SlXTH EDitiON
Ci I'!! I
.
.
1Y'
ELEMENTARY DIFFERENTIAL EQ.Ui\TI(): NS stXTHmmoN
I, 1 .,. I
~
'\_<
'
lo
...
EDWARDS & PENNEY
EDWARDS
&
PENNEY
ABOUT THE COVER. This image illustrates the trajectory of a moving point whose space coordinates satisfy (as functions of time) the Rossler system of differential equations that is discussed on page 553, and which originated in studies of oscillations in chemical reactions. In its motion along its trajectory the point may appear to spiral repeatedly around a set  the socalled Rossler bond  that somewhat resembles a (twisted) Mobius strip in space. To portray the progress of the moving point, we can regard its trajectory as a necklace string on which beads are placed to mark its successive positions at fixed increments of time (so the point is moving fa.stest where the spacing between beads is greatest). In order to aid the eye in following the moving point's progress, the color of the beads changes continuously with the passage of lime and motion along the trajectory. As the point travels around and around the band, it may be observed to drift radially back and forth across the band in an apparently unpredictable fashion . Two points that start from nearby initial positions may loop around and around the band somewhat in synchrony, while moving radially in quite different ways, so that their trajectories diverge appreciably with the passage of time. This illustrates the phenomenon of chaos, in which tiny differences in initial conditions can result in great differences in the resulting situations some time later. Further discussion of chaos associated with differential equations can be found in Section 7.6. Throughout this textbook computergenerated graphics are used to portray numerical and symbolic solutions of differential equations vividly and to provide additional insight.
r
m m m
C HENRY
~
EDWARDS
&
DAVID E.
z ~ :::0 <
:::0
"'" m
m
"'"
)> z
c
PENNEY
I
rm /:)
c: ~
0
V"
Upper Saddle River, NJ 07458 www.prenhall.com
.r i)Tioof ·.t .L... ~~~~~
ISBN~13
87S.0.13·23t730·$
•
...
SIXTH

z
ELEMENTARY DIF·FERENT'IAL EQUATIONS
ELEMENTARY Dl FFER ENTIAL EQUATIONS
Sixth Edition
C. Henry Edwards David E. Penney
The University of Georgia with the assistance of
David Calvis
BaldwinWallace College
PEARSO N
Pr<'nt icc
llall
Upper Saddle River, NJ 07 458
Library of Congress CataloginginPublication Data on file.
Editorial Director, Computer Science, Engineering, and Advanced Mathematics: Marcia J, Horton Senior Editor: Holly Stark Editorial Assistant: Jennifer Lonschein Senior Managing Editor: Scott Disanno Production Editor: Irwin Zucker Art Director and Cover Designer: Kenny Beck Art Editor: Thomas Benfatti Manufacturing Manager: Alan Fischer Manufacturing Buyer: Lisa McDowell Senior Marketing Manager: Tim Galligan
© 2008, 2004, 2000, 1996 by Pearson Education, Inc.
Pearson Education, Inc. Upper Saddle River, New Jersey 07458
All rights reserved. No part of this book may be reproduced, in any form or by any means, without permission in writing from the publisher.
TRADEMARK INFORMATION MATLAB is a registered trademark of The MathWorks, Inc. For MATLAB product information, please contact: The MathWorks, Inc. 3 Apple Hill Drive Natick, MA, 017602098 USA Tel: 5086477000 Fax: 5086477101 Email: info@mathworks.com Web: www.mathworks.com/ Maple is a registered trademark of Waterloo Maple, Inc. Mathematica is a registered trademark of Wolfram Research, Inc.
Printed in the United States of America 10 9 8 7 6 5 4 3 2 1
ISBN 0132397307
Pearson Education LTD., London Pearson Education Australia Pty, Limited, Sydney Pearson Education Singapore, Pte. Ltd Pearson Education North Asia Ltd., Hong Kong Pearson Education Canada, Ltd., Toronto Pearson Education de Mexico, S.A. de C.V. Pearson Education Japan, Tokyo Pearson Education Malaysia, Pte. Ltd.
Preface vu
CHAPTER
FirstOrder Differential Equations
1.1 1.2 1.3
1.4
1
10
1
CHAPTER
Differential Equations and Mathematical Models Integrals as General and Particular Solutions Slope Fields and Solution Curves 19 Separable Equations and Applications Linear FirstOrder Equations 46 32 59
1.5
1.6
1.7
1.8
Substitution Methods and Exact Equations Population Models 74 AccelerationVelocity Models 85
Linear Equations of Higher Order
2.1
2.2
100
100 124 148
2
c
Introduction: SecondOrder Linear Equations General Solutions of Linear Equations 113
i
2.3
2.4
Homogeneous Equations with Constant Coefficients
2.5 2.6
Mechanical Vibrations 135 Nonhomogeneous Equations and Undetermined Coefficients Forced Oscillations and Resonance Electrical Circuits 173 Endpoint Problems and Eigenvalues 162 180
2.7
2.8
CHAPTER
Power Series Methods
3.1 3.2 3.3
3.4
194
Introduction and Review of Power Series 194 Series Solutions Near Ordinary Points 207 Regular Singular Points 218 Method of Frobenius: The Exceptional Cases Bessel's Equation 248 Applications of Bessel Functions 257 233
3.5 3.6
v
vi
Contents
CHAPTER
Laplace Transform Methods
4.1 4.2 4.3 4.4 4.5 4.6
266
266
4
CHAPTER
Laplace Transforms and Inverse Transforms
Transformation of Initial Value Problems 277 Translation and Partial Fractions 289 Derivatives, Integrals, and Products of Transforms 297 Periodic and Piecewise Continuous Input Functions 304 Impulses and Delta Functions
316
Linear Systems of Differential Equations
5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 FirstOrder Systems and Applications The Method of Elimination
338 326
326
5
CHAPTER
Matrices and Linear Systems 347 The Eigenvalue Method for Homogeneous Systems 366 SecondOrder Systems and Mechanical Applications 381 Multiple Eigenvalue Solutions
393
Matrix Exponentials and Linear Systems 407 Nonhomogeneous Linear Systems 420
Numerical Methods
6.1 6.2 6.3 6.4
430
430
6
CHAPTER
Numerical Approximation: Euler's Method A Closer Look at the Euler Method 442 The RungeKutta Method 453 Numerical Methods for Systems
464
Nonlinear Systems and Phenomena
7.1
7.2
480
7
7.3 7.4 7.5 7.6
Equilibrium Solutions and Stability 480 Stability and the Phase Plane 488 Linear and Almost Linear Systems 500 Ecological Models: Predators and Competitors Nonlinear Mechanical Systems Chaos in Dynamical Systems
526 542 513
References for Further Study 555 Appendix: Existenc e and Uniqueness of Solutions Answers to Selected Problems 573 Index 11
559
PR'EFACE
he evolution of the present text in successive editions is based on experience teaching the introductory differential equations course with an emphasis on conceptual ideas and the use of applications and projects to involve students in active problemsolving experiences. At various points our approach reflects the widespread use of technical computing environments like Maple, Mathematica, and MATLAB for the graphical, numerical, or symbolic solution of differential equations. Nevertheless, we continue to believe that the traditional elementary analytical methods of solution are important for students to learn and use. One reason is that effective and reliable use of computer methods often requires preliminary analysis using standard symbolic techniques; the construction of a realistic computational model often is based on the study of a simpler analytical model.
T
While the successful features of preceding editions have been retained, the exposition has been significantly enhanced in every chapter and in most individual sections of the text. Both new graphics and new text have been inserted where needed for improved student understanding of key concepts. However, the solid classtested chapter and section structure of the book is unchanged, so class notes and syllabi will not require revision for use of this new edition. The following examples of this revision illustrate the way the local structure of the text has been augmented and polished for this edition.
Chapter 1: New Figures 1.3.9 and 1.3.10 showing direction fields that illustrate failure of existence and uniqueness of solutions (page 24); new Problems 34 and 35 showing that small changes in initial conditions can make big differences in results, but big changes in initial conditions may sometimes make only small differences in results (page 30); new Remarks 1 and 2 clarifying the concept of implicit solutions (page 35); new Remark clarifying the meaning of homogeneity for firstorder equations (page 61); additional details inserted in the derivation of the rocket propulsion equation (page 95), and new Problem 5 inserted to investigate the liftoff pause of a rocket on the launch pad sometimes observed before blastoff (page 97). Chapter 2: New explanation of signs and directions of internal forces in massspring systems (page 101); new introduction of differential operators and clarification of the algebra of polynomial operators (page 127); new introduction and illustration of polar exponential forms of complex numbers (page 132); fuller explanation of method of undetermined coefficients in Examples 1 and 3 (page 149150); new Remarks 1 and 2 introducing "shooting" terminology, and new Figures 2.8.1 and 2.8.2 illustrating why some endpoint
vii
viii
Preface
value problems have infinitely many solutions, while others have no solutions at all (page 181 ); new Figures 2.8.4 and 2.8.5 illustrating different types of eigenfunctions (pages 183184).
Chapter 3: New Problem 35 on determination of radii of convergence of power series solutions of differential equations (page 218); new Example 3 just before the subsection on logarithmic cases in the method of Frobenius, to illustrate first the reductionoforder formula with a simple nonseries problem (page 239). Chapter 4: New discussion clarifying functions of exponential order and existence of Laplace transforms (page 273); new Remark discussing the mechanics of partialfraction decomposition (page 279); new muchexpanded discussion of the proof of the Laplacetransform existence theorem and its extension to include the jump discontinuities that play an important role in many practical applications (page 286287). Chapter 5: New Problems 2023 for student exploration of threerailwaycars systems with different initial velocity conditions (page 392); new Remark illustrating the relation between matrix exponential methods and the generalized eigenvalue methods discussed previously (page 416); new exposition inserted at end of section to explain the connection between matrix variation of parameters here and (scalar) variation of parameters for secondorder equations discussed previously in Chapter 3 (page 427). Chapter 6: New discussion with new Figures 6.3.11 and 6.3.12 clarifying the difference between rotating and nonrotating coordinate systems in moonearth orbit problems (page 473). Chapter 7: New remarks on phase plane portraits, autonomous systems, and critical points (page 488490); new introduction of linearized systems (page 502); new 3dimensional Figures 6.5.18 and 6.5.20 illustrating Lorenz and Rossler trajectories (page 552553).
Throughout the text, almost 550 computergenerated figures show students vivid pictures of direction fields, solution curves, and phase plane portraits that bring symbolic solutions of differential equations to life. About 15 application modules follow key sections throughout the text. Their purpose is to add concrete applied emphasis and to engage students is more extensive investigations than afforded by typical exercises and problems. A solid numerical emphasis provided where appropriate (as in Chapter 6 on Numerical Methods) by the inclusion of generic numerical algorithms and a limited number of illustrative graphing calculator, BASIC, and MATLAB routines.
Organization and Content
The traditional organization of this text still accommodates fresh new material and combinations of topics. For instance: • The final two sections of Chapter 1 (on populations and elementary mechanics) offer an early introduction to mathematical modeling with significant applications. • The final section of Chapter 2 offers unusually early exposure to endpoint
Preface
ix
• •
•
•
•
problems and eigenvalues, with interesting applications to whirling strings and buckled beams. Chapter 3 combines a complete and solid treatment of infinite series methods with interesting applications of Bessel functions in its final section. Chapter 4 combines a complete and solid treatment of Laplace transform methods with brief coverage of delta functions and their applications in its final section. Chapter 5 provides an unusually flexible treatment of linear systems. Sections 5.1 and 5.2 offer an early, intuitive introduction to firstorder systems and models. The chapter continues with a selfcontained treatment of the necessary linear algebra, and then presents the eigenvalue approach to linear systems. It includes an unusual number of applications (ranging from brine tanks to railway cars) of all the various cases of the eigenvalue method. The coverage of exponential matrices in Section 5.7 is expanded from earlier editions. Chapter 6 on numerical methods begins in Section 6.1 with the elementary Euler method for single equations and ends in Section 6.4 with the RungeKutta method for systems and applications to orbits of comets and satellites. Chapter 7 on nonlinear systems and phenomena ranges from phase plane analysis to ecological and mechanical systems to an innovative concluding section on chaos and bifurcation in dynamical systems. Section 7.6 presents an elementary introduction to such contemporary topics as perioddoubling in biological and mechanical systems, the pitchfork diagram, and the Lorenz strange attractor (all illustrated with vivid computer graphics).
This book includes adequate material for different introductory courses varying in length from a single term to two quarters. The longer version, Elementary Differential Equations with Boundary Value Problems (0136006132), contains additional chapters on Fourier series methods and partial differential equations (including separation of variables and boundary value problems).
To sample the range of applications in this text, take a look at the following questions: • What explains the commonly observed lag time between indoor and outdoor daily temperature oscillations? (Section 1.5) • What makes the difference between doomsday and extinction in alligator populations? (Section 1.7) • How do a unicycle and a twoaxle car react differently to road bumps? (Sections 2.6 and 5.5) • Why are flagpoles hollow instead of solid? (Section 3.6) • If a mass on a spring is periodically struck with a hammer, how does the behavior of the mass depend on the frequency of the hammer blows? (Section 4.6) • If a moving train hits the rear end of a train of railway cars sitting at rest, how can it happen that just a single car is "popped" off the front end of the second train? (Section 5.5)
X
Preface
• How can you predict the time of next perihelion passage of a newly observed comet? (Section 6.4) • What determines whether two species will live harmoniously together, or whether competition will result in the extinction of one of them and the survival of the other? (Section 7.4) • Why and when does nonlinearity lead to chaos in biological and mechanical systems? (Section 7.6)
The answer section has been expanded considerably to increase its value as a learning aid. It now includes the answers to most oddnumbered problems plus a good many evennumbered ones. The 605page Instructor's Solutions Manual (0136006140) accompanying this book provides workedout solutions for most of the problems in the book, and the 345page Student Solutions Manual (0136006159) contains solutions for most of the oddnumbered problems. The approximately 15 application modules in the text contain additional problem and project material designed largely to engage students in the exploration and application of computational technology. These investigations are expanded considerably in the 320pageApplications Manual (0136006795) that accompanies the text and supplements it with about 30 additional applications modules. Each section in this manual has parallel subsections "Using Maple," "Using Mathematica," and "Using MATLAB" that detail the applicable methods and techniques of each system, and will afford student users an opportunity to compare the merits and styles of different computational systems.
Acknowledg~~_!ltS ____
 Irfan UlHaq, University of WisconsinPlatteville Carl Lutzer, Rochester Institute of Technology Sigal Gittlieb, University of Massachusetts, Dartmouth
In preparing this revision we profited greatly from the advice and assistance of the following very capable and perceptive reviewers: Raymond A. Claspadle, University of Memphis Semion Gutman, University of Oklahoma Miklos Bona, University of Florida
It is a pleasure to (once again) credit Dennis Kletzing and his extraordinary TJ3Xpertise for the attractive presentation of both the text and the art in this book. Finally, but far from least, I am especially happy to acknowledge a new contributor to this effort, David Calvis, who assisted in every aspect of this revision and contributed tangibly to the improvement of every chapter in the book.
C. H. E.
h.edwards@rnindspring.corn
: E· ····· M r EL
.
'
v
;
,:
E·~···· ·· N 1 r· ·A ~ R; · · y~ .
'
"
'
·
~
.
.
·,
.
D .r· I•f: ... F :. . . E ..·.R ; · E :_ ..
. ...... . . ,.
J .
,
~.
N : ;r: ·~. A L :" . . . I!
'
. .. · .·AT ..· . t. 0 . . N· .S EQ
'
u ·
•
1
;
!
.
:
l
l
i
.
<>·

',
''".
. ··'' ;
_. JI
j'
»
,,
··~"
i .. '
it is natural that equations involving derivatives are frequently used to describe the changing universe. 3. Algebra is sufficient to solve many static problems. 1 . To findeither exactly or approximately. An equation relating an unknown function and one or more of its derivatives is called a differential equation. Because the derivative dxjdt = f'(t) of the function f is the rate at which the quantity x = f (t) is changing with respect to the independent variable t. y of the independent variable x and the first two • The study of differential equations has three principal goals: 1. To discover the differential equation that describes a specified physical situation. Example 1 T The differential equation dx . The differential equation d 2y dy dx2 + 3 dx + 7 y = 0 involves the unknown function derivatives y' and y" of y.the appropriate solution of that equation. To interpret the solution that is found. 2. but the most interesting natural phenomena involve change and are described by equations that relate changing quantities.FirstOrder Differential Equations he laws of the universe are written in the language of mathematics.=x2 +t2 dt involves both the unknown function x(t) and its first derivative x'(t) = dxjdt.
Eq.11x +41 = 0.4) it can be shown that every solution of the differential equation in (2) is of the form in Eq.=2xy dx for all x. (3) FIGURE 1. dT dt = k(T . if possible. That is.A). then dTjdt < 0. where k is a positive constant. one for each choice of the arbitrary constant C.1). • ···· Example 4 Torricelli's law implies that the time rate of change of the volume V of water in a draining tank (Fig. Example 2 If C is a constant and y(x) = Cex . (4) . 2 (1) then ~~ = c (2xex2 ) = (2x) ( cex 2 ) = 2xy.1. we are challenged to find the unknown functions y = y (x) for which an identity such as y'(x) = 2xy(x)that is.1. Eq. so that Tis increasing. in solving a differential equation. Example 3 Newton's law of cooling may be stated in this way: The time rate of change (the rate of change with respect to timet) of the temperature T(t) of a body is proportional to the difference between T and the temperature A of the surrounding medium (Fig. we will want to find all solutions of the differential equation. the differential equation  dy =2xy dx holds on some interval of real numbers. By contrast. so the temperature is a decreasing function of t and the body is cooling. but we will see numerous examples in which some quantity other than time is the independent variable.1. (3). describes the cooling of a hot rock in water. In each of these examples the independent variable is time t. If we are given the values of k and A. Observe that if T > A. 1. Thus the physical law is translated into a differential equation. 1. we should be able to find an explicit formula for T(t). • Differential Equations and Mathematical Models The following three examples illustrate the process of translating scientific laws and principles into differential equations. In particular.1. Ordinarily. (1). and thenwith the aid of this formulawe can predict the future temperature of the body. Newton's law of cooling. Thus every function y(x) of the form in Eq. But if T < A.2 Chapter 1 FirstOrder Differential Equations In algebra. (1) satisfiesand thus is a solution ofthe differential equation dy (2) .2) is proportional to the square root of the depth y of water in the tank: dV dt = ky'Y. By the method of separation of variables (Section 1. then dT jdt > 0. we typically seek the unknown numbers that satisfy an equation such as x 3 + 7x 2 . (1) defines an infinite family of different solutions of this differential equation.
Because substitution of each function of the form given in (7) into Eq. then V = Ay. the predicted number of bacteria in the population after one and a half hours (when t = 1.ik. In this case Eq.! ch~ng~~T~p~p{. Torricelli's law of draining. If the tank is a cylinder with vertical sides and crosssectional area A. (6) produces an identity. • dP dt = kP. We can use this particular solution to predict future populations of the bacteria colony.1.nst~tbirthandd.~~P(t)~ith c.~th_r _ atesi s. This is typical of differential equations.!II]tflll!llr.!!!lljl!!ll!ii"J.5) = 1000 · 2312 ~ 2828. Substitution of k = ln 2 and C = 1000 in Eq. • . Let us discuss Example 5 further. h) was 1000.ti~~ ~~. one for each choice of the "arbitrary" constant C. For instance. Suppose that P (t) = C ekt is the population of a colony of bacteria at time t.1 Differential Equations and Mathematical Models 3 where k is a constant. (6). Note first that each function of the form P(t) = Cek1 • (7) is a solution of the differential equation FIGURE 1.. so dVjdt =A· (dyjdt). because it may allow us to use additional information to select from among all these solutions a particular one that fits the situation under study.693147)P. That is. the differential equation dPjd t = k P has infinitely many different solutions of the form P (t) = C ekt. describes the draining of a water tank. dP =kP dt in (6). Eq. 2000 = P(l) = Cek. It follows that C = 1000 and that ek = 2. This additional information about P(t) yields the following equations: = Ce 0 = C. (4) takes the form dy dt = hy'y.693147. even if the value of the constant k is known.I.~!llfb. Thus. (7) yields the particular solution P(t) = 1000e(ln 2)1 = 1000(e1n2 ) 1 = 1000 · 21 (because e1n2 = 2) that satisfies the given conditions.. all such functions are solutions of Eq. We verify this assertion as follows: P'(t) = Ckekt = k (Cek1 ) = kP(t) for all real numbers t.5) is P(l.2. proportional to the size of the population. (4). (6) ll~!":l. that the population at timet = 0 (hours. and that the population doubled after 1 h. It is also fortunate.1. so k of k the differential equation in (6) is 1000 = P(O) = ln2 ~ 0. in many simple cases. where k is the constant of proportionality. With this value dP dt = (ln2)P ~ (0.... (5) where h = kjA is a constant.
Because exactly one solution passes through each such point. If.5) ~ 2828 bacteria after 1. answering the question originally posed." Figure 1. The analysis or solution of the resulting mathematical problem. The interpretation of the mathematical results in the context of the original realworld situationfor example. the realworld problem is that of determining the population at some future time. P(O) = 1000. describing the bacteria population of Example 6.4). 3.3 shows several different graphs of the form P (t) = C ek1 with k = ln 2. we see in this case that an initial condition P (0) = Po determines a unique solution agreeing with the given data. which involves the following: 1. P(O) = Po) that are known or are assumed to hold. that is. Finally. For instance. the selection of any one point Po on the Paxis amounts to a determination of P(O).4. As an example of this process. The mathematical analysis consists of solving these equations (here. 2. our prediction may be quite accurate.5 hours. Moreover. The formulation of a realworld problem in mathematical terms.1. The graphs of all the infinitely many solutions of dPjdt = kP in fact fill the entire twodimensional plane. FIGURE 1.1. together with one or more equations relating these variables (dPjdt = kP . for no choice of the constants C and k does the solution P (t) = C ek1 in Eq. In the population example. for P as a function oft). A mathematical model consists of a list of variabies (P and t) that describe the given situation. The process of mathematical modeling. 6 4 2 2 4 6 8~~~~r~~7~ The condition P(O) = 1000 in Example 6 is called an initial condition because we frequently write differential equations for which t = 0 is the "starting time. in which case we conclude that the mathematical model is quite adequate for studying this particular population.4 Chapter 1 FirstOrder Differential Equations C= 12 C= 6 C = 3 8.1.. C=12 2 Mathematical Models Our brief discussion of population growth in Examples 5 and 6 illustrates the crucial process of mathematical modeling (Fig. FIGURE 1.3. 1. For an interpretation in terms of our realworld situationthe actual bacteria populationwe substituted t = 1.r. On the other hand.1. Graphs of P(t) = Cekt withk = ln2. (7) accurately . Then our mathematical analysis there consisted of solving for the solution function P(t) = 1000eCinZ)t = 1000 · 21 as our mathematical result.. and no two intersect.5 to obtain the predicted population of P(1. the construction of a mathematical model.rr. the bacteria population is growing under ideal conditions of unlimited space and food supply.. we apply these mathematical results to attempt to answer the original realworld question. it may turn out that no solution of the selected differential equation accurately fits the actual population we're studying. for instance. think of first formulating the mathematical model consisting of the equations dPjdt = kP.
The remainder of this introductory section is devoted to simple examples and to standard terminology used in discussing differential equations and their solutions. . As indicated in Fig.x) 2 = y 2 if x ::j:: C . then dy dx 1 (C. (8) defines a oneparameter family of solutions of d yjdx = y 2 . With sufficient insight. Mathematical models are discussed throughout this book. Ways must be found to simplify the model mathematically without sacrificing essential features of the realworld situation. We must conclude that the differential equation dPjdt = kP is inadequate for modeling the world population. This made the mathematical analysis quite simple. If the model is so detailed that it fully represents the physical situation.1.1. then the mathematical analysis may be too difficult to carry out. perhaps unrealistically so. we might formulate a new mathematical model including a perhaps more complicated differential equation.4 in a counterclockwise manner. yet it must be sufficiently simple to make the mathematical analysis practical. If we can solve the new differential equation. we get new solution functions to compare with the realworld population. The construction of a model that adequately bridges this gap between realism and feasibility is therefore the most crucial and delicate step in the process.3. But in Example 6 we simply ignored any complicating factors that might affect our bacteria population. the results may be so inaccurate as to be useless. Thus 1 y (x ) = Cx defines a solution of the differential equation =y (8) dy dx 2 (9) on any interval of real numbers not containing the point x = C. 1. Indeed. we may attempt to traverse once again the diagram of Fig. one for each value of the arbitrary constant or "parameter" C.x).x that satisfies the initial condition y (0) = 1. 1) but has a vertical asymptote at x = 1.oo.1.1 Differential Equations and Mathematical Models 5 describe the actual growth of the human population of the world over the past few centuries. With the formulation of this new mathematical model.5. Thus there is an inevitable tradeoff between what is physically realistic and what is mathematically possible. 1. Eq. a successful population analysis may require refining the mathematical model still further as it is repeatedly measured against realworld experience. one that that takes into account such factors as a limited food supply and the effect of increased population on birth and death rates. A satisfactory mathematical model is subject to two contradictory requirements: It must be sufficiently detailed to represent the realworld situation with relative accuracy. Actually. this solution • is continuous on the interval ( . With C = 1 we get the particular solution 1 y(x) = 1. If the model is too simple.which in recent decades has "leveled off" as compared with the steeply climbing graphs in the upper half (P > 0) of Fig. 1. Examples and Terminology Example 7 If Cis a constant and y(x) = 1/(C.1.
(13) where F is a specific realvalued function of n + 2 variables. For the sake of brevity. and = is a fourthorder equation. Our use of the word solution has been until now somewhat informal.1 12 Inx and y"(x) = ix.6 Chapter 1 FirstOrder Differential Equations Example 8 Verify that the function y (x) = 2x 112  x 112 In x satisfies the differential equation (10) for all x > 0. This is why only a continuous function can qualify as a (differentiable) solution of a differential equation on an . . • (11) The fact that we can write a differential equation is not enough to guarantee that it has a solution. = u (x ) satisfies the Remark: Recall from elementary calculus that a differentiable function on ~~~. uCn) exist on I and ' U 11 .. For example. For a variation on this theme. ••• .3 12 • Then substitution into Eq.. The differential equation of Example 8 is of second order. it is clear that the differential equation (y') 2 +l = 1 has no (realvalued) solution. those in Examples 2 through 7 are firstorder equations. because the sum of nonnegative numbers cannot be negative. The most general form of an nthorder differential equation with independent variable x and unknown function or dependent variable y = y(x) is ' y " . we may say that u differential equation in ( 13) on I. for all x in I.. U. Solution First we compute the derivatives y'(x) = ~x. U. (10) yields if x is positive. . u". an open interval is necessarily continuous there.y. . we say that the continuous function u = u (x) is a solution of the differential equation in (13) on the interval I provided that the derivatives u'. note that the equation (12) obviously has only the (realvalued) solution y(x) 0. U (n)) = 0 F (X ..y (n)) = 0 ' F (x. To be precise. . The order of a differential equation is the order of the highest derivative that appears in it. In our previous examples any differential equation having at least one solution indeed had infinitely many. . so the differential equation is satisfied for all x > 0.312 tnx.~x. y.
1. In addition. oo). 1 Differential Equations and Mathematical Models Example 7 Continued 7 Figure 1.5 shows the two "connected" branches of the graph y = 1/(1 .. In Chapters 1 through 7 we will be concerned only with ordinary differential equations and will refer to them simply as differential equations. . if they are. Figure 1. we will ordinarily assume that any differential equation under study can be solved explicitly for the highest derivative that appears. Eq. The solution of y' = y 2 defined by y(x) = 1/(1. If the dependent variable is a function of two or more independent variables. y).1. we will see that an nthorder differential equation ordinarily has an nparameter family of solutionsone involving n different arbitrary constants or parameters. (11) and (12). . The righthand branch is the graph of a different solution of the differential equation that is defined (and continuous) on the different interval (1. All the differential equations we have mentioned so far are ordinary differential equations.6 shows the graphs of several such solutions. meaning that the unknown function (dependent variable) depends on only a single independent variable.1.9Bsin3x = 9y(x) for all x. = 9Acos3x...5. So the single formula y(x) = 1/(1 . then partial derivatives are likely to be involved.. The lefthand branch is the graph of a (continuous) solution of the differential equation y' = y 2 that is defined on the interval ( oo.x) actually defines two different solutions (with different domains of definition) of the same • differential equation y' = y 2 • ~ .ax 2 ' where k is a constant (called the thermal diffusivity of the rod).x).1. t) of a long thin uniform rod at the point x at time t satisfies (under appropriate simple conditions) the partial differential equation k au a2 u at . In this chapter we concentrate on firstorder differential equations of the form dx = f(x.6.. that is.y 1 11 y (n) . 5 ~~r~ on the whole real number line. . then two successive differentiations yield (14) y'(x) y"(x) = 3A sin 3x + 3B cos 3x. For example.x). we will always seek only realvalued solutions unless we warn the reader otherwise. 1). The three solutions y 1(x) = 3 cos 3x. • Although the differential equations in (11) and (12) are exceptions to the general rule. In both Eqs. Consequently. ···· Example 9 If A and B are constants and y(x) =A cos 3x + B sin 3x. where G is a realvalued function of n + 1 variables.y. Y2(x) = 2 sin 3x. the temperature u = u (x. the equation is called a partial differential equation. (14) defines what it is natural to call a twoparameter family of solutions of the secondorder differential equation 57~a~ ~ 0 5 y" + 9y = 0 (15) X FIGURE 1.y (n1)) ' (16) 5~L~ 3 0 3 X FIGURE 1.y. For this reason.1. dy (17) . the appearance of y' as an implicitly defined function causes complications. that the equation can be written in the socalled normal form _ G (x. and y3(x) = 3 cos 3x + 2 sin 3x of the differential equation y" + 9y = 0.
Yt = e3x. oo) of the solution of the different initial value problem y' 5 5 = y 2 .2x). The righthand branch passes through the point (2. y' = 3x 2. In Problems 1 through 12. • 0 X 5 FIGURE 1. y 2 =sin 2x y" = 9y.y = lnx. Yz = .5). 3. y(2) = 2.2x) .. and hence C = ~.lox.2y' + 2y = 0.1. + 2ex. y 1 = 1 x . I I I I r.JI :x= 3/2 I = 3 . Y2 = e. 7.7 shows the two branches of the graph y = 2/(3. y). 2) and is therefore the graph on ( ~.. elementary substitution methods (Section 1.6)are enough to enable us to solve a variety of firstorder equations having impressive applications. A typical mathematical model of an applied situation will be an initial value problem.x) of the differential equation dyjdx = y 2 discussed in Example 7. x 2 y" + xy' .x) satisfies the initial condition y(1) = 2. y = ex .. 4.2 solution 1.4).= 3 x 2 1 2 2 . solve the initial value problem dx  dy = y ' 2 y(1) = 2. y = 3e2x y" + 4y = 0. (18) on some interval containing x 0 • Example 10 Given the solution y(x) = 1/(C . primes denote derivatives with respect to x. The solutions of y' = y 2 defined by y(x) = 2/(3.1' so 2C. Substitution of the values x = 1 andy = 2 in the given solution yields 1 2 = y(l) = c . Solution We need only find a value of C so that the solution y(x) = 1/(C. solution of linear equations (Section 1. To solve the initial value problem dy dx = f(x.' 1 +x 2 10. Y2 = xe2x y" .O \(1. what can we do with it? We will see that a relatively few simple techniquesseparation of variables (Section 1. Y2 = ex sinx = cosxcos2x. ~) of the solution of the given initial value problem y' = y 2 . The central question of greatest immediate interest to us is this: If we are given a differential equation known to have a solution satisfying a given initial condition. consisting of a differential equation of the form in (17) together with an initial condition y(xo) = yo. y 1 9. y2 = 1 sinxcos2x 1. Throughout these problems. Yt =ex cosx. y' + 2xy 2 = O· y = .1. y = x 3 + 7 y' + 2y = 0.2x. y(l) = 2.2x). once found.8 Chapter 1 FirstOrder Differential Equations We also will sample the wide range of applications of such equations. y(xo) =Yo (18) means to find a differentiable function y = y(x) that satisfies both conditions in Eq. 5. y' = y 6.ex y" + 4y' + 4y = 0. Note that we call y(xo) = Yo an initial condition whether or not xo = 0. y 1 = cos 2x.3x 8. Yt = e. The lefthand branch is the graph on ( oo..7. x !C I I I Figure 1. y"+y = 3cos2x.. how do we actually find or compute that solution? And. 2. verify by substitution that each given function is a solution of the given differential equation. With this value of C we obtain the desired y(x) 5 y =2/(3.lox X .
28. 24. y(x) = cex. y) is the sum of x and y . y(x) = ln(x +C). How long does it take for the velocity of the boat to decrease to 1 m/s? To ro mls? When does the boat come to a stop? In Example 7 we saw that y(x) = 1/(C . y" + y = 0 43. 25. y(O) = 3 y' = y + 1. y) having the function g as its solution (or as one of its solutions).4y = 0 35. (b) Determine by inspection a solution of the initial value problem x' = kx 2 . xy' + y = 3x 40. y(rr) = 0 44. 32. 22. y" = 0 38. (a) Continuing Problem 43. the differential equation dyjdx = y 2 has exactly one solution y(x) satisfying the condition y (a) = b? = 10 23. Every straight line normal to the graph of g passes through the point (0. given any point (a. y(O) = 5 y' = X . 37. the time rate of change of the number N of those persons who have heard a certain rumor is proportional to the number of those who have not yet heard the rumor. 20. where Cis an arbitrary constant. y(O) = 0 dy x . (b) How would these solutions differ if the constant k were negative? Suppose a population P of rodents satisfies the differential equation dPjdt = kP 2 • Initially. 34. and vis decreasing at the rate of 1 m/s2 when v = 5 m/s. y(x) = ce. y(x) = cex +x.x. _ • _ 1 _ In x = 0.1 Differential Equations and Mathematical Models 9 11. x 2 y". 27.x) defines a oneparameter family of solutions of the differential equation dyjdx = y 2 • (a) Determine a value of C so that y(lO) = 10. 0). 1). 18. The time rate of change of the velocity v of a coasting motorboat is proportional"to the square of v.3y = x 3 . y(x) = (x +C) cosx. Then determine a value of the constant C so that y(x) satisfies the given initial condition. y 1 = x cos(lnx). (y') 2 + y 2 = 1 41. y2 = x sin(lnx) In Problems 13 through 16. 45. y' = y 2 39.in the manner of Eqs.x).2y =0 14. y) intersects the xaxis at the point (x/2. 21. 3y' = 2y 15. write. In Problems 32 through 36. In a city having a fixed population of P persons. y(x) = y(O) = 7 eYy' = 1. x). 31.+ 3y = 2x 5 . y(x) = tan(x 3 +C).3 . In Problems 27 through 31. 36. (a) If k is a constant. a function y = g(x) is described by some geometric property of its graph. y(l) = 17 y' = 3x 2 (y 2 + 1). y' + y = 0. How long will it take for this population to grow to a hundred rodents? To a thousand? What's happening here? Suppose the velocity v of a motorboat coasting in water satisfies the differential equation dvjdt = kv 2 • The initial speed of the motorboat is v(O) = 10 meters per second (m/s). first verify that y(x) satisfies the given differential equation. y(x) = ix 5 + cx. The acceleration dvjdt of a Lamborghini is proportional to the difference between 250 kmlh and the velocity of the car.. use your knowledge of derivatives to make an intelligent guess. show that a general (oneparameter) solution of the differential equation dx 2 =kx dt is given by x(t) = 1/(C. Does it appear that these solution curves fill the entire xyplane? Can you conclude that. (3) through (6) of this sectiona differential equation that is a mathematical model of the situation described. 3y" + 3y'. Use a computer or graphing calculator (if desired) to sketch several typical solutions of the given differential equation.1. y(O) cex 3 .2 X X 12. y' + y = ex 42. Write a differential equation of the form d y jdx = f (x. x 2 y +5xy +4y O. 13. 33. In Problems 37 through 42. 4y" = y 16. y(x) = x 3 (C + lnx). b) in the plane. .xy' + 2y . Then test your hypothesis. y) passes through the point ( y.yt. and their number is increasing at the rate of dPjdt = 1 rodent per month when there are P = 10 rodents. The time rate of change of a population P is proportional to the square root of P. and highlight the one that satisfies the given initial condition.y. The line tangent to the graph of g at the point (x. determine by inspection at least one solution of the given differential equation. 47. and then sketch graphs of solutions of x' = kx 2 with several typical positive values of x(O). The slope of the graph of g at the point (x.2·Y2. y' + 3x 2 y = 0.8 shows typical graphs of solutions ofthe form y(x ) = 1/(C. 29. 26.1. The line tangent to the graph of g at (x. 17. assume that k is positive. In Problems 17 through 26. . the time rate of change of the number N of those persons infected with a certain contagious disease is proportional to the product of the number who have the disease and the number who do not. Can you guess what the graph of such a function g might look like? 30. y(O) = 2 y' = 2y. (b) Is there a value of C such that y(O) = 0? Can you nevertheless find by inspection a solution of d yjdx = y 2 such that y(O) = 0? (c) Figure 1. substitute y = e'x into the given differential equation to determine all values of the constant r for which y = e'x is a solution of the equation.kt). y(x) = Ce2x. That is. y(O) = 1 y' + ytanx = cosx. 46. y(2) = 1 dx xy'. x(O) = 0.1. there are P(O) = 2 rodents. 19. In a city with a fixed population of P persons. y" + y' . The graph of g is normal to every curve of the form y = x 2 + k (k is a constant) where they meet.1.
. The graph y = Cx 4 for various values of C.2.. (c) Given any two real numbers a and b. if G' (x) f (x )then = y(x) = G(x ) +C. meaning that it involves an arbitrary constant C.1..1. (a) Show that y(x) = Cx 4 defines a oneparameter family of differentiable solutions of the differential equation xy' = 4y (Fig.2... (1) In this special case we need only integrate both sides of Eq.10 Chapter 1 FirstOrder Differential Equations 3~~~~~~~~~ C=2 C=1 C=O C= 1 C=2 C=3 100 80 60 40 20 . we obtain the particular solution of Eq. 1. The firstorder equation dyjdx = f(x. but is not of the form y(x) = Cx 4 . If G (x) is a particular antiderivative of fthat is.. Graphs of solutions of the equation dyfdx = y 2 • FIGURE 1. There we see that the constant C is geometrically the vertical distance between the two curves y(x) = G(x ) and y(x ) = G(x) +C.ry~.G(xo).8. so that C = Yo .. so dy dx = f(x). With this choice of C. (1). we need only substitute x = x o and y = Yo into Eq. (1) to obtain y(x) = J f(x) dx +C. explain whyin contrast to the situation in part (c) of Problem 47there exist infinitely many differentiable solutions of x y' = 4 y that all satisfy the condition y(a) =b. (3) The graphs of any two such solutions YI (x) G(x) + C1 and Y2(x) = G (x) + C2 on the same interval I are "parallel" in the sense illustrated by Figs. 1.2.. . y (xo) =YO· . To satisfy an initial condition y(x0 ) = y 0 .9. 48. and for every choice of C it is a solution of the differential equation in ( 1). (1) satisfying the initial value problem dy dx = f(x). (b) Show that y(x) = ~x 4 x4 if x < 0. (3) to obtain Yo = G(xo) + C.9). if x ~ 0 defines a differentiable solution of xy' = 4y for all x. y) takes an especially simple form if the righthandside function f does not actually involve the dependent variable y.r1 2 C=4 0 jJ~~~tH~~~y 2 20 40 60 80 1oo~~~~~~~~~ 54321 0 1 2 3 4 5 X FIGURE 1.1.1 and 1. (2) This is a general solution of Eq.
2. we call it the general solution of the differential equation.1. (2) immediately yields the general solution y(x) 2 0 = J (2x + 3) dx = x 2 + 3x +C. Solution 4 dy dx y(l) = 2. A natural question is whether a given general solution contains every particular solution of the differential equation. When this is known to be true. a general solution of a firstorder differential equation is simply a oneparameter family of solutions. It follows that C = 2.2 Integrals as General and Particular Solutions 4rT~~~~~~~~. 11 6 3 2 C=1 C=2 4 2 1 ~ 0 1\~~____::::::. Thus Eq. • Example 1 Solve the initial value problem .. Solution curves for the differential equation in Example 1.2.3.. (1) is of the form in (2)..2. it follows that every solution of Eq.3 shows the graph y = x 2 + 3x + C for various values of C..10 6 0 2 4 y(l) = (1) 2 + 3. thereby satisfying the initial condition 4 2 X . Graphs of y = sin x + C for various values of C. • .. Graphs of y = ~ x 2 + C for various values of C. FIGURE 1. We will see that this is the typical pattern for solutions of first order differential equations. by appropriate choice of C. so the desired particular solution is y(x) = x 2 + 3x . The particular solution we seek corresponds to the curve that passes through the point (1. 2 ~ 4 6 8 Figure 1.. Integration of both sides of the differential equation as in Eq.= 2x + 3.1. 2). (2) serves to define the general solution of (1). (1) + c = 2.2. we will first find a general solution involving an arbitrary constant C.. because any two antiderivatives of the same function f (x) can differ only by a constant.t. FIGURE 1.2.. We can then attempt to obtain.2. a particular solution satisfying a given initial condition y(xo) = YO· Remark: As the term is used in the previous paragraph. For example.::::::~HY 1 2 2 3 4 3 X 4 0 X 2 4 6 FIGURE 1. Ordinarily.
The motion of a particle along a straight line (the xaxis) is described by its position function X= j(t) (5) giving its xcoordinate at time t. dx V=. dx Velocity and Acceleration Direct integration is sufficient to allow us to solve a number of important problems concerning the motion of a particle (or mass point) in terms of the forces acting on it. The velocity of the particle is defined to be v(t) = f' (t). We simply integrate once to obtain ~. The observation that the special firstorder equation d y jdx = f (x) is readily solvable (provided that an antiderivative off can be found) extends to secondorder differential equations of the special form (4) in which the function g on the righthand side involves neither the dependent variable y nor its derivative dyjdx. where Cz is a second arbitrary constant.dt 2 • dv d 2x (7) J v(t) dt or in the definite integral form Equation (6) is sometimes applied either in the indefinite integral form x(t) = x(t) = x(t0 ) +it to v(s) ds. In effect. a . that is. which you should recognize as a statement of the fundamental theorem of calculus (precisely because d x jdt = v). dt (6) Its acceleration a (t) is a(t) = v'(t) = x"(t). = j y"(x) dx = j g(x) dx = G(x) + Ct.dt . Then another integration yields y(x) = J y' (x) dx = J [G(x) + Ct] dx = J G(x) dx + C 1x + C 2 . the secondorder differential equation in (4) is one that can be solved by solving successively the firstorder equations dv dx = g(x) and dy = v(x). in Leibniz notation. where G is an antiderivative of g and C 1 is an arbitrary constant. .12 Chapter 1 FirstOrder Differential Equations Secondorder equations.
So dx v(t) = .2 Integrals as General and Particular Solutions 13 Newton's second law of motion says that if a force F(t) acts on the particle and is directed along its line of motion. Its retrorockets. with Eq. suppose that the force F. of the particle at any time t in terms of its constant acceleration a.5 meters per second per second (mjs 2 ) (the gravitational acceleration produced by the moon is assumed to be included in the given deceleration). its initial velocity v0 . and substitution of this information into the preceding equation yields the fact that C 1 = v0 . (8) where m is the mass of the particle. Constant acceleration. These two arbitrary constants are frequently determined by the initial position x 0 = x(O) and the initial velocity v0 = v(O) of the particle.= at dt + vo . (11) Thus. and its initial position x 0 • Example 2 A lunar lander is falling freely toward the surface of the moon at a speed of 450 meters per second (mjs). x (t) = ~at 2 + vot + xo. Therefore. Then we begin with the equation dv = a dt (a is a constant)  (9) and integrate both sides to obtain v(t)= Jadt=at+C 1 • We know that v = v0 when t = 0. For instance. (10) A second integration gives X(t) = J V(t) dt = J (at+ Vo) dt = ~at 2 + Vot + C2.1. then ma(t) = F(t). (10) we can find the velocity. when fired. then the equation x"(t) = F(t)jm can be integrated twice to find the position function x(t) in terms of two constants of integration. that is. If the force F is known. and the substitution t = 0. and with Eq. X = Xo gives c2 = xo. At what height above the lunar surface should the retrorockets be activated to ensure a "soft touchdown" (v = 0 at impact)? . (11) the position. F = ma. provide a constant deceleration of 2. and therefore the acceleration a = F jm. are constant.
more precise values are 9.8 m/s2 The last line of this table gives values for the gravitational acceleration g at the surface of the earth.2. negative because the height x(t) is decreasing). where xo is the height of the lander above the lunar surface at the time t = 0 when the retrorockets should be activated. the footpoundsecond (fps) and meterkilogramsecond (mks) unit systems are used more generally in scientific and engineering problems.5 = 180 s (that is.500 metersthat is.7805 m/s2 and 32. However.5t .4.2. (13) yields x0 = 0. Thus the retrorockets should be activated when the lunar lander is 40.) . • Physical Units Numerical work requires units for the measurement of physical quantities such as distance and time. Although these approximate values will suffice for most examples and problems. We let t = 0 denote the time at which the retrorockets should be fired. From Eq. 1.25)(180) 2 + 450(180) = 40.450t + xo. Thus 1 N is (by definition) the force required to impart an acceleration of 1 mjs2 to a mass of 1 kg.5 kilometers above the surface of the moon. as indicated in Fig. Both systems are compatible with Newton's second law F = ma. Then v0 = 450 (m/s. Similarly. while mks units constitute the standard international system of scientific units. (12) we see that v = 0 (soft touchdown) occurs when t = 450/2. 1 slug is (by definition) the mass that experiences an acceleration of 1 ft/s 2 under a force of 1 lb. Force Mass Distance Time g pound (lb) slug foot (ft) second (s) 32 ft/s 2 newton (N) kilogram (kg) meter (m) second (s) 9.450 (12) (13) and Lunar surface x(t) = 1. and a = +2.4. x = 0 into Eq.14 Chapter 1 FirstOrder Differential Equations Solution We denote by x (t) the height of the lunar lander above the surface. FIGURE 1. and it will touch down softly on the lunar surface after 3 minutes of decelerating descent. (10) and (11) become v(t) = 2.088 ft/s 2 (at sea level at the equator). because an upward thrust increases the velocity v (although it decreases the speed Ivi).(1. 3 minutes). We sometimes use ad hoc unitssuch as distance in miles or kilometers and time in hoursin special situations (such as in a problem involving an auto trip). then substitution oft = 180. xo = 40. The lunar lander of Example 2. fps units are commonly used only in the United States (and a few other countries). Then Eqs.25t 2 .5 km ~ 25t miles. (We will use mks units in all problems requiring mass units and thus will rarely need slugs to measure mass. In fact.5.
54.8 N 2 9. Yo denotes the initial (t = 0) height of the body and vo its initial velocity.S. For conversions between fps and mks units it helps to remember that 1 in. x 2.48 em.4 kg.448 Njlb) = 444. Successive integrations (as in Eqs. speed limit of 50 mi/h means thatin international termsthe legal speed limit is about 50 x 1. 1 m1 = em 12 in. so its mass is m =.= g W 444. a mass m weighing 100 pounds has mks weight W = (100 lb)(4.45 kmjh. a mass of m = 20 kg has a weight of W = (20 kg)(9. To discuss vertical motion it is natural to choose the yaxis as the coordinate system for position.609 km.8 mjs ~ 45.54 em (exactly) and 1 lb ~ 4.gt + vo and y(t) = . then the effect of gravity on a vertically moving body is to decrease its height and also to decrease its velocity v = dyjdt . Vertical Motion with Gravitational Acceleration The weight W of a body is the force exerted on the body by gravity. (17) . = 30. = 2.4 em~ 1.8 mjs2 ). Substitution of a= g and F =Win Newton's second law F = ma gives W=mg (14) for the weight W of the mass mat the surface of the earth (where g ~ 32 ftjs 2 ~ 9. dt This acceleration equation provides a starting point in many problems involving vertical motion. Similarly.48 em ft = 160934. For instance. Thus a posted U .2 Integrals as General and Particular Solutions 15 Inches and centimeters (as well as miles and kilometers) also are commonly used in describing distances.1. (10) and (11)) yield the velocity and height formulas (16) v(t) = .m. if we ignore air resistance.448 N. Consequently. frequently with y = 0 corresponding to "ground level." If we choose the upward direction as the positive direction. For instance. = 5280 ft x 30.609 ~ 80. then the acceleration a = dvjdt of the body is given by dv (15) =g.8 N.8 mjs2 ) = 196 N. 1 ft and it follows that .tgt 2 + Vot + YO· Here.
Hence the maximum height that the ball attains is y(3) = ~ . substitution using (18) gives the differential equation (19) for the swimmer's trajectory y = y(x) as he crosses the river. ~~~~·~~~~. 3 + 0 = 144 (ft) and thus when t = 3 s. his velocity vector (relative to the riverbed) has horizontal component Vs and vertical component VR.= 3(1.5 shows a northwardflowing river of width w = 2a. (16)) is zero. then it returns to the ground when y(t) = ~.9)t( t + 10) = 0. (17)).·   Integration yields y(x) = J (3 .2. 32 (with the aid of Eq. then Eq.·~.16 Chapter 1 FirstOrder Differential Equations Example 3 (a) Suppose that a ball is thrown straight upward from the ground (y 0 = 0) with initial velocity v0 = 96 (ftjs.~~~ Example 4 Suppose that the river is 1 mile wide and that its midstream velocity is vo = 9 mijh.12x 2 )dx = 3x . (19) takes the form dy 2 . (9. + 96. so we use g = 32 ftjs 2 in fps units). 1. Suppose that a swimmer starts at the point (a. Hence the swimmer's direction angle a is given by tana = Vs VR FIGURE 1.8)t 2 + 49t = (4. 32. The lines x =±a represent the banks of the river and the yaxis its center.5. where VR = vo.2.8 mjs2 in mks units). Because tan a = d y jdx. so we use g = 9. and thus after 10 s in the air. (18) to verify that the water does flow the fastest at the center. 0) on the west bank and swims due east (relative to the water) with constant speed v5 .4x ). A swimmer's problem (Example 4). Then it reaches its maximum height when its velocity (Eq. Suppose that the velocity v R at which the water flows increases as one approaches the center of the river. and that VR = 0 at each riverbank.4x 3 +C . and indeed is given in terms of distance x from the center by (18) You can use Eq. v(t) = 32t + 96 = 0.5.~. If the swimmer's velocity is v 5 = 3 mi/h. dx ·H•· ~~. • A Swimmer's Problem yaxis Figure 1. (b) If an arrow is shot straight upward from the ground with initial velocity vo = 49 (m/s. As indicated in Fig.2.
=cos 2x.2. a(t) = 21. dy 1 4. Then y G)= 3 (D. a(t) = 17. . 1. v0 = .6 through 1. 2.xo = 1 . Sketch the graph of the resulting position function x (t) for 0 ~ t ~ 10. + 1. dx = 2x + 1. a(t) = 50.9. y(O) = 1 8 + 1 .4 (D 3 + 1 = so the swimmer drifts 2 miles downstream while he swims 1 mile across the river. dy dx = = = + 9. y(O) = 1 dx "' (5 . so + 1.2. . dx dy dy = (x  2) 2 . . v0 = 15. a(t) = 2t 15.= xex. v0 = 5. a particle starts at the origin and travels along the x axis with the velocity function v(t) whose graph is shown in Figs. Jorrr~. xo = 0 18. ..2 Integrals as General and Particular Solutions for the swimmer's trajectory.. a(t) = 3t. The initial condition y ( y(x) = 3x . v0 = 1.. dx 9.1.2. .4x 3 17 D = 0 yields C = 1. . find a function y = f (x) satisfying the given differential equation and the prescribed initial condition. vo = 10. dx = . 6.. . xo = 0 14.. r. a(t) = 16. 10 .= . y(2) =1 "' 4 3. y(2) = 1 20. 10 In Problems 1 through /0. y(O) dy dx 6 ~ =0 dy 10. initial position xo = x(O). dy dx dy 7.7. y(4) = 0 8. x 0 =8 In Problems 19 through 22.. y(l) dx x2 dy 1 dx = =5 2 4 6 8 10 Jx +2 xJx 2 10 x2 . x 0 = 4 4(t + 3)2 . xo = 20 12. FIGURE 1. (5. y(O) = 3 8 6 2. 5) 4 2 In Problems 11 through 18. . y(4) = 0 2 0 0  . Graph of the velocity function v(t) of Problem 20. and initial velocity v0 = v(O).2.find the position function x(t) of a moving particle with the given acceleration a(t). dy 1. 5) + 1)3 v0 = 0.7 . 6 8 10 FIGURE 1.8. Vo = 10. Graph of the velocity function v(t) of Problem 21.6. x 0 = 5 13. a(t) = 50sin5t. FIGURE 1. y(O) = 0 1 . x 0 = vt+4 1 (t ~· 1 8 6 1 Vo = 1. s.. 11.. Graph of the velocity function v (t) of Problem 19.ji.2. a(t) = 20. • E Problems 19.
How high could this person throw the ball on the planet Gzyx of Problem 29? 35.5 mi and v0 = 9 mijh as in Example 4. Chapter 1 FirstOrder Differential Equations 10 . The car in question is known to have a constant deceleration of 20 m/s2 under these conditions. 35 miles away. How far does the car travel before coming to a stop? 26. a projectile is fired straight upward toward the bomb. 5) (7. At noon a car starts from rest at point A and proceeds at constant acceleration along a straight road toward point B. It strikes the ground with a speed of 60 mjs. a ball dropped from a height of 20 ft hits the ground in 2 s. 30. and Vs = 3 mi/h as in Example 4.2. when fired. If the car starts from rest (x 0 = 0. If a ball is dropped from the top of a 200fttall building on Gzyx. The skid marks made by an automobile indicated that its brakes were fully applied for a distance of 75 m before it came to a stop. (18). v 0 = 0). 36. What is the maximum height attained by the arrow of part (b) of Example 3? 24. what is the distance from A to B? 38. (c) its total time in the air. A projectile is fired straight upward with an initial velocity of 100 mjs from the top of a building 20m high and falls to the ground at the base of the building. (b) when it passes the top of the building.was the car traveling when the brakes were first applied? rather than the quadratic function in Eq. A baseball is thrown straight downward with an initial speed of 40 ft/s from the top of the Washington Monument (555 ft high). how far will it skid if it is moving at 100 km/h when the brakes are applied? 33. From the ground directly beneath the helicopter. If a = 0. how long will it take to hit the ground? With what speed will it hit? 34.J'Iih.in km/h. 23. Find (a) its maximum height above the ground.000 mi/h2 • At what height above the lunar surface should the astronauts fire the retrorockets to insure a soft touchdown? (As in Example 2. 27. but that the velocity of the river is given by the fourthdegree function . With what initial velocity should the projectile be fired. How long does it take to reach the ground? With what speed does the ball strike the ground? 25. Show that the speed with which it strikes the ground is v = . At noon a car starts from rest at point A and proceeds with constant acceleration along a straight road toward point C.5 mi. A ball is thrown straight downward from the top of a tall building. How long does it take to reach the ground. provide a constant deceleration of 20. ignore the moon's gravitational field. If the car reaches B at 12:50 P.. (3.M. How fast.18 22. 41. 8 6 . Now find how far downstream the swimmer drifts as he crosses the river. v0 = 9 mijh. 5) 4 6 8 10 FIGURE 1. Suppose a woman has enough "spring" in her legs to jump (on earth) from the ground to a height of 2. A stone is dropped from rest at an initial height h above the surface of the earth. Graph of the velocity function v(t) of Problem 22.6)t (ft/s 2 ). exactly 2 seconds after the bomb is released. On the planet Gzyx. A ball is dropped from the top of a building 400 ft high.= dv dt (0. what is that deceleration? For how long does the skid continue? 31. Suppose that a car skids 15 m if it is moving at 50 km/h when the brakes are applied. If the constantly accelerated car arrives at C with a velocity of 60 mijh. what must the swimmer's speed v 5 be in order that he drifts only l mile downstream as he crosses the river? 40. in order to hit the bomb at an altitude of exactly 400 feet? 42. Its retrorockets.~. The initial speed of the ball is 10 mjs. Suppose that a = 0. A car traveling at 60 mi/h (88 ftj s) skids 176 ft after its brakes are suddenly applied. A bomb is dropped from a helicopter hovering at an altitude of 800 feet above the ground. A spacecraft is in free fall toward the surface of the moon at a speed of 1000 mph (mi/h). How tall is the building? 28.25 feet. find the distance it has traveled at the end of the first 10 s and its velocity at that time. Under the assumption that the braking system provides constant deceleration. A diesel car gradually speeds up so that for the first 10 s its acceleration is given by 32. at what time does it arrive at C? 39. The brakes of a car are applied when it is moving at 100 km/h and provide a constant deceleration of 10 meters per second per second (mjs 2) . Assuming that the car has the same constant deceleration.3 ft/s 2how high above the surface will she rise? 37.. If she jumps straight upward with the same initial velocity on the moonwhere the surface gravitational acceleration is (approximately) 5. with a velocity of 60 mijh.~.9. and with what speed does the baseball strike the ground? 29.12)t 2 + (0. A person can throw a ball straight upward from the surface of the earth to a maximum height of 144 ft.) .
and therefore cannot be evaluated explicitly. 19 and how far will it have traveled by then? 44. y) of the . y(x)) through which it passesthat is. y) (1) y where the righthand function f (x. A solution curve for the differential equation y' = x . y). they found that when its brakes were applied at 25 mph. and hence write y(x) = J f(x.0098 rn/s 2 • Suppose this spacecraft starts from rest at time t = 0 and simultaneously fires a projectile (straight ahead in the same direction) that travels at onetenth of the speed c = 3 x 108 rn/s of light. y) involves both the independent variable x and the dependent variable y. . the car skidded only 45 feet before coming to a stop. At each point (x. the graphical and numerical methods of this and later sections can be used to construct approximate solutions of differential equations that suffice for many practical purposes. Example 1  ········ . How long will it take the spacecraft to catch up with the projectile. y2).1 shows a solution curve of the differential equation y' = x . the value of f(x.y 1 at the point (xJ. y 1) . But the driver's skid marks at the accident scene measured 210 feet.2 (a)(d) show slope fields and solution curves for the differential equation d y =ky dx (2) with the values k = 2.y 3 at the point (x3. y(x)) dx +C. the solutions of such a simplelooking differential equation as y' = x 2 + y 2 cannot be expressed in terms of the ordinary elementary functions studied in calculus textbooks.xyplane whose tangent line at each point (x. y(x)). .y together with tangent lines having • slope m 1 = x 1 . a spacecraft propelled by the solar wind.1. y'(x) = f(x. A driver involved in an accident claims he was going only 25 mph.OOlg = 0. Slope Fields and Graphical Solutions There is a simple geometric way to think about solutions of a given differential equation y ' = f(x . Assuming the same (constant) deceleration.3. y) . ·· · ~···· FIGURE 1. and • slope m 3 = x 3 . Indeed. Fig. Its aluminized sail provides it with a constant acceleration of O..  Figures 1. We might think of integrating both sides in (1) with respect to x. This geometric viewpoint suggests a graphical method for constructing approximate solutions of the differential equation y' = f(x . Nevertheless. y) determines a slope m = f (x.1. 1.3. Thus a solution curve of the differential equation y' = f (x. . 1.3 Slope Fields and Solution Curves 43. y3). Note that each slope field yields important qualitative information about the set of all solutions . • slope m2 = x 2 . Actually. Arthur Clarke's The Wind from the Sun (1963) describes Diana. y).xyplane.. there exists no straightforward procedure by which a general differential equation can be solved explicitly. because the indicated integral involves the unknown function y(x) itself. When police tested his car. and 3 of the parameter k in Eq. All these line segments constitute a slope field (or a direction field) for the equation y ' = f(x. y) . this approach does not lead to a solution of the differential equation. y) has slope m = f(x. 0. determine the speed he was actually traveling just prior to the accident.Y2 at the point (x2. For instance. (2). A solution of the differential equation is simply a differentiable function whose graph y = y(x) has this "correct slope" at each point (x. ·· ··. E Sl~pe Fields and Solution Curves Consider a differential equation of the form dy dx = f(x. However. Through each of a representative collection of points (x .5.·····. y) in the plane we draw a short line segment having the proper slope m = f (x .. y )the graph of a solution of the equationis simply a curve in the .3. y).y together with its tangent lines at three typical points. y).
The numerical slope m = x . All this is apparent from slope fields like those in Fig. Through each point a solution curve should proceed in such a direction that its tangent line is nearly parallel to the nearby line segments of the slope field..2.. 4).. 1.5)y..l_~~~3ill~i!!!i!l~ 1 2 .3.nTrnr>oTrn~.2(a) Slope field and solution curves for y' = 2y.2(a) and (b) suggest that each solution y(x) approaches ±oo as x ~ +oo if k > 0.2(c) Slope field and solution curves for y' = .. 1. (2) is given explicitly by y (x) = C ekx. although the sign of k determines the direction of increase or decrease of y(x). 1.3. its absolute value lkl appears to determine the rate of change of y(x). Moreover. even without knowing that the general solution of Eq.3.or.~~~~ 4~~~~~~~~70 3 3 2 2 1 1 2 3 2 0 X 2 3 4 X FIGURE 1. 1.rvr.3.y appears at the intersection of the horizontal xrow and the vertical ycolumn of the table. we can attempt to sketch freehand an approximate solution curve that threads its way through the slope field. of the differential equation.3.3 shows a table of slopes for the given equation.2(b) Slope field and solution curves for y' = (0. (Of Solution .y. • A slope field suggests visually the general shapes of solution curves of the differential equation. FIGURE 1. 3 2 .. Starting at any initial point (a.2(d) Slope field and solution curves for y' = 3y . 4 3 4 .:. Figs. 0 ~. you can see that it was easily and quickly constructed.3..2(c) and (d) suggest that y (x) ~ 0 as x ~ +oo if k < 0. Example 2 Construct a slope field for the differential equation y' = x . b). If you inspect the pattern of upperleft to lowerright diagonals in this table.. For instance. 0 1 I I 2 2 3 X X FIGURE 1.20 Chapter 1 FirstOrder Differential Equations 4. whereas Figs.:.3. Solution Fig..3.4. following the visible line segments as closely as possible. . FIGURE 1.y and use it to sketch an approximate solution curve that passes through the point (.
y for 4 2 1 0 1 x. together with typical solution curves treading through this slope field. 4).y of Example 2..4. FIGURE 1.3.3. Although a spreadsheet program (for instance) readily constructs a table of slopes as in Fig.6. such commands are illustrated in the application material for this section..1.5 shows an approximate solution curve sketched through the point (4. If you look closely at Fig.3.j.':7f\J 1 2 3 4 4 . This inference illustrates the fact that a slope field can suggest tangible information about solutions that is not at all evident from the differential equation itself.. 0 Jii\i+C:':. 1. a more complicated function f(x.6. 1.:>.3. and Fig. you can verify that the linear function y = x . At each point it appears to proceed in the direction indicated by the nearby • line segments of the slope field.r:. ..y:.4 shows the corresponding slope field. FIGURE 1.3.3.3. 4) so as to follow as this slope field as closely as possible. the more accurately solution curves can be visualized and sketched. Values of the slope y' 5 =x \ \ I I I \ \ .y corresponding to the table of slopes in Fig..y.3..5.3 Slope Fields and Solution Curves 21 x\Y 4 3 4 0 1 3 1 0 1 2 2 1 0 1 3 0 1 5 2 6 5 3 4 3 7 6 5 2 1 4 3 4 8 7 6 5 2 1 0 1 2 3 2 1 0 1 4 3 2 3 4 5 6 1 2 3 0 1 2 1 4 3 4 5 6 2 3 0 2 1 0 4 3 2 3 4 5 6 2 3 4 7 8 4 5 1 2 3 . \ \ \ \ \ \ \ \ \ / \ " " \ \ " / 5~~~r~ / I I I 4 3 2 I 0 \ \ \ "/ "I I I 0 X "" I I O~r*~~r~++~ ".'j!.3.1 is a solution of the equation y' = x . 1.3. most computer algebra systems include commands for quick and ready construction of slope fields with as many line segments as desired.) Figure 1. y ~ 7 4 1 2 ~ 2 1 0 4.~. by tracing the 4 3 I .3. and it appears likely that the other solution curves approach this straight line as an asymptote as x ~ +oo. 1.>. you may spot a solution curve that appears to be a straight line! Indeed.3. Can you. Figure 1. Slope field for y' = x . The more line segments are constructed. y) on the righthand side of the differential equation would necessitate more complicated calculations. Slope field and typical solution curves for y' = X .3. course.3. The solution curve through ( 4.6 shows a "finer" slope field for the differential equation y' = x .3 2 1 0 X I 2 3 4 FIGURE 1./ 5 5 / I I I I I I I I I I I I I I 5 1 2 3 4 / 5~~~ 5 0 X 5 FIGURE 1. However. it can be quite tedious to plot by hand a sufficient number of slope segments as in Fig. 1.4..>.y.
This implies thathowever you throw itthe baseball should approach the limiting velocity v = 200 ft/s instead of accelerating indefinitely (as it would in the absence of any air resistance).0.22 Chapter 1 FirstOrder Differential Equations appropriate solution curve in this figure.16v. y(4) = 4? ~ 2 for the solution y(x) of Applications of Slope fields The next two examples illustrate the use of slope fields to glean useful information in physical situations that are modeled by differential equations. infer that y(3) the initial value problemy'= x. then the initial acceleration given by Eq. h Perhaps a catcher accustomed to 100 milh fastballs would have some chance of fielding this speeding ball. (4) is negative.kv. Note that all these solution curves appear to approach the horizontal line v = 200 as an asymptote. because of air resistance. The handy fact that 60 milh = 88 ft/s yields V ft = 200s X 60 milh 88 ft/s ~ mi 136.in the absence of air resistance. 1. then Eq.(0. (3) A typical value of the air resistance proportionality constant might be k = 0. Its velocity therefore remains unchanged.3. If the yaxis is directed downward.7. It therefore seems quite reasonable that.0.36 . you can use your laptop's computer algebra system to construct a slope field for the differential equation dv . • Comment If the ball's initial velocity is v(O) = 200. while its acceleration due to air resistance is negative. then the ball's velocity v = dyjdt and its gravitational acceleration g = 32 ft/s 2 are both positive.16)(200) = 0. But if the initial velocity is less than 200.16. You might like to verify that. Example 3 is based on the fact that a baseball moving through the air at a moderate speed v (less than about 300 ft/s) encounters air resistance that is approximately proportional to v.y. then it experiences both the downward acceleration of gravity and an upward acceleration of air resistance. Slope field and typical solution curves for v' = 32. Hence its total acceleration is of the form dv dt = g. dt (4) 5 10 15 20 25 FIGURE 1. the baseball will approach a limiting velocity of 200 ft/s. (4) gives v' (0) = 32 . so the ball slows down as it falls.= 32. and hence v(t) 200 is a constant "equilib = rium solution" of the differential equation. If the baseball is thrown straight downward from the top of a tall building or from a hovering helicopter.16v. Example 3 Suppose you throw a baseball straight downward from a helicopter hovering at an altitude of 3000 feet. so the ball speeds up as it falls. If the initial velocity is greater than 200.7. The result is shown in Fig.whatever initial velocity it starts with.. together with a number of solution curves corresponding to different values of the initial velocity v(O) with which you might throw the baseball downward. • . then the initial acceleration given by (4) is positive. You wonder whether someone standing on the ground below could conceivably catch it. so the ball experiences no initial acceleration.3.this ball would hit the ground at over 300 milh. In order to estimate the speed with which the ball will land.
= 0. then the initial rate of change given by (6) is positive. We may also want to know whether there is only one solution of the equation satisfying a given initial conditionthat is. X 1 y(O) = 0 (7) has no solution. dt 250 200 (6) ~150~~~~~~~~~ 100 50 00~~~~~ 25 50 t 75 100 The positive term 0.06P.0004P 2 • Comment If the initial population is P(O) = 150.3. We see this graphically in Fig.0.3.8 shows a slope field for Eq. (6) gives P' (0) = 0.06P.0004?(150. then the initial rate of change given by (6) is negative. This implies thatwhatever the initial populationthe population P(t) approaches the limiting population P = 150 as t + oo. so the population experiences no initial (instantaneous) change. and hence P(t) 150 is a constant "equilibrium solution" of the differential equation. because no solution y(x) = j(ljx) dx =In lxl + C of the differential equation is defined at x = 0. then Eq. for instance). it is wise to know that solutions actually exist. Figure 1.8.3 Slope Fields and Solution Curves 23 In Section 1. Example 5 (a) [Failure of existence] The initial value problem y I = . Note that all these solution curves appear to approach the horizontal line P = 150 as an asymptote. This means that M is the maximum population that this environment can sustain on a longterm basis (in terms of the maximum available food.0004? 2 .0004? 2 represents the inhibition of growth due to limited resources in the environment. It therefore remains unchanged.0. 1.150) = 0.1. 7 we will discuss in detail the logistic differential equation dP = kP(M.P) = 0. But if the initial population is less than 150. Slope field and typical solution curves for P' = 0. 0).0004 and M = 150. so the population immediately begins to decrease.0004(150)(150. so the population immediately begins to increase.P) dt (5) that often is used to model a population P (t) that inhabits an environment with carrying capacity M. whether its solutions are unique. It therefore seems quite reasonable to conclude that the population will approach a limiting value of 150whatever the (positive) initial population. which shows a direction field and some typical solution curves for the equation y' = ljx. (6). then the logistic equation in (5) takes the form dP .06P on the right in (6) corresponds to natural growth at a 6% annual rate (with timet measured in years). It is apparent that the indicated direction field "forces" all solution curves near the yaxis to plunge downward so that none can pass through the point (0. If the initial population is greater than 150. . • FIGURE 1. Example 4 If we take k = 0. • = Existence and Uniqueness of Solutions Before one spends much time attempting to solve a given differential equation. together with a number of solution curves corresponding to possible different values of the initial population P(O).9. The negative term 0.3.
