 384 9.2.1 Legendre Functions of the Second Kind 387 9.2.2 Special Properties of Pn(x) 389 9.3 Gamma Function 394 09.3.1 Additional Properties 396 9.3.2 Digamrna Function 399 9.4 Bessel Functions 403 904.1 Bessel Functions of the Second Kind 407 9.4.2 Series Expansions for the Functions Yn(x) 408 9.4.3 Basic Identities and Recurrence Formulas 409 9.5 Eigenvalue Problems Associated with Bessel Functions 09.5.1 Equations Reducible to Bessel's Equation 419 9.6 Modified Bessel Functions 425 9.7 Generalized Fourier Series. 429 9.7.1 Fourier-Legendre Series 430 9.7.2 Fourier-Bessel Series 433






__ ..


.. _-_ ..

__ _--_ .. _ _-------.--------_
.. ..

... _.-,...---..._. '/'.



Chapter 10 Problems in Several Dimensions
Introduction 438 10.1.1 Two-Dimensional Heat Equation 439 010.1.2 Two-Dimensional Wave Equation 441 10.2 Rectangular Domains 443 10.2. I Double Fourier Series 446 10.3 Circular Domains: Bessel Functions 451 10.3.1 Radial Symmetric Vibrating Membrane 451 10.3.2 General Vibrating Membrane 455 lOA Cylindrical Domains: Bessel Functions 460 1004.1 Radial Symmetric Potential Problem 460 1004.2 A More General Potential Problem 463 10.5 Spherical Domains: Legendre Polynomials 467 10.5.1 Electric Potential Due to a Sphere 468 10.6 Poisson's Equation 473 10.6.1 Rectangular Domains 474 Bibliography 477 10.1






Appendix A Appendix B Appendix C Index

Table of Laplace Transforms Table of Fourier Transforms Cramer's Rule

478 481

Answers to Selected Odd-Numbered Problems


who made numerous suggestions while teaching from early drafts of the manuscript.'" -_ -r! . a two-course sequence in boundary value problems has been offered at the University of Central Florida to a mix of mathematics. The integral transforms of Laplace and Fourier are introduced in Chapter 5. Potchen. The basic concepts of partial differential equations are presented in Chapter 3.'_'?""~i!~. and ends with Fourier integral representations. To help in this regard. which deal with special functions (primarily Legendre polynomials and Bessel functions) and their role in solving partial differential equations in coordinate systems other than rectangular. Santa Cruz. The lecture notesdeveloped for that sequence of courses have finally led to this textbook. University of California. I am extremely grateful to Philip Crooke. who served as reviewers and provided many helpful comments. Chapter 4 begins with Fourier trigonometric series. I wish to express my appreciation to Wesley Lawton. Frank L. However. there are other sections and chapters that can be omitted. These sections can be easily bypassed for a shorter course without any loss of continuity.. Roland DiFranco. Vanderbilt University. Herman Gollwitzer. Mathematics Editor. Because of the frequent need throughout the text to discuss the solutions of linear systems of equations. Landesman. Part III contains Chapters 9 and 10.9. a short review of Cramer's rule and its role in developing the theory of simultaneous equations is presented in Appendix C.~-'~_"-:--:'. and parts of 1. and 10. wave. Brigham. semester or one-term course. Salzmann and Robert C. respectively. depending on the instructor.5. and Monty J. Strauss. Their patience. and the entire production staff of Academic Press for their efforts in producing this text. and Chapters 6 through 8 provide a systematic coverage of the heat. understanding. xii PREFACE for both initial value and boundary value problems is presented in Chapter 2. and physics students.In some cases Chapter I need not be covered except for a short review on certain selected topics before Chapter 3. 6 through 8. As supplementary material. I have marked optional sections with an open box (0). with a similar table of Fourier transforms in Appendix B. and potential equations. Chapters 3 through 8 make up Part II of the text. University of the Pacific. I wish also to thank my colleagues and friends. I have tried to arrange the material so that it is equally well-suited for a more traditional one- .'. Edward M. and helpful suggestions are very much appreciated. The difficult task of checking and correcting the answers was done by Lisa M. engineering. Drexel University. Naturally. a table of some of the most common Laplace transforms and their inverses is provided in Appendix A. 4. moves on to generalized Fourier series. Texas Tech University. A standard one-semester or one-term course can be developed from material in Chapters 3. lowe many thanks to all the students who sat through my course over the last ten years while this text was being written. Finally. Since 1974.

---.--....:' ~--. ~- ELEMENTARY PARTIAL DIFFERENTIAL EQUATIONS with BOUNDARY VALUE PROBLEMS .-.-----.'-.- .--.ao..' ---------.-------.-. .-- .-..--- .--.


In such problems the auxiliary conditions are usually specified at only two points of the interval. initial value problems are "well-behaved" in that they almost always lead to unique solutions. and SCi the general theory of them is inherently more complicated (and more interesting) than that of initial value problems. If the auxiliary conditions are specified at more than one point on the interval of interest. and we refer to these as two-point boundary value problems. In sharp contrast with initial value problems. Some additional physical applications leading to boundary value problems' are discussed in Section 1. look at the small displacements of an elastic string and an elastic beam. a class of problems for which the solution depends upon a parameter Aoccuring in the equation.5_In particular. We begin our discussion of beundary value problemsin . In general. This situation is further examined in Section 1. we illustrate how boundary value problems can arise naturally in the formulation of a mathematical model of a physical problem. since first-order DEs have only one auxiliary condition and are therefore classified as initial value problems. the same is not true of boundary value problems. both of which are supporting a distributed load. This class of problems is characterized by auxiliary conditions all specified at a single value of the independent variable. some boundary value problems possess nontrivial solutions even when the equation and boundary conditions are all homogeneous.3. Unfortunately.Section l. By deriving the governing DE and describing some possible boundary conditions.CHAPTER 1 BOUNDARY VALUE AND EIGENVALUE PROBLEMS Most of the theoretical discussions and applications presented in a first course in differential equations (DEs) involve initial mille problems. and related eigenvalue we 3 . By appropriately selecting A.k with an introductory application problem involving steady-state heat conduction in a thin rod. It is primarily this latter class of problems that concern us here. an infinite collection of nontrivial solutions (called eigenfunctions) can often be found corresponding to an infinite number of values of A (called eigenvalues).4 in connection with eigenvalue problems. The DEs associated with boundary value problems are therefore at least secondorder. the resulting problem is called a boundary mille problem. We also use this introductory section for motivation in studying the general theory concerning the solutions of boundary value problems in Section 1.

x). at x = XI and x = X2). the number of which is usually two (equal to the order of the DE). 0. Unless the boundary conditions are carefully chosen.)"(XI) = a. Physical problems associated with boundary value problems often suggest the correct type of I 1- . Typical boundary conditions are of the general form aIlY(XI) G2IY(X2) + GI2. and singular Sturm-Liouville systems. (2) The problem of solving (I) subject to the auxiliary conditions (2) is called a boundary value problem.6 + (22)"(X2) = 13. we call them boundary conditions. which has far-reaching consequences. Uniqueness-uuece is at most one solution 3.se e ¥SA ++ IAW4 4 CHAPTER I BOUNDARY VALUE AND EIGENVALUE PROBLEMS problems problem including the famous buckling of Euler. 1. < x < x~. at most. A.. + where the a's. whereas when they are specified at two points (viz. Eristeilce-there is at least one solution 2.1 INTRODUCTION A linear differential equation (DE) of second-order has the form (I) x. it is a useful method for approximating the eigenvalues and eigenfunctions of various SturmLiouville problems. In most applications the unknown function y in (I) must satisfy certain restraints or auxiliary conditions. a boundary value problem may not have a unique solution. Foremost among these properties is the orthogonality of the eigenfunctions. and A~(x) is not identically zero there. . o . periodic.6 we investigate the general properties shared by the eigenvalues and eigenfunctions belonging to all regular. The functions A o(X) .(.6 0. to small changes in the solution. and 13 are constants. The solution is said to be stable if small errors in these measurements lead. and Alx) are called the coefficients of the DE and are generally assumed to be continuous in the interval (x" X~). When these conditions are both specified at a single point we call them initial conditions. The method of weighted residuals is introduced in the final section of the chapter as an approximation technique for solving boundary value problems. ail + ai2 (I~I (/~2 . or in fact any solution at all! We say the mathematical problem is properly-posed (or well-posed) if it satisfies the following requirements: I. In addition to handling nonhomogeneous equations. and small errors are naturally introduced in the measurement process. In Section 1. Stability-the solution depends continuously on the boundary data The last requirement is important in practice since the boundary data are usually measured.

the famous Eulcr formulas relating sines and cosines to exponentials with imaginary arguments. at the ~age of76. number theory. An important feature of a linear. Vertical columns have been used extensively in Greek and Roman structures throughout the centuries. columns were designed according to empirical formulas developed by the architects of the early civilizations. t LEONHARD EULER (1707-1783). The DE (I) is said to be homogeneous if F(x) == 0 in the interval of interest. ami the discovery of the gamma function (Section 9. such empirical formulas give only a crude estimate of the weight a particular column design can support without collapsing. In this chapter we will address mostly homogeneous DEs. Still other variations occur from time to time.r = XI and the second only at the endpoint x = X2. he still continued his work. so is Cy for any constant C. and optics. considered 10 be one of the five greatest mathematicians of all time. by today's standards. astronomy. Some of his main contributions to mathematics are the addition theorem for elliptic integrals. was born in Basel. Switzerland. when Ct = 13 = 0 we say the boundary conditions are homogeneous. since the first is specified only at the endpoint . fluid dymimics. Most boundary conditions arising in practice are specializations of (2). One of the oldest applications involving -a boundary value problem concerns the buckling of a long. = [3. .) (called mixed boundary conditio liS) are sometimes prescribed. when possible. so is the linear combination CrY. Needless to say. and similarly. the analytic treatment of trigonometric functions as numerical ratios. boundary value problems are closely associated with problems of static equilibrium configurations or steady-state phenomena. He made significant contributions in mechanics. (3) aZI)'(x2) + b2Iy(xl) + b22y'(X. subject to both homogeneous and nonhomogeneous boundary conditions. and he developed much of the theory of differential equations. As a measure of safety such columns were often made much larger than necessary to ensure su pport of the structure above them. and )'2 are both solutions of this DE.INTRODUCTION 5 boundary conditions under which the problem is properly posed. More general boundary conditions of the form GII)-(XI) + GIZy'(XI) + bIlY(X2) + b12y'(_'d + aZ2y'(x2) = a. and although lie was blind during the last 17 years of his life. His collected works fillmore than 70 volumes. Euler developed the first truly mathematical model that can rather accurately predict the "critical com- • We also say a linear DE or boundary condition is homogeneous if whenever y is a solution of the DE or boundary condition. and C2• Generally speaking. but these are less common than (2).3). He died on September 18. and for this reason it is helpful to interpret a problem physically. * All other specifications of the DE or boundary conditions are called nonhomogeneous. + CZ)'2 for any choice of constants C. geometry. for guidance in the choice of appropriate boundary conditions. slender rod or column supporting a compressive load. 1783. homogeneous DE is that if y. Prior to the time of Eulert in the eighteenth century. called unmixed or separated boundary conditions.

that we wish to determine.r ::.e. we choose the steady-state conduction of heat in a cylindrical rod of uniform material. and they are mathematically yeO) = Tit y(b) = h at the endpoints of the rod are called described by (4) • We assume the medium surrounding the lateral surface is insulated. which we denote by Y(X) .2 STEADY-STATE HEAT CONDUCTION: AN EXAMPLE Because boundary value problems are closely associated with physical applications. a steady-state temperature distribution will have been reached. the rod is not changing temperature in lime.. i. a column that weighs more than necessary is as unacceptable as one that cannot support its intended load. Such a mathematical model is essential today in the design of buildings. )I----X x=b FIGURE A cylindrical The temperatures prescribed boundary conditions. For simplicity. if . Because the total weight of the structure is often a factor. it may be useful to start our discussion of them by showing how a mathematical model arises out of a given physical problem. * It is this steady-state temperature distribution. ---+-(~) ~( JI~ x=o x x + AX 1...b (see Figure 1.1). Consider a long cylindrical rod or wire coinciding with a portion of the x-axis over 0 ::.--. since slender columns playa fundamental role in the plan of these structures.1 rod. suppose the end of the rod at x = 0 is held at constant temperature TI and the end at x = b is held at constant temperature Tl• Physical intuition or experience suggests that after a sufficiently long period of time. aircraft.. Also. For an illustrative example. 1. To begin.g. It belongs to a particular class of boundary value problems called eigenvalue problems. e. and marine frameworks.'7 6 CHAPTER I BOUNDARY VALUE AND EIGENVALUE PROBLEMS pressive load" that a column can wi thstand before deformation or' 'buckling" takes place. it is assumed that the temperature distribution we seek is the same throughout any cross section of the rod. this development may help motivate some of the mathematical theory in subsequent sections... the temperature inside the rod will no longer change in time. The mathematical model developed by Euler is still being used in many designs. In this way we wiII have an immediate physical interpretation of the terms in the governing DE and of the boundary conditions.. but may changefrom cross section to cross section.

If Q(x) is continuous. Combining terms. the rate at which heat is generated in the slice is AuxF(x). through the lateral surface). if the lateral surface of the rod is insulated and there are no heat sources within the rod.Q(x) . the assumption of heat balance implies that AQ(x) + AuxF(x) = AQ(x + 6. the rate at which heat leaves the slice at (x + ~x) isAQ(x + ~x). (5) which can be rewritten as Q(x + ~x) . we need to relate Q(x) and the temperature y(x).:. which we wish to derive. 0< x < b. 0< x < h. where F is the rate of heat generation per unit volume. . (7) Finally. To do so.STEADY-STATEHEAT CONDUCflON:AN EXAMPLE 7 At other points of the rod the temperature rex) will be governed by a certain DE..\x (6) Observe that the cross-sectional area A drops out. then in the limit . then AQ(x) is the rate at which heat enters the slice at x. we obtain the boundary value problem 0< x < b. (8) where k is a positive constant.x). Q(x) = -ky'(x).. = Fix). In addition. From experimental evidence. For example.. Basically. the heat balance requirement is the amount of heat entering the slice at x plus that generated inside the slice. If A is the cross-sectional area of the rod.:. (The negative sign reflects the fact that heat flows from hotter to cooler regions. Let Q(x) denote the horizontal rate of heat flo II' at the point x.) Hence.\X -> 0 we find that (6) becomes Q'(x) = F(x). measured in units of heat per unit time per unit area. as shown in Figure 1. the substitution of (8) into (7) leads to the governing DE Y" = -k I F(x). we will apply the principle of heat balance (conservation of thermal energy) to a small slice of the rod between x and (x + ~x). if heat enters the slice by means other than through the two faces (e.g.1. * In symbols. yeO) = TI.. Similarly. then F(x) "'" 0 and the governing DE (9) reduces to • This is knownas Fourier's I". (10) Other varieties of governing DE and boundary conditions for this heat conduction problem are also possible. we know that the heat flow rate Q(x) is proportional to the temperature gradient y'(v). equaling the amount of heat leaving the slice at (x + ~x). (9) By combining (4) and (9).

EXAMPLE I Find the steady-state temperature distribution in a rod of unit length whose lateral surface is insulated and ends are maintained at temperatures TI and T2• SOLUTION Because the lateral surface is insulated. according to Newton's law of cooling. ( 15) Still other variations of the DE and boundary conditions are possible. . (14) Finally..-BOUNDARY VALUE b. Ti.----8 CHAPTER I = -----------_. AND EIGENVALUE PROBLEMS (II) y" 0. and we write the boundary conditions as )"(0) = 0. Applying the boundary conditions. from which we deduce CI = TI and C2 = T2 .. if one end of the bar (at x = b) permits loss of heat due to radiation into the surrounding medium while the other end is maintained at constant temperature.e.y'(b) = h (y(b) - TuJ. yeO) = TI. the governing DE is (II) and the problem is therefore characterized by v" = 0.!. Suppose both ends of the rod are insulated instead of being held at constant temperatures. [Note that heat enters the rod if y(x) < To._ yeO) = CI = y(l) = CI + C2 = T2.TI• Hence. . the solution we seek is the linear function mt5Rl __ &$ HieBd . and leaves the rod if y(x) > To.. By integrating the I?E two successive times. the boundary conditions take the form yeO) = TI.] In this case the governing DE becomes . we get • j'. no change in temperature). we write F(x) = -h[y(x) . 0< x < If the lateral surface is not insulated but heat exchanges occur at the lateral surface by convection into the surrounding medium. (13) -'.-----. o <x< I. 0 < x < b.. )"(b) = o. then the rate of change of heat transfer in the rod is known to be proportional to the difference between the temperature in the rod and the temperature of the surrounding medium.--------------:. we obtain the general solution y = CI + C2x. . Thus.To1. ky" -Izy =-/~To. This means there is no heat loss at the endpoints (i.. (12) where lt is a positive constant and To is the temperature of the surrounding medium.

are constants. d' o.. If 1/ = 0 on the inner surface and 1/ = To on the outer surface.. 3.l where I" To. y'(O) = T" y'(I) 1. and for that reason it is instructive to study the general theory of boundary value problems common to all such applications. 4... y(b) = 1'. --. y(b) = T: 2. . y" y(l) = y(l) T: y(l) 0._oo _. __ ---- •• - -_ . y(l) = 0 = 6.1'"' .--. y(O) = O. Express the solution in terms e")12 and sinh x = (c' . 5.! *8. '..Tn).. y'(O) "" 0.y Hint: (e' + O. y(O) = T ..3 BOUNDARY VALUE PROBLEMS The heat conduction problems discussed in the previous section are specializationsof the more general second-order DE A2Cx)y"+ A I(. y. I..~- BOUNDARY VALUE PROBLEMS 9 Although we have confined our attention here to a simple problem of heat conduction.. That is.I"). . x. y(O) . and T.y'(O) = T" y'(O) = 0 = -r. where To is constant.."" 0.. • __ • ~. y" = = 0.--~. = T2 Under what conditions will the following problem have a solution? y" = 0.t)y' + AoCx)y "" F(x). (a) Find the solution y(. the same mathematical problem may arise in twoor more entirely different situations. (b) What solution is obtained in the limit b .Y = . y(O) = T. = 0. Let lI(r) denote the steady-state temperature distribution in a solid bounded by two concentric spheres of radii r = I and r = 2. =0 of hyperbolic functions cosh x = y"' . XI <x < Xz (16) and unmixed boundaryconditions ~ . . The temperature in a cooling fin is governed by the boundary value problem y" = II(y . EXERCISES 1.2 In Problems 1-6..To. determine the steady-state temperature distribution for the given problems.1'(0) = T" -y'(b) = "[y(b) - T..--. y..e-'1/2. ._____ o __ • . *7. find the temperature distribution in the region between the concentric spheres if the governing DE is -d r: (fit) = ' *9. there are many similarities that carry overto other applications.