Figure 1.24 Chapter 1 FirstOrder Differential Equations I I I I I I I I I I I I I I I I I I I I I I I I / I I / I I / I / I / I I / / / . the solution interval I may not be as "wide" as the original rectangle R of continuity./.) .y. The rectangle R and xinterval I of Theorem I. you can readily verify that the initial value problem (8) y' = 2y/y.9. see Remark 3 below. and the solution curve y = y(x) through the point (a./.11. This raises an immediate question as to whether the mathematical model adequately represents the physical system. The theorem stated below implies that the initial value problemy' = f(x. Direction field and typical solution curves for the equation y' = ljx. for some open interval I containing the point a./. Methods of proving existence and uniqueness theorems are discussed in the Appendix. 0) 0 X FIGURE 1. We see that the curve y 1 (x) = x 2 threads its way through the indicated direction field./. but that our proposed mathematical model involves a differential equation not having a unique solution satisfying those conditions. Questions of existence and uniqueness of solutions also bear on the process of mathematical modeling.1 0 shows a direction field and these two different solution curves for the initial value problem in (8)..3. y) are continuous on some rectangle R in the xyplane that contains the point (a. FIGURE 1./. Then. y(O) = 0. Suppose that we are studying a physical system whose behavior is completely determined by certain initial conditions. the initial value problem X dy dx = f(x. y).9. has one and only one solution that is defined on the interval I. y(a) = b has one and only one solution defined near the point x =a on the xaxis./ 0~~~~~~ Y2(x) =0 (0. Direction field and two different solution curves for the initial value problemy' = 2JY. whereas the differential equation y' = 2JY specifies slope y' = 0 along the xaxis y 2 (x) = 0. y(O) = 0 has the two different solutions y 1 (x) = x 2 and y 2 (x) = 0 (see Problem 27). before we can speak of "the" solution of an initial value problem. (As illustrated in Fig.(x)=x2 /.3 .10. we need to know that it has one and only one solution.a y are continuous near the point (a. y) and its partial derivative D y f(x. • Example 5 illustrates the fact that. y(a) = b (9) FIGURE 1. y R THEOREM 1 Existence and Uniqueness of Solutions b Suppose that both the function f(x. (b) [Failure of uniqueness] On the other hand. / / / / / / .3./.11. b) in its interior.3. 1. provided that both the function f and its partial derivative a f . y). b) in the xyplane. b)./.3.
y) and/or its partial derivative 8ff8y fail to satisfy the continuity hypothesis of Theorem 1. b)..1 we examined the especially simple differential equation dyjdx = y 2 • Here we have f(x. each of which satisfies the initial condition solutions y 1 (x) = x 2 and y2 (x) y(O) = 0. (0.x) is discontinuous at x = 1. Both of these functions are continuous everywhere in the xyplane.3. dx dy (11) Applying Theorem 1 with f(x. y) = 2...3 Slope Fields and Solution Curves 25 Remark 1: In the case of the differential equation dyfdx = y of Example 1 and Fig.._ •····· ···~··~·· . (11) must have a unique solution near any point in the xyplane where x =f. .JY of Example 5(b) and Eq. each solution y(x) = cex actually is defined for all x.12). so our unique continuous solution does not exist on the entire interval 2 < x < 2. and hence at the point (0. Theorem 1 guarantees a unique solutionnecessarily a continuous functionof the initial value problem 6 Remark 3: dy dx 4 y = 1/(l x) =y 2 ' y (0) =1 (10) ~ 2 r l __. 1) leaves the rectangle R before it reaches the right side of R. Example 6 Consider the firstorder differential equation x ... 1) lies in the interior of this rectangle. Although the theorem ensures existence only on some open interval containing x =a. (8). The solution curve through the initial point (0. 0).= 2y. and in particular on the rectangle 2 < x < 2. if the function f (x. the function f(x. 0 < y < 2... 0. ··········•· . but the partial derivative 8fj8y = 1/..3.JY is continuous wherever y > 0. so Theorem 1 implies the existence of a unique solution for any initial data (a. Remark 2: In the case of the differential equation dyjdx = 2.. we see immediately by substitution in (11) that y(x) = Cx 2 (12) .2(c).3. 1. both the function f(x. y) = y 2 and ajjay = 2y. This is why it is possible for there to exist two different 0. y) = y and the partial derivative 8fj8y = 1 are continuous everywhere. the reason is that the solution curve provided by the theorem may leave the rectanglewherein solutions of the differential equation are guaranteed to existbefore it reaches the one or both ends of the interval (see Fig. that we discussed in Example 7.. Indeed. Indeed this is the solution 1 y(x) = 1 IR : I x 4 2 0 X 2 4 FIGURE 1. But y(x) = 1/(1 . Geometrically. l) on some open xinterval containing a = 0.1.. then the initial value problem in (9) may have either no solution or manyeven infinitely manysolutions. y) = 2yjx and 8ff8y 2/x .12.JY is discontinuous when y = 0. Because the point (0. • The following example shows that. 1. 0 2 I I i . Thus the solution interval I of Theorem 1 may not be as wide as the rectangle R where f and 8ff8 y are continuous. = In Example 7 of Section 1. .. we conclude that Eq.
Then for any value of the constant C the function defined by X x2 y(x) = { FIGURE 1. but none of them passes through any other point on the yaxis. then the initial value problem dy xdx = 2y. • no solution if a = 0 but b =1= 0. We therefore see that Theorem 1 (if its hypotheses are satisfied) guarantees uniqueness of the solution near the initial point (a.3.3.) Observe that all these parabolas pass through the origin (0. the solution y (x) = x 2 of the initial value problem in (17) exists on the whole xaxis. 1).26 Chapter 1 FirstOrder Differential Equations (0. if a =I= 0. • infinitely many solutions if a = b = 0. y(a) =b (15) has a unique solution on any interval that contains the point x = a but not the origin x = 0 In summary.13. Finally. It follows that the initial value problem in (13) has infinitely many solutions. but this solution is unique only on the negative x axis .14. 0) satisfies Eq. if X> 0 (16) Still more can be said about the initial value problem in (15). 1.3. Cx 2 is continuous and satisfies the initial value problem dy xdx=2y. but the initial value problem dy xdx=2y. There are infinitely many solution curves through the point ( 1. whose solution curves are the parabolas y = Cx 2 illustrated in Fig. has no solution if b =I= 0.3. b) may eventually branch elsewhere so that uniqueness is lost. Thus a solution may exist on a larger interval than one on which the solution is unique. Thus the unique solution curve near ( 1. 0).14. 1. b) if a =I= 0. (In case C = 0 the "parabola" is actually the xaxis y = 0. (17) For a particular value of C.13. note that through any point off the yaxis there passes only one of the parabolas y = C x 2 . b) (0. 1) indicated in Fig.14. There are infinitely many solution curves through the point (0. y(O)=b (14) FIGURE 1. the solution curve defined by ( 16) consists of the left half of the parabola y = x 2 and the right half of the parabola y = C x 2 . (11) for any value of the constant C and for all values of the variable x. but no solution curves through the point (0. the initial value problem dy xdx =2y. b) if b :f: 0. 1) branches at the origin into the infinitely many solution curves illustrated in Fig. 1.3. y(1)=1. 0). For instance.oo < x < 0. Consider a typical initial point off the yaxisfor instance the point ( 1. the initial value problem in (15) has • a unique solution near (a. . b). In particular. • if X S 0. Hence. but a solution curve through (a. y(O)=O (13) X has infinitely many different solutions.
.
>.. dy 18..1. y(O) y(O) . 7. y' = X+ y. 21. ___ 0 X 2 2 3 X FIGURE 1. y(O) = 0.1.=Jx y. y dx = x . use this solution curve to estimate the desired value of the solution y(x). _ =x2 _ y2.{/Y. determine whether Theorem 1 does or does not guarantee existence ofa solution ofthe given initial value problem.24. dy 14.28 dx Chapter 1 FirstOrder Differential Equations 10. = Jx.3. dy 20. dy 2 8. y(l) = 1 y(O) = 1 y(O) = 0 y(2) = 2 y(2) = 1 y(O) = 1 y(l) .4) =? =? . dy 19. 0 I 2 2 1 I . . Finally. y dx dy 2 9.23. dx = .=xlny· dx ' 13. dx I I I I I I I = I 2 1.>.1 I I I I I FIGURE 1. first use the method of Example 2 to construct a slope field for the given differential equation.y. . .=x y2 dx 3 = x.. If existence is guaranteed.4) y(.. X dy 15.3. y(1) = 1 dx dy 12. y' = y .21.  dy .{/Y.. dx FIGURE 1. . .X.>. 0 X 2 3 In Problems 21 and 22. .. dx =0 =0 1 2 = ln(l + y2).  I 2 dy dx = . . ________ ___ .3. dy 11. dx dy 16. Then sketch the solution curve corresponding to the given initial condition. =smx+smy dy dx • = X 2 + Slll y 2 .>. determine whether Theorem 1 does or does not guarantee uniqueness of that solution. . FIGURE 1. y(4) = 0.3. . 22.22.. y(.= 2x2y2. =X y dx 3 2 In Problems I 1 through 20. dy 17.
A suggestion for Problem 29. tial value problem y' = 31. +Jf=Y2.3 Slope Fields and Solution Curves Problems 23 and 24 are like Problems 21 and 22. 30. You bail out of the helicopter of Example 3 and pull the ripcord of your parachute. Carry out an investigation similar to that in Problem 30. 1. or infinitely many solutions. (ii) a unique solution that is defined for all x? 28. 24.) Sketch a variety of such solution curves. construct a slope field for this differential equation and sketch the appropriate solution curve. Verify that if c is a constant.0225 p . Then determine (in terms of a and b) how many different solutions the initial value problem y' = 4xJY.1. none. Verify that if k is a constant.l. Then determine (in terms of a and b) how many different solutions the iniy (a) = b has. y(a) = b hasone.c) { 3 for x ~ c. then the function y(x ) = kx satisfies the differential equation xy' = y for all x . lfyou wish (and know how). then the function defined piecewise by y(x ) ~ { ~x. how long will it take the number of deer to double? What will be the limiting deer population? The next seven problems illustrate the fact that. if X~ C + 7r 27. y (a) = b has.= 0.3. y(2) =? 29 29. 23. Verify that if c > 0. FIGURE 1.I. (a) Verify that if c is a constant. for x > c 2? 25. or infinitely many. y' = x 2 + y 2 . if the hypotheses of Theorem 1 are not satisfied. finitely many solutions. y(O) = 0. b) of the xyplane such that the initial value problemy' = 3y 213 .) Sketch a variety of such solution curves. Is there a point (a.25. Sketch a variety of such solution curves for different values of c. you can check your manually sketched solution curve by plotting it with the computer. except with the differential equation y' = Does it suffice simply to replace cos(x. y(O) = b have (i) no solution.c) with sin(x. Then determine (in terms of a and b) how many different solutions the initial value problem x y' = y. if x 2 > c satisfies the differential equation y' = 4x JY for all x . Can you also use the "left half" of the cubic y = (x . for x > c satisfies the differential equation y' = 2JY for all x (including the point x = c). (b) For what values of b does the initial value problemy' = 2JY. then the function defined piecewise by JI=Y2 JI=Y2. c. y(x) = { 0 (x 2  c) 2 if x 2 ::s.25.c) in piecing together a solution that is defined for all x ? 32. What will your limiting velocity be? Will a strategically located haystack do any good? How long will it take you to reach 95% of your limiting velocity? 26.c) 3 in piecing together a solution curve of the differential equation? (See Fig. y(a) = b has either no solution or a unique solution that is defined for all x? Reconcile your answer with Theorem 1. y ). but now use a computer algebra system to plot and print out a slope fieldfor the given differential equation. y' =X+ y(2) = 0. v(O) = 0. satisfies the differential equation y' = for all x . y (a) = b may have either no solutions. Construct a figure illustrating the fact that the initial value problemy' = 2JY. Verify that if c is a constant. then the function defined piecewise by y(x) ~ ~x . Suppose the deer population P(t)in a small forest satisfies the logistic equation dP . (3). then the function defined piecewise by y(x) = 1 +1 1 cos(x. satisfies the differential equation y' = 3y 213 for all x.0003 P 2 • dt c Construct a slope field and appropriate solution curve to answer the following questions: If there are 25 deer at time t = 0 and t is measured in months. then the initial value problem y' = f (x.0. y(O) = 0 has infinitely many different solutions.c) 2 for x ~ c . Now k = 1.6 in Eq. (Perhaps a preliminary sketch with c = 0 will be helpful. so your downward velocity satisfies the initial value problem dv dt = 32.c) ifx~c.6v. ifc < x < c +n. Construct a slope field and several of these straight line solution curves. y In order to investigate your chances of survival. . y(2) =? 1 2.3.
5 .2. Mathematica™. b = 3.x + 1 with initial values y(1) = 1. 1. 1..1).. " " .. '"" "\ II II \ _. Slope field and solution curves for the differential equation  dy dx = sm(x .8.2 and y(1) = 0. y(a) = b has.5.27. 2. b). as do some graphing calculators (see Fig. (a) Use the direction field of Problem 6 to estimate the values at x = 2 of the two solutions of the differential equation y' = x .01 and y( 3) = 2. (b) Use a computer algebra system to estimate the val ues at x = 3 of the two solutions of this differential equation with initial values y( 3) = 3. determine (in terms of a and b) how many different solutions the initial value problem x 2 y' + y 2 = 0.5 :S x . Sketch a variety of such solution curves for different values of c.. 1. verify that the function defined by y(x) = xj(cx. 35.x) =sin(x._ .3. y =. .y(x)).3. 4 and window . If c Chapter 1 FirstOrder Differential Equations "I 0. Also.. ' ______ .2 andy( 3) = +0. 2.3. with initial points (0. The applications manual accompanying this textbook includes discussion of Maple™.27). \ \ " " " "" I \ \ I \ \ I I \ "" (b) Use a computer algebra system to estimate the val' \ \ I \ \ ' "" ues at x = 3 of the two solutions of this differential equation with initial values y(. 5). FIGURE 1. 1. Slope field for x 2 y' + y 2 = 0 and graph of a solution y(x) = x j(cx . FIGURE 1.. \ \ I \ ___ .30 33.3.1) (with graph illustrated in Fig.y ) .26..26) satisfies the differential equation x 2 y' + y 2 = 0 if x "I 1jc. The lesson of this problem is that big changes in initial conditions may make only small differences in results. and MATLAB™ resources for the investigation of differential equations. The lesson of this problem is that small changes in initial conditions can make big differences in results. For instance.. 34.99. y(x). 0. x=5 .3) = 0.5 and y( 3) = +0. the Maple command with(DEtools): DEplot(diff(y(x). note the constantvalued function y(x) = 0 that does not result from any choice of the constant c. 5.y + 1 with initial values y( 3) = 0. Finally.3 Application Widely available computer algebra systems and technical computing environments include facilities to automate the construction of slope fields and solution curves. y ::: 5 on a TI89 graphing calculator. (a) Use the direction field of Problem 5 to estimate the values at x = 1 of the two solutions of the differential equation y' = y .
28. Upper Saddle River. a number of apparent straight line solution curves should be visible.3. Use a graphing calculator or computer system in the following investigations.math. {y. produce slope fields similar to the one shown in Fig. 1.3 Slope Fields and Solution Curves and the Mathematica command << Graphics\PlotField.y).eduldfield). FIGURE 1. 0 l7::i~~~~~~~~'! 1 31 2 3 4 X FIGURE 1.29). 10 ~ y ~ 10. but with a larger window than that of Fig.2 x.3.3. Sin[xy]}.3.2 tan.29.y). 5.rice. you can (with mouse button clicks) plot both a slope field and the solution curve (or curves) through any desired point (or points).m PlotVectorField[{l.n /2 corresponding to the initial condition y(n/2) = 0? Are there any values of C for which the corresponding solution curves lie close to this straight line solution curve? .3.28.3.28 itself was generated with the MATLAB program dfield [John Polking and David Arnold. Ordinary Differential Equations Using MATLAB. INVESTIGATION A: Plot a slope field and typical solution curves for the differential equation dyjdx = sin(x. 5}] .28. for instance.. MATLAB dfield setup to construct slope field and solution curves for y' = sin(x . You might warm up by generating the slope fields and some solution curves for Problems 1 through 10 in this section. (a) Substitute y = ax + b in the differential equation to determine what the coefficients a and b must be in order to get a solution. When a differential equation is entered in the dfield setup menu (Fig. 5}.edu/iode)... 1.y). NJ: Prentice Hall. Can you see that no value of C yields the linear solution y = x .1. 1. 5. 2nd edition. {x. 1.. Figure 1.uiuc. (b) A computer algebra system gives the general solution y (x) = x . Computergenerated slope field and solution curves for the differential equation y' = sin(x ..3.C c) • Plot this solution with selected values of the constant C to compare the resulting solution curves with those indicated in Fig. With 10 ~ x ~ 10.28. Another freely available and userfriendly MATLABbased ODE package with impressive graphical capabilities is lode (www. 1999] that is freely available for educational use (math.1 ( x .
(b) Then generate a slope field for this differential equation. . dx n (a) First investigate (as in part (a) oflnvestigation A) the possibility of straight line solutions.ny). State your inference as plainly as you can. Can you make a connection between this symbolic solution and your graphically generated solution curves (straight lines or otherwise)? The firstorder differential equation dy dx = H(x. y) (1) is called separable provided that H (x . try to predict (perhaps in terms of y 0 ) how y(x) behaves as x + +oo.32 Chapter 1 FirstOrder Differential Equations INVESTIGATION B: For your own personal investigation. In this case the variables x and y can be separatedisolated on opposite sides of an equationby writing informally the equation f(y) dy = g(x ) dx . which we understand to be concise notation for the differential equation dy f( y ) dx = g(x ). and consider the differential equation dy 1 = cos(x. (c) A computer algebra system gives the general solution y(x) = ~ [ x + 2 tan 1 (x ~C) J. plus a sufficient number of nonlinear solution curves that you can formulate a conjecture about what happens to y(x) as x + +oo. (2) It is easy to solve this special type of differential equation simply by integrating both sides with respect to x: I f( y (x)) ~~ dx = I g(x ) d x + C. y) can be written as the product of a function of x and a function of y: dy dx = g(x)h(y) = . with the viewing window chosen so that you can picture some of these straight lines. Given the initial value y (0) = y0 . let n be the smallest digit in your student ID number that is greater than 1. f(y) g(x ) where h (y) = 1/f (y )..
(2) and (3) are equivalent. • . Hence ld: = l(6x)dx. (0. This is the upper emphasized solution curve shown in Fig.4. we divide both sides of the differential equation by y and multiply each side by dx to get dy y = 6xdx. The condition y(O) = 7 yields A = 7. y(O) = 7. (3) All that is required is that the antiderivatives I 1 f(y) dy G(x) =I g(x) dx can be found. (4) because two functions have the same derivative on an interval if and only if they differ by a constant on that interval. Solution Informally.3x2' = ec. I F(y) = f(y)dy =I and g(x)dx +C. 7) ln IYI = 3x 2 +C.4. Example 1 Solve the initial value problem dy dx = 6xy.1. so the desired solution is y(x) X FIGURE 1. We see from the initial condition y(O) = 7 that y(x) is positive near x may delete the absolute value symbols: lny = 3x 2 8 I I I I I I 6 = 0.1. 4 and hence where A 6 8 ~2~~L~~~~2 = e3x2+C = e3x2ec = Ae. 1.1.4 Separable Equations and Applications 33 equivalently. To see that Eqs. note the following consequence of the chain rule: Dx[F(y(x))] =F (y(x))y (x) 1 = dy f(y)dx = g(x) = Dx[G(x)].6xy in Example 1. so we 4 2 2 "' 0 f*~~E1f3:~~*l + C. which in tum is equivalent to F(y(x)) = G(x) + C. Slope field and solution curves for y' = .
To solve the initial value problem lli·l. . dx dy 4. we see that each solution curve y = y(x) lies on a contour (or level) curve where the function FIGURE 1.34 Chapter 1 FirstOrder Differential Equations Remark: Suppose. 1.. 7).5y is constant.. (6) explicitly in terms of x. The corresponding solution curve y = y (x) lies on the upper contour curve in Fig.5y = 4x .. • 2 4 61''.. instead. We should therefore replace IYI with y in the integrated equation In IYI = 3x 2 + C to obtain ln( . and hence  • (5) ~ Solve the differential equation dy 4. Although it is not convenient to solve Eq. 1. (6) gives an implicit solution of the differential equation in (5). 5) but not on the interval ( . then we call (4) an implicit solution of the differential equation in (2). . The initial condition then yields C = ln 4.2x dx.4.3.y) = . As Example 2 illustrates. (6) and get C solution y(x) is defined implicitly by the equation = 9.______J 6 4 2 0 2 4 6 8 X ~~. (7) we substitute x = 1 andy = 3 in Eq.5) in Example 2.:_ l. H(x. 5) dy = I (4.· :._______.2x)j(3y 2 .3x 2 y(x) = 4e3x2.3y 2 . . that the initial condition in Example 1 had been y(O) = 4.:.:. . ..2x ) dx . so In(. we get I (3i4 _.5 ' Solution When we separate the variables and integrate both sides. it appears from the figure that this particular solution is defined on the interval ( 1.l. Example 2 + In 4. Thus Eq.. it may or may not be possible or practical to solve Eq. Thus the desired particular (8) l  5y = 4x  x2 + 9. (6) This equation is not readily solved for y as an explicit function of x . If not..2x y(l) = 3..__'_. This is the lower emphasized solution curve in Fig._:_ (1. y) = x 2  4x + l.5 . Slope field and solution curves for y' = (4.: ~.. Figure 1..:.4. Then it would follow that y(x) is negative near x = 0.. 3) ...y) = .x 2 +C.4..·_:. 3).4....:.2. .2 shows several of these contour curves. • . (4) explicitly for y in terms of x . ..3x 2 +C. ..__  ..ltfU = 3y 2 .1. Because the graph of a differentiable solution cannot have a vertical tangent line anywhere... ..2the one passing through ( 1.
5)..J4 . and Singular Solutions The equation K (x. differentiation of x 2 + y 2 = 4 yields dy x + y dx = 0. 1. ~ 0 5 10 15 20 Figure 1. It appears that this particular solution through (1.J4 .= x(y .2x) dy dx (9) .x 2 and y = .2.x 2 of the differential equation x + y y' = 0. with the initial condition y(1) = 2 we get the lower contour curve in Fig.4. satisfies the initial condition y (0) = 2 (Fig.8552 of the particular solution in Example 3. With a graphing calculator we can solve for the single real root y :::::::: 2.4 Separable Equations and Applications 20 15 35 Remark I: When a specific value of x is substituted in Eq. For instance. then we get the new implicit solution (y. if we multiply the implicit solution x 2 + y 2 . 4) but not on the interval ( 1. Remark 2: If the initial condition in (7) is replaced with the condition y(l) = 0. consider the differential equation (y .4 = 0 by the factor (y. (8) for xvalues at desired increments from x = 1 to x = 5 (for instance). y) = 0 may or may not satisfy a given initial condition. 1.4.2. x = 4 yields the equation 10 5 f (y) = l  5y . But note that a particular solution y = y (x) of K (x. Such a table of values serves effectively as a "numerical solution" of the initial value problem in (7). • Implicit. 1.2x).x 2 FIGURE 1.3.4). 0) is defined on the interval (0. then the resulting particular solution of the differential equation in (5) lies on the lower "half" of the oval contour curve in Fig.9.9 = 0.2x ) y .4. Remark 2: Similarly.1. Thus the initial condition can determine whether a particular solution is defined on the whole real line or only on some bounded interval.5y. General. For instance. Slope field and solution curves for y' = xjy.x 2 and y(x) = )4.4) = 0 that yields (or "contains") not only the previously noted explicit solutions y +.8552. On the other hand. so x 2 + y 2 = 4 is an implicit solution of the differential equation x only the first of the two explicit solutions X + yy' = 0.4. Remark I: You should not assume that every possible algebraic solution y = y (x) of an implicit solution satisfies the same differential equation. but also the additional function y = 2x that does not satisfy this differential equation. For instance. we can attempt to solve numerically for y. But y(x) = +)4. y) = 0 is commonly called an implicit solution of a differential equation if it is satisfied (on some interval) by some solution y = y(x) of the differential equation.4.2x)(x 2 +i .4. solutions of a given differential equation can be either gained or lost when it is multiplied or divided by an algebraic factor. For example. 4 2 0 y 6 2 4 6 FIGURE 1. This yields the value y(4) :::::::: 2. With a computer algebra system one can readily calculate a table of values of the ysolutions of Eq.3 shows the graph of f. Graph of f(y) = y 3 .4. (8). This particular solution is defined for all x.
The argument preceding Example 1 actually suffices to show that every particular solution of the differential equation f(y)y' = g(x) in (2) satisfies the equation F(y(x)) = G(x) + C in (4). By contrast. 1. whereas negative values yield those that dip below it.4. any particular choice of a specific value for C yields a single particular solution of the equation. then we get the previously discussed differential equation dy ydx=x.1) 213 1 dy = J 2x dx . we "gain" this new solution when we multiply Eq. Positive values of the arbitrary constant C give the solution curves in Fig. (10) of which y = 2x is not a solution.5. • A solution of a differential equation that contains an "arbitrary constant" (like the constant C in the solution of Examples 1 and 2) is commonly called a general solution of the differential equation.6 that lie above the line y = 1.2x).6. but the general solution of (2). dx 15 10 5 0 X 5 10 15 FIGURE 1. but the possibility of loss or gain of such "extraneous solutions" should be kept in mind. • = . and to observe that the constantvalued function y(x) = 0 is a singular solution that cannot be obtained from the general solution by any choice of the arbitrary constant C . Thus we "lose" the solution y = 2x of Eq.1)1/3 = xz + C. Example 4 Solution Separation of variables gives f 3(y.= 6x(y. it is appropriate to call (4) not merely a general solution of (2). In Section 1.1) 3 both satisfy the initial condition y(l) = 1. (y . Note that the two different solutions y(x) 1 and y(x) = 1 + (x 2 . y (x) = 1 + (x 2 + C) 3 • 2 1 0 X 2 FIGURE 1.C) 2 and the singular solution curve y = 0 of the differential equation (y')2 = 4y. In Problem 30 we ask you to show that the general solution of the differential equation (y') 2 = 4y yields the family of parabolas y = (x .5 we shall see that every particular solution of a linear firstorder differential equation is contained in its general solution. These exceptional solutions are frequently called singular solutions. (9) upon its division by the factor (y . y) = 6x (y .C) 2 illustrated in Fig. The value C = 0 gives the solution y (x ) = 1 + x 6 . or x + y dx dy = 0. General and singular solution curves for y' = 6x (y .2x).4.4. The general solution curves y = (x . (10) by (y .4. alternatively.1)213 is not differentiable. 1. Find all solutions of the differential equation dy . Indeed. Such elementary algebraic operations to simplify a given differential equation before attempting to solve it are common in practice.5.1) 213 . but no value of C gives the singular solution y(x) = 1 that was lost when the variables were separated.2x). the whole singular solution curve y = 1 consists of points where the solution is not unique and where the function f (x.1)2/ 3. But if we divide both sides by the common factor (y. Consequently.36 Chapter 1 FirstOrder Differential Equations having the obvious solution y = 2x. it is common for a nonlinear firstorder differential equation to have both a general solution involving an arbitrary constant C and one or several particular solutions that cannot be obtained by selecting a value for C.
10.= l i m . we use Eq. and with f3 = 0.t+0 b. (Note that 10% annual interest means that r = 0.t (12) where k = f3. It has been observed that a constant fraction of those radioactive atoms will spontaneously decay (into atoms of another element or into another isotope of the same element) during each unit of time. This same ratio permeates all life.P ~ ({3.8)P(t) b. Then. This proportion remains constant because the fraction of 14 C in the atmosphere remains almost constant. and living matter is continuously taking up carbon from the air or is consuming other living matter containing the same constant ratio of 14 C atoms to ordinary 12C atoms.4 Separable Equations and Applications 37 Natural Growth and Decay The differential equation dx =kx dt (k a constant) (11) serves as a mathematical model for a remarkably wide range of natural phenomenaany involving a quantity whose time rate of change is proportional to its current size. because organic processes seem to make no distinction between the two isotopes.t deaths occur.A = r A(t) b. and therefore dP b. with k > 0 in place of 8. POPULATION GROWTH: Suppose that P(t) is the number of individuals in a population (of humans. We thus get the differential equation dN =kN. approximately f3P(t) b. dt fl.t. To write a model for N(t). Consequently. or insects.8.t + 0 b. Here are some examples.=rA. Let A (t) be the number of dollars in a savings account at time t (in years). The key to the method of radiocarbon dating is that a constant proportion of the carbon atoms in any living creature is made up of the radioactive isotope 14 C of carbon.t births and 8P(t) b. dt fl.) Continuous compounding means that during a short time interval b.P =lim =kP.A . so that COMPOUND INTEREST: dA b. dt (14) The value of k depends on the particular radioactive isotope.1. during a short time interval b. (12) with N in place of P.t.t. so the change in P(t) is given approximately by b. . and suppose that the interest is compounded continuously at an annual interest rater. the amount of interest added to the account is approximately b. the sample behaves exactly like a population with a constant death rate and no births.t (13) RADIOACTIVE DECAY: Consider a sample of material that contains N(t) atoms of a certain radioactive isotope at time t.t. or bacteria) having constant birth and death rates f3 and 8 (in births or deaths per individual per unit of time).
38 Chapter 1 FirstOrder Differential Equations The ratio of 14C to normal carbon remains constant in the atmosphere because.4.8. (Matters are not as simple as we have made them appear. (16) Because of the presence of the natural exponential function in its solution. 7 shows a typical graph of x (t) in the case k > 0.) In many cases the amount A(t) of a certain drug in the bloodstream. measured by the excess over the natural level of the drug. Then we solve for x : Because C is a constant. will decline at a rate proportional to the current excess amount. For 14 C. it is known that k ~ 0. the differential equation dx (17) =kx dt is often called the exponential or natural growth equation. although 14C is radioactive and slowly decays. That is. Over the long history of the planet. and consequently the ratio of 14C to normal carbon begins to drop. so the particular solution of Eq. the cosmic ray levels apparently have not been constant. researchers in this area have compiled tables of correction factors to enhance the accuracy of this process. The Natural Growth Equation The prototype differential equation dxjdt = kx with x(t) > 0 and k a constant (either negative or positive) is readily solved by separating the variables and integrating: lnx = kt +C. the amount of time elapsed since the death of the organism can be estimated.4. DRUG ELIMINATION: dA = AA. it ceases its metabolism of carbon and the process of radioactive decay begins to deplete its 14C content. Figure 1. There is no replenishment of this 14C. the amount is continuously replenished through the conversion of 14 N (ordinary nitrogen) to 14C by cosmic rays bombarding the upper atmosphere. In addition.0001216 if tis measured in years. this decay and replenishment process has come into nearly steady state. . the case k < 0 is illustrated in Fig. (11) with the initial condition x(O) = x 0 is simply x (t) = xoekt. It is also clear that A = x (0) = x 0 . so is A = ec . dt (15) where A > 0. In applying the technique of radiocarbon dating. 1. By using independent methods of dating samples. extreme care must be taken to avoid contaminating the sample with organic matter or even with ordinary fresh air. so the ratio of 14C in the atmosphere has varied over the past centuries. The parameter A is called the elimination constant of the drug. Of course. when a living organism dies. By measuring this ratio. For such purposes it is necessary to measure the decay constant k.
~ ~ 0. Natural decay. This value of k gives the world population function P(t) = 6eO. when t = In 10 0. P(O) 6 t = 0 we now Thus the world population was growing at the rate of about 1. FIGURE 1.07743 k = . and thus in the year 2177.1. so Po = 6. and was then increasing at the rate of about 212 thousand persons each day.29% annually in 1999.OI 29t .0129.000212 billion.0129 ~ 178.8. or 0.25) ~ 0. • . Natural growth. The fact that P is increasing by 212. From the natural growth equation P' = k P with obtain P' (O) 0. that is.000212)(365. persons per day at timet = 0 means that P' (O) = (0.4. (b) With t = 51 we obtain the prediction 0 ·01 29)(S l) ~ 11.07743 Solution billion per year.4 Separable Equations and Applications X X 39 x =xo ekt (k > 0) FIGURE 1.gov. the world's total population reached 6 billion persons in mid1999. We take t = 0 to correspond to (mid) 1999.OJ 29t .58 (billion) P(51) = 6e< for the world population in mid2050 (so the population will almost have doubled in the just over a halfcentury since 1999). we want to answer these questions: (a) What is the annual growth rate k? (b) What will be the world population at the middle of the 21st century? (c) How long will it take the world population to increase tenfoldthereby reaching the 60 billion that some demographers believe to be the maximum for which the planet can provide adequate food supplies? (a) We measure the world population P(t) in billions and measure time in years.census. Assuming that natural population growth at this rate continues. (c) The world population should reach 60 billion when 60 = 6 eo.7. Example 5 According to data listed at www.000.4.
1 billion.63)N0 now. What is the age of the sample? We take t = 0 as the time of the death of the tree from which the Stonehenge charcoal was made and No as the number of 14C atoms that the Stonehenge sample contained then. because this parameter is more convenient. The decay constant of a radioactive isotope is often specified in terms of another empirical constant.63) 0_ 0001216 ~ 3800 (years).T.40 Chapter 1 FirstOrder Differential Equations Note: Actually. (3) of Section 1. Thus the sample is about 3800 years old. so we solve the equation (0. To find the relationship between k and T . approximately 5700 Example 6 A specimen of charcoal found at Stonehenge turns out to contain 63% as much 14C as a sample of presentday charcoal of equal mass. ·········· . initially at 50°F. the time rate of change of the temperature T(t) of a body immersed in a medium of constant temperature A is proportional to the difference A .T) . or templewhichever it may bedates from 1800 B. We are given that N = (0.M.63)No = N 0 ekt with the value k = 0. • Cooling and Heating According to Newton's law of cooling (Eq. is placed in a 375°F oven at 5:00 P. A simple mathematical model cannot be expected to mirror precisely the complexity of the real world. the rate of growth of the world population is expected to slow somewhat during the next halfcentury. we set t = T and N = 4 in the equation N (t) = N 0 ekt.1 ). and the best current prediction for the 2050 population is "only" 9. the halflife of years. or earlier. After 75 minutes it is found that the temperature T(t) of the roast is 125°F. we find that (18) For example. Thus we find that t = Solution ln(0. so that 4 No = N 0ekr: .. monument. That is. When will the roast be 150°F (medium rare)? . The halflife T of a radioactive isotope is the time required for half of No it to decay. = k(A .C. _ _ ____ _ Example 7 A 4lb roast. When we solve for r. dt (20) It includes the exponential equation as a special case (b = 0) and is also easy to solve by separation of variables. This is an instance of the linear firstorder differential equation with constant coefficients: dx =ax +b.0001216. 14C is r ~ (In 2)/(0. our computations suggest that this observatory. the halflife of the isotope. If it has any connection with the builders of Stonehenge. dt dT (19) where k is a positive constant.0001216).
the total cooking time required.4 Separable Equations and Applications Solution 41 We take timet in minutes. We have T(t) < A = 375.T dT=fkdt· ' ln(375. Denote by y(t) the depth of water in the tank at timet. As a consequence ofEq. from which water is leaking.oo3s)t for t = [ln(225j325)]/(0. and T(75) = 125. (22b) This is a statement of Torricelli 's law for a draining tank. under ideal conditionsthat the velocity of water exiting through the hole is v = fiiY. (21) which is the velocity a drop of water would acquire in falling freely from the surface of the water to the hole (see Problem 35 of Section 1.T) = kt + C. 375. Let A (y) denote the horizontal crosssectional area of the tank at height y. Hence we finally solve the equation 150 = 375 . f 375. One can derive this formula beginning with the assumption that the sum of the kinetic and potential energy of the system remains constant. (21). Substitution of these values in the preceding equation yields k = .0035.M.6 for a small continuous stream of water). We also know that T = 125 when t = 75.J'liY.T). so T(t) = 375. dt (22a) equivalently. we have dV = av = afiiY.. Because the roast was placed in the oven at 5:00P. applied to a thin horizontal .325e(o. it should be removed at about 6:45P. v = c.fii. = kJy dt dV where k= a. Under real conditions.T = Bek1 • Now T(O) =50 implies that B = 325. For simplicity we take c = 1 in the following discussion. T(O) = 50.71 5 ln (~~~) >=:::: 0. • Torricelli's Law Suppose that a water tank has a hole with area a at its bottom.1.325ekr. Then. Hence 1 dT dt = k(375.2).M. with t = 0 corresponding to 5:00P. and by V(t) the volume of water in the tank then. taking into account the constriction of a water jet from an orifice. We also assume (somewhat unrealistically) that at any instant the temperature T(t) of the roast is uniform throughout.0035) >=:::: 105 (min). It is plausibleand true. where cis an empirical constant between 0 and 1 (usually about 0.M.
dt dy dt dt  (23) From Eqs. so FIGURE 1.. Draining a hemispherical tank.(4. . (24) becomes dy I )2 I n(8y._l. The tank is empty when y = 0.y) 2 ] = n(8y. • . 45/ 2  5  448 15.t +c.i)..4. (8yl f2 .l/2) dy = 3 5  f 7i 72 dt. So it takes slightly less than 36 min for the tank to drain. thus when t = 72 · 448 15 ~ 2150 (s) · ' that is. 4 3/ 2 3 ~ . (22) and (23) we finally obtain dy ~ A(y).= n ( 24 y 2 · 32y.= a. With g = 32 ftjs 2 .. C  16 . we see that A(y) Solution = nr 2 = n: [16.42 Chapter 1 FirstOrder Differential Equations slice of water at height y with area A (y) and thickness dy. Eq. A(y) and The fundamental theorem of calculus therefore implies that dVjdy hence that dV dV dy dy = . • Example 8 A hemispherical bowl has top radius 4 ft and at time t = 0 is full of water.2gy = kyly.4.. dt f Now y(O) = 4. 1..~y5!2 . about 35 min 50 s. How long will it take for all the water to drain from the tank? From the right triangle in Fig. At that moment a circular hole with diameter 1 in. l&y3!2 . the integral calculus method of cross sections gives V(y) =loy A(y) dy.9.9.y 2 ) . is opened in the bottom of the tank. dt (24) an alternative form of Torricelli 's law.=A(y).
16' dy 22. (x 2 + 1)(tan y)y' = x dx x 2 (2y 3 . (c) Inspect a sketch of typical solution curves to determine the points (a.x2 ) dy = 2y dx 11. . . (Radiocarbon dating) Carbon taken from a purported relic of the time of Christ contained 4.3 dy 23. (Continuously compounded interest) Suppose that you discover in your attic an overdue library book on which your grandfather owed a fine of 30 cents 100 years ago. dy 25. 33. 1. (b) a unique solution. y(O) = 2e y(O) = 1 dy 20. a couple deposited $5000 in an account that pays 8% interest compounded continuously. Assume that its population will continue to grow exponentially at a constant rate. y(l) = 1 dy 24. (1 + 2xy 2 = dy 0 + x) dx = 4y 5. The dog is anesthetized when its bloodstream contains at least 45 milligrams (mg) of sodium pentobarbital per kilogram of the dog's body = 3fo dy 8. Discuss the difference between the differential equations (dyjdxf = 4y and dyjdx = 2JY.= (64xy) 113 dx !f=Y2 dy 6. = 6e 2xy. b) in the plane for which the initial value problemy'= yJYT=l. Compute the approximate age of the relic. Carbon extracted from a presentday specimen of the same substance contained 5. b) in the plane for which the initial value problem (y') 2 = 4y. explicit if convenient) of the differential equations in Problems 1 through 18./X . Find a general solution and any singular solutions of the differential equation dymyslashdx = yJY2=!". dy (x. ~~ y(O) =0 dy 28. dy 19. dx = ysmx dy 2.jX 1 + . Then determine the points (a.0 x 10 10 atoms of 14 C per gram. 2. How much will the account contain on the child's eighteenth birthday? 38. y(a) = b has (a) no solution. 3.) 18. y' = xy 3 dy 13./X d y = dx dy 7. y(l) = . Primes denote derivatives with respect to x. (b) infinitely many solutions that are defined for all x.= 2x secy dx 9.+2xy = 0 dx dy . dx 30. 2. How old is the skull? 36. Determine the points (a. y(4) = n/4 dx 29. dy 1. x 2 y' = 1 . (Drug elimination) Suppose that sodium pentobarbital is used to anesthetize a dog. . 31. b) in the plane for which the initial value problemy'= 2JY. . (1 . (c) infinitely many solutions. y 3 .= cos2 y. b) for which the initial value problem y' = y 2. the number of bacteria increased sixfold in 10 h.y = 2x2y. only finitely many solutions. y(a) = b has (a) no solution. What is your opinion as to its authenticity? 37. (Radiocarbon dating) Carbon extracted from an ancient skull contained only onesixth as much 14 C as carbon extracted from presentday bone. What population can its city planners expect in the year 2000? 34. (1 + x)2 dy 12. dy dx + 1) cosx = = 1 + . If an overdue fine grows exponentially at a 5% annual rate compounded continuously. (Continuously compounded interest) Upon the birth of their first child. y(a) = b has (a) no solution. (c) infinitely many solutions.1)y5 16.y) 17.= (y 4 dx 10. yy' 14.x 2y 2 Find explicit particular solutions of the initial value problems in Problems 19 through 28. dx = yex. dx 27. Solve the differential equation (dyjdx) 2 = 4y to verify the general solution curves and singular solution curve that are illustrated in Fig. x dx ..4. how much would you have to pay if you returned the book today? 39. y(4n) = 1n y(l) = 1 y(l)=1 = 2xy2 + 3x2y2. 2ydy = y(5) = 2 dx Jx2 . (tanx) dx = y. dx + 1 = 2y.. (a) Find a general solution of the differential equation dyjdx = y 2 • (b) Find a singular solution that is not included in the general solution. (Population growth) A certain city had a population of 25000 in 1960 and a population of 30000 in 1970.JY = (1 dx x(y 2 + 1) + y)2 15.= 3x 2 (y 2 + 1). . dx 4.x 2 + y 2 . (Population growth) In a certain culture of bacteria. y' = 1+x+y+xy (Suggestion: Factor the righthand side. The interest payments are allowed to accumulate.4 Separable Equations and Applications 43 Find general solutions (implicit if necessary.5.1. y(a) = b has a unique solution. dx X 21. (c) on some neighborhood of the point x = a. Do they have the same solution curves? Why or why not? Determine the points (a.6 x 10 10 atoms of 14 C per gram. dy 26. dx =4x3y y. How long did it take for the population to double? 35. 32. (b) a unique solution.
44 Chapter 1 FirstOrder Differential Equations weight. (a) At what depth is the intensity half the intensity / 0 at the surface (where x = 0)? (b) What is the intensity at a depth of 10 m (as a fraction of / 0 )? (c) At what depth will the intensity be 1% of that at the surface? 46. (b) Without prior conditioning.51 x 109 years for 238 U and 7. of mercury. there were equal amounts of the two uranium isotopes 235 U and 238 U at the creation of the universe in the "big bang. Suppose that the temperature of the buttermilk has dropped to l5 °C after 20 min. (a) If the initial amount is 10 pu (pollutant units). What is the age of the rock. There are now about 3300 different human "language families" in the whole world. and 5 months later it is still 10 su.will it be until A = 1 su. A cake is removed from an oven at 210oF and left to cool at room temperature. it initially contains water to a depth of 9 ft.5 years. how long will this take? 40.000 ft and again at 30. An accident at a nuclear power plant has left the surrounding area polluted with radioactive material that decays naturally. A certain moon rock was found to contain equal numbers of potassium and argon atoms. After 1 h the depth of the water has dropped to 4 ft. A pitcher of buttermilk initially at 25°C is to be cooled by setting it on the front porch.10 x 108 years for 235 U.2) p. How . few people can survive when the pressure drops to less than 15 in." Assume (as in Problem 52) that the number of these language families has been multiplied by 1. (b) What amount of radioactive material will remain after 8 months? (c) How longtotal number of months or fraction thereof.000 people had heard this rumor. with a halflife of 5 h. write a formula for A(t) giving the amount (in pu) present after t years.kA (k > 0)." At present there are 137. The halflife of radioactive cobalt is 5.92. Suppose that a mineral body formed in an ancient cataclysmperhaps the formation of the earth itselforiginally contained the uranium isotope 238 U (which has a halflife of 4.4)/.5 language families every 6 thousand years. How long does it take for all the water to drain from the tank? 55. The initial amount of radioactive material present is 15 su (safe units). About how long ago was the single original human language spoken? 53. How long will it be until half the population of the city has heard the rumor? 51. How high is that? 47. The intensity I of light at a depth of x meters below the surface of a lake satisfies the differential equation dljdx = (1.51 x 109 years) but no lead. measured from the time it contained only potassium? 43. p(O) = 29. The barometric pressure p (in inches of mercury) at an altitude x miles above sea level satisfies the initial value problem dpjdx = (0. When will it be at 5°C? 44.27 years. If today the ratio of 238 U atoms to lead atoms in the mineral body is 0. Assume that the rate of increase of the number who have heard the rumor is proportional to the number who have not yet heard it. If 25% of the sugar dissolves after 1 min.000 ft. (a) Calculate the barometric pressure at 10. how long does it take for half of the sugar to dissolve? 45. when did the cataclysm occur? 42. (b) What will be the amount (in pu) of pollutants present in the valley atmosphere after 5 years? (c) If it will be dangerous to stay in the valley when the amount of pollutants reaches 100 pu. Using the halflives 4. About when did the ancestors of today's Native Americans arrive? 54. and that a language family develops into 1. A tank is shaped like a vertical cylinder. and a bottom plug is removed at timet = 0 (hours). calculate the age of the universe.7 atoms of 238 U for each atom of 235 U. Within a week. the end product of the radioactive decay of 238 U. Thousands of years ago ancestors of the Native Americans crossed the Bering Strait from Asia and entered the western hemisphere.) 41.28 x 109 years) and that one of every nine potassium atom disintegrations yields an argon atom. There are now 150 Native American language families in the western hemisphere. 10. Suppose also that sodium pentobarbital is eliminated exponentially from the dog's bloodstream. A certain piece of dubious information about phenylethylamine in the drinking water began to spread one day in a city with a population of 100. Suppose that the tank of Problem 48 has a radius of 3 ft and that its bottom hole is circular with radius 1 in. 49. According to one cosmological theory. Since then. What single dose should be administered in order to anesthetize a 50kg dog for 1 h? 48. the amount A that remains undissolved after t minutes satisfies 'the differential equation dAjdt = . where the temperature is ooc. When will it be 100° F? 50. which is 70°F. Suppose that a nuclear accident has left the level of cobalt radiation in a certain region at 100 times the level acceptable for human habitation. When sugar is dissolved in water. Assume that all these are derived from a single original language. so it is safe for people to return to the area? 52.5 every 6000 years.9. The single language that the original Native Americans spoke has since split into many Indian "language families. (a) Write a formula giving the amount A(t) of radioactive material (in su) remaining after t months. they have fanned out across North and South America. How long will it be until the region is again habitable? (Ignore the probable presence of other radioactive isotopes. The amount A(t) of atmospheric pollutants in a certain mountain valley grows naturally and is tripling every 7. After 30 min the temperature of the cake is 140° F. Assume that all the argon is the result of radioactive decay of potassium (its halflife is about 1.000.
show that dx 1 k=dt t where k is a constant.4. it had moved an additional 3 miles. What time did it start snowing? This is a more difficult snowplow problem because now a transcendental equation must be solved numerically to find the value of k.M. John Bernoulli proposed this problem as a public challenge. it had traveled 2 miles. how long wi11 it take for the liquid to drain completely? 61. (The clepsydra.the curve of minimal descent time is an . When will the tank be empty? 57. In June of 1696. (b) What time did it start snowing? (Answer: 6 A. received Bernoulli's challenge on January 29. A plug at the bottom is removed at 12 noon. (a) Find the depth y(t) of water remaining after t hours.M. A snowplow sets off at 7 A./h)? 65. What should be this curve. What was the time of death? 66.M.M.J28Y" (taking constriction into account) to determine when the tank will be empty. At 12 noon the temperature of the body is 80°F and at 1 P.4.M. it had traveled 4 miles and that by 9 A. but it took two more hours (until 10 A. and what should be the radius of the circular bottom hole. After 1 h the water in the tank is 9 ft deep.1 0. A cylindrical tank with length 5 ft and radius 3 ft is situated with its axis horizontal. How long will be required for all the gasoline to drain from the tank? 62. Suppose that a cylindrical tank initially containing V0 gallons of water drains (through a bottom hole) in T minutes.M. shaped like the surface obtained by revolving the curve y = f (x) around the yaxis. what is the radius of the circular hole in the bottom? 60.M.M. 58. (Answer: 4:27A. in order that the water level wi11 fall at the constant rate of 4 inches per hour (in.M. If a circular bottom hole with a radius of 1 in. (b) When will the tank be empty? (c) Ifthe initial radius of the top surface of the water is 2 ft. or water clock) A 12h water clock is to be designed with the dimensions shown in Fig. Assuming that the snowplow clears snow from the road at a constant rate (in cubic feet per hour. circular bottom hole with radius 1 in. The very next day he communicated his own solution. Just before midday the body of an apparent homicide victim is found in a room that is kept at a constant temperature of 70°F. when will the tank be empty? 63.M. when a circular plug in the bottom of the tank is removed. 1. A water tank has the shape obtained by revolving the parabola x 2 = by around the yaxis. The clepsydra.6)rrr 2 . the depth of the water is 1 ft. At 7 A. At timet = 0 the bottom plug (at the vertex) of a full conical water tank 16 ft high is removed. it is 75°F.) 67. the bottom hole is opened and at 1:30 P.10.6°F and that it has cooled in accord with Newton's law. At 1 P. the depth of water in the tank is 2ft. The brachistochrone problem asks what shape the wire should be in order to minimize the bead's time of descent from P to Q. At 1 P. (a) Lett = 0 when it began to snow and let x denote the distance traveled by the snowplow at time t. By 8 A.11 shows a bead sliding down a frictionless wire from point P to point Q. a snowplow set off to clear a road. Early one morning it began to snow at a constant rate. Suppose that an initially full hemispherical water tank of radius 1 m has its flat side as its bottom.M. say). If this bottom hole is opened at 1 P. except that the radius r of its circular bottom hole is now unknown.£ T minutes is V = V0 [1 (tiT) f .M. Suppose now that by 8 A. (b) What is the radius of the bottom hole? 64. Consider the initially full hemispherical water tank of Example 8. the depth of the water is 6 ft.1. is opened. The water depth is 4 ft at 12 noon.M.) 68.. At 1 P. with a 6month deadline (later extended to Easter 1697 at George Leibniz's request). as in Problem 66.) for the snowplow to go an additional 2 miles. Use Torricelli's law to show that the volume of water in the tank after t . A water tank has the shape obtained by revolving the curve y = x 413 around the yaxis. Assume that the temperature of the body at the time of death was 98. It has a bottom hole of radius 1 em.M. is opened and the tank is initially half full of xylene. (a) Use Torricelli's law in the form dV/dt = (0. A spherical tank of radius 4 ft is full of gasoline when a y = f(x) or X FIGURE 1. Isaac Newton.4 Separable Equations and Applications long will it take the water (initially 9 ft deep) to drain completely? 45 56. 1697. Figure 1.4. When will the tank be empty? 59. then retired from academic life and serving as Warden of the Mint in London. when the depth of water in the tank is 12ft.
. Principles of physics can be used to show that the shape y = y (x) of the hanging cable satisfies the differential equation where the constant a = TIp is the ratio of the cable's tension T at its lowest point x = 0 (where y' (0) = 0 ) and its (constant) linear density p. d vjdx = d 2 yjdx 2 in this secondorder differential equation.4. H) where a is an appropriate positive constant.11. y = a(l cos 2t) (iii) Yo X for which t = y = 0 when x = 0.· . which . 7th edition (Upper Saddle River.] . y (L. For a modem derivation of this result.4 we saw how to solve a separable differential equation by integrating after multiplying both sides by an appropriate factor.= 2xy (y > 0).4. dy = 4a sin t cost dt in (ii) to derive the solution x = a(2t . y dx that is. (i) the differential equation Solve this differential equation for y'(x) = v(x) sinh(x/a). Then integrate to get the shape function y (x) = a cosh ( ~) + C dy= dx Vy ~ (ii) of the hanging cable. the substitution of e = 2a in (iii) yields the standard parametric equations X = a(e sin e). This curve is called a catenary. H) at equal heights located symmetrically on either side of the xaxis (Fig.= 2x. from the Latin word for chain.J2iY is the bead's velocity when it has descended a distance y vertically (from KE = ~mv 2 mgy = PE).12. to solve the equation dy (1) . Finally.12). For instance. all that remains are two simple integrations. we get the firstorder equation dv a= dx Q v 1 + v2 • ~ FIGURE 1. 1. Suppose a uniform flexible cable is suspended between two points (±L.4. The catenary.= constant.4 of Edwards and Penney. NJ: Prentice Hall. [See Example 5 in Section 9.sin 2t) . (a) First derive from Eq. dx we multiply both sides by the factor 1/y to get 1 dy . If we substitute v = dymyslashdx . suppose the bead starts from rest at the origin P and let y = y(x) be the equation of the desired curve in a coordinate system with the yaxis pointing downward.46 Chapter 1 FirstOrder Differential Equations arc of an inverted cycloidto the Royal Society of London. A bead sliding down a wirethe brachistochrone problem. y = a(l COS e) FIGURE 1. Calculus: Early Transcendentals. p 69. Dx (ln y) = Dx (x). 2 (2) Because each side of the equation in (2) is recognizable as a derivative (with respect to the independent variable x ). 2008). (b) Substitute y = 2a sin2 t. E Linear FirstOrder Equations In Section 1. v sin a (i) where a denotes the angle of deflection (from the vertical) of the tangent line to the curveso cot a = y' (x) (why?)and v = . Then a mechanical analogue of Snell's law in optics implies that of the cycloid that is generated by a point on the rim of a circular wheel of radius a as it rolls along the xaxis.
+ P(x)y = dy Q(x) (3) on an interval on which the coefficient functions P(x) and Q(x) are continuous. (6) This formula should not be memorized. Next. (3) with the coefficient functions P(x) and Q(x) displayed explicitly. 2. . Begin by calculating the integrating factor p (x) = ef P(x ) dx. the function p (y) = 1/y is called an integrating factor for the original equation in (1). the lefthand side is the derivative of the product y(x) · ef P(x) dx. For this reason. ef P(x)dx J = Q(x)ef P(x)dx. there is a standard technique for solving the linear firstorder equation dx . METHOD: SOLUTION OF FIRSTORDER EQUATIONS 1. = Q(x)ef P(x)dx. in order to solve an equation that can be written in the form in Eq. solving for y. 3. you should attempt to carry out the following steps. With the aid of the appropriate integrating factor. An integrating factor for a differential equation is a function p (x. y) such that the multiplication of each side of the differential equation by p(x. (3) by the integrating factor p(x) = ef P(x)dx. We multiply each side in Eq. recognize the lefthand side of the resulting equation as the derivative of a product: D x [p(x)y(x)] = p(x) Q(x).5 Linear FirstOrder Equations 47 yield ln y = x 2 + C. (5) is equivalent to Dx [y(x). Then multiply both sides of the differential equation by p (x). y) yields an equation in which each side is recognizable as a derivative. That is. we obtain the general solution of the linear firstorder equation in (3): y(x) = e. In a specific problem it generally is simpler to use the method by which we developed the formula.J P(x)dx [! ( Q(x)ef P(x) dx) dx + C J. Finally. Integration of both sides of this equation gives y(x)ef P(x)dx = J( Q(x)ef P(x)dx ) dx +C. (4) (5) The result is ef P(x)dx dy dx + P(x)ef P(x)dx y Because Dx [! P(x) dx J= P(x).1. so Eq.
Remark 2: You need not supply explicitly a constant of integration when you find the integrating factor p(x) . Example 1 Solve the initial value problem dy II .!e . 8 so the integrating factor is p(x) = e f(l)d x = e . p(x)y(x) = J p(x)Q(x) dx + C. (4).xy _  J . integrate this equation. • . Finally..x y) = .e y = 8 e .48 Chapter 1 FirstOrder Differential Equations 4. so the desired particular • . so we might as well take K = 0.x/3 dx .y= 8 e . without bothering to add a constant of integration. then solve for y to obtain the general solution of the original differential equation. (7) _ d (e .!. the result is with j P(x) dx +K p(x) = eK+ f P(x)dx = eKef P(x)dx. y(0)=1.x yields e which we recognize as x dy x II 4x/3 dx.4x/3 .x .4x/3 8 32 + C' (8) and multiplication by e x gives the general solution y (x) = C ex . Solution Here we have P(x) = 1 and Q(x) = 11 ex/3. For if we replace j P(x) dx in Eq.!.!e. Substitution of x = 0 and y = 1 now gives C = solution is 1 32 . Remark I: Given an initial condition y(x 0 ) = y0 . you can (as usual) substitute x = xo andy = Yo into the general solution and solve for the value of C yielding the particular solution that satisfies this initial condition. But the constant factor eK does not affect materially the result of multiplying both sides of the differential equation in (3) by p(x). Multiplication of both sides of the given equation by e . 8 dx Hence integration with respect to x gives e.•. You may therefore choose for J P(x) dx any convenient antiderivative of P(x).~~e x/3 .4x/ 3 dx __ 33 e .
+ dy 3x x2 +1 y = :x2 +1 + 1) and Q(x ) = 6xj(x 2 as a firstorder linear equation with P(x) = 3xj(x 2 Multiplication by p(x) = exp ( / x 2 : 1 dx) + 1). we recognize the result 6x dx . (9)..5. This constant = .1.1 shows a slope field and typical solution curves for Eq.5 Linear FirstOrder Equations y= ~ exp(x/3) 32 49 "'1 2 3 4 1 0 2 X 3 4 5 FIGURE 1.~ 1 o~~~~~~~~LL~ and thus Integration then yields . while others grow rapidly in the negative direction..~~ ex/3 for which C = 0 in Eq. then C > 0 in Eq. 3 = exp (~ ln(x 2 + 1)) = (x 2 + 1) 312 yields 7 6 5 4 3 "' 2 ~.~~ on the yaxis grow in the positive direction. Note that some solutions grow rapidly in the positive direction as x increases. Note that. (8).1.~~ grow in the negative direction as x + +oo. • Remark: Figure 1. 1).2. 1.. while solutions that start lower than . The behavior of a given solution curve is determined by its initial condition y(O) = YO· The two types of behavior are separated by the particular solution y(x) = . and hence y(x) + +oo as x + +oo. all other solution curves approach the constant solution curve y(x ) 2 that corresponds to C = 0 in Eq. (10)..=:::=F:!!ii.1. If y0 > . Thus the initial condition y 0 =~~is critical in the sense that solutions that start above . (9). (9) Solution After division of both sides of the equation by x 2 + 1. Remark: Figure 1. so both terms in y(x) are negative and therefore y(x) + oo as x+ +oo. But if y0 < ~~'then C < 0. including the one passing through the point (0. so the term ex eventually dominates the behavior of y(x).5.3/ 2 .2 shows a slope field and typical solution curves for Eq. (8). as x + +oo.. (10) y (x) = 2 + C(x 2 FIGURE 1. • Find a general solution of (x 2 Example 2 + 1) dx + dy 3xy = 6x.1 2 Multiplication of both sides by (x 2 X + 1)3/ 2 gives the general solution + 1). The interpretation of a mathematical model often hinges on finding such a critical condition that separates one kind of behavior of a solution from a different kind of behavior.5. (7). Slope field and solution curves for y' = y + ¥exf3 . Slope field and solution curves for the differential equation in Eq.5.~~..~~ for the solution curve that is dashed in Fig. so Yo = .~.5..
then y(x) is given by the formula in Eq. Conversely. because y(O) = 2 implies that y(x) = 2 for all x (and thus the value of the solution remains forever where it starts). there isas previously noteda unique value of C such that y(xo) = Yo· Consequently. f xo . Note that substitution of y' = 0 in the differential equation (9) yields 3xy = 6x. (11)can be selected "automatically" by writing p(x )=exp(1: P(t)dt) . for which it follows that y'(x) 0. (12) y(x ) =  p (x ) 1 . Remark 2: Theorem 1 tells us that every solution of Eq. Then the anti derivatives J P(x) dx and J( Q(x)ef P(x)dx) dx exist on I. Remark 1: Theorem 1 gives a solution on the entire interval I for a linear differential equation.3. Our derivation of Eq. Finally. in contrast with Theorem 1 of Section 1.as needed to solve the initial value problem in Eq. (6). given by the formula in Eq. Thus a linear firstorder differential equation has no singular solutions. we have proved the following existenceuniqueness theorem. as seems visually obvious in Fig. (6) of the linear firstorder equation y' + P y = Q bears closer examination. • = = = A Closer Look at the Method The preceding derivation of the solution in Eq.2. then the initial value problem dy dx + P(x)y = Q(x).50 Chapter 1 FirstOrder Differential Equations solution can be described as an equilibrium solution of the differential equation. Remark 3: The appropriate value of the constant C in Eq.[Yo + { x p(t)Q(t) dt]. you may verify by direct substitution (Problem 31) that the function y(x) given in Eq. (3)." so by an equilibrium solution of a differential equation is meant a constant solution y(x) c. (3) is included in the general solution given in Eq. (6) for some choice of the constant C. (6) with an appropriate value of C. given a point xo of I and any number y 0 . Hence we see that y (x) 2 is the only equilibrium solution of this differential equation. 1. so it follows that y = 2 if x :j= 0. the word "equilibrium" connotes "unchanging. (6) satisfies Eq. (3) on I. which guarantees only a solution on a possibly smaller interval. y(xo) = y 0 (11) has a unique solution y(x) on I. More generally. Suppose that the coefficient functions P (x) and Q (x) are continuous on the (possibly unbounded) open interval I. (6) shows that if y = y(x) is a solution of Eq.5. (6). THEOREM 1 The Linear FirstOrder Equation If the functions P(x) and Q(x) are continuous on the open interval I containing the point x 0 .
Withx0 = 1 theintegratingfactorin(12) is p(x) = exp (lx ~ dt) = exp(ln x) = x. (6) that guarantees in advance that p(x0 ) = 1 and that y(x0 ) = y0 (as you can verify directly by substituting x = x 0 in Eqs.1..5.5 Linear FirstOrder Equations 51 The indicated limits x 0 and x effect a choice of indefinite integrals in Eq. (14) reduces to y(x) ( 1.when a solution of a differential equation can be expressed in terms of elementary functions. S1(x)S1(1)]..3) I 1 [ Yo+ =X 1x 0 sin t dt t 1 1 sin t dt t 2 3 0 J= [Yo+ 1 . for instanceto find the value y(x) of the solution at x. We will study various devices for obtaining good approximations to the values of the nonelementary functions we encounter. (12)). It appears that on each solution curve. an integral such as the one in Eq. • Comment: In general.3. Typical solution curves defined by Eq.. ···· ··· ~« ' . (15).. y (x ) ~ 0 as x ~ + oo.. ... . . . Then the particular solution in Eq. The sine integral function is available in most scientific computing systems and can be used to plot typical solution curves defined by Eq. In Chapter 6 we will discuss numerical integration of differential equations in some detail. (14) would (for given x) need to be approximated numericallyusing Simpson's rule. • In the sequel we will see that it is the exceptionrather than the rule.3 shows a selection of solution curves with initial values y (l) = y0 ranging from y0 = 3 to y0 = 3. '"" " '"""'~ •· ··. 1965).· Example 3 Solve the initial value problem x dx 2dy + xy = smx. and this is in fact true because the sine integral function is bounded.5. X (15) 0 5 10 X 15 FIGURE 1... In this case. . • . t J (14) In accord with Theorem 1. however. Handbook of Mathematical Functions (New York: Dover. A good set of tables of special functions is Abramowitz and Stegun. (15).. this solution is defined on the whole positive xaxis. Figure 1. y( ) =YO· 1 (13) Solution Division by x 2 gives the linear firstorder equation +y=2 dx x x dy 1 sinx with P(x) = 1/x and Q(x) = (sinx)jx 2 . 1 0 t 3 2 which appears with sufficient frequency in applications that its values have been tabulated.. we have the sine integral function x sin t Si(x) = dt.. so the desired particular solution is given by 1 [ Yo+ y(x) = ~ lx 1 sint dt .
{grams output}~ rici ~t: ~X ~t r0 c0 ~t. but c0 denotes the variable concentration x(t) C0 (t) = V (t) (17) of solute in the tank at time t.c. Suppose that solution with a concentration of ci grams of solute per liter of solution flows into the tank at the constant rate of ri liters per second.examination of the behavior of the system over a short time interval [t . Important: Equation ( 18) need not be committed to memory. = r0 ) of solution in the tank at time t. dt v (18) If V0 = V(O). where V(t) denotes the volume (not constant unless r. note how the cancellation of dimensions checks our computations: Amount x(t) Volume V(t) Concentration c0 (t) Output: r0 Lis. then the error in this approximation also approaches zero. we estimate the change ~x in x during the brief time interval [t. given the amount x (0) = x 0 at time t = 0. The amount of solute that flows out of the tank during the same time interval depends on the concentration c0 (t) of solute in the solution at time t. then V(t) = V0 + (r.c. Remark: It was convenient for us to use giL mass/volume units in deriving Eq.roCo . In the following example we measure both in • cubic kilometers. liters ) ( grams) c. There is both inflow and outflow. ( 18).= rici ..r0 C0 • Finally.5. because it is a very useful tool for obtaining all sorts of differential equations.x . . and we obtain the differential equation . we take the limit as ~t + 0.that you should strive to understand. dt dx (16) in which ri. (18) is a linear firstorder differential equation for the amount x (t) of solute in the tank at time t. The singletank mixture problem.4. t + ~t]. yields a quantity measured in grams. It is the process we used to obtain that equation.4. c. Then ~x ={grams input}. so Eq. 1.. c 0 (t) = x(t)jV(t).5. The amount of solute that flows into the tank during ~t seconds is r. C0 giL = f.. But as noted in Fig. ( r. To check this. Thus the amount x (t) of solute in the tank satisfies the differential equation dx ro . and we want to compute the amount x(t) of solute in the tank at time t.52 Chapter 1 FirstOrder Differential Equations Mixture Problems As a first application of linear firstorder equations.=r. hter second (~t seconds) FIGURE 1. . We now divide by  ~t ~ riCi . . we consider a tank containing a solutiona mixture of solute and solventsuch as salt dissolved in water.. and that the solution in the tankkept thoroughly mixed by stirringflows out at the constant rate of r 0 liters per second. and r 0 are constants. if all the functions involved are continuous and x (t) is differentiable. ~t grams. t + ~t ].r0 )t. To set up a differential equation for x(t). But any other consistent system of units can be used to measure amounts of solute and volumes of solution.
3 ( 90 + t x ) .t.6. and the wellstirred mixture flows out of the tank at the rate of 3 galjmin. due to the differing rates of inflow and outflow.+ px =q dt dx (20) with constant coefficients p = r jV.x dt v .x in the amount x of salt in the tank from time t to time t + . Eq. Brine containing 2lbjgal of salt flows into the tank at the rate of 4 galj min. and the question is this: When is x (t) = 2c V? With this notation.~ (ert/V 1) J.6. (21) x(t) = cV +4cverr. If the outflow henceforth is perfectly mixed lake water.v_ To find when x (t) = 2c V.901 (years) .1.v [5cV + . we therefore need only solve the equation cV + 4cve.5 Linear FirstOrder Equations 53 Example 4 Assume th~t L~e Erie has a ~~lume of 480 km3 and that its rate of inflow (from Lake Huron) and outflow (to Lake Ontario) are both 350 km3 per year.pt [xo +~ (eP 1 .t (minutes) is given by .x Solution ~ (4)(2) .= rc. Suppose that at the timet = 0 (years). . The change . = ro = r = 350 (km3 jyr). How much salt does the tank contain when it is full? The interesting feature of this example is that.rt/V = 2cV for t = ln4 r v = 480 ln4 350 ~ 1.6. the volume of brine in the tank increases steadily with V (t) = 90 + t gallons. the pollutant concentration of Lake Eriecaused by past industrial pollution that has now been ordered to ceaseis five times that of Lake Huron.I) J = err. how long will it take to reduce the pollution concentration in Lake Erie to twice that of Lake Huron? Here we have Solution V = 480 (km3 ). q = rc.t. and integrating factor p = eP1 • You can either solve this equation directly or apply the formula in (12). and xo = x(O) = 5cV. (19) which we rewrite in the linear firstorder form .6. ri ci = c (the pollutant concentration of Lake Huron). The latter gives x(t) = ept [xo + 1 1 qeP1 dt] = e . (18) is the separable equation dx r .6. • Example 5 A 120gallon (gal) tank initially contains 90 lb of salt dissolved in 90 gal of water.