HI. solutions of . and Cz so that (24) also satisfies the boundary conditions in (23) is our remaining task.H[y] H2[Y] = = (3.6 0). (a~.B. x. and B 2 all satisfy (22) is left to the exercises (see Problem 17 in Exercises 1. let us introduce the differential operator (D = dldx) M = A2(x)D2 + A. First. we introduce BIIY] B2IY] OIlY(X. we can express (16) and (17) more compactly B2[Y] (20) I I In terms of these operators. (25) from which we deduce the values ('(BZlY2] ""-I3B.1 HOMOGENEOUS DEs When F(x) == 0 in (21). Verifying that AI.) 022y'(X2) = n.)'(X2) + + 0'2Y'(X.: . are linearly independent of solutions is given by 0. + oh .(x)D + Ao(x) + AI(x)y' + Ao(x)y. it is useful to introduce simplifying notation. we sayan operator M is linear ifand only if M[CJ(x) + C2g(x)] = CIAi[Jtx)] + C2M[g(x)] (22) where flx) and glx) are given functions and CI and C2 are any constants. + C2H2[Y2] == 13. OI1Y(X. (23) If y. Imposing the prescribed boundary conditions upon the solution function (24) yields the system of linear equations C.b'2] 11 (26) . Determining the values of C. we say the DE is homogeneous and the problem is then characterized by M(y] = 0. Operators M. BI• and B2 are examples of what we call1illcar operators.) 02.B2[y. as (21) I == 13.ly.] + C2HIL}'2]= o .6 O) oi2 .3. In general.-------------------- ----.3). Mlyj = F(x). To discuss the general theory concerning solutions of'( 16) and (17). boundary operators B. < x < X2. (ar. ----- -- --- --~ JO CHAPTER I BOUNDARY VALUE AND EIGENVALUE PROBLEMS i (17) .} 02IY(X2) + 012y'(XI) + 022y'(X2). and C2 are arbitrary constants. 1. andy. and B2 defined by (i8) I ! I:" for which MI)'] == == == A2(x)y" (19) I Similarly. + = (3. the general family (24) where C.] C.

_-------------" BOUNDARY VALUE PROBLEMS II ~81[)'IJ .1 If )'1 and )'2 are linearly independent solutions homogeneous boundary value problem M[y] = of M[y] = 0. which then leads to the indeterminate form 0/0 for both CI and Cz• This indeterminate form for both constants suggests that (26) and (27) are equivalent.mcr's rule. can sometimes be zero. THEOREM 1. In summary.hasnot been rigorcusly verified. • Sec Appendix Cfor a review orCr.. -. . --------. If tl ¥. has a nontrivial solution if and only if Ll.. = 0.aB2[yd 11. if Ll. can be expressed as a multiple of the other through use of either equation. and in this case it always has the solution y = O. 0< x < -rr. = O. is the coefficient determinant BI[)'2] B2(Y2] BI[yd B2[yd I Here we see that the solution of (23) is formally! given by (24). (27) defined by (28) (via Cramer's 11. + y = 0. and then the problem has an infinite number of solutions.-:-------- . then the 0._ . For nontrivial solutions to exist when ex = 13 = 0. the evaluation of the constants C. where the constants CI and C2 are explicitly defined by (26) and (27).----_. Owing to the fact that 11. For instance. Y(7i) = o. and one constant. For example. The resulting problem is then called a homogeneous boundary value problem. we have the following theorem.~---._---_. either CI or C2. called the trivial solution. ? 0. yeO) = 0.~. the problem (23) generally has no solution at all when Do = 0 unless the numerators of (26) and (27) are also both zero. and y = 0 is the only solution of (23). --. XI < X < X2.1 with some examples. = I rule*) where 11._-------_ .0. y = 0. then C1 = Cz = 0. suppose ex = 13 = 0 in (23). and C2 can lead to certain difficulties._------_. Let us illustrate Theorem EXAMPLE 2 Solve: y" 1. t By/tJrmai solution wemean onethai. the problem has only the trivial solution.. we must also have Ll. SOLUTION The general solution of the DE is y = CI cos X + C~sin x. Solutions that arc not identically zero are called nontrivial solutions.

yielding . SO'lUTION This time our DE does not have constant coefficients. ~n most simple problems like Example 2. )'(II) = 0. C~sin.ora nontrivial aid in the actual determination of the solution. we see thai A = I costO). and also )'(1T) = C2~. However. by y = where C2 may assume any value. the general solution is v= C:1 cOS.3 for more details on the solution or Cauchy-Euler DEs. )'(1) :::.12 CHAPTER I BOUNDARY VALUE AND EIGENVALUE we find . 0 Hence.. 0 :::. O. = 0.. =::. EXAMPLE 3 °1 =0 o solurion=it does not O<:r SOLUTION < .1 is not necessary. . y(O) :::. As in Example 2.12. compuling thevalue of the coefficient determinant ." + A)" + 707 = 0" I <x< e.* The transformation .1( + C1 sin z. ... cos(-.) sin(-.lll :::. PROIlU!:MS Applying the prescribed boundary conditiens. = C! sin(O) sin('lT12) 0..e' yields the consrant-coeffieiem equation CP)I d/ 2 + 'IT'Y .) sin(O) II = I -I which merely informs us of the existence . EXAMPLE 4 Solve: xl. y(r. This determinant is introduced primarily for theoretical purposes.C1 •.only the trivia! solution Y = 0. • See Problem 34 in Exercise '1...t = . relative to Example 2.. but Is of 'the type of DE called a Cauchy-ElIler <"qllaliolt... In this case A = costO) -- I 'cos('lTl2) I =1# O. Tilis time the boundaryccnditicns demand that C. 0 y(e) = o.. 1 + C2 0 "" 0 ami deducethat C1 = '0.e this problem has an infinite which suggests nnmber of soirrliOIlS described that C~ is arbitrary.

we mean the naturo! {og(lritlrllJ. we get x). = C1 • + C2 '0= O. But if 0 = 0 in this case. but situations can arise that lead to two families of solutions corresponding to two linearly independent solutions <jJ(x) and ljJ(x).. That is. Hence we obtain the infinite and thus conclude that C1 = 0 and C2 is arbitrary. corresponding to the nontrivial solution q. since both C1 and C2 are uniquely determined by these conditions (and are not both zero). both cos .r and sin x satisfy the homogeneous boundary value problem y" +y = 0. boundary X y( -rr) = 2. y'(O) + y'( . log . . y(O} + y( .* + C2 sin( . then (23) will have a unique nontrivial solution.·also commonly denotedby In s. EXAMPLES Solve: y" + )' = 0. ) = 0. SOLUTION the prescribed y conditions x on the general solution =C 1 Cos + Cz sin • By logx. there will be either 110 solution or infinitely mallY solutions. t + C2 sin . each member of which is proportional to <D(x).. For most problems only one such family of nontrivial solutions exists. For example. or in terms of the original variable x. yeO) = 0. it follows that if)' = <jJ(x) is a nontrivial solution of the homogeneous DE and homogeneous boundary conditions. From the linear homogeneous nature of these examples... where t y(x) = C1 cos( rr log x) = log x. so is y = C<jJ(x) for any value of the constant C. Again let us illustrate with some examples. The general solution is yet) = C1 cos ..r)..BOUNDARY VALUE PROBLEMS 13 whose independent variable is now t.) = 0 (29) and the most general solution is the linear combination y = C1 cos X + C2 sin x (30) where C1 and C2 are both arbitrary constants. . y = C~ sin( .6 0.. collection of solutions (0 = 0). t. log • Applying the boundary conditions. Imposing O<x< . y(l) = C1 y(e) • we have I + C2 (-I) 0 = 0. When a and 13 are not both zero and 0 .(x) is a family of solutions.

.-----. is yeo) = 0. upon imposition of the boundary + C2 sin x leads to conditions. cos x which._. = 1 y('jj) = .1 = 0) In this case we y = cos x + Cz sin x.we know that the general solution of I . because values for both C. conditions yeO) = 1. cos x + Cz sin x i I i ~ yeO) = C. = 0. we must choose C. = .2 NONHOMOGENEOUS DEs in I I I Based on our knowledge of nonhomogeneous DEs from a first course differential equations. Examples 5-7 illustrate that the solution of a boundary value problem depends crucially upon the interval on which it is defined as well as on the values of a and 13. O<x<TI. y('jj) = -I. we conclude the problem has /10 solution (u = 0). EXAMPLE Solve: 6 y" +Y = 0. yeO) = C1 = 0 Y(TI/2) = C2 = 3. SOLUTION Again the general solution y = C. r SOLUTION The boundary applied to the general solution y lead to the relations = C. Hence. y('jj) I = Cz sin 71 = 2.C.------_-_. 0< x < TI/2. EXAMPLE Solve: 7 y" +y = 0. obtain an infinite number of solutions given by (.I. 1.3._-- ----_-- 14 CHAPTER we find I BOUNDARY VALUE AND EIGENVALUE PROBLEMS ----l I yeO) = C. Therefore we obtain the unique solution (u ""' 0) I i i j y = 3 sin x._-----_. = I while C2 is left arbitrary. ~ i i i l: . and C2 cannot be found. Y(TI/2) = 3. which is impossible for any choice of C2• Hence.

.• .. yeO) = -I. . )'r = x - SOLUTION By inspection...::: ---.. the problem (31) and (33) will have either no solution or an infinite number of solutions as before. a particular solution is found to be and thus the general solution is )' = C. _. Again._--_ . Foresman and Co. then (34) and (35) provide a unique determination of C.---. BOUNDARY VALUE PROBLEMS M[y] = F(x).(x) )'p(x) where Yu = mogeneous DE.. and hence a unique solution of (31) and (33).B2[YI'] (34) (35) where the coefficient determinant Il is that defined by (28). _.B.] C.6 and 4.[y. o <x < I.._----.. see Sections 4._-_ .---- _ _-.-. The particular solution y» can usually be found by either the method of undetermined coefficients or variation of parameters.: Glenview. (Scott.----_. and yp is any particular solution of (31).-:-. y(l) = I. Andrews. cos x I.. C. by t For a review of these techniques. and C2. Ordinary DijJcmriiotEqlloliolls witl: Applimliolls.. + C2 sin x + x .(x) + C2Y1(X)* is the general solution of the associated ho- (33) upon the solution (32). 15 (31) is given by the sum of solutions Y = Y/f(x) + = C.-.p(x).I. If Il ? 0.[y.y. • The homogeneous solution YII is oflen culled the 'complementary solution and designated the symbol Yc in many texts.B2[yd + C2BdY21 == a + C1B2[Y2] Bd)'2) B2[Y2] I . = ~---------------~ ~ 1 [3 . C. __ .. t By imposing the boundary conditions (32) + C2Y2(X) + ).y.-~.-.-.] C. we consider some examples.. M[y] = 0.[}'.. EXAMPLE 8 Solve: y" + Y = x .7 in L.B.I.-..] = [3 .-.-. _ .)1982. we are led to the system of equations C. 111. But if!:l = 0.B2[YI'] with solution ('( - B...

-- -~~~---. from which we find C~ "" If(sin 1). SUmmarizing the results of this section. . we obtain an both of which are satisfied by C1 = "IT and arbitrary il/fillite Ill/mba of solutions given by (~ = 0) y= 11" cos X + C1 sin x + x.~J '. Hence. sin x + x. and therefore C. ) '" o. sin 1 EXAMPLE 9 y" Sin x Solve: +y :::= x-I. we have the following theorem. y(n) = I. Our final example illustrates that infinitely many solutions arc also possible. = C1• Thus.~. r! 0) y". :::= yeo) + = "IT. and the second condition leads to )'(1) = C. sin 1 '" I.. ' 16 CHAPTER I BOUNDARY VALUE AND EIGENVALUE PROBLEMS The or C1 firsl boundary condition requires that .~~ . SOLUTtO:"! A particular solution is yy '" C1 cos X x.(0) :::= - I. y( . 0< x < "IT. ).-_ .. sin . EXAMPLE 10 Solve: _v" + y '" x. Hence the unique solution is (t!.--+x-1. the solution satisfying the first boundary dition is y = C. SOLUTION From Example 8.v con- + . ~-~.r < 'IT. .\'(0) = C1 ~ I = .r ~ I =' I 1. Y("IT) -C1 = + "IT 0. O<. The second boundary condition requires that y( rr ) '" 'IT - which (~'" is impossible. ~ 0. we conclude the problem has I/O solution Examples 8 and 9 show once 'lgain how critical the prescription of boundary conditions is in order to have a unique solution of the boundary value problem. 0).I..'- The boundary conditions lead to the system of equations y(O) = C1 = "IT.

. = 0.'-'.J = 0 y Trr) = 0 y(. EXERCISES 1. yCO) = I.y = 0.1"(1) = 0 )""-y=o.3 y(l)=el + 9y = 0.1' + y" + y = O. + y(1) =1 14. (a) Using (22). the problem either has no solution or infinitely many solutions.. 4.and B I defi ned by (20) are both linear operators. y" . = 0. )'(1) = 1 y(r.1'(1) = ° > +Y +s +Y )'(-1) = 0.- ~--~~. yeO) = O. 12. I. y·-)"'-6y=0. y'(O) + 3)"(0) = 0. yO . yO . + 2y = 0. solutions. 13. 17. .(.3 In Problems [-16. yeO) = 3. y" . y'(OJ =0. 10. =: 0. . y. verify that the differential operator (D = dfdx) At = A. )"(1)=0 + . 6. +y .y = O. y" y" y(O} =0. = O.}y' yO y" yO 7.-~-~-EXERCISES J. y(l) = 0 y(l) "'. the boundary operators B. y. where B![Y~l B[ 1 Y21 1· If 6. find nontrivial is unique or not. )"(0) = O.6y' + 2y = 0. )"(0)=0.) = -2 y{h) = -3 J(I) = 2. if they exist.1"(0) = O. has a unique solution ifand only if 11 ~ 0.3y' 15. II. 8.2 M{y] If}'1 and y. = O. )'(1) '" -I y'(I) (k 0) yeO) = 3.3 17 THEOREM 1. are linearly independent solutions of the homogeneous =: 0. then the boundary value problem M[y] =: DE. '" O. 9. 16. y(O) = 0.r)D' + A. )"(0)=0. yeO) = I. .) = + 9. y'(r.~. State whether the solution y(I)=O 2. F(x). y" "IT)" 0 )"(0) 5. 3. y"=O. _rIO) = O.1"(0) + y( I) = 0.~---~~. = 0. + k'y = 0.~ .(x)D + A"'r) (b) is a linear operator.

= e". y. y" = 3 sin rr. )"(0) = 0. > 0) has no solution. 1/ - b . y(O) = 0. . yea) = 0. >"(0) = 0. /(1) = 0 y(l) . b. yeO) '" 1. For which values of k does the problem y" have nontrivial 20. y(b) = «(I 13 b). The DE . find the solution (if possible).18 CHAPTER 18. . y(b) = I. y" . (k" solutions? 0) solutions? What are the corresponding Determine the values of b for which the boundary y" value problem (b + 4y = 0.r.6y = e". ! ·:3. cannot have a nontrivial solution for real values of k. State whether the solution is unique or not. J~6) = -2. ° ° 27.y'(1) = y'( 1) = 26..rx. + Y 0..6/ + 9y = 9. (c) no solution? exactly one solution? more than one solution? I In Problems 24-33. yeO) = . y" = I. )"(1) = 0. y" . 31. y(b) = 3. 30.1'" + have (a) (b) _I" b (if any) docs the boundary y«(I) = I. y'(b) = 13 (I "" has a unique solution for every choice of a and b such that u== :!:J. y(O) = 0. 33.I y'(O) = 0. .. 1 BOUNDARY VALUE AND EIGENVALUE PROBLEMS Show that the boundary value problem yeO) '" 0.~2. = -x. + 4y . y" '" sin -.y = y. Y(II) = 0. . ¢ Show that the boundary value problem = y. 24. 28.. y(l) = 3 y(l) '" 1 y" .I. 19. 32.y' . y(O) = . 25. !lrol2. e=. I .1'(0) '" I. )"(1) = 0. + k'y = 0. y" = x.. *34. 21. ror which values of /I and . y'(I) y'(1) = tt« =I y'(""!2) = .3. = + 2y' + y x.=J. yeO) = 0. yeO) = 0.Py '" 0. y. value problem (" < h) = 0.. Show that the boundary value problem y' '" 0. y. 29.e : ""/5 y'(I) =I y(l) = . has a unique solution for every choice of a and b 22.

(x) dx J dx.t'y· . ~O. 38. WRONS KI (J 778 -1853) who studied Olathematics in Germany but lived most of his life in France..\". and solve this DE for IV.1" _\(1) '" y(e) '" e'/~ ! (Abets [ormulat" If. and transform back to the original variable to obtain the desired solution.d solution. 'NEI LS H.5. y(l) = y(e') '" 0 + 5.ry" 3. lise the technique of Problem 34 to transform the following Cauchy-Euler equations to constant-coefficient equations. 0.1""'. Stricken with tuberculosis in 1827. b.ty' + 25y = 0. *-"9. (b) Using the result of part (a). he died two years later at Ihe age of26.5xy' x'y" . .~p[ - J u .)(x) = C exp( - f a.1'(1) '" 0 1 1 I I I 37. = 4.EXERCISES where a. *42.') '" f!12 .\. y(l) == I. y(O) = 0..1' = y(J) = () y(e') '" 0 )"(1. . yO) '" 0.(x) (1"(. equation can be In Problems 35-40.y '" . W(y. . Show that if y.(X»'l = W(x) for)" using Problem 41.(x) clX] . x'y~ . for some constant C..(X)} e. •) Yi(X Hint: Solve the first-order DEy. *4 J. .'"(0) '" 0. t R ccall th at the Wronskian is defined by W(y" y. 35. is call ed a Cauchy-Euler equation. Him: Show that dlVldx + aM) \I' = 0.) '"" y. . H.IY '" D(D - I)y. 0. It i~roamed after JOZEF M.Y. solve (he constant-coefficient equation. y. and care consta nts .y.y. are continuous. *36. show that the Cauchy-Euler transformed into the constant-coefficient equation faD' + (b .y>. ABEL (t802-1829) was one of six children born into a poor Norwegian family. ! + .\)"' . and y.(x) is a nontrivial solution of yN + ill(X»" + flo(x)y == 0 a second (linearly independern) solution is given by _ Y1 = y.a)D + elY == O. (a) Show (hat the change of variable x = e' leads to (D = dldr) xy' '" Dy.f)Y = on some interval where (/J{x) and satisfies.3 19 . 2x'y" + . x'y· + xy' . =.. One of his clIrty accomplishments was provi ng that Ihe fifth-degree al geb raic equation has no radic. 1.(x».xy' + y = 0. O. are linearly independent ol-der DE y" solutions of the second0 show (hill the Wron skiant + a/I))" + (/.\".r.











In the previous section we found that homogeneous problems, consisting of homogeneous DEs and homogeneous boundary conditions, can sometimes have nontrivial solutions. Here we wish to examine this special class of problems in more detail. Frequently the solutions of a DE depend upon a pararneter x which may assume various values during a given discussion. This parameter can appear in the coefficients of the DE, in the boundary cond itions, or in bot h. For a large class of problems of practical significance, the typical equation is of the form M(y]

+ hy




where M = Alx)D~ + A l(x)D + A1,(x). In the problems we wish to discuss, we will generally assume that A appears only in the DE, as in (36), but not in the boundary conditions. Clearly, the general solution of (36) must depend upon both .r and the parameter :i.. Thus, if)'1 and y~ constitute linearly independent solutions of (36). we write the general solution as (37) Subjecting this solution function to the


conditions (38)

leads to a coefficient determinant A (given by (28) in Section 1.3] that must also depend upon A in this situation. Problems of this type arise routinely in solving partial differential equations (Chapters 6-8, !OJ, but also come up in other applications (e.g., see Sections 1.5.2 and 1.5.4). The basic problem is to determine al! values of A for which t..(I-..) '" 0; that is, determine all values of A for which the homogeneous boundary value problem (36) and (38) admits nontrivial solutions, and then find the solutions corresponding to these values of 1-... These special values of A are called eigenvalues, and the corresponding nontrivial solutions are called eigenfunctions. The terms characteristic values and characteristic [unctions are also commonly used. The general problem described here is called an eigenvalue problem, or Sturm-Liouville problem after the two mathematicians who pioneered its investigation. * Special equations (minus boundary conditions) that fall into this classification are the following:

• JACQUES C. F. STURM (L803-L8SS) was a French mathematician of Swiss origin. He is recognized for mal;.ing the first accurate determination of the velocity of sound in water. his essay on compressible fLuids. and his work in differential equations. J05 EPH LlOUVI LLE (1809-1882) provided the first proofs of the existence of transcendental numbers. His other achievements include his research in boundary value problems, the theory of nu mbers, and differentiaL geometry.





y" (I -


AY = 0, = 0,






(Helmholtz I (Legendre (Bessel (Laguerre (Hermite

equation) equation) equation) equation) equation)


X)V" - 2xy'
l -

+ A)'


<.r <


+ xy' +

,,2)y = 0,

O<x<b O<x<""

.tf + (I

+ AY =0, y" - 2xy' + AY = 0,
- x)y'


t· ,





The above DEs are common to many areas of application. For example, Helmholtz's equation appears in vibrating-string problems and in finding the temperature distribution in a rod. Legendre's equation arises in certain problems displaying spherical geometry, Bessel's equation is closely associated with problems involving circular or cylindrical-shaped regions, and the equation s of Laguerre and Hermite are con ventiona lin certa in quantu m mechanics problems. In appl [cation s the eigen va lues and e ige nfu nctio ns have irnporta nt ph ysical in te rpreta tion s. For exa rnp!e , in vibrat ion probl e ms the eigen values are proportional to the squares of the natural frequencies of vibration, while the eigenfunctions provide the nat ural configuration modes of the system. In buckling problems the eigenvalues <Ire related to the critical loads a column can support before deformation may lake place. and the eigenfunctions describe the possible shapes of such deformations. The eigenvalues denote the possible energy states of a system in quantum mechanics problems, lind !h" eigenfunctions are the wave functions. The wave function for a given energy level will yield the electronic distribution for the state corresponding to that energy, as we II as at he r prope rties. Th us, the in te rpreta tio n s vary wide ly , as do the areas of application. EXAMPLE II and eigenfunctions of yeO) '" 0,
),(1) ""

Find the eigenvalues


<x< I,




..:..:., {~ '

SOLUTION We first attempt to find the eigenvalues, which we tacitly assume are real. (More will be said about this assumption very shortly.) Because the solution of the DE may assume different functional forms, depending upon the value of A, we mu st consider three separate cases corres pond ing to these different functional forms."