Problems Find general solutions of the differential equations in Problems 1 through 25.y = x. 26. r eJo 1 2 30. y(l) =0 as an integral as in Example 3 of this section. (1 .2 y' = (1. (1 +2xy)dy = 1 + y 2 dx 29. y'. 17. find the corresponding particular solution. y(O) = 0 3. y(O) = 5 xy' + (2x.2y = 3e2x. y'. (x + yeY)  dy = 1 dx 28. 7. 18. 6. 19. which gives Dr [(90 + t) 3 x] = 8(90 + t) 3 .3y = 9x 3 xy' + y = 3xy. Express the general solution of dyjdx of the error function 2 erf(x) = .3)y = 4x 4 (x 2 + 4)y' + 3xy = x. Throughout. y(O) = 1 xy' . (90 + t) 3x = 2(90 + t) 4 +C. y' + 3y = 2xe. 904 The tank is full after 30 min.2xy = ex xy' + 2y = 3x. 9. = exp ( / 3 dt) 90+ t = e 31n(90+t) = (90 + t) 3 . primes denote derivatives with respect to x.y)cosx. 23.3 120 of salt in the tank. 13.54 Chapter 1 FirstOrder Differential Equations so our differential equation is .+3x 3 y = 6xexp(~x 2 ). 8.+ An integrating factor is p(x) dx dt 3 90 + t X= 8. y(l) = 0 xy' + 3y = 2x 5 . y(2rr) = 0 y' = 2xy + 3x 2 exp(x 2 ). y (0) = 0 2 2. y(2) = 5 2xy' + y = 10. 10. y(O) = 1 dx 21.4xy 2 ) dy dx = y3 27. 1.3x 4.Jx' 3xy' + y = 12x xy'. 15. (x 2 + 1) . 5. 24. y(n) = 2 (1 + x)y' + y = cosx.. 22. y' + y = 2. 14. 20. 11. Substitution of x(O) timet is = 90 gives C = (90)4 . y(l) = 10 y' + 2xy = x. y(O) = 0 Solve the differential equations in Problems 26 through 28 by regarding y as the independent variable rather than x. y(l) = 5 xy' + 5y = 7x 2 . . y(l) = 7 2xy' . y(O) = 1 dy 25. If an initial condition is given.fir = 1 + 2xy in terms dt.( 90 + t) 3 . y(O) = . and when t = 30. 16. Express the solution of the initial value problem dy 2xdx = y + 2x COSX. y(O) = 1 xy' = 2y + x 3 cosx y' + y cotx = cosx y' = 1 + x + y + xy. y(2) = 1 y' + y =ex. 202 (lb) • xy' = 3y + x 4 cos x. 12. we have ~ 904 x(30) = 2(90 + 30).3y = x 3 . so the amount of salt in the tank at x(t) = 2(90 + t).
5. (b) Find the maximum amount of ethanol ever in tank 2. A cascade of two tanks. 33. 1. Show that y(x) = Yc(x) + Yp(x) is a general solution of dyjdx + P(x)y = Q(x).1.5. with V 1 = 100 (gal) and V2 = 200 (gal) the volumes of brine in the two tanks. A multiple cascade is shown in Fig. 35. The three flow rates indicated in the figure are each 5 gal/min. 39. 1. The only differences are that this lake has a volume of 1640 km3 and an inflowoutflow rate of 410 km3 /year. is a particular solution of dyjdx + P(x)y = Q(x). Pure water flows into tank 1 at 10 galjmin. and the wellmixed brine in the tank flows out at the rate of 3 galjs. all the remaining tanks contain 2 gal of pure water each. find the maximum amount of salt ever in tank 2. and the other two flow rates are also 10 gal/min. = e. Pure water is pumped into tank 0 at 1 galjmin. There is a daily inflow of 500 million ft3 of water with a pollutant concentration of 0. A tank initially contains 60 gal of pure water. How long will it take to reduce the pollutant concentration in the reservoir to 0. Each tank also initially contains 50 lb of salt. FIGURE 1. 55 time t. tank 0 contains 1 gal of ethanol and 1 gal of water. 1. Pure water is pumped into the tank at the rate of 5 Ljs. How much salt will the tank contain when it is full of brine? Consider the cascade of two tanks shown in Fig.200' 31. (b) Suppose that y (t) is the amount of salt in tank 2 at FIGURE 1. A multiple cascade. with pure water flowing into tank 1.5.10%? Rework Example 4 for the case of Lake Ontario. 40. 37. (b) What is the maximum amount of salt ever in the tank? A 400gal tank initially contains 100 gal of brine containing 50 lb of salt. (c) Suppose that yc(x) is any general solution of dyjdx + P (x) y = 0 and that Yp (x) is any particular solution of dyjdx + P(x)y = Q(x). y(O) = 1.5.5. Suppose that in the cascade shown in Fig.6. Lawrence River and receives inflow from Lake Erie (via the Niagara River). .is pumped out at the same rate.f P(x)dx [! ( Q(x)ef P(x)dx) dx J 32. Brine containing 1 lb of salt per gallon enters the tank at the rate of 5 galj s.05% and an equal daily outflow of the wellmixed water in the reservoir. Show first that dy dt 5x 5y 100. How long will it be until only 10 kg of salt remains in the tank? Consider a reservoir with a volume of 8 billion cubic feet (ft3 ) and an initial pollutant concentration of 0.5 Linear FirstOrder Equations Problems 31 and 32 illustratefor the special case of firstorder linear equationstechniques that will be important when we study higherorder linear equations in Chapter 3. A tank contains 1000 liters (L) of a solution consisting of 100 kg of salt dissolved in water. 36. (a) Find constants A and B such that Yp(x) = A sinx + B cosx is a solution of dyjdx + y = 2 sinx.5. (a) Show that Yc(X) = ce. and the mixturekept uniform by stirring. 34. (c) Finally.25%.5. thus the tank is empty after exactly 1 h. (b) Use the result of part (a) and the method of Problem 31 to find the general solution of dyjdx + y = 2sinx. (a) Find the amount of salt in the tank after t minutes. (a) Find the amount x(t) of salt in tank 1 at time t. (c) Solve the initial value problem dyjdx + y = 2 sin x. (a) Find the amounts x(t) and y(t) of ethanol in the two tanks at time t ~ 0. Brine containing 1 lb of salt per gallon enters the tank at 2 galjmin. (b) Show and then solve for y(t). using the function x(t) found in part (a). tank 1 initially contains 100 gal of pure ethanol and tank 2 initially contains 100 gal of pure water. and the (perfectly mixed) solution leaves the tank at 3 galjmin.6. 38. which empties into the St. At time t = 0.5.f P(x)dx is a general solution of dyjdx that Yp(X) + P(x)y = 0.
Meanwhile.5.5. the amount available for her retirement at age 70.9. which accumulates interest at a continuous annual rate of 6%. Then show that dvjdt = g/4.112 • (b) Show by induction on n that tnet/2 Xn(t) = . 10 8 6 4 2 :>. How long does it take the pollutant concentration in the reservoir to reach 5 Llm 3 ? 25 X 20 15 10 5 ~~ 10 20 Probl<m 46 Problem 45 30 40 50 60 FIGURE 1.. = mg. 46. Slope field and solution curves for y' = x . Slope field and solution curves for y' = x + y. 12% of her salary is deposited continuously in a retirement account. Suppose that a falling hailstone with density 8 = 1 starts from rest with negligible radius r = 0. The wellmixed water in the reservoir flows out at the same rate.y. (a) Show that every solution curve approaches the straight line y = x . Verify that the graph of x(t) resembles the steadily rising curve in Fig.9.998. Let Xn (t) denote the amount of ethanol in tank n at time t.7. that the mixtures are kept perfectly uniform by stirring.I asx+ oo. The incoming water has a pollutant concentration of c(t) = 10 liters per cubic meter (L/m3 ) . (a) Estimate L'lA in terms of L'lt to derive the differential equation satisfied by the amount A(t) in her retirement account after t years.5.J2irii that Mn ~ (2rrn). 5.999. Assume. 43. Figure 1.002.n.1 as x + +oo. and the positive yaxis points downward.56 Chapter 1 FirstOrder Differential Equations 44. where m is the variable mass of the hailstone. (b) Foreachofthefivevaluesy 1 = 10. n! 2n (c) Show that the maximum value of Xn(t) for n > 0 is Mn = Xn(2n) = nnenjn!. Thereafter its radius is r = kt (k is a constant) as it grows by accretion during its fall. (a) Show that every solution curve approaches the straight line y = x . Does it seem predictable that the lake's polutant content should ultimately oscillate periodically about an average level of 20 million liters? . (b) For each of the five values y 1 = 3. 4. with an average concentration of 10 L/m 3 and a period of oscillation of slightly over 6 ± months. 0 2 4 6 8 10 X FIGURE 1.5. Thus the hailstone falls as though it were under onefourth the influence of gravity.000 per year.112 • 41. You r first task is to find the amount x (t) of pollutant (in millions of liters) in the reservoir after t months. determine the initial value y0 (accurate to five decimal places) such that y(5) = y 1 for the solution satisfying the initial condition y( 5) =Yo· 10 and the varying mixture in each tank is pumped into the one below it at the same rate.8. 42. 0 2 4 6 8 10 X 45. (b) Compute A(40). 4. The incoming water has pollutant concentration c (t) 10(1 + cos t) L/m3 that varies between 0 and 20. Initially it is filled with fresh water. which approaches asymptotically the graph of the equilibrium solution x(t) = 20 that corresponds to the reservoir's longterm pollutant content. as usual.001. v = dyjdt is its velocity. (d) Conclude from Stirling's approximation n! ~ nne. (a) Show that x 0 (t) = e.8 shows a slope field and typical solution curves for the equation y' = x + y. and 4. but at time t = 0 water contaminated with a liquid pollutant begins flowing into the reservoir at the rate of 200 thousand cubic meters per month.y . Figure 1.000.5) = Yo· FIGURE 1. Her salary S(t) increases exponentially. Use Newton's second lawaccording to which the net force F acting on a possibly variable mass m equals the time rate of change dpjdt of its momentum p = m vto set up and solve the initial value problem d (mv) dt 8 6 4 2 :>. Problems 45 and 46 deal with a shallow rese rvoir that has a one square kilometer water suiface and an average water depth of 2 meters. 5. A 30yearold woman accepts an engineering position with a starting salary of $30. 1. v(O) = 0.7 shows a slope field and typical solution curves for the equation y' = x . 3.5.5. determine the initial value y0 (accurate to four decimal places) such that y(5) = y 1 for the solution satisfying the initial condition y( . with S(t) = 30e 1120 thousand dollars after t years. 0. and I 0.for n > 0. Graphs of solutions in Problems 45 and 46.
If we write Newton's law of cooling (Eq. For instance. GA with a minimum temperature of 70°F when t = 4 (4 A.M. but with the outside temperature A(t) given by Eq. then these oscillations have a period of 24 hours (so that the cycle of outdoor temperatures repeats itself daily) and Eq. and we cannot afford to have it repaired until payday at the end of the month.M.5v'3 sin wt . Suppose that our air conditioner fails at time to = 0 one midnight.kt + c1 cos wt + d1 sinwt.2 for a wellinsulated building with tightly sealed windows). (1) instead of a constant ambient temperature A.1. we would take A(t) = 80. Typical values of the proportionality constant k range from 0.2 to 0. or less than 0.5 for a poorly insulated building with open windows. indeed. (1). (4) .5 cos wt. du .4) = 80.5. (1) provides a realistic model for the temperature outside a house on a day when no change in the overall daytoday weather pattern is occurring. consider indoor temperature oscillations that are driven by outdoor temperature oscillations of the form A(t) = a0 + a 1 coswt + b 1 sinwt.9. You may want to use the integral formulas in 49 and 50 of the endpapers. resemble the oscillatory curve shown in Fig. You should get the solution SCENARIO: u(t) = ao + coe.5 Application For an interesting applied problem that involves the solution of a linear differential equation.5 Linear FirstOrder Equations Verify that the graph of x(t) does.{3) = cos a cos f3 +sin a sin f3 to get ao = 80. Begin your investigation by solving Eq.+ ku = k(ao + a 1 coswt + dt b1 sinwt) (3) with coefficient functions P(t) = k and Q(t) = kA(t).). we get the linear firstorder differential equation dt = that is. We therefore want to investigate the resulting indoor temperatures that we must endure for the next several days. or possibly a computer algebra system. du k(u. and b1 = 5v'3 in Eq. (1) If w = n /12.5 (although k might be greater than 0. (2) by using the identity cos(a . (3) with the initial condition u(O) = u 0 (the indoor temperature at the time of the failure of the air conditioner). 1. (2) We derived Eq.1) for the corresponding indoor temperature u(t) at timet. a 1 = 5.) and a maximum of 90°F when t = 16 (4 P.A(t)). How long does it 57 take the pollutant concentration in the reservoir to reach 5 L/m 3 ? 1. (3) of Section 1.10 cosw(t. for a typical July day in Athens.
5.. 71.. Solution curves given by Eq....3351) cos 12 ...the indoor temperature "settles down" within about 18 hours to a periodic daily oscillation. Comparison of indoor and outdoor temperature oscillations. Indeed. b 1 = 5. = rr/12. 90 :S 80 ::! eo 85 75 = 80.. 1.. (What is the wintersummer difference for the indoor temperature problem?) You may wish to explore the use of available technology both to solve the differential equation and to graph its solution for the indoor temperature in comparison with the outdoor temperature. JTt (5) 90 :3 80 ::! eo 85 as t ~ Observe first that the "damped" exponential term in Eq. and k = 0. But the amplitude of temperature variation is less indoors than outdoors. comparison of Eqs.(5.11. '92. ..J3 (as in Eq.5 hours.6036) sin 12 . so the hottest part of the day inside is early evening rather than late afternoon (as outside).5.50824 ~ 3. a 1 = 5. Consequently..(6.M..2 JTt u(t) = 80 + er. With a0 = 80..0707) cos 12 (t  7T (7) 70 65 600 FIGURE 1..5082). Do you see that this implies that the indoor temperature varies between a minimum of about 74 oF and a maximum of about 86 oF? Finally.. this solution reduces (approximately) to 1()() .s (u 0  95 82.11. (6) can be rewritten (verify this!) as l 00 . Eq.5.10.(5... leaving the longterm "steady periodic" solution JTt JTt 75 70 65 u 5p(t) = 80 + (2... the longterm indoor temperatures oscillate every 24 hours around the same average temperature 80° F as the average outdoor temperature. (5) approaches zero +oo.r. For a personal problem to investigate.. (5) with u 0 = 65.3351) + (2. each evening..3351) cos.6036) sin. (2)).(6.. Thus the temperature inside the house continues to rise until about 7:30 P.5. Figure 1. 68.. w (for instance).9656) 7. Outdoor 95 u(t) = 80. You might also consider a winter day instead of a summer day..58 Chapter 1 FirstOrder Differential Equations where dl = kwa1 + k 2 b1 k2 + w2 with w = rr/12. carry out a similar analysis using average July daily maximum/minimum figures for your own locale and a value of k appropriate to your own home. (2) and (7) indicates that the indoor temperature lags behind the outdoor temperature by about 7. . as illustrated in Fig. Observe thatwhatever the initial temperature. using the trigonometric identity mentioned earlier..0707) cos ( 7~  1.10 shows a number of solution curves corresponding to possible initial temperatures u 0 ranging from 65°F to 95°F. 95. 12 12 (6) 600 FIGURE 1.
y). (5) is one we can solve. (5). If the substitution relation in Eq. V) and a{Jjav = fJv(X. If this new equation is either separable or linear. v). If we substitute the righthand side in (4) for d y jdx in Eq. Example 1 Solve the differential equation . (1) with dependent variable y and independent variable x. v(x )) will be a solution of the original Eq. But many applications involve differential equations that are neither separable nor linear. V) are known functions of x and v.6 Substitution Methods and Exact Equations 59  Substitution Methods and Exact Equations The firstorder differential equations we have solved in the previous sections have all been either separable or linear. v ) (5) with new dependent variable v. (2). the result is a new differential equation of the form dv dx = g(x . the differential equation dy dx = f(x. The trick is to select a substitution such that the transformed Eq. then we can apply the methods of preceding sections to solve it. (1). For instance. Thus the differential equation dy dx = (x + y + 3)2 practically demands the substitution v = x + y + 3 of the form in Eq. If v = v(x) is a solution of Eq. y) (2) of x and y that suggests itself as a new independent variable v. it may require a fair amount of ingenuity or trial and error. Even when possible. (1) and then solve for dvjdx .1. then y = fJ(x. may contain a conspicuous combination v = a(x . this is not always easy. (2) can be solved for y = fJ(x. In this section we illustrate (mainly with examples) substitution methods that sometimes can be used to transform a given differential equation into one that we already know how to solve. (3) then application of the chain ruleregarding v as an (unknown) function of xyields (4) where the partial derivatives a{Jjax = fJx (X.
We see that. dy dv = v + x .1 v +C. dx dx (8) . This situation is fairly typical of nonlinear differential equations. y = v . in contrast with linear differential equations. whose solutions are continuous wherever the coefficient functions in the equation are continuous.x . dx x If we make the substitutions (7) v= . and we have no difficulty in obtaining its solution 10 8 6 1111111 I I I I I I dv Ill I I II I I I I I x = I I I I I I I I I I 11111111111 1111111 Ill I dv = tan.. 1+v 2 4 2 . n/2). although the function f (x. 0 Hil"7'::::::=':':7t!77f+++H.x .60 Chapter 1 FirstOrder Differential Equations Solution As indicated earlier.3. each solution is continuous only on a bounded interval. y(x) = tan(x. • Example 1 illustrates the fact that any differential equation of the form  dy dx = F(ax +by +c) (6) can be transformed into a separable equation by use of the substitution v = ax + by + c (see Problem 55). the general solution of the original equation dyjdx = (x + y + 3) 2 is x + y + 3 = tan(x. y) = (x+ y+3) 2 is continuously differentiable for all x andy.6. Homogeneous Equations A homogeneous firstorder differential equation is one that can be written in the form dy = F (~). The paragraphs that follow deal with other classes of firstorder equations for which there are standard substitutions that are known to succeed. Slope field and solution curves for y' = (x + y + 3) 2 • Remark: Figure 1.C) . • 5 X 5 FIGURE 1. C n/2 < x < C + n j2. because the tangent function is continuous on the open interval ( n/2.:H 2 4 6 8 I I I JO I I I I I I So v = tan(x . that is..C < n j2. Because v = x + y + 3. that is. let's try the substitution V =X+ y + 3.1. X y y = vx .3. dx dx so the transformed equation is = 1 + v 2• dx This is a separable equation. In particular.1 shows a slope field and typical solution curves for the differential equation of Example 1.C) ..  that is. the particular solution with arbitrary constant value C is continuous on the interval where n /2 < x..6. Then dy dv = 1.C).
y ) y ' = Q(x . a differential equation of the form P (x . and so forthis the equation dy _ (y)q A B + C(y)s ( y)n x dx X X which evidently can be written (by another division) in the form of Eq.xm+n = (yfxt. but we recognize it as a homogeneous equation by writing it in the form dy dx 4x2 + 3y2 = 2 2xy (~) + ~ (~). then the resultbecause xmyn." Consider a differential equation of the form whose polynomial coefficient functions are "homogeneous" in the sense that each of their terms has the same total degree.= 4x 2 + 3l. y ) with polynomial coefficients P and Q is homogeneous if the terms in these polynomials all have the same total degree K. If we divide each side of(*) by xK. dy dx dv dx y v = .6 Substitution Methods and Exact Equations then Eq. X and 1 X v y v+ x and hence dv x dx dv 2 =dx v + v .1.dv = v2 +4 J 2_ dx. dx Solution This equation is neither separable nor linear. (7) is transformed into the separable equation dv x . m + n = p + q = r + s = K.= F(v). v v2 + 4 2v ' 3 2 = ~+2= 2 f : 2v. ··········· ······· Example 2 Solve the differential equation dy 2xy. The differential equation in the following example is of this form with K = 2.v. Remark: A dictionary definition of "homogeneous" is "of a similar kind or nature. More generally. =v+x. . x ln(v 2 + 4) = ln lxl + ln C. y 2 x The substitutions in (8) then take the form y = These yield VX. (7). dx 61 Thus every homogeneous firstorder differential equation can be reduced to an integration problem by means of the substitutions in (8).
Indeed. y(xo) = 0.4x 2 that are defined for x > 4jk if the constant k is positive. Now note that v(xo) y(xo)fxo = 0. is the desired particular solution.. It follows that k > 0 in the case of solutions that are defined for x > 0. Solution curves for xy' = y + Jxz _ y2.62 Chapter l FirstOrder Differential Equations We apply the exponential function to both sides of the last equation to obtain X FIGURE 1. Figure 1.2 exhibits symmetry about both coordinate axes..2. Because of the radical in the differential equation..1 v = lnx +C. there are positivevalued and negativevalued solutions of the forms y(x) = ±Jkx 3 .. the family of solution curves illustrated in Fig. • Example 3 Solve the initial value problem dy x dx = y + Jx 2  r::: y2. soC= sin. x sin. 1.6.1 0. Note that the lefthand side of this equation is necessarily nonnegative. and for x < 4/ kif k is negative.6.6.ln xo = ln xo.3 shows some typical solution curves. these solution curves are confined to the indicated triangular region x ~ IYI· You can check that the boundary lines y = x andy = x (for x > 0) are singular solution curves that consist of points of tangency with the solution curves found earlier. where xo > 0. Slope field and solution curves for 2xyy' = 4x 2 + 3y 2 .. Hence v = l/ / / / / / / / / / ~=sin (lnx lnx0 ) X =sin (In~). Actually. while k < 0 for solutions where x < 0. we get v+xdv =v+JIv 2 • dx ' I Jl .20 30 40 y =x sin(ln(x/20)) y=x y =x sin(ln(x/ 10) We need not write In lxl because x > 0 near x = x 0 > 0. XQ and therefore y(x) = x sin (In : 0 ) 0 10 20 X 50 30 40 50 FIGURE 1. • . Solution We divide both sides by x and find that so we make the substitutions in (8).=I 1 v2 dv 2_ dx. 0 10 .6.3. 50 40 30 20 10 .
1.6 Substitution Methods and Exact Equations 63 Bernoulli Equations A firstorder differential equation of the form . If either n = 0 or n I. it is more efficient to make the substitution in Eq. Rather than memorizing the form of this transformed equation.v =4x dx x with integrating factor p = ef<. (10) explicitly.n)P(x)v = dx (1.n)Q(x). dx 2x y we see that it is also a Bernoulli equation with P (x) n = 1. Example 4 If we rewrite the homogeneous equation 2xyy' = 4x 2 + 3 y 2 of Example 2 in the form dy 3 2x y=. v112 _ 2 2 dx 2x Then multiplication by 2v 112 produces the linear equation dv 3 .. as we ask you to show in Problem 56. then Eq.3 f x ) dx = x.3 • So we obtain • . (9) is linear. (9) into the linear equation dv + (1. the substitution (10) transforms Eq. y = vlf2. as in the following examples. Otherwise. and dy dy dv 1 _1 12 dv ==v dx dv dx 2 dx This gives 1 dv _ 3v i / 2 = xv112. Q(x) 2x. Hence we substitute v = y' 2 3/(2x). and 1 .+ P(x)y = dy dx Q(x)yn (9) is called a Bernoulli equation.n = 2.
nor is it a Bernoulli equation.2f x ) dx = x. nor homogeneous. dx x After multiplying by the integrating factor p = ljx . nor linear. we find that ~v = and hence J 3x 2 d x = x 3 + C.+ 6v. 1 • (11) Example 6 The equation is neither separable. (11) into the linear equation x v'(x) = 3x 4 + v(x ).. But we observe that y appears only in the combinations e 2Y and Dx(e 2Y) = 2e 2Yy' .64 Chapter 1 FirstOrder Differential Equations ·. so e 2Y = v = x 4 + Cx. dv 1 3 .3 = 3xv. This prompts the substitution dv 2Ydy =2edx dx that transforms Eq.yields the linear equation v=1 transform it into dv dx 2 x with integrating factor p = ef<. • . y (x ) = 1In lx 4 + Cx j . This gives and finally. 1 . y(x) = (x + Cx2)3.4 • dx 4 Division by 3xv.2 .v = 3x .4 . v = yl /3' and dy dy dv _4 dv ==3v dx dv dx dx dv 3xv. but it is a Bernoulli equation with n = ~.·· Example 5 The equation x dy + 6y = 3xy4f3 dx is neither separable nor linear nor homogeneous.n = The substitutions t. that is.
we finally obtain _a y(x) 2 [(x)lk . The components of the velocity vector of the airplane. ( 16) and then solving for v is (18) Because y = xv. The plane travels with constant speed vo relative to the wind.VoY. As indicated in Fig. (12) takes the homogeneous form (14) The substitution y = xv. Figure 1.1. (16) = 0 yields (17) C=klna. y' = v + xv' then leads routinely to J.k In x + C.5.wJx2 dx dxjdt v0 x If we set X k= . As we ask you to show in Problem 68.::=::::::::=::::: dt )x2 + yz'  dy . 0) located due east of its intended destinationan airport located at the origin (0. (12) dy = dyjdt 1 ( . They are (a. = vo sm () dt +w =  VoY )x2 + y2 + w. the ratio of the windspeed to the plane's airspeed.6. then Eq. 0). Hence the trajectory y  = f (x) of the plane satisfies the differential equation + y2 ) .= .o= .J1 + In ( v and the initial condition v(a) . we assume that the plane's pilot maintains its heading directly toward the origin.. vo w (13) FIGURE 1. 1.4.5 helps us derive the plane's velocity components relative to the ground. The airplane headed for the origin.6. the result of substituting ( 17) in Eq. we find that + J1 + v2) = y(a)ja = .(x)l+k] a a (19) for the equation of the plane's trajectory.v2  J~ x dx · (15) By trigonometric substitution.6.  dx vox = . which is blowing due north with constant speed w.6 Substitution Methods and Exact Equations 65 Flight Trajectories y y =f(x) Suppose that an airplane departs from the point (a.vo cos () = .6. 0) x FIGURE 1. or by consulting a table for the integral on the left. .4.== d v===..
Eq. y ). On the other hand. If w = v 0 (so that k = 1). so the plane's trajectory approaches the point (0._) 1/5] 200 5 200 ' and we readily solve the equation y ' (x) = 0 to obtain (x /200) 215 = ~. ax ay d x that is. (19) pass through the origin. aj2) rather than (0. Provided that Eq. The three cases w < v0 (plane velocity exceeds wind velocity). (19) yields y(x) = 100 [ ( 200 X )4/5. y ) and N(x . y) = Fy(x. w = v0 (equal velocities).::.6. (20) yields dy = dx 2 ! [~ 5 (..6. 0) X Ymax = 100 2) . The vertical scale there is exaggerated by a factor of 4. FIGURE 1.6.) • Exact Differential Equations We have seen that a general solution y(x) of a firstorder differential equation is often defined implicitly by an equation of the form F(x.. so that the plane reaches its destination. (19) that y + +oo as x + 0.6. With these values. The situation is even worse if w > v0 (so k > 1)in this case it follows from Eq.::_)1/5. w < v0 ) does the curve in Eq. and w > v0 (wind is greater). y ) + N(x.6.=0. (21) where C is a constant.= 0. y ) = Fx(x . 1.66 Chapter 1 FirstOrder Differential Equations Note that only in the case k < 1 (that is. 0). 0). v0 = 500 mijh. y(x)) = C. (The graph of the function in Eq. y). 1. (19) takes the form y(x) = ~a(l . this gives the original differential equation in the form aF aF dy + . what is the maximum value of y(x) for 0 ~X~ 200? Solution y Differentiation of the function in Eq. dx (22) where M (x.(200 )6/5] · X (20) Now suppose that we want to find the maximum amount by which the plane is blown off course during its trip.~ (. Example 7 If a= 200 mi. (20) is the one used to construct Fig. then k = wjv0 = ~'so the plane will succeed in reaching the airport at (0.(2) 3 [( 3 2 3 ] = 400 27 ~ 14.4. then Eq. given the identity in (21). . (21) implicitly defines y as a differentiable function of x . Hence (a.x 2ja 2 ). we can recover the original differential equation by differentiating each side with respect to x. The three cases are illustrated in Fig. That is.. and w = 100 mijh. Thus the plane is blown almost 15 mi north at one point during its westward journey. dy M(x .81...
y) such that aF =M ax then the equation F(x. y) dx + N(x. then the differential equation in question is not exact. aN ax aM ay aN ax (24) is a necessary condition that the differential equation M dx + N dy = 0 be exact. Eq. The general firstorder differential equation y' = f (x. The preceding discussion shows that. (23). if My =F Nx. Natural questions are these: How can we determine whether the differential equation in (23) is exact? And if it is exact. if you prefer. (22) in the more symmetric form M(x. it then follows that Thus the equation aM ay = Fxy = Fyx = .113 • • . (23) is called an exact differential equationthe differential dF = Fxdx + Fydy of F(x.1. (23) is exact and M and N have continuous partial derivatives.6 Substitution Methods and Exact Equations 67 It is sometimes convenient to rewrite Eq. if there exists a function F (x. Example 8 The differential equation y 3 dx + 3xl dy = 0 (25) is exact because we can immediately see that the function F (x. (23) called its differential form. implicitly defines a general solution of Eq. y) such that Fx = M and Fy = Nthere is no such function. y) = C and =N aF ay . y) dy = 0. That is. so we need not attempt to find a function F (x. y) = x y 3 has the property that Fx = y 3 and Fy = 3xy 2 . y) and N = 1. In this case. (25) is xi =C. how can we find the function F such that Fx = M and Fy = N? To answer the first question. then they are equal: Fxy = Fyx· If Eq. let us recall that if the mixed secondorder partial derivatives Fxy and Fyx are continuous on an open set in the xyplane. y) is exactly M dx + N dy. y(x) = kx . Thus a general solution of Eq. y) can be written in this form with M = f (x.
y)= J M(x. (24) holds on R. y) defined on R with aF jax = M and aF jay= N if and only ifEq. In other words. y)dy = 0 (23) is exact in R if and only if aM ay aN ax (24) at each point of R.) We plan to choose g (y) so that aF = N = ay (a J ay ay M(x.68 Chapter 1 FirstOrder Differential Equations But suppose that we divide each term of the differential equation in Example 8 by y 2 to obtain + 3xdy = 0. whether a given differential equation is exact or not is related to the precise form M dx + N dy = 0 in which it is written. (24) is not satisfied. That is. y) with respect to x. y)dx + N(x. (27). the notation J M(x. The differential equations in (25) and (26) are essentially equivalent. ay ax (26) Hence the necessary condition in Eq. then the differential equation M dx + N dy = 0 is exact. for any function g(y). y) and N (x. so that g I (y) = N. Theorem 1 tells us that (subject to differentiability conditions usually satisfied in practice) the necessary condition in (24) is also a sufficient condition for exactness. and they have exactly the same solutions. y) dx denotes an antiderivative of M (x. Then the differential equation M(x. (28) . (In Eq. we must show that if Eq. Proof: We have seen already that it is necessary for Eq. c < y < d. with M = y and N = 3x.y)dx ) I +g(y) as well. (24) to hold if Eq. We are confronted with a curious situation here. then we can construct a function F(x. if My = Nx. In brief. yet one is exact and the other is not. y) such that aF. y) are continuous and have continuous firstorder partial derivatives in the open rectangle R: a < x < b. there exists a function F (x. that is.ax = M and aFjay = N. To prove the converse. This equation is not exact because. Note first that. we have ydx aM aN =1#3=. (24) holds. (23) is to be exact.y)dx+g(y) (27) satisfies the condition aFjax = M. the function F(x. a/ M(x. y)dx. THEOREM 1 Criterion for Exactness Suppose that the functions M (x.
6 Substitution Methods and Exact Equations 69 To see that there is such a function of y. y)dx ==0 by hypothesis.. First we integrate M (x . (27) to obtain F(x.l ) dx = 3x 2 y . ay ax Integrating a F . Then we differentiate with respect toy and set aFjay = N(x. This yields a general solution in the implicit form F(x. we get F(x. indeed. y) = J M(x. y) dx + g(y). y) with respect to x and write F(x._!_ ~ ax ax ay aN a a = . This yields aF = 3x 2 ay 3xl + g'(y) = 4y + 3x 2  3xl. Then we determine g (y) by imposing the condition that aFjay = N(x. it suffices to show that its derivative with respect to x is identically zero. (28). y). y) = 6xy. y) =C. (30) Solution Let M(x.ax ay ax aN ax aM ay J J M (x. We substitute this result in Eq. y) with respect to x. y). .1. y) dx M(x. We can then find g(y) by integrating with respect toy. But _!_ ax (N  _!_ ay J M (x. (28) is a function of y alone.aay J M(x. and hence on an interval as a function of x.xl + g(y).3xy 2 • The given equation is exact because aM aN 2 =6x3y = . (29). Because the righthand side in Eq. y) = J  (6xy . y) = 4y + 3x 2 .ax = M (x. y) dx + J( N(x. (27) and (28). y). Example 9 Solve the differential equation (6xy . So we can. • Instead of memorizing Eq. find the desired function g(y) by integrating Eq. y) = J M(x. (28) is defined on a rectangle. it is usually better to solve an exact equation M dx + N d y = 0 by carrying out the process indicated by Eqs. y) dx) = aN . y) dx) dy (29) as the desired function with Fx = M and Fy = N. it suffices to show that the righthand side in Eq.l ) dx + (4y + 3x 2  3xl) dy = 0.y 3 and N(x. thinking of the function g(y) as an "arbitrary constant of integration" as far as the variable x is concerned.
and thus has the general form F(x . Hence g(y) = 2y 2 + C 1 . a general solution of the differential equation is defined implicitly by the equation 5 4 3 2 (31) (we have absorbed the constant C 1 into the constant C)..6.75. 2. then it is easily reduced by a simple substitution to a firstorder equation that may be solvable by the methods of this chapter. y'. y") = 0. If y is missing. p. If we can solve this equation for a general solution p(x. The other two special points in the figureat (0..7. (32) takes the form F(x. (32) If either the dependent variable y or the independent variable x is missing from a secondorder equation.12)are ones where both coefficient functions in Eq. C. y. (30) vanish. y'. The solution satisfying a given initial condition y(xo) = Yo is defined implicitly by Eq. Dependent variable y missing.70 Chapter 1 FirstOrder Differential Equations and it follows that g'(y) = 4y. and thus Therefore. Slope field and solution curves for the exact equation in Example 9.7 shows a rather complicated structure of solution curves for the differential equation of Example 9. C 1 ) involving an arbitrary constant C 1 . Reducible SecondOrder Equations A secondorder differential equation involves the second derivative of the unknown function y(x). (33) that involves two arbitrary constants C 1 and C2 (as is to be expected in the case of a secondorder differential equation).)dx + Cz to get a solution of Eq. (33) Then the substitution I dy p = y = dx' II y dp dx (34) results in the firstorder differential equation F(x. (31). p') = 0. the particular solution satisfying y(O) = 1 is defined implicitly by the equation 3x 2 y . For instance. 0 ~ ~~~~~~~~~ I • 2 3 4 5UL~L~~LLL~~~ X Remark: Figure 1. . with C determined by substituting x = x 0 and y = y 0 in the equation.3 does not guarantee a unique solution. then we need only write y(x) = J y'(x)dx = J p(x . . 0) and near (0. so the theorem of Section 1. • FIGURE 1.6.:.xy 3 + 2y2 = 2. y") = 0. then Eq.
dp that is.8.dy dx  dp dy (36) results in the firstorder differential equation for pas a function of y. Solution curves of the form y = x 2 +for c.1.dy = dyjdx f 1 dy = p f dy +C2. then (assuming that y' =!= 0) we need only write x(y) = f dx dy = dy f 1 . (32) takes the form F( y. we multiply by its integrating factor p = exp (J (2/x) dx) = e 21 n x = x 2 and get Dx(x 2 p) = 6x 2 . = c2 = 0). of the secondorder equation xy" + 2y' = 6x . ±3. y'..6. ± 20. X c. C 1) + C2 of our secondorder differential equation. • Independent variable x missing. y") = 0. The substitution defined in (34) gives the firstorder equation dp x dx + 2p = 6x. Solution curves with C 1 = 0 but c2 =I= 0 are simply vertical translates of the parabola y = x 2 (for which c. ± 35. Observing that the equation on the right here is linear. p==2x+. 1.8 shows this parabola and some typical solution curves with C2 = 0 but C 1 =I= 0. then Eq.~~nm 20 15 10 5 ~ 0~~~~~~~~~ 5 A final integration with respect to x yields the general solution y(x) X 2 =X c. If we can solve this equation for a general solution p(y .6. p(y . x 2 p = 2x 3 + C 1 .8. ±10.6 Substitution Methods and Exact Equations Example 10 Solution 71 Solve the equation xy" + 2y' = 6x in which the dependent variable y is missing. If x is missing.6. = 0. Solution curves with C 1 and C2 both nonzero are vertical translates of those (other than the parabola) shown in Fig. ++ C2 X FIGURE 1. C 1) involving an arbitrary constant C 1. the result is an implicit solution x (y) = P (y . dx +p x 2 = 6. dx x2 25~~~~nn. C 1) If the final integral P = J(ljp) d y can be evaluated. ±60. (35) Then the substitution I dy p = y = dx' y 11 p dp dp dy . . Figure 1. ±100. dy c.dx .
25. x 2 y' + 2xy = 5y 3 3 2 21.9. ± 1 are the horizontal lines y = 0.= p2 dy Then separation of variables gives f d: f d. The upper half of Fig. and then point out at the end that this restriction is unnecessary. The exponential curves withE> OandA = ±1 are in color.JXY x(x + y)y' = y(x xy 2y' = x 3 + y3 (x y) x 2y'=xy+i (x 2 . = ln p = ln y p = C1y +C (because y > 0 and p = y' > 0).1 0 X I 2 3 4 5 FIGURE 1.6. 26. 27. we see that we get solutions of yy" = (y') 2 . 24. 20. In particular. Hence dx 5~~~~nT~~~n~ 1 1 4 3 2 II== "" 0 1 I____.Y . 12. 18. 4. 6. 1. With B = 0 and different values of A. 5. 28. + J x2 + y2 yy' +x = Jx2 + y2 15. 11. those with B < 0 and A = ± 1 are black.3 .y 2 )y' = 2xy xyy' = y 2 + xJr 4x2::+y=2 2. which imply that the constants A and B are both positive. Primes denote derivatives with respect to x throughout. 3. where A = eCz and B = C 1• Despite our temporary assumptions.1 . x(x + y)y' + y(3x + y ) = 0 y' = Jx + y + 1 17. The solution curves y = Ae 8 x with B = 0 and A = 0. 29.2x = 2xy y 2(x y' + y)(l + x 4 ) 112 = x 3y2y' + y3 =ex 3xy 2 y' = 3x 4 + y 3 xeYy' = 2(eY + x 3 e 2x) (2x sin y cosy) y' = 4x 2 + sin2 y (x + eY )y' = xe. 7.9 shows the solution curves obtained with A = 1 (for instance) and different positive values of B. 9. y' = (4x + y) 2 (x + y )y' = 1 19. 30.. 2xyy' = x 2 + 2y 2 (x. y' = y + y 3 y y' + 2xy = 6x 23. The substitution defined in (36) gives the firstorder equation dp yp.2 .72 Chapter 1 FirstOrder Differential Equations Example 11 Solution Solve the equati~n yy" = (y') 2 in which the l~dependent variable xis missing. 10. ± 1. With A = 1 these solution curves are reflected in the xaxis. we get all horizontal lines in the plane as solution curves.54 . allowing both positive and negative possibilities for both y and y'. + y) y' = x . where C 1 = ec. 22.~ The resulting general solution of the secondorder equation yy" = (y') 2 is 2 3 4 5ww~~~~~~~~w .y)y' = x + y (x + 2y)y' = y x 2 y' = xy + x 2eYix xyy' = x 2 + 3y 2 13. xy' = y 14. We assume temporarily that y and y' are both nonnegative. 16.y xy' = y + 2. • Problems Find general solutions of the differential equations in Problems 1 through 30.6. and with negative values of B they are reflected in the yaxis. 1. we readily verify that y(x) = Ae 8 x satisfies yy" = (y') 2 for all real values of A and B. xy' + 6y = 3xy413 x 2y' + 2xy = 5y 4 2xy' + y 3 e. 8.
Show that the substitution v = ax + by + c transforms the differential equation dyjdx = F(ax +by+ c) into a separable equation. is called a Clairaut equation.1 0.1 0.18 61. given that y 1 (x) =xis a solution of each. Suppose that one particular solution y 1 (x) of this equation is known.6. The equation dyjdx = A(x)y2 + B(x)y + C(x) is called a Riccati equation. (x 39. for which g (y') y = xy' ~(y') 2 59. + tan. Explain why this implies that y = x 2 is a singular solution of the given Clairaut equation. Solve the differential equation X = ~ (y') 2 in Eq.6 Substitution Methods and Exact Equations In Problems 31 through 42. 67. 63. Show that the substitution v = y~ n transforms the Bernoulli equation dyjdx + P(x)y = Q(x)yn into the linear equation dv dx + (B + 2Ayt)v =A.x 3) 37. 41. (37). 34. This singular solution and the oneparameter family of straight line solutions are illustrated in Fig.. xy" = y' 44. Make an appropriate substitution to find a solution of the equation dyjdx = sin(x . y" = 2yy' 54. An equation of the form y (37) dv dx + (1. Use the method of Problem 63 to solve the equations in Problems 64 and 65. 2x5/2y2!3 dx + x3 y2 ~ 3 y5!3 _ 2x5f2 3x3/2y5/3 dy = 0 _ 1 _) are of the form x 3 + y 3 = 3Cxy.3y2) dx + (2y .v u +v . then solve it. Does this general solution contain the linear solution y(x) = x . Suppose that n "I 0 and n "I 1.x) dy = 0 (3x 2 + 2y 2 ) dx + (4xy + 6y 2 ) dy = 0 (2xy 2 + 3x 2 ) dx + (2x 2y + 4y 3) dy = 0 (x 3 + ~) dx + (y 2 + lnx)dy = 0 (1 + yexY) dx + (2y + xexY) dy = 0 + lny)dx + (~ + eY) dy = 0 dx 4x. yy" + (y') 2 = yy' 51. y 3y" = 1 53. 35. (cosx 38.n/2 that is readily verified by substitution in the differential equation? 62. y" = (x + y'f 49. y = v + k transform it into the homogeneous equation dv du u. y" + 4y = 0 46. 43.n)P(x)v(x) = (1. 73 31. yy" + (y') 2 = 0 45. by finding h and k so that the substitutions x = u + h. Assume x. dy 64. 56. x 2y" + 3xy' = 2 50. y and/or y' positive where helpful (as in Example 11 ).y). (4x. ~C 2 ). Show that the line x +y+3 is tangent to the parabola y = x 2 at the point (~C. Use the idea in Problem 57 to solve the equation dy 2 x . (2x+3y)dx+(3x+2y)dy=0 32. 36. 40.1.y) dx 33. verify that the given differential equation is exact.1 y) dx + x + y dy = 0 1 + y2 (3x 2y 3 + y 4 ) dx + (3x 3y 2 + y 4 + 4xy 3) dy = 0 (ex sin y + tan y) d x + (ex cos y + x sec2 y) d y = (2x . yy" = 3(y') 2 55. y(x) = Cx + g(C) (38) 58. xy" + y' = 4x 47. dx dy 65. y" = (y') 2 48. Show that the substitution v = ln y transforms the differential equation dyjdx + P(x)y = Q(x)(y In y) into the linear equation dvjdx + P(x) = Q(x)v(x). Show that the substitution y = YI +v 1 transforms the Riccati equation into the linear equation Find a general solution of each reducible secondorder differential equation in Problems 4354. Show that the solution curves of the differential equation dy dx y(2x 3 l) x(2y3 .n)Q(x) . 60. Show that the oneparameter family of straight lines described by 57. 1. y" = 2y(y') 3 52.3y.xz + dy = 0 y x4 0 2x5/2 _ 3 y5/3 42.4x y dx dy dx is a general solution of Eq. (37). dx + y2 = 1 + x 2 + 2xy = 1 + x 2 + y 2 = xy' + g(y') 66. Use the method in Problem 59 to solve the differential equation dy 2y x +7 + (6y. Consider the Clairaut equation + 2y In y = y.
Solutions of the Clairaut equation of Problem 67. suppose that a = 100 mi. (c) If w = 6 ftjs. which of course changes only by integral incrementsthat is. substitute p = y' to derive a general solution of the secondorder differential equation (with r constant) in the form y+ Jx + y 2 2 = 2(200x 9 ) 1110 • Thus a circle of radius r (or a part thereof) is the only plane curve with constant curvature 1/r. by one birth or death at a time. always heading toward a tree at (0. . (16) and (17). 0) on the west bank directly across from the dog's starting point.4 we introduced the exponential differential equation dPjd t = k P. • 8 (t) is the number of deaths per unit of population per unit of time at time t. the parabola at the point iC C!C. one learns that the curvature K of the curve y = y(x) at the point (x. A dog starts at (100. 7th edition (Upper Saddle River. (18) in this section from Eqs. y) is given by ly''(x)l K = ':::::(1 + y'(x)2]3/2.10. and w = 40 mijh.6. t + ~t] of length ~tis ~p = {births}  {deaths} ~ {J(t) · P(t) · ~t  o(t) · P(t) · ~t. iC 68. In the calculus of plane curves. 0) and swims at v0 = 4 ftjs. (a) If w = 2 ftjs. Derive Eq. In the situation of Example 7. The "typical" straight 2 is tangent to line with equation y = Cx2). It is customary to track the growth or decline of a population in terms of its birth rate and death rate functions defined as follows: • f3 (t) is the number of births per unit of population per unit of time at time t. A river 100ft wide is flowing north at w feet per second. with solution P (t) = P0 ekr. t + ~t] is given (approximately) by deaths: 8(t) · P(t) · ~t. show that the dog reaches instead the point on the west bank 50ft north of the tree. As in the text discussion. as a mathematical model for natural population growth that occurs as a result of constant birth and death rates.] Conversely. [t. 2008). 69. Here we present a more general population model that accommodates birth and death rates that are not necessarily constant.6 of Edwards and Penney. 150). Suppose that the population changes only by the occurrence of births and deathsthere is no immigration or emigration from outside the country or environment under consideration. Calculus: Early Transcendentals. and the plane begins at the point (200. Hence the change ~ P in the population during the time interval [t. v0 = 400 mijh. FIGURE 1. our population function P(t) will be a continuous approximation to the actual population. show that the dog never reaches the west bank. As before. show that the dog reaches the tree. Now how far northward does the wind blow the airplane? 70. (b) If w = 4 ft/s. NJ: Prentice Hall. D P'?pulation Models I n Section 1. show that its trajectory is described by and that the curvature of a circle of radius r is K = 1j r. however. 72. Then the numbers of births and deaths that occur during the time interval births: {J(t) · P(t) · ~t.74 Chapter 1 FirstOrder Differential Equations 71. suppose that an airplane maintains a heading toward an airport at the origin. If v0 = 500 mijh and w =50 mijh (with the wind blowing due north). [See Example 3 in Section 11.
then Eq. Eq.0005) P 2 in Example 1.B = .B = (0.. Slope field and solution curves for the equation dP/dt = (0.1. .·~iii·g. 1 . 0 10 20 30 40 50 FIGURE 1. But it also includes the possibility that f3 and 8 are variable functions of t. dt P(O) = 100 (with t in years). the direction field and solution curves shown in Fig.8. Then upon separating the variables we get f. Equation (1) is the general population equation.7.0005) Pand thus increases as the population does. If the birth rate is .0005)? .7 Population Models 75 so !1P !1t ~ [. p = 0. and then we readily solve for P(t) = 20.finite period of time. In particular. 1.B and 8 are constant.t.ti~~~~~b~~. and P = P(t) for brevity. Indeed. 8 = 8(t). If .= (0. they may well depend on the unknown function P(t). it appears that the • population always becomes unbounded in a . But we see that P + +oo as t + 20. so after 10 years the alligator population has doubled.B(t) . (1) in which we write .th. The birth and death rates need not be known in advance.7. Substitution oft 2 dP = f (0. ( 1) gives the initial value problem dP 2 .8(t)] P(t).1..1 indicate that a population explosion always occurs. sotaking the limitwe get the differential equation Tt dP = (... P (1 0) = 2000 j 10 = 200. (1) reduces to the natural growth equation with k = . For instance.t~~p~p~i. lllt'·'l·'·'tsl' ··s~pp~.~ ·th~t· .~at~is 8 = 0 (so none of the alligators is dying).8.:· ~~ctrh~tits d~.B .8)P.= (0.0005) =2000 dt. The error in this approximation should approach zero as !1t + 0.0005)t +C. P = 100 gives C 1/100. so a real "population explosion" occurs in 20 years. whatever the size of the (positive) initial population P(O) = P0 .1ooi~iti~ii.B(t).
.f3t P .o6r ___ = ±ec eo.Po).{3 1 P.0004P(150. (2) that is.0004P 2 .~" · '•·· . it is useful to rewrite the logistic equation in the form dt where k m"''''""" " .PI= 0. • · . ..p 150.0.P). (2) is called the logistic equation. so that f3 = {30 .0._._. this equation is easy to solve for the population 150Po P(t) = .Po Finally. .. Hence [where B = ±ec ].· •··•• ""' "' •··. 0. then Eq. we find that Poe0.~ dP = kP(M.76 Chapter 1 FirstOrder Differential Equations Bounded Populations and the Logistic Equation In situations as diverse as the human population of a nation and a fruit fly population in a closed container.p p f f # 0. ...h. dt (4) To solve this differential equation symbolically.Po )e. We get 1 150 f (_!__ + p f dP = P(150.8o)P. then Eq. = Po p 150 into this last equation. Suppose.06t 150. that the birth rate f3 is a linear decreasing function of the population size P. In IPI In 1150. • ··• •.u.=aPbP dt ' where a= f3o. we separate the variables and integrate.•. ·•.8o and b = f3t· If the coefficients a and b are both positive. " " ' . where f3o and {3 1 are positive constants.0004dt .. . ." """ .. ~ ... The reasons may range from increased scientific or cultural sophistication to a limited food supply.P) = 0._. Example 2 In Example 4 of Section 1.0004dt [partial fractions] . If the death rate 8 = 80 remains constant.3 we explored graphically a population that is modeled by the logistic equation  dP = 0.Po+ (150 ..o6r = If we substitute t = 0 and P B = Po/ (150 . for example. For the purpose of relating the behavior of the population P(t) to the values of the parameters in the equation. dP 2 .p + C. (1) takes the form dP dt = (f3o.. it is often observed that the birth rate decreases as the population itself increases.P) 1 ) dP = 150. (3) = b and M = ajb are constants.06r (5) .06r 150. B eo.06P..
it approaches the finite limiting population M as t ~ +oo.< .0004P 2 • The finite limiting population noted in Example 2 is characteristic of logistic populations. Typical solution curves for the logistic equation P' = kP(M. if Po > M. If Po = M.Po)ekMr .0. Po + 0 (8) FIGURE 1. but steadily decreases and approaches M from above if Po > M.2 shows a number of solution curves corresponding to different values of the initial population ranging from Po = 20 to Po = 300. Typical solution curves for the logistic equation P' = 0. Po However. Indeed. the behavior of a logistic population depends on whether 0 < Po < M or Po> M. whatever the initial value Po > 0. P(O) = Po (6) P(t) = MP0 Po + (M .7 Population Models p 77 300 240 1801 150r 120 60 :::=§3111~!!!!!!!~==== at time t in terms of the initial population Po = P (0). Example 3 . Assume that this population satisfies the logistic equation.Po)ekMt (7) Actual animal populations are positive valued. you should be able to see directly from Eq. then we see from (6) and (7) that P' < 0 and P(t) = p MPo Po + (M . 7.3. It follows that lim P(t) = t + +oo MPo = M.> MPo Po + {neg. Note that all these solution curves appear to approach the horizontal line P = 150 as an asymptote. In Problem 32 we ask you to use the method of solution of Example 2 to show that the solution of the logistic initial value problem dt IS dP = kP(M. the population P(t) steadily increases and approaches M from below ifO < Po< M.) Thus a population that satisfies the logistic equation does not grow without bound like a naturally growing population modeled by the exponential equation P' = k P.P). Sometimes M is called the carrying capacity of the environment. IfO < Po< M.06P. Suppose that in 1885 the population of a certain country was 50 million and was growing at the rate of 750.2. As illustrated by the typical logistic solution curves in Fig. 1.7. (See Problem 34. • 20 Limiting Populations and Carrying Capacity 25 50 r~~~~1 75 100 FIGURE 1. (5) that lim1+oo P(t) = 150. Po + MPo {pos. Suppose also that in 1940 its population was 100 million and was then growing at the rate of 1 million per year.Po)ekMt . then (7) reduces to the unchanging (constantvalued) "equilibrium population" P(t) M. Each solution curve that starts below the line P = M/2 has an inflection point on this line.3. considering it to be the maximum population that the environment can support on a longterm basis.. Instead. Figure 1.7. Otherwise. Determine both the limiting population M and the predicted population for the year 2000. number} M MPo Po  · In either case.7.1.000 people per year at that time.P). the "positive number" or "negative number" in the denominator has absolute value less than Po andbecause of the exponential factorapproaches 0 as t + +oo. number} MPo =M. then we see from (6) and (7) that P' > 0 and = P(t) = MPo Po+ (M.
78 Chapter 1 FirstOrder Differential Equations Solution We substitute the two given pairs of data in Eq. If we take Po = 5.212 (for 1900) in the logistic model formula in Eq.212 = 5..oooi)(200)(60) ' 100. Verhulst as a possible model for human population growth. P = 76.192 million.308)e(0. (7). we find that 76. population census data.121.S.03I55I )r (10) .192 (for 1850) and t = 100.308 + (M. .50). population during the 19th century is P(t) = (5.308 and substitute the data pairs t =50. F.7% per year. . (7)the population in the year 2000 will be P(60) = ::::====:::::::100 + (200.000167716.100)e< o. (7) yields the logistic model 998.026643)r (9) (with t in years and P in millions). (3) and find that 0. and with t = 0 corresponding to the year 1940 (in which Po = 100).(0. We solve simultaneously forM = 200 and k = 0. Thus the limiting population of the country in question is 200 million. Nonlinear systems like this ordinarily are solved numerically using an appropriate computer system.308e 100r. Thus our natural growth model for the U. M = 188. • Example 5 The U. Example 4 The U.75 = 50k(M.308 ~ 0.5.308)eIOOkM in the two unknowns k and M.100).00 = lOOk(M.S. With these values of M and k. we find thataccording to Eq. Because e 0·026643 ~ 1.308)e .02700. Substitution of these values in Eq.308 million and in 1900 was 76. But with the right algebraic trick (Problem 36 in this section) the equations in (1 0) can be solved manually for k = 0.813)e.026643. P = 76. population in 1800 was 5.212 5.0001.308 (with t = 0 in 1800) in the natural growth model P(t) = Poerr and substitute t = 100. • Historical Note The logistic equation was introduced (around 1840) by the Belgian mathematician and demographer P. then compare projections for the 20th century.7 million people.212 r = ln 100 5. 1. If we take Po= 5.192 5.S.546 (11) P(t) = · 5. In the next two examples we compare natural growth and logistic model fits to the 19thcentury U.308)M = 76.308 + (M. population in 1850 was 23.212 million.S. the average population growth between 1800 and 1900 was about 2.212. so 1 76. P = 23. we get the two equations (5308 )M = 23.308 + (182.50kM ' (5.200 about 153.5.
805 15.638 12.S.492 221.253 34.113 26. it is customary to define the average error (in the model) as the square root of the average of the squares of the individual errors (the latter appearing in the fourth and sixth columns of the table in Fig.849 169.000 0.4).237 288.593 0.972 25. ·.000 1.212 92.387 76. By the end of the 20th century the exponential model vastly overestimates the actual U.S.326 50.620 288.354 154.721 6.437 0.394 0. In order to measure the extent to which a given model fits actual data.079 5.052 161.308 7. the exponential error is considerably larger during the 19th century and literally goes off the chart during the first half of the 20th century.202 9.001 1.710 281.316 173.4 compares the actual 18001990 U.251 23.203 132.·.252 177.192 30.323 203.501 23.545 0. Both agree well with the 19thcentury figures.7. .240 9. while the average error in the logistic model is only 0. .459 4. But the exponential model diverges appreciably from the census data in the early decades of the 20th century.• Exponential .943 492.038 Logistic Error 0.271 41.422 5.834 132.4.730 58.5.192 31.155 0.594 1. Consequently.452. 1. populationpredicting over a billion in the year 2000whereas the logistic model somewhat underestimates it.'.7. ' odeli .780 376. even in 1900 we might well have anticipated that the logistic model would predict the U.818 Logistic Model 5.990 168.312 58.023 642.929 9.308 1094.861 17.228 106. Comparison of exponential growth and logistic models with U. where plots of their respective errorsas a percentage of the actual populationare shown for the 18001950 period.290 5.827 46.598 812.694 589.165 151.289 89.212 99.226 76.S. census population figures with those predicted by the exponential growth model in (9) and the logistic model in (11).000 7. '.479 129.022 123.834 105.213 90.980 76.056 1.~ . • .0.369 . We see that the logistic model tracks the actual population reasonably well throughout this 150year period.097 0.7.612 119. 1.044 11.384 Error 1800 1810 1820 1830 1840 1850 1860 1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 5.443 38.034 62.409 20.410 3.7 Population Models 79 The table in Fig.'.735 13.308 6. However.038 0.458 104.886 144.072 .435 76.162 for the average error in the exponential model.<::. The two models are compared in Fig.~~~~btlaJ(.S.064 23. FIGURE 1. census populations (in millions). population growth during the 20th century more accurately than the exponential model.655 3.095 17.234 0.189 62.236 838. 1.768 0.558 50.268 44.311 0.721 415.454 197.1. Using only the 18001900 data. whereas the logistic model remains accurate until 1940.000 0.5.:t::~::::: '.326 179.7. Error 0.137.542 248.308 7.7.190 4.240 FIGURE 1. • ·.. this definition gives 3.038 0. Percentage errors in the exponential and logistic population models for 18001950.302 226.405 39.
Limited environment situation. Then P' (t) should be proportional to the product of the number P of individuals having the disease and the number M . in which all deaths result from chance encounters between individuals. 3.P of those not having it. then . and therefore dPjdt = kP(M . when will 80% of the city's population have heard the rumor? Substituting Po= 10 and M = 100 (thousand) in Eq. Then f3 . It is then reasonable to expect the growth rate f3 . The mathematical description of the spread of a rumor in a population of M individuals is identical. we get 1000 P(t) = 10 + 90e. If the birth rate f3 is constant but the death rate 8 is proportional to P. Assuming that P(t) satisfies a logistic equation.P). Then substitution oft (12) Solution = 1. Of course.P as the potential for further expansion. More Applications of the Logistic Equation We next describe some situations that illustrate the varied circumstances in which the logistic equation is a satisfactory mathematical model.80 Chapter 1 FirstOrder Differential Equations The moral of Examples 4 and 5 is simply that one should not expect too much of models that are based on severely limited information (such as just a pair of data points). The classic example of a limited environment situation is a fruit fly population in a closed container.8 (the combined birth and death rates) to be proportional to M . Competition situation.P). because we may think of M . so that dP dt = ({3. 10 thousand people in a city with population M = 100 thousand people have heard a certain rumor. Again we discover that the mathematical model is the logistic equation. A certain environment can support a population of at most M individuals.JOOkr. The disease is spread by chance encounters. competition between individuals is not usually so deadly. Example 6 Suppose that at timet = 0. so that 8 = a P.P). Let P(t) denote the number of individuals in a constantsize susceptible population M who are infected with a certain contagious and incurable disease. nor its effects so immediate and decisive. P = 20 gives the equation 20 = 1000 10 + 90e.8)P = kP(M.8 = k(M . (7). Much of the science of statistics is devoted to the analysis of large "data sets" to formulate useful (and perhaps reliable) mathematical models. 2. This might be a reasonable working hypothesis in a study of a cannibalistic population. 1. Joint proportion situation.= dP dt ({3.aP)P = kP(M P).100k .P. After 1 week the number P(t) of those who have heard it has increased to P(1) = 20 thousand.
With P(t) = 80.06P. p ___ = ±ece.. lOOk In 4 thus after about 4 weeks and 3 days. We will see that the constant M is now a threshold population.150) (~p 1 p.150) = 0.. (14) We want to find P(t) if (a) P(O) = 200.9 ~ 4.42.0004dt .0004P(P. In IPIIn IP.150 ) dP =J 0.1.~ . We get f 1 .150 [where B = ±ec ].06t + C. Example 7 Consider an animal population P(t) that is modeled by the equation dP dt = 0. Note that the righthand side in Eq. with k constant). (b) P(O) = 100.0.0. (13) is the negative of the righthand side in the logistic equation in (3). then the general population equation in (1) yields the differential equation dP 2 . we separate the variables and integrate. (12) takes the form 1000 80=10 + 90elOOkt' which we solve for elOOkt = 3~. with the way the population behaves in the future depending critically on whether the initial population Po is less than or greater than M.7 Population Models 81 that is readily solved for eIOOk = ~.= kP .M) dt (13) (where M = 8/k > 0) as a mathematical model of the population. Solution To solve the equation in (14). Eq.= . • Doomsday versus Extinction Consider a population P(t) of unsophisticated animals in which females rely solely on chance encounters to meet males for reproductive purposes.06r p .. If the death rate 8 is constant. (15) . · ~ .J 150 dP = J0.0004dt [partial fractions]. hence at a rate proportional to P 2 • We therefore assume that births occur at the rate k P 2 (per unit time.06r = Be.8P = kP(P.008109.1501 = 0. P(P. so k = Ibo In ~ ~ 0. The birth rate (births/ time/population) is then given by f3 = kP. It is reasonable to expect such encounters to occur at a rate that is proportional to the product of the number P/2 of males and the number P /2 of females.0004P 2  0. It follows that 80% of the population has heard the rumor when In 36 In 36 t = .
as t increases and approaches T = ln(4)/0. depending on whether or not its initial size exceeds the threshold population M = 150. the positive denominator on the right in (16) approaches +oo.13). Use either the exact solution or a computergenerated slope field to sketch the graphs ofseveral solutions of the given differential equation.x 2 .  1 Note that.4x 2 . x(O) = 2 dx 7. as t increases without bound. dt = 3x(5.06t · (17) Note that. dt dx = x. P(t) + 0 as t + +oo. the positive denominator on the right in (16) decreases and approaches 0.  dx dt = ?x(x. (15) for 30oe0·061 300 P(t).x 2 . With this value of B we solve Eq. Consequently. this equilibrium situation is very unstable.x 2 x(O) = 3 4. Figure 1. (13). (b) If Po = 100 and after 6 months there are 169 fish in the lake.x).= 1. then the population remains constant. This is a doomsday situationa real population explosion. and highlight the indicated particular solution.6. how many will there be after 1 year? . (15) for 4e.105. Consequently P(t) + +oo as t + r. Thus the population in Example 7 either explodes or is an endangered species threatened with extinction. If Po exceeds M (even slightly). how long did it take all the fish in the lake to die? 11. If Po = M (exactly!). . x(O) = 17 9. This is an (eventual) extinction situation. x(O) = 2 2.2 + e0.2e0.0·061 oo61 . with the result that the fish cease to reproduce (so that the birth rate is f3 = 0) and the death rate 8 (deaths per week per fish) is thereafter proportional to 1/v'P. the birth and death rates f3 and 8 are both inversely proportional to v'P.~. At time t = 0 (months) where k is a constant.82 Chapter 1 FirstOrder Differential Equations (a) Substitution oft = 0 and P = 200 into (15) gives B = 4. such as the alligator population in certain areas of the southern United States. dt ' dt dx 5. However. dt=4x(7x). whereas if the initial (positive) population is less than M (however slightly).. x(O) = 0 = 3x(x  5).7. x(O) = 1 dx dx 3.M).7. See Problem 33. An approximation to this phenomenon is sometimes observed with animal populations. (a) Show that Separate variables and use partial fractions to solve the initial value problems in Problems 18. (b) Substitution oft = 0 and P = 100 into (15) gives B = 2. How many rabbits will there be one year later? 10. then it decreases (more gradually) toward zero as t + +oo.06t + 1 . dt = dx lOx.6 shows typical solution curves that illustrate the two possibilities for a population P(t) satisfying Eq. • p FIGURE 1.· 60oe. If there were initially 900 fish in the lake and 441 were left after 6 weeks. dt = 9. Typical solution curves for the explosion/extinction equation P' = kP(P . Suppose that the fish population P(t) in a lake is attacked by a disease at time t = 0. 1.x(0)=11 8. x(O) = 8 dx 6. B Problems the population numbers 100 rabbits and is increasing at the rate of 20 rabbits per month. The time rate of change of a rabbit population P is proportional to the square root of P. P(t) = With this value of B (16) we solve Eq. then P(t) rapidly increases without bound. Suppose that when a certain lake is stocked with fish.06 >:::::: 23.
Consider an alligator population P (t) satisfying the extinction/explosion equation as in Problem 18. Eq. and that the number of those who have heard it is then increasing at the rate of 1000 persons per day. 19. the number x(t) of grams of the salt in a solution after t seconds satisfies the differential equation dxjdt = 0.1. If the initial population is P (0) = P0 . Predict this country's population for the year 2000. The swamp contained a dozen alligators in 1988. Consider a rabbit population P (t) satisfying the logistic equation as in Problem 15.7. show that the threshold population isM = DoP0/B 0 .h) 2h with h = 1 to estimate the values of P'(t) when P = 25. where B = aP is the time rate at which births occur and D = b P 2 is the rate at which deaths occur. At time t = 0 the number N (t) of people who have developed Michaud's syndrome is 5000 and is increasing at the rate of 500 per day. Suppose that a community contains 15.04 FIGURE 1. The time rate of change of an alligator population P in a swamp is proportional to the square of P. Consider an alligator population P(t) satisfying the extinction/explosion equation as in Problem 18. (b) Suppose that P0 = 6 and that there are nine rabbits after ten months. f3 and 8. Consider a prolific breed of rabbits whose birth and death rates. How long will it take for this rumor to spread to 80% of the population? (Suggestion: Find the value of k by substituting P(O) and P' (0) in the logistic equation. If P(O) = 100 and P' (O) = 8. how many months does it take for P(t) to reach 10 times the threshold population M? 20. 26. are each proportional to the rabbit population P = P(t). 1.P(t.38 1974 1975 1976 47.0.00 25. Ifthe initial population is 120 rabbits and there are 8 births per month and 6 deaths per month occurring at time t = 0. (Suggestion: Take t = 0 to correspond to the year 1925.) (b) Use the values of k and M found in part (a) to determine when P = 75.000 persons have heard a certain rumor. Ifthe initial population is 110 alligators and there are 11 births per month and 12 deaths per month occurring at time t = 0. Assume that N ' (t) is proportional to the product of the numbers of those who have caught the disease and of those who have not. When does doomsday occur? 14.) 23. a contagious disease. 16.bP 2 .7.P) with k constant. and B0 births per month and Do deaths per month are occurring at time t = 0. Consider a rabbit population P (t) satisfying the logistic equation as in Problem 15. How long will it take for another 5000 people to develop Michaud's syndrome? 25. If the initial population is P (0) = Po and B0 births per month and Do deaths per month are occurring at time t = 0.00 and when P = 47. how many months does it take for P (t) to reach 10% of the threshold population M? 21. show that the limiting population is M = BoPo/Do. Its population in 1940 was 100 million and was then (a) What is the maximum amount of the salt that will ever dissolve in the methanol? (b) If x = 50 when t = 0. two dozen in 1998.bP. Repeat part (a) of Problem 13 in the case f3 < 8. (a) Show that 83 growing at the rate of 1 million per year. Note that P(t) + +oo as t + 1j (kP0 ). Consider a population P(t) satisfying the logistic equation dP/dt = aP. with f3 > 8. A population P(t) of small rodents has birth rate f3 (O. (3).000 people who are susceptible to Michaud's syndrome.OOI)P (births per month per rodent) and constant death rate 8.63 25.54.8x . Population data for Problem 25. The data in the table in Fig. how long (in . If the initial population is 100 alligators and there are 10 births per month and 9 deaths per months occurring at time t = 0.) Year 1924 1925 1926 P (millions) 24.004x 2 • P (t)  R 0 1kPot' k constant.54 48. Consider a population P(t) satisfying the extinctionexplosion equation dPjdt = aP 2 . When will there be four dozen alligators in the swamp? What happens thereafter? 13. how many months does it take for P (t) to reach 105% of the limiting population M? 18. Suppose that the population P(t) of a country satisfies the differential equation dP/dt = kP(200 . (a) What is the limiting population M? (Suggestion: Use the approximation p (t) ~ I P(t +h). half of a "logistic" population of 100. What now happens to the rabbit population in the long run? 15.7 are given for a certain population P(t) that satisfies the logistic equation in (3).04 47.7 Population Models 12. how many months does it take for P (t) to reach 95% of the limiting population M? 17. 22. If the initial population is 240 rabbits and there are 9 births per month and 12 deaths per month occurring at time t = 0. Suppose that at time t = 0. where B = aP 2 is the time rate at which births occur and D = b P is the rate at which deaths occur.7. As the salt KN0 3 dissolves in methanol. This is doomsday. Then substitute these values in the logistic equation and solve fork and M . how long will it take for an additional 50 g of salt to dissolve? 24.
symbolically by analyzing the solution given in Eq.M). and 1960. or numerically by substituting successive values of t. Use the method of Problem 36 to fit the logistic equation to the actual U. and hence dP dt = f3oe . (3).7.2 million.9 million to 123. 38. 1930. population through the year 1930 (long after his own death. With M now known.= 0. it follows that any solution curve that crosses the line P = ~ M has an inflection point where it crosses that line.S. Birth and death rates of animal populations typically are not constant. Make it clear how your derivation depends on whether 0 < P0 < M or Po > M . population P(t) (tin years) grew from 3. P1. A tumor may be regarded as a population of multiplying cells. solve this differential equation to determine what happens to the alligator population in the long run. 29.0001x 2 .0001489P 2 . population data (Fig. and 1950.JOOkM.e. and 2000. Upon equating the two resulting expressions for x 2 in terms of M . Consider two population functions P1(t) and P2 (t). except with 150 alligators initially. 1900. Observe that P (t) approaches the finite limiting population P0 exp ({30/a) as t + + oo. (a) Derive the solution p (t) months) will it take this population to double to 200 rodents? (Suggestion: First find the value of 8. 1. (b) Repeat part (a). the U. P(O) = P0 • = (b) How does the behavior of P(t) as t increases depend on whether 0 < Po < M or Po > M ? 34.01 (deaths per animal per month) and with birth rate f3 proportional to P.7. P (t) remained close to the solution of the initial value problem = MPo Ro_+_(_M__::__Po _)_ ek ::Mr of the extinctionexplosion initial value problem P' kP(P. and Pz corresponding to equally spaced times to = 0. suppose that at time t = 0 there are Po = 106 cells and that P(t) is then increasing at the rate of 3 x 105 cells per month.0. 35.) 27. and P" > 0 if P > M. dt (a) What 1930 population does this logistic equation predict? (b) What limiting population does it predict? (c) Has this logistic equation continued since 1930 to accurately model the U. Solve the resulting logistic equation. This technique can be used to "fit" the logistic equation to any three population values P0 .M). dP . t 1.4) for the years 1850. 32. both of which satisfy the logistic equation with the same limiting population M but with different values k 1 and k 2 of the constant k in Eq. (7). 31. Solve this initial value problem for Conclude that P" > 0 if 0 < P < ~M.S. For the tumor of Problem 30. so that f3(t) = f30 ear (where a and {3 0 are positive constants).P). dP . Derive the solution P(t) = Po + (M  MPo Po)ekMt of the logistic initial value problem P' = kP(M .03135P.0.9. begin by solving the first equation for the quantity x = eSO kM and the second equation for x 2 = e. Suppose that the number x(t) (with tin months) of alligators in a swamp satisfies the differential equation dx jdt = 0. If P(t) satisfies the logistic equation in (3). P(O) = Po. they vary periodically with the passage of seasons.S. population data (Fig.01x. population? [This problem is based on a computation by Verhulst. Assume that k 1 < k2 • Which population approaches M the most rapidly? You can reason geometrically by examining slope fields (especially if appropriate software is available).4) for the years 1900.3. Find P (t) if the population P satisfies the differential equation P(t) = P0 exp ( ~0 (1 . During the period from 1790 to 1930.S. P(O) = 3. 36.S.= (k dt + b cos27Tt)P. .84 Chapter 1 FirstOrder Differential Equations 33. and therefore resembles one of the lowerSshaped curves in Fig. 1990.] 30. (a) If initially there are 25 alligators in the swamp.S. P(O) = P0 . Solve the resulting logistic equation and compare the predicted and actual populations for the years 1990 and 2000.7. Suppose that P (0) = 200 and P' (0) = 2. Consider an animal population P(t) with constant death rate 8 = 0. Throughout this period. who in 1845 used the 17901840 U. either of the original equations is readily solved fork. 39. use the chain rule to show that P"(t) = 2e P(P ~M)(P. and t2 = 2t1 • 37.ar P . 1. you get an equation that is readily solved for M . P" = 0 if P = ~M. instead. of course). It is found empirically that the "birth rate" of the cells in a tumor decreases exponentially with time. To solve the two equations in ( 10) for the values of k and M. and then find the limiting population of the tumor. Fit the logistic equation to the actual U. population data to predict accurately the U. Solve numerically for a. In particular. After 6 months the tumor has doubled (in size and in number of cells). then compare the predicted and actual populations for the years 1980. P" < 0 if ~M < P < M.ar)) . (a) When is P = 1000? (b) When does doomsday occur? 28. 1.
+ 49 = 0.8.· · . How would the two populations compare after the passage of many years? In Section 1. p = 1 for relatively . (1) with g = 9. g ~ 32 ftjs 2 in fps units). The bolt returns to the ground when y seconds aloft.= 9.8)t + 49] dt = (4. Generally speaking.8 mjs2 (in mks units.8)t + 49.. The force F R exerted by air resistance on the moving mass m must be added in Eq.8 gives  ··· . = (4. Thus the growthrate function r(t) = k + b cos 2rrt varies periodically about its mean value k.8 AccelerationVelocity Models where t is in years and k and b are positive constants..9)t 2 + 49t + y 0 = (4.9)(5 2 ) + (49)(5) = 122.10) = 0. (1).9)t(t.2 we discussed vertical motion of a mass m near the surface of the earth under the influence of constant gravitational acceleration. Thus its maximum height is Ymax = y(5) = (4.8)t + v0 = (9.8)t t = 5 (s). ····· "" ·· ··     ···   ···· . and thus after 10 • Now we want to take account of air resistance in a problem like Example 1.=Fe+ FR. dt (2) Newton showed in his Principia Mathematica that certain simple physical assumptions imply that F R is proportional to the square of the velocity: F R = kv 2 • But empirical investigations indicate that the actual dependence of air resistance on velocity can be quite complicated.e dt ' (1) where Fe = mg is the (downwarddirected) force of gravity. Hence the bolt's height function y(t) is given by y(t) = J[ (9. Then Eq.9)t2 + 49t. where the gravitational acceleration is g ~ 9. Suppose that a crossbow bolt is shot straight upward from the ground (y0 = 0) with initial velocity v0 = 49 (m/s). If we neglect any effects of air resistance. dv dt so v(t) = (9. then Newton's second law (F = ma) implies that the velocity v of the mass m satisfies the equation dv m. . so now dv m.5 (m). hence when The bolt reaches its maximum height when v = (9.1. For many purposes it suffices to assume that where 1 ~ p ~ 2 and the value of k depends on the size and shape of the body.  Example 1 ·.= F. as well as the density and viscosity of the air. Construct a graph that contrasts the growth of this population with one that has 85 the same initial value Po but satisfies the natural growth equation P' = kP (same constant k).
it approaches afinite limiting speed. the net force acting on the body is then F = FR +Fa= kv. = lim v(t) = _ §__ t~oo p (6) Thus the speed of a body falling with air resistance does not increase indefinitely. subject to two forces: a downward gravitational force Fa and a force FR of air resistance that is proportional to velocity (so that p = 1) and of course directed opposite the direction of motion of the body. (3) makes FR positive (an upward force) if the body is falling (vis negative) and makes FR negative (a downward force) if the body is rising (vis positive). Thus air resistance is a complicated physical phenomenon. Vertical motion with air resistance. As indicated in Fig.1.) L.mg. 1. then Fa= mg and FR = kv. (3) m m (Note: F R acts upward when the body is falling. You should verify for yourself that if the positive yaxis were directed downward. (4) would take the form dvjdt = pv +g. fo><oF • F. or terminal speed. If we set up a coordinate system with the positive ydirection upward and with y = 0 at ground level. Note that the minus sign in Eq. where k is a positive constant and v = dyfdt is the velocity of the body. dt (4) where p = kf m > 0. and Newton's law of motion F = m(dvjdt) yields the equation dv mdt = kv mg. and its solution is v(t) = (vo + ~) e. Note that v. yields a tractable mathematical model that exhibits the most important qualitative features of motion with resistance.1.~· (5) Here.pt. But how slow "low speed" and how fast "high speed" are depend on the same factors that determine the value of the coefficient k.8. instead. then Eq.g. whereas 1 < p < 2 for intermediate speeds.. Thus  dv = pv.8.+ Fa Ground level FIGURE 1. But the simplifying assumption that FR is exactly of the form given here. Resistance Proportional to Velocity Let us first consider the vertical motion of a body with mass m near the surface of the earth.86 Chapter 1 FirstOrder Differential Equations low speeds and p = 2 for high speeds. (7) . Equation (4) is a separable firstorder differential equation. with either p = 1 or p = 2. v0 = v(O) is the initial velocity of the body.
5 meters without air resistance). y(t) = 7350.245 = 0 for tm = 25ln(294/245) : : : : 4. We substitute y0 obtain Solution = 0. We now rewrite Eq. We . (5) and (9). v0 = 49. + v.tf25 . I : : : : 21. Thus we find that C =Yo+ (vo. we must solve the equation y(t) = 7350.5. The two equations also involve the terminal velocity v.04 in Eq.7350etf 25 • To find the time required for the bolt to reach its maximum height (when v = 0). Example 2 We again consider a bolt shot straight upward with initial velocity v0 = 49 mj s from a crossbow at ground level.245t. With an unbuttoned overcoat flapping in the wind in place of a parachute. The formulas depend on the initial height y0 of the body.v. it even helps explain the occasional survival of people who fall without parachutes from highflying airplanes. pr (9) Equations (8) and (9) give the velocity v and height y of a body moving vertically under the influence of gravity and air resistance.( vo .3 ftjs. about 44 mijh. To find when the bolt strikes the ground. an unlucky skydiver might increase p to perhaps as much as 0.1125 = 0.v. which corresponds to a terminal speed of Iv.5. We ask how the resulting maximum height and time aloft compare with the values found in Example 1. and v. its initial velocity v0 .y(tn)/y' (tn) to generate successive approximations to the root. or about 14. But now we take air resistance into account. we solve the equation v(t) = 294e. (5) in the form dy pr . and v(t) = 294etf25  245. (8) Integration gives 1 y(t) = (vo. the constant such that the acceleration due to air resistance is aR = pv. Its maximum height is then Ymax = v(tm) : : : : 108.8 AccelerationVelocity Models 87 This fact is what makes a parachute a practical invention. )e dt + v. Or we can simply use the Solve command on a calculator or computer.)eP' p We substitute 0 fort and let y 0 = y(O) denote the initial height of the body. See Problems 10 and 11 for some parachutejump computations.245t. (4). For a person descending with the aid of a parachute.t 1 + (vov. with p = 0. )(1 p e.v. we can begin with the initial guess to = 10 and carry out the iteration tn+l = tn . Using Newton's method.)jp. defined in Eq. a typical value of p is 1. = gjp = 245 in Eqs. and so y(t) =Yo+ v.7350e. which gives a terminal speed of Iv..5 mi/h. (6). and the drag coefficient p..280 meters (as opposed to 122.1.558 (s). I : : : : 65 ft/ s.t +C.).
ln I cos u I + C .tm ~ 4.g ( 1 + ~ v2 ) • (12) In Problem 13 we ask you to make the substitution u = v v'P7i and apply the familiar integral 1 du = tan.853 s) than in ascent Ctm ~ 4. Then Eq./Pi) cos Ct (14) .. Then Newton's second law gives dv mdt (10') = Fa+ FR = mg.88 Chapter 1 FirstOrder Differential Equations find that the bolt is in the air for tt ~ 9. (11) with v > 0 gives the differential equation ~~ = . Thus the effect of air resistance is to decrease the bolt's maximum height. Taking the positive ydirection as upward. Note also that the bolt now spends • more time in descent (tf.g  p v2 = .kvlvl.227 mjs (as opposed to its initial velocity of 49 mjs). and its final impact speed. It hits the ground with a reduced speed of lv(tf)l ~ 43. Resistance Proportional to Square of Velocity Now we assume that the force of air resistance is proportional to the square of the velocity: (10) with k > 0. (11) where p = kfm > 0. so we can rewrite Eq.1 u + C 2 l+u to derive the projectile's velocity function f v(t) =/!tan (C 1  tv'iJi) with C1 = tan. The choice of signs here depends on the direction of motion. the total time spent aloft. Thus the sign of FR is always opposite that of v.1 ( vo/f) .411 seconds (as opposed to 10 seconds without air resistance). FR < 0 for upward motion (when v > 0) while FR > 0 for downward motion (when v < 0).558 s). that is. UPWARD MOTION: Suppose that a projectile is launched straight upward from the initial position y 0 with initial velocity v0 > 0. (13) Because J tan u d u = . dt = dv g . a second integration (see Problem 14) yields the position function y (t) = Yo + 1 ln p ''~ COS ( C 1  t .pvlvl. (10) as FR = kvlvl. which the force of resistance always opposes. We must discuss the cases of upward and downward motion separately.
. Then Eq. ~ · ·. . another integration (Problem 16) yields the position function y(t) =Yo.g + pv2=  g (1.0011 in Eqs..5 108.85 4.04)v (0. so v(t) = ..41 9.t v'fi8) cosh c2 (17) (Note the analogy between Eqs.J{i[i and apply the integral 1 d u = tanh . 0.49 .t.:. with p = 0. 0.00ll)v 2 122.. then C 2 = 0. (11) with v < 0 gives the differential equation dv dt = .) Ifvo = 0. (15) In Problem 15 we ask you to make the substitution u = v..gv P 2) .8 AccelerationVelocity Models 89 DOWNWARD MOTION: Suppose that a projectile is launched (or dropped) straight downward from the initial position Yo with initial velocity vo . In Problems 17 and 18 we ask you to verify the entries in the last line of the following table.61 5 4.. Example 3 . .l u + C I 1u2 to derive the projectile's velocity function v(t) =!!tanh ( c2.80 49 43.· ·· ·· ·············· We consider once more a bolt shot straight upward with initial velocity v0 = 49 m j s from a crossbow at ground level. it follows that in the case of downward motion the body approaches the terminal speed (18) (as compared with tv. Because sinhx coshx l (ex .47 10 9. But now we assume air resistance proportional to the square of the velocity.0 (0..Jg7/)tanh(tv7J8). (4)) . ...ln p 1 cosh ( C2 . (16) and (17) and Eqs.Jiii) with (16) Because f tanh u d u = ln I cosh u I + C. as in Example 2..41 5 4.ex ) lim tanhx = lim x+oo x+ oo = lim x+ oo 2 !Cex +ex ) = 1.1. (12) and (15). I = gjpin the case of downward motion with linear resistance described by Eq...56 4.28 108.23 43. (13) and (14) for upward motion.
provide a deceleration of T = 4 rn/s 2 • At what height above the lunar surface should the retrorockets be activated to ensure a "soft touchdown" ( v = 0 at impact)? Let r(t) denote the lander's distance from the center of the moon at time t (Fig. the difference is hardly visible.74 x 106 meters (or 1740 km. which has a radius of R = 1.2.8.90 Chapter 1 FirstOrder Differential Equations y 120 100 80 60 40 20 FIGURE 1. a little over a quarter of the earth's radius). but now we take account of lunar gravity.8. However. the atmospheric reentry and descent of a space vehicle. The following example is similar to Example 2 in Section 1.2. and Example 3 (with air resistance proportional to the square of the velocity) are all plotted. the distance r is measured between the centers of the spheres. Example 2 (with linear air resistance). and at an altitude of 53 kilometers above the lunar surface its downward velocity is measured at 1477 kmlh. The formula is also valid if either or both of the two masses are homogeneous spheres. The height functions in Example 1 (without air resistance).6726 x w. where the corresponding height functions are graphed.35 x 1022 (kg) is the mass of the moon.11 N · (m/kg)2 in mks units). the gravitational acceleration acting on it is not constant. And in Fig.3). in this case. when fired in free space. Its retrorockets. the difference between linear and nonlinear resistance can be significant in more complex situationssuch as. . Comparison of the last two lines of data here indicates little differencefor the motion of our crossbow boltbetween linear air resistance and air resistance proportional to the square of the velocity.8. The graphs of the latter two are visually indistinguishable. When we combine the (constant) thrust acceleration T and the (negative) lunar acceleration F/m = GMjr 2 of Eq. we get the (acceleration) differential equation (20) Solution where M = 7. for instance. (19).2. 1. " " Example 4 A lunar lander is freefalling toward the moon. 1. • Variable Gravitational Acceleration Unless a projectile in vertical motion remains in the immediate vicinity of the earth's surface. According to Newton's law of gravitation. the gravitational force of attraction between two point masses M and m located at a distance r apart is given by F=~ GMm r2 (19) where G is a certain empirical constant ( G ~ 6.
78187 x 106 and finally h = r.C2 ) = 1. r = R (at touchdown) into (21) gives 1 2 GM v = 4 r + .dr r2 FIGURE 1. After ignition: Substitution ofT = 4 and v = 0. Before ignition: Substitution ofT = 0 in (21) gives the equation 1 2 v =+C! GM r 2 (2la) where the constant is given by C 1 = v5/2.000 = 1.GMfro with km X Vo = . r = R at touchdown.450 rnls at the instant of ignition.1. Moreover.74 x 106 ) +53. 41.R = 41. The lunar lander with the new independent variable r.87 kilometersjust over 26 miles).3.1477 h 1000 km m X 1h 3600 S  14770m 36 s and r0 = (1.793 x 106 m (from the initial velocityposition measurement).8.GMjR is obtained by substituting the values v = 0. Therefore we can find its desired height h above the lunar surface at ignition by equating the righthand sides in (21a) and (21b).6) and obtain the firstorder equation dv GM v.870 meters (that is.= T . (36) of Section 1.+C2 2 r (2lb) where the constant C2 = 4R. we substitute dr v=dt' d 2r dv dv dr dv ==. At the instant of ignition the lunar lander's position and velocity satisfy both (21a) and (21b).· =vdt2 dt dr dt dr (as in Eq. • . Integration with respect to r now yields the equation 1 2 GM v =Tr++C 2 r descending to the surface of the moon. (21) that we can apply both before ignition (T = 0) and after ignition (T = 4). This gives r = iCC 1 .8 AccelerationVelocity Models 91 Noting that this secondorder differential equation does not involve the independent variable t. substitution of this value of r in (21a) gives the velocity v = .
A mass mat a great distance from the earth.680 ftjs. (20).378 x 106 (m). With r(t) denoting the projectile's distance from the earth's center at time t (Fig. so it continues forever to move away from the earth. Remark: Equation (24) gives the escape velocity for any other (spherical) planetary body when we use its mass and radius.180 (m/s) (about 36. Therefore.. then r (t) * oo as t * oo. about 6. Similarly. In particular. 1. when we use the mass M and radius R for the moon given in Example 4. R 2GM so v will remain positive provided that v5 ~ 2GMjR . this gives vo ~ 11.8. which has an equatorial radius of R = 6. • . but with T = 0 (no thrust) and with M = 5. v >v0 2 2  . we find that escape velocity from the lunar surface is v 0 ~ 2375 mjs..4). 2 r Now v = v0 and r = R when t = 0. With the given values of G and the earth's mass M and radius R./2G M j R.95 mijs. This will be so if its velocity v = drjdt remains positive for all t > 0. if the projectile's initial velocity exceeds .975 x 1024 (kg) denoting the mass of the earth. indeed. For instance.GMjR. (23) This implicit solution ofEq. a fact that greatly facilitates the return trip ("From the Moon to the Earth"). Substitution of the chain rule expression dvjdt = v(dvjdr) as in Example 4 gives GM r2 Then integration of both sides with respect to r yields v=. similar to Eq. "escape" from the earth. soC= !v5. about 25. and hence solution for v 2 gives 2 v 2 = v0 + 2GM ~ .8.000 mi/h).92 Chapter 1 FirstOrder Differential Equations Escape Velocity lr Velocity v(t) In his novel From the Earth to the Moon (1865). This is just over onefifth of the escape velocity from the earth's surface. we have the equation dv d 2r dt = dt 2 = GM r2 ' (22) FIGURE 1. (22) determines the projectile's velocity vas a function of its distance r from the earth's center.R 1 2 GM (1 1) . we can ask what initial velocity vo is necessary for the projectile to escape from the earth altogether. the escape velocity from the earth is given by Vo = J2~M· (24) In Problem 27 we ask you to show that. dv dr v =+C.4. Jules Verne raised the question of the initial velocity necesary for a projectile fired from the surface of the earth to reach the moon. so it does.
so that dvjdt = kv 2 • Show that v(t) and that = 1 + v0 kt Vo x(t) = xo + k ln(l + v0 kt). Suppose that a body moves through a resisting medium with resistance proportional to its velocity v. A woman bails out of an airplane at an altitude of 10. and that 10 s later the boat has slowed to 20 ft/ s. Then dv 1000. Suppose that a car starts from rest. (17) with initial condition y(O) = Yo· .5 with the parachute. 15. Assume that a body moving with velocity v encounters resistance of the form dvjdt = kv 312 • Show that m dv dt = W +B+ FR. (13) to obtain the upwardmotion position function given in Eq. so that dvjdt = kv. with the sole difference that the deceleration due to air resistance now is (0. Conclude that under a ~power resistance a body coasts only a finite distance before coming to a stop. Assume. Rework both parts of Problem 7. dt If the boat starts from rest.000 ft. while air resistance provides 0. The acceleration of a Maserati is proportional to the difference between 250 km/h and the velocity of this sports car.= 5000. a paratrooper survived a training jump from 1200 ft when his parachute failed to open but provided some resistance by flapping unopened in the wind. Which offers less resistance when the body is moving fairly slowlythe medium in this problem or the one in Problem 2? Does your answer seem consistent with the observed behaviors of x(t) as t * oo? 5. Assume that the water resistance is 100 pounds for each foot per second of the speed v of the boat.JVO +2) x(t)=xo+~. It is proposed to dispose of nuclear wastesin drums with weight W = 640 lb and volume 8 ft3by dropping them into the ocean ( v0 = 0). x (t) * +oo as t * +oo.fii!i to obtain the downwardmotion velocity function given in Eq. According to a newspaper account. (12) and substitute u = v.) 11. Test the accuracy of this account. where the buoyant force B is equal to the weight (at 62. (a) Find the car's maximum possible (limiting) velocity. How far will the boat coast in all? 4. (16) with initial condition v(O) = v0 • Integrate the velocity function in Eq. 1 Note that.JV0+2 14. and find that distance. (Suggestion: Find p in Eq. what is the maximum velocity that it can attain? 10.15 without the parachute and p = 1.jpjg to obtain the upwardmotion velocity function given in Eq. 7.000 lb and its motor provides a thrust of 5000 lb. 1 ftjs 2 of deceleration for each foot per second of the car's velocity. as in Problem 2. If the drums are likely to burst upon an impact of more than 75 ftjs. 2 )· kt. taking p = 0. Assuming resistance proportional to the square of the velocity (as in Problem 4). How long will it take her to reach the ground? Assume linear air resistance pv ftjs 2 . Then calculate the time required to fall 1200 ft. If this machine can accelerate from rest to 100 km/h in 10 s. what is the maximum depth to which they can be dropped in the ocean without likelihood of bursting? Separate variables in Eq. (b) Find how long it takes the car to attain 90% of its limiting velocity. in contrast with the result of Problem 2.kr and x(t) = Xo + (~0 ) (1. The force equation for a drum falling through water is 1. 16. v(t) and that = 4v0 2 (kt. Suppose that a motorboat is moving at 40 ftjs when its motor suddenly quits. (14) with initial condition y(O) = Yo· Separate variables in Eq. (4) by assuming a terminal velocity of 100 mi/h. (b) Conclude that the body travels only a finite distance. and how far it travels while doing so. A motorboat weighs 32. (13) with initial condition v(O) = v0 • Integrate the velocity function in Eq. 9. Allegedly he hit the ground at 100 mi/h after falling for 8 s.ekr). how far does the motorboat of Problem 3 coast in the first minute after its motor quits? 6.5 lb/ft3 ) of the volume of water displaced by the drum (Archimedes' principle) and FR is the force of water resistance.8 AccelerationVelocity Models 93 E Problems 8. (15) and substitute u = v.JV0(1k 13. found empirically to be 1 lb for each foot per second of the velocity of a drum.) 12. then opens her parachute. its engine providing an acceleration of 10 ft/ s2 . (a) Show that its velocity and position at time t are given by v(t) = voe. (Suggestion: First determine her height above the ground and velocity when the parachute opens. falls freely for 20 s. 3. how long will it take for the car to accelerate from rest to 200 km/h? 2. Consider a body that moves horizontally through a medium whose resistance is proportional to the square of the velocity v.100v. that the resistance it encounters while coasting is proportional to its velocity.001)v2 ft/s 2 when the car's velocity is v feet per second.1. (16) to obtain the downwardmotion position function given in Eq.
Consider the crossbow bolt of Example 3. Suppose that you are strandedyour rocket engine has failedon an asteroid of diameter 3 miles..R(R 2GMy + y) for the velocity v of the projectile at height y. (a) Show that if a projectile is launched straight upward from the surface of the earth with initial velocity v0 less than escape velocity J2GMJR.80 slater with an impact speed of about 43.1 2GM V~ Jr). Take g = 9." . the maximum possible speed of the boat). (b) If the projectile is launched vertically with initial velocity v0 > J2GM/R. g 22. suppose that the bolt is now dropped (v0 = 0) from a height of y0 = 108. respectively. In Jules Verne's original problem. (b) With what initial velocity v0 must such a projectile be launched to yield a maximum altitude of 100 kilometers above the surface of the earth? (c) Find the maximum distance from the center of the earth.~ max . If you =+ (S. (a) Suppose that a body is dropped ( v0 = 0) from a distance r 0 > R from the earth's center. can you blast off from this asteroid using leg power alone? 27. Then use Eqs. r(O)=vo 1 where Me and Mm denote the masses of the earth and the moon. r y'r Why does it again follow that r(t) + oo as t + oo? 20. How long will it now take him to reach the ground? 24. how long does it take to fall and with what speed will it strike the earth's surface? 29.94 Chapter 1 FirstOrder Differential Equations upward from the ground (y = 0) at time t = 0 with initial velocity v0 = 49 mjs. Continuing Problem 17. respectively. expressed in terms of earth radii. so its acceleration is dvjdt = GMjr 2 • Ignoring air resistance. An arrow is shot straight upward from the ground with an initial velocity of 160 ftjs. (15) for a paratrooper falling with parachute open. attained by a projectile launched from the surface of the earth with 90% of escape velocity.Rv 2 ' 0 (Suggestion: Substitute r = r0 cos 2 e to evaluate J r j(r0 . (23) in this section) explicitly to deduce that r(t) + oo as t + 00.49 mjs.47 min about 4. the projectile launched from the surface of the earth is attracted by both the earth and the moon. (a) To what radius (in meters) would the earth have to be compressed in order for it to become a black holethe escape velocity from its surface equal to the velocity c = 3 x 108 m/s of light? (b) Repeat part (a) with the sun in place of the earth. R is the radius of the earth and S = 384.0011 in Eq. Then use Eqs.61 s. 19. y(O) = 0.000 ft and opens his parachute immediately. (16) and (17) to show that it hits the ground about 4.00075 before opening his parachute.!.. y' (0) = vo.8 mjs 2 and p = 0.400 km is the distance between the centers of the earth and the moon. and falls from there to the lunar surface.. shot straight 18.r) d r.320 times that of the earth and its radius is 109 times the radius of the earth. 25. If he jumps from an altitude of 10._. with g = 32 ftjs 2) in Eq. deduce from Eq. Suppose that a projectile is fired straight upward from the surface of the earth with initial velocity v0 < J2GMJR . It experiences both the deceleration of gravity and deceleration v 2j800 due to air resistance. 28.r) 2 . what will be his terminal speed? How long will it take him to reach the ground? 23. Then its height y(t) above the surface satisfies the initial value problem J (y GM + R)2.47 m. Then solve the differential equation drjdt = kjy'r (from Eq. To reach the moon. so its distance r(t) from the center of the earth satisfies the initial value problem where M and R are the mass and radius of the earth. dt 2 r2 d 2r GMe GMm r(O)=R. 26. deduce that dr = dt Jk2 +a>. have enough spring in your legs to jump 4 feet straight up on earth while wearing your space suit. Suppose that p = 0. the projectile must only just pass the point between the moon and earth where its net acceleration vanishes. 1 ( = 2 p In 1 + pv~) .075 (in fps units. with density equal to that of the earth with radius 3960 miles. Thereafter it is "under the control" of the moon.r 2 + rocos. show that it reaches the height r < r0 at time 21. ( 14) that the maximum height it attains is Ymax 17. Suppose that the paratrooper of Problem 22 falls freely for 30 s with p = 0. (a) Suppose a projectile is launched vertically from the surface r = R of the earth with initial velocity v0 = J2GM/R so vJ = k 2 fR wherek 2 = 2GM. (13) and (14) to show that the bolt reaches its maximum height of about 108. A motorboat starts from rest (initial velocity v(O) = v0 = 0).) (b) If a body is dropped from a height of 1000 km above the earth's surface and air resistance is neglected. then the maximum distance from the center of the earth attained by the projectile is 2GMR r . The mass of the sun is 329. Substitute dvfdt = v(dvjdy) and then integrate to obtain vz = v6 . How high in the air does it go? If a ball is projected upward from the ground with initial velocity v 0 and resistance proportional to v 2 . Its motor provides a constant acceleration of 4 ftjs 2 . (12). and also find the limiting velocity as t + +oo (that is. but water resistance causes a deceleration of v 2 j 400 ft/ s2 • Find v when t = 10 s. Find the minima/launch velocity v0 that suffices for the projectile to make it "From the Earth to the Moon.2GM . t = J ro (Jrro. What maximum altitude does it reach if its initial velocity is 1 kmfs? 30.
8. where Fa = mg is a constant force of gravity and FR = kv is a force of air resistance proportional to velocity. Then the change in the momentum of the rocket itself is 11P ~ (m + 11m)(v + 11v). But here m is not constant. Hence the total change in momentum during the time interval 11t is 11P ~ = + v 11m+ 11m 11v) + (11m)(v.). To derive the equation of motion of the rocket.mv = m 11v + v 11m+ 11m 11v. m(tJ) = mr. Because of the combustion of its fuel. m(t) = mo. we use Newton's second law in the form dP (1) =F dt where P is momentum (the product of mass and velocity) and F denotes net external force (gravity. 1.1.c) m 11v + c 11m+ 11m !1v. with mass !:lm and approximate velocity v c. during which time the mass of the rocket decreases from m 0 to m 1• Thus m(O) = mo. But the system also includes the exhaust gases expelled during this time interval. so 11m ~ 0. (1) gives F= d(mv) dv dm dv =m+v=mdt dt dt dt' FIGURE 1. dm = {3 dt fort ::S tr.8 Application Rocket Propulsion y F l Suppose that the rocket of Fig. t 1]. for instancethen Eq. Suppose m changes to m + 11m and v to v + 11 v during the short time interval from t to t + 11t. An ascending rocket. The rocket is propelled by exhaust gases that exit (rearward) with constant speed c (relative to the rocket).5 blasts off straight upward from the surface of the earth at timet = 0. (m 11v Now we divide by 11t and take the limit as 11t ~ 0.8. The substitution of the resulting expression for dPjdt in (1) yields the rocket propulsion equation dv dm m+c=F. dm (3) Constant Thrust Now suppose that the rocket fuel is consumed at the constant "bum rate" fJ during the time interval [0. air resistance. (2) finally gives dv m dt + c dt = mg. the mass m = m(t) of the rocket is variable.5. (4) with burnout occurring at time t = t 1• .8 AccelerationVelocity Models 95 1.{Jt. Ifthe mass m of the rocket is constant so m'(t) = 0when its rockets are turned off or burned out. We want to calculate its height y and velocity v = dyjdt at time t. which (with dvjdt =a) is the more familiar form F = ma of Newton's second law.kv. then Eq. etc. assuming continuity of m(t). dt dt (2) IfF = Fa + FR.
{Jt)g.gt 1 ln...f3 mo.{Jt) ln . .{Jt) dv dt + kv = {Jc.{Jt 1 = m 1.{Jt (7) Because mo. (5) Solve this linear equation for (6) where vo = v(O) and M = _m_(t_) = _m_o_. it follows that the velocity of the rocket at burnout (t = t 1) IS (8) PROBLEM 3 Start with Eq. set k = 0 in Eq. (5) and integrate to obtain v(t) = vo.gt + cln .850 kg. Assume it encounters air resistance of 1.gt 2 y(t) = (vo c mo (mo. No Resistance PROBLEM 2 For the case of no air resistance.. of which 68.45 N per rn/s of velocity.{J_t mo mo denotes the rocket's fractional mass at timet.{Jt (9) It follows that the rocket's altitude at burnout is 1 2 cm1 mo + c)t1.96 Chapter 1 FirstOrder Differential Equations PROBLEM 1 Substitute the expressions in (4) into Eq.mo mo. 2 YI = y(tJ) = (vo f3 (10) m1 PROBLEM 4 The V2 rocket that was used to attack London in World War II had an initial mass of 12.5% was fuel. (3) to obtain the differential equation (m. This fuel burned uniformly for 70 seconds with an exhaust velocity of 2 km/s.(mo..:. (7) and integrate to obtain 1 2 + c)t. Then find the velocity and altitude of the V2 at burnout under the assumption that it is launched vertically upward from rest on the ground...
4).. if the rocket blasts off from rest (vo = 0) and c = 5 km/s and m 0 fmi = 20.8 AccelerationVelocity Models PROBLEM 97 5 Actually. and that the time t 8 which then elapses before actual blastoff is given by ts = mog.6). our basic differential equation in (3) applies without qualification only when the rocket is already in motion.{3c . its increase in velocity is ~ v = VI  mo v0 = c In . linear. when a rocket is sitting on its launch pad stand and its engines are turned on initially. but does not depend on the bum rate {3. {3g Free Space Suppose finally that the rocket is accelerating in free space. With the notation introduced to described the constantthrust case. show that the rocket initially just "sits there" if the exhaust velocity c is less than mog/{3. In this chapter we have discussed applications of and solution methods for several important types of firstorder differential equations. Given a differential equation f(x . y)] = 0. But the rocket does not descend into the ground. mi (11) Note that ~ v depends only on the exhaust gas speed c and the initialtofinal mass ratio m 0 jm I. The reason is that if v = 0 in (3). Thus if a rocket initially consists predominantly of fuel. including those that are separable (Section 1.6 we also discussed substitution techniques that can sometimes be used to transform a given firstorder differential equation into one that is either separable. With g = 0 in Eq. In Section 1. it just "sits there" while (because m is decreasing) this calculated acceleration increases until it reaches 0 and (thereafter) positive values so the rocket can begin to ascend. linear (Section 1. it is observed that a certain time interval passes before the rocket actually "blasts off" and begins to ascend. (1) we attempt to write it in the form d x [G(x. then it can attain velocities significantly greater than the (relative) velocity of its exhaust gases. For example. y. (8) we see that. so g = k = 0. Lest it appear that these methods constitute a "grab bag" of special and unrelated techniques. as the mass of the rocket decreases from m 0 to m I. it is important to note that they are all versions of a single idea. or exact.1. then the resulting initial acceleration =g dv dt c dm m dt of the rocket may be negative. or exact (Section 1. then its velocity at burnout is VI = 5ln 20 ~ 15 km/s. However.5). d (2) . where there is neither gravity nor resistance. y') = 0.
separate the variables and integrate (Section 1. Primes denote derivatives with respect to x. 23. • Is it linear? That is. y) such that Eqs.x 2y' = 0 xy + y 2 . xy' + 2y = 6x 2vy 10. 2x 2y + x 3y' = 1 8. Chapter 1 Review Problems Find general solutions of the differential equations in Problems 1 through 30. 2xy + x 2y' = y 2 9.l)y = 1 xy' = 6y + 12x 4 y 213 eY + ycosx + (xeY + sinx)y' = 0 9x 2y2 + x312y' = yz 2y + (x + l) y' = 3x + 3 . is there a plausible substitution that will make it so? For instance. 16. Several numerical methods for the appropriate solution of differential equations will be discussed in Chapter 6. 18. 20.6)? • If the equation as it stands is not separable. (1) by an appropriate integrating factor (even if all we are doing is separating the variables). linear. multi ply by the integrating factor p = exp (J P dx) of Section 1.l)y' + (x . 15.xy' = 0 xy 2 + 3y 2 . (1) and (2) are equivalent.5. we can attack it by means of the following steps: • Is it separable? If so.3x y' = x 2 .x 2y' = 0 2xy3 +ex+ (3x 2y 2 +sin y)y' = 0 3y + x 4 y' = 2xy 2xy2 + x2y' = y2 7. y) = C (3) that one obtains by integrating Eq. 14. Because of the wide availability of computers. numerical techniques are commonly used to approximate the solutions of differential equations that cannot be solved readily or explicitly by the methods of this chapter. or exact. Given a specific firstorder differential equation to be solved. can it be written in the form . 22. 6xy3 + 2y 4 + (9x 2y 2 + 8xy 3 )y' = 0 1. 5. is aMjay = aNjax (Section 1.x 3y' = i 3x 5 y 2 +x 3 y' = 2y2 xy' + 3y = 3x . when the equation is written in the form M dx + N d y = 0. is it homogeneous (Section 1. 21. 4.312 (x 2 .2xy + y 2 ex+ yexy + (eY + xeYx )y' = 0 2x 2y . But once we have found a function G(x. 3. 17. • Is it exact? That is.98 Chapter 1 FirstOrder Differential Equations It is precisely to obtain the form in Eq.4). 13. 4xy2 + y' = 5x 4 y 2 x 3 y' = x 2 y. y' = 1 +x2 + y2 +x2y2 11. most of the solution curves shown in figures in this chapter were plotted using numerical approximations rather than exact solutions. a general solution is defined implicitly by means of the equation G(x. (2) that we multiply the terms in Eq. 2. x 2y' = xy + 3y 2 12. Nevertheless. Indeed. 19.6)? Many firstorder differential equations succumb to the line of attack outlined here.+ P(x)y = dx dy Q(x)? If so. many more do not. 6. (2). x 3 + 3y .y 3 y' + 3y = 3x 2e. 24. 25.
35.y dy = dx tanx . 3y + x 3y 4 + 3xy' = 0 28..= 3(y +7)x 2 dx dy 32.jY. dy dx 3x 2 + 2y 2 4xy dy dx  X +3y y 3x dy 2xy + 2x = dx x2+ 1 36. y + xy' = 2e2x 29. 33. (2x + 1)y' + y = (2x + 1)312 30. then reconcile your results. exact. derive gen 99 era/ solutions for each ofthese equations in two different ways.8 AccelerationVelocity Models 26. homogeneous. 9xlf2y4f3 _ 12xlfSy3/2 + (8x3f2yl f3 _ 15x6f5ylf2)y' = 0 27. dy 31. etc. Bernoulli.=xlxy dx 34. y' = Jx + y Each of the differential equations in Problems 31 through 36 is of two different types considered in this chapterseparable. . Hence. .1. linear.
y'. B(x). We now tum to equations of higher order n ~ 2. 100 . By contrast. beginning in this chapter with equations that are linear. which we outline in this initial section. y.Linear Equations of Higher Order BJ Introduction: SecondOrder Linear Equations n Chapter 1 we investigated firstorder differential equations. Recall that a secondorder differential equation in the (unknown) function y (x) is one of the form I G(x. and F(x) are continuous on some open interval I (perhaps unbounded) on which we wish to solve this differential equation.! X is linear because the dependent variable y and its derivatives y' and y" appear linearly. Thus a linear secondorder equation takes (or can be written in) the form A(x)y" + B(x)y' + C(x)y = F(x). (2) Unless otherwise noted. C(x). we will always assume that the (known) coefficient functions A(x). The general theory of linear differential equations parallels the secondorder case (n = 2). (1) This differential equation is said to be linear provided that G is linear in the dependent variable y and its derivatives y' and y". but we do not require that they be linear functions of x. Thus the differential equation ex y" + (COSX)y' + (1 + Vx)Y =tan. the equations y" = yy' and y" + 3(y') 2 + 4l = 0 are not linear because products and powers of y or its derivatives appear. y") = 0.
We assume that the dashpot force FR is proportional to the velocity v = dxjdt of the mass and acts opposite to the direction of motion. we will return to this problem in detail in Section 2.1. In general. Of course.+c.2. it is nonhomogeneous. then we call Eq. dx FR = cv = . (2) is A(x)y" + B(x)y' + C(x)y = 0. Then X. 2. (2) a homogeneous linear equation. the secondorder equation x 2 y" + 2xy' + 3y = cosx is nonhomogeneous. the homogeneous linear equation associated with Eq. (3) In case the differential equation in (2) models a physical system. dt 2 dt (5) Thus we have a differential equation satisfied by the position function x(t) of the mass m. then Newton's law F = ma gives (4) that is.+kx = 0.1 ).6).1 Introduction: SecondOrder Linear Equations 101 If the function F(x) on the righthand side of Eq. (2) vanishes identically on I. This homogeneous secondorder linear equation governs the free vibrations of the mass. suppose that a mass m is attached both to a spring that exerts on it a force Fs and to a dashpot (shock absorber) that exerts a force FR on the mass (Fig. . the nonhomogeneous term F (x) frequently corresponds to some external influence on the system. If FR and Fs are the only forces acting on the mass m and its resulting acceleration is a = dvjdt. For example.1. For example. Directions of the forces acting on m.1. d 2x dx m . Assume that the restoring force Fs of the spring is proportional to the displacement x of the mass from its equilibrium position and acts opposite to the direction of displacement.2. V > 0 FIGURE 2. otherwise. A Typical Application Spring Mass Dashpot ~ ==~ x=O x>O Equilibrium position Linear differential equations frequently appear as mathematical models of mechanical systems and electrical circuits.4. it is not unusualeither in mathematics or in the English language more generallyfor the same word to have different meanings in different contexts.c dt (with c > 0) so FR < 0 if v > 0 (motion to the right) while FR > 0 if v < 0 (motion to the left). its associated homogeneous equation is x 2 y" + 2xy' + 3y = 0. A massspringdashpot system. Then FIGURE 2. Fs = kx (with k > 0) so Fs < 0 if x > 0 (spring stretched) while Fs > 0 if x < 0 (spring compressed).1. Remark: Note that the meaning of the term "homogeneous" for a secondorder linear differential equation is quite different from its meaning for a firstorder differential equation (as in Section 1.
If c 1 and c2 are constants. + p(x)y' + q(x)y = 0. 0 at each point of I. (6) This nonhomogeneous linear differential equation governs the forced vibrations of the mass under the influence of the external force F (t). (7) by A(x) and write it in the form y" + p(x)y' + q(x)y = f(x). so we can divide each term in Eq. in addition to Fs and FR. (8) We will discuss first the associated homogeneous equation y" (9) A particularly useful property of this homogeneous linear equation is the fact that the sum of any two solutions of Eq. (9) is again a solution. THEOREM 1 Principle of Superposition for Homogeneous Equations Let y 1 and y 2 be two solutions of the homogeneous linear equation in (9) on the interval I.102 Chapter 2 Linear Equations of Higher Order If. then the linear combination (10) is also a solution of Eq. (7) where the coefficient functions A. Here we assume in addition that A (x) =J=.+ kx = dt F(t). B. . (4)the resulting equation is d 2x mdt 2 dx +c . C. as is any constant multiple of a solution. (9) on I. This is the central idea of the following theorem. the mass m is acted on by an external force F(t)which must then be added to the righthand side in Eq. which gives Then y" + py' + qy = (CJYl + c2y2)" + p(CIYI + c2y 2)' +q(CIYI + c2y2) = (CJY~ + c2y~) + p(CJY~ + c2y~) + q(CJYI + C2Y2) = CJ (y~ + PY~ + qyi) + c2(y~ + PY~ + qy2) = CJ · 0 + C2 · 0 = 0 because Yl and Y2 are solutions. Proof: The conclusion follows almost immediately from the linearity of the operation of differentiation. Homogeneous SecondOrder Linear Equations Consider the general secondorder linear equation A(x)y" + B(x)y' + C(x)y = F(x). Thus y = c1 Y1 + c2Y2 is also a solution. and F are continuous on the open interval I.
4. FIGURE 2. Solutions y(x) = c 1 cosx of y" + y = 0.5.. We will see later that.1. (6) ought to have a unique solution satisfying these initial conditions.. 2. More generally. Solutions of y" + y = 0 with c 1 and c2 both nonzero.1 .. every solution of y" + y = 0 is a linear combination of these two particular solutions y 1 and Y2· Thus a general solution of y" + y = 0 is given by y(x) = c 1 cos x + Cz sinx. Earlier in this section we gave the linear equation mx" +ex'+ kx = F(t) as a mathematical model of the motion of the mass shown in Fig.1.1 . Physical considerations suggest that the motion of the mass should be determined by its initial position and initial velocity.>. FIGURE 2.1 Introduction: SecondOrder Linear Equations ~ 103 Example 1 We can see by inspection that Yt (x) = cos x and Y2 (x) = sin x are two solutions of the equation y II +y= 0. 8 6 4 2 .. It is important to understand that this single formula for the general solution encom passes a "twofold infinity" of particular solutions. in order to be a "good" mathematical model of a deterministic physical situation. Hence. Theorem 1 tells us that any linear combination of these solutions.3 through 2. a differential equation must have unique solutions satisfying any appropriate initial conditions.>.2 sin x y = 3 cos x + 4 sin x I 0 0 0 2 2 4 6 8 :rt 0 X :l't 4 6 2:rt 8 :rt 0 X :l't 2:rt 2 4 6 8 10 :rc 0 :l't X 2:rc 3n: FIGURE 2.3. Solutions y(x) = c2 sin x of y" + y = 0. because the two coefficients c 1 and c2 can be selected independently. Figures 2. given any preassigned values of x(O) and x' (0). y = 6 cos x . is also a solution. . • 8 6 4 2 . / c. The following existence and uniqueness theorem (proved in the Appendix) gives us this assurance for the general secondorder equation.5 illustrate some of the possibilities. 8 = 10 5 6 4 2 . such as y(x) = 3yt(x).2.>.1.1. with either c 1 or c2 set equal to zero. or with both nonzero. conversely.2y2(x) = 3cosx..2sinx.1.. Eq.
7. y'(a) = b 1• That is.. Theorem 2 tells us that any such initial value problem has a unique solution on the whole interval I where the coefficient functions in (8) are continuous. given any two numbers bo and b 1.6. Example 1 continued c2y2(a) = bo.3 that a nonlinear differential equation generally has a unique solution on only a smaller interval. (11) FIGURE 2. every nonvertical straight line through (a. one for each (real number) value of the initial slope y' (a) = b1• That is. bo) is tangent to some solution curve of Eq. (8). Figure 2. First. also as guaranteed by Theorem 2. Solutions of y" + 3y' + 2y = 0 with the same initial slope y' (0) = 1 but different initial values. we might actually find the solution y(x) whose existence is assured by Theorem 2. we find two "essentially different" solutions y 1 and y2. The application at the end of this section suggests how to construct such families of solution curves for a given homogeneous secondorder linear differential equation. the equation . More generally. b).7 shows a number of solution curves all having the same initial slope y' (0) = 1. b0 )namely. It has the initial values y(O) = 3. Then. • We saw in the first part of Example 1 that y (x) = 3 cos x .2 sin x is a solution (on the entire real line) of y" + y = 0.6 shows a number of solution curves of the equation y" + 3y' + 2y = 0 all having the same initial value y (0) = 1. y'(a) = bt . we attempt to solve the CtYi (a)+ 2 . 0 1 y" + p(x)y' + q(x)y = f(x) (8) 2 " 0 y'(O) = 6 has a unique (that is.firstorder differential equation dyjdx = F(x. 5 4 3 y(O) = 3 Remark I: Equation (8) and the conditions in (11) constitute a secondorder linear initial value problem. and f are continuous on the open interval I containing the point a.104 Chapter 2 Linear Equations of Higher Order THEOREM 2 Existence and Uniqueness for Linear Equations Suppose that the functions p. given a homogeneous secondorder linear equation.1. Solutions of y" + 3y' + 2y = 0 with the same initial value y(O) = 1 but different initial slopes. this illustrates the existence of such a solution. the solution y(x) = bo cosx + b1 sinx satisfies the arbitrary initial conditions y(O) = b 0 . 2. instead of there being only one line through (a. y) generally admits only a single solution curve y = y(x) passing through a given initial point (a. . 0 1 2 3 4 5 2 1 0 X 2 3 4 FIGURE 2. Theorem 2 tells us that this is the only solution with these initial values. • Example 1 suggests how. one and only one) solution on the entire interval I that satisfies the initial conditions 4 1 2 X 3 5 y(a) = bo.1. Recall from Section 1. while Fig.. second.1. Remark 2: Whereas a.>. y' (0) = b 1 . bo) tangent to a solution curve. CtY~ (a)+ c2y~(a) = b1 (13) for the coefficients c 1 and c 2 .>. we attempt to impose on the general solution (12) the initial conditions y(a) simultaneous equations bo. y'(O) = 2. q. Theorem 2 implies that the secondorder equation in (8) has infinitely many solution curves passing through the point (a.1 .
The following definition tells precisely how different the two functions y 1 and Y2 must be. c2 = 2. =C) = 4 6 8 10 2 The resulting solution is c 1 value problem is 1 0 X = 3. no matter what the initial conditions bo and b 1 might be. we omit it. x2· x+ l X lxl.. Hence the solution of the original initial y(x) = 3ex . e. Two functions are said to be linearly dependent on an open interval provided that they are not linearly independent there. y' (O) = 1.1. one of them is a constant multiple of the other.8 shows several additional solutions of y". Example 3 Thus it is clear that the following pairs of functions are linearly independent on the entire real line: sinx ex ex and and and and and cos x.8. and then find a solution satisfying the initial conditions y(O) = 3. 10 8 6 4 2 .1 Introduction: SecondOrder Linear Equations 105 Example 2 Verify that the functions Yl (x) =ex and Y2(x) = xex are solutions of the differential equation y".2y' +y = 0. ' xex. that is. We can always determine whether two given functions f and g are linearly dependent on an interval I by noting at a glance whether either of the two quotients f jg or gjf is a constantvalued function on I.. Figure 2. • In order for the procedure of Example 2 to succeed.2. to obtain the simultaneous equations y(O) 0 2 3. Solution The verification is routine. We impose the given initial conditions on the general solution for which y'(x) = (cl +c2)ex +c2xex. y'(O) = c1 +c2 = 1.2xex. the two solutions y 1 and Y2 must have the elusive property that the equations in (13) can always be solved for c 1 and c2 .2y' + y = 0 with the same initial value y(O) = 3. all having the same initial value y(O) = 3. FIGURE 2. DEFINITION Linear Independence of Two Functions Two functions defined on an open interval I are said to be linearly independent on I provided that neither is a constant multiple of the other. Different solutions y (x) = 3ex + c2 x ex of y" . ' .1.2y' + y = 0.2x..
2 + sm x = 1 and W( e x . x smx = I ex ex sm x cosx I = cos xe x 2 x . W (cos x. y~(a) = 1.ex is a constantvalued function. But the identically zero function j(x) 0 and any other function g are linearly dependent on every interval. . y2. depending on whether we wish to emphasize the two functions or the point x at which the Wronskian is to be evaluated. It is then impossible that either y 1 = ky2 or Y2 = ky 1 because k · 0 =!= 1 for any constant k. the determination of the constants c 1 and c2 in (12) depends on a certain 2 x 2 determinant of values of y 1 . and so forth. Given two functions f and g . the functions f (x) = sin 2x and g (x) = sin x cos x = are linearly dependent on any interval because f(x) = 2g(x) is the familiar trigonometric identity sin 2x = 2 sin x cos x. Also. that given any two linearly independent solutions y 1 and Y2 of the homogeneous equation y"(x) + p(x)y'(x) + q(x)y(x) = 0. finally.f 'g . • General Solutions But does the homogeneous equation y" + py' + qy = 0 always have two linearly independent solutions? Theorem 2 says yes! We need only choose y 1 and Y2 so that Yt (a) = 1. (9)it provides all possible solutions of the differential equation. and in greater detail beginning in Section 2. These are examples of linearly independent pairs of solutions of differential equations (see Examples 1 and 2).2x = e 3x nor e2x. (9) every solution y of Eq. neither sinxjcosx = tanx nor cosxjsinx = cotx is a constantvalued function. We want to show. Yi (a) =0 and Y2(a) = 0. Theorem 2 tells us that two such linearly independent solutions exist. As suggested by the equations in (13). the Wronskian of f and g is the determinant W= f g ! ' g' = fg'.cos . We write either W(f.106 Chapter 2 Linear Equations of Higher Order That is. because 0 · g(x) = 0 = j(x). g) or W(x ). neither ex je. Note that in both cases the Wronskian is everywhere nonzero. sm x) . actually finding them is a crucial matter that we will discuss briefly at the end of this section. and their derivatives. = I. (9) can be expressed as a linear combination (12) of Yt and Y2· This means that the function in ( 12) is a general solution of Eq. For example.3. xe x) ex X~ + I = e 2x .
In Section 2. then there exist numbers c 1 and c2 such that . (9)) ·.:_.J. then W(f. By ._ . If y is any solution whatsoever ofEq. The determinant of the coefficients in this system of linear equations in the unknowns c1 and c2 is simply the Wronskian W(y 1. y 2) evaluated at x = a. with f I.2 we will prove that. (9) if Yt and Y2 are linearly independent solutions. (14) CtYi (a)+ c2y~(a) = Y'(a). Y2) ":f 0 at each point of I.. the Wronskian is never zero if the solutions are linearly independent. . y" + p(x)y' + q(x)y =0 on an opeJ. " · · . The latter fact is what we need to show that y = Ct Yt + c2Y2 is the general solution of Eq. .: . THEOREt.A4 General Solutions of Homogeneous Equations Letyrandy2 '6 e·twolinearlyindependent solutions of the homogeneous equation {Eq. y2) 0 on I.. if the two functions Yt and Y2 are solutions of a homogeneous secondorder linear equation.kg'g = 0. Y(x) = CtYt (x) + c2y2(x) for allx in L In essence.intei"Vali on which p and q are continuous..___ · . We therefore call the linear combination Y = c1 YI + c2Y2 a general solution of the differential equation.atld are Hneady independent. :' Suppo~~ttl~f)it an<. then w(YI. then we have found all of its solutions. (9)on I.1<ly2 arelinearly dependent. then the strong converse stated in part (b) of Theorem 3 holds. '' .ly2are two solutions of the homogeneous secondorder linear equation(Eq.1 Introduction: SecondOrder Linear Equations 107 = On the other hand. Yi = Thus. . and consider the simultaneous equations CtYl (a)+ c2y2(a) = Y(a). then W(y 1.(9)) y"+p(x)y' +q(x)y = 0 with _p andq'i¢bntil1UOUS on the open interval I.. Thus the Wronskian of two linearly dependent functions is identically zero. given two solutions of Eq. g)= f and g are linearly dependent. THEOREM 3 Wronskians of Solutions ·: ..l. (9). Theorem 4 tells us that when we have found two linearly independent solutions of the secondorder homogeneous equation in (9)... there are just two possibilities: The Wronskian W is identically zero if the solutions are linearly dependent.. (a}Ifyra. Proof of Theorem 4: Choose a point a of I. '. " (b) :Ifyt .2. I= kgg'. if the functions kg (for example).
because cosh 2x = ~e 2x + ~e. (9).2 (cosh2x) = (2sinh2x) = 4cosh2x dx dx and." • .2x and sinh 2x = ~e 2x  d2 d ~e. Example 4 y~ = (2)(2)e 2x = 4e2x = 4yl and y~ = ( 2)( 2)e2x = 4e2x = 4Y2· Therefore. (15).4y = 0. sinh x are two different pairs of linearly independent solutions of the equation y" . so do Y' and G'. With these values of c 1 and c2. • Remark: Because e 2x.2x by the definitions of the hyperbolic cosine and hyperbolic sine. because . Theorem 4 implies that every particular solution Y (x) of this equation can be written both in the form and in the form Y(x) = a coshx + b sinhx. so by elementary algebra it follows that the equations in (14) can be solved for c 1 and c2. this is no surprise. (sinh 2x )" = 4 sinh 2x. we define the solution of Eq. By the uniqueness of a solution determined by such initial values (Theorem 2). as desired. Of course. y 1 and Y2 are linearly independent solutions of y".2x and cosh x. this determinant is nonzero. Hence each of these two linear combinations is a general solution of the equation. Thus these two different linear combinations (with arbitrary constant coefficients) provide two different descriptions of the set of all solutions of the same differential equation y" .108 Chapter 2 Linear Equations of Higher Order Theorem 3. then G(a) = CJYI (a)+ c2y2(a) = Y(a) and G'(a) = CJYi (a)+ c2y~(a) = Y'(a). (15) But y3(x) =cosh 2x and y4 (x) =sinh 2x are also solutions of Eq. It therefore follows from Theorem 4 that the functions cosh 2x and sinh 2x can be expressed as linear combinations of YI (x) = e 2x and y2(x) = e.4 y = 0. it follows that Y and G agree on I. Thus we see that Y(x) = G(x) = CJYl (x) + c2y2(x). likewise. Thus the two solutions Y and G have the same initial values at a.4y = 0 in (15). Indeed. similarly.2x. e. this is why it is accurate to refer to a specific such linear combination as "a general solution" rather than as "the general solution.
(18) has two distinct (unequal) roots r 1 and r 2 . we actually have found two solutions: y 1 (x) = e 2x and y2 (x) = e3x. then each term would be a constant multiple of erx. We first look for a single solution ofEq. b. if we substitute y = erx in the equation y". and c. Hence. (18) This quadratic equation is called the characteristic equation of the homogeneous linear differential equation ay" + by' + cy = 0.3)erx = 0. (16). then y = erx will be a solution ofEq. . b. THEOREM 5 Distinct Real Roots If the roots r 1 and r 2 of the characteristic equation in ( 18) are real and distinct. Hence y = erx will be a solution if either r = 2 orr = 3. For example. This suggests that we try to find a value of r so that these multiples of erx will have sum zero. we discuss the homogeneous secondorder linear differential equation ay" + by' + cy = 0 (16) with constant coefficients a. we find the result to be Because erx is never zero. we substitute y = erx in Eq. then (19) is a general solution ofEq. (16) and begin with the observation that (17) so any derivative of erx is a constant multiple of erx.1 Introduction: SecondOrder Linear Equations 109 Linear SecondOrder Equations with Constant Coefficients As an illustration of the general theory introduced in this section.2. we conclude that y(x) = erx will satisfy the differential equation in ( 16) precisely when r is a root of the algebraic equation ar 2 + br + c = 0. (16). with the constant coefficients dependent on r and the coefficients a. (16). we obtain Thus (r 2  5r + 6)erx = 0. So. (16) If Eq. With the aid of the equations in (17). (16). (r . if we substituted y = erx in Eq. (Why?) This gives the following result.5y' + 6y = 0. If we succeed. To carry out this procedure in the general case. in searching for a single solution. then the corresponding solutions YI(x) = er 1x and y 2 (x) = erzx of(16) are linearly independent.2) (r . and c.
• = Remark: Note that Theorem 5 changes a problem involving a differential equation into one involving only the solution of an algebraic equation. we get the general Figure 2. (16). Solution We can solve the characteristic equation 2r 2 7r +3 = 0 by factoring: (2r . so Theorem 5 yields the general • Example 6 The differential equation y" + 2y' = 0 has characteristic equation r2 ····~ . all appearing to approach the solution curve y(x) 1 (with c2 = 0) as x + +oo. (20). Because e0 ·x solution = 1. The problem in this case is to produce the "missing" second solution of the differential equation. A double root r = r 1 will occur precisely when the characteristic equation is a constant multiple of the equation Any differential equation with this characteristic equation is equivalent to (20) But it is easy to verify by direct substitution that y = xer'x is a second solution of Eq.x of Eq. It is Clear (but you should verify) that are linearly independent functions.9 shows several different solution curves with c 1 = 1. we get (at first) only the single solution y 1(x) = er. + 2r = r(r + 2) = 0 with distinct real roots r 1 = 0 and r2 = 2. so the general solution of the differential equation in (20) is .1.2x of y" + 2y' with different values of c2 • =0 If the characteristic equation in (18) has equal roots r 1 = r2 . The roots r 1 solution = ~ and r 2 = 3 are real and distinct.1.7y' + 3y = 0.. Solutions y(x) = I+ c 2 e.1)(r.9.3) = 0. • X FIGURE 2. .11 0 Chapter 2 Linear Equations of Higher Order Example 5 Find the general solution of 2y".
This particular solution. two functions y 1 and yz. y' (0) = 0 6. (16). lfll Problems In Problems 1 through 16. Yz = e.ex + 2xex of y" + 2y' + y = 0 with different values of c 1 • which imply that c 1 = 5 and c2 = 2. so the initial conditions yield the equations y(O) = y'(O) = 5~~~~~~~~~~ CJ CJ + Cz = 5. together with several others of the form y(x) 2xe. Yz = e. y(O) = 3.3 y' + 2 y = 0. y'(O) = 5 2. then (21) is a general solution of Eq. Solutions y(x) = c. y' (O) = 8 4. and a pair of initial conditions are given. = c 1e . y' (O) = 10 5. 3.1. y . .6y = 0.1 Introduction: SecondOrder Linear Equations 111 THEOREM 6 Repeated Roots If the characteristic equation in (18) has equal (necessarily real) roots r 1 = r2. y . y" + 25y = 0.9y = 0. y" + y' . Yl =ex. 1. y" + 4y = 0. y (0) = 1. y(O) = 10. y" .10. Yz = sin 5x.Y = 0. y" . y 1 = cos2x.3. Example 7 To solve the initial value problem + 2y' + y = 0.x.x. y'(O) = 15 3.1. Primes denote derivatives with respect to x. y(O) = 7. y'(O) = 3. First verify that y 1 and y 2 are solutions of the differential equation. Yz = e2x. y(O) = 5. Thus the desired particular solution of the initial value problem is y(x) =sex + 2xex . y" we note first that the characteristic equation r2 + 2r + 1 = (r + 1) 2 = 0 has equal roots r 1 = r2 = 1.x + • The characteristic equation in (18) may have either real or complex roots. Yt + CzYz that satisfies the given initial conditions. a homogeneous secondorder linear differential equation. = cos 5x. y 2 = sin2x. Hence the general solution provided by Theorem 6 is Differentiation yields y ' ( x ) =cle . y(O) = . y(O) = 0. y' (O) = 1 . Y1 = e 2X.4 3 2 1 0 X 1 2 3 4 5 FIGURE 2. Yz = e. Y1 = ex. 2. y" . is illustrated in Fig.X + 5 c2e  X czxe X .10. Then find a particular solution of the form y = c. = e3 X.3x.1.2.3x. The case of complex roots will be discussed in Section 2.
2. A(x) where K is a constant. find a solution of y" + y = 1 satisfying the initial conditions y(O) = 1 = y' (0). but that if c "I 0 and c 1 1. Show that Y1 = I and Yz = Jx are solutions of yy" + (y') 2 = 0. 2 y" + 3 y' = 0 =0 35. g(x) = cos2 x + sin2 x f(x) = x 3 . y". Primes denote derivatives with respect to x. Yz) is either zero everywhere or nonzero everywhere (as stated in Theorem 3)? Apply Theorems 5 and 6 to find general solutions of the differential equations given in Problems 33 through 42. 7. y(l) = 7. y" . Yz = xe 5x. Find the thirdquadrant point of intersection of the solution curves shown in Fig.112 Chapter 2 Linear Equations of Higher Order 31. Find the highest point on the solution curve with y(O) and y' (0) = 6 in Fig. 9y"I2y'+4y=0 42. g(x) =I+ lxl f(x) = xex. y(l) = 2. 6y" . y(O) = 2. Yz = x . Yt = e 5x. 24. y(2) = 10. (a) Show that Y1 = x 3 and y 2 = lx 3 l are linearly independent solutions on the real line of the equation x 2 y" . Show that dW A(x)dx " ) .3x y' + 3 y = 0.1. 7. y'(l) = 1 14. y" 2y' +2y = 0. x 2 y" + 2xy'. Then substitute for Ay~ and Ay. =1 50. y'(O) = 1 10. 28. Why does this imply that there is no differential equation of the form y" + p(x) y' + q (x) y = 0.20y = 0 38. having both Yt and Yz as solutions? 32. Yz =ex sinx.3 . Show that y 1 = sinx 2 and y 2 = cosx 2 are linearly independent functions.(yz)(Ay " = (yJ)(Ay2 1 ).y' . Y1 = x 2 . Y1 = x. g(x) = 1 . Show that y = 1I x is a solution of y' + y 2 = 0. Yz = x 2 .7. (b) Solve this firstorder equation to deduce Abel's for 17. y 2 = sin(lnx). then y = cx 3 is not a solution. 4y" + 4y' + y = 0 41. yz). 23. 35 y" . Why do these facts not contradict Theorem 3? mula W(x) = K exp ( J B(x) dx). y2 ) is identically zero. 2y" . Yz = xex. Yt = x. g(x) =ex cosx f(x) = 2cos x + 3 sinx. y" + 6y' + l3y = 0. g(x) = lxl ex f(x) = sin2 x. but that their sum y = y 1 + y 2 is not a solution.2 sinx Let Yp be a particular solution of the nonhomogeneous equation y" + py' + qy = f(x) and let Yc be a solution of its associated homogeneous equation. y(O) = 3. y 1 = cos(lnx). y(x) = C J + CzX y(x) = ex ( Ct exVl + CzexVl) Problems 49 and 50 deal with the solution curves of y" + 3 y' + 2y = 0 shown in Figs.y = 0 39. with both p and q continuous everywhere. 21. 30. y'(O) = 5 12. . y" . y(x) = C t e IOx + CzXeIOx c1e 10x 46. y'(l) = 3 The following three problems illustrate the fact that the superposition principle does not generally hold for nonlinear equations. Y1 =ex cosx. y'(O) = 0 13. but that if c 2 11.cos 2x f(x) = ex sinx. y(O) = 2. y'(O) = 8 8.3y' = 0. y" + 5y' = 0 37. + Cze 100x 47. 27. y'(2) = 15 15.lOy' + 25y = 0.6. y(x) = 48.6y = 0. g(x) = 3 cosx . 19.15y 36.6 and 2. Find such an equation. y'(O) = 2 9. (c) Why does Abel's formula imply that the Wronskian W(y 1 . and C are continuous and A(x) is never zero.x. then y = cjx is not a solution. Yz =e. Yz = e 3x. 2.3x cos 2x. Show that y 1 = x 2 and y 2 = x 3 are two different solutions of x 2 y" . 2. Explain why these facts do not contradict Theorem 2 (with respect to the guaranteed uniqueness). y(O) = 4. Yt = l. x 2 y". y(O) = 0.y'. Show that y = x 3 is a solution of yy" = 6x 4 .2xy' + 2y = 0. but that their Wronskian vanishes at x = 0. 33. y(l) = 3. 29. 49. (a) Let W = W(yJ.3x sin 2x. y(O) = 2. both satisfying the initial conditions y(O) = 0 = y'(O) . 25. With Yp = I and Yc = c 1 cos x + Cz sin x in the notation of Problem 27. x 2 y" + xy' + y = 0.1.3y' + 2y =0 34. (b) Verify that W (y 1 . x 2 y".7y' .' from the original differential equation to show that dW A(x)dx = B(x)W(x).1.xy' + y = 0. y'(O) = 13 11. Yz = xlnx. f(x) = rr. y" + y' = 0. 26. y" + 2 y' . y" + 2y' + y = 0. Determine whether the pairs of functions in Problems 20 through 26 are linearly independent or linearly dependent on the real line. Yt = 1. y'(l) = 2 16. 18. Yz = e.1. Let y 1 and y 2 be two solutions of A(x)y" + B(x)y' + C(x)y = 0 on an open interval I where A. 20. B.4xy' + 6y = 0. Show that y = Yc + Yp is a solution of the given nonhomogeneous equation. 4y" + 8y' + 3y = 0 40.12 y =0 Each of Problems 43 through 48 gives a general solution y(x) of a homogeneous secondorder· differential equation ay" +by' + cy = 0 with constant coefficients. 22. Yt = ex. Y1 = e. 45. g(x) = x 2 1xl f(x) =I+ x.