CASE I: Let us first assume


= 0, which leads to the general solution y


+ C2x.


• The auxiliary equal ion is m' + ), ~ O. with roots

III '"


spond to ), ". O. ), < O. and), > O.

v'=1.: . Thus.

I he

three ca ses








The first boundary condition ),(0) = 0 requires that C1 == 0, and the second condition leads to C1 + Cl == O. Hence, C1 = C2 == 0, so that the only possible solution is the trivial solution), = O. Thus, ).. == 0 is not an eigenvalue of the problem. CASE 11: If).. is negative, it is helpful to set x = -k1 < 0, and then the general solution can be expressed in either the form
y == C1tI'x

+ C1e-h
+ C1 sinh

y = C1 cosh kx

where the hyperbolic cosh

functions are defined by

z = He' + e-'),

z = He: - e-:).

In solving eigenvalue problems on finite domains, it is usually preferred to express the general solution in terms of hyperbolic functions rather than exponential functions (although either solution form is acceptable). The reason for this is that certain simplifications take place in determining the constants when using hyperbolic functions; for example, cosh{O) == I and sinh(O) = O. Other basic properties of these functions are taken up in Problem [6 in Exercises 1.4. By applying the boundary conditions to the above general solution written in terms of hyperbolic functions. we find
y(O) = C1 =


yO) but since sinh k ;.0: 0 for k only the trivial solution y this problem.
;.0: = O.


C1 sinh k = 0,

0, we deduce that C1 = C1 = 0, which leads to That is, there are no negative eigenvalues of

CASE ll/: Next we set x = kl > 0, and the general solution is
}' =

C1 cos kx + C1 sin kx, leads to C1 =.0 and


of the boundary conditions

C1 sin k == O. If we req uire C I ;.0: 0, this last cond ition can be satisfied only if k is an integral multiple of n, i.e., k = n"lr {n = I, 2,3, .. J.* Hence, the eigenvalues are given by n and the corresponding eigenfunctions





are any (nonzero) multiples of

• Since k .. k', w e can take Ie positive without loss of generality.

= <Pn(X) ==



sin mr.r,




(i.e., we set C! == 1 for convenience). REMAR K: We could also determine the eigenvalues in Case III by considering the coefficient determinant
d(}.) =


cos «





which vanishes, of course, for k '" mr (n = 1,2, 3, ... ). However, the actual calculation of J.(}.) is not necessary ill IIIOS/ simpie problems, bur if may prove useful ill more difficfllt problems.




II '"


" ,_/







1.2 <t>.(x) = sin
1177.1', II


== I. 2,3, and their zeros on the interval

(0. 1).

The first three eigenfunctions of Example 1I are sketched over the interval [0, I J in Figure 1.2. All eigenfunctions are necessarily zero at the endpoints .r = 0 and x = I because of the prescribed boundary conditions. Notice that the first eigenfunction <!> I(X) == sin 11X has no additional zeros On the open interval (0, I). However, the second eigenfunction <!>~(x) "" sin 2'll'x has one zero on (0, I) at x = 1/2, and the thi rd eigenfun ct ion !f>J (x) = si n 3rrx has zeros at x = ]/3 and x = 2/3. In fact, it becomes clear that for any n ~ 1, the eigenfunction !f>n(.t) = sin mrx has exactly (1/ - I) zeros on the open interval (0, I), which are located at


23 ....... n n 1/
1 -

(Ii •••• ., ---.



Not only is it true for this special set of eigenfunctions, but a general property of eigenfunctions is that the nth eigenfunction ~A.t) has exactly (1/ - I) zeros on the open in terval for which the proble m is formulated. This is a remarkable property that also has important physical significance. For instance, in vi-

- ---------------------








bration problems the zeros of the eigenfunctions identify the nodes (stationary points) for each vibrational mode of the system. Fina IIy, we point out that, in sol ving for the eige n values in Example II, we assumed that they were real, The possibility of complex eigenvalues exi sis, but subsequently we will show t hat for a certain class of problem s (to which Example I I belongs), only real eigenvalues are possible. Also, because eigenfunctions are determined only to within a multiplicative constant, it is customary in most situations to set the constant to unity. In some problems the determination of the eigenvalues requires us to find the roots of a transcendental equation. Such equations can often be solved numerically by a simple procedure such as Newton's method from calculus. This particular method relies on initial approximations to the roots, the accuracy of which may then be improved upon by an iterative process. The number of iterations necessary for a des ired accurac y will de pend upon the ace uracy of the initial estimate s of the roots. We will illu strate Newton' s method in finding the eigenvalues of Example 12.


12 and eigenfunctions 0<.\"<1, belonging to
)"(0) = 0, 10 consider y(1)

Find the eigenvalues


"Y = 0,





Again it is necessary '" 0, the general

three separate


CASE I: For"

solution of the DE is


= C1

+ C~x.

The two boundary conditions demand that C1 '" 0 and C1 + 2C1 = 0, respectively, forcing the trivial solution. Therefore, " = 0 is not an eigenvalue.

CASE II: For negative '" we again set"


-/!- <

O. The general



and the boundary

= C1 cosh


+ C1 sinh kx,
cosh k)


lead to C1 = 0 and

Cl(sinh or, for Ct



= 0,


k = +tanh k.
To see if there graphs of /I = it is clear that conclude that are any solutions of this transcendental equation. we plot the

k and u = - tanh k and look for intersections. From Figure 1.3 no intersections of these curves occur for k > 0, and hence we
there are no negative eigenvalues.



.. '
-·:r ..... , ~~.-~
. .

FIGURE 1.3 CASE III: If A = kl > 0, then the general solution is
)' 0=

C) COs kx + C~sin kx.
that C1 == 0, and the condition

The boundary condition .r == I leads to

at x == 0 requires




+ k COs k)


:,;r ...

.:;.".; .~ . .-.{













Once again we plot the graphs of /I "" k and 1/ "" - tan k to see if any intersec tions occur. Looking at Figure I .4, we see that there are an infin ite number of values k; (II "" I, 2, 3, ... ) where intersections occur, but their exact values must be determined numericall y. To utilize Newton's method in finding these values, we look for zeros of the function
F(x) ~ x

+ tan


Each value of x for which F(x)


0 can be approximated

by the sequence" ,

== s.>

F(xj) F'(xo)'


0, 1,2, ...

where Xo is our initial estimate of a root. For instance. from Figure 1.4 we esti mate the first zero, denoted by k I, to be ap proxi rna re Iy 2. Sell ing r, = .2, we find
F'(2) = I

+ sec2(.2)



and thus
Xj Xj+1 '" Xj -

+ tan 6.7744





0, 1,2 .....

Successively substituting j '" 0, 1,2, ... sequence of approximations

into this last result. we obtain the

= 2.0273,

Xl =




and therefore conclude that k, = 2.0287. Consequently, Al = ki = 4.1 [58. Using Newton's method for finding additional intersections of the curves in Figure 1.4 will involve more iterations than the first value because of the steepness of the tangent curves near the intersections (see Problem I7 in Exercises 1.4). We do note that after the first few in tersections, the re mai ning intersections of /I = k with the graph of /I = - tan k are close to the vertical asymptotes of - tan k. Hence, we deduce that the larger eigenvalues satisfy the approximate relation A. Corresponding eigenfu nction (set C2


H2n -





to e~ch eigenvalue

A" '" k~ (n

I, 2, 3, ... ...

J is the

n: 1,2,3,

* Newton',

metbod is generally given by x", .. sr: F(x,)lF'(x/l, but it is more convenient in many cases to usexJ" = XJ- F(x/lIF'(xtJ so Ihat F'(xj) does not have 10 be evalualed al cach step, However, if F'(x) is rapidly changing, then Ihc rate of convergence of the sequence of esrtmaies may be considerabty slower by this simplcr formula (e.g., see Problem 17 in Exercises 1.4).



The three cases identified in the hunt for eigenvalues in Examples II and 12 are direct] y related to the form of the equation y~ + AY "" O. For more general DEs, the cases are usually different. For instance, in solving the general second-order, constant-coefficient DE ay"

+ by' +

).y '"

0, am2 + bm +).


we are led to the aux illary equation (by setting y "" r)


The roots of thi s auxiliary equation are given by -b± vb~ -4aA 2a and th us th e three cases in vol ved in the search for eigenvalues to b -4aA=0, b
2 2


(40) 0
= - 1.2

4aA "" J:! > 0, 4aA

b1 -


(rather than A = 0, A < 0, and A > 0). Sometimes the eigenvalue problem involves a higher-order DE. The solution procedure is basically the same as that used in solving second-order DEs, as illustrated in our nex example. EXAMPLE 13 Find the eigenvalues and eigenfunctions of



yeo) = y~(O)



= y'(I)

= 0.*

Because we are dealing with a fourth-order DE, four separate bou ndary condl tions are prescribed rather than the customary two associated with second-order equations.


Suppose we start by assuming A "" J? > O. The resulting DE then possesses the general solution
y == C1


+ C~x + C) cos lex + C4 sin kx,

the first two derivatives

of which are C~ + k(-C) sin kx



+ C4

cos kx),

y~ "" -k1(C) cos kx

+ (:4 sin

• For higher·order


we use

y" ~ d'y{dx"

in place of primes.







The two boundary

conditions at .r

° yield the eqUations

Cl + C]


C] =0, from which we readily deduce Cl ee CJ = O. The remaining boundary tions at x = I lead to the set of equations C2 condi-

+ C. sin

k = 0,

C! + C~k cos k


This system of equations has a nonzero solution for C2 and C. if and only if the de terrninant of coefficients satisfies sin k k cos k


= k cos k - sin k == 0.

Hence, k must be selected such that k = Ian k, which is similar to the situation in Example 12 (see Problem 18 in Exercises 1.4). Once again it can be shown that an infinite number of values k; satisfy this last relation, giving the set of eigenvalues A. = k~ (II == I. 2, 3, ... ). Observe that when k is one of the values k,,, the ' w ·0 equations above in C2 and C. are then equivalent, and either one can be used to eliminate one of the constants. For example, the first equation gives C2'" -C. sin ko, (set C. == I)
11 =

from which we obtain the eigenfunctions


== sin k"x - x sin k",



In this example it is a routine matter to show that when A .", 0, the DE possesses only the trivial solution:" Thus the problem has no additional e igenvalu es. Eigen value problems are characterized by bot h a homogeneous DE and boundary conditions. As we ha ve al read y observed, such problems always possess the trivial solution (y = 0), but it is the possibility of nontrivial solutions that interests us most. The .exarnples illustrated here are rather elementary, but the techniques used apply to more difficult problems. In fact, the eigenvalues and eigenfunctions belonging to certain types of Sturm-Liouville systems have many properties in common, so that solving a simple example of a general class reveals much information characteristic of the whole class. We will discuss these ideas in more detail in Section 1.6.




In Problems 1-14, find all real eigenvalues and corresponding eigenfunctions given boundary value problem. I. y. + J..y '" 0, 2.

of the

y(O) '" 0, y'(O) .. 0, y'(O) '" 0,
y(O) = 0,

y(2) Y'(lI)
y(1) -



+ +

J..y = 0,


y' + J..y = 0,
J..y = 0,


4. y" 5. 6. 7.
y' y" y"

° ° °




+ 2)" +

(I - )..)y = 0, 9)..)y = 0,

yeO) = 0, y(O) = 0,


+ 4/ + (4 + + y' + J..y

= 0,

r(O) '" 0,

y{l) =

8. y' + 2)"' + )..y = 0,
9. y" j-"


r(2) '"

° °

y(O) '" 0,
yeO) '" 0,

3.1" +

2)..y = 0,


yO) = 0 .1"(0) - )"(7":)



lIy = 0,

y(O) - .1'(7\") '" 0, r(!) '" 0.

.x\-" + xv' + ).._\' 0, '"
.l~yjj -


yk) =



... J..y = 0, A

)"(1)=0, .1"(1) =0,

y(t") = 0





+ - v '" 0, .r: +
IIw '" O. .

r(2) =

y(]} = 0,

*14. *15.





y'(e~) '" 0

Given the eigenvalue problem y' + AY = 0, (a) (b) (c)

show that if >-. = is an eigen value, the n h = I. If h > I, show that there is exactly one negative eigenvalue and that this eigenvalue decreases as h increases. If h '" 2, show that the positive eigenvalues for large values of n satisfy the relation J... ""




y'(O) = 0,


(d) 16.





P I.


if II

< 0,

show that there are no negative eigenvalues. and sinh x =

Show (a) (b) (c) (d) (e) (f)

the hyperbolic functions cosh x = (1" + r')12 - I' -')12 satisfy Ihe relations (see figu re) cosh(O) = I. sinh{O) = O. cosh! X - sinh' x = I. dldx cosh .r = sinh .r, dldx sinh x = cosh x, cosh x =- sinh x "" e'tz for x,. I. that

). + + AY = 0. yeO) = 0.). approximate lion cosh X cos x + 1 "" O. 2. )'(2) . + AY y'(I) = O. "" 0. Consider the eigenvalue y' problem = 0. y{'TI) y'('TI)" 0 0.J. 2. '" Xj .138. I.)IF'(x.. Using Newton's method.. + Y'('TI) "" 0 )'(1) *22. approximate the second 18. Using Newton's method.. y" + Ay = O. *24. (b) Xj" "" x) F(x.)lF'(xo). '" 24. equa- 19. yH Y" for the first nonzero + Ay + AY = 0. j = 0.-------~ 30 CHAPTER I BOUNDARY VALUE AND EIGENVALUE PROBLEMS PROBLEM 16 17. Also determine the form of the eigenfunctions.y) '" 0.y'{O) '" 0. 2.'(2) = 0 >. *25. use Newton's method to find an estimate eigenvalue A. I. 20. . • The actual eigenvalue is >. y'(I) . j = 0. eigenvalue in Example 12 by employing the formula" (a) Xj. 3. Find all real eigen values and eigenfunctions in each problem: y'(O) "" 0.AYO) = 0 )'(1) = 0..F(x.. . Using Newton's method with Xo = 5 and four iterations.. j = 0. yeO) . . yH y(1) '" 0 Xl. *26. + y'(l) = 0 Problems 23-25 are nonstandard eigenvalue problems because of the way in which A appears in the DE or boundary condition.(xy' . 3. 21. approximate the first two eigenvalues to three signficant figures by employing the formula Xj" ill Example 13 = Xi - F(xj)IF'(x.. 1. the first root of the transcendental In Problems 20-22./(0) yeO) = 0. yeO) = 0. :0 yeO) .

show that if 4>~(. 28. y'(lJ = y". y'(I) *}o.jl".. y'" ~ h)' = U. 31. y{ I) = ] r" + + '\y = 0. 0. y(. r( -11") = r.".) =0 = 32. finding the steady-state temperature distribution in a long slender rod with prescribed end conditions. +}o... we wish to derive the equation of equilibrium for a string stretched tight Iy bet ween two supports. r-. 27.. and subjec ted to a distributed vertical force of intensity q(.~~: . y.y = 0. = O.. ) =I .I)HYSICAL APPLICATIONS 31 Without explicitly solving for them. . respectively. )"(. 2 In Problems 31-33.r) = y"(. subtract the two equations. then . for which the given boundary value problem has a unique solution.' -~. r'" y'" -). and then integrate the result from 0 to ] using integration by parts and taking into account the boundary conditions. v" 4.t) per unit length (see Figure 1. Ii nd Ihe real eigenvalues and eigenfunctions of the fou rt h-erder DEs by assuming only non-negative eigenvalues (sct x = 0 amI)" '" k' > 0).jl. " '" 0. while others will be taken up in the exercises. In this section we will discuss several elementary examples. - ..\.! }o. and find the solution in these cases. y(O) = 0. vIOl = y '(0) = y(I)=y'{I)=O 1. It . y(O) = l . '(OJ'" y.~ Hint: Observe that 4>~ + }o.1"(2) = 0 = ~9.'.m4>. In Problems 21-30... located at x = 0 a nd x = b."...\" + }o.0) "" )"'{O) y = O. 4>: + ).y -. • JJ. the second by 4>~.. MUltiply the first of these equations by .".5. determining the possible buckling modes of a long column under axial loads.t) and ..jl..(1} 0 = 0. :'. )"(0) = 0. with }o. y" + 4y' + (4 + 9'\)y = 0.."(O) "" 0. and }o.(x) are eigenfunctions corresponding 10 the eigenvalues ~.1" '" 0.m .5 PHYSICAL APPLICATIONS Boundary value problems are associated with determining the equilibrium configuration of some physical system such as a stretched string or a beam supporting an external load.y = 0..1 DEFLECTIONS OF AN ELASTIC STRING To begin. )"(. determine the values of ).~ _. and so forth. 1.5).