. Yn be n solutions of the homogeneous linear equation in (3) on the interval I. we will always assume that the coefficient functions Pi (x) and F (x) are continuous on some open interval I (perhaps unbounded) where we wish to solve the equation.3x . A secondorder Euler equation is one of the form ax 2 y" 113 + bxy' + cy = 0 (22) conclude that a general solution of the Euler equation in (22) is y (x) = c 1x'~ + c 2x'2 • where a. + 8xy'.12y = 0 55. we can divide each term in Eq.2. and y 3(x) = sin 2x . Under the additional assumption that P0 (x) =!= 0 at each point of I. 54. 52. x 2 y" + 2xy'. (22) into the constantcoefficient linear equation Make the substitution v = ln x of Problem 51 to find general solutions (for x > 0) of the Euler equations in Problems 5256. c are constants. (1) by Po(x) to obtain an equation with leading coefficient 1.y d2y a dvl + (b. c2 . The proof of the following theorem is essentially the samea routine verificationas that of Theorem 1 of Section 2. a homogeneous nthorder linear differential equation has the valuable property that any superposition. en are constants. (3) on I. If c 1 . or linear combination.(x)y<nl) + · · · + PnI(x)y' + Pn(x)y = F(x). then the linear combination Y = CJYI + C2Y2 + · · · + CnYn (4) is also a solution of Eq.3xy' + 4y = 0 We now show that our discussion in Section 2. 4x 2 y" 56. THEOREM 1 Principle of Superposition for Homogeneous Equations Let y.I) + · · · + Pn1 (x)y' + Pn (x)y = f(x). then the substitution v = In x transforms Eq. b.a) dv + cy = dy 0 (23) =0 53. of solutions of the equation is again a solution. y 2 (x) = cos2x..3y = 0 2 x y" . + PnJ(x)y' + Pn(x)y = 0. x 2 y" + xy' = 0 with independent variable v. (1) Unless otherwise noted.1. (a) Show that if x > 0. (b) If the roots r 1 and r2 of the characteristic equation ofEq... Example 1 It is easy to verify that the three functions y 1 (x) = e. (23) are real and distinct. x 2 y" + xy'. (2) is yen)+ PI(x)y(n!) + .. (3) Just as in the secondorder case. ••.. . of the form yen) +PI (x)y(n . y2.2 General Solutions of Linear Equations 51. . (2) The homogeneous linear equation associated with Eq.1 of secondorder linear equations generalizes in a very natural way to the general nthorder linear differential equation of the form Po(x)y(n) + P.
2. Theorem 1 tells us that any linear combination of these solutions. ••• . It tells us nothing.3 we will see how to construct explicit solutions of initial value problems in the constan. a particular solution of an nthorder linear differential equation is determined by n initial conditions. y' (0) = 5.. and y 3 • Thus a general solution is given by y(x) = c 1e. 2. conversely.3e.1) looks periodic on the right. y'(a)=bt.2. we see that y (x) ~ 3 cos 2x . We saw earlier that y(x) = 3e. The particular solution y(x) = .2 sin 2x.2 sin 2x for large positive x.tcoefficient case that occurs often in applications. Pn. (2)) y<n) +PI (x)y(nl) + · · · + Pn1 (x)y' + Pn(x)y = f(x) has a unique (that is. We will see that.3x + 3 cos 2x  2 sin 2x. ' (5) 0 2 4 . Theorem 2 tells us that any such initial value problem has a unique solution on the whole interval I where the coefficient functions in (2) are continuous . . In Section 2. however. proved in the Appendix. .X 6 10 FIGURE 2. and Theorem 2 implies that there is no other solution with these same initial values.114 Chapter 2 Linear Equations of Higher Order are all solutions of the homogeneous thirdorder equation y(3) + 3y" + 4y' + 12y = 0 on the entire real line. is also a solution on the entire real line. Equation (2) and the conditions in (5) constitute an nthorder initial value problem. and f are continuous on the open interval I containing the point a.3x Example 1 Continued + 3 cos 2x  2 sin 2x 0 is a solution of y<3 ) + 3y" + 4y' + 12y = on the entire real line.3x + 3 cos 2x . is the natural generalization of Theorem 2 of Section 2. b 1 . because of the negative exponent. Note that its graph (in Fig. THEOREM 2 Existence and Uniqueness for Linear Equations Suppose that the functions p 1 . about how to find this solution. This particular solution has initial values y(O) = 0. such as y(x) = 3y 1 (x) + 3yz(x)  2y3 (x) = 3e. one and only one) solution on the entire interval I that satisfies the n initial conditions y(a)=bo.1. given n numbers b0 .1.1 that a particular solution of a secondorder linear differential equation is determined by two initial conditions. Indeed. every solution of the differential equation of this example is a linear combination of the three particular solutions y 1 . Similarly. The following theorem. p 2 . and y"(O) = 39. • .3x + Cz cos 2x + C3 sin 2x. • Existence and Uniqueness of Solutions We saw in Section 2. y 2 . bnl• the nthorder linear equation (Eq.. ••. Then.
Example 2 It is easy to verify that are two different solutions of x 2 y " . (3) with coefficient functions continuous on an open interval containing the point x = 0.3.2. Solutions of y< 3l + 3y" + 4y' + 12y = 0 with y (O) = y ' (O) = 0 but with different values for y" (0)... Remark: Because its general solution involves the three arbitrary constants and c 3.4. y'(O) = b 1. and = c 3 sin 2x (with c 1 = c 2 = 0).l (x)y' + Pn(x)y = 0 that satisfies the trivial initial conditions y(a) = y'(a) = · · · = y<nl )(a) = 0. Why does this not contradict the uniqueness part of Theorem 2? It is because the leading coefficient in this differential equation vanishes at x = 0.. and y" (0) = b2 .2 General Solutions of Linear Equations 115 c1. FIGURE 2. Note that Theorem 2 implies that the trivial solution y(x) solution of the homogeneous equation y<n) = 0 is the only (3) + PI (x)y<n1) + . Alternatively.8 0..4 0. two of these three initial values are zero..2. Solutions of y (3l + 3y" + 4y' + 12y = 0 with y(O) = y " (O) = 0 but with different values for y' (0). FIGURE 2. c 2. = c 2 cos 2x (with c 1 = c 3 = 0). and that both satisfy the initial conditions y(O) = y' (0) = 0.4 0.2.1 2 1 3 4 5 1 2 0 X y '(O) ""' = 3 2 3 4 5 3 2 1 0 X 2 3 4 0. so this equation cannot be written in the form ofEq. • 5 4 1 3 2 1 2 y'(O) = 3 / .2.. 0 0. Figures 2. the thirdorder equation in Example 1 has a "threefold infinity" of solutions.3x (obtained from the general solution with Cz = c3 = 0).4 illustrate three corresponding families of solutionsfor each of which.2..2.. + Pn.8 1 y "(O) = 3 / y"(O) = 3 3 4 5 1 0 X 2 FIGURE 2.2. Theorem 2 suggests a threefold infinity of particular solutions corresponding to independent choices of the three initial values y(O) = b0 . including three families of especially simple solutions: • y (x) • • y(x) y(x) = c 1e..2 through 2. Solutions of 3l + 3y" + 4y' + 12y = 0 with y< y' (0) = y" (0) = 0 but with different values for y (O).4xy ' + 6y = 0.2 0.6 0.. • .2 .6 0. 0 .
If we write these equations as (l)!J + (k)h = 0 or (k)j. we say that n functions f 1. c2 . . (6). if c 1 and c2 are not both zero.. . DEFINITION Linear Dependence of Functions The n functions !I. Yn are particular solutions of Eq. . h . c2 . ••. Cn not all zero such that (7) on I.. But these n particular solutions must be "sufficiently independent" that we can always choose the coefficients ci.. . . ••. .. if either !I = kh or h = kfi for some constant k. we anticipate that a general solution of the homogeneous nthorder linear equation y<n) + P1 (x)y(nI) + · · · + Pn .I (x)y' + Pn(x)y = 0 (3) will be a linear combination (4) where YI. that is. then clearly we can solve for at least one of the functions as a linear combination of the others.. (7) are zero. c 2 . .116 Chapter 2 Linear Equations of Higher Order Linearly Independent Solutions On the basis of our knowledge of general solutions of secondorder linear equations. fn are said to be linearly dependent on the interval I provided that there exist constants CI. + (l)h = 0. that is.. Y2 • . Thus the functions !I.. (6) certainly implies that f 1 and h are linearly dependent. In analogy with Eq.. h. we see that the linear dependence of f 1 and c 1 and c2 not both zero such that h implies that there exist two constants (6) Conversely. Cn are zero (although some of them may be zero). (3). fn are linearly dependent provided that some nontrivial linear combination of them vanishes identically.. The question is this: What should be meant by independence of three or more functions? Recall that two functions !I and h are linearly dependent if one is a constant multiple of the other.. then Eq. . h . . f n are linearly dependent if and only if at least one of them is a linear combination of the others. . If not all the coefficients in Eq. Cn in (4) to satisfy arbitrary initial conditions of the form in (5). nontrivial means that not all of the coefficients c 1. for all x in /. . . and conversely.
. Then their Wronskian is the n x n determinant ft W= fn f{ ~~ (8) We write W(j1 . fn are called linearly independent on the interval I provided that they are not linearly dependent there. h . . obtaining the n equations cifi(x ) + c2h(x) CJf{(x) + c2j. and j)(x) =ex are linearly dependent on the real line because (l)JI + (2)h + (O)j) = 0 (by the familiar trigonometric identity sin 2x = 2 sin x cos x ). Wronski (17781853). M. We then differentiate this equation n . But in order to show that n given functions are linearly independent. h. . nontrivial values of the coefficients so that Eq. More generally. To prove this. the functions f 1 . . • (7) Then functions f 1. h(x) = sinx cos x. that is. h • .(x) + ··· + +···+ Cnfn(X) = 0. in the case of n solutions of a homogeneous nthorder linear equation. .. We saw in Section 2.1. Cn not all zero. .. they are linearly independent on I provided that the identity Ctfi + c2h + · · · + Cnfn = = C2 0 holds on I only in the trivial case Ct = · · · = Cn = 0. no nontrivial linear combination of these functions vanishes on I. . (7) holds on the interval I for some choice of the constants c 1 ... we must prove that nontrivial values of the coefficients cannot be found. there is a tool that makes the determination of their linear dependence or independence a routine matter in many examples. fn) or W(x). c2 . . . Suppose that the n functions !I. The Wronskian is named after the Polish mathematician J... as in Example 3.1 times differentiable. fn is identically zero. fn are each n . Fortunately. H. This tool is the Wronskian determinant. Put yet another way. . h . the Wronskian of n linearly dependent functions fi.. Cnf~(x) = 0.1 times in succession. assume that Eq. which we introduced (for the case n = 2) in Section 2. h ...1 that the Wronskian of two linearly dependent functions vanishes identically.2 General Solutions of Linear Equations 117 Example 3 The functions !1 (x) =sin 2x. depending on whether we wish to emphasize the functions or the point x at which their Wronskian is to be evaluated.2. Equivalently. (9) .. fn are linearly independent if no one of them is a linear combination of the others. (7) holds. .. . and this is seldom easy to do in any direct or obvious manner. (Why?) Sometimes one can show that n given functions are linearly dependent by finding.
en and the determinant of coefficients is simply the Wronskian W (JI . fn) evaluated at the typical point x of I. When we compute the Wronskian of the three given solutions. to show that the functions !I. 0 X 2 . . y' (1) = 2.. We recall from linear algebra that a system of n linear homogeneous equations in n unknowns has a nontrivial solution if and only if the determinant of coefficients vanishes. Yz(x)=xlnx. In Eq. y 2 (x) Example 1) are linearly independent. and y3(x) sin 2x (of Solution sin2x 2cos2x 4 sin 2x W = 3e. (11) Solution Note that for x > 0. y 2 . it suffices to show that their Wronskian is nonzero at just one point of I.3x cos2x sin2x 4 sin 2x 4 cos 2x cos2x = 26e. (9) the unknowns are the constants c 1 . .:.3x i= 0.3x 2sin2x 4 cos 2x 2cos2x 4 sin 2x sin 2x 2 cos 2x + 3e.x 2 y" + 2xy'.Jx. fn are linearly independent on the interval I. . it follows that y 1 . . Then find a particular solution of Eq.. it follows that W (x) 0. . 2 sin 2x Because W i= 0 everywhere. Their Wronskian is cos2x cos 2x. h. Because we know that the ci are not all zero.118 Chapter 2 Linear Equations of Higher Order which hold for all x in I. (10) that satisfies the initial conditions y(l) = 3. h. Example 5 Show first that the three solutions YI(x)=x.3x 2 sin 2x 4 cos 2x 9e. and y 3 are linearly independent • on any open interval (including the entire real line). .2y = 0 (10) are linearly independent on the open interval x > 0. we could divide each term in ( 10) by x 3 to obtain a homogeneous linear equation of the standard form in (3). and Y3(x)=x 2 of the thirdorder equation x 3 y( 3) . y" (1) = 1. = Example 4 Show that the functions y 1 (x) ~. ••. . .. as we wanted to prove. c2 . Therefore. we find that X W= x lnx 1 + lnx 1 x2 1 2x =X.
it is sufficient to assume that W (a) = 0 at some point of I .. y2.2 General Solutions of Linear Equations 119 Thus W (x) =I= 0 for x > 0. Yn). Theorem 3 provides the necessary nonvanishing of W in the case of linearly independent solutions.1(x)y' + Pn(x)y = 0 (3) on an open interval I. . where each Pi is continuous. . . . c 2 = 3.2. THEOREM 3 Wronskians of Solutions Suppose that YI.. Y2· . . Yn are linearly dependent. Yn are n solutions of the homogeneous nthorder linear equation y(n) + PI (x)y(nl) + · · · + Pn.. and y 3 are linearly independent on the interval x > 0. . Thus there are just two possibilities: Either W = 0 everywhere on I .. . . so y 1. we impose the initial conditions in (11) on y"(x) = 0 This yields the simultaneous equations y(l) = CJ y' (1) = CJ + C3 = 3. it turns out (Theorem 4) that we can always find values of the coefficients in the linear combination Y = CIYI + C2Y2 + · · · + CnYn that satisfy any given initial conditions of the form in (5). .. then W =!= 0 at each point of I. · .. Y2. y2 . To find the desired particular solution. Yn are linearly dependent. • Provided that W (Yt. Y2. . y"(l) = we solve to find c 1 question is 1. Let W = W(yl. But W(a) is simply the determinant of . Thus the particular solution in y(x) = x . y2. + C2 + 2c3 = 2. . = Proof: We have already proven part (a). . . Yn) =I= 0. To prove part (b).. and c3 = 2. . Y2. and show this implies that the solutions Yt. (a) If Yl.3x lnx + 2x 2 . (b) If YI. or W =1= 0 everywhere on I. Yn are linearly independent. then W 0 on I.
Yn ben linearly independent solutions of the homogeneous equation yen)+ Pt(x)yCn. We now use these values to define the particular solution Y(x) = CJYJ (x) + C2Y2(x) + · · · + CnYn(x) (13) of Eq. In view of (13) and the fact that c 1. .1) + · · · + PnI(x)y' + Pn(x)y = 0 (3) on an open interval I where the Pi are continuous. Cn are not all zero. THEOREM 4 General Solutions of Homogeneous Equations Let Yt.. This completes the proof of Theorem 3. Y2. Yn are linearly dependent. (3). c 2 . . every (other) solution of the equation can be expressed as a linear combination of those n particular solutions. . A = General Solutions We can now show that. (3). then there exist numbers c 1 . y 2.1 \a) = 0.1 (the case n = 2). Cn. en are not all zero. . the basic fact from linear algebra quoted just after (9) implies that the equations in (12) have a nontrivial solution.. CnY~(a) = 0. . . If Y is any solution whatsoever ofEq.. c 2 . . . The equations in (12) then imply that Y satisfies the trivial initial conditions Y(a) = Y' (a) = · · · = yCn. Thus every solution of a homogeneous nthorder linear differential equation is a linear combination Y = Ct YI + C2Y2 + · · · + Cn Yn of any n given linearly independent solutions.1)(a) + · · · + CnY~nl)(a) = 0 in then unknowns c 1 . Using the fact from Theorem 3 that the Wronskian of n linearly independent solutions is nonzero. Cn such that Y(x ) = CtYI (x ) + c2y2(x ) + · · · + CnYn(x) for all x in I.• .120 Chapter 2 Linear Equations of Higher Order coefficients of the system of n homogeneous linear equations CtYI (a) + CtY~ (a) + C2Y2(a) + · ·· + c2y~(a) + · · · + CnYn(a) = 0. That is. •.. . On this basis we call such a linear combination a general solution of the differential equation. this is the desired conclusion that the solutions y 1 . given any fixed set of n linearly independent solutions of a homogeneous nthorder equation..•• . the numbers c 1 . Theorem 2 (uniqueness) therefore implies that Y(x) 0 on I. (12) C1Y~nl)(a) + C2Yin. the proof of the following theorem is essentially the same as the proof of Theorem 4 of Section 2. ••• . c2 . c 2 . Because W(a) = 0.
and that Y is any other solution of Eq.) + 2c3 = bt. .3x. b 1 .Yp is a solution of the associated homogeneous equation in (3).[( Yp + PlYp · · · + PnlYp + PnYp J 1 = f(x). (14) + CzY2 + · · · + CnYn.yP.4cz (See the application for this section.. (15) where y 1 . the particular solutions y 1 (x) = e.2 General Solutions of Linear Equations 121 Example 6 According t~. (2). .lY~ + PnYc = [cr<nl + Ply<nl) + · · · + Pn. . Now Theorem 2 says thatgiven bo.. Upon successive differentiation and substitution of x = 0. Then Y and it follows from Theorem 4 that Yc = C1Y1 = Yc + Yp.f(x) = 0. y' (0) = bh and y" (0) = b2 . (2). there exist coefficients c 1.3x + Cz cos 2x + C3 sin 2x. we discover that to find these coefficients. and y3 (x) =sin 2x of the linear differential equation y(3l +3y" +4y' + 12y = 0 are linearly independent. c2 . (3) Suppose that a single fixed particular solution Yp of the nonhomogeneous equation in (2) is known. We call Yc a complementary function of the nonhomogeneous equation and have thus proved that a general solution of the nonhomogeneous equation in (2) is the sum of its complementary function Yc and a single particular solution Yp of Eq.2. Yn are linearly independent solutions of the associated homogeneous equation. and b2there exists a particular solution y(x) satisfying the initial conditions y(O) = bo.1(x)y' + Pn(X)y = with associated homogeneous equation y<n) + Pt(x)yCnl) + · · · + Pn . 3cl 9cl .l) + · · · + Pn . then subsitution of Yc in the differential equation gives (using the linearity of differentiation) y~n) + PlY~n . yz.tY' + PnY] (n) (n1) + . Hence Theorem 4 implies that this particular solution is a linear combination of y 1. y 2 (x) = cos 2x. and c3 such that y(x) = c 1e. Thus Yc = Y.Ex~mple 4. we need only solve the three linear equations Ct + Cz = bo.l(x)y' + Pn(x)y = 0. and y 3 • That is.l)+ · · · + Pn . y 2 . • f(x) (2) Nonhomogeneous Equations We now consider the nonhomogeneous nthorder linear differential equation y<n) + Pl (x)yCn . = bz. If Yc = Y.
1. c2. y(3l +2y" .122 Chapter 2 Linear Equations of Higher Order THEOREM 5 Solutions of Nonhomogeneous Equations Let Yp be a particular solution of the nonhomogeneous equation in (2) on an open interval I where the functions Pi and f are continuous. . h(x) = 10 + 15x 2 = 0. Cn such that Y(x) = CIYI (x) + C2Y2(x) + · · · + CnYn(x) + Yp(x) (16) for all x in I. In Problems 13 through 20.g(x) = xex. + 2c2 cos 2x + 3. Now y' (x) = 2c 1 sin 2x Hence the initial conditions give y(O) = C] y' (0) = 2c2 +3= 7. y'(O) = 7.3x 2 . Solution The general solution of Eq. find a nontrivial linear combination of the given functions that vanishes identically.2X Y1 = e . f(x) 7. show directly that the given functions are linearly dependent on the real line. h(x) = sinx.2 Inx. Thus the desired solution is y(x) = 5cos2x +2sin2x +3x. 1. . h(x) = x. 11. Yn be linearly independent solutions of the associated homogeneous equation in (3).X . If Y is any solution whatsoever of Eq. h(x) = e 3 x. g(x) = 2sin 2 x . h(x) = 5x . g(x) = cos(lnx). . use the Wronskian to prove that the given functions are linearly independent on the indicated interval. h(x) =ex = 17. We find that c 1 = 5 and c2 = 2. h(x) = x 2 . That is. a thirdorder homogeneous linear equation and three linearly independent solutions are given. y'(O) = 2. x > 0 =x. 9. 3. y"(O) = 0.8x 2 = 5. x > 0 = = 17. g(x) = coshx. 4. h(x) = sin(lnx). then there exist numbers c 1. 10. 13. Y2 = e . g(x) = x. 6. h(x) = 3cos2 x f(x) f(x) f(x) f(x) f(x) 12. y(O) = 1. 8. g(x) = e2X. S. . 2. g(x) = cosx. Example 7 It is evident that Yp = 3x is a particular solution of the equation y" + 4y = 12x. f(x) f(x) f(x) f(x) f(x) f(x) = 2x.h(x) =x 2 ex. the real line =ex. = 5. therealline = x. the real line = ex. • In Problems 1 through 6. (17) and that Yc(x) = c 1 cos 2x +c2 sin 2x is its complementary solution. g(x) = 3x 2 .. Find a particular solution satisfying the given initial conditions.2 .2y = 0.y' . (17) that satisfies the initial conditions y(O) = 5.. X .. the real line = ex. y2. g(x) = 2. (2) on I. Y3 = e . g(x) = sinx. h(x) = sinhx In Problems 7 through 12.. g(x) = x . (17) is y(x) = c 1 cos 2x + c2 sin 2x + 3x. Let Yt. h(x) = cos2x =eX. g(x) = cos2 x. Find a solution ofEq.
x 2 ) vanishes at x = 0..yl =x. Yz = e2x.2y' + 2y = 2x. are distinct. Generalize the method of Problem 27 to prove directly that the functions fo(x) rl V= r2 I rz r2 2 r. f. and r3 are distinct. y'(O) = 0. y"(O) = 0. a complementary solution Yc· and a particular solution Yp are given. fz(x) = x 2 .2 = 1. Show that the three functions exp(r 1x). Y3 =ex sinx. y(O) = 0. fz(x) = x. Assume as known that the Vandermonde determinant are linearly independent on the whole real line. 27. r 2 .) and Lyz = g(x). Yz = e 2x. exp(r2 x). Yc = Cjex cosx + Czex sinx. y'(O) = 4. (a) Find by inspection particular solutions of the two nonhomogeneous equations y" 33. y 1 = 1. 123 16. Yl =ex. Yl =ex. .1 and Y. y.2y'.(x ) = exp(r.2 General Solutions of Linear Equations 14. y(O) = 1.2. Verify that y 1 = x and y2 = x 2 are linearly independent solutions on the entire real line of the equation x 2 y" .(k)(a) =0 if k =!= i . y ' (O) = 0. ••. y"(O) = 3. Yz = cos 3x. y"(O) = 0. Y3 = sin 3x y< 3 >3y" +4y' 2y = 0. (a) Given the equation y" + py' + qy = 0. the functions is nonzero if the numbers r 1 .1.. Why do these observations not contradict part (b) of Theorem 3? 31. Yz = xex. Yz = x 2 .l n.3y" + 3y'. y". Suppose that the three numbers r 1 . y'(O) = 11.3x 2 y" 6xy'. y'(O) = 2.y3 =x2 lnx 30. y'(O) = 0. y(O) = 3. y(l) = 6.y = 0.x). + but that W (x. and conclude from the equations you get that C1 = Cz = C3 = 0.yz =x2 . Y3 = xe 2x yO> +9y' = 0. Show that their sum y = y 1 + y 2 satisfies the nonhomogeneous equation Ly = f(x) + g(x). 34. (b) Use the method of Problem 25 to find a particular solution of the differential equation y" + 2y = 6x + 4. y'(O) = 1. (Suggestion: Assume that c 1 + c 2x + c 3x 2 = 0.. y"(l) = 22. y'(l) = 5. 18. Let Ly = y" + py' + qy. are linearly independent. 29.(x) = x" rl n.l rz n. y'(O) = 0. Y3 = e3x y< 3>. Yz =ex cosx. Yp = 3x 22.. Find a solution satisfying the given initial conditions. are linearly independent on the real line. rz n = 1. Yc = CJ cosx + Cz sinx. y(O) = 2. y".l r. . 26. (b) Prove that the equation y" . y"(O) = 2. 19.6y = 0. 32. In Problems 21 through 24. Y1 = x.3y = 6. Yp = .2xy' + 2y = 0.2y'. for any constant r. y(O) = 0. y"(l) = ll. + 11y' . y'(l) = 14.) 28. x 3y<3>..6y = 0. Suppose that y 1 and y2 are two functions such that Ly1 = f(x) 21. y" + y = 3x. . Y1 =ex. y(l) = 1. Yl =ex. Y3 = x 2 ex y<3>5y" +8y' 4y = 0. y(O) = 3. explain why the value of y" (a) is determined by the values of y(a) and y'(a). (Suggestion: Let y.(x) = x. . y"(O) = 0.6y" 15. Prove by the method of Problem 33 that the functions j. 1~ i ~ n are linearly independent on the whole real line. j. y(O) = 2. Yp =X+ 1 25. 17. y'(O) = 10. .4y = 0. Prove directly that the functions !1Cx) = 1. yz. and exp(r3 x) are linearly independent by showing that their Wronskian + 2y = 4 and y" + 2y = 6x. r2 .. Differentiate this equation twice. . _ 2x + _ 3 YcCJe Cze 2x. be the unique solution such that (i1)( ) . This problem indicates why we can impose only n initial conditions on a solution of an nthorder linear differential equation. y(O) = 1.4y = 12. Prove that an nthorder homogeneous linear differential equation satisfying the hypotheses of Theorem 2 has n linearly independent solutions y 1 . y".yp23. Yc = CJex + Cze 3x. a nonhomogeneous differential equation. and y" (O) = C if and only if C = 5.. Y3 = x 3 x 3y< 3> + 6x 2 y" + 4xy' . and /3(x) = x2 is nonzero for all x. y'(O) = 8. y(O) = 4. a  Y. 20. Use the result of Problem 28 and the definition of linear independence to prove directly that. yO> .5y =0 has a solution satisfying the conditions y(O) 24. r.
It can be shown that the Wronskian of n solutions y 1.x 2)y" + 2xy'. but we said little about how actually to find even a single solution. n linearly independent solutions of a given nthorder linear equation if it has constant coefficients. 39. Use the method of reduction of order as in Problem 37 to find a second linearly independent solution Yz· W' = y. Yt (x) = ex/2 x 2 y" .4y' 0 (x > 0). z cos x is one solution (for x > 0) of Bessel's equation of order ~. y 1(x) = x First note that y 1 (x) = x is one solution of Legendre's equation of order 1. Thus. Thence solve for v(x) = C ln x.2 we saw that a general solution of an nthorder homogeneous linear equation is a linear combination of n linearly independent particular solutions. a differential equation and one solution y 1 are given.(x + 2)y' + y = 0 (x > . + (2y. According to Problem 32 of Section 2. Prove this for the case n = 3 as follows: (a) The derivative of a determinant of functions is the sum of the determinants obtained by separately differentiating the rows of the original determinant. 40..x 2)y" .1 12 sinx.2xy' + p(x)y' + q(x)y = 0 (18) + 2y = 0..124 Chapter 2 Linear Equations of Higher Order 37.2y = 0 (1 < x < 1). First verify by substitution that y 1(x) = x t. use the fact that y 1(x) is a solution to deduce that Yt v" Then use the method of reduction of order to derive the second solution Yz(x) = 1. Then derive by reduction of order the second solution Y2(x) = x . Conclude that YI substitute y = vx 3 and deduce that xv" + v' = 0. The . 43.l n . and yj3 ) from the equation y< 3) +PI y" + P2y' + P3Y = 0. the equation must first be written in the form of ( 18) with leading coefficient 1 in order to correctly determine the coefficient function p(x). . explicitly and in a rather straightforward way. The method of reduction of order consists of substituting y 2(x) = v(x)y 1(x) in (18) and attempting to determine the function v(x) so that y 2(x) is a second linearly independent solution of ( 18). In each of Problems 38 through 42. 36.2 1. y 1(x) = x (x + 1)y". y 1(x) = ex (1.)v' = 0.. Y2. x 2y " +xy'. Y1 (3) y~ Y2 (3) y~ Y3 (3) (b) Substitute for y} 3). . (19) with a given homogeneous secondorder linear differential equation and a known solution y 1 (x). (1. (x) dx) for some constant K.p 1 W. Suppose that one solution y 1(x) of the homogeneous secondorder linear differential equation y" 38.1). starting with the readily verified solution y 1 (x) = x 3 of the equation x 2y".1. + py. But we can now show how to find. Yt(X) = x 3 + y = 0. is known (on an interval I where p and q are continuous functions). Yn of the nthorder equation y<n) +PI (x)y(nl) + · · · + Pn1 (x)y' + Pn(X)y = 0 + 9y = 0 (x > 0).X x 1 +x (for1 < x < 1). satisfies the same identity. and thereby obtain (with C = 1) the second solution y 2 (x) = x 3 In x. yi3 ). If y 1 (x) is known. and then show that W' = . Integration now gives Abel's formula. 42. (18). Frequently it is more convenient to simply substitute y = v (x) y 1(x) in the given differential equation and then proceed directly to find v(x).x(x + 2)y' + (x + 2)y = 0 (x > 0).9y = 4y" . (19) 44. Before applying Eq. lfiJ Homogeneous Eq]JatiOJ:lS with Constant Coef¥ic~ents In Section 2. After substituting y = v(x)y 1(x) in Eq. Integration of v' (x) then gives the desired (nonconstant) function v(x). the Wronskian W(y 1. 41. y 2 ) of two solutions of the secondorder equation y" +PI (x)y' + P2(x)y = 0 is given by Abel's's formula W(x) = K exp ( j p. then ( 19) is a separable equation that is readily solved for the derivative v' (x) of v(x).Sxy' 35. The solution of a linear differential equation with variable coefficients ordinarily requires numerical methods (Chapter 6) or infinite series methods (Chapter 3).
but for equations of higher degree we may need either to spot a fortuitous factorization or to apply a numerical technique such as Newton's method (or use a calculator j computer solve command). . . (1) precisely (3) This equation is called the characteristic equation or auxiliary equation of the differential equation in (1).1 we substituted y = erx in the secondorder equation ay" + by'+ cy = 0 to derive the characteristic equation ar 2 + br + c = 0 that r must satisfy. Finding the exact values of these zeros may be difficult or even impossible. the quadratic formula is sufficient for seconddegree equations.. For example. in Section 2. every nthdegree polynomial. This suggests that we try to find r so that all these multiples of erx will have sum zero. we substitute y = and with the aid of Eq. that is. if we substituted y = erx in Eq. ( 1).3 Homogeneous Equations with Constant Coefficients 125 general such equation may be written in the form (1) where the coefficients a0. is 'reduced to the solution of this purely algebraic equation. . 0.. (1). + a2r 2 + azr + ao) = 0. with the constant coefficients depending on r and the coefficients ak. (3). a2. erx Because erx is never zero. The Characteristic Equation We first look for a single solution of Eq. To carry out this technique in the general case. According to the fundamental theorem of algebra. Our problem. in which case y = erx will be a solution of Eq. (1). Hence. erx (anrn + anlrnl + . (2) we find the result to be erx in Eq. though not necessarily distinct and not necessarily real.2. (1).such as the one in Eq. . then. each term would be a constant multiple of erx. an are real constants with an f. we see that y = when r is a root of the equation will be a solution ofEq.has n zeros. and begin with the observation that (2) so any derivative of erx is a constant multiple of erx. a 1 . .
and so the characteristic equation has the three distinct real roots r and r = 2. . . so c 2 = 2. 70 in the coefficients c 1. so let us first examine the simplest casein which the characteristic equation has n distinct (no two equal) real roots r 1 . r 2. + c2e . Then the second equation gives c 3 = 5. (1).. • . we have proved Theorem 1. THEOREM 1 Distinct Real Roots If the roots r1. In summary.lOr = 0.2) these n solutions are linearly independent on the entire real line. The situation is slightly more complicated in the case of repeated roots or complex roots ofEq. Then the given initial conditions yield the linear equations y(O) = CJ + y'(O) = y"(O) = + C3 = Sc2 + 2c3 = 25c2 + 4c3 = C2 7. Thus the desired particular solution is y(x) = 2e.126 Chapter 2 Linear Equations of Higher Order Distinct Real Roots Whatever the method we use. r n of the characteristic equation in (3) are real and distinct.Sx + 5e2x. . y' (0) = 0.Sx + c3e2x . (1).lOy'= 0. then (4) is a general solution of Eq. . r2. Theorem 1 gives the general solution y(x) = CJ = 0. Because e0 = 1. r n.. Then the functions are all solutions of Eq. y" (0) = 70. y(O) = 7. and (by Problem 34 of Section 2. c2 .10) = r(r + 5)(r. Then we can always write a general solution of the differential equation.2y' (0) 35c2 = 70. (3). We solve by factoring: r(r 2 + 3r.. and c3 • The last two equations give y" (0) . Solution The characteristic equation of the given differential equation is r 3 + 3r 2 . r = 5.. Example 1 Solve the initial value problem y<3 ) + 3y". let us suppose that we have solved the characteristic equation.2) = 0. and finally the first equation gives c 1 = 0. 0.
Consequently.a2 ) • • • (x .a)(D ..b)(y' .b)(D . then.by).a)(D .ay.b) = D 2 . it can be shown by induction on the number of factors that an operator product of the form (D .a) y. The proof of the formula in (7) is (D.(a + b)D + ab.a)(D ..by) = y" . A firstdegree polynomial operator with leading coefficient 1 has the form D .b(y'.ay = y' . 2. Similarly. The problem. . We also denote by D differentiation with respect to x . the algebra of polynomial differential operators closely resembles the algebra of ordinary real polynomials.2.a 1)(x . We see here also that (D .a)y = Dy. + a2 / 2) + a1Y 1 + aoy of y and its first n derivatives. is to produce the missing linearly independent solutions.by) = D(y' .in the same way as does an ordinary product (x . For example.a) y (7) for any twice differentiable function y the following computation: = y(x). it is a polynomial differential operator. It operates on a function y = y (x ) to produce (D.3 Homogeneous Equations with Constant Coefficients 127 Polynomial Differential Operators If the roots of the characteristic equation in (3) are not distinctthere are repeated rootsthen we cannot produce n linearly independent solutions of Eq. the operator Lin (5) may be written (6) and we will find it useful to think of the righthand side in Eq. In terms of D.(a + b)y' + bay = D(y' . it is convenient to adopt "operator notation" and write Eq.an) of linear factors. where a is a real number.b)y = (D. we obtain only the two functions ex and e 2 x. 2. (1) in the form Ly = 0. where the operator operates on the ntimes differentiable function y(x) to produce the linear combination Ly = an/n) + an! yen!) + .a)(y' .ay ) = (D . The important fact about such operators is that any two of them commute: (D.a(y'.ay) = (D . (6) as a (formal) nthdegree polynomial in the "variable" D.b)(D. and 2.a .a2 ) · · • (D an) expandsby multiplying out and collecting coefficients. so that = djdx the operation of and so on.a 1)(D.b)y = (D . (1) by the method of Theorem 1. with x denoting a real variable. For this purpose.(b + a)y' + aby = y" . if the roots are 1.ay) .
r 1)ky ] = 0.1 > 1. For example. .. But this is so if and only if U(X ) = CJ + C2X + c 3x 2 + · · · + Ck XkI.r1) [ ue' 1x ] = (Du)e' 1x + u(r1 e' 1x )  r1 (ue' 1x ) = (Du) e' 1x. xkl e'lx of the original differential equation Ly = 0.1 solutions. Consequently. because the equation is of order k + 1. suppose that Eq. 1 (11) where u (x) is a function yet to be determined. (1 0) is y (x ) = U e ' 1x = ( CJ In particular. ro of multiplicity 1 and r 1 of multiplicity k = n . however.128 Chapter 2 Linear Equations of Higher Order Repeated Real Roots Let us now consider the possibility that the characteristic equation (3) has repeated roots. This is. we need k + 1 linearly independent solutions in order to construct a general solution. To find the missing k .. we see here the additional solutions x e'l x. Hence y = ue' 1x will be a solution of Eq. a polynomial of degree at most k . + C2X + C3X 2 + · · · + Ck Xk l) e ' 1x . The fact that y 1 = e'I x is one solution of Eq. Observe that (D .ro)[(D. (10) suggests that we try the substitution y(x) = u(x )y 1(x) = u(x)e' x . the corresponding operator Lin (6) can be written as (9) the order of the factors making no difference because of the formula in (7). (3) has only two distinct roots. it follows that (13) for any sufficiently differentiable function u(x ). (12) Upon k applications of this fact. Two solutions of the differential equation Ly = 0 are certainly Yo = e'0 x and y 1 = e'Ix.r 1 replaced with an arbitrary polynomial operator. not sufficient. (3) can be rewritten in the form (8) Similarly. . Then (after dividing by an) Eq. Hence our desired solution of Eq. The preceding analysis can be carried out with the operator D .1. When this is done. we note that Ly = (D. (10) if and only if Dku = u(k) = 0. x 2 e'l x . every solution of the kthorder equation (10) will also be a solution of the original equation L y = 0. Hence our problem is reduced to that of finding a general solution of the differential equation in (10). . the result is a proof of the following theorem.
1... "  .  . Thus a root of multiplicity k corresponds to k linearly independent solutions of the differential equation. i 3 = .i. The characteristic equation is 9r 5 contributes  6r 4 + r 3 = r\9r 2  6r + 1) = r\3r 1) 2 = 0. according to Problem 29 of Section 2. while the double root r = t contributes c4 exl 3 + c5 xexl 3 • general solution of the given differential equation is Hence a • ComplexValued Functions and Euler's Formula Because we have assumed that the coefficients of the differential equation and its characteristic equation are real... .3 Homogeneous Equations with Constant Coefficients 129 THEOREM 2 Repeated Roots If the characteristic equation in (3) has a repeated root r of multiplicity k. we get eiB =L oo n= O = 1+ t + . i 4 = 1. xerx. x 2erx. and xk!erx involved in (14) are linearly independent on the real line.2. ) . The triple root r + c2xe0·x + c3x 2e 0·x = cr + c2x + c3x 2 to the solution. any complex (nonreal) roots will occur in complex conjugate pairs a±bi where a and b are real and i = .+ · · · .2... we recall from elementary calculus the Taylor (or MacLaurin) series for the exponential function et If we substitute t so on. the k functions erx. and = t n= O (W)n n! =1+W++··· 2! 3! 4! 5! = g2 ie3 g4 ies (1  g2 2! + g4 4! . then the part of a general solution of the differential equation in (1) corresponding to r is of the form (14) We may observe that..J=T... Example 2 Find a general solution of the fifthorder differential equation 9/5) Solution  6/4) + y(3) = 0. This raises the question of what might be meant by an exponential such as e(a+bi)x. . . 0 It has the triple root r = 0 and the double root r = cre 0·x t..+ . n! 2! 3! 4! tn t2 t3 t4 = i() in this series and recall that i 2 = .. . ) + i (e  g3 3! + es 5! . To answer this question. ··· ·········· .+ .
A complexvalued function F of the real variable x associates with each real number x (in its domain of definition) the complex number F(x) = f(x) + ig(x). The proof of this assertion is a straightforward computation based on the definitions and formulas given earlier: Dx(erx) = D x (eax cosbx) + iDx (eax sinbx) = (aeax cosbx.beax sinbx) + i (aeax sinbx + beax cosbx) = (a+ bi)(eax cosbx + ieax sinbx) = rerx. we define the derivative F' of F by F'(x) = J'(x) + ig'(x). we define the exponential function eZ. (17) The realvalued functions f and g are called the real and imaginary parts. for z = x + iy an arbitrary complex number. to be (16) Thus it appears that complex roots of the characteristic equation will lead to complexvalued solutions of the differential equation. where r = a ± bi. If they are differentiable. respectively. We note from Euler's formula that (19a) and (19b) The most important property of erx is that (20) if r is a complex number. (15) This result is known as Euler's formula. Because of it. . this implies that e 10 =cost/+ i sine. We say that the complexvalued function F(x) satisfies the homogeneous linear differential equation L[F(x)] = 0 provided that its real and imaginary parts in (17) separately satisfy this equationso L[F(x)] = L[f(x)] + iL[g(x)] = 0. respectively. of F. The particular complexvalued functions of interest here are of the form F (x) = er x. (18) Thus we simply differentiate the real and imaginary parts ofF separately.130 Chapter 2 Linear Equations of Higher Order Because the two real series in the last line are the Taylor series for cost/ and sine.
So Theorem 3 (with a = 0) gives the general solution y(x) = c 1 cosbx + c 2 sinbx. ! THEOREM 3 Complex Roots If the characteristic equation in (3) has an unrepeated pair of complex conjugate roots a ± bi (with b :f.4 y' + 5 y = 0 for which y(O) = 1 and y'(O) = 5.2) 2 +1= 0. while the choice cl = !i. This yields the following result.3 Homogeneous Equations with Constant Coefficients 131 Complex Roots It follows from Eq. the choice C1 = C2 = gives the realvalued solution Yt (x) = eax cos bx. For instance. so r .2.2 = ±.C2)eax sinbx. • Example 4 Find the particular solution of y" . Hence Theorem 3 with a = 2 and b = 1 gives the general solution y(x) = e2x(c 1 cosx + c2 sinx).bi are simple (nonrepeated). Solution Completion of the square in the characteristic equation yields r2  4r +5 = (r . then the corresponding part of a general solution of Eq. (1) has the form (21) Example 3 The characteristic equation of y" + b2 y = 0 (b > 0) is r 2 + b 2 = 0.J=T = ±i. with roots r = ±bi. (1) is y(x) = Cter!x = + C2er2X = Cle(a+bi)x + C2e(abi)x C1eax(cosbx + i sinbx) + C2eax(cosbx. then the corresponding part of a general solution of Eq. . (20) that when r is complex Uust as when r is real). If the complex conjugate pair of roots r 1 = a + bi and r2 = a . erx will be a solution of the differential equation in ( 1) if and only if r is a root of its characteristic equation. Thus we obtain the complex conjugate roots 2 ± i (which could also be found directly using the quadratic formula).i sinbx) + C2)eax cosbx + i(CI . y(x) = (Ct where the arbitrary constants C 1 and C 2 can be complex. 0). c2 = !i gives the independent realvalued solution Y2(X) = eax sinbx.
It follows that c2 = 3.1. y) + iy = reiB of the complex number z. in terms of the modulus r = x 2 + y2 > 0 of the number z and its argument e indicated in Fig. we find that and ~= (2e i3rr/2)1 /2 = Jle i3rr / 4 = J2 ( cos 3: + 3:) i sin = 1 + i.3. Similarly. J Example 5 Solution 4S + 4 y = 0.1 ± i . Thus the four (distinct) roots of the characteristic equation are r = ± (± 1+ i ). This form follows from Euler's formula upon writing z= r X (~ + i~) = r(cose + i sine)= rei 8 FIGURE 2. 1 ± i and .132 Chapter 2 Linear Equations of Higher Order Then so the initial conditions give y(O) = c1 = 1 and y'(O) = 2ct + c2 = 5. give a general solution of the differential equation yC4 ) + 4y = 0. Modulus and argument of the complex number X+ iy. 2.Jr denotes (as usual for a positive real number) the positive square root of the modulus of z. i = e 3rr12 . so i = eirr/ 2 . Find a general solution of y < The characteristic equation is and its four roots are ±J±2T. • .1. For instance. the imaginary number i has modulus 1 and argument n j2. and so the desired particular solution is y(x) = e2x(cosx + 3 sinx).3. • (22) In Example 5 below we employ the polar form y z =x (x. These two pairs of complex conjugate roots. Another consequence is the fact that the nonzero complex number z = reie has the two square roots (23) where . Since i = eirr/ 2 and i = ei 3rrl 2 .
2. and ± 10 of the constant term 10. sin bx).. (24) are linearly independent.: k 1 that appear in Eq. The three roots we The roots of this quotient are the complex conjugates 1 have found now yield the general solution • . (24) It can be shown that the 2k functions xPeax cosbx. then the corresponding part of the general solution has the form (AI+ Azx + . and division of the former into the latter produces as quotient the quadratic polynomial r 2 + 2r +5 = (r + 1)2 + 4.lOy= 0 + r10 = 0. + Bkxkl)e(a. we see that the characteristic equation has as its roots the conjugate pair 3 ± 2i of multiplicity k = 2.. By trial and error (if not by inspection) we discover the root 2. cosbx p=O + d. ±5.3 Homogeneous Equations with Constant Coefficients 133 Repeated Complex Roots Theorem 2 holds for repeated complex roots. ±2. 0. Often the most difficult part of solving a homogeneous linear equation is finding the roots of its characteristic equation. + Akxkl)e(a+bi)x + (BI + Bzx + . Hence Eq..: p. The factor theorem of elementary algebra implies that r . the only possible rational roots are the factors ± 1.. If the conjugate pair a ± bi has multiplicity k.. ± 2i .bi)x k1 = L:xpeax(c.10. Example 7 illustrates an approach that may succeed when a root of the characteristic equation can be found by inspection. Example 6 Solution Find a general solution of (D 2 + 6D + 13)2 y = 0. xPeax sinbx. (24) gives the general solution In applications we are seldom presented in advance with a factorization as convenient as the one in Example 6. By completing the square.. Example 7 The characteristic equation of the differential equation / 3) + y'. is the cubic equation r3 By a standard theorem of elementary algebra.2 is a factor of r 3 + r .
Find the general solution.:rr < () ~ :rr. The solution can be read directly from the list of roots.2y = 0 29. y<4 J + 3y". y' (O) = 10. y 0 l + 3y". 5. y"(O) = 1 26.4y = 0 y" + 3y' . 9y" + 6y' + 4y = 0.iy' + 6y = 0 Find a general solution of y" = (2 + 2i . 41. 2.y< 3) + y". = 0. Solve the initial value problem In Problems 33 through 36. 36. (Suggestion: Impose the given initial conditions on the general solution . 48. 10. y'(O) = 4 23. y'(O) = 1 24.J3 in the form rei&. 2y<3l . 7. • Find the general solutions of the differential equations in Problems 1 through 20. y".3y". 2y(3) .12y' + 9y = 0 y" + 8y' + 25y = 0 y<4) . First find a small integral root of the characteristic equation by inspection.500y y(O) = 0. where r ~ 0 and . Find a general solution of y". 5. 40.J3.6y' + 25y = 0. 12. 47.8y = 0 32. 21. 13._ ~·~·~·.5y'. y". 0. and y<3 l(Q) = 7.7y' + 3y = 0 y" + 5 y' + 5 y = 0 y" . Note in each case that the roots are not complex conjugates.x sinx yC 3l = y. y = e 2xfJ 6y<4 l + 5y<3 l + 25y" + 20y' + 4y = 0. 3. 44.2y = 0 43. y'(O) = 11 22. 2 y" . y = e 3x 3/3) . y =cos 2x 9y<3 J + lly" + 4y' . 1 + i. y"(O) = 13.2 + 2i . y' (O) = 4. Use the quadratic formula to solve the following equations. 39. y<4) . 33.3 y' = 0 2y" .·•• .3y". 1.5y' .y" . 8. Write a general solution of this homogeneous differential equation.2y" + 12y' .y' .5y" + lOOy'.8y" + 16y = 0 6y(4 ) + lly" + 4y = 0 y< 3l + y" . 46. 42. Solve the initial value problem y< 3l .2iy' + 3y = 0. y'(O) = 12. y<4l + y(3).2y' = 0.3/3 ) + 3y" . y(O) = 1. then factor by division. 0.y = 0 y<4J + 2y< 3l + 3y" + 2y' + (r 2 + r + 1) 2 . find a linear homogeneous constantcoefficient equation with the given general solution. 2 ± 3i. y'(O) = 1. 20. y(O) = 1.Jrei&/2 • Find the square roots of the numbers 2 . 27. 35.J3) y.y' = 0 9/3) + 12y" + 4y' = 0 y<4 J . It is y(x) = Solution c. + CzX + c3x 2 + c4x 3 + cse 3x + c6eSx + C7XeSx + e2x (cs cos 3x + cg sin 3x) + xe2x (cw cos 3x + c11 sin 3x). Find general solutions ofthe equations in Problems 27 through 32. 9. y(O) = 1. y(O) = 3. Solve the initial value problems given in Problems 21 through 26. and 2 ± 3i. y<4J = 16y y = 0 (Suggestion: Expand In Problems 39 through 42. y(3) + 3y" + 4y' . y"(O) = 5 y(x) y(x) y(x) y(x) =(A+ Bx + Cx 2 )e2x = Ae 2x + Bcos2x + Csin2x =A cos 2x + B sin2x + C cosh2x + D sinh2x = (A+ Bx + Cx 2 ) cos 2x + (D +Ex+ Fx 2 ) sin 2x Problems 43 through 47 pertain to the solution of differential equations with complex coefficients. y<4 J + 18y" + 81y = 0 18. (a) Use Euler's formula to show that every complex number can be written in the form rei&. = 14. Find a general solution of y". and 1 + i . 2.6y' + 13y = 0 5y(4 ) + 3y(3) = 0 37. y<3 l +lOy"+ 25y' = 0. 15.l4y = 0. y" (O) = 250 given that y 1 (x) e5x is one particular solution of the differential equation. y". y(O) = 3. y(3) + 27y = 0 30. 11.8y<3l + 16y" = 0 / 4) . 5.3y'. 0. 38.) 4. 0. (c) The two square roots of rei& are ±. y'(O) = 0. y< 3) + 3y" . 17.J3 and .4y = 0 16. y(O) = 7. y(O) = 3. (a) x 2 + ix + 2 = 0 (b) x 2  2ix +3 = 0 45. 34.2i . 6. Find a function y(x) such that y<4 l(x) = y< 3l(x) for all x and y(O) = 18.4y' + 3y = 0.lOy = 0 y" + 6y' + 9y = 0 4y". one solution of the differential equation is given. (b) Express the numbers 4. 19. 3i.54y = 0. 3y<3l + 2y" = 0. y'(O) = y" (O) = 0.·• •••• • • •V• ·o•   Example 8 The roots of the characteristic equation of a certain differential equation are 3.4y = 0 28.6y = 0 31. y = e . y"(O) = 3 25.134 Chapter 2 Linear Equations of Higher Order •• •••• .8y = 0.
57. 2. x 2 y" + xy' + 9y = 0 x 2 y" + 7xy' + 25y = 0 x 3 y"' + 6x 2 y" + 4xy' = 0 x 3 y"'. 51. (0) = 0. A massspringdashpot system. .3x 2 y" + xy' = 0 x 3 y"' + 6x 2 y" + 7xy' + y = 0 Mechanical Vibrations The motion of a mass attached to a spring serves as a relatively simple example of the vibrations that occur in more complex mechanical systems.2.) 49. 54. We take x > 0 when the spring is stretched. so that it can move only back and forth as the spring compresses and stretches. 52.1. Because this is the same as the displacement x of the mass m from its m c I I Equilibrium position FIGURE 2.3.2. Solve the initial value problem FIGURE 2. a . We consider a body of mass m attached to one end of an ordinary spring that resists compression as well as stretching. Y1 (0) = y~(O) = 1 and Yz(O) = y.4. According to Hooke's law. (25) nevertheless has two linearly independent solutions Y1 (x) and Yz (x) defined for all x such that • • • Make the substitution v = ln x of Problem 51 to find general solutions (for x > 0) of the Euler equations in Problems 52 through 58. the substitution v = ln x (x > 0) transforms the secondorder Euler equation ax 2 y" + bxy' + cy = 0 to a constantcoefficient homogeneous linear equation. For many such systems. The differential equation y" + (sgnx)y =0 ax 3 y"' + bx 2 y" + cxy' + dy = 0 has the discontinuous coefficient function (where a. the other end of the spring is attached to a fixed wall. Graphs of y 1 (x) and Yz(x) in Problem 50.+ dy = dv ~ 0.1 = 0. and thus x < 0 when it is compressed.x 2 y" + xy' = 0 x 3 y"' + 3x 2 y" + xy' = 0 x 3 y"'.) The graphs of these two solutions are shown in Fig. if X < 0.+ (b. 56. the analysis of these vibrations is a problem in the solution of linear differential equations with constant coefficients.1. (Suggestion: Each Yi (x) will be defined by one formula for x < 0 and by another for x ~ 0.4.2. c. as shown in Fig. 58.b + 2a). Show that Eq. 53. (25) at each point x =1= 0. d are constants) into the constantcoefficient equation sgnx = 11 +1 if X > 0.4 Mechanical Vibrations where a and f3 are the complex conjugate roots of r 3 . the restorative force Fs that the spring exerts on the mass is proportional to the distance x that the spring has been stretched or compressed. y(O) = y' (0) = y" (0) = 0.3a)dv 3 dv 2 ~y ~y + (c. Each satisfies Eq. to discover that 135 1 ( ex+ 2exfZ cos xv'3) y(x) = 3 2is a solution. b. Denote by x the distance of the body from its equilibrium positionits position when the spring is unstretched.3. 55. Show similarly that this same substitution transforms the thirdorder Euler equation (25) y<4) = y<3) + y" + y' + 2y. 2. Each has a continuous derivative at x = 0. Assume that the body rests on a frictionless horizontal plane. 50. According to Problem 51 in Section 2. 2y<3 \0) = 30.1.
We refer to the motion as free in this case and forced in the case F(t) "/:. (3). the mass is subjected to a given external force FE = F(t).2.=mx ' dt 2 we obtain the secondorder linear differential equation mx" +ex'+ kx = F(t) (3) that governs the motion of the mass.2. That is. it follows that Fs = kx.1 shows the mass attached to a dashpota device. 0. For an alternative example. . we might attach the mass to the lower end of a spring that is suspended vertically from a fixed support. If there is no dashpot (and we ignore all frictional forces).136 Chapter 2 Linear Equations of Higher Order equilibrium position. Using Newton's law d2x " F=ma = m . then we ask you to show in Problem 9 that y satisfies Eq. we replace F(t) with 0 in Eq. specifically. (1) The positive constant of proportionality k is called the spring constant. then we set c = 0 in Eq. dx FR = cv =c. Note that Fs and x have opposite signs: Fs < 0 when x > 0.4.6. (1) with Fs = Wand x = s0 . 2.4. We will defer discussion of forced motion until Section 2. Figure 2. dt (2) The positive constant c is the damping constant of the dashpot. that my"+ cy' + ky = F(t) (5) if we include damping and external forces (meaning those other than gravity). so that so = mgjk.4. like a shock absorber. A mass suspended vertically from a spring. then the total force acting on the mass is F = Fs + FR + F£. mg = ks0 . Fs > 0 when x < 0. we may regard Eq. If there is no external force. describes free motion of a mass on a spring with dashpot but with no external forces applied. measured downward from its static equilibrium position. This gives the static equilibrium position of the mass. (2) as specifying frictional forces in our system (including air resistance to the motion of m). More generally. If y denotes the displacement of the mass in motion. in addition to the forces Fs and FR. as in Fig. that provides a force directed opposite to the instantaneous direction of motion of the mass m. that is. In this case the weight W = mg of the mass would stretch the spring a distance s0 determined by Eq. it is damped motion if c > 0. (3). If. Thus the homogeneous equation mx" + ex' + kx = 0 (4) FIGURE 2. We assume the dashpot is so designed that this force FR is proportional to the velocity v = dxjdt of the mass. (3) and call the motion undamped.
~mv2 = ~m (ds)2 = ~mL2 (d8)2 2 2 dt 2 dt We next choose as reference point the lowest point 0 reached by the mass (see Fig. The distance along the circular arc from 0 to m is s = L8.3.3. (7) . The fact that the sum of T and V is a constant C therefore gives 2 (d8) 1 mL +mgL(lcos8)=C.4. However. In fact. the result is an equation in the form of Eq. a simple pendulum consists of a mass m swinging back and forth on the end of a string (or better.cos8) . It therefore seems reasonable to simplify our mathematical model of the simple pendulum by replacing sin 8 with e in Eq. sin 8 and 8 agree to two decimal places when 181 is at most n/12 (that is. (6). and it may be instructive to see the energy method in a simpler application like the pendulum. 2. The simple pendulum. then sin 8 ~ 8 (this approximation obtained by retaining just the first term in the Taylor series for sin 8).4. so V = mgL(l. Now recall that if 8 is small. We may specify the position of the mass at time t by giving the counterclockwise angle 8 = 8(t) that the string or rod makes with the vertical at timet. derivations of differential equations based on conservation of energy are often seen in more complex situations where Newton's law is not so directly applicable. 2 dt 2 We differentiate both sides of this identity with respect to t to obtain so (6) after removal of the common factor mL 2 (d8jdt). we will apply the law of the conservation of mechanical energy. In a typical pendulum clock.4. so the velocity of the mass is v = dsjdt = L(d8jdt). 2. and therefore its kinetic energy is T = FIGURE 2. If we also insert a term c8' to account for the frictional resistance of the surrounding medium. according to which the sum of the kinetic energy and the potential energy of m remains constant.3). as shown in Fig. (3) and (5) stems from the fact that it describes the motion of many other simple mechanical systems. (4): e" + c8' + k8 = 0. This differential equation can be derived in a seemingly more elementary manner using the familiar second law F = rna of Newton (applied to tangential components of the acceleration of the mass and the force acting on it). 8 would never exceed 15°. To analyze the motion of the mass m.4 Mechanical Vibrations 137 The Simple Pendulum The importance of the differential equation that appears in Eqs. for example. For example. Then its potential energy V is the product of its weight mg and its vertical height h = L (1 .cos 8) above 0.2. 15°). a massless rod) of length L.
(12) Thus the mass oscillates to and from about its equilibrium position with . In any event. Free Undamped Motion If we have only a mass on a spring. then Eq. Instead. although tan a= BjA. n/2) given by a calculator or computer. we choose constants C and a so that cosa = c' A and sma = . (8) as x II 2x = 0 . We might. where either A or B or both may be negative. .4. (7) will probably not describe accurately the actual motion of the pendulum over a long period of time. (8) It is convenient to define wo=ff (9) and rewrite Eq. expect the effects of the discrepancy between () and sin() to accumulate over a period of time. tan.4. 2. we first analyze free undamped motion and then free damped motion. we find that x(t) = C cos(wot a). if A > 0. c' B (II) as indicated in Fig. Note that. (8') is x(t) = Acoswot + Bsinwot.n/2 < x < n/2). however. B > 0 (first quadrant). The angle a. With the aid of the cosine addition formula. (10) To analyze the motion described by this solution.n/2. so that Eq.138 Chapter 2 Linear Equations of Higher Order where k = gjL. if A < 0 (second or third quadrant). a is the angle between 0 and 2n whose cosine and sine have the signs given in (11).4. the angle a is not given by the principal branch of the inverse tangent function (which gives values only in the interval . In the remainder of this section. with neither damping nor external force. Note that this equation is independent of the mass m on the end of the rod. ) +C smwot = C(cosa coswot .1 (8/A) 2n + tan.1 (8/A) is the angle in ( . (3) takes the simpler form mx" + kx = 0. Thus A FIGURE 2.1 (B/A) if A > 0. where tan.1 (B/A) a= { n + tan.4. ) + sma smwot . from (I 0) and ( 11) we get A x(t ) = C ( C cos wot B . B < 0 (fourth quadrant). +w0 (8' ) The general solution of Eq.
cvo.= T 2n 2rr ~ 1. (10). that is. period of oscillation. and the time lag a o=cvo are indicated.6283 s and with frequency 1 cvo 10 v= .4.4. then find the amplitude C and phase angle a by carrying out the transformation of x(t) to the form in Eq. and Such motion is called simple harmonic motion. If the initial position x (0) = x 0 and initial velocity x' (0) = v0 of the mass are given. The period of the motion is the time required for the system to complete one full oscillation.. Solution The spring constant is k SOx = 0. which measures the number of complete cycles per second. Amplitude 139 2. as indicated previously. (8) yields ~x" x" + + 100x = 0. 0 )) = C cos(cvo(t. Note that frequency is measured in cycles per second.= . If time t is measured in seconds. its frequency is 1 cvo V==T 2rr (14) in hertz (Hz). whereas circular frequency has the dimensions of radians per second.) Find the position function of the body as well as the amplitude. Hence it will oscillate with period 2rr 2rr T = cvo = lO ~ 0. we first determine the values of the coefficients A and Bin Eq. the circular frequency cv0 has dimensions of radians per second (radjs). Consequently. frequency. Circular frequency 3. Phase angle C. = (100 N)/(2 m) = 50 (N/m).5. a. the circular frequency of the resulting simple harmonic motion of the body will be cv0 = .5915 Hz.5. so is given by T=X 2rr cvo (13) seconds.JIOO = 10 (rad/s).a)= C cos ( cvo (t. .2. is shown in Fig. so Eq. (Note that these initial conditions indicate that the body is displaced to the right and is moving to the left at time t = 0.4 Mechanical Vibrations 1.8)) FIGURE 2. Example 1 A body with mass m = ~ kilogram (kg) is attached to the end of a spring that is stretched 2 meters (m) by a force of 100 newtons (N). A typical graph of a simple harmonic position function 1T 1 x(t) = C cos(cvot. Simple harmonic motion. where the geometric significance of the amplitude C. It is set in motion with initial position x 0 = 1 (m) and initial velocity v0 = 5 (rnls). (12). and time lag of its motion. 2. the period T.
~. the position function of the body takes the form x(t) ~ ~Jscos(lOt.8195).6. we write x (t) = v'5(2 1 ) 2 v'5 cos lOt.. = 1 and B = .5820 s.140 Chapter 2 Linear Equations of Higher Order We now impose the initial conditions x (0) function x(t) =A cos lOt+ B sin lOt It follows readily that A = 1 and x' (0) = 5 on the position with x'(t) =lOA sin lOt+ lOB cos lOt. 2. Hence its amplitude of motion is To find the time lag. where the phase angle a satisfies 2 .< 0. 1 cosa = v'5 > 0 and sma = . • . With the amplitude and approximate phase angle shown explicitly.5.4. v'5 Hence a is the fourthquadrant angle and the time lag of the motion is a 0= wo ~ 0. and its graph is shown in Fig. 1 .v'5 sin lOt = v'5 2 cos(lOt  a) . so the position function of the body is x(t) =cos lOt  2 sm lOt.
118.5 c .J4kiri.JK7iii.2. Graph of the position function = C cos(w0 t . r2 + w6 = 0 of Eq. Solution curves are graphed with the same initial position x 0 and different initial velocities.628.a) in Example 1.4.5 2.6. (15) .7 shows some typical graphs of the position function for the overdamped case.4. both of which are negative. (16) The characteristic equation r 2 + 2pr r1.4. It is easy to see that x (t) + 0 as t + +oo and that the body settles to its equilibrium position without any oscillations (Problem 29). we chose xo a fixed positive number and illustrated the effects of changing the initial velocity v0 .5 3 0. Then ( 17) gives distinct real roots r 1 and r2 . the differential equation we have been studying takes the form mx" + ex' + kx = 0. we are dealing with a strong resistance in comparison with a relatively weak spring or a small mass. Free Damped Motion With damping but no external force. x" where w0 = + 2px' + w5x c 2m = 0. Overdamped motion: x(t) = c 1er 11 + c2 e'2 1 with r 1 < 0 and r2 < 0.4 Mechanical Vibrations X 141 T 0. alternatively.7. with amplitude C ~ 1. period T ~ 0. (19) FIGURE 2. and we distinguish three cases.582. Because cis relatively large in this case. The position function has the form 0 OVERDAMPED CASE: c > Ccr(c2 > 4km). In every case the wouldbe oscillations are damped out. . and time lag 8 ~ 0. (15) has roots = p ± (p 2  w6) 112 (17) that depend on the sign of pw=0 2 2 c2 4m 2 k m c2 4km 4m 2  (18) The critical damping Ccr is given by Ccr = according as c > Ccn c = Ccr• or c < Ccr· .5 1 x(t) FIGURE 2. Figure 2.> 0. is the corresponding undamped circular frequency and p = .
4. The motion is not actually periodic. Most of these quantities are shown in the typical graph of underdamped motion in Fig. but it is nevertheless useful to call cv 1 its circular frequency (more properly.4. In the critically damped case. Solution curves are graphed with the same initial position x 0 and different initial velocities.p ± i . 2. and they resemble those of the overdamped case (Fig. in accord with the timevarying FIGURE 2.4. the body passes through its equilibrium position at most once.142 Chapter 2 Linear Equations of Higher Order CRITICALLY DAMPED CASE: c Ccr(c 2 4km).P 1 an integral multiple of rr . and C e . we may rewrite Eq.pt its timevarying amplitude. 2. 2.8. that is. (20) as so x(t) = cepr cos(cv1t. Thus the action of the dashpot has at least two effects: 1.4. 2. its pseudofrequency). The graph of x(t) lies between the "amplitude 1 and touches them when cv t . but even a slight reduction in resistance will bring us to the remaining case. . Underdamped oscillations: X(t) = Cept COS(Wtt  a).jw6  The characteristic equation now p 2 . .P1 and x = ce.a) (23) where cosa .j4km .a is envelope" curves x = ce.c 2 2m (22) Using the cosine addition formula as in the derivation of Eq. / X= + CePt 0 The solution in (22) represents exponentially damped oscillations of the body around its equilibrium position.a) '/ / '.7). (17) gives equal roots r 1 = r 2 = . It exponentially damps the oscillations. (21) that in this case cv 1 is less than the undamped circular frequency cv0 . = c· B a I x = CeP1 cos(w 1t . It slows the motion. = . T1 = 2rr 1cv 1 its pseudoperiod of oscillation. and the general solution is (21) FIGURE 2. the one that shows the most dramatic behavior. (12). the resistance of the dashpot is just large enough to damp out any oscillations. so T1 is larger than the period T of oscillation of the same mass without damping on the same spring.9. so the general solution is = = (20) Because e .8. UNDERDAMPED CASE: c < Ccr(c2 < 4km). 0 has two complex conjugate roots . Critically damped motion: x(t) = (ct + c2t)ept with p > 0. Note from Eq. Some graphs of the motion in the critically damped case appear in Fig. the dashpot decreases the frequency of the motion. c A and sma .pt > 0 and c 1 + c2 t has at most one positive zero.p of the characteristic equation. amplitude.4.9. In this case.w. where CVJ _J w0 2  p2  _ . and it is clear that x(t) + 0 as t + +oo.
1 (cos 59 t. • ~ · '" " '••~'o. damping typically also delays the motion furtherthat is. It follows that x(O) =A= 1 whence we find that A the body is and x'(O) =A+ B59 = 5. Recall that m = and k = 50.5915 > v 1 in Example 1).. = 1 and B = 4/ . ·  Example 2 The mass and spring of Example 1 are now attached also to a dashpot that provides 1 N of resistance for each meter per second of velocity.5836 Hz (as compared with T ~ 0. The characteristic equation r 2 + 2r + 100 = (r r2 = 1 ± . The new (pseudo)period and frequency are 2rr T1 = . its new frequency and pseudoperiod of motion. The mass is set in motion with the same initial position x(O) = 1 and initial velocity x'(O) = 5 as in Example 1.1 (A cos J99 t + 1) 2 + 99 0 has roots r 1. its new time lag.1 (A cos J99 t + B sin J99 t) + J99 e. so the general solution is x(t) = e.2. Thus the new position function of x(t) = e. (4) is ~x" + x' +SOx = 0. that is. increases the time lagas compared with undamped motion with the same initial conditions. Hence Eq.•N. Now find the position function of the mass.J99 ~ 9. the new circular (pseudo)frequency is w 1 = . and the times of its first four passages through the initial position x = 0. 1 v99 IN'\ ~ 0. + B sin J99 t).6315 s and v1 =  T1 =  WI 2rr =  _J99 2rr ~ 1.6283 < T 1 and v ~ 1.4 Mechanical Vibrations 143 As the following example illustrates. we are now given c = 1 in mks units.9499 (as cornpared with wo = 10 in Example 1).1 (A sin J99 t + B cos J99 t).J99 i. it is better practice in a particular case to set up the differential equation and then solve it directly. We now impose the initial conditions x (0) = 1 and x' (0) = 5 on the position function in (23) and the resulting velocity function x' (t) = e. o.= 2rr w.J99.~ sin 59 t) . . (24) Consequently.. Hence its timevarying amplitude of motion is . Solution 1 x" + 2x' + 100x = 0. Rather than memorizing the various formulas given in the preceding discussion.
and the time lag of the motion is (as compared with 8 ~ 0. (25) and its graph is the damped exponential that is shown in Fig.1 ( vf99 v'Ii5 99/ 115 = 2n.1 cos(u)Jt. y 99 vi99 ~ 4 sina 1 = .< 0.5820 < 81 in Example 1).10..10 (in comparison with the undamped oscillations of Example 1).tan.9009. 2..4. Graphs of the position function x(t) = C 1e. and the envelope curves x(t) = ±C1e. With the timevarying amplitude and approximate phase angle shown explicitly. the position function x(t) = C cos(w0 t .a) of Example 1 (undamped oscillations).r cos( 59 t  5. where the phase angle a 1 satisfies cosa1 = > 0 v 115 and v'Ii5 Hence a 1 is the fourthquadrant angle a1 = 2n 4/v'IiS) + tan.1 • . the position function of the mass takes the form x(t) ~ y{ill 99 e .1 ( IN\ 4 ) v99 ~ 5.9009).4.144 Chapter 2 Linear Equations of Higher Order We therefore write x(t) = v'Ii5 IN\ et ( vi99 ~cos 59 t v99 vll5 4 ~sin 59 t ) vll5 = fillet cos( 59 t .a I).aJ) of Example 2 (damped oscillations). X FIGURE 2.
X • x (t) = C cos (w0 t . Determine the period and frequency of the simple harmonic motion of a body of mass 0.. 8o+. A body with mass 250 g is attached to the end of a spring that is stretched 25 ern by a force of 9 N. and thus when Jr Jr 3n 2' 2' 2 . period. 8o+. t 4 we calculate for the 1 0.1107 0. A mass of 3 kg is attached to the end of a spring that is stretched 20 em by a force of 15 N. . that is. 2.a 1) = 0. 1. respectively.1195 2 0.11.0667 Accordingly.. 3n 2w1 We see similarly that the undamped mass of Example 1 passes through equilibrium when Jr 3n 3n Jr t = 80 ...11 (where only the first three equilibrium passages are shown) we see the damped oscillations lagging slightly behind the undamped ones .4 Mechanical Vibrations 145 From (24) we see that the mass passes through its equilibrium position x = 0 when cos(w 1t ..a) 1 FIGURE 2. and frequency of the resulting motion. Determine the period and frequency of the simple harmonic motion of a 4kg mass on the end of a spring with spring constant 16 N j m.4352 3 0. n tn (undamped) tn (damped) t 1.0532 1. when t =01.2. t3. t2. It is set in motion with initial position x 0 = 0 and initial velocity v0 = 10 mj s.7509 4 1...8 illustrating the additional delay associated with damping. 3. 4. At time t = 0 . 8o. in Fig.4. Graphs on the interval 0 ~ t ~ 0. 2.7390 0. Find the amplitude.4. .75 kg on the end of a spring with spring constant 48 N j m. 2wo 2wo 2wo 2wo The following table compares the first four values undamped and damped cases.4249 0.
when located at the respective distances R 1 and R 2 from the center of the earthhave periods p 1 and p 2 .x 0 in this differential equation. with mass M and radius R = 3960 (mi).4. with a period of about 84 min.2 ft j s2 .5 (recall that the density of water is 1 gj cm3 ). 7..146 Chapter 2 Linear Equations of Higher Order the body is pulled 1 m to the right.4. and uniform density p . A mass m falling down a hole through the center of the earth (Problem 12). (Suggestion: First denote by x(t) the displacement of the mass below the unstretched position of the spring. has the same period as does a pendulum of length 100. (a) Show that F. Most grandfather clocks have pendulums with adjustable lengths. where M. Conclude from Newton's second law and part (a) that r " (t) = k 2 r(t). 2. Conclude that the buoy undergoes simple harmonic motion around its equilibrium position X e = ph with period p = 2:rr ~· Compute p and the amplitude of the motion if p = 0. How do you explain the coincidence? Or is it a coincidence? (e) With what speed (in miles . where the radius of the earth is R = 3956 (mi). (b) Now suppose that a small hole is drilled straight through the center of the earth. Two pendulums are of lengths L 1 and L 2 and. The buoy of Problem 10.00 in. compare with the result in part (c). 8. A certain pendulum keeps perfect time in Paris.. M denotes the mass of the earth).10 in. Show that FIGURE 2. (5) describing the motion of a mass attached to the bottom of a vertically suspended spring. Assume that friction is negligible. 2. But this clock loses 2 min 40 s per day at a location on the equator. where k 2 = GMjR 3 = gjR. and g = 980 cmj s2 • FIGURE 2.13). PI Pz 6. and set in motion with an initial velocity of 5 mjs to the left. 11. A pendulum of length 100. = GMmrjR 3 . (a) Find x(t) in the form C cos(w0 t . set up the differential equation for x .) 10. The buoy is initially suspended at rest with its bottom at the top surface of the water and is released at time t = 0. Find the radius of the buoy. h = 200 em. Use the result of Problem 5 to find the amount of the equatorial bulge of the earth.5 g j cm 3 . assume that the differential equation of a simple pendulum of length L is L()" + g() = 0. When depressed slightly and released. Consider a floating cylindrical buoy with radius r. Derive Eq. (d) Look up (or derive) the period of a satellite that just skims the surface of the earth. is the mass of the part of the earth within a sphere of radius r. Let a particle of mass m be dropped at time t = 0 into this hole with initial speed zero. One such clock loses 10 min per day when the length of its pendulum is 30 in.12). atop a nearby mountain. A cylindrical buoy weighing 100 lb (thus of mass m = 3. Thereafter it is acted on by two forces: a downward gravitational force equal to its weight mg = :rr r 2 hg and (by Archimedes' principle of buoyancy) an upward force equal to the weight :rr r 2 xg of water displaced. For a particle of mass m within the earth at distance r from the center of the earth. Use the result of Problem 5 to find the height of the mountain. height h. With what length pendulum will this clock keep perfect time? 9. (c) Take g = 32. and conclude from part (b) that the particle undergoes simple harmonic motion back and forth between the ends of the hole. where x = x(t) is the depth of the bottom of the buoy beneath the surface at time t (Fig. it oscillates up and down four times every 10 s. 0. stretching the spring. Then substitute y = x .4. Assume that the earth is a solid sphere of uniform density. located at a point at sea level where the radius of the earth is R = 3960 (mi). the gravitational force attracting m toward the center is F.mjr 2 . thus connecting two antipodal points on its surface.4.a). where g = G M / R 2 is the gravitational acceleration at the location of the pendulum (at distance R from the center of the earth.12. In Problems 5 through 8. = GM.13. and let r(t) be its distance from the center of the earth at timet (Fig. (b) Find the amplitude and period of motion of the body. 5.125 slugs in ftlbs (fps) units) floats in water with its axis vertical (as in Problem 10). 12.
{a) Find the position function x(t) and show that its graph looks as indicated in Fig. findthe undampedposition 25. c = 30. write the position function in the f orm x(t) = C1 e. Also. 24. If it is underdamped. and wf = w5 . (Critically damped) Deduce from Problem 24 that the mass passes through x = 0 at some instant t > 0 if and only if x 0 and v0 + px 0 have opposite signs. vo = 4 m = 1. k = 40.15. How do you explain the coincidence? Or is it a coincidence? 13. and phase angle of the motion.a 1) . 0 5 10 15 20 FIGURE 2.4. and k = 226 is set in motion with x(O) = 20 and x' (0) = 41. or underdamped. v0 = 16 m = 2. The position function x(t) of Problem 13. and the resulting damped vibrations have a frequency of 78 cycles/min. (a) If the weight is pulled down 1 ft below its static equilibrium position and then released from rest at timet = 0.14. as specified in each problem. x 0 = 2. 18. and to a dashpot that provides 3 lb of resistance for every foot per second of velocity. 20. (Overdamped) Show in this case that . 15. Vo =50 A 12lb weight (mass m = 0. Suppose that the mass in a massspringdashpot system with m = 25. The system is critically damped. timevarying amplitude. c = 9. After how long will the timevarying amplitude be 1% of its initial value? t. xo = 6. (a) Find the stiffness coefficient k of the spring if the car undergoes free vibrations at 80 cycles per minute (cycles/ min) when its shock absorbers are disconnected.4. c = 8. compare with the result in part (e). 2. Suppose that the mass in a massspringdashpot system with m = 10.2. x 0 = 5. Vo = 10 m = 2. x 0 = 0. (b) Find how far the mass moves to the right before starting back toward the origin. k = 50. find its position function x(t). but with the dashpot disconnected (so c = 0). the car is set into vibration by driving it over a bump. 5 4 147 function u(t) =Co cos(wot. The remaining problems in this section deal with free damped motion. 17. 27. 14. c = 16. c = 12.14. v0 = 2 m = 1. overdamped.4. ' ' 10 ~ 0 10 / Problems 24 through 34 deal with a massspringdashpot system having position fun ction x (t) satisfying Eq. In Problems 15 through 21. The mass is set in motion with initial position Xo and initial velocity v0 • Find the position fun ction x(t) and determine whether the motion is overdamped. 22. critically damped. c = 20. (4 ).4 Mechanical Vibrations per hour) does the particle pass through the center of the earth? (f) Look up (or derive) the orbital velocity of a satellite that just skims the surface of the earth. xo = 2. k = 169. w5 = k j m . k = 16. 2. c = 10. k = 125.pt cos(w 1t . Finally. k = 4. x 0 = 4. We write x 0 = x(O) and v0 = x' (O) and recall that p = cj(2m) . and k = 2 is set in motion with x(O) = 0 and x ' (O) = 5. 16. Xo = 5. (b) With the shock absorbers connected. construct a figure that illustrates the effect of damping by comparing the graphs of x(t) and u(t). 20 m = c = 3. vo = 8 m = 4.p 2 . c = 10. 21. Assume that the suspension system acts like a single spring and its shock absorbers like a single dashpot.4. (Critically damped) Deduce from Problem 24 that x (t) has a local maximum or minimum at some instant t > 0 if and only if v0 and v0 + px 0 have the same sign. 3 2 ~ 0 I 2 23. (a) Find the position function x(t) and show that its graph looks as indicated in Fig. 19. (b) Find the frequency. (Critically damped) Show in this case that 5 10 / 20 0 I 15 20 x (t) = (xo + vot + p xot)e .15. The position function x(t) of Problem 14.ao) that would result if the mass on the spring were set in motion with the same initial position and velocity. a mass m is attached to both a spring (with given spring constant k) and a dashpot (with given damping constant c). (b) Find the pseudoperiod ofthe oscillations and the equations of the "envelope curves" that are dashed in the figure. k = 63. (4) with appropriate values of the coefficients. so that its vertical vibrations satisfy Eq.375 slugs in fps units) is attached both to a vertically suspended spring that it stretches 6 in. or underdamped. This problem deals with a highly simplified model of a car of weight 3200 lb (mass m = 100 slugs in fps units). v0 = 0 m = 3.P1 • FIGURE 2. 26.
. Thus the future motion of an ideal massspringdashpot system is completely determined by the differential equation and the initial conditions. Suppose that m = 1 and c = 2 but k = I . D Nonhomoge~eous Eq~~~ions and Undetermined Coefficients We learned in Section 2. y x(t) 29. = 2rr pjw 1 is called the logarithmic decrement of the oscillation.. c. 30. Wt has a unique solution for t ~ 0 satisfying given initial conditions x (0) = x 0 . + a!y' + aoy = f(x). a spherical body of radius a moving at a (relatively slow) speed through a fluid of viscosity fL experiences a resistive force FR = 6rr fLaV. (1) . Differential Equations and Determinism Given a mass m. Of course in a real physical system it is impossible to measure the parameters m.. = 1. Note also that c = mw 1 ll.17 s.4. Conclude that on a given finite time interval the three solutions are in "practical" agreement if n is sufficiently large. respectively. which is an important parameter in fluid dynamics but is not easy to measure directly. 33. (Underdamped) Show that in this case 34. The final formula in Problem 33 then gives c and hence the viscosity of the fluid. 8mk 32.2n.JSmk.4.. but we saw in Section 2. (26).4 that an external force in a simple mechanical system contributes a nonhomogeneous term to its differential equation. 37. The frequency Wt and logarithmic decrement ll.!Y(n . Note: The result of Problem 33 provides an accurate method for measuring the viscosity of a fluid. Suppose that m = 1 and c = 2 but that k = 1 + 10.125 slugs in fps units) is oscillating attached to a spring and a dashpot. Whereas the graphs of x 1(t) and x 2 (t) resemble those shown in Figs. (26) with x(O) = 0 and x' (0) = 1 is x2(t) = lOne.1) + .~). The general nonhomogeneous nthorder linear equation with constant coefficients has the form GnY(n) + Gn . a dashpot constant c. According to Stokes's drag law. Show that the solution of Eq. 2.34 s and 1.8.4. and k precisely.a ) = . Compute the damping constant (in poundseconds per foot) and spring constant (in pounds per foot). c = 2. :::::o wo(l . Thus if a spherical mass on a spring is immersed in the fluid and set in motion. 35.73 in. this drag resistance damps its oscillations with damping constant c = 6rr a fL.7 and 2. (Underdamped) If the damping constant cis small in comparison with .46 in. 2.148 Chapter 2 Linear Equations of Higher Order 28. the graph of x 3 (t) exhibits damped oscillations like those illustrated in Fig.1 implies that the equation 31. Deduce from the result of Problem 32 that lnx' = 2rrp x2 w. x' (0) = v0 . Its first two maximum displacements of 6.1o2 n.t 1 = 2rrjw 1 if two consecutive maxima occur at times t 1 and t2. (Underdamped) A body weighing 100 lb (mass m = 3. show that for each fixed t > 0 it is true that n+oo lim X2(t) = lim x 3 (t) n~ oo = X t (t). Show that the solution with x (0) = 0 and x' (0) = 1 is x 1 (t) = te 1• 36. and 1. are observed to occur at times 0. (Underdamped) Show that the local maxima and minima of occur where p tan(wtt. Suppose that m Conclude that t 2 . apply the binomial series to show that mx" + ex' + kx = 0 (26) w. The constant ll. and k = 1 in Eq. (Overdamped) Prove that in this case the mass can pass through its equilibrium position x = 0 at most once. (26) with x(O) = 0 and x'(O) = 1 is 38.9. but with a very long pseudoperiod. of the oscillations can be measured by direct observation./rr because p = cj(2m). Nevertheless. (Underdamped) Let x 1 and x 2 be two consecutive local maximum values of x(t).1 sinh 10nt. and a spring constant k. (Overdamped) If x 0 = 0.3 how to solve homogeneous linear equations with constant coefficients. Theorem 2 of Section 2. deduce from Problem 27 that = vo ept sinh yt. Show that the solution of Eq. Problems 35 through 38 explore the resulting uncertainty in predicting the future behavior of a physical system.
suppose that f (x) = a cos kx + b sin kx. we illustrate it with several preliminary examples. (1) is sufficiently simple that we can make an intelligent guess as to the general form of Yp. Example 1 Find a particular solution of y" + 3 y' + 4 y = 3x + 2. Then. a linear combination with undetermined coefficients A and B . Then it is reasonable to expect a particular solution of the same form: Yp(x) =A cos kx + B sin kx. Similarly. that is. a general solution ofEq. substitute this expression for Yp into Eq. so Yp will satisfy the differential equation provided that (0) + 3(A) + 4(Ax +B) = 3x + 2. (1). Here f(x) = 3x Solution + 2 is a polynomial of degree 1. but with as yet undetermined coefficients. Thus our remaining task is to find Yp· The method of undetermined coefficients is a straightforward way of doing this when the given function f(x) in Eq. We may therefore substitute this form of Yp in Eq. be a particular solution of Eq. ••• .2. (1) has the form Y = Yc + Yp (2) where the complementary function Yc(x) is a general solution of the associated homogeneous equation (3) and Yp(x) is a particular solution of Eq. A 1 . We may. For example. suppose that f (x) is a polynomial of degree m. (4A )x + (3A + 4 B) = 3x +2 . (1).!Xm! + · · · + A1x + Ao that is also a polynomial of degree m. Then y~ = A and y~ = 0. Before describing the method in full generality. be a particular solution.5 Nonhomogeneous Equations and Undetermined Coefficients 149 By Theorem 5 of Section 2. Am so that Yp will. It turns out that this approach does succeed whenever all the derivatives of f (x ) have the same form as f (x) itself. it is reasonable to suspect a particular solution Yp(X) = AmXm + Am. The reason is that any derivative of such a linear combination of cos kx and sin kx has the same form. (1). indeed. because the derivatives of a polynomial are themselves polynomials of lower degree.2. so our guess is that Yp(x) =Ax+ B. (1). and thenby equating coefficients of like powers of x on the two sides of the resulting equationattempt to determine the coefficients A 0 . indeed. and thenby equating coefficients of cos kx and sin k x on both sides of the resulting equationattempt to determine the coefficients A and B so that y P will. therefore.
2(A cosx + B sinx) = 2cosx. So we try Yp(x) = y~(x) = y~(x) Solution A cos x + B sinx.1~. A. • Example 3 Find a particular solution of 3 y" + y' . (SA+ B) cosx +(A.B sinx) + (A sinx + B cosx).~~ • Example 2 Solution Any derivative of e 3x is a constant multiple of e3x. A sinx + B cosx. Thus our particular solution is Yp (x) = ~ e 3x. that is (collecting coefficients on the left). A first guess might be Yp(x) =A cosx.SB = 0 with readily found solution A = 5 13 . which superficially resembles Example 2.2 y = 2 cos x. 5 Yp ( x ) = .150 Chapter 2 Linear Equations of Higher Order for all x.SB) sinx = 2cosx. B = /3 • Hence a particular solution is I · + 13 smx. This will be true if the xterms and constant terms on the two sides of this equation agree. This will be true for all x provided that the cosine and sine terms on the two sides of this equation agree. but the presence of y' on the lefthand side signals that we probably need a term involving sin x as well. It therefore suffices for A and B to satisfy the two linear equations SA+ B = 2. Then y~ = 9Ae 3x. indicates that the method of undetermined coefficients is not always quite so simple as we have made it appear. Thus we have found the particular solution Yp(X) =~X. SA = 2. Example 4 . It therefore suffices for A and B to satisfy the two linear equations 4A = 3 and 3A + 4B = 2 that we readily solve for A= ~and B = . so the given differential equation will be satisfied provided that that is. so it is reasonable to try Yp(x) = Ae3x. so that A = ~.B sinx. =A cosx . Then substitution of Yp and its derivatives into the given differential equation yields 3( A cosx.13 cosx • The following example.
4Ae2 x = 0 =/= 2e 2x. we find that y~. (1) is a linear combination of (finite) products of functions of the following three types: 1. so that The terms involving xe2x obligingly cancel. . In fact. The method of undetermined coefficients applies whenever the function f (x) in Eq. Then take as a trial solution for Yp a linear combination of all linearly independent such terms and their derivatives.5 Nonhomogeneous Equations and Undetermined Coefficients 151 Solution If we try Yp (x) = Ae2X. and Rule 2 tells what to do when we do have it. Therefore.4yp = 4Ae 2 x . Thus. no matter how A is chosen. • The General Approach Our initial difficulty in Example 4 resulted from the fact that f(x) = 2e 2x satisfies the associated homogeneous equation. Rule 1. An exponential function erx. cos kx or sin kx. 2e 2x. leaving only 4Ae2x A = Consequently. RULE 1 Method of Undetermined Coefficients Suppose that no term appearing either in f (x) or in any of its derivatives satisfies the associated homogeneous equation Ly = 0. 3. A reasonable guess is Yp(x) = Axe2 x. a particular solution is !. the preceding computation shows that Ae 2x satisfies instead the associated homogeneous equation. A polynomial in x. Substitution into the original differential equation yields (4Ae 2x + 4Axe2x)  4(Axe 2 x) = 2e 2 x. given shortly. Any such functionfor example. tells what to do when we do not have this difficulty.2. we should begin with a trial function Yp(x) whose derivative involves both e 2x and something else that can cancel upon substitution into the differential equation to leave the e2x term that we need. Ae 2x cannot satisfy the given nonhomogeneous equation. Then determine the coefficients by substitution of this trial solution into the nonhomogeneous equation Ly = f(x). In Rules 1 and 2 we assume that Ly = f(x) is a nonhomogeneous linear equation with constant coefficients and that f (x) is a function of this kind. 2. for which y~(x) = Ae2x + 2Axe2x and y~(x) = 4Ae2x + 4Axe2x. (4) has the crucial property that only finitely many linearly independent functions appear as terms (summands) in f(x) and its derivatives of all orders.
x. Because none of these appears in Yc. 2B 4B =0.) In practice we check the supposition made in Rule 1 by first using the characteristic equation to find the complementary function Yc. that this procedure will always succeed under the conditions specified here. (5) is = ~' B = 0. Example 5 Find a particular solution of (5) Solution The (familiar) complementary solution of Eq. (It can be proved. x 2 . 6A +4C = 0.10cos3x.8 x. The function f (x) = 3x 3 and its derivatives are constant multiples of the linearly independent functions x 3 . Hence a particular solution of 3 3 9 Yp ( x ) = 4 x . it is merely a procedure to be followed in searching for a particular solution Yp· If we succeed in finding yP. so the . + 2Bx + C. and 1. we try Yp = Ax 3 + Bx 2 + Cx y~ = 3Ax 2 y~ + D.  = 2. Substitution in Eq. • (6) Example 6 Solve the initial value problem y". = 6Ax +2B. then we proceed with Rule I.152 Chapter 2 Linear Equations of Higher Order Note that this rule is not a theorem requiring proof. If none of the terms in this list duplicates a term in Yc. (5) is Yc(x) = Ct cos 2x + c2 sin 2x. then nothing more need be said. 0 has roots r Solution The characteristic equation r 2 complementary function is 3r +2 = = 1 and r = 2. (6Ax We equate coefficients of like powers of x in the last equation to get 4A = 3. however. (5) gives y~ + 4yp = = + 2B) + 4(Ax 3 + Bx 2 + Cx + D) 4Ax 3 + 4Bx 2 + (6A + 4C)x + (2B +D)= 3x 3 . and then write a list of all the terms appearing in f (x) and its successive derivatives.3y' y(O) + 2y = y'(O) 3ex . C = ~ . = 1. and D 0. +D = = 0 with solution A Eq.
3y~ + 2yp = 6Ae .x + (78. and C = {3 • This gives the particular solution 7 9 · 3 + 13 cos 3x + 13 sm x.2.10 cos 3x and its derivatives are ex.9C = 10. The result is the system 6A = 3. 13 • (7) Example 7 Find the general form of a particular solution of . 1B. cos 3x .~ ' c 2 = 1~.x.10cos3x. 3B sin 3x + 3C cos 3x.9C)cos3x + (98 7C) sin3x = 3ex. 9B 7C = 0 with solution A = ~. B = 7 13 . Bcos3x Aex .9C sin 3x. After we substitute these expressions into the differential equation in (6) and collect coefficients. e + 13 with derivative The initial conditions in (6) lead to the equations y (O) = c 1 + Cz + ~ + 7 13 = 1.5 Nonhomogeneous Equations and Undetermined Coefficients 153 The terms involved in f (x) = 3ex . To satisfy those initial conditions. we get y~. those involving cos 3x.9B cos 3x . The desired particular solution is therefore y (x ) 1 ex + 6 e2x + I e x + 7 cos 3x + 9 sm =2 2 13 13 · 3x . y' (0) = c 1 + 2c2  ~ + g=2 with solution c 1 = . we try Yp = y~ y~ Aex + = Aex = + Csin3x. Because none of these appears in Yc. We equate the coefficients of the terms involving e. however. and those involving sin 3x. and sin 3x. we begin with the general solution y(x) = Yc(x) = + Yp(X) 1 x 7 9 · 3 c 1ex + c2 e2x + 2 cos 3x + 13 sm x . I x Yp ( x ) = 2 e which. does not have the required initial values in (6).
To see how to amend our first guess. x 2 e 2x . and r = 3i. the trial solution takes the form Yp(x) =A cosx + B sinx + Cx cosx + Dx sinx + Ee2x + Fxe 2x + Gx 2 e 2x. (8) is (10) so substitution of (9) in the lefthand side of (8) would yield zero rather than (2x. F. we observe that by Eq. suppose that we want to find a particular solution of the differential equation (8) Proceeding as in Rule l. ' X The derivatives of the righthand side in Eq. we get seven equations determining the seven coefficients A.154 Chapter 2 Linear Equations of Higher Order Solution The characteristic equation r 3 + 9r the complementary function is Yc(x) = 0 has roots r = 0. D. sinx. If y(x ) is any solution of Eq.3)erx . e 2x x cosx. our first guess would be (9) This form of Yp (x) will not be adequate because the complementary function of Eq. we see that y (x) is also a solution ofthe equation (D . and G. e 2x and Because there is no duplication with the terms of the complementary function. (8) and we apply the operator (D. Upon substituting Yp in Eq. + Cz cos 3x + c3 sin3x. (13) of Section 2. r = 3i. E.r) 2 to both sides. (7) and equating coefficients oflike terms. • The Case of Duplication Now we turn our attention to the situation in which Rule 1 does not apply: Some of the terms involved in f (x ) and its derivatives satisfy the associated homogeneous equation. C .3 . ' x sinx. The general solution of this homogeneous equation can be written as Thus every solution of our original equation in (8) is the sum of a complementary function and a particular solution of the form (11) .r ) 5 y = 0. So = c. For instance. B. (7) involve the terms cosx.
The table in Fig. where f (x) is a linear combination of products of the elementary functions listed in (4 ). (12) it suffices to find separately particular solutions Y1 (x) and Y 2 (x) of the two equations Ly = ft (x) and Ly = h(x). it is common to have j(x ) =!I (x) + h(x) . (12). Thus f (x ) can be written as a sum of terms each of the form Pm(x)erx cos kx or Pm (x)erx sin kx. we observe that to find a particular solution of the nonhomogeneous linear differential equation L y = j. corresponding to the possibilities m = 0. (11) can be obtained by multiplying each term of our first guess in (9) by the least positive integral power of x (in this case.5 Nonhomogeneous Equations and Undetermined Coefficients 155 Note that the righthand side in Eq. (14) where Pm(x ) is a polynomial in x of degree m. In practice we seldom need to deal with a function f (x) exhibiting the full generality in (14). RULE 2 Method of Undetermined Coefficients Ifthe function f(x) is of either form in (14). (15) by substituting Yp into the nonhomogeneous equation.5. The procedure by which we arrived earlier at the particular solution in (11) of Eq. (8) can be generalized to show that the following procedure is always successful. This procedure is illustrated in Examples 8 through 10. This procedure succeeds in the general case. (This is a type of "superposition principle" for nonhomogeneous linear equations. choosing s separately for each part to eliminate duplication with the complementary function. where !1 (x) and h (x) are different functions of the sort listed in the table in Fig. To simplify the general statement of Rule 2. For linearity then gives and therefore Yp = Y1 + Y 2 is a particular solution of Eq. Note that any derivative of such a term is of the same form but with both sines and cosines appearing. 2. r = 0. On the other hand.x + B zx 2 + · · · + Bmxm)erx sinkx] .5.(x) + h(x) . x + A zx 2 + · · · + Am x m)erx coskx + (Bo + B. (15) where s is the smallest nonnegative integer such that such that no term in Yp duplicates a term in the complementary function Yc · Then determine the coefficients in Eq. In this event we take as Yp the sum of the trial solutions for j 1(x ) and h(x). (13) respectively. . and k = 0.) Now our problem is to find a particular solution of the equation Ly = f(x) .1 lists the form of Yp in various common cases.2.1. take as the trial solution Yp(x ) = xs[(Ao + A. 2. x 3 ) that suffices to eliminate duplication between the terms of the resulting trial solution Yp(x) and the complementary function Yc(x) given in (10).
156
Chapter 2 Linear Equations of Higher Order
Pm = bo + b1x
+ b2x 2 + · · · + bmxm a cos kx + b sin kx
Pm(x)e'x
x 5 (Ao
+ A 1x + A 2x 2 + · · · + Amxm) x (A coskx + B sinkx)
5
e'x(acoskx + bsinkx) Pm(x)(acoskx +bsinkx)
x 5 e'x(Acoskx + Bsinkx) xs(Ao + A 1x + A 2x 2 + · · · + Amxm)e'x xs[(Ao
+ A 1x + · · · + Amxm) coskx + (Bo + B 1x + · · · + Bmxm) sinkx]
FIGURE 2.5.1. Substitutions in the method of undetermined coefficients.
"'
• m•·~ ·· ••• •••
•
v '"
'••••••.••
••·••·•r·~ · r •·•m
''"' ,,,,.,, '""."''" ·•• ••••••
Example 8
Find a particular solution of (16)
Solution
The characteristic equation r 3 complementary function is
+ r 2 = 0 has roots r 1 = r 2 = 0 and r3 = 1, so the
As a first step toward our particular solution, we form the sum
The part Aex corresponding to 3ex does not duplicate any part of the complementary function, but the part B + Cx + Dx 2 must be multiplied by x 2 to eliminate duplication. Hence we take
Yp = Aex + Bx 2 + Cx 3 + Dx 4 ,
y~ = A ex + 2Bx + 3Cx 2 +4Dx 3 , y~ = A ex + 2B + 6Cx + 12Dx 2 ,
and
y~3 )
= A ex + 6C + 24Dx.
Substitution of these derivatives in Eq. (16) yields
2Aex + (2B + 6C) + (6C + 24D)x + 12Dx 2 = 3ex + 4x 2 .
The system of equations
2A = 3, 2B +6C = 0,
6C + 24D has the solution A solution is
= 0,
~, and D
12D = 4
=
~, B
= 4, C = 
=
1· Hence the desired particular
•
2.5 Nonhomogeneous Equations and Undetermined Coefficients
157
Example 9
Determine the appropriate form for a particular solution of
y"
+ 6y' + 13y =
e 3x cos 2x.
Solution
The characteristic equation r 2 + 6r + 13 = 0 has roots 3 ± 2i, so the complementary function is Yc(x) = e 3x(c, cos2x + Cz sin2x). This is the same form as a first attempt e 3 x (A cos 2x + B sin 2x) at a particular solution, so we must multiply by x to eliminate duplication. Hence we would take
Yp(x) = e 3x (Ax cos 2x
+ Bx sin 2x).
•
Example 10
Determine the appropriate form for a particular solution of the fifthorder equation
Solution
The characteristic equation (r 2) 3 (r 2 + 9) so the complementary function is
= 0 has roots r = 2, 2, 2, 3i, and 3i,
As a first step toward the form of a particular solution, we examine the sum
[(A+ Bx + Cx 2 )e2x] + [(D +Ex) cos 3x + (F + Gx) sin 3x].
To eliminate duplication with terms of Yc (x), the first partcorresponding to x 2e2x must be multiplied by x 3 , and the second partcorresponding to x sin 3xmust be multiplied by x. Hence we would take
Variation of Parameters
Finally, let us point out the kind of situation in which the method of undetermined coefficients cannot be used. Consider, for example, the equation
y"
+y
= tan x,
(17)
which at first glance may appear similar to those considered in the preceding examples. Not so; the function f (x ) = tan x has infinitely many linearly independent derivatives sec2 x, 2 sec2 x tan x, 4 sec 2 x tan 2 x
+ 2 sec4 x,
Therefore, we do not have available a finite linear combination to use as a trial solution.
158
Chapter 2 Linear Equations of Higher Order
We discuss here the method of variation of parameters, whichin principle (that is, if the integrals that appear can be evaluated)can always be used to find a particular solution of the nonhomogeneous linear differential equation
y(n)
+ Pn1 (x)yCnl) +· · ·+PI (x)y + po(x)y =
1
f(x),
(18)
provided that we already know the general solution
Yc = CJYI
+ CzYz + · · · + CnYn
(19)
of the associated homogeneous equation
y(n)
+ Pn1 (x)yCnI) + · · · + Pl (x)y + po(x)y =
1
0.
(20)
Here, in brief, is the basic idea of the method of variation of parameters. Suppose that we replace the constants, or parameters, c 1, c2 , ..• , Cn in the complementary function in Eq. (19) with variables: functions u 1, u 2 , ... , un of x. We ask whether it is possible to choose these functions in such a way that the combination
Yp(x) = Ut(X)YI (x)
+ uz(x)yz(x) + · · · + Un(X)Yn(x)
(21)
is a particular solution of the nonhomogeneous equation in ( 18). It turns out that this is always possible. The method is essentially the same for all orders n ~ 2, but we will describe it in detail only for the case n = 2. So we begin with the secondorder nonhomogeneous equation
L[y] = y"
+ P(x)y + Q(x )y =
1
f(x)
(22)
with complementary function
Yc(x) = CtYI(x)
+ CzYz(x)
(23)
on some open interval I where the functions P and Q are continuous. We want to find functions u 1 and u 2 such that
Yp(x) = u1 (x)yi (x)
+ uz(x)yz(x)
(24)
is a particular solution of Eq. (22). One condition on the two functions u 1 and u 2 is that L[yp] = j(x). Because two conditions are required to determine two functions, we are free to impose an additional condition of our choice. We will do so in a way that simplifies the computations as much as possible. But first, to impose the condition L[yp ] = f(x), we must compute the derivatives y~ and y~. The product rule gives
Yp =
I (
UJ YJ + UzYz
I
I )
+
(
U1Y1
I
+ UzYz ·
I )
To avoid the appearance of the second derivatives u 1{ and u ~ , the additional condition that we now impose is that the second sum here must vanish:
(25)
2.5 Nonhomogeneous Equations and Undetermined Coefficients
159
Then (26) and the product rule gives (27) But both Yt and Y2 satisfy the homogeneous equation
y"
+ Py' + Qy =
0
associated with the nonhomogeneous equation in (22), so
I Q Yi YiII =  p Yi
(28)
for i = 1, 2. It therefore follows from Eq. (27) that
In view of Eqs. (24) and (26), this means that
hence
(29)
The requirement that Yp satisfy the nonhomogeneous equation in (22)that is, that L[yp] = f(x)therefore implies that (30) Finally, Eqs. (25) and (30) determine the functions u 1 and u2 that we need. Collecting these equations, we obtain a system (31) of two linear equations in the two derivatives u~ and u;. Note that the determinant of coefficients in (31) is simply the Wronskian W (y 1 , y2 ). Once we have solved the equations in (31) for the derivatives u'1 and we integrate each to obtain the functions u 1 and u 2 such that
u;,
(32) is the desired particular solution of Eq. (22). In Problem 63 we ask you to carry out this process explicitly and thereby verify the formula for Yp(x) in the following theorem.
160
Chapter 2 Linear Equations of Higher Order
THEOREM 1
Variation of Parameters
If the nonhomogeneous equation y " + P(x)y 1 + Q(x)y = f(x) has complementary function Yc(x) = CtYt(X) + c2y2(x), then a particular solution is given by
Yp X 
( )__ ( )f
Yt X
Y2(x)f(x) d
VV(x)
X
+ Y2 (X)
f
Yt(x)f(x) d
VV(x)
X'
(33)
where VV = VV (y 1 , Y2) is the Wronskian of the two independent solutions Y1 and Y2 of the associated homogeneous equation.
Example 11
Find a particular solution of the equation y" + y
··· ·  · •• •• ••• ··."
¥
= tan x.
. a
¥
. ..
Solution
The complementary function is Yc (x) = c 1 cos x + c2 sin x, and we could simply substitute directly in Eq. (33). But it is more instructive to set up the equations in (31) and solve for u1 1 and u;, so we begin with
Yt = cosx, y 1 =  smx,
I •
Y2 = sinx,
Y2
I
= cosx.
Hence the equations in (31) are
+ cu;)(sinx) = cu;)( sin x ) + cu;)(cosx) =
(u 1,)(cosx) We easily solve these equations for
0,
tanx.
1 • sin2 x u 1 =  smx tanx =    = cosx  secx, cosx
u2
I
= cosx tan x = smx.
•
Hence we take u, = and
u2 =
J
(cosx  secx) dx = sinx  In I secx
+ tan x l
J
sin x dx =  cos x .
(Do you see why we choose the constants of integration to be zero?) Thus our particular solution is
Yp(x ) =
u, (x )y, (x ) + u2(x ) y2 (x )
= (sinx ln lsecx
+ tanxl) cosx + ( cosx)(sin x);
that is,
Yp(x) =  (cos x) ln lsecx
+ tan x l.
•
2.5 Nonhomogeneous Equations and Undetermined Coefficients
161
Ill) Pro~l~~s
In Problems I through 20, find a particular solution Yr of the given equation. In all these problems, primes denote derivatives with respect to x.
1. y" + 16y = e 3x 2. 3. y" y' 6y = 2sin3x 4. 5. y" + y' + y = sin2 x 6. 7. y" 4y = sinhx 8. 9. y" + 2y' 3y = 1 + xex 10. y" + 9y = 2 cos 3x + 3 sin 3x 11. y<3J + 4y' = 3x 1 12. 13. y" + 2y' + 5y =ex sinx 14. 15. y(5l + 5y<4J  y = 17 16. 17. y" + y = sinx +xcosx 18. / 4 )  5y" + 4y =ex xe 2x 19. y<5J + 2y< 3J + 2y" = 3x 2  1 20. y<3 J  y = e' + 7
(b) Use the result of part (a) to find a general solution of
y" + 4 y
= cos3 x.
y" y' 2y = 3x + 4 4y" +4y' + y = 3xex 2y" + 4y' + 7y = x 2 y" 4y =cosh 2x
yOl + y' = 2 sinx
Use trigonometric identities to find general solutions of the equations in Problems 44 through 46.
44. y" + y' + y = sin x sin 3x 45. y" + 9y = sin4 x
46. y" + y = x cos3 x
y(4l  2y" + y = xex
y" + 9y = 2x 2 e 3x
+ 5
In Problems 47 through 56, use the method of variation ofparameters to find a particular solution of the given differential equation.
47. y" + 3y' 49. y" 4y'
In Problems 2I through 30, set up the appropriate form of a particular solution Yr· but do not determine the values of the coefficients.
21. 22. 23. 24. 25. 26.
+ 2y = 4ex + 4y = 2e2 x
48. y"  2y'  8y = 3e 2x 50. y" 4y =sinh 2x
51. y" + 4 y = cos 3x 53. y" +9y = 2sec3x 55. y" +4y = sin 2 x
52. y" +9y = sin3x 54. y" + y = csc 2 x
56. y" 4y
27. 28.
29. 30.
y" 2y' + 2y = e' sinx y< 5J  y<3l = e' + 2x 2  5 y" + 4 y = 3x cos 2x yOl  y" 12y' = x 2xe 3x y" + 3y' + 2y = x(ex  e2x) y" 6y' + 13y = xe3x sin 2x y<4J +5y" +4y = sinx +cos2x y<4J + 9y" = (x 2 + 1) sin 3x (D 1) 3 (D 2  4)y = xex + e 2x + e 2x y<4 J 2y" + y = x 2 cosx
= xex
57. You can verify by substitution that Yc = c 1x + c2x 1 is a complementary function for the nonhomogeneous secondorder equation
Solve the initial value problems in Problems 31 through 40.
31. 32. 33. 34. 35. 36.
But before applying the method of variation of parameters, you must first divide this equation by its leading coefficient x 2 to rewrite it in the standard form
37.
38. 39.
40.
41.
y" + 4y = 2x; y(O) = 1, y'(O) = 2 y" + 3y' + 2y =ex; y(O) = 0, y'(O) = 3 y" + 9y =sin 2x; y(O) = 1, y'(O) = 0 y" + y = cosx; y(O) = 1, y'(O) = 1 y" 2y' +2y =X+ 1; y(O) = 3, y'(O) = 0 y<4J 4y" = x 2; y(O) = y'(O) = 1, y"(O) = y(3l(O) = 1 y<3l  2y" + y' = 1 + xex; y(O) = y'(O) = 0, y"(O) = 1 y" + 2y' + 2y = sin3x; y(O) = 2, y'(O) = 0 y< 3l + y" = x +e x; y(O) = 1, y'(O) = 0, y"(O) = 1 y<4J y = 5; y(O) = y'(O) = y"(O) = y Ol (O) = 0 Find a particular solution of the equation
y<4J y<3J y" y' 2y = 8xs.
y
II
+ 1 y
x
I
1 3   y = nx . x2
Thus f(x) = 72x 3 in Eq. (22). Now proceed to solve the equations in (31) and thereby derive the particular solution Yr = 3x5.
In Problems 58 through 62, a nonhomogeneous secondorder linear equation and a complementary function Yc are given. Apply the method of Problem 57 to find a particular solution of the equation.
58. x 2 y"  4xy'
42. Find the solution of the initial value problem consisting of the differential equation of Problem 41 and the initial conditions
y(O)
+ 6y = x 3 ; Yc = c,x 2 + CzX 3
= y'(O) = y"(O) = y<3l(O) = 0.
43.
(a) Write
cos3x + i sin3x = e 3ix = (cosx + i sinx) 3
by Euler's formula, expand, and equate real and imaginary parts to derive the identities cos 3 x = ~ cos x + ~ cos 3x, sin3 x = ~ sinx  ~ sin3x.
59. x 2 y"  3xy' + 4y = x 4 ; Yc = x 2 (c1 + c2lnx) 60. 4x 2 y"  4xy' + 3y = 8x 4 13 ; Yc = c,x + c2x 3/4 61. x 2 y" +xy' + y = lnx; Yc = c 1 cos(lnx) +czsin(lnx) 62. (x 2  l)y" 2xy' + 2y = x 2  1; Yc = c1x + c2(1 + x 2 ) 63. Carry out the solution process indicated in the text to derive the variation of parameters formula in (33) from Eqs. (31) and (32). 64. Apply the variation of parameters formula in (33) to find the particular solution Yr (x) =  x cos x of the nonhomogeneous equation y" + y = 2 sin x.
162
Chapter 2 Linear Equations of Higher Order
BJ
Forced Oscillations and Resonance
In Section 2.4 we derived the differential equation
mx" +ex'+ kx = F(t)
(1)
that governs the onedimensional motion of a mass m that is attached to a spring (with constant k) and a dashpot (with constant c) and is also acted on by an external force F(t). Machines with rotating components commonly involve massspring systems (or their equivalents) in which the external force is simple harmonic:
Equilibrium
F(t) = F0 cos wt
or
F(t) = F0 sin wt,
(2)
FIGURE 2.6.1. The cartwithflywheel system.
where the constant F0 is the amplitude of the periodic force and w is its circular frequency. For an example of how a rotating machine component can provide a simple harmonic force, consider the cart with a rotating vertical flywheel shown in Fig. 2.6.1. The cart has mass m  m 0 , not including the flywheel of mass m 0 • The centroid of the flywheel is off center at a distance a from its center, and its angular speed is w radians per second. The cart is attached to a spring (with constant k) as shown. Assume that the centroid of the cart itself is directly beneath the center of the flywheel, and denote by x(t) its displacement from its equilibrium position (where the spring is unstretched). Figure 2.6.1 helps us to see that the displacement x of the centroid of the combined cart plus flywheel is given by _
X
=
(m  mo)x
+ mo(x +a cos wt)
m
moa = X + COS Wt. m
Let us ignore friction and apply Newton's second law mx" =  kx, because the force exerted by the spring is kx. We substitute for x in the last equation to obtain
mx"  m 0 aw 2 cos wt = kx;
that is,
mx" + kx = m 0 aw2 cos wt.
(3)
Thus the cart with its rotating flywheel acts like a mass on a spring under the influence of a simple harmonic external force with amplitude F0 = m 0 aw2 . Such a system is a reasonable model of a frontloading washing machine with the clothes being washed loaded off center. This illustrates the practical importance of analyzing solutions of Eq. (1) with external forces as in (2).
Undamped Forced Oscillations
To study undamped oscillations under the influence of the external force F(t) F0 cos wt, we set c = 0 in Eq. (1), and thereby begin with the equation
mx"
+ kx
= Fo cos wt
(4)
whose complementary function is Xc =
c 1 cos w 0 t
+ c2 sin w0 t. Here
wo=/f
2.6 Forced Oscillations and Resonance
163
(as in Eq. (9) of Section 2.4) is the (circular) natural frequency of the massspring system. The fact that the angle w 0 t is measured in (dimensionless) radians reminds us that if t is measured in seconds (s), then w0 is measured in radians per secondthat is, in inverse seconds (s 1). Also recall from Eq. (14) in Section 2.4 that division of a circular frequency w by the number 2n of radians in a cycle gives the corresponding (ordinary) frequency v = wj2n in Hz (hertz= cycles per second). Let us assume initially that the external and natural frequencies are unequal: w =!= w0 . We substitute Xp = A cos wt in Eq. (4) to find a particular solution. (No sine term is needed in Xp because there is no term involving x' on the lefthand side in Eq. (4).) This gives
mw2 A cos wt
+ kA cos wt =
F0 cos wt,
so
Fo A=k mw2
(5)
and thus
Xp(t) =
2 w0
Fofm coswt. w2
(6)
Therefore, the general solution x = Xc + x P is given by x(t) =
c 1 cos wot
+ c2 sin wot +
2 w0
Fofm cos wt, w2
(7)
where the constants c 1 and c2 are determined by the initial values x(O) and x'(O) . Equivalently, as in Eq. (12) of Section 2.4, we can rewrite Eq. (7) as x(t) = C cos(wot a)+
Fofm
2 w0 2
w
coswt ,
(8)
so we see that the resulting motion is a superposition of two oscillations, one with natural circular frequency w0 , the other with the frequency w of the external force.
Example 1
Suppose that m = 1, k = 9, F0 = 80, and w = 5, so the differential equation in (4) is x" + 9x = 80cos St. Find x(t) if x(O) = x'(O) = 0.
Solution
Here the natural frequency w0 = 3 and the frequency w = 5 of the external force are unequal, as in the preceding discussion. First we substitute x P = A cos St in the differential equation and find that  25A + 9A = 80, so that A = 5. Thus a particular solution is xp (t) =  5cos5t. The complementary function is Xc = c 1 cos 3t + c2 sin 3t, so the general solution of the given nonhomogeneous equation is x(t) = c 1 cos 3t + c2 sin 3t  5 cos 5t, with derivative
x' (t)
=  3c 1 sin 3t + 3c2cos 3t + 25sin5t.
164
15 r10 5
~
Chapter 2 Linear Equations of Higher Order

Period= 2n
I I I I I I I I 1
The initial conditions x(O) = 0 and x'(O) = 0 now yield c 1 = 5 and c2 = 0, so the desired particular solution is
x(t) = 5 cos 3t 5 cos St.
0
5
vv
Jt
M
v~
M
vv
5Jt 6Jt
1
As indicated in Fig. 2.6.2, the period of x(t) is the least common multiple 2n of the periods 2n /3 and 2n /5 of the two cosine terms. •
10 r15 0 2Jt 3Jt 4Jt
Beats
If we impose the initial conditions x (0) = x' (0) = 0 on the solution in (7), we find that
c1
x(t)
FIGURE 2.6.2. The response = 5 cos 3t  5 cos 5t in Example 1.
= 
Fo
m(w6 w 2 )
and
c2
= 0,
so the particular solution is
x(t) =
Fo
2 m(w0 
w)
2
(cos wt cos w 0 t).
(9)
The trigonometric identity 2 sin A sin B = cos(A  B)  cos(A + B), applied with A= ~(wo + w)t and B = ~(w0  w)t, enables us to rewrite Eq. (9) in the form
x(t) =
2Fo
2 m(w0 
w )
2
. 1 sm 2 (w0

w)t sm 2 (w0
.
1
+ w)t.
(10)
Suppose now that w ;: : :;: wo, so that wo + w is very large in comparison with lwo wl . Then sin ~(w0 + w)t is a rapidly varying function, whereas sin ~(wo  w)t is a slowly varying function. We may therefore interpret Eq. (10) as a rapid oscillation with circular frequency ~(w0 + w),
x(t) = A(t) sin ~(wo
+ w)t,
but with a slowly varying amplitude
A(t) =
2Fo
2 m(w0 
w)
2
. I sm 2 (w0

w)t.
Example 2
1.5 ,.,1,.,.., x = sin 5t 1.0 I
\ I
With m = 0.1, F0 =50, w0 =55, and w = 45, Eq. (10) gives
x (t) = sin5t sin50t.
I
I I I
I
I I I
I
0.5
~
I
~\ N o.o rutfl1Hffiyv:ltlfi+HIIM ,fi+HI#,I:mfj ¥11fmJ~Jt11 11
4 1
1
m,. 1
Figure 2.6.3 shows the corresponding oscillation of frequency ~ (w0 + w ) = 50 that is "modulated" by the amplitude function A(t) = sin 5t of frequency ~(w0 w) = 5.
I!+HjiHJ
 0.5 1.0
I I I I
I
1.,~
I I
I I I I
I
I
' I ~·
I
I I
I
I
x = sin 5t sin 50t
1.5 0.0 0.5
1.0 1.5 2.0 2.5 3.0
A rapid oscillation with a (comparatively) slowly varying periodic amplitude exhibits the phenomenon of beats. For example, if two horns not exactly attuned to one another simultaneously play their middle C, one at w0 j(2n) = 258 Hz and the other at wj(2n) = 254 Hz, then one hears a beat an audible variation in the amplitude of the combined soundwith a frequency of
(wo  w)/2
•
FIGURE 2.6.3. The phenomenon of beats.
2JT
258  254 = 2 (Hz). 2
2.6 Forced Oscillations and Resonance
165
Resonance
Looking at Eq. (6), we see that the amplitude A of Xp is large when the natural and external frequencies w0 and w are approximately equal. It is sometimes useful to rewrite Eq. (5) in the form A=,kmw2
Fo
__h_o_/k_= _ ± _Ph_o
1  (wfwo) 2

k
'
(11)
where Folk is the static displacement of a spring with constant k due to a constant force Fo, and the amplification factor p is defined to be
1 p .,. 11 (wfwo) 2 1. (12)
It is clear that p + +oo as w + w0 . This is the phenomenon of resonancethe increase without bound (as w + w 0 ) in the amplitude of oscillations of an undamped system with natural frequency w0 in response to an external force with frequency w ~ w 0 . We have been assuming that w i= w 0 . What sort of catastrophe should one expect if wand wo are.precisely equal? Then Eq. (4), upon division of each term by m, becomes
x"
2 + w0 x
=cos w0 t. m
Fo
(13)
1.5 1.0 0.5
"<
ffX =
I
I
I
.'
Because cos wot is a term of the complementary function, the method of undetermined coefficients calls for us to try
.'
__ .
t
.'
Xp(t)
= t(Acoswot + Bsinw0 t).
Fo/(2mwo).
0.0 0.5
1.0 1.5
fktl_/}
·x ....v....
ff
v
l_1f_f
 ~
I
We substitute this in Eq. (13) and thereby find that A = 0 and B Hence the particular solution is
' ' ... '
'
~
I
x p(t) =   t
Fo
x
f
=
I
t sin SOt
2mwo
smw0 t.
.
(14)
o.oo
0.25
o.so
' 0.75 1.00 1.25 1.50
I I
FIGURE 2.6.4. The phenomenon of resonance.
The graph of xp(t) in Fig. 2.6.4 (in which m = 1, F0 = 100, and w 0 = 50) shows vividly how the amplitude of the oscillation theoretically would increase without bound in this case of pure resonance, w = w 0 . We may interpret this phenomenon as reinforcement of the natural vibrations of the system by externally impressed vibrations at the same frequency. Suppose that m = 5 kg and k = 500 Njm in the cart with the flywheel of Fig. 2.6.1. Then the natural frequency is w 0 = ,Jkliii = 10 radjs; that is, 10/ (2n) ~ 1.59 Hz. We would therefore expect oscillations of very large amplitude to occur if the flywheel revolves at about (1.59)(60) ~ 95 revolutions per minute (rpm). • In practice, a mechanical system with very little damping can be destroyed by resonance vibrations. A spectacular example can occur when a column of soldiers marches in step over a bridge. Any complicated structure such as a bridge has many natural frequencies of vibration. If the frequency of the soldiers' cadence is approximately equal to one of the natural frequencies of the structure, thenjust as in our simple example of a mass on a spring resonance will occur. Indeed, the resulting
Example 3
166
Chapter 2 Linear Equations of Higher Order resonance vibrations can be of such large amplitude that the bridge will collapse. This has actually happenedfor example, the collapse of Broughton Bridge near Manchester, England, in 1831and it is the reason for the nowstandard practice of breaking cadence when crossing a bridge. Resonance may have been involved in the 1981 Kansas City disaster in which a hotel balcony (called a skywalk) collapsed with dancers on it. The collapse of a building in an earthquake is sometimes due to resonance vibrations caused by the ground oscillating at one of the natural frequencies of the structure; this happened to many buildings in the Mexico City earthquake of September 19, 1985. On occasion an airplane has crashed because of resonant wing oscillations caused by vibrations of the engines. It is reported that for some of the first commercial jet aircraft, the natural frequency of the vertical vibrations of the airplane during turbulence was almost exactly that of the massspring system consisting of the pilot's head (mass) and spine (spring). Resonance occurred, causing pilots to have difficulty in reading the instruments. Large modem commercial jets have different natural frequencies, so that this resonance problem no longer occurs.
Modeling Mechanical Systems
The avoidance of destructive resonance vibrations is an everpresent consideration in the design of mechanical structures and systems of all types. Often the most important step in determining the natural frequency of vibration of a system is the formulation of its differential equation. In addition to Newton's law F = rna, the principle of conservation of energy is sometimes useful for this purpose (as in the derivation of the pendulum equation in Section 2.4). The following kinetic and potential energy formulas are often useful. 1. Kinetic energy: T = ~mv 2 for translation of a mass m with velocity v; 2. Kinetic energy: T = ~I ui for rotation of a body of a moment of inertia I with angular velocity w; 3. Potential energy: V = 4kx 2 for a spring with constant k stretched or compressed a distance x; 4. Potential energy: V = mgh for the gravitational potential energy of a mass m at height h above the reference level (the level at which V = 0), provided that g may be regarded as essentially constant.
Example 4
Find the natural frequency of a mass m on a spring with constant k if, instead of sliding without friction, it is a uniform disk of radius a that rolls without slipping, as shown in Fig 2.6.5. With the preceding notation, the principle of conservation of energy gives
Solution
Equilibrium
_2lmv2
+
_ 2 1
I.,2 ""' + _21kx2 = E
where E is a constant (the total mechanical energy of the system). We note that v = aw and recall that I = ma 2j2 for a uniform circular disk. Then we may simplify the last equation to
~mv 2 + 4kx 2 =E.
x
=
o
Because the righthand side of this equation is constant, differentiation with respect to t (with v = X 1 and V 1 = x ") now gives
3 I II k I 0 2mx x + xx = .
FI Gu RE 26 . .5. The rolling disk.
2.6 Forced Oscillations and Resonance
16 7
We divide each term by ~mx' to obtain
II 2k x +x =0. 3m
Thus the natural frequency of horizontal backandforth oscillation of our rolling disk is J2kj3m, which is ~ ~ 0.8165 times the familiar natural frequency .Jk7iii of a mass on a spring that is sliding without friction rather than rolling without sliding. It is interesting (and perhaps surprising) that this natural frequency does not depend on the radius of the disk. It could be either a dime or a large disk with a radius of one meter (but of the same mass). •
Example 5
Suppose that a car oscillates vertically as if it were a mass m = 800 kg on a single spring (with constant k = 7 x 104 Njm), attached to a single dashpot (with constant c = 3000 N·s/m). Suppose that this car with the dashpot disconnected is driven along a washboard road surface with an amplitude of 5 em and a wavelength of L = 10 m (Fig. 2.6.6). At what car speed will resonance vibrations occur?
m
2ns y  a cos L
s=O
position
FIGURE 2.6.6. The washboard road surface of Example 5.
FIGURE 2.6.7. The "unicycle model" of a car.
Solution
We think of the car as a unicycle, as pictured in Fig. 2.6.7. Let x(t) denote the upward displacement of the mass m from its equilibrium position; we ignore the force of gravity, because it merely displaces the equilibrium position as in Problem 9 of Section 2.4. We write the equation of the road surface as y
= a cos L
2ns
(a
= 0.05 m, L =
10m).
(15)
When the car is in motion, the spring is stretched by the amount x  y , so Newton's second law, F = ma, gives
mx" =  k(x  y);
that is,
mx" + kx = ky
(16)
Ifthe velocity of the car is v, then s = vt in Eq. (15), so Eq. (16) takes the form
mx "
+ kx =
2nvt ka cos£·
(16')
168
Chapter 2 Linear Equations of Higher Order
This is the differential equation that governs the vertical oscillations of the car. In comparing it with Eq. (4), we see that we have forced oscillations with circular frequency w = 2nvjL. Resonance will occur when w = w0 = ,Jkliii. We use our numerical data to find the speed of the car at resonance:
v
=
2n
Lff  = 10~x104
m 2n
800
~
14.89 (m/s); •
that is, about 33.3 mijh (using the conversion factor of 2.237 mi/h per mjs).
Damped Forced Oscillations
In real physical systems there is always some damping, from frictional effects if nothing else. The complementary function Xc of the equation
mx" + ex' + kx = Fo cos wt
(17)
is given by Eq. (19), (20), or (21) of Section 2.4, depending on whether c > Ccr = ,J4kfli., c = Ccr. or c < Ccr· The specific form is not important here. What is important is that, in any case, these formulas show that xc(t) + 0 as t + +oo. Thus Xc is a transient solution of Eq. (17)one that dies out with the passage of time, leaving only the particular solution xp. The method of undetermined coefficients indicates that we should substitute
x(t) = A cos wt
+ B sin wt
in Eq. (17). When we do this, collect terms, and equate coefficients of cos wt and sin wt, we obtain the two equations
cwA + (k mw2 )B = 0
(18)
that we solve without difficulty for
(k  mw2 ) F 0 A=~~ (k  mw 2 ) 2 (cw ) 2 '
+
(19)
If we write
A cos wt
+ B sin wt =
C(cos wt cos a+ sin wt sin a) = C cos(wt a)
as usual, we see that the resulting steady periodic oscillation
Xp(t) = C cos(wt  a)
(20)
has amplitude (21) Now (19) implies that sin a = BjC > 0, so it follows that the phase angle a lies in the first or second quadrant. Thus tan a
==kA
B
cw mw2
with
0 <a < n ,
(22)
2.6 Forced Oscillations and Resonance
169
so tan 1
a = {
;r
cw k mw 2 cw + tan 1 k mw2
ifk > mw2 , if k < mw2
(whereas a = rr /2 if k = mw2 ). Note that if c > 0, then the "forced amplitude"defined as a function C(w) by (21)always remains finite, in contrast with the case of resonance in the undamped case when the forcing frequency w equals the critical frequency wo = ,Jkfm. But the forced amplitude may attain a maximum for some value of w, in which case we speak of practical resonance. To see if and when practical resonance occurs, we need only graph C as a function of w and look for a global maximum. It can be shown (Problem 27) that C is a steadily decreasing function of w if c ~ v'?JWi. But if c < v'?JWi, then the amplitude of C attains a maximum valueand so practical resonance occursat some value of w less than wo, and then approaches zero as w ~ +oo. It follows that an underdamped system typically will undergo forced oscillations whose amplitude is • Large if w is close to the critical resonance frequency; • Close to F0 jk if w is very small; • Very small if w is very large.
Example 6
Find the transient motion and steady periodic oscillations of a damped massandspring system with m = 1, c = 2, and k = 26 under the influence of an external force F (t) = 82 cos 4t with x (0) = 6 and x' (0) = 0. Also investigate the possibility of practical resonance for this system. The resulting motion x(t) problem
Solution
= Xtr(t) +
Xsp(t) of the mass satisfies the initial value
x " + 2x' + 26x = 82cos4t;
x(O) = 6,
x'(O) = 0.
(23)
Instead of applying the general formulas derived earlier in this section, it is better in a concrete problem to work it directly. The roots of the characteristic equation
r 2 + 2r + 26
= (r +
1)2 + 25
=0
are r = 1
± 5i , so the complementary function is
When we substitute the trial solution
x (t) = A cos 4t + B sin4t
in the given equation, collect like terms, and equate coefficients of cos 4t and sin 4t, we get the equations lOA+ 8B = 82,  8A + lOB = 0
170
Chapter 2 Linear Equations of Higher Order
with solution A = 5, B = 4. Hence the general solution of the equation in (23) is
x(t) = et (c 1 cos 5t
+ c2 sin 5t) + 5 cos 4t + 4 sin 4t.
At this point we impose the initial conditions x (0) = 6, x' (0) = 0 and find that c1 = 1 and c 2 = 3. Therefore, the transient motion and the steady periodic oscillation of the mass are given by
X1r(t) = et (cos 5t  3 sin 5t)
and . 4t = .J4i ( ,J41 5 cos 4t + ,J41 4 sm . 4t ) = 5 cos 4t + 4 sm
=
Xsp (t)
.J4i cos(4t a)
where a = tan 1 (~) ~ 0.6747. Figure 2.6.8 shows graphs of the solution x(t) = x 1r(t) value problem
x"
+ Xsp(t) of the initial
(24)
+ 2x' + 26x =
82 cos 4t,
x (0) = x 0 ,
x' (0) = 0
for the different values x 0 = 20, 10, 0, 10, and 20 of the initial position. Here we see clearly what it means for the transient solution x 1r(t) to "die out with the passage of time," leaving only the steady periodic motion Xsp(t). Indeed, because X~r(t) + 0 exponentially, within a very few cycles the full solution x(t) and the steady periodic solution Xsp(t) are virtually indistinguishable (whatever the initial position xo).
X
xo= 20
20
X =
Xsp(f)
10
5
 10
 20
FIGURE 2.6.8. Solutions of the initial value problem in (24) with x 0 =  20,  10, 0, 10, and 20.
2.6 Forced Oscillations and Resonance
10
9
171
Practical resonance
8 7
6
(..)
To investigate the possibility of practical resonance in the given system, we substitute the values m = 1, c = 2, and k = 26 in (21) and find that the forced amplitude at frequency w is C(w) = 82 .J676 48w2 + w4
.
5 4 3 2 I
The graph of C(w) is shown in Fig. 2.6.9. The maximum amplitude occurs when
5
00
10
w
15
20
, 41 (4w 3  96w) C (w)    =:::::: (676 48w 2 + w 4 ) 312
164w(w2  24) ::;;;;::  0 (676 48w2 + w4 ) 312  ·
FIGURE 2.6.9. Plot of amplitude C versus external frequency w.
Thus practical resonance occurs when the external frequency is w = .J24 (a bit less than the massandspring's undamped critical frequency of w0 = .Jk1iii = .J26 ).
•*'
•
Problems
Each of Problems 15 through 18 gives the parameters for a forced mass springdashpot system with equation mx" +ex'+ kx = Fo cos wt. Investigate the possibility of practical resonance of this system. In particular, find the amplitude C(w) of steady periodic forced oscillations with frequency w. Sketch the graph of C (w) and find the practical resonance frequency w (if any).
In Problems 1 through 6, express the solution of the given initial value problem as a sum of two oscillations as in Eq. (8 ). Throughout, primes denote derivatives with respect to time t. In Problems 14, graph the solution function x(t ) in such a way that you can identify and label (as in Fig. 2.6.2) its period.
1. x" + 9x = 10cos2t; x(O) = x'(O) = 0 2. x" + 4x = 5 sin 3t; x(O) = x'(O) = 0 3. x" + lOOx = 225 cos St + 300 sin St; x(O) 375, x'(O) = 0 4. x" + 25x = 90cos4t; x(O) = 0, x ' (O) = 90 5. mx" + kx = F0 cos wt with w =I= w0 ; x (0) = x 0 , x ' (0) = 0 6. mx " + kx = Fo cos wt with w = w0 ; x (0) = 0, x' (0) = Vo
In each of Problems 7 through I 0, find the steady periodic solution Xsp(t) = C cos(wt  a) of the given equation mx" + ex'+ kx = F(t) with periodic forcing function F(t) of frequency w. Then graph Xsp(t) together with (for comparison) the adjusted forcing function F1 (t) = F(t)/ mw.
15. m = 1, c = 2, k = 2, F0 = 2 16. m = 1, c = 4, k = 5, Fo = 10 17. m = 1, c = 6, k = 45, F0 =50 18. m = 1, c = 10, k = 650, F0 = 100 19. A mass weighing 100 lb (mass m = 3.125 slugs in fps units) is attached to the end of a spring that is stretched 1 in. by a force of 100 lb. A force F0 cos wt acts on the mass. At what frequency (in hertz) will resonance oscillations occur? Neglect damping. 20. A frontloading washing machine is mounted on a thick rubber pad that acts like a spring; the weight W = mg (with g = 9.8 mjs2 ) of the machine depresses the pad exactly 0.5 em. When its rotor spins at w radians per second, the rotor exerts a vertical force F0 cos wt newtons on the machine. At what speed (in revolutions per minute) will resonance vibrations occur? Neglect friction. 21. Figure 2.6.10 shows a mass m on the end of a pendulum (of length L) also attached to a horizontal spring (with constant k). Assume small oscillations of m so that the spring remains essentially horizontal and neglect damping. Find the natural circular frequency Wo of motion of the mass in terms of L, k , m, and the gravitational constant g .
7. 8. 9. 10.
x"+4x' +4x x" + 3x' + Sx 2x" + 2x' + x x" + 3x' + 3x
= = = =
l0cos3t 4cos St 3 sin lOt 8 cos lOt+ 6 sin lOt
In each of Problems 11 through 14, find and plot both the steady periodic solution Xsp(t) = C cos(wt a) of the given differential equation and the transient solution x1r(t) that satisfies the given initial conditions.
11. 12. 13. 14.
x" + 4x' + Sx = lOcos 3t; x(O) = x'(O) = 0 x" +6x' + 13x = 10sin5t; x(O) =x'(O) =0 x" + 2x' + 26x = 600cos lOt; x (O) = 10, x'(O) = 0 x" + 8x' + 25x = 200cost + 520sint; x (O) =  30, x' (O) =  10
Show that there are two values of w for which resonance occurs..172 Chapter 2 Linear Equations of Higher Order In particular.a).6..10. k )(k.:. (21). The first floor is attached rigidly to the ground.) 27. Assume that in an earthquake the ground oscillates horizontally with amplitude Ao and circular frequency w. show that C attains a maximum value (practical resonance) when w = Wm = J~ . except with sin(wt a) in place of cos(wt.6. 2. The pendulumandspring system of Problem 21. ( 17)) is given by (b) Suppose that e 2 < 2mk. it requires a horizontal force of 5 tons to displace the second floor a distance of 1 ft. The mass. a.Z 0 2 FIGURE 2. as shown in Fig. A building consists of two floors.000 lb).Jkfiii. 22. and g. Find the natural circular frequency of the system in terms of m. FIGURE 2. (Suggestion: Add the steady periodic solutions separately corresponding to E 0 cos wt and F0 sin wt (see Problem 25). (a) Show that the amplitude of the steady periodic solution of the differential equation mx" +ex' + kx = mAu} cos wt (with a forcing term similar to that in Eq. A mass m hangs on the end of a cord around a pulley of radius a and moment of inertia I.2~2 < Wo = [f. The elastic frame of the building behaves as a spring that resists horizontal displacements of the second floor. (a) What is the natural frequency (in hertz) of oscillations of the second floor? (b) If the ground undergoes one oscillation every 2. k. Thus the resonance frequency in this case is larger (in contrast with the result of Problem 27) than the natural frequency w0 = ..J2. ·~ ·~ ·~ ' k with both cosine and sine forcing termsderive the steady periodic solution Xsp(t) = j E0 + F. .4iWi. According to Eq.. Given the differential equation L mx" + ex' + kx = Eo cos wt + Fo sin wt . and find both..25 s with an amplitude of 3 in.{3). what is the amplitude of the resulting forced oscillations of the second floor? 24.'. Assume small oscillations so that the spring remains essentially horizontal and neglect friction. the amplitude of forced steady periodic oscillations for the system mx" + ex' + kx = F0 cos wt is given by (a) If e . .11. show that it is what one would expectthe same as the formula in (20) with the same values of C and w. 23. where a is defined in Eq. ecr/. and the second floor is of mass m = 1000 slugs (fps units) and weighs 16 tons (32.) . Derive the steady periodic solution of mx" +ex'+ kx 28. (Suggestion: Maximize the square of C. Show that the maximum amplitude occurs at the frequency Wm given by Wm = k ( 2mk ) m 2mk.. 25.springpulley system of Problem 22.a . I. 26. resulting in an external horizontal force F(t) = mA 0 w 2 sin wt on the second floor. A mass on a spring without damping is acted on by the external force F (t) = F0 cos 3 wt. The rim of the pulley is attached to a spring (with constant k). As indicated by the cartwithflywheel example discussed in this section.e 2 • = Fo sinwt.1(F0 jE0 ). an unbalanced rotating machine part typically results in a force having amplitude proportional to the square of the frequency w.11..6. (22) and {3 = tan.mw 2 ) 2 + (cw) 2 cos(wt. where ecr = steadily decreases as w increases. show that C (b) If e < ecr/h.
this gives C as a function of v.2.7.7. Maximum resonance vibrations with amplitude about 14 em occur around 32 mij h.1.7.1 satisfy the basic circuit equation Ldt FIGURE 2. As a consequence.1. dt (1) R FIGURE 2. but then subside to more tolerable levels at high speeds. (2) . Verify these graphically based conclusions by analyzing the function C(w). • • •. and A capacitor with a capacitance of C farads in series with a source of electromotive force (such as a battery or a generator) that supplies a voltage of E (t) volts at time t. it initially oscillates with 3 Velocity (mi/h) FIGURE 2. The series RLC circuit..7 Electrical Circuits 173 Automobile Vibrations Problems 29 and 30 deal further with the car of Example 5.6. . 7.1 is closed.. As shown in Fig. dl + RI + 1 Q = c E (t) . Table of voltage drops. An inductor with an inductance of L henries.1 with the aid of this table and one of Kirchhoff's laws: The (algebraic) sum of the voltage drops across the elements in a simple loop of an electrical circuit is equal to the applied voltage.E o 11) = cwa and Fo = ka.2. 2.7. as the car accelerates gradually from rest. 15 12 + kx = Eo cos wt + Fo sin wt 8 . Figure 2. the voltage drops across the three circuit elements are those shown in the table in Fig. it consists of L A resistor with a resistance of R ohms. 7. this differential equation becomes mx" + ex' where Eo amplitude slightly over 5 em. 2. in which time is measured in seconds. 6 E <t: Because w = 2nvjL when the car is moving with velocity v. find the practical resonance frequency and the corresponding amplitude. The relation between the functions Q and I is dQ =l(t). 2.2. CIr(!Uit.12 shows the graph of the amplitude function C(w) using the numerical data given in Example 5 (including c = 3000 N·s/m). . With y = a sin wt for the road surface.12. Here we examine the RLC circuit that is a basic building block in more complicated electrical circuits and networks. Amplitude of vibrations of the car on a washboard surface. Element · Inductor Resistor Capacitor L1 d! dt Rl CQ We will always use rnks electric units. the current and charge in the simple RLC circuit of Fig. 2.7. 30. It indicates that. this results in a current of /(t) amperes in the circuit and a charge of Q(t) coulombs on the capacitor at time t . ~ 29. Its upward displacement function satisfies the equation mx" + ex' + kx = cy' + ky when the shock absorber is connected (so that c > 0 ).6. According to elementary principles of electricity. In particular. If the switch shown in the circuit of Fig. 2. Apply the result of Problem 26 to show that the amplitude C of the resulting steady periodic oscillation for the car is given by 9 0. We can analyze the behavior of the series circuit of Fig.