6.----~-----------. Let us focus our attention on a small element of the string from x to as shown in Figure [.• -----y FIGURE 1.---------- -VALUE AND EIGENVALUE PROBLEMS force T which remains 32 CHAPTER I BOUNDARY is assumed that the string is under a large tensile constant for small displacements y . Because q(x) is a distributed load.6 Small element of the elastic string. That is. T on the string is so large thai gravitational effects can displacement 3.r-axis in a vertical plane. t That is. a torce per unit length. OUT basic assumptions include the following: L The mass of the string is constant 2. Further simplifying assumptions must be made so that the resulting equation of equilibrium does not become too complicated. I q'(x) t t I T I I I I ~as y ~ I T FIGURE 1. The tension be ignored. and the string is perfectly elastic.-axis downward. .t the total (.5 Displaced position __ . • II is cuslomary in such problems 10 choose the positive .t + ~x). we wish to approximate nonlinear behavior with a linear model. The equilibrium position of the SITing is a "small" from the . a common practice in applications.-- Elastic string under a load.

al the ends may be nonzero [e. II is common to lind the same DE occurring in entirely different applications..~). at (x + ax) we find the v-cornponent of the rensil e force to be Tsin Hence. equation (41) of Finally.Ty'(.. . * The string may be attached to various types of supports at the ends .. = T tan O~ == Ty'(x + l!.6. This condition could be realized. These include the external force and the ycomponents of the tensile force at x and at (x + l!. (42) called the one-dimensional POiHO/I equation.• see Problems I and 7 in .. by considering [he free end looped over a frictionless peg that maintains a zero slope on the string at this end while allowing small vertical movements (see Figure ]. Likewise.r) . ) 7 (/IX) "" O.."'(x [ by j. where we are using the small angle approximation sin 81 = tan 81 and the fact that tan 0] is simply the slope y'(x) of the string at . which yields zero displacement (y = 0) at the endpoint. '(Xl] + -. the free end must be restricted from appreciable movement .e. it is not really "free").r = ~x and write the external force over the small element of the string as q(x)l!.PHYSICAL APPLICATIONS 33 force acting on the elemental length of string is simply q(x)~s (i..r. Most curnrnon are the fixed support. or.r.r = 0 and x-b.e. For small displacements we only need to consider forces acting in the vertical plane. and equilibrium conditions demand that the sum of all such forces algebraically add to zero. t and the "free end.\" :lx .g .r < b.\"). ali vertical .7). we find + Ty'(x + !u) + q(x):1x == 0. equivalent to an elastic ·Obscrve that (42) has the same functional form as (I 2)forthe steady-state heat conduction problem.T sin (II ~ - T tan 81 = -lY(x). Moreover.. summing 8:. the y-cornponent of the tensile force at x is given by as .x and T.I") forces. + l!.r-axis (i.x). along the .'.x lends to zero leads to the desired equilibrium )" = --. upon division . In order to maintain nearly uniform tension. tin some cases. for example. the displacement Exercises I. the force is essentially constant over the short length l!oS). the limit of (41) as . A third type of bou nd ary cond ilion arises when one end of the string is attached 10 a yielding support.. for small vertical displacernen ts t he slope of t he displaced st ring rernai ns small so that we may use the approximation l!. " q(x) T 0< ." which requires the slope to vanish (y' = 0) at the end point. Referring to Figure 1.

8 Yielding support. ). that C1 = QI}'lT/2Tand C2 '" and the boundary conditions require O. if 0 Y'''' =====:----x Y FIGURE 1. Thus we .7 Free end support.---___ ~_~~~_~ _ or' r~_~~~· ---------------------- 34 CHAPTER I BOUNDARY VALUE AND EIGENVALUE PROBLEMS spring Chat exerts a restoring force proportional to its stretch.3qI}.8). Such a condition is mathematically described by Ty' + ky '" Oat the appropriate endpoint. T Two integrations o <x < 'IT. y FIGURE 1. SOLtmON The problem is characterized by y(O) = 0. where k corresponds to the spring constant (see Figure 1. and supporting a constant distributed load of intensity ql}per unit length. EXAMPLE 14 Find the shape of a string stretched tightly between fixed supports (zero displacements) at x ee 0 and x : 'IT.O. of the DE lead to the general solution y= -2Tx 2 + CIX+CZ. Also de terrnine an expression for the rnaxirnu m displacement of the string and the tension necessary to keep the maximum displacement no more than. qo o.('IT) '" qo y " = --.

1..1 J 2... and v ~ "'}"' it follows that 0. . and is given by· Y"""=sy' To keep }. ~- .. Hence the problem of finding these deflection modes is mathematically equivalent to determining the static equilibrium position of a tightly stretched string subject to the distributed load q(x) = pw!y. such action generates a distributed inertia force (force/unit length) of magnitude pwly in a direction transverse to the string. = pw2/T. the maximum displacement occurs at the midpoint x = 'If/2. 'In general. t Under proper conditions. Since R '" Iyl.x)... we can then write (42) in the form of an eigenequation 'y" +).4 = 4..5.. '" here \. is the linear velocit y and R is the radius..r(1T . we then obtain the inert i~ fon:e P'" '_.9 Rota! ing string.5. 'y. O. Making this replacement in (42) and introducing the parameter).9).. (43) referred to as the one-dimensional HelmhollZ equation.1 is now can nected at .2 ROTATiNG STRING FIGU":t~ ]. Multiplied hy the mass d~nsily p. the te nsion T must satisfy the relation qo7r! 81 :s.. Suppose Ihe st ri ng disc ussed in Section 1. Because the load is symmetrically distributed over the string.3qo from which we deduce T..3qo.hose axes of rotation coincide with the axis of the taut string (see Figure 1.r = 0 and .. o <x < b... . t The radial acceleration of a point x On t he string is a '= v'IR.. • • :s. the maximum displacement point is a solution ofy'«' ~ O.PHYSICAL APPLICATIONS 35 obtain the parabolic deflection curve qo Y '" 2T .y = 0. qO'lf 2 O.r = b to the centers of two synchronized rotating shafts . th is inertial force displaces the string away from its initial rest configuration.!12. lfwe assume the string has linear density p (mass/unit length) and is rotating with uniform angular speed w.

deflection mode arc found to be The first critical speed and corresponding WJ=b\/p' 7r fi· $lr) =C sin b' 7tX (46) Notice that only the shape of the string is determined by 6J(x). . and so forth. . This nondetermination of the amplitude is a direct consequence of using a linear mode! to describe nonlinear behavior.3..36 CHAPTER I BOUNDARY VALUE AND EIGENVALUE PROBLEMS The eigenval ue problem consisting of (43) along with a prescribed set of homogeneous boundary conditions has nontrivial solutions only when A is one of the eigenvalues. * (44) For angular speeds smaller than WJ. must be positive. 2. leading to the set of eige nvalues and eigenfunctions given by ~"(x) =: C ..• (45) where C is any (nonzero) constant. In spite of this particular shortcoming.= J¥. 3.. o 1. of the beam.U Sin b' /I '" I. th e boundary conditions are yeO) =: y(b) = 0.e. Each eigenvalue An corresponds to a definite value of the angular speed W. and the curve of its resulting cen troidal ax is is called the el astic cu rve. the firs! deflection mode may be obtained. when both ends of the string are fixed along the x-axls . This load causes the beam to bend or be displaced from its equilibrium configuration along the z-axis.. For example. the amplitude C is apparently arbitrary.3 DISPLACEMENT CURVE OF AN ELASTIC BEAM A uniform elastic beam of length b supports a distributed load of intensity q(x) (see Figure 1. To do so. /I....10). we will assume that such displace- • The form of (44) suggests that aU eigenvalues )". or displacement curve. the linear model used here gives good qualitative information about the rotating string. a new configuration mode is possible. An important problem in strength of materials is to find the equation of this displacement curve. the only stable configuration of the string is its undeformed position (y ee 0) along the x-axis.5. W.5). To remove this indeterminacy.. we would have to analyze the true behavior of the string more exactly (see Problem 16 in Exercises 1. If the angular speed is increased until W =: Wj. /I =: 1.2. called the criticalspeed for that mode of deflection. If W increases further to Wl. i.

A teach end of the beam element is a shear force Vand a bending moment M.lx(ld.f) ::= O..--~--J!i-(X--'+ L) \" '-... N ext . Iak ing moments about the end x..l_.M(x + Ax) + Vex + d. rnents are small and wefl within the elastic limit of the beam. (48) . Our sign convention is such that V is positive downward and M is positive COUI1terciockwise. II). I· I y I r------ -~--x I M(X(J. and also that the force q(x) is approximately constant over this same arc lengt h.11 r-I q(x) I ~x M(x + 'xl FIGUR.dar - q(x). Let us begin by exarni ni ng Ih e equ ilibri um COn ditions of a small e Iemenl of th e beam over the interval from .E Small ele men r or the beam.. Summing forces in the vertical direction gives us vex) + q(X)dx - vex + a. we find M(x) .10 Elastic beam with load.. Equilibrium conditions require all forces to sum to zero and all moments taken about one end of the beam to vanish. by the use of some basic principles of elementary beam theory..:1.< 1..".X) (see Figure I. we will derive the DE satisfied by the displacement curve.f) '= 0. (47) where we are using the approximation thai the elemental arc length is ds = dX.r to (x + !!.+ .PHYSICAL APPLICATIONS 37 ------Displaced beam y - FIGURE 1. Then.

\) _ J21/_( _t_)' . since EI is constant (4) _ 0< x < b. Free end: No moment or shearing force exists at the[ree • For example.:. y = 0. and llR is the curvature of the beam. ---->- 0. ~ __c...F(x) _____:. we have the approxi mation * M == Ii EI = [I + (r')~V' Ely" = EI . for a uniform beam. we obtain the standard equations of (49) lvI'(x) = Vex)..(S modulus. Thus..38_-. Timoshenko. (Van Nostrand: New York) 1955.e. (53) s' == 0. 3rd ed. see S.CHAPrER I BOUNDARY VALUE AND EIGENVALUE PROBLEMS -. The term EI can physically be interpreted as it measure of the stiffness of the beam. ' or. but one for which the general solution can be obtained through four integrations of the right-hand side of (52). At this point we wish to relate the bending moment M to the displacement y.u. _ ax and by passing to the limit ~x elementary beam theory V'(x) = q(x). Combining both equations in (49) yields the result (50) M"(xl = q(x).-----_------. . .'}. From the elementary theory of small deflections.x M(x ~. and slope must van is h at the rued end poi nt. endpoint.- terms in (47) and (48) leads to Vex + ax) .. (Ely") = q(x). I v' I q I. i.- + ~x) __ . writing M = Ely". 0< x < b. Strength of Materials.e. I is the moment of inertia. (5 I) where E is YOllll." -). Rearranging .----== q(x). Several different types of end conditions are commonly prescribed in the study of beam deformations: Fixed end: Both displacement i. Once again we are led to a nonhomogeneous DE. \ _.H(x) _ V( x_ + . (50) becomes d'-d):. (52) q(x) y - 1:::1 ' 0< x < h.1.

SOLUTION The problem is characterized by yeo) = y"(O) = 0. d (56) EXAMPLE IS x = 0 and x = I. = Cl = 0.12). we obtain the general solution x == where the C's are constants of integration.12 x=1 Beam supporting a load. Hence.PHYSICAL APPLICATIONS Ely" = 0. where the boundary conditions reflect the fact that the beam is simply supported. 39 (54) d (t1y'1 d. i. x=O FIGURE 1. Sliding clamped end: No slope or shear force exists at the sliding damped endpoint. Also find the bending moment and shear force at each point along the beam. y(l) = y"(I) = 0. Ely" == O. or moment exists at the (55) y = 0. and the maximum displacement. . the displacement curve of the beam is given by ° sc- .r = O.e... By performing four successive integrations of the DE. The two boundary conditions at require that C. Simple support or hinged end: No displacement simpiy-s upported end point. and subject Determine the displacement curve of a beam that is simply supported at (0 the distributed load q(x) = Elx (see Figure 1.116and C) = 7/360. while the remaining conditions at x = 1 lead to C1 = .(Elv") = () dx .e.y" == 0. i.

from which we calculate . where .5. . I. ]. By p x=b \ \ I I FIGURE 1... 2. approximately..5 [9.. we find that . the shear force is related to the bending moment by Vex) "" M'(x) .0065.40 CHAPTER I BOUNDARY VALUE AND EIGENVALUE PROBLEMS 1 Recalling from (51) that Al(x) ee Ely"(x)... value can also be round in this case by the quadratic rormula. . so that Vex) "" ~ E/(3x! .I'm» = 0." That is. as shown in Figure I.1'0 '" 0. Successively substituting j '" O._ + of the beam occurs at 24 12 - 7 360 "" O. . Finally. i= 0. Also. ·Thi.r "" 0.5 is our initial estimate of the point of maximum displacement.I).. .13.2 . the bending moment of the beam at any point x along the beam is found to be M(x) "" 6" EIx(x· I .! = :. method to approximate this By defining F(x) = y'(x).1') t' \.I). into this expression. o 1. we note that the maximum displacement the point x for which )"(..4 BUCKLING OF A LONG COLUMN Let us now consider a long column or rod of length b that is subjected to an axial compressive force P applied at one end. we can use Newton's value. we form the sequence F(x) Xj+! = Xj - F'(Xo) ._ :.13 I J Long column under axial load P.

. mak ing it probably the oldest known eigenvalue problem. the column is stable and will remain in its undeformed equilibrium position. where we have introduced p }.3.Py" were acting on it. .3.) leads to the elgenequarion o < . If the axial load P applied to the column is less than the Euler load.5.. F. (57) . > 0 for buckli ng problems. These columns normally support the weight and loads of the structure above them. Thus. For sufficiently large values of P. bridges. y(b) = y"(b) = 0. /I = 1. and this weight induces a compressive force P on the ends of the column.! (58) Equation (57). buildings.. The allowed values of" corre spond to critical buckling loads given by t ions. called the [undamental buckling mode.2. EXAMPLE 16 Determine the possible deflection modes that a simply-supported column of length b will assume. Slender rods or columns are conventional in many structures such as aircraft.. )"(0) = y"(O) = • Ph)·sical considerations sliggesilhal ). (59) The first critical load PI is also known as the EIII<'I" load. along with prescribed homogeneous boundary condicon stit utes an eigen value pro bl em. the theory. the column will suddenly bow out of its equilibrium state-caHed a state of buckling.~"'-' PHYSICAL APPLICATIONS 41 long we mean 1h at the len gth of the col umn is rnuc h greater than its diameter. whenPreaches the value PI. The substitution of q{x) = +Py" into Equation (5]. Due to Euler's mathematical analysis of columns.r < b. and so forth.=-. We wish to determine the possible modes of lateral deflection from the equilibrium position (y = 0) of the column along the x-axis. However. This problem was di scovered by Eu Ie r in the eighteenth cent u ry. we can use the principles of elementary beam theory as applied in Section 1. engineers were finally able to develop efficient designs on support columns in order to control buckling. . in determining the possible deformations.' The axial load P affects the column as though a distributed lateral loading of intensity q(x) = . SOLUTION The problem is characterized by 0. and calculate the largest load the column can support before deformation may take place.. predicts that the column will assume the shape (under a small disturbance) predicted by the first eigenfunction <!> I(r) .

42 CHAPTER I BOUNDARY VALUE AND EIGENVALUE PROBLEMS where we are recalling the boundary conditions (55) for a simply supported column or beam. The largest load the column can support loud" given above.(x) = C• Sin . can be increased if the length of tbe column or beam is decreased. • Notice that t he critical load P. higher-order buckling modes may sometimes be realized. It is for this reason that we often find cross-sectional shapes in beams thar re semble tho: lcue rs I and H.3. C4 sin kb For nontrivial solutions from which we deduce to exist. and the remaining conditions at the end x = b lead to C:b + C. Although the mathematical theory predicts an infinite number of possible critical loads and deflection modes. However.. 90X =C () = r: . only the Euler load and corresponding fundamental buckling mode are actually realized by the column in most situations (if buckling does indeed occur). C) = CJ = 0. :£1 ¢. . we must select k = nuib and set C~ = 0... of the deflection. The critical load can also be increased if the lIIoment of inertia I of the beam is increased.. particularly if the column is restrained from movement at some point between its endpoints . Hence. The moment of inertia is [arge for cross-sections Ihat have their masses "far" from the centroid of the beam. before possible buckling is the Euler P'=Y' corresponding to the deflection shape .2. . the critical With X defined as one of the eigenvalues buckling loads become n = 1.. 3. The general solution of this DE for X .. mrx n = I.. b' TIX Once again the linear theory does not provide us with a value of the amplitude C.. kl > 0 is y = C! + C2x + C) cos kx + C4 sin kx. O. and applying the boundary conditions at the end x = 0 requires that C! + C) = 0 and C) = O. 2.. . sin kb = = 0.

y(O) =0.+ -. 0. I.1"<1 I < .@. = --. (h) If greater accuracy is required in the displacements erning DE should be replaced by the nonlinear DE of the cable.4. solve the resulting DE by separation show that the solution satisfying the prescribed boundary (l '" i3 = 0 is y= (c) of variables and conditions when _I__ /IIg {COSh mg ::. (a) Show that the cable hangs in a parabolic arc. = j'..(OJ = O. . 0<.5 In Problems 1-6. which acts like a uniformly distributed load. the gov- By letting \. 0 )'(~l = 0 .cosh [mg 2T approximation cosh . EXERCISES 1. in.t _ !)]} . 4. = !J'. 0. \.: 1" H y(O) = u. Also find the maximum displacement of the string.8.I" - 5. )'('1:)=0 y(l) = 0 )'(1) = y" y" >= -xl I -(. Under the assu mption that /IIglT = I. . *6. A heu V)' cable bet ween supports at .r < 2. 2 Using the polynomial I + . . O. 3. 0. reO) = 0. * 7. compare the numerical values obtained by the solution in (a) when a = i3 = 0 and the solution in (b) at the points x = 0. y'(O) = O.2. The stalk displacements of t he ca ble therefore sati 5fy [he bou ndary va lue preble m .. determine the displacement curve of a taut string subject to the prescribed external load and boundary conditions. 0. l :>< T (.5 43 EXERCISES 1. y(l) = 13· where IIIglT is const ant.r}. I" • = {- I. at rest u ride r its own weight.r = 0 and x = J ha [lg. 2\ 4\ l~ z' estimate the displacement of the cable at x = 112 by the solution given in (b) and compare with the result obtained by using the solution in (a) when (d) a=fJ=O. y"=-sirl:r.6.

e .. Sol ve t he rotating preble m desc ribcd by -r v" + /--. Assume t he boundary conditions are )"(0) = 0. of equilibrium becomes d -d [T(. is characterized by v"> -x. (b) the physical conditions are as in .I"). (a) determine whether A '" 0 is an eigenvalue. fixed at the end x '" 0.r)y'] = -q(..1'(-(1) = O. connected to a spri ng wi th spring con srant K atr = I. . 14. Find the subsequent 9. show that the cigcn value s must sati sfy the transceride nln! equat ion . Sol ve t he rotating string problem when (a) one end of the string is fixed and the other is free. and subject to a transverse force of unity at x '" 1 and a transverse force q(. displacement..r(a) = O. . 1 BOUNDARY VALU E AN D EIGENVALUE PR01JLEMS The tra 11 sverse displacerne fit of a string under constant te nsion (T '" 1). . If the end x = h of a rotating string is attached 10 an elastic spring but fixed at the endr = 0. (<1). '" 16. . .«kb cos kb where F = h and the boundary by aby'{b) = sin /.1"(0) '" O. y(O) '" 0..1') = gi ven by Whcn the tension T in a taut stri ng is not constant.1"(1) + Ky(l) '" l.1' = 0. When a rotating st ring is subject to the nonconstant eigenequation has the form (see Problem 9) tension T(x) = Xl. (a) show thai the governing DE then takes the form as "" Ax is not valid for the .. 11. Assuming boundary conditions y(l) = y(e) = 0. u< X < <I.. 51 ri ng 12. but . Solve the equilibrium equation in Problem 9 when the tension I/( I + r) and q(x) '" I. described condition = -J(b) at x = b is mathematically (a constant). Under the assumption that the approximation rotating string problem.44 CHAPTER 8. 13.. (b) Find an approximate expression for the larger eigenvalues. 1 <X < e. the ! (x'y') + pwly = 0. x 10. i. --..b.:' . find the first critical speed w. For Problem 13. y(o) = 0.:.'.r) = x per unit length along the stri ng."(h) '" 0: (e) both ends lire free. r(l) = O. and corresponding deflection mode. 15. show that the linear equation ')j .~ is 1"(..