All we need to know about these constants and functions is that they satisfy Eqs.6 for mechanical systems can be applied at once to electrical circuits. Mechanical. When the switch is open. (3) and substitute I for Q' to obtain LI" + RI' +I= E ' (t). This idea is the basis of analog computerselectrical models of mechanical systems. or even dangerous. The table in Fig. More concretely. The MechanicalElectrical Analogy It is striking that Eqs. under the assumption that the voltage E(t) is known. there is a current I (t) in the circuit and a charge Q(t) on the capacitor. our mathematical model for the RLC circuit. 2 . no current flows in the circuit.174 Chapter 2 Linear Equations of Higher Order If we substitute (1) in Eq. We can then learn a good deal about electricity by studying this mathematical model. and capacitor in an electrical circuit as "black boxes" that are calibrated by the constants R.3 details this important mechanicalelectrical analogy. when the switch is closed. Massm Damping constant c Spring constant k Position x Force F Inductance L Resistance R Reciprocal capacitance 1I C Charge Q (using (3) (or current I using (4))) Electromotive force E (or its derivative E' ) FIGURE 2. In most practical problems it is the current I rather than the charge Q that is of primary interest. (1) through (4). we get the secondorder linear differential equation LQ" + RQ' + Q = c 1 E(t) (3) for the charge Q(t). It suffices to regard the resistor. 2. .7. A battery or generator is described by the voltage E(t) that it supplies.7. The performance of the mechanical system can then be predicted by means of accurate but simple measurements in the electrical model. most of the results derived in Section 2. 1 c (4) We do not assume here a prior familiarity with electrical circuits. inaccurate.electrical analogies.3 can be used to construct an electrical model of a given mechanical system. so we differentiate both sides of Eq. L. This is especially useful when the actual mechanical system would be expensive to construct or when measurements of displacements and velocities would be inconvenient. The fact that the same differential equation serves as a mathematical model for such different physical systems is a powerful illustration of the unifying role of mathematics in the investigation of natural phenomena. Analog computers modeled the first nuclear reactors for commercial power and submarine propulsion before the reactors themselves were built. As a consequence. the correspondences in Fig. inductor.7. and C. using inexpensive and readily available circuit elements.3. (2). (3) and (4) have precisely the same form as the equation mx" +ex'+ kx = F(t) (5) of a massspringdashpot system with external force F(t).
7 Electrical Circuits 175 In the typical case of an alternating current voltage E(t) =Eo sinwt.LCw o ::s. kmw 2 0 ~a~ rr. and a steady periodic current / 5p. Then the steady periodic current /5 p(t) = Eo z cos(wt. Eq. (5) with F(t) = Fo cos wt is X 5 p(t) = Fo cos(wt. z= J R 2 + ( wL ~C J (ohms).a) j(k. 1/ C fork. If we make the substitutions L form.a) (11) has amplitude lo= reminiscent of Ohm's law. thus (7) Recall from Section 2. so the roots of the characteristic equation have negative real parts). (19) through (22) there) that the steady periodic solution of Eq. the solution of Eq. a ::. R for c.2.6 (Eqs. we get the steady periodic current (8) with the phase angle a=tan _1 wRC 2' 1. JT. (6) As in a massspringdashpot system with a simple harmonic external force. (6) is the sum of a transient current 11r that approaches zero as t * +oo (under the assumption that the coefficients in Eq. I = EI R. (9) Reactance and Impedance The quantity in the denominator in (8).mw2 ) 2 + (cw)l . (6) are all positive. (4) takes the form 1 LI" + RI' + CI = wE0 coswt. where a= tan 1 cw . z' Eo (12) . (10) is called the impedance of the circuit. and wE0 for F0 .
5). and voltage. Eo . and C 5 x 104 farad (F). The differential equation in (6) takes the form (O. Isp(t) = .sm(wt .7. Example 1 Consider an RLC circuit with R = 50 ohms (Q).J R 2 + S2 . To do so. (14) z where S LCw2 . Find the current in the circuit and the time lag of the steady periodic current behind the voltage. Initial Value Problems FIGURE 2. When we want to find the transient current. (9) that a is as in Fig.7.= tan. approximately 377 radjs.4.1 8 = tan. Solution . Reactance and delay angle. So we must first find I' (0). Time lag of current behind imposed voltage..sin 8 cos wt).7. So we take E(t) = 110 sin 377t and use equality in place of the symbol for approximate equality in this discussion.1 henry (H). we first introduce the reactance 1 S = wL. To convert Isp to a sine function..5.176 Chapter 2 Linear Equations of Higher Order Equation (11) gives the steady periodic current as a cosine function. wC S (13) Then Z = .8). the circuit is connected to a 110V.1 . charge. 2. A frequency of60 Hz means thatw = (2n)(60) radjs.1 . 2. Equation (11) now yields Isp(t) = ~0 (cos a cos wt +sin a sin wt) S cos wt + Z R sin wt ) Z R Eo ( =Z  = Eo (cos 8 sin wt . with delay angle 8 =a 1n. when both I (0) and Q(O) are zero.7. At time t = 0. 60Hz alternating current generator.l)I" + 501' + 2000I = (377)(110) cos 377t. z Therefore. whereas the input voltage E(t) = Eo sin wt was a sine function. (2) to obtain the equation LI'(O) + RI (0) + C Q(O) = E(O) 1 (16) to determine I' (0) in terms of the initial values of current. and we see from Eq.4. L = 0. FIGURE 2.. we substitute t = 0 in Eq.R wRC (15) This finally gives the time lag 8jw (in seconds) of the steady periodic current Isp behind the input voltage (Fig. we are usually given the initial values I (0) and Q(O).
..1. Eq.575).4561 + (1. The observation that after onefifth of a second we have 1Itr(0. their solution is c 1 = 0. The characteristic equation (0.575).648) Thus the time lag of current behind voltage is = 0....1 (0..846) sin(377t .441 + (1.yo. . 59. .. Now find the current in the circuit..0.2. I' (0) = 44c 1 . C. (16) gives I ' (O) = 0 as well.0.000047 A (comparable to the current in a single human nerve fiber) indicates that the transient solution dies out very rapidly. c2 = 1.575. 307)e.846) sin(377t.······ . still with I (0) = Q(O) = 0. (10) to find that the impedance is Z = 59.0. . .... indeed..0.4561 . is connected at timet = 0 to a battery supplying a constant 110 V. Isp = (1..311. With these approximations.004 = 0.456c2 + 584 = 0.4561 + 696cos(377t. == w 377 and the steady periodic current is 8 0.. ····· ····. (15) gives the sine phase angle: 8 = tan. and w = 377 in Eq..456c2 e.575). • .311)e.7 Electrical Circuits 177 We substitute the given values of R.•····• ···. Eq. ·········· ······ Example 2 Suppose that the RLC circuit of Example 1. with derivative I'(t) = 44c 1e441 .846 amperes (A)... L.0015 S..58 Q. Because I (0) = Q(O) = 0.. we obtain the equations I(0) = Ct + C2. With these initial values substituted. the general solution is I (t) = c 1e441 + c2 e. so the steady periodic amplitude is 110 (volts) = 1.575 0..307...2)1 < 0. Thus the transient solution is Itr(t) = ( .···· • ········.1)r 2 + 50r + 2000 = 0 has the two roots r 1 ~ 44 and r2 ~ 456..58 (ohms) Io = With the same data.
Therefore. L. In Section 2.6. • Electrical Resonance Consider again the current differential equation in (6) corresponding to a sinusoidal input voltage E (t) = Eo sin cvt. Suppose that we wanted to pick up a particular radio station that is broadcasting at frequency cv. The radio is a familiar example.1)I" + 50I' + 2000I = E ' (t) = 0. so Eq. but its capacitance C is varied as one operates the tuning dial. I (t) = (2. L 0.670. For typical values of the constants R. . and thereby (in effect) provides an input voltage E(t) = Eo sin cvt to the tuning circuit of the radio. Its inductance L and resistance R are constant.1 and the differential equation is (0. the graph of Io as a function of w resembles the one shown in Fig.JLC and then approaches zero as cv + +oo. I'(O) = 44c 1 .670)(e44r e456t) . Note that I (t) + 0 as t + +oo even though the voltage is constant.178 Chapter 2 Linear Equations of Higher Order Solution We now have E(t) =110.456c2 = 1100 for c1 = c2 = 2. By contrast.7. many common electrical devices could not function properly without taking advantage of the phenomenon of resonance.6. It reaches a maximum value at cvm = 1/. The resulting steady periodic current Isp in the tuning circuit drives its amplifier. with the volume of sound we hear roughly proportional to the amplitude I 0 of Isp · To hear our preferred station (of frequency cv) the loudestand simultaneously tune out stations broadcasting at other frequencieswe therefore want to choose C to maximize Io. Its general solution is the complementary function we found in Example 1: We solve the equations I (0) = C) + C2 = 0.6 we emphasized the importance of avoiding resonance in most mechanical systems (the cello is an example of a mechanical system in which resonance is sought). 2. (16) gives I (0) = 1 £(0) 110 = = 1100 (A/s).7. The effect of frequency on lo. A highly simplified model of its tuning circuit is the RLC circuit we have discussed. and in tum its loudspeaker. We have seen that the amplitude of its steady periodic current is (17) FIGURE 2. and E0 . the critical frequency Wm is the resonance frequency of the circuit. C.
so that a steady current of 4 A is flowing in the circuit.7. (b) Write the solution in the form l sp(t) = C cos(wt .Q= c 1 E(t) for the charge Q = Q(t) on the capacitor at timet. and I (0) = 0. 3. the switch is thrown to position 2. Note that I(t) = Q'(t). when c= 1 2. In the circuit of Fig. Given the same circuit as in Problem 1. Problems 7 through 10 deal with the RC circuit in Fig.7. Suppose that the battery in Problem 2 is replaced with an alternatingcurrent generator that supplies a voltage of E (t) = 100 cos 60t volts. wC 1 that is. suppose that the switch is initially in position 2. the second controls the frequency of a signal that the radio itself generates. but modem AM radios have a more sophisticated design.8. This is the way that old crystal radios worked. 2.=0. a series circuit containing an inductor with an inductance of L henries. . known as the intermediate frequency. Lw (18) So we merely turn the dial to set the capacitance to this value.. Find I (t) and show that I (t) + 4 as t + +oo. a resistor with a resistance of R ohms. Find the maximum current in the circuit for t ~ 0. so that I(O) = 0 and E = 100 fort ~ 0. a capacitor (C farads). 2. but is thrown to position 1 at timet = 0. 2. and I (0) = 0. take L = 1. R = 20. (a) Substitute l sp(t) = A cos 60t + B sin 60t and then determine A and B to find the steadystate current l sp in the circuit. 1. containing a resistor ( R ohms).2. R = 40. with the switch in position 1. E(t) = WOe .7. This technique has the advantage that the several RLC circuits used in the amplification stages easily can be designed to resonate at 455 kHz and reject other frequencies. We see at a glanceno calculus requiredthat / 0 is maximal when w L .a). 2. Suppose also that the switch has been in position 1 for a long time. With everything else the same. suppose that E(t) = WOe . A pair of variable capacitors are used.7. now find I (t). (2) reduces to the linear firstorder equation LI' + RI = E(t). resulting in far more selectivity of the receiver as well as better amplification of the desired signal. but no inductor.7. In the circuit of Fig. and the source E of emf is a battery supplying 100 V to the circuit. 2. In the circuit of Fig. The circuit for Problems I through 6. 6. with the switch in position 1. thinking of w as a constant with C the only variable.7. 7.101 cos60t.7. is then amplified in several stages. 4. The resulting beat frequency of 455 kHz. and a source of electromotive force (emf). 2. Find I (t). and E(t) = 30cos60t + 40sin60t. ( 3) gives the linear firstorder differential equation dQ dt R +.7. but no capacitor. L R FIGURE 2. 7. L = 2. R = 10. The first controls the frequency selected as described earlier. 2. a switch. 5. suppose that L = 2. kept close to 455 kilohertz (kHz) above the desired frequency. In this case Eq. so that I (0) = 4 and E = 0 for t ~ 0.101 .7.7.7. Problems 1 through 6 deal with the RL circuit of Fig. In the circuit of Fig. with the switch in position 1.7 Electrical Circuits 179 But examine Eq. R = 25 Q . a source of emf. At time t = 0. Find I (t). suppose that L = 5 H. Substitution of L = 0 in Eq. (17).
2.8 at timet = 0 (with the switch closed).0025 F. E(t) = 50sin2t V = 200 Q. C. express this frequency in terms of Land In Problems 11 through 16. Suppose that in the circuit of Fig. using the appropriate value of w. R = 50 E(t) R L = 10 H. R = 200. if R. R 12. Q(O) = 0. L = 2 H.02 F. = 100 V. R = 25 Q. Suppose that in the circuit of Fig. 2. (a) Find Q(t) and I(t). The circuit for Problems 7 through 10. 25. 2. on the uniqueness of solutions of linear initial value problems (as guaranteed by Theorem 2 of Section 2. C = 0. Q(O) = 0. (4 ). Q(O) = 0 19. an RLC circuit with input voltage E(t) is described. to find the steady periodic current in the form Isp(t) = Io sin(wt. an RLC circuit with R = 0with input voltage E(t) = E 0 sinwt.8). directly or indirectly. 24. Consider an LC circuitthat is.1(wRC). The circuit and input voltage of Problem 15 with I (0) = 0 and Q(O) = 0 23. (a) Find Q(t) and I(t). c = 2.2 H. IEIJ Endp?int Problems and Eigenvalues You are now familiar with the fact that a solution of a secondorder linear differential equation is uniquely determined by two initial conditions. R = 20 14. E(t) = 120cos 377t V = 300cos lOOt+ 400sin lOOt V 15. then any solution of LI" + RI' +I j C = 0 is a transient solutionit approaches zero as t * +oo. L = 2 H. Q(O) = 5 18.8.8..51 (volts). Prove this. (a) Find the charge Q(t) and current I (t) in the RC circuit if E(t) = Eo (a constant voltage supplied by a battery) and the switch is closed at time t = 0. C = 0.1 V.{3) = tan. R = E(t) In each of Problems 20 through 22. and C are positive. The circuit and input voltage of Problem 11 with I (0) = 0 and Q(O) = 0 21.02 F. It was stated in the text that. C = 0. 20. L = 2 H. C = 5 x 1o6 F. I(O) = 0. E(t) = 100sin lOt V Q. E(t) = 110 sin 60rrt V 16. R = 60 Q. and E(t) = 100cos 120t./IT.7. Substitute Qsp(t) = A cos wt + B sin wt in the differential equation to show that the steady periodic charge on the capacitor is Qsp(t) = where f3 EoC 16 Q. L = 0.180 Chapter 2 Linear Equations of Higher Order 11.8. = . and Q(O) = 0.101 V. y'(a) = 0 (1) is the trivial solution y (x) 0. (b) What is the amplitude of the steadystate current? 10. (b) What is the maximum charge on the capacitor for t ~ 0 and when does it occur? 9. Find the current I (t) using the given initial current (in amperes) and charge on the capacitor (in coulombs). The circuit and input voltage of Problem 13 with I (0) = 0 and Q(O) = 3 22.01 F. C = 0. FIGURE 2. L = 5 H. E(t) = 200cos5t V L = 5 H. R = 60 Q. C = 0. y (a) = 0. the parameters of an RLC circuit with input voltage E(t) are given. Substitute Isp(t) = A cos wt + B sin wt in Eq. C = 5 x 104 F. and E(t) = 100e. lim Q(t) = E 0C and that t>+oo lim I (t ) = 0.7. Q. (b) Show that t> +oo 100 Q. C = 0. R = 30 Q. Jl + w 2 R 2 C 2 cos(wt .001 F. R = In Problems 17 through 22. 7.0025 F. C = 0. 13. L. Most of Chapter 2 has been based.7.02. Prove that the amplitude I 0 of the steady periodic solution of Eq. 8.2). In particular. L = 2 H. the only solution of the initial value problem y" + p(x)y' + q(x) y = 0. Show that unbounded oscillations of current occur for a certain resonance frequency. I (0) = 0. plot both the steady periodic current Isp(t) and the total current I (t) = Isp(l) + I 1r(t). L = 10 H. (6) is maximal at frequency w = 1/. so that Q (0) = 0. E(t) = lOOe.5 X 104 . C = 0. we haveR = 10. I(O) = 0. Q(O) = 1 17.005 F. 7. An emf of voltage E(t) = Eo cos wt is applied to the RC circuit of Fig. E(t) = lOOe.
2.2. y(a) = 0.J3 with B =I= 0 hits the desired target value y = 0 when x = n.2. Thus the endpoint value problem in (4) has infinitely many different nontrivial solutions. so the condition y(O) = 0 implies that A = 0. ·''' """' ""~"'"'''''•"'~""'. .2. Examples 1 and 2 illustrate the sorts of complications that can arise in endpoint problems. Now y(O) = A. • Jr.8 Endpoint Problems and Eigenvalues 181 In this section we will see that the situation is radically different for a problem such as y" + p(x)y' + q(x)y = 0.1 illustrates the fact that no possible solution y (x) = B sin x . In (2) we are to find a solution of the differential equation on the interval (a. so the condition y(O) = 0 implies that A = 0. Various possible solutions y(x) = B sinxv'3 ofthe endpoint value problem in Example 1. Example 1 Consider the endpoint problem y" + 3y = 0. No matter what the coefficient B is. Perhaps this does seem a bit surprising.74588. y(b) = 0. Graphically.8. But then FIGURE 2. y(rr) = 0. b) that satisfies the conditions y(a) = 0 and y(b) = 0 at the endpoints of the interval. every possible solution y(x) = B sin 2x hits automatically the desired target value y = 0 when x = n (whatever the value of B). Therefore the only possible solutions are of the form y(x) = B sin 2x. y(O) =A. the solution automatically hits the target value y = 0 for X = Again. Various possible solutions y (x) = B sin 2x of the endoint value problem in Example 2. 2. Thus the only solution of the endpoint value problem in (3) is the trivial solution y(x) = 0 (which probably is no surprise). Therefore the only possible solutions are of the form y (x) = B sin x . so the other condition y(n) = 0 requires that B = 0 also. FIGURE 2.J3. But now y(n) = B sin 2n = 0 no matter what the value of the coefficient B is. y(n) = B sin nJ3 ~ 0.8. with the coefficient 3 in one replaced by the coefficient 4 in the other. We consider a possible solution which starts at the left endpoint value and ask whether it hits the "target" specified by the right endpoint value. In mathematics as elsewhere. Remark 1: Note that the big difference in the results of Examples 1 and 2 stems from the seemingly small difference between the differential equations in (3) and (4 )."''"'"' • " "' Example 2 Consider the endpoint problem y" + 4y = 0.8. Such a problem is called an endpoint or boundary value problem. (4) The general solution of the differential equation is y(x) =A cos 2x 1t X + B sin 2x. (3) The general solution of the differential equation is y(x) =A cosxJ3 + B sinxJ3.1. y(n) = 0. as illustrated graphically in Fig. y(O) = 0." Remark 2: The "shooting" terminology used in Examples 1 and 2 is often useful in discussing endpoint value problems.8. Fig. For no B =I= 0 does the solution hit the target value y = 0 for x = rr. y(O) = 0. (2) The difference between the problems in Eqs. (1) and (2) is that in (2) the two conditions are imposed at two different points a and b with (say) a < b. Hence. sometimes "big doors tum on small hinges.
One might think of such a value as a "proper" value of A for which there exist proper (nonzero) solutions of the problem. (5) with p(x) = 0. Thus we saw in Example 2 that A = 4 is an eigenvalue of the endpoint problem y" + Ay = 0. so eigenvalues are sometimes called proper values (or characteristic values). If we take A = 3. y(O) = 0. (7) Solution We must consider all possible (real) values of Apositive. note that the problem in (5) is homogeneous in the sense that any constant multiple of an eigenfunction is again an eigenfunctionindeed. we get the equations in (3). That is. the prefix eigen is a German word that (in some contexts) may be translated as the English word proper. The number A is a parameter in the problem (nothing to do with the parameters that were varied in Section 2. Suppose that A* is an eigenvalue of the problem in (5) and that y*(x) is a nontrivial solution of the endpoint problem that results when the parameter A in (5) is replaced by the specific numerical value A*. q(x) = 1. Indeed. If A = 0. Thus we saw in Example 2 that y* (x) = sin 2x is an eigenfunction associated with the eigenvalue A* = 4.e. a = 0. so Then we call y* an eigenfunction associated with the eigenvalue A*. y(O) = 0. and negative. Examples 1 and 2 illustrate the typical situation for an endpoint problem as in (2): It may have no nontrivial solutions. and b = JT. More generally. then the equation is simply y" = 0 and its general solution is y (x) = Ax + B. as is any constant multiple of sin 2x. one associated with the same eigenvalue. y(rr) = 0. y(b) = 0.. nonzero) solution of the endpoint value problem? Such a value of A is called an eigenvalue of the problem. y(a) = 0. (6) whereas Example 1 shows that A = 3 is not an eigenvalue of this problem. Note that the problems in (3) and (4) both can be written in the form y" + p(x)y' + Aq(x)y = 0.182 Chapter 2 Linear Equations of Higher Order Eigenvalue Problems Rather than being the exceptional cases. Examples 1 and 2 show that the situation in an endpoint problem containing a parameter can (and generally will) depend strongly on the specific numerical value of the parameter. or it may have infinitely many nontrivial solutions. The question we ask in an eigenvalue problem is this: For what values of the parameter A does there exist a nontrivial (i. with A = 4.5). then so does any constant multiple cy*(x). y(L) =0 (L > 0). if y = y*(x) satisfies the problem in (5) with A = A*. An endpoint value problem such as the problem in (5)one that contains an unspecified parameter Ais called an eigenvalue problem. we obtain the equations in (4). Example 3 Determine the eigenvalues and associated eigenfunctions for the endpoint problem y" + Ay = 0. zero. It can be proved (under mild restrictions on the coefficient functions p and q) that any two eigenfunctions associated with the same eigenvalue must be linearly dependent. .
8 Endpoint Problems and Eigenvalues 183 Then the endpoint conditions y(O) = 0 = y(L) immediately imply that A = B = 0. (Recall that cosh ax = (eax that sinh ax = (eax . ' nn.2. nnx L n = 1. • . 9n 2 L2.3. f . If A < 0. This implies that B = 0. if n2 A =a 2 = L 2' y 4n 2 L2 . and its general solution is where A = c 1 + c2 and B = c 1 . (9) FIGURE 2. The hyperbolic sine and cosine graphs.. B sin a x .4 shows graphs of the first several of these eigenfunctions. and we may therefore conclude that the problem has no negative eigenvalues.c2 . . let us then write A = a 2 (with a > 0) to be specific. •• 0' n2n2 L2 ' Thus we have discovered that the problem in (7) has an infinite sequence of positive eigenvalues (8) L X With B = 1.L forn = 1. 0? Yes. A= 0 is not an eigenvalue of the problem in (7). so y(x) y(L) = 0 then gives y(L) = B sinaL= 0. but only provided that aL is a (positive) integral multiple of n: aL =n. Then the differential equation takes the form = II y . The condition y(O) = 0 implies that A = 0.a2 y =0 . gives y(L) = B sinhaL = 0. the eigenfunction associated with the eigenvalue An is Yn(X) = sin . and sinhx = 0 only for x = 0 (examine the graphs of y = sinhx andy = coshx in Fig. 3. 2n..8.8. The only remaining possibility is that A = a 2 > 0 with a > 0. y(L) = 0. FIGURE 2. Figure 2. But now the second endpoint condition..8. + eax)/2 and X so that y(x) = B sinh ax. 2. because a f. 3. Therefore.3..) Thus the only solution of the problem in (7) in the case A < 0 is the trivial solution y 0. . nnx etgen unctwns Yn (x ) = sin .8. 0. The . 3n. The condition Can this occur if B f.) The condition y(O) = 0 then gives y(O) =A coshO + B sinhO =A= 0. We see visibly how the endpoint conditions y(O) = y(L) = 0 serve to select just those sine functions that start a period at x = 0 and wind up at x = L precisely at the end of a halfperiod..eax)/2. .4. 2. 4. 2. •• 0 ' that is. In this case the differential equation is = with general solution y(x) =A cos ax+ B sin ax. so the only solution in this case is the trivial function y(x) 0..
a1y(a) + a2y'(a) = 0..8. The second endpoint condition y' (L) = 0 now gives y' (L) = BacosaL = 0.1)2rr 2 9rr 2 A=. The eigenfunctions . y(O) = 0. we get the problem of Example 4 (in which p(x) 0 and q(x) 1. With a 1 = 1 = b2 and a2 = 0 = b1. Then the differential equation is with general solution y(x ) = A cos ax+ B sin ax. in which the endpoint conditions involve values of the derivative y' as well as values of y: + p(x)y' + Aq(x)y = 0. (2n..8. so y(x) = B sin ax and y'(x) = Ba cos ax. so we take A = a 2 > 0 (a > 0) to be specific. b1y(b) + b2y'(b) = y" 0. According to a theorem whose precise statement we will defer until Section 9.. under the assumption that q (x) > 0 on the interval [a. y' (L) = 0. as in the previous example).. • .l)rrx Yn(x ) = sm '2L for n = 1. 3. (11) Solution Virtually the same argument as that used in Example 3 shows that the only possible eigenvalues are positive. = = Example 4 Determine the eigenvalues and eigenfunctions of the problem y" + AY = 0.1)rr 7r 3rr aL=. (10) where a1..5. ' 2' 2 ' 2 that is. y This will hold with B =1= 0 provided that a L is an odd positive integral multiple of rrj2: (2n .. 2. We see visibly how the endpoint conditions y(O) = y' (L) = 0 serve to select just those sine functions that start a period at x = 0 but wind up at x = L precisely in the middle of a halfperiod. This is also true of the following more general type of eigenvalue problem. b 1. 4. .l)rr x 2L (1 2) FIGURE 2.. if rr 2 (2n .. for n = 1. any eigenvalue of the form in (5) has a divergent increasing sequence )q < A2 < A3 < · · · < An < · · · + +oo of eigenvalues. b ].5 shows graphs of the first several of these eigenfunctions. a2.184 Chapter 2 Linear Equations of Higher Order Example 3 illustrates the general situation. Figure 2. The condition y(O) = 0 immediately gives A = 0. each with an associated eigenfunction. ' . and b2 are given constants. 2.1. ' 4L 2 ' 4L 2 ' 4L 2 Thus the nth eigenvalue An and associated eigenfunction of the problem in (11) are given by I and Yn(x ) = sin (2n . 3. .
JIL) + B ·0 = + B sin(. The remainder of this section is devoted to three such applications.Jix). A) because y 1 and y 2 will depend on A. (as we saw in Example 3). A)+ By2(X. When we collect coefficients of A and B in the resulting pair of equations. Most of the interest in eigenvalue problems is due to their very diverse physical applications. We assume that the portion of the string to one side of any point exerts a constant tension force T on the portion of the string to the other . We first write the general solution of the differential equation in the form y = Ayt(X. (13) Now A is an eigenvalue if and only if the system in (13) has a nontrivial solution (one with A and B not both zero). 3. 0 (in the unknowns A and B) which correspond to the equations in (13). . We write Yi (x. But such a homogeneous system of linear equations has a nontrivial solution if and only if the determinant of its coefficients vanishes. which implies that .. The Whirling String Who of us has not wondered about the shape of a quickly spinning jump rope? Let us consider the shape assumed by a tightly stretched flexible string of length L and constant linear density p (mass per unit length) if it is rotated or whirled (like a jump rope) with constant angular speed w (in radians per second) around its equilibrium position along the xaxis. as in Examples 3 and 4. in which y 1 (x) = cosax = cosx~ and y 2 (x) = sinax = sinx~. (14) in a concrete problem. and hence also linear in A and B.JIL = nn. Then we impose the two endpoint conditions. If A > 0.JIL) = 0. let's revisit the eigenvalue problem of Example 3. 0. then the differential equation y" + AY = 0 has the general solution y(x) = A cos(.. noting that each is linear in y and y'. The endpoint conditions y(O) = 0 and y(L) = 0 then yield the equations y(O) =A· 1 y(L) =A cos(. We therefore conclude that the eigenvalues of the problem in (10) are the (real) solutions of the equation (14) To illustrate Eq.2.8 Endpoint Problems and Eigenvalues 185 A general procedure for determining the eigenvalues of the problem in ( 10) can be outlined as follows.)B = at (A)A 0. The determinant equation D(A) = 0 corresponding to (14) is then simply the equation sin(.JIL) = 0. Numerous additional applications are included in Chapters 8 and 9 (on partial differential equations and boundary value problems).. 2. the solution of the equation D(A) = 0 in (14) may present formidable difficulties and require a numerical approximation method (such as Newton's method) or recourse to a computer algebra system.Jix) + B sin(. we therefore get a system of the form + f3t (A)B = a2(A)A + f32(J. or A= n 2 n 2 jL 2 for n = 1. A). For more general problems.
6(c). substitution of this and ( 15) in F = ma yields Ty'(x + 6.8. (16) x If we write FIGURE 2.Ty'(x). then the only solution of the problem . 2. x + 6. But what does all this mean in terms of the whirling string? It means that unless A in ( 17) is one of the eigenvalues in (8). Here we have r = y. each point moves in a circle centered at that point's equilibrium position on the xaxis.7 we see that the net vertical force in the positive ydirection is F = T sin(() x=O Whirling string (b) x=L + 6.8. (15)  8 ' Vertical component: 1 Tsin e 1 String (c) FIGURE 2.8. T (17) and impose the condition that the ends of the string are fixed. Next we recall from elementary calculus or physics the formula a = ro} for the (inward) centripetal acceleration of a body in uniform circular motion (r is the radius of the circle and w is the angular velocity of the body). we assume that the deflection of the string is so slight that sin() ~ tan() = y' (x) in Fig. t ) Equilibrium position (a) X side of the point.x string: / / 0 to get the differential equation of motion of the Ty" x . the minus sign because the inward direction is the negative ydirection.x).8)  T tan(). so the vertical acceleration of our piece of string is a = w2 y. We further assume that. Forces on a short segment of the whirling string.x + ~ pw 2 y 6. Because m = p 6. From Fig. so that as it whirls it also stretches to assume a curved shape. y so that T T · y'(x y'(x) ~ .7.6.x). so that F ~ Ty'(x + 6. we finally get the eigenvalue problem y" + AY = 0.pw y.x. We plan to derive a differential equation for y(x) by application of Newton's law F = ma to the piece of string of mass p 6. We found there that the eigenvalues of the problem in (7) are (8) with the eigenfunction Yn(x) = sin(mrx/L) associated with An.8. y(L) =0 (7) that we considered in Example 3. 2. y(O) = 0. Denote by y(x) the displacement of the string from the point x on the axis of rotation. Thus the string is elastic.186 Chapter 2 Linear Equations of Higher Order T ' I l .x.8)  T sin() ~ T tan(() + 6. 2 We now take the limit as 6. Finally. with the direction of T tangential to the string at that point.X x + L1x + pw2 y pwz A= = 0.x ]. The whirling string.x) 6.Ty'(x) + 6. as the string whirls around the xaxis. The only forces acting on this piece are the tension forces at its two ends.x corresponding to the interval [x.
We want to investigate the shape y = y(x) of this curve. But when w = w1. And when w is increased still further.4.. Suppose that we start the string rotating at speed = then gradually increase its speed of rotation. A consequence of the theory of elasticity is that for relatively small deflections of such a beam (so small that [y' (x)] 2 is negligible in comparison with unity). . with the positive yaxis directed downward. 2.8. the string pops back into its undeflected position along the axis of rotation! = The Deflection of a Uniform Beam We include now an example of the use of a relatively simple endpoint value problem to explain a complicated physical phenomenonthe shape of a horizontal beam on which a vertical force is acting. Consider the horizontal beam shown in Fig.. In this case the string remains in its equilibrium position with zero deflection. where • • E is a constant known as the Young 's modulus of the material of the beam. I denotes the moment of inertia of the cross section of the beam around a horizontal line through the centroid of the cross section. and • F(x) denotes the density of downward force acting vertically on the beam at the point x .2. So long as w < w 1. so the numerical value of en would necessarily be significant! y smaller than 1. then the force of its own weight distorts its longitudinal axis of symmetry into the curve shown as a dashed line in the figure.8.9. 2. 2.8. At angular speed w it assumes a shape of the form Yn = en sin(mr xj L) illustrated in Fig. we get a sequence of critical speeds of angular rotation.9. an adequate mathematical model of the deflection curve is the fourthorder differential equation EI/4 ) = F(x). 2. (19) FIGURE 2. 2. the string pops into a whirling position y = c 1 sin(n xj L). but it assumes much smaller deflections than those observed in Fig.8. The deflection curve.4 (where Cn 1).8. 3.8.8.8.8 Endpoint Problems and Eigenvalues 187 in (7) is the trivial solution y (x) 0. = (18) for n = 1. . the string remains in its undeflected position y 0. Distortion of a horizontal beam. L X Positive yvalues FIGURE 2. But if we equate (17) and (8) and solve for the value Wn corresponding to An. . Only at these critical angular speeds can the string whirl up out of its equilibrium position. uniform both in cross section and in material. Our mathematical model is not sufficiently complete (or realistic) to determine the coefficient en. If it is supported only at its ends. the deflection curve of the beam_ We will use the coordinate system indicated in Fig.
which depends only on the material of the beam. where A. which depends only on the shape of the cross section of the beam. w pounds per foot.10. For instance. The following table shows the boundary or endpoint conditions corresponding to the three most common cases. Eq. Figure 2. x + ~x] of the beam is approximately F(x) ~x. Thus we obtain a solution of Eq. (21) FIGURE 2. C3 . C2 . Values of the Young's modulus E of various materials can be found in handbooks of physical constants. B. (20) Note: We assume no previous familiarity with the theory of elasticity or with Eq. We will consider here the case in which the only force distributed along the beam is its own weight. (20) yields a second yields another yields Ely'= iwx 3 a final integration gives x0 + iC1x 2 + Czx + C3. (20) is a fourthorder differential equation. two constants: E. although a discussion here of their origin would take us too far afield.188 Chapter 2 Linear Equations of Higher Order Density of force? Yes. This proportionality involves.8. A beam might also be supported one way at one end but another way at the other end. X=L Simply supported or hinged where C 1 .. 2. its solution involves only the solution of simple firstorder equations by successive simple integrations. I. Fig.x 2 + Ax 3 24£/ w + Bx 2 + Cx + D. such as pounds per foot. .10 shows two common types of support. (20) of the form x=L x=O I Built in I! y(x) = . (19) takes the form Ely<4 l = w where E. (19) or (20) here. and w are all constant.8. and C4 are arbitrary constants. I = ~ n a 4 for a circular cross section of radius a. this means that the force acting downward on a very short segment [x. and Dare constants resulting from the four integrations. Two ways of supporting a beam. These last four constants are determined by the way in which the beam is supported at its ends. thus we develop an understanding of the equation by examining its solutions. so that F(x) = w. Although Eq. We will see that these conditions are applied readily in beam problems. Observe that. Then Eq. and I. where x = 0 and x = L. It is important to be able to begin with a differential equation that arises in a specific applied discipline and then analyze its implications.11 shows a cantilevera beam firmly fastened at x = 0 but free (no support whatsoever) at x = L. in essence. C. One integration of Eq. (20) implies that the fourth derivative y<4 l is proportional to the weight density w. however.8. The units of F (x) are those of force per unit length.
(x 4 24£/ w  2Lx 3 + L 3x ) (24) . It follows that C = w L 3/24.2. Example 5 Solution Rather than imposing then directly on Eq. and D determined by the conditions y(O) = y'(O) = 0 and y"(L) = y< 3\L) = 0. let us begin with the differential equation E I y<4 l = w and determine the constants as we proceed with the four successive integrations. The first two integrations yield Ely(3l = wx +A. I L 2 w x.11.. The cantilever. Hence y"(O) = 0 implies that B = 0. and then y " (L) = 0 gives 0 = ~wL 2 It follows that A = wL/2 and thus that + AL . We have the endpoint conditions y(O) = y" (O) = 0 = y(L) = y"(L). the deflection curve of the cantilever in Fig.8. with the coefficients A. 2. B. Determine the shape of the deflection curve of a uniform horizontal beam of length L and weight w per unit length and simply supported at each end. Ely" = ~wx 2 +Ax+ B. (23) Now y(O) = 0 implies that D = 0. Hence from Eq. (21 ). then. E ly(x) = i wx 4 4  / 2 wLx 3 + Cx +D.11 would be given by Eq. (22) FIGURE 2. (23) we obtain y (x) = .8 Endpoint Problems and Eigenvalues Endpoint Condition Simply supported Builtin or fixed end Free end y = y" = 0 y = y' = 0 y " = y(3) = 0 189 For example. (21). because y (L) = 0. C. corresponding to the fixed end at x = 0 and the free end at x = L. The conditions in (22) together with the differential equation in (21) constitute an endpoint value problem.8. EI y II Then two more integrations give Ix2 =2 and finally.
. .. ~ ... .27 ~) (980 c~) ~ 38484..190 Chapter 2 Linear Equations of Higher Order as the shape of the simply supported beam..60) 4 ~ 16_ 96 em (384)(2 X 1012 )(2.75 gjcm 3 and that its Young's modulus is E = 2 x 10 12 gjcms 2 . .68 in. so it will be more convenient to work in cgs units.54 ~m) = 1. . .) = 609. It is interesting to note that Ymax is proportional to L 4. . so I = = 4n(l. . . (25) that the same reduction in maximum deflection • could be achieved by doubling the radius a of the rod. ..·  v . ..48 c. .04) ' Ymax about 6. 5wL 4 = 384 EI · Ymax (25) •  . It is apparent from symmetry (see also Problem 17) that the maximum deflection Ymax of the beam occurs at its midpoint x = L/2. ._. suppose that we want to calculate the maximum deflection of a simply supported steel rod 20ft long with a circular cross section I in.._.6 dyn. (25) yields ~ (5)(38484. _ · · · · · ·· Example 6 For instance.. its mass per unit length) is so w = pg = (39. .) ( 2. Its linear mass density (that is.·~~· . its maximum deflection would be only onesixteenth as much.. . _ . Because I = ~na\ we see from Eq.. 4 Therefore Eq. . ··~···'<· '"' ' 4"~· • ~ ~~··· ~· . . and thus has the value L) y(2 Ymax = = 24 E I w ( 1 16 L  4 2 4 1 4) S L + 2L .27) 1 4 ~ 2. Thus our rod has length: and radius: a L = (20ft) ( 30.6)(609.. that is.. as the maximum deflection of the rod at its midpoint.42 in. in diameter. 2 m...60 em = (~ in..04 em . em s em The area moment of inertia of a circular disk of radius a around a diameter is I ~na4.only about 0. so if the rod were only 10ft long.. From a handbook we find that typical steel has density 8 = 7.~ .27 em. .
that has been "buckled" by an axial force of compression P applied at one end. y(O) = y(L) = 0 (7) that we considered in Example 3.E I. .. Only when the compressive force P is one of these critical forces should the rod "buckle" out of its straight (undeflected) shape. y(O) = y(L) = 0 (26) is used to model the actual (nonlinear) behavior of the rod. hinged at each end. (In practice a rod may fail at a significantly smaller force due to a contribution of factors not taken into account by the mathematical model discussed here. 3.8 Endpoint Problems and Eigenvalues 191 The Buckled Rod y y = y(x) p I p X x=O lx=L Figure 2. The forces n = 1. it is the upper bound for those compressive forces to which the rod can safely be subjected without buckling. recall from Eq. The buckled rod.12 shows a uniform rod of length L. If we write p 'A=EJ' (27) then the problem in (26) becomes the eigenvalue problem y" + A. (27) that P = A. E denotes the Young's modulus of the material of the beam and I denotes the moment of inertia of each cross section of the beam around a horizontal line through its centroid.y = 0.£ L. We found that its eigenvalues {'An} are given by (8) with the eigenfunction Yn = sin(mrx/L) associated with 'An.) . The smallest compressive force for which this occurs is (29) This smallest critical force P1 is called the Euler buckling force for the rod.8.8. In the theory of elasticity the linear endpoint boundary value problem FIGURE 2. Ely"+ Py = 0.12.2.. We assume this buckling to be so slight that the deflection curve y = y(x) of the rod may be regarded as defined on the interval 0 . (28) are the critical buckling forces of the rod.) To interpret this result in terms of the buckled rod. (Thus whirling strings and buckled rods lead to the same eigenvalues and eigenfunctions.£ x . 2. As in our discussion of the deflection of a uniform beam.
y" + AY = 0. 12.'"······ Example 7 For instance.. respectively.rr) = y'(rr) . • The eigenvalues in Problems 1 through 5 are all nonnegative.448 x 105 dynjlb. y(O) = 0. (c) Draw a sketch indicating the roots {an}f as the points of intersection of the curves y = tan z and y = .48 c. y'(rr) = 0 y" + AY = 0. (a) Show that Ao = 0 is an eigenvalue with associated . y(rr) = 0 y" + AY = 0. (b) Show that y = A cos ax + B sin ax satisfies the endpoint conditions if and only if B = 0 and a is a positive root of the equation tan z = 1/z.192 Chapter 2 Linear Equations of Higher Order . (a) Show that A = 0 is not an eigenvalue. (10ft) ( 30. (b) Show that the eigenfunctions are the functions {sinanx}f. all its eigenvalues are nonnegative. y' (. Use a method similar to that suggested in Problem 10 to show that the eigenvalue problem in Problem 6 has no negative eigenvalues. All the eigenvalues are nonnegative. y(2) = 0. y'(O) = 0. which is not of the type in (1 0) because the two endpoint conditions are not "separated" between the two endpoints. 8.54 4 ~m) ~ 2. y'(O) = 0. Deduce from this sketch that f3n ~ (2n + l)rr /2 when n is large. y'(O) = 0. Deduce from this sketch that a n ~ (2n . in diameter. Consider the eigenvalue problem y" +AY FIGURE 2. (Suggestion: Show graphically that the only root of the equation tanh z = . Draw a sketch showing these roots. y(l) = 0 y" + AY = 0. 2.. where f3n is the nth positive root of the equation tan z = z. where a n is the nth positive root of the equation tan z = .) = 304. y" +AY = 0. ln. 4. . In cgs units we have E L =2x = 10 12 gjcm·s 2 . First determine whether A = 0 is an eigenvalue.. 1. · .l)rr /2 when n is large.. 3.z. y (I) = y'(l).. (b) Show that the remaining eigenfunctions are given by Yn(x) = sin f3nx . Consider the eigenvalue problem = 0. y (l) + y' (l) = 0 has no negative eigenvalues. 2.04 cm4 .34 x 108 dyn ~ 976 lb.13.. y'(2) = 0 Consider the eigenvalue problem y" + AY all its eigenvalues are nonnegative. as indicated in Fig..) (2.13. Prove that the eigenvalue problem of Example 4 has no negative eigenvalues. and I= rr [co.) 11.. (a) Show that A = 0 is an eigenvalue with associated eigenfunction Yo(x) = x. Prove that the eigenvalue problem y" + AY = 0. y(. y (l) + y'(l) = 0. ·~··. 5. = 0. These roots {an }f are the abscissas of the points of intersection of the curves y = tan z andy= ljz. using the conversion factor 4.8 em. y(O) = 0.z. suppose that we want to compute the Euler buckling force for a steel rod 10 ft long having a circular cross section 1 in. (29) we find that the critical force for this rod is P 1 ~ 4.5 in. then find the positive eigenvalues and associated eigenfunctions. y'(rr) = 0. y'(rr) = 0 y" + AY = 0. so write A = a 2 where a ~ 0. Consider the eigenvalue problem y" + AY = 0. 10.8.8. 9. (a) Show that A = 0 is not an eigenvalue.. y(O) = 0. The eigenvalues are determined by the intersections of the graphs of y = tan z and y = l j z (Problem 6). y(l) + y'(I) = 0.rr ) = y(rr). y(rr) = 0. r Upon substituting these values in Eq. 6. Thus the eigenvalues and eigenfunctions of this problem are the numbers {a~}f and the functions {cosanx}f. 7.z is z z = 0.
y'(l) = 0. cos nx and sin nx.( x4 w (b) Show that its maximum deflection occurs where x = ( 15 . 14. Show that its deflection curve IS y(x) = . y(O) = 0. For the simply supported beam whose deflection curve is given by Eq.6% of the maximum deflection that would occur if the beam were simply supported at each end. 24£/ (b) Show that y' (x) = 0 only at x = 0. and thus that it follows (why?) that the maximum deflection of the cantilever is Ymax = y(L) = wL 4/(8£ /). Show that its shape is given by = y" + 2y' + AY = 0. (c) Show that the nth positive eigenvalue is n 2 and that it has two linearly independent associated eigenfunctions. Consider the eigenvalue problem (a) Show that A (b) Show that the roots of y'(x) = 0 are x = 0. y(O) = y(l) = 0. show that the only root of y'(x) = 0 in [0. (b) Show that there are no negative eigenvalues. so it follows (why?) that the maximum deflection of the beam is Ymax = Y( 2 L) = 384£/ ' wL 4 y" + 2y' + AY = 0. (b) Show that there is no eigenvalue A such that A < 1. and x = Lj2. (a) Suppose that a beam is fixed at its ends x x = L. 17. 15.. (a) A beam is fixed at its left end x = 0 but is simply supported at the other end x = L. (25). 13. (c) Show that the nth positive eigenvalue is A11 = n 2 rr 2 + 1. 18. . where an is the nth positive root of tan z = z. Show that its shape is given by . (24).4Lx 3 + 6L 2 x 2 ).. onefifth that of a beam with simply supported ends. (a) A uniform cantilever beam is fixed at x = 0 and free at its other end. Consider the eigenvalue problem 193 0 and = 16. so it follows (why?) that the maximum deflection is indeed that given in Eq. Show that the eigenvalues are all positive and that the nth positive eigenvalue is An = a~ + I with associated eigenfunction Yn(x) =ex sinanx.8 Endpoint Problems and Eigenvalues eigenfunction y0 (x) 1. where x = L. L] is x = L/2.J33) Lj16 and is about 41 . = 1 is not an eigenvalue. with associated eigenfunction Yn(x) = ex sin nrr x.2. x = L .
at+··· .3 we saw that solving a homogeneous linear differential equation with constant coefficients can be reduced to the algebraic problem of finding the roots of its characteristic equation.2xy' +. state (without proof) several theorems that constitute a review of the basic facts about power series. These techniques suffice for many of the nonelementary differential equations that appear most frequently in applications. n=O 00 (2) 194 . It has the form (1 .Power Series Methods  Introduction and Review of Power Series I n Section 2.a is an infinite series of the form L Cn(X n=O 00 at =co+ c1 (x. Perhaps the most important (because of its applications in such areas as acoustics. There is no similar procedure for solving linear differential equations with variable coefficients.a) 2 + · · · + Cn(X . such as the occasional equation that can be solved by inspection. heat flow. (1) If a = 0.x 2 )y" . at least not routinely and in finitely many steps. With the exception of special types.a)+ Cz(x. Recall first that a power series in (powers of) x . In this section we introduce the power series method in its simplest form and.n(n + l) y = 0. and electromagnetic radiation) is Bessel's equation of order n: Legendre's equation of order n is important in many applications. this is a power series in x : L CnXn = Co+ C1 X + CzX 2 + · · · + CnXn + · · · . linear equations with variable coefficients generally require the power series techniques of this chapter. along the way.
=L 1x and (1 1 00 xn = 1 + x + x 2 + x 3 + ··· . 3! 5! x2 · ' (7) cosh x = n=O x 2n x4 L = 1 +.1)x 2 2! + a(a.a by replacement of x with x. is the binomial series.+ . 2'. In this case the sum 00 f(x) = LCnXn n=O (4) is defined on I...1 Introduction and Review of Power Series 195 We will confine our review mainly to power series in x.+ . including x = 0.+. The following power series representations of elementary functions should be familiar to you from introductory calculus: (5) oo COS X (l)nx2n x2 = "'\"" ~ n=O (2n)'' = 1. The power series in (2) converges on the interval I provided that the limit (3) exists for all x in I. · · (2n)! 2! 4! ' oo (8) sinh x = L x2n+l n=O (2n + 1)! = x + . and we call the series L CnXn a power series representation of the function f on I. + ..·· · ' 4'. (What if lx I = 1?) The series in (11) is the geometric series.a. with a an arbitrary real number.2)x 3 3! + ··· . smx = n=O L oo oo ( l)nx2n+1 (2n + 1)! = x . Otherwise. If a is a nonnegative integer n. ' x3 xs x4 (6) .3. (11) n=O + x) a = 1 + ax + a(a. In contrast.+ . the series in ( 10) and ( 11) converge if lx I < 1 but diverge if lx I > 1. The series in (5) through (9) converge to the indicated functions for all x . but every general property of power series in x can be converted to a general property of power series in x .1)(a.. then the series in (1 2) terminates and the binomial series reduces to a polynomial of degree n which converges for all x.. · · 3! 5! ' x3 xs (9) (10) .+. . The series in (12). we observe the usual conventions that 0! = 1 and that x 0 = 1 for all x. its behavior for lx I = 1 depends on the value of a.. the series is actually infinite and it converges if lxI < 1 and diverges if lxI > 1. (12) In compact summation notation..
power series may be manipulated algebraically in much the same way as polynomials. then the series in (13) is the Maclaurin series 00 j(n) (0) L n=O n! xn = f(O) + j'(O)x + .( x . (13) are all defined). suppose that f (x) = ex. f (x) for all x in some open f is analytic at x = a. If a = 0. (13') reduces to the exponential series in (5). Power Series Operations If the Taylor series of the function f converges to interval containing a. if 00 00 f(x) = LanXn n=O and g(x) = L bnxn. It is rather awkward to compute the Taylor series of the tangent function using Eq. The Taylor series with center x = a of the function f is the power series > oo f(nl(a) L n=O (x. (6) and (7)). and hence f(n) (0) = 1 • for all n ~ 0. (13) because of the way in which its successive derivatives grow in complexity (try it!).196 Chapter 3 Power Series Methods Remark: Power series such as those listed in (5) through (12) are often derived as Taylor series.a)+ . · (13) in powers of x a. • every rational function is analytic wherever its denominator is nonzero. For instance.. if the two functions f and g are both analytic at x = a. Fortunately. then we say that the function example. n=O (14) then 00 f(x) + g(x) = L(an + bn)Xn (15) n=O and 00 f(x)g(x) = L CnXn n=O ..a)n = f(a) n! + f"(a) j'(a)(x. Then f(n) (x) = eX. For • every polynomial function is analytic everywhere. then so are their sum f + g and their product f · g. • more generally. the function h (x) = tan x = (sin x) 1(cos x) is analytic at x = 0 because cos 0 = 1 f 0 and the sine and cosine functions are analytic (by virtue of their convergent power series representations in Eqs. For example. In this case Eq..x 2 + 2! J"(O) f(3) (0) 3! x3 + · · ·. (13') For example. as is their quotient fIg if g(a) f 0. under the hypothesis that f is infinitely differentiable at x =a (so that the coefficients in Eq.a) 2 2! + .
Similarly... The Power Series Method The power series method for solving a differential equation consists of substituting the power series (18) in the differential equation and then attempting to determine what the coefficients co.1. The series in (15) is the result of term wise addition and the series in ( 16) is the result of formal multiplicationmultiplying each term of the first series by each term of the second and then collecting coefficients of like powers of x. but when it is we obtain an infinite series representation of a solution..1. we must first know what to substitute for the derivatives y'. This method is not always successful. .3. 3. on any open interval on which both the series in (14) converge.2 1 X4 + 24 ••. This division of the Taylor series for cos x into that for sin x yields the first few terms of the series tanx = x + 3x 3 + 15 x 5 + 315 x 7 + · · · . y".x 6 120 =2 1[ (2x) 3 (2x). 6 2 24 12 120 4 3 16 5 =x. in contrast to the "closed form" solutions that our previous methods have yielded.) The series in (15) and (16) converge to f(x) + g(x) and f(x)g(x).1 Introduction and Review of Power Series 197 where cn = aobn + a 1bn I + · · · + an bo. the sine and cosine series converge for all x.~) x3 + (_!_ + _1 + _1_) x5 + .. This is reminiscent of the method of undetermined coefficients. cz. but the tangent series in (17) converges only if lx I < rr /2. must be in order that the power series will satisfy the differential equation. the quotient of two power series can be computed by long division.x + . For example. . I 2 17 (17) Division of power series is more treacherous than multiplication. respectively.~ (2x) 1 . Before we can substitute the power series in (18) in a differential equation. + s!· · · J= 2 sm 2x 5 for all x.. X COS X Slll = (X  1X 3 + 1 X 5 120 6 • · · ) (1 1X 2 . . as illustrated by the computation shown in Fig. . . . ) = x + (~ . (Thus the processes strongly resemble addition and multiplication of ordinary polynomials. but now we have infinitely many coefficients somehow to determine. c. . The following theorem (stated without proof) tells us that the derivative y' of y = L cnxn is obtained by the simple procedure of writing the sum of the derivatives of the individual terms in the series for y. For example. the series thus obtained for fIg may fail to converge at some points where the series for f and g both converge.
. 00 and j'(x) = :Lncnxnl = c1 + 2c2x + 3c3x 2 + · · · n=! (20) at each point of I...198 Chapter 3 Power Series Methods X + x3 1~2 +~: ~> · ··)x X 3 x3 6 x3 2 x3 + + + 2x 5 15 xs 120 xs + 17x 7 315 x7 5040 x7 + ... differentiation of the geometric series 1 =~ L. 3 x3 3 + + x7 840 x7 72 4x7 315 x7 2x 5 15 2x 5 15 + . + .. For example.tx n =1 +x+x 2 + x 3 + ..... 24 xs 30 xs 6 720 + . then f is differentiable on I.1.. 1x n=O (11) . + ...1... Obtaining the series for tan x by division of series. THEOREM 1 Termwise Differentiation of Power Series If the power series representation f(x) = L CnXn =Co+ CtX + n=O 00 c2x 2 + C3X 3 + · · · (19) of the function f converges on the open interval I. + ... 15 17x 7 315 FIGURE 3. + .
that is. we shift the index of summation in the first sum. then an = bn for all n 0. Example 1 Solution Solve the equation y' + 2y = 0. then they are the same series. Thus we can replace n with n + 1 if.l' n=l (21) 00 L ncnXn. . In particular. we need the general term in each sum to be the term containing xn.1 Introduction and Review of Power Series 199 gives 1 o (1 . The result of making this shift in Eq. we start counting one step lower.3. To accomplish this. This theoremalso stated without prooftells us that if two power series represent the same function. In particular. (21) is the identity 00 :L)n n=O + 1)cn+1Xn + 2 L CnXn = 0.x)2 =L 00 n=l nxnl = 1 + 2x + 3x 2 + 4x 3 + · · · . at n = 0 rather than at n = 1. it follows from Theorem 2 that an = 0 for all n ~ 0. at the same time.a) that represents the function f. n=O 00 that is. THEOREM 2 If Identity Principle :L: anxn = :L: bnxn n=O n=O ~ 00 00 for every point x in some open interval I. note that 00 00 'L:ncn xnl = c1 + 2c2x + 3c3x 2 + · · · = L(n n=l n=O + 1)cn+lxn . The process of determining the coefficients in the series y = L CnXn so that it will satisfy a given differential equation depends also on Theorem 2. 00 :L)Cn + 1)cn+1 + 2cn]Xn = n=O 0. the Taylor series in (13) is the only power series (in powers of x. if L anxn = 0 for all x in some open interval. This is a shift of + 1 in the index of summation. To see how to do this. n=l n=O 00 00 To compare coefficients here. We substitute the series and and obtain y' = ~ L ncnXn.l + 2 L CnXn = 0.
consequently..x n =co L n=O n=O · n=O · 2x n' n' .) Consequently.X = coe y(x) = L CnX n = L (1) n . thereby obtaining the series on the right. (22) for all n 2: 0. 3 23 co = .' n n ~ l. 1 With n = 1. In the final step we have used the familiar exponential series in Eq. our solution takes the form oo oo 2n oo ( 2 )n co . we will have n 2nco c = ( . n. Eq.=. (22) gives 2co c. in terms of c0 . then it follows from the identity principle that (n + l)cn+t + 2cn = 0 for all n ~ 0. 2. (22) gives C3 2cz 23co = . With n = 0. Eq..200 Chapter 3 Power Series Methods If this equation holds on some interval. from n = 1 to n = 0.1 ) I. the latter will tum out to be the arbitrary constant that we expect to find in a general solution of a firstorder differential equation. (22) gives With n = 2. c2 .= 3 1 .l = n=l 00 00 L(n n=O + 1)cn+tXn (23) by shifting the index of summation by + 1 in the series on the left. •. (This is easy to prove by induction on n. • Shift of Index of Summation In the solution of Example 1 we wrote L ncnxn . Eq. c3 . That is. n + n + 1) and decreased the starting point by 1. This procedure is valid because each infinite series in (23) is simply a compact notation for the single series (24) .. 3! By now it should be clear that after n such steps. (5) to identify our power series solution as the same solution y(x ) = c0 e .2x we could have obtained by the method of separation of variables. we simultaneously increased the index of summation by 1 (replacing n with n + 1. Equation (22) is a recurrence relation from which we can successively compute c 1 .
(b) If 0 < p < oo. then L CnXn converges if lx I < p and diverges if lx I > p. then the series diverges for all x =!= 0. More commonly. a power series solution is not recognizable in terms of the familiar elementary functions. Then (a) If p = 0. then the series converges for all x. THEOREM 3 Radius of Convergence Given the power series L cnxn. Hence this formal process is justified only if in the end we can show that the power series we obtain converges on some open interval. The following theorem (which we state without proof) may be used for this purpose. n=l 00 00 n=3 we have decreased the index of summation by 2 but increased the starting point by 2. (25) will be quite sufficient for computing the radius of convergence. suppose that the limit p = hm Cn n*oo Cn+l . Eq. After all. we interpret a "decrease by k" as an increase by k = shift by 2 (n + n . The procedure illustrated in Example 1 for finding the coefficients {en} is merely a formal process and may or may not be valid. Its validityin applying Theorem 1 to compute y ' and applying Theorem 2 to obtain a recurrence relation for the coefficientsdepends on the convergence of the initially unknown series y = L cnxn. I (1t2n cofn! I= h . we have p = hm L .2) in the index of summation yields lkl. y = L CnXn is merely an assumed form of the solution.l)n+12n+lco/(n + 1)! n*OO 2 ' and consequently the series we obtained in Example 1 converges for all x. there always will be a number p such that exactly one of the three alternatives in Theorem 3 holds. The number p in (25) is called the radius of convergence of the power series CnXn. If so. Even if the limit in (25) fails to exist. I I (25) exists (p is finite) or is infinite (in this case we will write p = oo ).m n+1 .2Xn3. it then represents a solution of the differential equation on that interval. (c) If p = oo. Thus a L nCnXn1 = L(n. we can shift the index of summation by k in an infinite series by simultaneously increasing the summation index by k (n + n+k) and decreasing the starting point by k. You should check that the summation on the right is merely another representation of the series in (24). For instance. for the power series obtained in Example 1.2)Cn. from n = 1 ton = 3.1 Introduction and Review of Power Series 201 More generally. a shift by +2 (n + n + 2) yields If k is negative. we need a way of finding where it converges.3.. We know that the power series obtained in Example 1 converges for all x because it is an exponential series. For instance. This number may be difficult to find.= oo n*oo (. When we get an unfamiliar power series solution. . but for the power series we will consider in this chapter.
I = hm Cn + l n'> 00 n + 2 = 3. in turn. we substitute 00 Solution and to obtain 00 y' = L n=l ncnxn.Cn .1 00 (x . that is. In the first sum we can replace n = 1 with n = 0 with no effect on the sum.1 + 2 L CnXn = 0. Hence our proposed power series solution is (26) Its radius of convergence is p = n '> 00 . . I . from which we obtain the recurrence relation cn+l We apply this formula with n n+2 c 3 (n + 1) n for n 2:: 0. it is not difficult to show by induction on n that Cn =co 3n n+ I if n 2:: I.3) L n= l ncnxn. As before.3(n n=O + I )en+ I+ 2cn] Xn = 0. This yields 00 00 00 L nCnXn. n=O 00 so that 00 00 L n=l nCnXn . This is almost enough to make the pattern evident. and n = 2.202 Chapter 3 Power Series Methods Example 2 Solve the equation (x. 2 32 . n = 1. = 0. and find that 3 3 c2 = c 1 = co 3.3)y' + 2y = 0. In the second sum we shift the index of summation by + 1. 00 L [ncn.3(n + l)cn + l + 2cn = 0. 3n + 3 hm .3 L(n n=O n=O + l)cn+tXn + 2 L n=O CnXn = 0. The identity principle then gives ncn.3 L ncnXnl n=1 +2L n=O CnXn = 0.
1 (replace n = 1 with n = 2 and n with n . c0 + c 1x. An elementary solution (obtained by separation of variables) of our differential equation is y = 1j (3 .1 Introduction and Review of Power Series 203 Thus the series in (26) converges if 3 < x < 3 and diverges if lxl > 3.X+ Co+ C]X + L n=2 C11 X11 • Because the lefthand side contains neither a constant term nor a term containing x to the first power.I X11 n=2 = 1. which yield 11 x2 LnCnXnl = 1. in general. Thus we obtain the power series 00 y(x ) = 1 +x + L(n.x)2 on the interval 3 < x < 3. It follows that C2 = 1 · CJ = 1!. Thus this series (with the arbitrary constant c0 appropriately chosen) represents the solution 1 y(x) = (3. and the singularity at x = 3 is the reason why the • radius of convergence of the power series solution turned out to be p = 3.x on the righthand side.J.X+ LCnX 11 n=i n=O 00 so that 00 L:nc11 xn+l = 1x + LCnX 11 • n=i n=O Because of the presence of the two terms 1 and .l)! x n=2 11 • . of the series on the right for comparison. C3 = 2 · C2 = 2! . c 1 = 1. the identity principle now yields co = 1. C4 = 3 · C3 = 3! .1)! for n ~ 2.1)cnJ for n ~ 2.x) 2 • If we differentiate termwise the geometric series we get a constant multiple of the series in (26). we need to split off the first two terms. we get co co L(n. Example 3 Solution We make the usual substitutions y = 00 Lc 11 x 11 and y' = 00 L nc xn.1). and Cn = (n . In this particular example we can explain why.1)Cn.3. If we also shift the index of summation on the left by . that C11 = (n . and.