and y' = -d . . X=-. )"(0) = y··(O) == 0.~.. (b) the bending moment. y(O) = )"'(0) = 0. show that the buckling p modes are also The beam in Problem The beam in Problem By integrating Equation (57) twice. 21.y '" 0. the boundary When both ends of a column (of u nit length) under an axial compressive conditions are }"(O) "" )"(0) (a) = 0. P are filled. 19.r. S constant. Show t hat each set of eigenvalues in (a) leads to a different type of deflection mode by findi ng two sets of eigenfunctions. 23.5) to the location uf and maximum displacement of the following beams.1'"(1) == y"'(1) zo. /" = I. y([) y(l) = y"(1) = 0 = y'(I) =0 y'" =.. 19. y(O) = y"(O) = o. . 22 and 13. Using this result. )"(1) '" y'(I) = O. 0< x < b. r.r. = With the added assumption in (a) becomes yH that T ~f = T. .f ily' In Problems 17-21.. )"(1) == ). 18. y"J =x.'(1) = 0 . (a) determine the maximum load P that the rod can support without buckling. 0< x < b.) == y"( . ylO) == y'W) = 0. where 0 is a constant.. y'" = . ) = 0 In Problems approximate 22. respectively. when when the ends of the column are simply supported.-:. solutions of the second-order DE y" + I. == 0 r" = sin . assuming EI constant.. EI i. :r::-.r. EXERCISES -(I U 45 tis (b) ( T- JY) tis + pw'y "" 0. 18. show that the DE + AyVI + (y')l = 0. 24. of the equations SIIl- for this problem which are (b) where k' = X.-. y( . and (e) the shear force of an elastic beam subject to the prescribed external load and boundary conditions. 17. (b) If a load P = (lEI is applied.k "" 0 Z ' tan '2 = '2' k k .e. use Newton's method (see Example 15 in Section 1. where A = pwllT. determine (a) the displacement curve. what is the length of the shortest rod for which buckling is likely to occur? *25.. )'(b) = y"(b) = O. y(O) = yOlO) = 0. y(O) = y'(O) = 0. force Show that there are two sets of eigenvalues solutions.

jfj. (II "" !.46 CHAPTER 26. Solve Problem 26 for the characteristic speeds when the origin is located at the center of the shaft. and Al(x) are continuous functions in the interval [XI. Xl].. 2. DEFINITION I. q(x) and rex) are all con tin- where p(x) > 0 and rex) > 0 in (x). For purposes of developing this theory. second-order adjoint form if and onl y if it h as the form p(x)y" or equivalently. If the rotat ing shaft described in Problem 26 is simply-supported at x = 0 while the end at . 27. 2. . (a) show that the critical angular speeds are given by w.6 STURM-LIOUVILLE THEORY In this section we wish to examine more closely some of the general theory concerning eigenvalue problems. which is one of the deepest and richest theories in all mathematics. For example. the self-adjoint form is general enough to embrace virtually all second-order linear DEs. .4) to approximate the ~r~1critical speed w" 1.r = 2. 0< x < b. that is simply supported <Iteach end. *28. are solutions of the eigenequation y''' . where A = pwllEI. :r [p(x)y'] + [q(x) + Ar(x)]y = 0. use Newton '5 method (see Example 12 in Section 1. (b) (c) where k. Find the deflection modes corresponding to the critical speeds given in (a). i. /I = 1. and p'(x). A I(X). I BOUNDARY VALUE AND EIGENVALUE PROBLEMS The deflection modes of a straight shaft of constant density p. ..J A homogeneous. .. For the case b = I. ) are the roots of tanh bt: = tan bk. it is helpful to introduce the notion of a self-adjoint DE...2 and .AY "" 0. linear DE is said to be in self + p'(x)y' + [q(x) + Ar(x)]}' = 0.r = b is free.. Find the characteristic speed s and deflection modes of a shaft of length 4 ft. rotating with uniform angular speed ~) about its equilibrium position along the . the ends are located at x = . XlI. Xl). e.. the general eigenequation of concern here is of the form (60) where Ao(x). "" k~. uous functions in the interval [x.r-axis. 3. -i-_ Despite its appearance.3.

x > 0. (61) Also. * This DE is nOI in self-adjoint form unless A I(. (66) called a se/f--lI(/joim operator.{xl.(x)y' + /l(x)[Ao(x) + >. it is notationally convenient to introduce the special differential operator (where D = dld.I..(.~"" STURM-LIOUVILLE THEORY 47 and where Al(X) > 0 on this interval. (61) Eq ua tion (6 J) is now in self-adjoin t form provided we pick p(x) such that p '(x) ".-1. we identify the functions with the self-adjoint form of Definition (64) and rex) "" j. (67) In all the remaining sections of the text. Al(x (62) By solving this first-order DE for p(x) ~ p(x). A.r{x)y = o. EXAMPLE 17 ff Put the DE X2y + :cy' + >.\") L = D[p(x)Dj + q(x).6.y "" 0. • If A . In terms of L. homogeneous second-order equations.. However. producing p(x)yW + /-L(x)A.l) == _-.(xj p(x) (65) In addition to providing a "standard form" for all linear. in self-adjoint form... we can easily transform (60) into self-adjoint form by multiplying through by the function /lex) = p(x)/A2(x).(x) '" 0 at one or both endpoin ts of the interva] of intcrest are discussed in Section 1. we findt (63) exp[f ~:~:~ dX].. Those cases for which A.. ) < 0.t) ee Ai(x). we 5inlply multiply (60) by . by further comparison of 1. t An y (nonzero] mull iple of (63) can also be used for .jy = O. the self-adjoint form enjoys certain operational ad-vantages over other forms. . we can express a homogeneous self-adjoint DE in the compact form L[ yJ + >. = /l(x)AI(x) = p(x)A I(X) ).. Since much of the subsequent discussion is centered around self-adjoint equations.L(. we will use the symbol L exclusively 10 denote a self-adjoint differential operator.1.

48 CHAIYfER SOLUTION I BOUNDARY VALUE AND EIGENVALUE PROBLEMS From (63).2 A self-adjoint [. as illustrated in the next example. we first determine p(x) "" exp(f . In addition to being self-adjoint. elf) ~ x.2 relates the symmetry property to the operator L. I] (b) yeO). In this regard it is possible that a given operator L is symmetric with one set of boundary conditions but not with a different set.IL[lIjJ dx '" 0 [or any functions II and I' having continuous second. an eigenvalue problem can have a self-adjoin: operator bUI!lOIbe a selfadjoint problem.tl. Thus.y( I) "" 0.t:] if and only jf operator L is said 10 be symmetric on the interval F' " (ilL [1'1 . Because this terminology call be confusing. and multiplying or equivalently. lL{f) = p(x)IA!(x) /J. REMARK: Eigenvalue problems for which the operator L is symmetric are also referred to as self-adjoint problems ill inucl: of the literature. DEFINITION!. ~ (xy') dx + (~) x y = O. . y( I) = 0 y'(l) = 0 on [0. .order derivatives on the interval and satisfying the prescribed boundary conditions associated withL. we will subsequently find out that this property is closely related to the kind of boundary conditions prescribed along with L. EXAMPLE 18 Determine whether the self-adjoint operator L := D! is symmetric with respect 10 the following boundary conditions: (a) y(O):::: 0. there is another property of the operator L that we desire in much of the following discussion. we have opted to lise the term symmetric operator (ill place of self-adjoint problem). which has a certain appeal because 'his same term is used ill matrix theory 10 refer to a special matrix possessing properties similar to those to be discussed. Although Definition 1.(x) leads to = 11x. the original DE through by Thus.

we find [u(x)v"(x) .v(l)u'(l)J . [p(X)W(If. (pv') dx = + :. it follows that u and v satisfy 11(0) = 1'(0) = 0 and lIe I) = \.\.(11(0) 1"(0) . .v(x)/I"(x)] dx "" u(x)v'(x) = [1I(I)v'(J) .2 and usin. 1':.\'(1) = 0. the righ t -hand side of the above int egra I vanishes and we conclude that L '" D~is symmetric in this case.(I) '" O... (X d which is what we wanted to prove. Lugl"(lllgc identity:" If L '" D[p(.. who is known for hi s many cuntributions in mechanics.1. " !!.1/(0)\"(0). and partial differential equations.(. r)(x»).I'(x)u'(x) 1. For the boundary conditions in (a).~)D) .. acoustics. j: . He also developed til e method of variation of parameters used in sutvi ng non homogeneous DEs. we have = u. In the case of (b).Yl" .STURM-UOUVILLE SOLUTION THEORY 4 integration f Substituting L = D~ into the integral in Definition by parts.1 the right-hand side of this last expression is not necessarily that L = D~ is not symmetric in this case. PROOF With L IIL[I') = D[p(.!_ (pu') . number t henry. I') d '= - dx [p(xJ W(II.\L /11] where \Y(/(.. zero. reO) . = Ill"~ - I/'l' is the Wronskian function. Based on these relations. v!. dx . algebra. Hence.\'L[II) + q(x).. and if II and \. the boundary conditions lead to 11(0) .yt.!)DI + q(x) is defined on the interval [x I./I(!) = O. the above integral reduces 10 1 \ [II(X)\'''(X) - r(x)/I"(x)] dx '" 1"(0)11'(0) . are any functions with continuous second-order derivatives on [x" x~l. and 11'(1) = v'( I) = O. then IIL[1"1 . Because we deduce LE~I~IA 1. -.r(O)u' (0)1. pit 'I') Fl(x)]. calculus of varia! ions.-h '" .:!._ dx {PI" /l - '" -. : "_: -.. /'. 'This identily is named afler the French mathematician lOS EPH LOUIS LAGRANGE (1736-t8 13).

1 PROPERTIES OF A SYMMETRIC OPERATOR Once we have established that a given operator L is symmetric..50 CHAPTER I BOUNDARY VALUE identity. X1]' \. x. I'l(x)] dx. (ilL{1'] .. - . I from which we deduce f' (uL{v] vL[/I]) dx = p(. there are several important consequential properties associated with the eigenvalues and eigenfunctions of such an operator. v)(x) I:: ' . (68) of one-dimensional calculus. which is just one of several similar formulas all referred to as Gru"·$/or"'''/(l or Grecn:« iJentiry.)(X)r~ = 0 for any functions /I and I' that satisfy the prescribed boundary conditions associated with L and have continuous second-order derivatives on the interval {XI.3 A self-adjoint operator L = D[p(x)D] the interval (Xl. + q(x) is a symmetric operator on 1.6. - • Equation (68) is also called the . ." f·" x. symmetric form" of Gree n 's formula.x [p(x) W(u. We refer to (68) as Green's formula! which is a kind vers ion of Gree n' s theorem in the plane from ad vanced grange's identity and Green's formula are fundamental of boundary value problems. AND EIGENVALUE we find that PROBLEMS Based on Lagrange's f·'. . Both Lato the general study Green's formula. Perhaps the single most important of these properties is the orr/lOgonaliry of the eigenfunctions (see Theorem 1. X1] if and only if p(X)\Y(II. Let us start by defining what we mean by saying two functions are orthogonal.r) W(II.1-[[11]) dx . The following theorem is a simple consequence of THEOREM 1.4).

r.. j" j(x)g(x)dx = 0. = 2 sin x cos . they are said to be orthogonal on this interval if and only if . ee) with Weight . we find -o-x)e-{ = 1-) =0. x 3 _.3 Iff and g are integrable functions on the interval (x" X2).:_:. ~I REMAR K: The interval of orthogonality in Definition 1.. EXAMPLE 20 ~ ~:_ _ _ Show thatf(x) = I and g(x) "" I .2 cos. EXAMPLE 19 Show that fix) = sin Zr and b'(x) = cos x are orthogonal weight function rex) = I. .[e-"dx from which our desired conclusion . follows._ . SOLUTION on (-n.~ SOLtrrlON Using integration [e-"O-X)dX= by parts. we obtain I" which is what we want to show.3 rna)' be of infinite extent ill some cases..STURM-LIOUVILLE THEORY 51 DEFINITION J. To) with By using the identity sin Zr J:" sin 2x cos x dx '" 2f:~ x cos! x dx sin '" -. or it may be either open or closed 01 one end or both ends of the finite interval. they are said to be orthogonal with respect to a weighting [unction r(x) > 0 if and only if jX' r(x)f(x)g(x) dx = o. '" 0.x are orthogonal function r(x) = eX • on (0.

'. In the discussion to follow concerning orthogonal functions. If we multiply the first of these DEs by tJ!.(x) satisfy the relations . much of the development in Chapter 4 concerning orthogonal functions has a vector analog in three-dimensional space. r(. and integrate over the interval of interest. respectively. LeI L be a symmetric associated with the eigenequation t[y] operator on the interval [XI.T)4>. L[ 4>.thus. however.(x). /I oF k . it is the orthogonality of the set of eigenfunctions {<jln(X)} of a symmetric SturmLiouvi!le operator that is of primary interest.\").product definition of orthogonality from vector analysis. are any two distinct eigenvalues eigenfunctions <jl.2 that the integral on By hypothesis.(x) and <jJdx) are If ~" and ~/. and hence 1. I'ROOF The eigenfunctions tJ!. we obtain Because L is symmetric. orthogonal. that our given definition of orthogonality is a generalization of the familiar dot .\)] = ... THEOREM 1..(x)] '" . r . and for which we have the following important theorem.~"r(x)¢ L[ 4>. In fact. it follows from Definition the left is zero.. where we discuss Fourier series. of L with corresponding then !IJ. ~n # ~~.(.: . We previously assumed that all eigenvalues are real without showing .".T) dx = 0.(. . subtract the resulting expressions.e. we deduce that the integral vanishes and the theorem is proved.r(x)4>J.52 CHAPTER I IlOUNDARY VALUE AND EIGENVALUE PROIlLEt. Please note.(.(x) and the second by $n(..~.(x) ant! !IJ.(x) and 4>..4 Orthogonality.IS The real significance of orthogonality will not be realized until Chapter 4.r)4>.(x). x!J + ~I~X)V := 0. i.(x)..

.(x) IIx '" r .. This means that f" oll r(x)tj).. however.(X) '= o.(x)tj). respectively. i.I(X) '" Ll(h(x)] + Atr(x)tj).. which leads to a contradiction._~.STURM·LlOUVILLE THEORY 53 this to be true.: '. t .lf~{)$. by forming th e com plex conju gate of the above equation. p.6 ~.Suppose there exists some complex eigenvalue At with corresponding eigenfunction cp. 5CC R.6 is generally proved..f 1.'.r) and ~). Courant and D. and where t. and hence are necessarily orthogonal due to the symmetry of L." -+ 00 as /I _ co. - form an infinite sequence or- .5 The eigenvalues of a symmetric operator are all real.1 ...(. Afd"<XIs of M<Jr/rematic. Although Theorem 1.(x») + A. Therefore. but since the integrand is positive... It follows now that ~. In more advanced works on DEs. < A~ < .. Vol. :. L(cp. Since the opertor L is composed of real functions.(x)J + t. it does not guarantee the existence of any eigenvalues.{x) belong to distinct eige nvalues. AI and AI. THEOREM J..-.e. t For a proof of Theorem t . Our assumption that a complex eigenvalue exists must be false.6.. 412.{x) .2 < . Hilbert. its complex conjugate I eq ual s L. ~~_. THEOREf.5 states that all eigenvalues of a symmetric operator are real.(.Y) '= O. The eigenvalues of a symmetric operator dered in increasing magnitude so that t. which asserts the existence of an infinite number of eigenval ues starting with a smallest value. and the theorem is proved. • The proof of chis Iheorem can be omitted for the readers not familiar wich complex variables. this proof is not presented here. this in! egrai can never be zero. Theorem 1..l < t. We are now ready 10 prove our assertion for symmetric operators. The eigenfunctions can also be shown to be real.("1 r(x) 1 <I!t(x) 12 dx = 0. . I (Wiley: New York) 1953. we find L[$.. PI/pies..r(x)cpA.. * PROOF .1 .

respectively. 0 the r inte rpretat ions of the se bou nd ary conditions concerni ng the supports for a taut string undergoing displacements were discussed in Section 1.t). LLE SYSTEMS Most of the eigenvalue separated L[y] B![y] problems studied thus far have featured unmixed or Problems of this type are characterized by + "r(x)y = 0.2 in connection with steady-state heat conduction in a rod. and third kinds." it but since all and aJ~cannot both be zero simultaneously. follows that the coefficient determinant vanishes.) By considering the boundary condition at x'" Xl. Each kind of boundary condition corresponds to a different type of physical constraint at the endpoint. 0). Unmixed homogeneous boundary conditions are of three distinct varieties. which at either endpoint of the interval may assume one of the following forms: y=o y' = 0 hy + r' = 0 (II constant). Any eigenvalue problem belonging to this general class is called a regular Sturm-Liouville system. r.. Some of these physical constrain Is were discussed in Section 1. we can apply a similar argument to show that W(u. . (/uY(x!) + a!lY'(x!) = = 0 0 (all + Orl ¥ 0). i. (69) Bl[ yJ "" ally{xl) + a22y'(xl) (a~1 + a~2 ¥ where L = D[p(x)D] + q(. These are called. by definition. I U(XI)· I )'(XI) II '(XI) V'(XI) ! = III I u(xJ) '(XI) V~X(I» v XI I = W{Il. we consider two functions II and \" with continuous second derivatives thai satisfy the prescribed unmixed boundary conditions in (69). ..6. Thus. = 0. 1')(X)jA' "" 0.54 CHAPTER I BOUNDARY VALUE AND EIGENVALUE PROBLEMS 1. + aI2v'(xl) = all I{XI) 0. At x == XI> this implies that a II I t{x I) + GI21l'(XI) = 0.2 REGU LAR STURM-LiOUVI boundary conditions. V)(X2) := o. To show that the operator L associated with a regular Sturm-Liouville system is symmetric. boundary conditions of J/Jc first. we have shown that p(X)W(II. v)(x.3 it follows that L is symmetric.5. second..e. ·lntercllan8i n8 the rows and col umns of a determinant does nOI alter its value. and by Theorem 1.

e.4 in L. Also see Problem 41 in Exercises J . 1.n(x) and q. At X = XI.. Such problems are called periodic Sturm-Liouville systems. I Periodic Sturm-Liouville systems are a particular member of a larger class of St urrnliouville systems featuring mixed boundary condit ions. - The eigen values of a regular Sturm-Liouville system are slmple-i. (70) where L = D(p(x)D] + q(x) and where P(XI) = P(Xl). Since (ill and al1 cannot both be zero. -.3 PERIODIC STURM-LIOUVILLE SYSTEMS problems that we consider are those The second of the form major class of eigenvalue L[ yJ + Ar(x)y '= 0. i//I Applica/iQns Foresman: GI en view. If the Wronskian of two solutions of it DE vanishes at one point on the solution interval. each eigenfunction must satisfy the prescribed boundary condition so that PROOF a"lfJn(xd + ad)~(x.) = 0. :. Ill.) J 982.?~'-.7 ~I. it must be identically zero on this interval.4. .6. t We leave it to the reader this time to show that the operator L is symmetric for this class of problems (see Problem 7 in Exercises 1. * Hence.6) .:"'_ ..3. a lI<h(xd + (/L~4>~(XI) = o. the eigenvalues and eigenfunctions of such a system have the important properties stated in Theorems 1. the general forms of which are given by Equation (3) in Section 1. In addition. we have the following result for these systems.. we claim that the coefficient nant must vanish. • L . C. and thus (as above) determi- W(lfJm <jlJJ(. THEOREM ]. and 1..5.t) '= O. 1.(xj are proportional (linearly dependent) and therefore represent the same eigenfunction. only Assume 1fJ"(_r) nd lfJix) are di stinct eigenfunctions a belonging to the same eigenvalue An. q. Ord/mlT)' OjjJe'tII//ul Eqlta/iOlu . Andrews.~ • See Theorem 4. (Scon. one eigenfunction belongs to each eigenvalue.STURM-LlOUVILLE THEORY 55 Because a regular Sturm-Liouville system has a symmetric operator..1.6.

AG = 0.. Y( . I}o(e) = 1 constitute an eigenvalue-eigenfunction solution takes the familiar form y = C1 cos ke pair. for k = n ('I "" I.. we see that both C1 and C2 remain arbitrary. therefore. = p(11").y'(71) = C1 - C2 = 0..vn(e) = C) cos nO + C~ sin 110. Thus. kl > 0.56 CHAPTER I BOUNDARY VALUE AND EIGENVALUE PROBLElvlS A prime distinction of periodic Sturm-Liouville systems is that the eigenvalues are not necessarily simple. y'( -71") = Y'(11"). the periodic boundary conditions )'( .2. To insure that the deflections remain single-valued. .3. We conclude. : . . which in the general case we may n = 1.2. J. n=I.. the general which implies that Cl = 0 while C1 is left arbitrary.. EXAMPLE 21 At a particu lar instant of ti me. 2. SOLUTION We wish to determine = the possible modes of deflection.3. = 1. and thus conform to physical considerations.TI) . . 1IS illustrated in the next example..To) = Y(71"). the transverse deflections of a vibrating eire ular membrane at a fixed distance from the center of the membrane are related to solutions of the eigenequation -n< e < rr. are imposed. so that p( . O..2. the general C1 + Cleo we must satisfy In order to accommodate the periodic boundary conditions.2Cl7l" = 0.. and .3 •. 3.yeTI) = .n) y = When >-. where y = y(e) is a function of the polar angle e. that to each eigenvalue n there corresponds write as two eigenfunctions. This time the boundary conditions lead to (upon simplification) k71 = :~ Hence. Here pee) solution of the DE is I. = 0.. For A = + C2 sin C1 sin i«. y'( + rr] .

1") and ~I.4 A Sturm-Liouville system is said to be singular if one or more of the foliowi ng event s OCCli r on the interval (XI.STURM-LIOUVILLE THEORY 57 where C" Ci. we observe that . 2.r ) ~ 0 in Eqll~!i\>n «(. we not e that for second-orde rOEs no more th an t B'O linea rly i ndepe nde nt eigenfunctions can exist for a single eigenvalue.. 1. . .6.{O) are linearly independent. by making this choice of eigenfunctions J:" cos I/O sin /]0 do '" 0. =0 • Th is Sil uat ion corresponds to A . These s i ngul arities change the general nature of the system. DEFINITION 1. For instance. .(6) = cos l1a.(0) and !J!.3.(. ~'"(O) = sin /I /I = I. Verifying that no negative eigenvalues exist is left to the exercises. /]=1.."' (b) p(x). .2. ensure that the resulting eigenfunctions are mutually orthogonal.. As a final comment here concerning the multiplicity of eigenvalues. if two linearly independent eigenfunctions <"P.2. . 1n general.(. /I '" Moreover.(I).(I"(x) can be found corresponding to a single eigenvalue A. i. by choosing Cl = C) = 0 (and CI = C4 '" I) in the result of Example 2 J.. especial! y in the form of boundary conditions necessary to ensure that the operator L is symmetric.... 3. a. Xl]: (a) P(XI) 0 and/or P(X2) = 0.6. or rex) becomes infinite at x "" XI or x = Xl (or both). and (c) either XI or Xl (or both) are infinite .. q(x).. not only arc cos n 0 and sin 116 the simplest combination of linearly independent eigenfunctions in Example ~ I.. but they are also orthogonal. 1. CJ and C are constants such that <1>. In situations like that in Example 21 we sometimes select the simplest combination of linearly independent eigenfunctions. A general procedure for obtaining orthogonal eigenfunctions in such cases is di sc ussed in Problem 25 in Exercises 1.r) that arc also orthogonal on the specified interval.3 a result that is easy to verify by simple integration methods.4 SING ULAR STU RM-LJOUYI LLE SYSTEMS Many of the most interesting and most common Sturm-Liouville systems in prac lice are classified as singular as def ned below.( . we obtain the linearly independent eigenfunctions 4>.." it is always possible to find linear combinations of 4>.(x) and \. Hence.e.

. where also p(x) = e-': .. To ensure that a singular Sturm-Liouville system has a symmetric operator. • Note also that p(..... we require Legendre's f': .."..3) where u and v are any continuous.. we denote a left-hand limit by x . but also one endpoint of the interval is infinite. x...xy = 0.. by prescribing the condition y(x). .] . we find p(x) = x vanishes at x = 0.. and for v¥.x2 is zero at both endpoints For Bessel's equation.. (75) Of course.(xe-'y') + ~e-Xy 0. twice differentiable functions satisfying the prescribed boundary conditions of the Sturm-Liouville system. any condition of the form and (73) is satisfied by prescribing alIY(xZ) + anY'(xv = O.58 CHAPTER I BOUNDARY VALUE AND EIGENVALUE PROBLEMS Some DEs that fall into this category include the following: !!_ [(I dx d dx (xy') - X2)y'] + ~Y = 0.. Hence. 0 as x . (74) finite as x . then (72) is satisfied. we select this weaker modified type of boundary condition in most instances. For example.... righi-hand limit.0. '{ .. conditions other than (74) could be prescribed in order to satisfy (72).." In the case of H erm ire' s e quat ion. we mean a. f By x ... \')(X)I" XI = 0. as Ixl .. y + >. the s ingula rit y is due entirely to the infinite endpoints of the interval. (ilL [\. = O<x<b O<x<o:: -x<x<oo (Bessel equation) (Laguerre (Hermite equation) equation) dx d . ..d = xe:? . equ ation is singular si nee p (x) = 1 ...T... the function q(x) = '-vI/X becomes infinite atr == O... but any other condition is normally too restrictive in practice to allow for a nontrivial solution of the problem. -1<x<1 (Legendre equation) VI . y'(x) P(XI) = 0. xt . ee. v)(x) = ot (72) (73) When the singularity arises specifically from for instance. ..r = Xl> we impose boundary conditions such that ~-Ai lim p(x)W(u. 0 ""... goes to zero... (71) (from Theorem 1.r = :!:: 1.. if there is a singularity at .-. xi. Laguerre's equation is singular at x = 0 since p(x) = xe -.. Similarly.I'L[u])dx = p(X)W(II.-:-. .

the eigenvalues are not discrete as is the case for symmetric operators. we impose the boundary conditions y(x).. For example.4-1. we will require that y(x) arid y'(x) remain finite as x --> Xl-' Finally. That is. 0.. For example. but form what we call a continuum of eigenvalues.. The corresponding eigenfunction is <!lex) = sin kx. y' o <x < =.t --> x.UTlO. for Example 22 iJlustrates that nonsymrnetric operators can lead to results quite d ifferen t from those associated with symmetric operators." In these cases we do not expect the eigenfunctions (if any) and eigenvalues to satisfy the conditions of Theorems 1. Consider the following example.. This is the case. SOl.fPLE 22 Find the eigenvalues Y" + AY = 0. '" The first boundary condition requires that C1 y = C1 sin kx.. Sometimes the singular eigenvalue problem does not involve a symmetric operator. finite as x --> ec... b.• (76) Singularities arise quite often in practice when a boundary point of the problem is not truly a physical boundary. y'(x) finite as . and thus Both y = C2 sin kx and y' "" kCl cos kx remain finite as x --> "".: O. since p(x) = I for all x.6. if p(x I) = P(Xl) "" 0. The point r = 0 is not a physical boundary of the circular domain. for example. and x --> X2. and so we conclude that A = kl is an eigenvalue for VII)' rea! (positive) number k. • This problem is discussed in Sec tion 6.'! and eigenfunctions of y.4. EX:\)o.. STURM-LIOUVILLE THEORY 59 The case of a singularity at x = X2 is similarly treated. For A = k2 > y 0. The operator is nonsyrnmetric in this example because the boundary conditions are not appropriate for (71) to be satisfied.[d ---~------~-~ - - . . For example.. and yet it is clearly a mathematical boundary. In such cases we generally have 10 impose the condition that the solution of the problem remain bounded at r = O.. sin h. We leave it to the reader to show that there are no further eigenvalues A'. if it should happen that P(X2) = 0. the general solution is = c] cos kx + C. it does not vanish in the limit X_"". when we consider the temperature distribution in a long rod that is mathematically modeled as infinite in extent. inside a circular domain of radius b we restrict the radial variable r such that 0 .r i. )"(0) = 0.

.2. put each equation in self-adjoint AY '" 0.(xj'" C. .~). . sin tu.• becomes normalized on (0. J.y' + AY AY' + (I .1.(xWdx '" 0. 'IT)..3 . becomes normalized on (0. (b) (c) .(. satisfying the conditions 11. = I..60 CHAPTER I BOUNDARY VALUE AND EIGENVALUE PROBLEMS EXERCISES 1. . Repeat Problem 10 for the set of polynomials (a) p . and p.. H.P. An orthogonal form an orthogonal 9. Determine the constants C.. . 1. verify that the following functions on the interval (0.t + <. 7. . 'IT): (a) $.. xy" 1-6. .t1lY· - + Ay '" 0.(x) '" b.. + >0 . Determine the first three mernbe rs of this set. (b) $. is said to be normalized on the given interval.r'y· + xy' + (Ax' . 10. show that the Lagrange identity takes the form /lL[I" - d I'L!Il] '"' dx (III'"' . 11 = 0. /I r' " = 1.. so that /I '" I. n yO k.1/'1'" + /1"1").11"'\' . 2. 1. Use this result to show that if of. (b) p.(x} are eigenfunctions order eigenvalue problem of the fourth- . ..2 •. and *12. .1I')y = 0.2 .Liouville system has a symmetric Given that r(x) = 1. 1/ = O.. 6. 2. (b) 4>..(XldX'" O. 3. I. x> 0 operator. + c..~)= cos /IX. 4. Set PG(x) = a" p.xlY' 2l)" + AY '" O..(x) = c.(xj) satisfying the relation [4>. (I . + bsx.and satisfying the three (cl Hint: f_'.(x)l' dx=. x =0 form.. y" ..(x) and of.(x) = C cos IIX. and solve for the unknown constants..I <x <I X> 0 .\") = sin /IX. It ¢ k. I. set Verify that a periodic Sturm..6 In Problems I. If L = DO. set of functions {$.OI'" I. be an infinite set of polynomials conditions (a) p .(.(x) is of degree It. 2. . II = O.r > 0 2. 1. 5... . .x'. 'IT): (a) 4>.(x) is of degree It. + 2r'y' + AY = 0. . x'(x' + I)y.(x)P'(X)dX e-'[p. /I '" 1. II = O.f [e-'p. Let p"(.(X)P. 3 .

'" 0.~ -"<. and X. .y"(I) =. so that 1'. . LC!..~) are orthogonal on (0.~)g.. find the eigenvalues and eigenfunctions of the regular SturmLiouville system.' . In Problems 17-19.4) that the eigenfunctions of the given Sturm-Liouville problem are orthogonal without explicitly solving for them.(x). r(x)[.~". then $"t~) and 4>.- X . . d. "" 0.- _" In Problems 13-15.(1).. 18.ry )"(0) = 0. Show that there are no negative eigen values in Example 21..'.1) = ).{x) '" ""tt) - a$. find the eigenvalues and eigenfunctions of the periodic SturmLiou ville system.. = O. correspondi ng to distinct eigenvalues X." See Problem 41 in Exercises 1.3. y( t) 1.. y(O) = y(I). 14. . show that for some J'l}(X} = C..~) '" ¢'~(X).r(x)D1J + D{p(x)Dj + q(xl.r /(0) = 0..) '" 0 y(5) = ° )"(1) '" 19. y'( . .1'" .Jy' + Xy = 0. y( . y(!) ° ° ° '" '" 16.I) = y'( I) y'(O) = y'{2-n') y'(O) '" y'( I) y.r' + 1 y = 0. -d ICx. < of < x: corresponding to the same eigenvalue X •• Given the linear combinations f. 8. 15.6 61 =. -. and state the orthogonality condition.. 25. . ]'(0) = 0. J). = Show that X 0 is an eigenvalue of the regular Sturm-Liouville system dx [p(x)y'] d + Xr(x)y = 0. y'( . where L is self-adioint. . Note: A fourth-order differential operator L is called self-adjoint if it has the form L = D~[..(. y" + X)' = 0.)13. ° y'(I) = O... Hint: 20. y'([) . p(x)W(y" Him.EXERCISES y"'-Xy=o. y" + X(I + x)y + X)' = 0. J(O) = 0. y(l) '" 0. 21.. y'(2) = xy.4. + Xy "" 0. . Let $. and)'l are solutions of L[ constant C yJ - 0. 22. ~'.(x)dx = o. xly.0. + y' + X.. ~.. In Problems 21-23.. ". )'(0) = y(2Tr). show directly (without Theorem 1. If)". . n )'" Xy + 24. xly" y(1) = 0. . 17. (See Problem 26 in Exercises 1..t = tan O.t) and "'~tt) be linearly independent eigenfunctions on the interval x. + 3xy' + Xy = 0. ". find a value of Q -~.~ . yeO) = 0.+ l)y'J + -.

62 CHAPTER I nOUN DARY VALUE AND EIGENVALUE PROBLEMS 26. y'(O) '" y.)y'(x.y = '" 0.. (b) If Ihe boundary cond itions are such Ihat p(x)"}" 'I" = " 0. -I <x < 1 d. are strictly positive unless q(x) "" 0.. _. O<x<! Given th e eigenequ31ion d [p(x)}"·) dx (a) multiply by .7 APPROXIMATION METHODS: WEIGHTED RESIDUALS Perhaps it is correct to say that most mathematical problems encountered in practice are either difficult or impossible to solve exactly by known analytical techniques..1"1 r p(X)(. y'(O) = y'(I). o 1..t~ly'l + xy = necessary 0. they frequently are too cumbersome or complex for interpretation and numerical . ui« Use Problem 3I(a). = O.. are strictly positive. x •~ I. deterrni ne the proper types of bou ndary conditions to ensure the orthogonality of the resulting eigenfunctions.I" '): - q(.------. + I.f·' jj r(x)y' dx + p(x)yy' I" rL = o. 30.. 0. Show that)" '"" is an eigenvalue y. finite as In Problems 29 and 30. show that the eigenvalues ·32.~»)"'l dx + I.1) and integrate + [4(. y' 0" • x- ·28. y' finite as y. show that all eigenvalues in which case they are non-negative.y Find an orthogonal Hilll: See Problem = 0...._-_. ° of multiplicity two for y'tO) + yeO) = y(ll. and if q(x) . 0. find the eigenvalues Liouville system. x'yff + xy' + I. boundary conditions such that the and q(x) < 0. -d (xy) + )"xy '" 0 . Even in cases where exact solutions are explicitly obtained. _. by parts sho«: that . 0.f" .I". pair of eigenfunctions belonging to I.-------------_.. • 27. If a Sturm-Liouville system has prescribed eigenfunctions satisfy the ineq ualities )"(..) .. and eigenfunctions x- In Problems 27 and 28. .y I. of the singular Sturm)'(1) = 0 "" 25.1) + hr(xlh' 10 '" O.'"(. ~ [(I . 29. y" + 0.

(77) where L is a self-adjoint operator" and 81 and Bl arc unmixed boundary operators. Essentially.. we sec that the trial solution lj! has been chosen to satisfy the prescribed boundary conditions exactly for all choices of the unknown c's. the technique is based upon a procedure of selecting a finite series of "standard functions. N are linearly independent functions chosen so that they satisfy the homogeneous boundary conditions j = I. even though the numerical solution may be more accurate. by our choice offunctions Ko(X} .7.3. Therefore.. Because of this. EQUATIONS 1l01l/JO/IlOgCll(. Historically. PU1 into . is first self-adjoint form. REMARK: 1/ a = 13 = 0. approximation methods are often preferred in practice 10 numerical techniques. The c's in (78) are unknown constants." frequently simple polynomials... which are all based upon minimizing the "error of the fit. We start by construct ing the linear trial solution I!r "" go(. g/r) . J. It is then forced to satisfy other physical or mat hernatical characteristics of the problem by anyone of several method s.O. '. . (78) which approximates the true solution y of (77). • Greater accuracy is achieved in 'omt approxirnarion methods when the OF. 2. with unspecified coefficients.. Analytical "solutions" generally provide more qualitative information and possibly a better understanding of the physical problem being studied than numerical "solutions.r) + j~1 L N Cj gAr) . .I' To begin. it:. -. but is selected in such a way that it satisfies exactly the prescribed boundary conditions. .U' go(. and the gAx)." l. j = 1. .( yJ = /(x) . 2.N. it is for this reason that approximation and nurneri cal techniques have received considerable attention in the applied areas of mathematics. \I·e clIOO." which consist of a set of numerical values. .2. J. (79) The remaining function g~(x} is selected as any suitable function for which (80) Thus. let us consider the x) boundary value problem = < X < x:. . N.1 NONHO~lOGENEOUS I. B.r) "". it becomes either necessary or convenient to employ <In approximation or numerical method that will yield accurate numerical estimates of the true solution of the problem. j = I.[y) 13.. The most widely used approximation methods for boundary value problems have bee n unified in to one sta ndardized tech niq ue called t he method a/weighted residuals.APPROXIMATION I\IET!lODS: WEIGHTED RESIDUALS 63 evaluation. This finite series represents a trial solution or approximate solution of the DE.0I/.

..• N. but here we choose the N constants Clo C2. i. c].' .. . ln many cases we sim pi y choose th em so that they life evenly distributed throughout the interval. We will discuss two ways in which this is accomplished-the collocation method and Galerkin's method.e.. Here we simply set E. equations from which the c's REMARK: The collocation method and Galerkin's method are specific examples of weighted residual methods.{. (81) The object is to minimize E. 2.r in the interval if ~ were XI picked as the exact solution y. Galerkin in 1915 is perhaps the most widely used of the approximation methods. III general.{x) in some meaningful fashion. In physically motivated problems the general nature of the solution is often known in advance lind can be used in selecting these functions.. . (83) will also lead to N simultaneous may be computed.e.". weighted residual methods rely all setting the weighted integral oj the residual to z era.1. • The N point s should be chosen so t hat none of the trial functions g. . CN in such a way that k = I.. X. or practical considerations such as computational simplicity. Collocation Me/hod: The collocation method is the simplest \0 apply. (82) This action leads to N simultaneous (nonhomogeneous) equations in ("I. This problem is somewhat alleviated by taking N "sufficiently large..." Galerkin's Method: The method developed by B. physical considerations. set E. .(x) = 0 at . In determining the c's.. Accuracy in the technique is enhanced in general by taking larger values of N. we first define the error of the fir or residual EN(x) = L[~J .J(x) . X: . c:. choosing the g's is to a large extent arbitrary. (".r < X2. * A disadvantage of the collocation method is that the approximate solution ~ may vary greatly with the location of the collocation points. The general procedure is similar to the collocation method.. i. The location of the N points is frequently determined by intuition.dx) = 0 at N points distributed throughout the open interval XI < X < x:.. •••• Cs· Solving these tV equations for the c's yields our solution given by OR). 3. REMARK: Observe that Jar all .64 CHAPTER I BOUNDARY VALUE AND EIGENVALUE P)WIlLEII1S Other than the conditions cited. then E".(x) = 0 < . . (83) Hence.r = XI.:r)has a zero at any of the collocation points. G..

For a dre..2). IjI ~ 2x ~ I + C1. N.e).N lire collocation mel/rod Ihe weight =:! a~{- .Y cor!' 1912. Thus.).'"110:-> We fir.r .-1.r "" O.r). . 2. the general trial solution takes the Icrm ~ == 2x ..\JO Using the set of linearly independent - .Ct} .x)..d" . find an approximate solution of the boundary .. .~ r dis~us~jon of the w eightcd r~~idu:. N.API'ROXIMATfON METHODS./ \'I11""'". A function Ko(X) satisfying the prescribed f!onltomogt. i= I. For 11'. 3.\....I'{Oj"" -I. ...d For the = 1)1" + \)I -x +1 = x + c. So:..r) ~. see 11. . value problem ). = 0 at . where 0 is Ille Dirac delta function (. such as Gu(. i. Other choices for !. = I.lee Sectlo» 2J).e.(x) = .t methods. Using our first choice.t. The .t·hu·J H. 501.! + 2: To begin..t(l . collocaricn method _.. WEIGHTED RESIDUALS k".'lIeolls boundar)' conditions is given by Co(x) '" 2r .(1) = 1..smislie!> the IIOIIIOCI?Jleou./l'r. = In. leading 10 (he residual E1(. and VI (Ire case of Galerkill's meillod the weigh' finwiousll.·. 1.(x)}.s bound- arycondi i= 1.J(1 . Finlayson.(1') 65 .aM. -CIlS'iU (see Problem I in Exercises 1. .{II' c .2." ipkJ (Ac:"kll'lic rrr . let U5 Lake rhe case N N I-' cj..'t). J. 3..1.(X) ore chosen 10 be tire trioljullc/iolls C.7)..'u(x} are also possible..1:...1 + c ::' I (!_! _2) 2 4 we will s<!\ E.t observe t ions thateach g.(ir -.s.. Anolher weighted residual method Ihol is conunonly IIsed is the least squares melnodjor which EXAIIIPLE n polynomials g. i\. t ..2. .II <'IhrnJ . for SOllie S('I of weight jilllc/ions [unctians are {II' .



CI =

'1 "" 0.286
+ 0.286x(l x).


and thus ~ = 2x - I

The Galerkin method requires that


E1(x)x(l - x) dx ='0.



[Xl - X'

+ CI(X~ - 2x' + 3Xl- 2"(")Jdx



Evaluating the integral, we get

from which we determine
CI =

18 = 0.278.



Therefore, our approximate solution is
IjJ = 2x - I + 0.278x(1 - x).

If we retain two terms in the trial solution, then ~ = 2"(" - 1 and in this case E2(x)

+ clx(1

- x)

+ c1X2(1

- .r),


+ CI(x _x_2 -,2) + Cl(xl


+ 2 - 6x).

Collocating at x = 113andx

W, 'we obtain the set of equations
. 2- . .I




,_27,._.<" 3
C2 '" -,


16 -.5 -CI + 9, ,,27~ . ..

with solution

CI ""

+ "" 2x -

0.195 and


=-- 0.173; hence, in this case
- .r),

1 + O.195.i(1 .;:.x) +0. I 73x1(l

..... ;..


APPROXIMATION METHODS; WEIGHTED RESIDUALS TA BlE I. I Approximate Solutions of y~ + y by the Method of Weighted Residuals x _~. 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 COLLOCATION (ONE TERM) -I -0.774 -0.554 -0.340 -0.131 0.072 0.269 0.460 0.646 0.826
I =


x - I, ),(0)



I, y(1) '" I, EXACT VALUE -1 -0.781 -0.564 -0.349 -0.137 0.070 0.271 0.466 0.653 0.831

GALERKIN (ONE TERM) -I -0.775 -0.556 -0.342 -0.133 0.070 0.267 0.458 0.644 0.825

COLLOCATION (TWO TERMS) -1 -0.781 -0.563 -0.348 -0.137 0.070 0.272 0.466 0.653 0.832 1




All three approximate solutions obtained in Example 2] are compared in Table 1.1 with the exact solution (see Example 8)

sin x v"'-.-+x-l.



Find an approximate solution 1.5.1 using C}x) = sinjx,j= assume qolT = I.


the siring problem of EXample 14 in Section 1,2,3, ... ,N For the sake of simplicity,

The string problem is characterized
),(0) = 0,

y(-11') = O.

Because the prescribed boundary conditions are homogeneous, choose go(x) .,. 0, and thus our trial solution. is

we can.




sinjx .. - .

For illustrative the form

purposes we will select N = 2 so that the trial solution. takes '" = CI sin x

+ Cl sin lx.
sinx - 4c2 sin 2x

The corresponding

residual is

El(x) = "'~+ 1=

+ I.








We will set E1(.t)

0 at .r

= 7[/3

and x

= 27[13, - I,

which leads to

V3 - 2"" Cl

vJ Cl =



Solving for

CI and Cl

CI =

V3 =



so that for the collocation


1.155 sin .r.

For Galerkin 's method, we require that





dx= 0,







which leads to

CI = - = '<l'






Thus, for the Galerkin method our solution is
lji = 1.273 sin .r.




by the Method of Weighted Residuals x COLLOCATION
0 0.442 0.817

solutions ofy~:=: -I, )'(0) = 0, )"(1l') :=:0, with N == 2.
0 0.487 0.900 1.273 0.900 0.487 0

0 0.540 0.925 1.234 0.925 0.540


1T18 TIn.


0.817 0.442 0





The approximate solutions found in Example 24 are compared with the exact sol u tion y := x( 1T - x)12 in Tabl e 1.2. Neither the collocation method nor the Galerkin method led to a very accurate solution in Example 24, even though we started with a two-term trial solution. The reason for this is that the second term ofthe trial solution was rejected by both methods (Cl '" 0), and thus we essentially had only a one-term trial solution, wh ich is not generally ex pee ted to give very accurate results. To understand why the second term did not contribute to the solution, we note from the solution values listed in Table 1.2 that the solution is symmetric about the midpoint x = 1T12, a fact that can be deduced from the formulation of the problem. A trial solution should be chosen so that each function gJx) reflects this symmetry property. Note that the function gl(x) "" sin x is symmetric about x'" rr/2 but Kl(X) = sin Zr is 1I0t, and hence the reason that the methods required Cl = O. Only those gj(x) ee sinjx with odd indexj have the proper symmetry, and so in Example 24 a better twoterm trial solution is



+ c) sin


(See Problem 2 in Exercises 1.7.) However. in using this trial solution we must select the collocation points other than .r = 1T/3 and x "" 21T/3 to avoid the zeros of sin Jr. For example, we might choose x = -rr/4 and x '" 1T12.

Consider a uniform beam of length b that is resting on an elastic foundation with constant modulus K (lb/ftldisplacement). If the beam is supporting a uniformly distributed load '10 and is simply supported at both ends (see Figure I. 14), t he problem is characterized by Elyl41 + Ky "'"qs,
)'(0) "" y"(O)

0 <x < b
y(b) = y"(b) :: O.



= qo


Elastic foundation

FIGURE 1.14 Beam resting on elastic foundation.








Use the collocation method with two terms solution for the displacement of the beam.


an approximate

sional by introducing the new parameters

In this case it may be best to first make the equation


lI'(z) = -b4 y(z),


The problem now assumes the form





w(l) = w"(l) = 0,

where differentiation is with respect to z. For a trial solution we take
ljr(;:) = Cl

sin To:: + c-' sin hz, arc symmetric about I,


where we are recognizing the fact that the displacements In. Calculating the residual, we obtain E:(~)
= 11''')

+ k'w - I

= (If·

+ k4)c) sin





I.:')c-, sin 3lfZ -

and by setting

0 at z +

1/4 lind z

= 1/2,

we are led to the set of \12". I.


+ (RI-:r" + k·)c,

= =

(-:r + /.:"'lc, - (81-:r + /.:"'k.l the simultaneous solution of which is


= 2(or.


I 1.:') ,

\12-1 c -----l - 2(3 l-:r' + k4) . our solution takes the form

In terms of the original parameters,

til (.r)


qob4 [ . EI CI Sin

b + C-'


• Sill

37lX] b'



are given above.

For a fixed value of N, the collocation method is generally less accurate than the Galerkin method. The main advantage of the collocation method is ease of implementation, but the Galerkin method is generall y prefe rred when using only one-term or two-term approximations because of its greater accuracy. 1.7.2 EIGENVALUE PROBLEMS

The weighted residual methods are also useful techniques for approximating the eigenvalues and eigenfunctions of Sturm-Liouville systems. For secondorder DEs, we can always put the DE in the self-adjoint form

+ >..r~t»)' = 0,




< Xl,


and define the residual by


71 (85)

E,\,(x) '" L[o.jJ]

+ Ar(x)!jl,
the eigenvalues of the given Sturm-Liouville


where A approximates and


"" Cjgj(x) ""2:
, i» I



approximates the eigenfunctions. Because the boundary conditions for eigenvalue problems are homogeneous, it is not necessary to include the term go(x) in the trial solution (86), We select the trial functions gAr) by primarily the same criteria listed in the last section. Basically, the technique from this point is essentially the same as before, except that now the system of equations in the unknown c's will be homogeneous. Hence. in order to obtain nontrivial solutions for the c's we must require the coefficient determinant of the c's to vanish (see Cramer's rule in Appendix C). Doing so leads to what is called the secular equation, an Nth degree polynomial in A. The roots of this polynomial furnish approximations to the first N eigenvalues of the given Sturm-Liouville system. *
EXAMPLE 26 A pproxi male the first two eigenvalues .1'''

and eigenfunctions
y(l) == 0,



).,y == 0,

y(O} = 0,

using the trial solution o.jJ = clxtl - .r) + c,x'(! - .r].

The residual is

£:(x) = .)r"

+ Ao.jJ


[2 - A(x - xl)]cl + [2 - 6x + A(x'

- x»lc~, the system of

and by setting £:(x) = 0 at .r = 1/3 and .r equations (for the collocation method)

2/3, we obtain

(2 - ~ A)

CI -


/l.Cl '"


(2 - ~A)



(2 - 2~A)

C~ '"

For nonzero values of




we require that

'The method is not uniformly accurate for in the larger eigenvalues for any N.

an eigenvalues,

Greater error is always observed




from which we calculate A, By substituting AI we find that C2 approximated by

= 9,

A~ "" 27.

= =

0 while

9 into the abo v e system of equations involving CI and CI is arbitrary. Hence, the first eigenfunction is Ifrl = x(1
- x)

(for CI = I). Similarly, substituting leads to from which we conclude setting CI = I

A) = 27 into the same system of equations
2Cl =

~4cl -



Cl =

-2el, leaving


and hence,

= .l(l - .r) - 2xl(l - x) == .t(1 -

3x + 2x').

Using the Galerkin method. we set

which yields the system of equations

(!- _!_ I (6" - I)

30 .


60 A


(!- _!_ /\)c, + (2 - I) c, 15
6 60 105 A
10 zero,




== O.

By equating the coefficient determinant equation with solutions

this time we obtain a secular

These eigenvalues lead to the same eigenfunction approximations found above by the collocation method. For the sake of comparison, we recall that the exact eigenvalues for this problem were previously found to be






= 4""ITl "" 39.478

(Example II in Section 1.4). Here we see that the Galerkin method has once again provided be tter approximations than the collocation method. Of course, greater precision can be achieved in both approximation schemes ifadditional terms are included in the trial solution.

Using a one-term trial solution and the Galerkin method, approximate the largest load P that a column of length b can support before buckling may take



_.".:_; ..

•• C




place. The problem is characterized

by 0<x <b
y(b) = yU(b) = 0

+ Py"

'" 0,

(see Example 16 in Section 1.5.4).

yeo) = y"(O) = 0,

SOLUTION To use a one-term polynomial approximation, the trial solution satisfy (we set CI = I for convenience) "' .. ==- x(x - b)




so that $"(0)

= ~"(h) '"

O. Therefore,

following two integrations

we obtain


= _!._ "6 !!. r 12
4 -

+ C 1X + C, -,

and by setting ~(O) = 0 and tjJ(b) = O. we find that C1 We thus have produced the tria! solution ~'" 12


and Cl

= 0.

x -6"x- +J2-'".
conditions. Writing A



which satisfies all four boundary residual to be

PIEI, we find the

and by setting

we obtain

b (I 4 £1(\") ru _" 12·


b yl 6 "'

b + - )x 12


dx = 0


The largest load is then
P ""



(the Euler load is proportional to the first eigenvalue), which is very dose to the exact value
P ==11"2 -



9 870 -.2 ." b


On the other hand, the collocation method with E1(x) lead to the less accurate result P "'"8Ellb1•


0 at x


bl2 would








In Problems 1-6, use the given trial solution find an approximate solution.

= I:

and the collocation
ot< = -

method to

yO + Y = .r - I,

yeO) = - I,





clx(1 - x)


y" = - I, yeO) = 0, y(r.) = 0: ot< = c, sin x + c, sin 3x Construct a table of numerical val ues like Table 1.2 and compare solution. [Set E,(.t) = 0 at x = ."./4 and x = -:./2.)

with the exact

, ,


i'l + Y

= x,

ot< = c,(.r·m -

r/6 + x/12)
'" 1,

yeO) = yOlO)





yO(O '" 0:



d [(1 + x)-'y'] l
CI.tO -




5. h:y" + 3xy' - y
~, '"



4xlr., yeO) c,r(l -.r)



= 0:


,d [(I + ayjy') = 0, yeo) = O. y(1) = 10: <to = lOx + c,x(1 - x) ,x (a) Assume a = 0.1 and compare your answer with the exact solution at .r = O.~. 0.4. 0.6, 0.8. (b) Repeat (a) when" '" 1. Comment on the accuracy of the approximate solution 5 as the parameter (/ inc reases. Note: The solution y represents the steady- state temperature acros s a slab with a temperature dependent conductivity key) '" 1 + ay. function and the Galerkin method

In Problems 7-12, use the given trial solution find an approximate solution.

7. Problem
8. 9.


Problem 2 Problem 3

Problem 4

II. ProblemS

Problem 6 If the uniform
qo(b - .r), the problem


beam in Example 2S is subject is characterized by

to the distributed

load q(x) =

yeO) = yOlO) = 0, y(b) "" y"(b) '" O.

Using the trial solution
ot< = CI

sin .".xfb + c, sin 2TrxJb,

find an approximation of the displacement (a) by the collocation method; (b) by the Galerkin method.

EXERCiSES 1.7 In Problems 14-17. use the given trial solution and the collocation approximation to the first eigenvalue.


method to find an

I' + >.y :


y'(O) : 0, y(O) : 0,

y(l) = 0; y'( I) = 0;

'" = c,(I - x') oj!

15. 16.

y. + >.y : 0,

c,x(1 - x(2)

xly" + 2xy' + >.y = 0,


y(1) = 0, y(O) finite,

J(3) '" 0;

= c,(x - 1)(3 -








ojs = c,(1 - x')

in Problems 18-21, use the given trial solution and the Galerkin method to find an approximation to the first eigenvalue. 18. 19. 20. 21. 22. Problem 14 Problem 15 Problem 16 Problem 17 By Galerkin's method, approximate

the first two eigenvalues



>.xy = 0.
<I< '" <I< '"

= O.


(bl 23.

using trial solution using trial solution

c,x( I - .r) + c,x'(1 - xl;

c, sin -:ox + Cz sin 2rrx. approximate the first two eigenvalues
y(l}: 0,

By the collocation


and eigen-

fu nc tions of

+ >..y '" 0,

y(O) = 0,

(a) (b) 24.

using trial solution':' = c,x(j - x) + c,.x'(I - xl; using trial solution Ij1 = c, sin lTX + c, sin 21rx. - 2t), approximate . the first

Using the trial solution ojs = c,(2t-' - 3.t) + clx' two eigen values and eigenfunctions of y~ (a) (b)


>.y = 0,

)'(0) = 0,






by the collocation method; by the Galerkin method. oscillator problem in quantum mechanics - EX'y is characterized by


The anharmonic -


y" - - ,"wl;;'y

1 2

+ >..y '"



lim y "'0, where all parameters are constant. The eigenvalues >... represent the energy levels possible for the system, the grou nd state being determined by the smallest eigenvalue 1.0• Using the trial solution'; = c, e1..p( -mw_~'/211) and Galerkin's method. approximate the ground state energy and show that for E = 0 it reduces 10 the well-known result 1.0"" wH12.

t)g.(.. leads to the system or equations y{x. - N Ab.) = 0.2. k= 1. 28.q(. . Y(X:) "" 0.(x)g. Use the results of Problem 26 to solve Proble m 22.76 CHAPTER I BOUNDARY VALUE AND EIGENVALUE PROBLEMS *26.(.r)d.. = ~ i" " (p(x)g.(x}g.r_ 27. Use the results of Problem 16 to solve Example 26 with N = 1.r)y "" 0.:~ .d] dx f' " r(r)g . Show that Galerkin's method for the special Sturm-Liouville problem L[y] + Ar(..~t)gax) . j-I 2: (a).N where aj• = and b.')cj = 0.. " . .

to the notion of a one-sided Green's function.4 to boundary vuluc problems. . which always exists when the coefficients of the DE.H1d the conditions under which the Green's function fails to exist arc examined. : " " .CHAPTER 2 THE METHOD OF GREEN'S FUNCTIONS Historically the use ofa Green's function to solve DEs grew out of a study of a special partial differential equation and boundary condition called the Dirichlet problem (see Chapter S)_ Later it was discovered that a similar function could also be used in the analysis of ordinary differential equations featuring nonhornogencit ics. It is th is failure to always exist that most clearly distinguishe s the Green's function associated with boundary value problems frOI11the Green's function associated with initial value problems.. today we collectivcl y refer to them all as Green'sfunctions.In general. in normal form. White this is perfectly acceptable from a purely mathematical point of view. a non homogenei ry arise s in the mathematical form u lat ion of the problem.: .'.. are conti nUOLJs. and a function )'". Anticipating the physical inl e rpreta lion of t he one-sided Green's fun ct ion.. which physically represents the response of a system to initial conditions alone. When the DEi s nonh ornogeneou s... In Section 2.3.. ei the r in the DEar in the a uxi Iiary cond itions .. Both the physical interpretation of the Green's function for boundary value problems .. we a re led . however. or both. a particular solution of the equation can be found by applying either the method of undetermined coefficients or the variation of parameters technique . ' 2. such techniques lead to a particular solution that has no special physical significance.1'11.1 INTRODUCTION When a physical system is subject to some external disturbance. A more general definition of a Green's function is presented here and the basic propert irs associated with this more general function arc also examined. In constructing _I"" by the method of variat ion of pararnet ers .2 we show how the solution of a second-order initial value problem is always the superposition of a function . we di scu 5S the concept of an unp IIlse function in Section 2.. it does not provide the kind of infer77 '. . which can be interpreted as the response of a system at rest that is later su bj ect ed to an ex te rnal forcing [unclion. Although some of these functions were originally given different names for different problems. The method of Green's functions is extended in Section 2.

5. (2) The solution of (2) is readily found to bet \'(1) = e-«'m (1'0 + ~ f e~. 0. Ill." whose name occurs so abundantly throughout classical analysis.1. the velocity at all later times is then a solution of the initia! value problem III d .I'"F(.. and all other solutions due to different source terms are found to be superpositions of the Green's function. Andrews. also called the fundamental solution in t his cont ext.! 78 CHArTEI{ 2 THE METHOD OF GREEN'S FUNCTIONS rnation about. or insight into.: Glenview. It is a powerful tool based upon the construction of a specific function known as the Green's function... Consider the motion of a particle of mass 11/ in resistive medium under the influence of an external force F(t).. Foresman and Ce. It is the purpose of this chapter to introduce a systematic procedure for solving nonhomogeneous DEs that not only has a special physical" significance. For instance. This function measures the cesponse of a system due to a point source somewhere on the fundamental domain. Not only that. Ifthe resistive force is proportional to the velocity \. when F(I) . then the equation of motion is m- .) . (3) The advantage of this solution formula is that it provides a solution of the problem for any initial velocity \'0 and external force F(I). but lends itself'as well to the development of general solution formulas. The method or attack is attributed to George Green..) d.. I> 0. the problem that is possible by constructing the particular solution in a more specific manner. 2. see Section 2.) t982. but it is composed of parts that each have important physical interpretations. the solution reduces to • G EQRGE GR EEN (1793-184 t) gnined recognition for his important works concerning the reflection and refrnClion of sound and light waves. On/inary Dijft'Mfial A. let us solve a simple introductory example. wilh .. £qUa/ions tFo( example.1 A PHYSICAL EXAMPLE To help motivate our discussion of the Green's function.-.' dl + C\· = F(t). Assuming the velocity of the particle at time r = 0 is \'0. C.2 in L. He also extended Ihe work of Poisson in lite theory of electricity and magnetism. of the particle.pplicQtiollS (ScOII.I dv dt = F(/) - ('1' ' (I) (via Newton's second law of motion) where c is a positive constant whose value is determined by the nature of the resistive medium..

it is helpful to put the DE (5) • Since initial value problems generally involve time.<tIm . rather than by . "and then impose the The conventional aux iliary conditions auxiliary conditions to find numerical values for the arbitrary constants.-. in subsequent sections. and yet such information is often vital in understanding the fundamental characteristics of the system.e..!_ e .. V J (4) 1'11(1) 1'1'(1) . Because of the physical interpretations derived from (4). we will attempt to develop a solution formula for higher order DEs that is comparable to (4).c/lm m [ 0 e'?" F( T) dr. Initial value problems involving second-order DEs have the general form" I> 10 )'(/0) = ko• For purposes of developing in normal fom: the following theory. to the notion of a Green's function. 2.(1) describes the motion of'a particle that is initially at rest and then subject at time I = 0 to the external force F(I).2 INITIAL VALUE PROBLEMS way we have been taught to solve DEs with prescribed is to first find a general solution.'. we see that 1'11(1) represents the "free velocity" of the particle in the absence of an external force. by rewriting (3) as the sum \'(1) "" I'oe' . ". i. \'0 "" VALUE PROilLEMS 79 0.INITIAL whereas when the initial velocity is zero. and general ize t he res ults \0 cquat ions of order 11 in the exe rcise s. it is desirable to obtain similar solution formulas for initial value problems involving DEs of higher order. and 1'1. We will initially address only second-order DEs. Our pursuit of such formulas is what leads us.-. since these are the most common DEs in practice. + . and the ease of using this solution formula for different values of 1'0 and different forcing functions F(/).. \. we will designate variable by I..____. (3) becomes Thus. In order to obtain a solution whose form readily provides important information about the physical system. The main disadvantage in this approach is that it usually gives very little insight as to how the system being studied responds to each input parameter independently.___. tile independent .

The homogeneous DE. (12) initial conditions we get the simultaneous C. which we will denote by YII. = ko. N[y) = 0.. on }'H. Y2)(t 0) (14) .kO)'I(lo) . C. = 0. )"(/0) = 0. Hence. denoted by Yr. Problem 26 in Exercises 2. (Io(r) = Ao(I)IA 1(1). described by (8) (9) Closely associated N[y] and N(y] =[(1).. i.v'(to) '" 0. 11. y(ro) (7) where D = dldt . with (8) are the problems y'(to) = k. represents the response of the same system.80 CHAPTER 2 THE METHOD y" or GREEN'S FUNCTIONS + OI(I)y' + (lo(t)y = [tt). Hence.k. their Wronskian I . y{ro) = ko. (6) obtained from (5) by division by A!(I). and 1(1) = F(/}IA2(t). The solution 0[(9). physically represents the response of the system described by (8) entirely due 10 the initial conditions in the absence of external disturbances. W(YI.yi(I}Jl(l) solutions. .(1) '" A 1(1)IA2(t).2 shows that the sum of solutions of (9) and (10) is the solution of (8). so that (5) can be expressed N[y] = [(f). C2)'2(tO) = ko. In this regard the function jll) is often referred to as a/arcing function. the sol ution of (I 0). in the more compact form I> to. which is at rest until lime I '" 10.Y2(IC) IW(YI. whereas in (10) it is the DE that is nonhomogeneous and the initial conditions homogeneous. We also find it convenient to introduce the /lOm1(11 differential operator N = D2 + a. Y2)(rC) .(t)D + Qa(t). )'1)(1} = )'1(t)Yl(t) Imposing equations the prescribed . in (9) has the genera! solution (J I) where y.)'I(/a) . On the other ha no. and }"1 are linearly independent is nonzero. 10. W(}"I.)'I(tO) the solution + C2yi(to) = kit of which leads to the unique determination (I]) C .koyi(to) .e. C 2 = k. at which lime it is subject to the external input I(tl. ( 10) In (9) we note that the DE is homogeneous while the initial conditions are nonhomogeneous. #< 0.yl(IO) +...

' + ao(t)yp ao(t)yd = u(I)[y'i + a.". zero v'---_'.(/)Y. we get C.\ solution of (15) of the form (16) where YI(I) and }"!(t) are (he linearly independent solutions of the associated homogeneous DE. + ao(t)yzJ "\. the substitution N[ ypJ t. Y'(lo) = 0: has only the trivial solution == O. 2. = 0. "" C.1 If N = D' + a.(/)y.{I)D + al)(t).(/) + v' (t)yj(/). then the homogeneous initial value problem N[y] =0.-. == 0.2. To solve (IS) we usc the method ofrarilltlim a/parameters. -' = yj..(I)y'. we concentrate problem (10). (18) ~{-: ~ :.IN!T!AL VALUE PROBLEMS 8I )'11 "" Notice that when ko = k. (18).-. Another differentiation y'. This begins with the assumption that there is . (20) . The physical implication of'this result is that a system which is initially at rest and not subject to any external disturbance must remain at rest.1 + /I'(I)y'. '. )'(/0) = 0. + zero + l'(/){yi + a. so that necessarily O. We now impose the assumption 1/ + )"(I)Y2(1) = 0. we obtain }' p = /I(t)y '(/)YI(I) :(1) + \"(I)y. and therefore conclude that )-'p = 11(1))-'.(1) + + v(/)y. THEOREM 2. (17) . (19) Next. which we renumber as N[y] =ftt). where ao(l) and a.(I) are continuous functions for 1 ~ 10.(1) + /I' (t)YJ(I) + v'(1))'2(1).(t). and (19) into the DE in (15) leads 10 + a.1 ON E~SIDED GREEN'S FUNCTION 011 Now that we have solved (9). = u(t}y')(/) of (18) yield s v(t)yi(l) + u'(I)yi(l) + v'(t)yi(I). By differentiating (16). of(16). I > j 10. v' ((o) "" solving the remaining (15) O. and 1/(1) and l'(I) are two functions to be determined.

if y. we wish to take definite integrals of (22) in finding the functions 11ft) and \'(1). Y2)(t) is the non vanishing Wronskian [recall Equation (12)].~ . in order to satisfy the prescribed homogeneous initial conditions in (15). then N(yp] '" fCl) .. when t'" to we get the immediate result Yr(to) = " f g 1(10.t = .)(. upon cornbi ni ng int egral s. )"1(7)/(. f) d. • y:(. by defining the function . which we will verify below.)/(. al we find yj.F(x. and we deduce from (20) that 1/'(I)y.(/o} I' l f' + F(f.) ds . (21) Using Cramer's 1/' rule. Finally.• _. af (28) . f).)'I(I)Y:(T) /(-) d .) dr. \1' • . is indeed a solution of the DE in (15). (23) The su bst itut ion of Ihese re suits into (16) yiel ds. -- - - &.) W()' (25) we can express our particular )'1' = solution in the form (26) I' " gl(r.)':)(-T) d r. . solution of (17) and (21) gives us (22) (r) = _ where W(YI. ag (10.) (24) However.. I. ). The proper choice.)!(. First.(x. )'2)(t) . using the Leibniz formula (from calculus) d .) == (' J.:)(. 82 CHAPTER 2 THE METHOD OF GREEN'S FUNCTIONS The bracketed expressions above are both zero because of the assumption that YI and Yl are both solutions of the associated homogeneous DE. (27) Next.. What remains now is to verify that this choice of)'p actually satisfies the homogeneous initial conditions in (IS). turns out to be lI(t) '" - I' . Thus.) W(YI. (0)[(10) = o.:(1) h ). '. I) dx aF d t I. W(_I'I . T)[(T) d« + C1(fo.) Wry\> )". the simultaneous y~(I)f(t) W(YI. r. .\'1(') 1 )"1(/) )"2(') I ).(1) + V'(l)y2(t) '" f(l). ve. = f" I.:)(. .I' I" ybly:{t) .)[(T) dt = O.

sin t) '" l. T) is called the one-sided Green's function. Using (26). we have . CI = I and C1 = -1.I ee 2" (Sin . ~ s» "" = L I KI(f. yeO) "" I.. EXAMPLE I .: .C1 . Hence. T) sin T dT sin I[ cos T sin T dT . SOLUTIO"'. T that W(cos I.. =- 0 + C2 1.1"(0) "" O. t~) = 0 by definition. )". it is important that the nonhomogeneous DE (15) is in normal form. Its construction depends only upon knowledge of the homogeneous solutions YI(/) and Y2(t). Hence. is unique.~os If sin? T dT I. so that )'11 = cos t . . problem we observe sin . the lnitial conditions are satisfied.1If"'"' -Y -< ". For the second consequently CI(f. And since these solutions always exist for a normal operator N with continuous coefficients. . and moreover. We first rewrite the problem as two problems: y" + y "" 0. y'(O) "" -I.". 111 the first case the general sol ution )'If = CI cos I + C! sin I. t . we find = and imposing the initial conditions. REMARK: For proper identification of the function jill (26). it follows that the one-sided Green's function always exists for such operators. Use the method of Green's function to solve _v"+y=sinl. ~ INITIAL VALUE PROBLEMS 83 where we recognize that CI(tO..cos t Sin T.(0) '" CI + C.sin I. . The function CI(t. l. yeo) = I. and T) = ! cos T cos t sin f I = cos T Sin . . sin r. y'(O) "" -I and y"+y= yeo) is = 0.. I> 0. cos f). 0 yif(O) = . :"-/ 7~ '.

.84 CHAPTER 2 THE ~IETHOD OF GREEN'S YII FUNCTIONS and by combining and )./l)e + 2/(1 By writing y = )'11 + s». We recognize the DE as a Cauchy.IJ. ). y(1) = 0. I~ 'I'J .) From the normal form of the DE we recognize /(1) solution of the second problem above is )'p = = 2Ile'.'(1) = 2. Imposing the iniual conditions in the first problem YII = " . y(l) = 0.. The two problems (ly" _ solve are y'(l) '" 2. t I. + 2: (Sin cos I ) [) "" (I .I. y( J) = 0 ..Jle t • • For solution techniques of Cauchy-Euler equations.1" we deduce 1Sin that I/ )' = cos . * whose homogeneous solution is )'11 = CII + C:I'.) = t1 ft e" ds - t r dt 'i'le' dr . and thus the fKI(t. we find thnt Cl = -C1 = I. ./ cos 22' EXAMPLE 2 I .1 le'. we get the intended solution )' = (tl - 1)(1 . and hence The Wronskian of )'1 = one-sided Green's function is gl(l.:' - .J 1= 1'1)-11'). r') = :!. Use the method of Green's I~Y" SOLlJflON 3/)" function to solve t> I.e) + 21(1 . Therefore the r) = 2'i'1. 2. 'I')(2'1'2e. see Problem 34 in E:ltcrcises 1. ). I. 31y' + 3)' "" 0. Y . I 3. we need 10 + 3)' = 21'e'. above. c .Euler equation.3.. I and }': 1 = 13 is Wet. w here we have di v iJed the 0 E in Ihe second preble m by /: to put iii n normal form.Sin .-I' \' II + -3 \' "" ')t:.'(1} = 0. = (. .

*9.o)"..)e'"'-·' (I (n e.TJ b ~[(.1. J. 3.2 85 TABLE 2.b' t'D' + ID .""-" sinh hit . 11=2.b' . T) I -T "" I.. D'+4D+4 4U-8D+S t'D' 6.r~'T)J (I . It = 2.. Table of One-Sided Green's OPERATOR Functions g. + b' ~ e""-" sin b J b(1 .b' (D . EXERCISES 2. all initial value problems with the same operator N have the same one-sided Green's function..Y-Gn Construction of the one-sided Green's function using (25) is mostly mechanical. ' (I .b). A few such entries are provided in Table 2.(I. D' . ar + (I~ a-b re~'-"t" T)'-' I)! ei1.4 sinh b(1 . 5. In solving initial value problems by this technique we find that the same DE appears many times.a)(D . D' ..(I 1 D{(I .T) 'j ': 4.J I 6. 9. *10. D' D'.5 3. +j + ID 16 4.I')DJ + 21')D . 10.T)'-' (n . 7. Dl D' 2. determine the one-sided Green's function for the given operator (D = didO.2 In Problems 1-10.3. .T) D' . Since the one-sided Green's function is not dependent upon the forcing function or initial conditions.I)! D' + b' D' .. S.a{I-tl (D .3. i -.2aD 8.EXERCISES 2... For this reason it is convenient to tabulate some of the most common differential operators and their corresponding one-sided Green's function. (D a#-b ~ sin b(1 . 7. tu . but with different forcing functions and initial conditions.1 .2(ID + a' .ID + 10' .4. 2. Dl_D-2 8..

26. 23.86 CHAPTER"1.6)"' + 9)" = I'l"'. + y~ is a solution of y'{tol = k" = J(I). + ty' + 4.1' = sirulog I). yeO) = 2.lT. 19. the function y(l) = g. and N[yl =j(I).1 28. y'(O) = I 12. )"(1) = 0.'+6)"=I+e-'. 14. 15. y(/. show that (a) g. I'y· 41'}"" yO) = I.(I)y = /(1). is .3y' . y'Oo) = 0.(t. y"+4).) = k•• show that it has the solution . 22. y" -)' '" I. y(l) = 0. log t'.2: cos 21. y"+y=l"-'.4y = e:'. N[yj=O. I'Y" . for a fixed value of T. respectively. yeO) = I. = 3Ie-'.) '" 0. I If y" and Yr satisfy. y(O) = 1. 27.. yeO) == 2. with respect to t. y(O) = 0. use the one-sid ed Green's problem. 24. y'O) = 0 y'(2) y" . THE METHOD OF GREEN'S FUNCTIONS I II Problem s I 1-25. Hint: Use the result of Problem 27. y'(I) y'(I) '" 0 = -116 y(l) '" 116. IJ. along with its first and second derivatives continuous. T) = o. 18. 2)"" + y' . Show that.5ty' + 8y = 21'. 29. + 8ty' + I'. I'y· + 71y' 5y = /. y'{T) = 1. fu nction to sol ve t he given initial value yeO) = 0. v'(I) = 0 )"'(1) = 25. y" + )" + 2y = sin 21 . Given the first-order initial value problem. 21. II. IY . 17. + 2)" +y + y(O) = 4. y'(O) = 3 )"'(0) = 0 )"(0)=-4 y'(O) '" 0 16. T). = ° Y'(TI/2) '" 1 + y' _ 2y = -4. For all T ~ t.Y = t + I.6}' = log r . T) is a solution of the initial value problem N[y] = 0. J(l) '" 0. 20_ y" . y" + Y == 2 esc 1 cot I. y'(O) = 2 y'(O) '" 6 r'(I) = 0 y. Y '" 3r yO) = I. Y("II"/2) = I. (b) g. y' + a. verify that the sum}" = N[y] )'H the initial value problems y(t.lt. yO )"(2) "" 3.

y'{O) = Vo.. find the response of the spring-mass system given that the system is initially at rest (i... is a solution of the associated )'.-·n".2 87 homogeneous DE and g. (b) c~rilically damped (cl = 4mk): g . 2.EXERCISES where )'. y. sinh[a(l a .(I. and F(t) is an external (driving) force. where a .T)l.e..e')"'l2m.4mk)'~nm.(I..-. where m is the mass. )'(0) "" 3. [iV. tIT. Show that the one-sided Green's function for the following cases of damping is given by (a) underdamped (C' < 4mk): g. T) = {I . ~ .= (4mk . (e) OI·t!rdamped (e' > 4mk): g. Using the one-sided Green's function given in Problem 31. (c) F(O = PeDs "-'ut. where c is a positive constant.!.(r)fy . the motions are called damped. "-' ~ 1010. '" "0 = 0) and then subject to the driving force (a) F(t) = P (constant) (b) F(O = P cos wI. • erf(t) is Ihe BTOI' /WlClimt defined by erf(1)" *[. where II. T) = 1e-oj. When resistive forces are taken into account for the spring-mass system in Problem 31.. In such cases the DE is modified to read my· + C)" + ky "" F{O.T)]. k is the spring constant. 31. T) = .(t. ~ . yeO) = Yo. erf(t) + 3] e' is the solution of the initial value problem y' . .(T). .(e' . T) '" *30.. Use the result of Probl em 29 to show that" y= :. e -<1'-'W'" II- sin[f1-(t.. 33.T)e -<1' -.~.2ty = I. Show that the one-sided Green's function associated with this system is 32. The small motions y{r) of an undamped spring-mass system are governed by the initial value problem my· + {y = F(t). (I.

U) function is defined by y. C.). use the one-sided Problem 35).I} *39. y"o. *35_ Consider the third-order y~ initial value problem ait)y· + + a.I 5 3 D' + .6Dl + liD ..88 CHAPTER 34.(T) y. (b) y. Establish the solution formula y= g. is called the steady-stale co SO/IIIiOIl. 0.. F(t) = (b) (d) F(t) '" P cos wI.. y~ .(.. 36..(r) g. *4]. £IT + C.) (c) and derive determinant expressions for C. that part of the particular solution y" that does not vanish in the limit as I .) y. 38.) = k.y.. 36-41.. yeO) = 0. D' (II = 2. and C. use the resul t of Problem 35 to construct function for the given operator. r) = W(}'I. y_..(1) + C.(T) y.) = ko. y. Let us imagine that the sudden excitation. y"(l) "" I y"«()) = -I In Problems 42 and 43. *43_ y" Green's y'(l) +y '" (t .Dl -2 2 D' .(J. }" are Iinearly inde pe nde nt sol ution s of the associated homo- geneous DE.f.. 0</<1 t > I. 40.){. Generalize the results in (a) and (b) to nth-order problems. 0<1<1t I:>. y'{O) = I.. the one-sided Green's r'. such as a sharp blow or voltage surge.) }'. D(DI + 4} D' . 2 THE METHOD OF GREEN'S FUNCTIONS When the motions ofa spring-mass system are damped. sin I. !n Problems 3. (a) Show that the one-sided Green's y. {I.{I.4y = I.y~ + 4y' .) where)". i . F (I) = { 0.(rl. similar to those in (3) and (14). *42.3 IMPULSE FUNCTIONS In certai n applications it is con venient to introduce (he concept of an impulse. which is the result of a sudden excitation administered to a system. 2.(tl YIlT) yj(.(I) + C. y{l) "" 0. D'(D' .)f(. .y. Use the one-sided Green's function given in Problem 33(a) 10 find the steady-state solution of the spring-mass system when the driving force is (a) (c) F(t) = P (constant). .4.l)l"-'.(T) y.6 function to find the solution (see = 0..(I)y' + y'(lo) = k" /10 (1»' '= /(1). y(t.. ) 37.

. otherwise.(t) dt = I. 0.. The defining properties of this function are therefore 8(1 .IMPULSE FUNCTIONS :-.1 Impulse function. Hence. 8(t . FIGURE 2. In order to provide a mathematical model of the function d.a) - r.• lim I . Furthermore. (29) The value of I is a measure of the strength of the sudden excitat ion.1).a).. Now leI us idealize the function d.e . . we still want I"" J... we write dAr) = {21E' 0.(t).E os: r os: a + E (see Figure 2.. which has the property of imparting a unit impulse to the system at time I "" a but be ing zero for all other values of t. but is otherwise zero. i. (31) We can use the result of this Iimit process to define an •'idealized" uni: impulse function." < ( < a + E.-0_..= lim [ 0 .... I ""'"a (32) 6(1 . 89 which we will denote by d.a) dl . h:IS a nonzero value over the short interval of time a .(t) by requiring it to act over shorter and shorter intervals of time by allowing E _ O.(t).. The total impl. it is convenient to think of it as having a constant value over the interval a .'se (force times duration) imparted to the system is thus defined by (E> 0). J.. (30) I 2.." I. Although the interval about t = a is shrinking to zero. we wish 10 choose this constant value in such a way that the total impulse given by (29) is unity.

Sign up to vote on this title
UsefulNot useful