. 2.. we get Cz = co 2! ... When we apply the recurrence relation in (27) with n = 0... CJ 7! ...204 Chapter 3 Power Series Methods But the radius of convergence of this series is p = n>oo hm .Z + L CnXn = n=O 00 0.. We shift the index of summation in the first sum by +2 (replace n = 2 with n = 0 and n with n + 2). and 4 in turn. and thus we obtain the recurrence relation c n+Z  .~·. 3.(n + 1)(n + 2) Cn (27) for n ~ 0.· · . C4 = 4! . (n.. we find that 00 00 y' = Z:ncnXnl n=l and y 11 = ' ""' { ~n n 1) CnX n. CJ co and C6 = .. The identity (n + 2)(n + 1)cn+Z + Cn = 0 now follows from the identity principle.. n=2 Substitution for y and y" in the differential equation then yields L n=2 00 n(n  l)cnxn . c0 and c 1 are not predetermined and thus will be the two arbitrary constants we expect to find in a general solution of a secondorder equation.= 0. n 1 so the series converges only for x = 0. This gives 00 L(n n=O + 2)(n + 1)cn+2Xn + L n=O 00 CnXn = 0. This example serves as a warning that the simple act of writing y = L CnXn involves an assumption that may be false. • . If we assume a solution of the form  ...5!' and C7 = .1)! n! = n>oo lim ... What does this mean? Simply that the given differential equation does not have a (convergent) power series solution of the assumed form y = L cnxn. Example 4 Solution Solve the equation y" + y = 0. It is evident that this formula will determine the coefficients Cn with even subscript in terms of co and those of odd subscript in terms of c 1 . _ _____ . co 6! Taking n = 1. we find that cs ... and 5 in turn.2 .
using only their initial values (at x = 0) and the derivatives in (30). and that y = S (x) is the unique solution that satisfies the initial conditions y(O) = 0 and y' (0) = 1.1 Introduction and Review of Power Series Again.3. the pattern is clear. the Taylor series for the sine and cosine functions converge for ali x. Indeed. • The solution of Example 4 can bear further comment. (Can you use the series in (28) and (29) to show that [C(x)] 2 + [S(x)f = 1 for all x?) This demonstrates that The cosine and sine functions are fully determined by the differential equation y" + y = 0 of which they are the two natural linearly independent solutions. Both of these power series converge for all x. the ratio test in Theorem 3 implies convergence for all z of the series I:< l)nzn/(2n)! obtained from (28) by writing z = x 2 . Suppose that we had never heard of the sine and cosine functions. x2 x4 x6 x3 xs x 7 ) + · · · ) + CJ ( X++··· . . andrecognizing the importance of the differential equation y" + y = Owe can agree to call C the cosine function and S the sine function. We would then have discovered the two power series solutions (28) and (29) of the differential equation y" + y = 0. c 2k+l  ( 1)kcl (2k + 1)!.. we have discovered that y = C (x) is the unique solution of y" +y = 0 that satisfies the initial conditions y (0) = 1 and y' (0) = 0. and termwise differentiation of the two series in (28) and (29) yields C' (x) = S(x) and S' (x) = C(x). Hence it follows that (28) itself converges for all x. It is clear that C(O) = 1 and S(O) = 0. It follows that C(x) and S(x) are linearly independent. as does (by a similar ploy) the series in (29).+2! 4! 6! 205 ~ 1. C' (0) = 0 and S' (0) = 1. Note that we have no problem with the radius of convergence here. y (x) = c0 cos x + c 1 sin x. 3! 5! 7! ' that is. Thus with the aid of the power series method (all the while knowing nothing about the sine and cosine functions). we leave it for you to show (by induction) that fork and Thus we get the power series solution y(x) =Co ( 1 . (30) Consequently. For instance. let alone their Taylor series. there is no need to refer to triangles or even to angles. all the usual properties of these two functions can be established.
identify the particular solution in terms offamiliar elementary functions. y(O) = 0.1. 21. y (l) = l . FIGURE 3. 6.l)y' + 2y = 0 2. x 3 y' = 2y has no power series solution of the form y = L CnX 11 • 24.2. y" = 4y +y = X Show (as in Example 3) that the power series method fails to yield a power series solution of the form y = L C11 X 11 for the differential equations in Problems 15 through 18.2y' + y = 0.. y" 14. 15. 10. use the method ofExample 4 to find two linearly independent power series solutions of the given differential equation. Establish the binomial series in ( 12) by means of the following steps. Taylor polynomial approximations to sinx.2)y' + y = 0 2(x + l)y' = y 2(x . y" + 4y = 0. y" +9y = 0 12. y'(O) y" + y' . y" = y 13. y n=8 n = 16 y n = 24 n=5 n= 13 n=21 n=6 n= 14 n = 22 n=7 n = 15 n = 23 FIGURE 3. y'(O) Show that the equation 1. y' = 4y y' + 2xy = 0 (x. finally. Taylor polynomial approximations to cosx. and identify the general solution in terms offamiliar elementary functions. y'(O) = 0 y". In Problems 1 through 10. Determine the radius of convergence of each series. y(O) = 2. 23. (Of course. This is by no means an uncommon situation. 22. 4. y' =y 2y' + 3y = 0 y' = x 2 y (2x l)y' +2y = 0 (x. 19. y(O) = l. xy' + y = 0 17. 9.1.4y = 0. (c) Explain why the validity of the binomial series given in (12) follows from parts (a) and (b). Next determine Cn (in terms of n. (a) Show that y = (1 + x )C/ satisfies the initial value problem (1 +x)y' = ay. 7.1. find a power series solution of the given differential equation. x 2 y' + y = 0 16. and that this series converges if lxl < 1. first derive a recurrence relation giving C11 for n ~ 2 in terms of co or c 1 (or both).1. y (O) = 1. 3. 11. y'(O) = 3 y" .2 and 3. 8. (5) through (12) to identify the series solution in terms of familiar elementary functions. Many important special functions of mathematics occur in the first instance as power series solutions of differential equations and thus are in practice defined by means of these power series. 25. 5. Then apply the given initial conditions to find the values of co and c.3. For the initial value problem y" = y' + y. y(O) = 0. 20. as in the text) and. In the remaining sections of this chapter we will see numerous examples of such functions. Determine the radius of convergence of the resulting series. and use the series in Eqs.206 Chapter 3 Power Series Methods Figures 3. 2xy' = y 18.3 show how the geometric character of the graphs of cos x and sin x is revealed by the graphs of the Taylor polynomial approximations that we get by truncating the infinite series in (28) and (29).l)y' = 3y =1 = 2 In Problems 11 through 14.2y = 0. no one can prevent you from checking your work by also solving the equations by the methods of earlier chapters!) In Problems 19 through 22. (b) Show that the power series method gives the binomial series in (12) as the solution of the initial value problem in part (a). y(O) = 0.
differential equations can sometimes be used to sum infinite series. consider the infinite series 1 1 1 1 1 1++++···· 1! 2! 3! 4! 5! ' note the + + .3. 13. F1 = 1.x 5 + . = 2c7 + 2c3 cs. Conversely. We saw in Example 3 of Section 3. see Problem 48 of Section 2. 5. but its most important applications are to homogeneous secondorder linear differential equations of the form A(x)y" + B(x)y' + C(x)y = 0. Given these facts. This section introduces the use of infinite series to solve differential equations. = 2cg + 2c3 c 7 + (c5 ) 2 . and C are analytic functions of x. Substitute this series in y' = 1 + y 2 and equate like powers of x to derive the following relations: 3c3 7c7 1lc11 = = 1. (1) where the coefficients A. and with P = B I A and Q = CIA. . (3) ._l for n > 1. 26.· · · pattern of signs superimposed on the terms of the series for the number e. in most applications these coefficient functions are simple polynomials. 3. 27.. Note that P(x) and Q(x) will generally fail to be analytic at points where A(x) vanishes. y(O) = y'(O) = 1. (a) It's possible to show that the power series given here converges for all x and that termwise differentiation is valid. The power series method introduced in Section 3. y(O) is y(x) = tanx.x + · · · 2! 3! 4! 5! ' because the sum of the numerical series in question is simply f(l). 2. (1) in the form y" + P(x)y' + Q(x)y = 0 (2) with leading coefficient 1. To discover when it does succeed.2 Series Solutions Near Ordinary Points derive the power series solution 207 y(x) = L _!!_xn n=l n! ()() F. 1.x 7 3 15 315 (b) Solve this initial value problem to show that 62 g 1382 II + 2835x + 155925x + · · · · (d) Would you prefer to use the Maclaurin series formula in (13) to derive the tangent series in part (c)? Think about it! For a suggestion. (b) Because tanx is an odd function with y' (0) = 1. Fn = Fn2 + F. y " (O) = 1.3. B. (c) Evaluate f (1) to find the sum of the numerical series given here. 1.1 that the series method does not always yield a series solution.x + x + x . we rewrite Eq. 2cs 5cs 9c9 + (c3 ) 2 . of Fibonacci numbers defined by F0 = 0. tanx = 2c3. (a) Show that the solution of the initial value problem y' = 1 + l. 8. =x 1 2 17 + x 3 + . For example.. (c) Conclude that yC3l = y. Indeed. For instance. its Taylor series is of the form =0 y(x) = Y =X + c3 x 3 + c5 x 5 + c7 x 7 + · · · . show that f (x) satisfies the initial value problem where {F~~}~0 is the sequence 0.+ + .1 can be applied to linear equations of any order (as well as to certain nonlinear equations). We could evaluate this series if we could obtain a formula for the function 12131415 f(x) = 1 + x. consider the equation xy" + y' +xy = 0.
• Theorem 2 of Section 2. A proof of the following theorem can be found in Chapter 3 of Coddington. 1 implies that Eq. It follows that. If A(x). (1) with analytic coefficients. these solutions will be power series in powers of x .) • ········ Example 3 The point x = 0 is an ordinary point of the equation because the coefficient functions A(x ). Thus the only singular point of Eqs. x = a is a singular point. Example 1 The point x = 0 is an ordinary point of the equation xy" + (sinx)y' + x 2y = 0. (2) has two linearly independent solutions on any open interval where the coefficient functions P(x) and Q(x) are continuous. . and C(x) are polynomials with no common factors. (Theorem 1 of Section 3.: Prentice Hall. (2)and of the equivalent Eq. Recall that a quotient of analytic functions is analytic wherever the denominator is nonzero. then x = a is an ordinary point.a.1 implies that an analytic function must be differentiable. and C(x) are polynomials with A(O) ::1 0. N. An Introduction to Ordinary Diff erential Equations (Englewood Cliffs. • Example 2 The point x = 0 is not an ordinary point of the equation For while P(x ) = x 2 is analytic at the origin. B(x) . Otherwise. 1961). if A(a) ::1 0 in Eq. despite the fact that A (x) = x vanishes at x = 0. The basic fact for our present purpose is that near an ordinary point a. then x = a is an ordinary point if and only if A(a) ::1 0. But in the form of (2) it is the equation 1 y" + . B(x). The point x = a is called an ordinary point of Eq. (3) and (4) is x = 0.y' X + y =0 (4) with P (x) = 1I x not analytic at x = 0. (1 )provided that the functions P (x) and Q (x) are both analytic at x = a.J. Q(x ) = x 112 is not.208 Chapter 3 Power Series Methods The coefficient functions in (3) are continuous everywhere. The reason is that is nevertheless analytic at x = 0 because the division by x yields a convergent power series. The reason is that Q(x) is not differentiable at x = 0 and hence is not analytic there.
a)n. Solution This example illustrates the fact that we must take into account complex singular points as well as real ones. The distance (in the complex plane) of each of these from 0 is 3. n=O (5) The radius of convergence of any such series solution is at least as large as the distance from a to the nearest (real or complex) singular point of Eq. The distance of each singular point from 4 is 5. Radius of convergence as distance to nearest singularity. . The coefficients in the series in (5) can be determined by its substitution in Eq. Then Eq.2. so a (x . so a series solution of the form L cnxn has radius of convergence at least 3. (1) that is. (1) has two linearly independent solutions.3.4. 3.4t has radius of convergence at least 5 (see series solution of the form L en Fig. Example 4 Determine the radius of convergence guaranteed by Theorem 1 of a series solution of (6) in powers of x.2 Series Solutions Near Ordinary Points 209 THEOREM 1 Solutions Near an Ordinary Point Suppose that a is an ordinary point of the equation A(x)y" + B(x)y' + C(x)y = 0. (6) are ±3i. (1). • y FIGURE 3. Repeat for a series in powers of x .2.1).1. the functions P = B/A and Q = CjA are analytic at x =a. (1). each of the form 00 y(x) = Lcn(X. Because X P(x) = x2 +9 and the only singular points of Eq.
so the series we get will have radius of convergence at least 2..4 L(n n=O n=O + 2)(n + 1)cn+2Xn + 3 L n=O ncnXn +L n=O CnXn = 0. 2.210 Chapter 3 Power Series Methods Example 5 Find the general solution in powers of x of (x 2  4)y" + 3xy' + y = 0. 4 4 2 . (7) Then find the particular solution with y(O) = 4. With n = 0. 4 .) Substitution of 00 00 00 y~CnX.1)cnxn. (See Problem 35 for the exact radius of convergence. We can begin the first and third summations at n = 0 as well. because no nonzero terms are thereby introduced. 2 . (7) are ±2. 4' and C6 5c4 3 · 5co = = . 6 Continuing in this fashion.1)cnxn .= . We shift the index of summation in the second sum by +2. which leads to the recurrence relation c for n ~ n+2   + (8) 0. Solution The only singular points of Eq.. After collecting coefficients of Cn and Cn+ 2. The identity principle yields (n + 1)2 cn  4(n + 2)(n + 1)cn+2 = (n 1)cn 4(n + 2) 0. This gives 00 00 00 00 L n(n. we evidently would find that C2n = 1 · 3 · 5 · · · (2n . replacing n with n + 2 and using the initial value n = 0.4 L n(n . '"""' n n=O in Eq. y' (0) = 1. 4n .6 43 .4.2 + 3 L ncnXn n=2 n=2 n=l +L n=O CnXn = 0. and 4 in turn. (7) yields 00 y1 = '"""' ~nCnX n 1 . 4 . we obtain L [(n n=O 00 2 + 2n + 1)cn.2.2n· n! . 2 · 4 · · · (2n) With the common notation (2n (2n + 1)! + 1)!! = 1 · 3 · 5 · · · (2n + 1) = .1)cnxn .4(n + 2)(n + 1)cn+2] xn = 0.1) co. we get 3c2 3co c4 = .2 n=l 00 00 00 L n(n . and y" = L n=2 n(n .l)cnxn.
n! Co.1)!! 23n . In Example 5. 5 ' and C7 =. ) 1024 (11') 1 3 1 1 ) +c1 ( x+x +x 5 + . 5 .x7 +·· · . 3. c3 . which • expresses the general solution in the form y = coyo + Ct Yt · .= 43 . 4. Using these values in Eq. we finally obtain C2n = (2n . 2n = 23n.2 Series Solutions Near Ordinary Points and the observation that 2 · 4 · 6 · · · (2n) = 2n · n!. the first few terms of the particular solution satisfying y(O) = 4 andy' (0) = 1 are y(x ) = 4 + x 121 3 34 1s +2 x + 6x + 32 x + 30 x + · · · . Yo(x) and y 1(x) are defined by the two series that appear on the righthand side in Eq. and let y 1(x ) be the solution obtained with c0 = 0 and c 1 = 1. 211 (9) (We also used the fact that 4n . 12 + 3x4 y(x) =co ( 1 + x 8 128 5 +. 5 42 • 3. c0 and c 1• In this event two linearly independent solutions are obtained as follows. and 5 in Eq. 2n · (2n + 1)!! (10) The formula in (9) gives the coefficients of even subscript in terms of c0 .x 6 + · . the formula in (1 0) gives the coefficients of odd subscript in terms of c 1 • After we separately collect the terms of the series of even and odd degree. Let y 0 (x) be the solution obtained with c0 = 1 and c 1 = 0. 7 . (12) • Remark: As in Example 5. 6 30 140 Because y(O) = c0 and y'(O) = c 1. (11) 2n . y~(O) = 1. (2n + 1)!! Alternatively. Then Y o(O) = 1.3.) With n = 1. (11). n! + CJ (x + E n= l n! x2n+l) . Yb(O) =0 and Yt(O) = 0.= 2 · 4ct 4... 3 . c4 . we get the general solution y(x) =co (1 + E(2~n= l 1)!! x2n) 2 n . so it is clear that Yo and y 1 are linearly independent. (8).7 6cs 2 · 4 · 6c1 It is apparent that the pattern is C2n+l = 2 · 4 · 6 · · · (2n) CJ 4n · 1 · 3 · 5 · · · (2n + 1) = n! CJ. ••• in terms of the first two. the given initial conditions imply that c0 = 4 and c 1 = 1. substitution of y = L CnXn in a linear second order equation with x = 0 an ordinary point typically leads to a recurrence relation that can be used to express each of the successive coefficients c2 . (11'). we get 4c3 cs = .
. dy dx and d 2y dt 2  [!!_ (dy)] dx dx dx dt  !!_ I _ II dx (y ) .3) dt2 + 3(t. we would have needed the general solution in the form 00 y(x) = LCn(X at. For only with a solution of the form in (13) is it true that the initial conditions y(a) =co and y 1 (a) = c1 determine the arbitrary constants c0 and c 1 in terms of the initial values of y and y 1 • Consequently.1) 2 + 6(t. Example 6 Solve the initial value problem 2 d2y (t . . so that we wind up looking for a series of the form L CnXn after all. We substitute t .1) dt + y = dy 0. it simplifies the computations if we first make the substitution x = t . y 1 (1) = 1.Y ' where primes denote differentiation with respect to x. (14) into one with the new independent variable x. Hence we transform Eq.a rather than in powers of x.212 Chapter 3 Power Series Methods Translated Series Solutions If in Example 5 we had sought a particular solution with given initial values y(a) and y 1 (a). we note that t2  2t .. To transform Eq. we need a series expansion of the general solution centered at the point where the initial conditions are specified.1 for x in Eq. ===y .4)y" + 3xy 1 + y = 0 with initial conditions y = 4 and y 1 = 1 at x = 0 (corresponding tot = 1). y(l) = 4.1)n. (14) into (x 2 .3 = (x dy dt + 1)2 dy dx dx dt 2(x + 1) 1 3 = x2  4. But instead of substituting this series in (14) to determine the coefficients. so the particular solution in (12) is available.1) 3 1) 5 1 + 32 (t  3 1) 4 + 30 (t  1 + . in powers of x . to solve an initial value problem. (12) and thereby obtain the desired particular solution 1 y(t) = 4 + (t.1.1) + 2(t. .2t. (13) n=O that is. (14) Solution We need a general solution of the form Len (t. This is the initial value problem we solved in Example 5.
A manyterm recurrence relation expresses each coefficient in the series in terms of two or more preceding coefficients.2 = ncn (n + 2)(n + 1) (16) . (Why?) A series such as this can be used to estimate numerical values of the solution.LnCnXn .8) ~ 3. n=2 n= l n=O We can start the second sum at n = 0 without changing anything else.8188.. so we must separate the terms corresponding to n = 0 and n = 1 in the first two sums before collecting coefficients of xn. • The last computation in Example 6 illustrates the fact that series solutions of differential equations are useful not only for establishing general properties of a solution.2 . Example 7 Find two linearly independent solutions of y If . n=2 n=O The common range of these three summations is n . that c3 = ic 1 . The identity principle now implies that 2c2 = 0. and the threeterm recurrence relation Cn+2 +Cn .2] xn = 0.. it is generally inconvenient or even impossible to find a formula that gives the typical coefficient en in terms of n.2) 3 1 1 + 332 ( 0. (15) Solution We make the usual substitution of the power series y = equation 00 00 00 L cnxn. (8) is an example of a twoterm recurrence relation.2).Cn. These shifts yield 00 00 00 L(n n=O + 2)(n + 1)cn+2Xn  L ncnxn .2)4 + 310 ( 0.2) 2 + 6(0.2Xn = 0. The next example shows what we sometimes can do with a threeterm recurrence relation. we shift the index of summation in the first sum by +2 (replace n with n + 2). 2. it expresses each coefficient in the series in terms of one of the preceding coefficients. but also for numerical computations when an expression of the solution in terms of elementary functions is unavailable. For instance. To make each term include xn in its general term.1)CnXn. Types of Recurrence Relations The formula in Eq. ' so that y(0.2) + 2(0. In the case of a manyterm recurrence relation.2 Series Solutions Near Ordinary Points 213 This series converges if 1 < t < 3.2)5 + . This results in the Ln(n.ncn.c1x + L n=2 [ (n + 2)(n + 1)cn+2..3.8) = 4 + ( 0. This gives 00 2c2 + 6c3x .L Cn..x 2 y = 0 . and we shift it by 2 in the third sum (replace n with n . y(0.LCnXn+ 2 = 0.xy I .
To get our first solution y 1 of Eq. so that c2 = 0 and c 3 = ~. The only singular points of the Legendre equation are at + 1 and 1. • The Legendre Equation The Legendre equation of order a is the secondorder linear differential equation (1 . cs = C7 3c3 + c1 20 5cs + c3 = 42 (17) Thus all values of Cn for n ~ 4 are given in terms of the arbitrary constants c0 and = 0 and c3 = ~cr. Then the formulas in ( 17) yield 3 cs = 40' so that C6 = 0.x + ··· . (20) where the real number a satisfies the inequality a > . (15). so it has two linearly independent solutions that can be expressed as power series in . so that c2 = c3 = 0. we choose c0 = 1 and c 1 = 0. we take co = 0 and c 1 = 1.x + . A general solution of Eq. it is clear from Eq.2xy' + a(a + 1)y = 0..1. (15).x + . (16) that this series contains only terms of even degree.x 2 )y". This differential equation has extensive applications. 12 90 1120 (18) Because c 1 = c 3 = 0.1120' 3 thus Yr (x) = 1 1 4 1 6 3 8 +x + .· . Then the formulas in (17) yield cr because c2 C4 = 12 . The solutions y 1(x) and y 2 (x) are linearly independent because YI (0) = 1 and y~ (0) = 0. In particular. C6 = 1 90 . 6 40 1008 (19) Because co = c2 = 0. 1 Cs = 0. C7 = 0. it is clear from Eq.x + ··· . so the power series representing y 1 (x) and y 2 (x) converge fur~lx.. whereas y 2 (0) = 0 and y~ (0) = 1.. 13 C? = 1008' Y2(x) = x 1 3 3 5 13 7 + x + . (16) that this series contains only terms of odd degree. c8 . (15) is a linear combination of the power series in (18) and (19).. Equation (15) has no singular points.214 Chapter 3 Power Series Methods for n ~ 2. To obtain a second linearly independent solution y2 ofEq. ranging from numerical integration formulas (such as Gaussian quadrature) to the problem of determining the steadystate temperature within a solid spherical ball when the temperatures at points of its boundary are known.
If a = n is even. The substitution y Eq. we see from Eq.2 Series Solutions Near Ordinary Points 215 powers of x with radius of convergence at least 1. y 1 (x ) is a polynomial of degree n and y 2 is a (nonterminating) infinite series. c5 = (a.4)· · ·(a.3) · · · (a . In terms of the arbitrary constants c0 and c 1.1)(a + 2) 3! 4! CJ. (23) that a 2m+ l = 0 when 2m + 1 > n . y 2(x) is a polynomial of degree n and y 1 is a (nonterminating) infinite series. one of the two solutions in (24) is a polynomial and the other is a nonterminating series. = a(a. If a = n is an odd positive integer. Now suppose that a = n.2)(a.1)(a. (20) leads (see Problem 31) to the recurrence relation (a.m)(a + m + 1) (m + 1)(m + 2) Cm = L cmxm in (21) Cm+2 = form ~ 0. Thus in either case. Eq.3)(a + 2)(a + 4) 5! c 1• We can show without much trouble that form > 0. In this case. we get two linearly independent power series solutions 00 Yt (x) =CoL( l)ma2mX 2m m =O and Y2(x) = Ct I:<1)ma2m+tX2m+l m =O 00 (24) of Legendre's equation of order a. C2m = ( 1) ma(a. In this case.2m + 1) (a + 2) (a ( 2m+ 1)! + 4) · · · (a + 2m) C I· (23) Alternatively.2)(a + 1)(a + 3) Co.1) (2m)! Co (22) and C  2m+l (1)m (a . We are using m as the index of summation because we have another role for n to play. a nonnegative integer. respectively. where a2m and a 2m+ I denote the fractions in Eqs. 2! C3=C4 (a. With this notation.3. (21) yields C2 =  a(a + 1) Co. .2m+ 2)(a + 1)(a + 3)· ··(a+ 2m. (22) that a 2m = 0 when 2m > n.1) (a . we see from Eq. (22) and (23).
12. y FIGURE 3.x 2 )y".x 2 )y". 11. 4. 1. (25) is denoted by Pn (x) and is called the Legendre polynomial of degree n.x 2 y' . 10. (x 2 . y" + xy' + y = 0 4. 14.4y = 0 Sy" .  P3(x) = 1 3 2csx  3x). Ps(x) = S(63x 5 1  70x 3 + 15x). The first six Legendre polynomials are Po(x)=1. the integral part of nj2. 2.2. (x 2 + 2)y" + 4xy' + 2y = 0 3. 3. (1 .2. and their graphs are shown in Fig. 3. 13. Graphs y = Pn (x ) of the Legendre polynomials for n = 1. and 5. II!J Problems 7. 8. 9.7xy' + 16y = 0 (2.3xy = 0 y" + x 2 y' + 2xy = 0 y" + xy = 0 (an Airy equation) y" +x 2 y = 0 (x 2 .1)y" .1)y" + 8xy' + 12y = 0 3y" +xy' .2k)! ' ( 1)k(2n (26) where N = [n/2].1).2xy' + lOy = 0 y".2. State the recurrence relation and the guaranteed radius of convergence in each case. P4(x) = S(35x 4 30x 2 + 3). 15. P2(x) = 2(3x 1 1 2 .2xy' + n(n + 1)y = 0. The graphs are distinguished by the fact that all n zeros of Pn (x) lie in the interval 1 < x < 1.3)y" + 2xy' = 0 6.2k X 2nk! (n.6xy' + 12y = 0 + 3)y".216 Chapter 3 Power Series Methods With an appropriate choice (made separately for each n) of the arbitrary constants c0 (n even) or c 1 (n odd). It then turns out that p X  n( )  L N k=O 2k)! n. (x 2 . the nthdegree polynomial solution of Legendre's equation of order n. (x 2 + l) y" + 6xy' + 4 y = 0 5.xy' + 16y = 0 (x 2 .2. It is customary (for a reason indicated in Problem 32) to choose the arbitrary constant so that the coefficient of xn in Pn(x) is (2n)!/ [2n(n!) 2 ]. Find general solutions in powers of x of the differential equations in Problems 1 through 15.k)! (n. (x 2 .l)y" + 4xy' + 2y = 0 2.
x 2 )v' + 2nxv = 0.6(x.2xy' + 2ay 18. 27. y'(O) = 1. H 2 (x) = 4x 2 .48x 2 + H5 (x) = 32x 5  Ht(X) = 2x. y(l) = 0.1)xv' + 2nv = 0. y(O) = 0. y(O) = 1. It is interesting to use a computer algebra system to investigate the conjecture that (for each n) the zeros of the Hermite polynomials Hn and Hn+l are "interlaced". State the interval of values of x for which Theorem 1 of this section guarantees convergence. The discussion following Example 4 in Section 3. 26.l)y" + 2xy' + 2xy = 0 y" +x 2 y' + x 2 y = 0 (1 + x 3 )y" + x 4 y = 0 Solve the initial value problem y" +xy' + (2x 2 + l)y (2m+ 1)! = 0.4y = 0. find the first three nonzero terms in each of two linearly independent solutions of the form y = L CnXn. The Hermite equation of order a is y".4(x . 160x 3 + 120x. 32.l)y'. the n zeros of Hn lie in the n bounded open intervals whose endpoints are successive pairs of zeros of Hn+l· 34.2. m= l 23. y(l) = 2.1 suggests that the differential equation y" + y = 0 could be used to introduce and define the familiar sine and cosine functions.x 2 )v" + 2(n . Solve the initial value problems in Problems 18 through 22.( x . First make a substitution of the form t = x . y(3) = 0. y'(1) = 0 19.a. Then find the first three nonzero terms in each of two linearly independent solutions._._. (Note that the coefficient of xn in Pn(x) is (2n)!/ [2n(n!) 2 ]. (2x. y'(3) = 0 21.that is. multiplied by a suitable constant so that the coefficient of xn is 2n. y" + exy = 0 (cosx)y" + y = 0 xy" + (sinx)y' + xy = 0 Derive the recurrence relation in (21) for the Legendre equation.2 Series Solutions Near Ordinary Points Use power series to solve the initial value problems in Problems 16 and 17. find a threeterm recurrence relation for solutions of the form y = L CnXn. y'( 2) = 0 22. A general formula for the Hermite polynomials is for the nthdegree Legendre polynomial.. y(O) = 1.2m+ 1) + L. in u is (2n)!/n!. 12. H4(x) = 16x4 . (x 2 +6x)y"+(3x+9)y'3y = 0. y'(O) = 0 = 1 2xv<n+l) + n(n + l)v(n) = 0. Verify that this formula does in fact give an nthdegree polynomial. (a) Show that v = (x 2 satisfies the differential equation on (1 . then state why this proves Rodrigues' formula. y" + (x . 217 (b) Differentiate each side of the last equation n times in succession to obtain (1 . the Airy equation y" = xy . (4x 2 + 16x + 17)y" = 8y.( 1)m m=l 2) · · ·(a 2m+ 2) X2m (2m)! and Y2 =X ~ 2m(a. Determine sufficiently many terms to compute y(l/2) accurate to four decimal places. Thus u = v<n) = Dn(x 2 . H3(x) = 8x 3  12x. (x 2 . y" + xy'.1) n! 2n dxn Show that y 1 is a polynomial if a is an even integer. Follow the steps outlined in this problem to establish Rodrigues's formula 28.) 33.x 2 )v<n+2)  16. 31. Differentiate each side of this equation to obtain (1 . y" + (1 + X) y = 0 (x 2 . then find a solution L Cntn of the transformed differential equation.1)(a. Substitute known Taylor series for the analytic functions and retain enough terms to compute the necessary coefficients. (1 + x 2 )y" + 2xy'. 25. 24.3) ··· (a..3. (a) Derive the two power series solutions Yt = 1 ~ 2ma(a + L. 30. 29.l)y' + y = 0.(1)m x2m+l.3)y' + 6y = 0. It is the nthdegree polynomial solution of Hermite's equation. (b) The Hermite polynomial of degree n is denoted by Hn(x). y( 2) = 1. y'(l) = 1 20. In a similar fashion.6x + lO)y" .1Y satisfies Legendre's equa(c) Show that the coefficient of xn tion of order n. y(3) = 2. = 0. y'(O) 17. y'(3) = 2 In Problems 23 through 26. 1 dn 2 n Pn(X) = . Show that the first six Hermite polynomials are Ho(x) = 1.2y = 0.x 2 )y". whereas y2 is a polynomial if a is an odd integer.2y = 0. In Problems 28 through 30.
y. y~(O) = 1.3... B l(X) . write the series of terms of even degree in Eq. has the two singular points x = 1 and x = 1. B. Then verify that your results agree with the formulas Yl (x) = 1 + f=( ~ 1 . whereas the Legendre equation of order n..3 exhibit trigonometriclike oscillatory behavior for x < 0.3 116r(~) o Y1 (x) + Y2(x) 3 lt6r(~) define the standard Airy functions that appear in mathematical tables and computer algebra systems. respectively..1) (3k + 1)! x3k+l FIGURE 3. Their graphs shown in Fig. For instance.. and C are polynomials having no common factors.218 Chapter 3 Power Series Methods y serves to introduce two new special functions that appear in applications ranging from radio waves to molecular vibrations.2xy' + n(n + l)y = 0. .32f3r(~) Yz(x) 31 /3 r(~) Yo(x) = 1 + LCznX 2n = 1 + Lanzn n=l n=l and . (3k . Y1 (x) A 1(x)o.. where an = c2n and z = x2 • Then apply the recurrence relation in Eq. The Airy function graphs y = Ai(x) andy= Bi(x). k=l ~ 2 .. Recall that if the functions A.. (1 . It turns out that some of the features of the solutions of such equations of the most importance for applications are largely determined by their behavior near their singular points. (1) are simply those points where A (x) = 0. (11) in the form to show similarly that its radius of convergence (as a power series in x) is also 2. 3. whereas Ai (x) decreases exponentially and Bi(x) increases exponentially as x + +oo. Derive the first three or four terms of two different power series solutions of the Airy equation..2) (3k)! x3k and Y2(x) = x + L. x = 0 is the only singular point of the Bessel equation of order n..1 to show that the radius of convergence of the series in z is 4. (0) = 0 and y 2 (0) = 0.2.3 (which is based on highprecision approximations to the Airy functions). (3k . How does this corroborate Theorem 1 in this section? (b) Write the series of terms of odd degree in Eq. (8) and Theorem 3 in Section 3.2.x 2 )y" . Hence the radius of convergence of the series in x is 2. It is interesting to use a computer algebra system to investigate how many terms must be retained in the y 1 . then the singular points of Eq. 35.n 2 )y = 0. (11) in the form 00 00 for the solutions that satisfy the initial conditions y 1 (0) = 1.and y2 series above to produce a figure that is visually indistinguishable from Fig. 5 . 3.2. 4 . IEIJ Regular Singular Points We now investigate the solution of the homogeneous secondorder linear equation A(x) y" + B(x)y' + C(x)y = 0 (1) near a singular point. x 2 y" + xy' + (x 2 . (a) To determine the radius of convergence of the series solution in Example 5. The special combinations .
Now it can be proved that each of the functions P(x) and Q(x) either is analytic or approaches ±oo as x + 0. This is a way of saying that P(x) and Q(x) have only "weak" singularities at x = 0. This new equation has the singular point t = 0 corresponding to x = 1 in the original equation. we assume that Eq. A differential equation having x = a as a singular point is easily transformed by the substitution t = x a into one having a corresponding singular point at 0.t 2 . x = 0 is a singular point of Eq. (1). To investigate the form that a solution of such an equation might take. we get the equation 2(t t(t + 2dt )2  d2y + 1)+ n(n + 1)y = dt dy 0. x2 (3) where p(x) = x P(x) and q(x) = x 2 Q(x). we rewrite Eq. as x + 0. if we rewrite the Bessel equation of order n in the form y II 1 + + y x I ( n) y = 0. Consequently.2 fails in this case. Types of Singular Points A differential equation having a singular point at 0 ordinarily will not have power series solutions of the form y(x) = L cnxn.(t + 1)2 = 2t. To state this more precisely. (2) where P = BjA and Q = CjA. if P (x) and Q(x) have convergent power series expansions in powers of x on some open interval containing x = 0. let us substitute t = x. Recall that x = 0 is an ordinary point (as opposed to a singular point) of Eq. (2) if the functions P(x) and Q(x) are analytic at x = 0. (1) has analytic coefficient functions and rewrite it in the standard form y'' + P(x)y' + Q(x)y = 0. 1x2 2 we see that P(x) = ljx and Q(x) = 1 . (4) . and Q(x) no more rapidly than 1jx 2 .1 into the Legendre equation of order n. For example. Because and 1 .x 2 = 1 . so the straightforward method of Section 3.(njx) 2 both approach infinity as x + 0. that is.3. (2). (2) in the form y" + p(x) y' x + q(x) Y = O. it has also the singular point t = 2 corresponding to x = 1.3 Regular Singular Points 219 We will restrict our attention to the case in which x = 0 is a singular point of Eq. provided that P (x) approaches infinity no more rapidly than 1/x. We will see presently that the power series method can be generalized to apply near the singular point x = 0 of Eq. (2) provided that either P(x) or Q(x) (or both) approaches ±oo as x + 0. For instance.
Because a quotient of polynomials is analytic wherever the denominator is nonzero.and Q(x) = x 2(1 + x ) x (l + x) both approach oo as x ~ 0. For instance.. consider the equation 2x 3 y" = + (1 + x)y + 3xy = 1 0. Thus p(x) = 1+ x 4 . We will not discuss the solution of differential equations near irregular singular points.~ as x ~ 0 (although q (x) ~ is a polynomial)..220 Chapter 3 Power Series Methods DEFINITION Regular Singular Point The singular point x = 0 of Eq. Hence x = 0 is a regular singular point of the given differential equation.. we see that x = 0 is an irregular singular point. . (3) is a regular singular point if the functions p(x) and q(x) are both analytic at x = 0. which has the singular point x = 0.x2 x(l + x )y 1 + 2 + 3x x 2(1 +x) y = O. .n 2 are both polynomials in x. the singular point x = 0 is a regular singular point if p(x) and q (x) are both polynomials. Because 1 +X 1 p(x) = . 0 """"' N N ·· Example 1 Consider the differential equation + x)y" + x (4.= 2x 2 2x 2 +....x 2)y + (2 + 3x)y = In the standard form y" + Py + Q y = 0 it is x 2 (1 1 1 0._N ' "' WW .~oOoo o = 1 2x oo N W .. Because 4 x 2 2+3x P(x). .. If we write this equation in the form of (3). • ..W.. we get (1 + x)/(2x 2) 1 ~ 11 0 y + X y +2 X Y= .._. Otherwise it is an irregular singular point. we see that x = 0 is a regular singular point of Bessel's equation of order n by writing that equation in the form 1 x2 y " +yl +  n2 x x2 y=O..x 2)j(l X + x) y 1 + (2 + 3x ) j (l 2 X +x) y = 0.x2 and q(x) = 2+3x 1+x .. " y + 4 . we see that x = 0 is a singular point. (3): y 11 + (4. this is a considerably more advanced topic than the solution of differential equations near regular singular points. By contrast. we see that p(x ) and q(x) are both analytic at x = 0. To determine the nature of this singular point we write the differential equation in the form ofEq. . noting that p(x) 1 and q(x) = x 2 . In particular..
. Alternatively. (1) and rewrite it in the form in (3). so the singular point x = 0 is regular.· H  H OH o '<'H oO. 0 = (6) If Po x 2 y 11 + xp(x)y' + q(x)y = = 0 = q 0 .. • If either limit fails to exist or is infinite. then x 0 may be an ordinary point of the differential equation 0 in (3).. Then l'H6pital's rule gives the values Po = xlim .. . we see that x = 0 is not an ordinary point.0 1 lim sin x cosx 1 1.• • H•ON H H Example 2 To investigate the nature of the point x = 0 for the differential equation x 4 y 11 + (x 2 sinx)y + (1.HO. we could write qo and = .= x + 0 x2 x + 0 2x 2 for the limits in (5) and (6).3..= lim . H<U• N " H   000 . ••~ · H"Ooo R..3 Regular Singular Points 221 It may happen that when we begin with a differential equation in the general form in Eq. "''''" '"N H H ...= + 0 X x . Since they are not both zero...·· ~~HoO • "·~Ooo h>OO<"" N H  . In this case p 0 = p (0) and qo = q(O) are simply the constant terms of these polynomials.0 (5) and qo = q{O) = x+0 lim q(x) = lim x 2 Q(x).. In this case the situation is determined by the limits Po= p(O) = lim p(x) = lim x P(x) x. .. Otherwise: • If both the limits in (5) and (6) exist and are finite. Remark: The most common case in applications.cosx sinx 1 = lim .. the functions p(x) and q(x) as given in (4) are indeterminate forms at x = 0.cosx)jx 2 X 2 y = 0. x x2 (3) is that the functions p (x) and q (x) are polynomials. Oo.cos x)y = 1 0.. But both limits are finite. then x = 0 is a regular singular point. for the differential equation written in the form y II+ p(x) I+ q(x) 0 y y= .. then x = 0 is an irregular singular point.0 x. (5) and (6). we first write it in the form in (3): y 11 + (sinx)/x X y 1 + (1. x . so there is no need • to evaluate the limits in Eqs.
(9) the product of xr and a power series. in which p(x) and q(x) are power series rather than constants.x2 x2 2! x .x6 ++ · ·· ) ] 4 4! 6! These (convergent) power series show explicitly that p(x) and q(x) are analytic and moreover that p 0 = p(O) = 1 and q0 = q (0) = ~.222 Chapter 3 Power Series Methods and 2 x q(x) = 1cosx = .1 [ 1 . An infinite series of the form in (9) is called a Frobenius series. This turns out to be a very fruitful conjecture. To investigate the possible existence of Frobenius series solutions. named for the German mathematician Georg Frobenius ( 18481917). In this case we can verify by direct substitution that the simple power function y(x) = xr is a solution of Eq. it is a reasonable conjecture that our differential equation might have a solution of the form y(x) = xr L n=O 00 CnXn =L n=O 00 CnXn+r = coxr + C!Xr+l + C2Xr+2 +.( 1 . + qo = (8) In the general case.. For instance. • The Method of Frobenius We now approach the task of actually finding solutions of a secondorder linear differential equation near the regular singular point x = 0. according to Theorem 1 (soon to be stated formally). indeed. thereby verifying directly that x = 0 is a regular singular point.~ the series in (9) takes the form it is not a series in integral powers of x. (3) reduces when p(x) p 0 and q(x) q0 are constants. Note that a Frobenius series is generally not a power series. have at least one such solution. with r = . (7) if and only if r is a root of the quadratic equation r(r.. every equation of the form in (1) having x = 0 as a regular singular point does.1) + por = = 0. This fact is the basis for the method of Frobenius. we begin with the equation x 2 y" + xp(x)y' + q(x)y = 0 (10) . who discovered the method in the 1870s. The simplest such equation is the constantcoefficient equidimensional equation x 2 y II + poxy'+ qoy = 0 (7) to which Eq.
(11) Suppose that Eq. (16) shows that if the Frobenius series y = xr L cnxn is to be a solution of the differential equation in ( 10). then the exponent r must be one of the roots r 1 and r2 of the indicia! equation in ( 16). so it follows that r must satisfy the quadratic equation r(r. it follows that there are two possible Frobenius series solutions. r2.3 Regular Singular Points 223 obtained by multiplying the equation in (3) by x 2 • If x = 0 is a regular singular point. (15) is co[r(r 1) + por +q0 ]xr. (10) now yields [r(r . n=O (12) We may (and always do) assume that c0 ::f. then the coefficient of this term (as well as those of the higherdegree terms) must vanish. But we are assuming that c0 ::f. (12) leads to 00 y' = L Cn(n n=O + r)xn+r. The exponents r 1 and r2 in the possible Frobenius series solutions are determined (using the indicia! equation) by the values Po = p(O) and qo = q(O) that we have discussed.1) + por + qo = 0 (16) of precisely the same form as that obtained with the equidimensional equation in (7).1)coxr + (r + 1)rcl x r+I + · · ·] +[pox+ PIX 2 + · · ·] · [rcoxr. Equation (16) is called the indicia! equation of the differential equation in (10). Our derivation ofEq.3. particularly when the coefficients in the differential equation . we see that the lowestdegree term in Eq. and its two roots (possibly equal) are the exponents of the differential equation (at the regular singular point x = 0). 0 because the series must have a first nonzero term. Termwise differentiation in Eq. IfEq.1 (13) and 00 y" = L Cn(n n=O + r)(n + r l)xn+r. q(x) = qo + q1x + q2x 2 + q3x 3 + · · · . whereas if r 1 = r2 there is only one possible Frobenius series solution. (14) Substitution of the series in Eqs.I + (r + 1)cix r + · · ·] 0. (10) has the Frobenius series solution 00 Y = LCnXn+r.2. (11) through (14) in Eq. 0. then p(x) and q(x) are analytic at x = 0. so p(x) = Po+ PIX+ P2x 2 + p3x 3 + · · · . If r 1 ::f. (15) + [qo + q1x + · · ·] · [coxr + c1xr+I + · · ·] = Upon multiplying initial terms of the two products on the lefthand side here and then collecting coefficients of x r. the second solution cannot be a Frobenius series. In practice. (15) is to be satisfied identically.
(12) through (14) in the differential equation.224 Chapter 3 Power Series Methods in the original form in (I) are polynomials. with roots r 1 = ~ and r 2 = 1.~) = 0. .~ . we need only replace x r 1 with lx lr1 to obtain a solution for x < 0. We also will seek solutions only for x > 0. essentially the same method as was used to determine coefficients in power series solutions in Section 3. If the exponents r 1 and r 2 are complex conjugates. Hence the indicia] equation is r(r . n=O 00 • Frobenius Series Solutions Once the exponents r 1 and r2 are known. We will restrict our attention here to the case in which r 1 and r2 are both real.x) 2 x2 y = 0. The two possible Frobenius series solutions are then of the forms Y1 (x) = x 112 L anxn n=O 00 and Y2(x) = x  1 L bnxn.2.~ = (r + l)(r. and thus see that Po = ~ and qo = .1) + ~r. then there always exist two linearly independent Frobenius series solutions. the simplest way of finding p 0 and q0 is often to write the equation in the form Then inspection of the series that appear in the two numerators reveals the constants Po and qo. Once such a solution has been found. Example 3 Find the exponents in the possible Frobenius series solutions of the equation Solution We divide each term by 2x 2 (1 + x) to recast the differential equation in the form y' y" + 1(1 2 + 2x + x 2 ) x + l(l . The following theorem is proved in Chapter 4 of Coddington's An Introduction to Ordinary Differential Equations. the coefficients in a Frobenius series solution are determined by substitution of the series in Eqs.~ = r 2 + ~r.
then there exists a second linearly independent solution for x > 0 of the form yz(x) = xr2 L bnxn n=O 00 (bo =1.r 2 is a positive integer. Example 4 Find the Frobenius series solutions of 2x 2 y" + 3xy' 3 (x 2 + 1)y = 0. and that p 0 = ~ and q0 = . It turns out that. (10) Let p > 0 denote the minimum of the radii of convergence of the power series 00 00 p(x) = LPnXn n=O and ~ q(x) = Lqnxn .4. then there can exist only one Frobenius series solution. of the indicia! equation r(r.1) + (a) For x > 0.r 2 is neither zero nor a positive integer. Examples 4 through 6 illustrate the process of determining the coefficients in those Frobenius series solutions that are guaranteed by Theorem 1. the Frobenius series we = ! . (18) and (19) are each at least p. The coefficients in these series can be determined by substituting the series in the differential equation x 2 y" + xp(x)y' + q(x)y = 0. These exceptional cases are discussed in Section 3. (1 0) of the form 00 YI(X) = Xq LanXn (ao =f. n=O Let r 1 and r2 be the (real) roots.3. with r 1 por + qo = 0. there exists a solution of Eq. Then r2 . (20) Solution First we divide each term by 2x 2 to put the equation in the form in (17): y" + ly' + x _l _ lx2 2 x2 2 y = 0. if r 1 . Because p (x ) ~ and q(x ) = ~x 2 are polynomials.~ .0) (19) corresponding to the smaller root r 2 • The radii of convergence of the power series in Eqs.3 Regular Singular Points 225 THEOREM 1 Frobenius Series Solutions Suppose that x = 0 is a regular singular point of the equation x 2 y" + xp(x)y' + q(x)y = 0. ( 19) corresponding to the smaller root r2 . (21) We now see that x = 0 is a regular singular point. 0) (18) n=O corresponding to the larger root r 1• (b) If r 1 . We have already seen that if r 1 = r2 . there may or may not exist a second Frobenius series solution of the form in Eq.
1)cnxn+r + 3 L(n + r)CnXn+r n=O 00 00 . whereas with r 2 = 1 it becomes a recurrence relation for the series for y 2 • When we substitute 00 00 Y = LCnXn+r. In this example. n=O y' = L(n n=O + r)cnxn+rl' and 00 y" = L)n + r)(n + rn=O 1)cnxn+r2 in Eq.~ ) co = 0. A good standard practice is to shift indices so that each exponent will be the same as the smallest one present. rather than Eq. Following our standard practice.226 Chapter 3 Power Series Methods obtain will converge for all x > 0. This gives 2 I)n n=O 00 00 + r)(n + r. n=O n=O (22) At this stage there are several ways to proceed.L CnXn+r+2 .L Cn Xn+r = 0. so we must treat n = 0 and n = 1 separately. They do not differ by an integer. n=2 n=O (23) The common range of summation is n ~ 2. (20). so Theorem 1 guarantees the existence of two linearly independent Frobenius series solutions.L CnXn+r = 0. We will then get a recurrence relation that depends on r. the terms corresponding to n = 0 will always give the indicia! equation [2r(r .~ = (r D(r + 1) = 0. it is more efficient to begin by substituting y = xr L CnXn. we shift the index of summation in the third sum by 2 to reduce its exponent from n + r + 2 to n + r. so the exponents are r 1 = ~ and r2 = 1. .l)cnxn+r + 3 L(n + r)cnxn+r n=O 00 00 .1) + ~r. The indicia} equation is r(r .1) + 3r  1]co = 2 (r 2 + ~ r .L Cn. (21)we get 00 00 2 L(n n=O + r)(n + r. Rather than separately substituting 00 00 Yt = x 112 Lanxn n=O and Y2 = xl L bnxn n=O in Eq.2Xn+r . With the value r 1 = ~ it becomes a recurrence relation for the series for y 1. (20)the original differential equation.
Cn = 0.1 ( 1 + 2 + .+ . 14 616 55..1]cl = (2r 2 + 5r + 2)cl = 0. (23) is 2(n + r)(n + r. (25) 1: rr = ~.. 4. (25). and 6 in (27). 40 x4 x6 2160 ) • .440 . .. (24) we see that an = 0 whenever n is odd. and 6 in Eq.. Because the coefficient 2r 2 + 5r + 2 of c 1 is nonzero whether r follows that CJ = ~ or r = 1.2(n + r) 2 + (n + r). . = ~ in Eq. 2 Hence the second Frobenius solution is x2 Y2(x) = box . it = 0 (24) in either case.3..+ . (26). We now write an in place of Cn and substituter This gives the recurrence relation for n 2: 2. Eq. (26) With this formula we can determine the coefficients in the first Frobenius solution y 1 • In view of Eq.. (25).. With n = 2. This gives the recurrence relation b _ n bn. we get and Hence the first Frobenius solution is Yl (x) a4 ao a6 = 90 = 55.1 CASE for n 2: 2.1)cn + 3(n + r)cn  Cn2. We solve for en and simplify to obtain the recurrence relation Cn2 Cn = . ....1 in x2 x4 x6 ) CASE 2: r2 = 1. The coefficient of xn+r in Eq. (27) Again.2 2b 2 . 4. (24) implies that bn = 0 for n odd.3n for n ~ 2. . We now write bn in place of Cn and substitute r Eq..3 Regular Singular Points 227 The terms corresponding to n = l yield [2(r + 1)r + 3(r + 1). With n = 2.440 · = aox 112 ( 1 + ++ + ··· . . we get bo b2 = .
1tco = . it is in fact a power series. Eq. denoted by lo(x). the indicia! equation is r(r. 22. it will not be a Frobenius series. 00 Thus we obtain only the single exponent r = 0. Substituting n 6 in Eq. Because p 0 = 1 and q 0 = 0. the Bessel function of order zero of the first kind.4. We will derive that solution in Section 3. and the term for xn yields the recurrence relation Cn. (28).1) = + r = r 2 = 0. (28) Solution In the form of ( 17). and so there is only one Frobenius series solution y(x) = x0 L CnX n n=O of Eq. Thus (30) In this example we have not been able to find a second linearly independent solution of Bessel's equation of order zero. (29). 4. 42 . we get = 2. the pattern is ~n =  ( l)nc0 ( . we see that Cn = 0 whenever n is odd. so our series will converge for all x > 0. Hence x = 0 is a regular singular point with p(x) 1 and q(x) = x 2 . The term corresponding to x 0 gives 0 = 0: no information... (2n)2 22n(n!)2 The choice c0 = 1 gives us one of the most important special functions in mathematics.n2 for n 2:: 2.228 Chapter 3 Power Series Methods Example 5 Find a Frobenius solution of Bessel's equation of order zero. the result is 00 00 00 Ln(n.2Xn = 0.. (29) Because c 1 = 0. • . Thus we substitute y = L CnXn in (28). and Evidently. The term corresponding to x 1 gives c 1 = 0. We combine the first two sums and shift the index of summation in the third by 2 to obtain 00 00 L n 2cnXn n=O +L n=2 Cn . (28) becomes y II + y + 2y = x x 1 I X2 0.1)cnxn n~ + L:ncnxn + LCnXn+2 = n~ n~ 0.2 Cn = .
which differ by an integer.3.r2 is a positive integer. we had obtained an equation such as 0 · c 1 = 3. The r (r . (31).2)cnxn. our computations will simultaneously yield both Frobenius series solutions. if r 1 .2 + Cn2Xn. then Theorem 1 guarantees only the existence of the Frobenius series solution corresponding to the larger exponent r1• Example 6 illustrates the fortunate case in which the series method nevertheless yields a second Frobenius series solution. we begin anew with the larger exponent r = r 1 to obtain the one Frobenius series solution guaranteed by Theorem 1. If the second solution does not exist. 2 and q0 so we see that x indicia} equation 0 is a regular singular point with Po = 0.3 Regular Singular Points 229 When r 1 . .1) + 2r = r (r + 1) = 0 has roots r 1 = 0 and r2 = 1.l)(n. this would have told us that no second Frobenius series solution could exist. which can be satisfied for no choice of c 1 . (31) Solution In standard form the equation becomes II y + y +y= x x2 2 I X2 0. Example 6 Find the Frobenius series solutions of xy" + 2y' + xy = 0. n=O n= 2 L 00 (32) = 0 and n = 1 reduce to 0 · co =0 and 0 · c 1 = 0. Hence we have two arbitrary constants c 0 and c 1 and therefore can expect to find a general solution incorporating two linearly independent Frobenius series solutions. In this case when r 1 . for n = 1.2 + Lcnxn = 0. As you will see. If. the recurrence relation will then tell us whether or not a second Frobenius series solution exists. The case in which the second solution is not a Frobenius series will be discussed in Section 3.r2 Is an Integer Recall that.l)cnxn. This gives 00 L(n.l)cnXn.2 + 2 L(n. n~ n~ n~ We combine the first two sums and shift the index by 2 in the third to obtain L The cases n 00 n(n .4.r 2 is an integer. it is better to depart from the standard procedure of Example 4 and begin our work with the smaller exponent. Hence we begin by substituting y = x1 L CnXn = L CnXn1 n=O n=O 00 00 00 00 in Eq.2 = 0. If it does exist.
a general solution of Eq. We have thus found a general solution expressed as a linear combination of the two Frobenius series solutions FIGURE 3. and y" the equation has two linearly independent power series solutions. The solutions cosx sinx Y1 (x) = and Y2(x) = .1. (31) is y(x) =xI L n=O 00 C11 X 11 = :o (1 _~~ + :~.1) for n. . one of these Frobenius series solutions is bounded but the other is unbounded near the regular singular point x = 0a common occurrence in the case of exponents differing by an integer..c3 =  5.4!' =  1 CJ c5 = . If P(x) and Q (x) are both analytic at x = 0.~~ + ~~.4 5!' C6 1 6.3.· · ·) y Thus y(x) = (CoCOSX X 1 + CJ SlllX). smx (34) in Example 6.3.x X YI(x) = . then x = 0 is an ordinary point. Therefore. = c2 4·3 = C4 1 co ..5 co 6!' C7 1 CJ = .3.:::: 2... in order to investigate the possible existence of series solutions we first write the equation in the standard form + P(x)y' + Q(x)y = 0.7. . 3.230 Chapter 3 Power Series Methods Now knowing that all is well.. C4 1 1 CJ=. As indicated in Fig. ) + ~ (x.C 6 = · 6 7!' evidently the pattern is C2n+ I ( 1) 11 CJ = (2n + 1)! for n ~ 1. from (32) we read the recurrence relation Cn2 Cn= n(n.x cosx and Y2(x) = X. • Summary When confronted with a linear secondorder differential equation A(x) y" + B(x)y' + C(x )y = 0 with analytic coefficient functions. .1. (33) The first few values of n give c2 =Nco. 2CJ.
xy" + (x. find the exponents of the differential equation at x = 0.3 Regular Singular Points 231 Otherwise. then the substitution t = x . 4xy" 18.x 2 )y' . x = 0 is a singular point.4xy = 0 4xy" + 8y' +xy = 0 xy" . If it is a regular singular point. 5. 8.(x 2 . 25. 29. Then construct a graph showing their graphs for x > 0. 2. 14. where p 0 = p(O) and q0 = q(O). 2x 2 y" + x(x + l)y'. and we next write the differential equation in the form y" + p(x) y' + q(x) y = O.2x 2 )y'.4xy = 0 If x = a f=.(cosx)y = 0 Note that x = 0 is an irregular point of the equation x 2 y" Find two linearly independent Frobenius series solutions (for x > 0) of each of the differential equations in Problems 17 through 26.y' +4x 3 y = 0 4x 2 y" .(2x + l)y = 0 (2x 2 + 5x 3 )y" + (3x .(x 2 + 2)y = 3x 2 y" + 2xy' + x 2 y = 0 2x y" + (1 + x) y' + y = 0 2xy" + (1. and if r 1 . 15.2) 2 y' + x 3 y = 0 4)y" + (x.2)y' + y = 0 x 2 )y" .x 2 )y = 0 x(l + x)y" + 2y' + 3xy = 0 x 2 (1 . determine whether x = 0 is an ordinary point. 16. a regular singular point.0 is a singular point of a secondorder linear differential equation. the existence of a second Frobenius series solution y 2 = xr2 L bnxn is also guaranteed.x)y" + (3x + 2)y' + xy = 0 In Problems 32 through 34. 24. 10.x 2 )y" + 2xy'. 32. 2x y" . In this case we find the two exponents r 1 and r2 (assumed real. In Problems 1 through 8. 31.2xy' + l2y = 0 2) 3 y" + 3(x . + (3x  1) y' + y = 0. 13.2y = 0 x 2 y" + (6 sin x)y' + 6y = 0 (6x 2 + 2x 3 )y" + 2lxy' + 9(x 2 . 4. 2x y" + 3 y'  + 2y' + y = 0 y = 0 (a) Show that y = xr L CnXn can satisfy this equation only if r = 0. What is the radius of convergence of this series? . (b) Substitute y = L CnXn to derive the "formal" solution y = L n !xn . There always exists a Frobenius series solution y = xr 1 L anxn associated with the larger exponent r 1 .1) + por + qo = 0.(1 + x)y = 0 2x 2 y" + (sinx)y'. 35. xy" +2y' +9xy = 0 xy" + 2y' .x 3 )y' + (sinx)y = 0 xy" + x 2 y' +(ex .2x 2 )y = 6x 2 y" + 7xy'. 22. We then attribute to the original equation at x = a the behavior of the new equation at t = 0. 3.4)y' + (x + 2)y = 0 . 6.4x 2 )y = 0 x)y" +xy' +x 2 y = 0 x) 2 y" + (2x . find the first three nonzero terms of each of two linearly independent Frobenius series solutions. 7.2)y' + (x + 2)y = 0 9) 2 y" + (x 2 + 9)y' + (x 2 + 4)y = 0 2) 2 y". 34. and with r 1 ~ r2 ) by solving the indicia! equation r(r.(1 + 2x 2 )y = 2x 2 y" + xy'. 33.4xy' + (3 .l)y = 0 19. 21. 27.y =0 0 0 0 3xy" + 2y' + 2y = 0 2x 2 y" + xy'.r2 is not an integer.3. 9.a transforms it into a differential equation having t = 0 as a singular point. Classify (as regular or irregular) the singular points of the differential equations in Problems 9 through 16. 30. 20. 12. 17. 23. 11. or an irregular singular point. (1(1 (1(x(x 2 (x 2 (xx 3 (1 Use the method of Example 6 to find two linearly independent Frobenius series solutions ofthe differential equations in Problems 27 through 31.y' . then x = 0 is a regular singular point. x x2 If p(x) and q(x) are both analytic at x = 0. 1.(3. 26. 28.l)y = 0 x 2 y" + (cosx)y' + xy = 0 3x 3 y" + 2x 2 y' + (1.
(35) has a power series solution Y+ = Xi """' ~PnX n and Y.lfx . Consider the equation x 2 y" + xy' + (1 . (c) Deduce from part (b) that the differential equation given in this problem has realvalued solutions of the form = n2 y 1 (x) y2(x) = = A(x) cos(lnx) .:. y =0 Yt that appeared in an advertisement for a symbolic algebra program in the March 1984 issue of the American Mathematical Monthly. Hence deduce (using the geometric series) that X has exponents r 1 = 1 and r2 = 1 at x = 0.3. that is. Substitute this series (with c 1 = 1) in the differential equation to show that c 2 = 2. (a) Show that its exponents are ±i.1)(x + 1) 2 y" + 2x(x. it follows (why?) that Eq.=xi L qnX n= O 00 11 y(x) = x0 L C X = L CnX 11 11 00 ()() 11 n=O n=O n=O . (a) Show that x = 0 is a regular singular point ofEq.3. This famous equation has wideranging applications in mathematics and physics. The solutions Y1 (x) = . where A(x) = L:a11 x n and B(x) = L:bnx". Conclude that Pn and qn are complex conjugates: Pn = an +ibn and qn = an ibn. (c) Show that the equation x 3 y" + Ax 2 y' + By = 0 has no Frobenius series solution. This problem is a brief introduction to Gauss's hypergeometric equation x(l . and FIGURE 3. with exponents 0 and 1 . and that the Frobenius series corresponding to r 1 = 1 is Jl(x) Yl(x) = (1 +x) 2 = 2 ~ n! (n + 1)!22n. andy are constants.1)y = 0.2(x .jX sinx .xy' + y = 0. Consider the differential equation x(x . then with r = i to obtain qn = Cn. 42.jX cosx and Yz(x) = .) 37. + 2rn Apply this formula with r =ito obtain Pn = Cn. 39. to derive its general solution for x > 0. A(x) sin(lnx) + B(x) cos(lnx). (b) Repeat part (a) with the equation x 3 y" + Axy' + By = 0.2 shows the graphs of the two indicated solutions. f>. (35) = x.B(x) sin(lnx).n)cn1 + (n 2 5n. (a) Use the method of Frobenius to derive the solution y 1 = x of the equation x 3 y".(a+ f> + l)x]y' .1 LC 11 = 0 00 11 X 11 • 40. (c) Use the recurrence relation in part (b) to prove by induction that en = (1)"+ 1n for n .232 Chapter 3 Power Series Methods with p 0 is 36. (n + l)(n + 2) x 2 y" + xy' + (x 2 ..:.x)y" + [y.y. (a) Show that x = 0 is a regular singular point with exponents r 1 = l and rz = 0. Apply the method of Frobenius to Bessel's equation of order~.2. show that it is impossible to determine the coefficients in Yz(x) for 0 < x < 1.2)c11  (n 2 + 7n + 4)cn+I Problem 38. (35). (b) If y is not zero or a negative integer. (a) Suppose that A and B are nonzero constants. x 2 y" +xy' + (x 2  Dy = 0. for n . Show that the equation x 2 y" + Ay' + By = 0 has at most one solution of the form y = xr Len xn.. (b) Verify by substitution the second solution y2 = xe. Cn+2 =  m (n 2 . 1 (!). so it has complexvalued Frobenius series solutions 00 where a. (Suggestion: In each case substitute y = xr L CnXn in the given equation to determine the possible values of r. (a) Show that Bessel's equation of order 1. (b) It follows from Theorem 1 that this differential equation has a power series solution of the form Y1 (x) =x + CzX 2 + c3x 3 + · · · .af>y = 0.3)(x + 1)y' . = q0 = 1.x)y = 0. where the numbers {a11 } and {bn} are real. (b) Show that the recursion formula Cn Cn1 . Does y2 have a Frobenius series representation? 38. 2. X oo (1tx2n (b) Show that there is no Frobenius solution corresponding to the smaller exponent r 2 = 1.1)y Figure 3. c3 = 3. 41.
Show that (1) F(l. 2. But we saw in Example 6 of Section 3.4 Method of Frobenius: The Exceptional Cases with c0 =I= 0.3. (d) The series in (36) is known as the hypergeometric series and is commonly denoted by F(a. For applications.3 guarantees that Eq. this is the most important secondorder linear differential equation with variable coefficients. 1.3 we derived the indicial equation by substituting the power series p (x) = L PnXn and q (x) = L qnxn and the Frobenius series y(x) = xr L CnXn = L CnXn+r n=O n=O 00 00 (co =I= 0) (3) in the differential equation in the form x 2y" + xp(x)y' + q(x)y = 0. with N a positive integer. x) = . (c) Conclude that with c0 part (b) is = 1 the series in 1 1x (the geometnc series). The Nonlogarithmic Case with r 1 = r 2 +N In Section 3. If the roots r 1 and r2 of the indicia! equation </J(r) = r(r 1) + por + qo = 0 (2) do not differ by an integer. and fJn and Yn are defined similarly.0 for n ~ 0. 1. (thebinomialseries). and thus there can be only one Frobenius series solution. ~. If r 1 = r 2. 1. then it is possible that a second Frobenius series solution exists. We consider now the more complex situation in which r 1 . F(k. 1. 1. and x = 0 is a regular singular point. (iii) xF (iv) (k. 0 (ii) xF(l. then Theorem 1 of Section 3. after collection of the coefficients of like powers of x. {3.x) = (l+x)k Method of Frobenius: The Exceptional Cases We continue our discussion of the equation y" + p(x) y' + q(x) Y = x2 x 0 (1) where p(x) and q(x) are analytic at x = 0. We will also see that it is possible that such a solution does not exist. (4) The result of this substitution. (1) has two linearly independent Frobenius series solutions. then there is only one exponent available. In fact.r 2 is an integer. the second solution involves ln x when it is not a Frobenius series. 1. Show that the recurrence relation for this series is (a+ n)(fJ + n) c n+l  233  (y + n)(1 + n) cn where an = a(a + 1)(a + 2) ···(a+ n . is an equation of the form L Fn(r)xn+r = 0 n=O 00 (5) . As you will see in Examples 3 and 4.3 that if r 1 = r 2 + N. x) (36) = ln(l + x). x 2 ) = tan.1) for n ~ 1.1 x. y . x). these exceptional cases occur in the solution of Bessel's equation.
. If it happens that then we can choose eN arbitrarily and continue to determine the remaining coefficients in a second Frobenius series solution. For in this case ¢(r2 + N) = 0. . .. Cn . (9) for en and continue to compute successive coefficients in the Frobenius series solution corresponding to the exponent r 1 • But when we use the smaller exponent r2 . . . (10) is not satisfied with any choice of eN. c1. so the indicia! equation is ¢ (r) = r (r  1) + 6r = r 2 + 5r = 0 (12) . Cnl· Although the exact formula is not necessary for our purposes.234 Chapter 3 Power Series Methods in which the coefficients depend on r.1) + por + qoJco = ::1 ¢(r)co. . ••. it follows that the exponent r and the coefficients c 0 . But if it happens that then Eq. there is a potential difficulty in computing eN. ••• . in this case there cannot exist a second Frobenius series solution corresponding to the smaller exponent r2 • Examples 1 and 2 illustrate these two possibilities. (7) Here Ln is a certain linear combination of c0. then the coefficient ¢(r 1 + n) of en will be nonzero for every n ~ 1 because ¢ (r) = 0 only when r = r 1 and when r = r2 < r 1. also. en must satisfy the equation ¢(r + n)cn + Ln(r. It turns out that the coefficient of xr is Fo(r) = [r(r. we therefore can solve Eq.J have been determined. (9) becomes (10) At this stage co. Cn . c 1 . Example 1 Consider the equation (11) Here Po = 6 and q 0 = 0. c 1..1· Now suppose that r 1 = r2 + N with N a positive integer. Cn1) = 0. (6) which gives the indicia! equation because c0 the coefficient of xn+r has the form 0 by assumption. for n > 1. (9). ••• . •. Co.. c 1. c1. c 1.. (9) This is a recurrence relation for Cn in terms of co. (4 ). CNI have already been determined. If we use the larger exponent r 1 in Eq. Once c0 . so Eq. . it happens that n1 Ln = :L)Cr + k)Pnk + qnkkk· k=O (8) Because all the coefficients in (5) must vanish for the Frobenius series to be a solution of Eq.
and we can compute additional coefficients.3. We now follow the recommendation in Section 3.4 Method of Frobenius: The Exceptional Cases 235 with roots r 1 = 0 and r 2 the Frobenius series y = 00 = 5.. (14) If n =I= 5. Eq. (14) takes the form 0 · c5 + 0 = 0. Hence c5 is a second arbitrary constant. and for n = 5 Eq.5)c I 11 = 0. it is always the coefficient c N that requires special consideration. whereas for n ~ 1 we get the equation [(n + r) 2 + 5(n + r)]c + (n + r)cn1 11 = 0. we can solve this equation for c11 to obtain the recurrence relation Cn1 Cn = . n=O n=O When we combine the first two and also the last two sums. We substitute c 11 xn+r and get 00 :L)n n=O + r)(n + r . Here N = 5. the roots differ by the integer N L = 5.= . n for n =I= 5. With r2 = 5.3 for the case r 1 = r2 + N: We begin with the smaller root r 2 = 5. (15) This yields Ct =Co. The terms corresponding to n = 0 give the indicial equation in (12). (16) and C4 = . and in the latter shift the index by 1.5)c11 + (n. (13) reduces to n(n.. Note that the coefficient of en is ¢ (n + r). the result is 00 00 L[(n n=O + r) 2 + 5(n + r)]CnXn+r + L(n + r)cn!Xn+r = n=l 0. 4 24 C3 CO In the case r 1 = r 2 + N. . still using the recursion formula in (15): Cg =  S C7 Cs = 6 · 7 · 8' (17) and so on.1)c 11 xn+r + 6 :L)n + r)c n=O 00 11 xn+r 00 + L(n + r)cnxn+r+l + LCnXn+r+l = 0. (13) which in this example corresponds to the general solution in (9)..
so the indicial equation is ¢(r) = r(r.8]cnxn+r + L n=2 Cn.x ++2 6 24 x3 x2 x4) and ofEq.1). x6 6 + .8 = r2  2r. The coefficient of xr gives the indicial equation. Solution Here p0 = 1 and q0 = 8.5 (1.236 Chapter 3 Power Series Methods When we combine the results in (16) and (17).r .· ··· • (18) Example 2 Determine whether or not the equation has two linearly independent Frobenius series solutions. we get 00 00 L[(n n=O + r) 2  2(n + r).2Xn +r = 0. On substitution of y L Cn Xn+r in Eq. (18). . .8] C J = 0. we get 00 00 L(n n=O + r)(n + r 1)cnx n+r. (11).6·7 x7 x8 6·7·8 + ·· · ) in terms of the two arbitrary constants c0 and c5 • Thus we have found the two Frobenius series solutions Yl (x) = x.8 =0 with roots r 1 = 4 and r 2 = 2 differing by N = 6.X+ 2x3 + C5X 5 ( X5 . and the coefficient of xr+l gives [ (r + 1) 2  2(r + 1) .8 L CnX n+r = 0. we get 00 y =x5Lcnxn n=O = CoX5 x2 6 + 24 x4) (1.L(n n=O 00 + r)cnxn+r 00 +L CnXn+r+2 . n=O n=O If we shift the index by 2 in the third sum and combine the other three sums.
For n ~ 2 we get the equation [Cn + r) 2  2(n + r). (19) to obtain the recurrence relation c 1!n(n+6) This gives co C2 Cn . (22) = . We work first with the smaller root r = r 2 = 2.3. 8' C2 C4= .2 (n ~ 2). Thus there is no Frobenius series solution corresponding to the smaller root r 2 = 2. note that the coefficient of c11 is ¢(n + r). and hence there is no way to choose c 6 so that this equation holds. 4·10 = Co 2·4·8·10 . (18). Then Eq.2.8] Cn + Cn2 = 0. n f= 6). • . To find the single Frobenius series solution corresponding to the larger root r 1 = 4. = 0. it follows that c 1 = 0 in each case.6) Cn2 (n ~ 2. The general pattern is ( 1) 11 co ~ = 2 · 4 · · · (2n) · 8 · 10 · · · (2n + 6) = ( 1) 11 6Co ..2 = 0 (20) for n ~ 2. For n f= 6 we can solve for the recurrence relation Cn =n(n. (20) with n = 6 reduces to 0 · C6 + 64 co = 0. 2211 n! (n + 3)! This yields the Frobenius series solution (1t x2n ) 1 + 6 ~ 22nn! (n + 3)! 00 YI(x) = x4 ( of Eq. (21) Because c 1 = 0. (19) which corresponds in this example to the general equation in (9). this formula gives C3 = 0. But co f= 0 by assumption. c4 864' C2 Co and cs Now Eq. we substituter = 4 in Eq. (19) becomes n(n 6)cn + Cn.4 Method of Frobenius: The Exceptional Cases 237 Because the coefficient of c 1 is nonzero both for r = 4 and for r = 2.
(27) is p thus = exp ( / (2~: + P(x)) dx) = exp (21n IYII + JP(x)dx). + PyJ)v 1 = 0. (25) We begin by substituting the expression in (25) in Eq. and rearrangement gives " v [ Y1 ]" + P Y1 + Q Y! + v Y1 + 2 v Y1 + P v Y! = 0· But the bracketed expression in this last equation vanishes because y 1 is a solution of Eq. Y2 will then be given by Y2(x) = v(x)yJ (x). using the derivatives I I I d /! /! 2 I I /! Y2 = vy 1 + v YI an Y2 = vy 1 + v y 1 + v YI· We get [vy~1 + 2v y. we need an additional technique. If we write u = v 1 and assume that y 1 (x) never vanishes on I. (26) yields u1 + (2~: + P(x)) u = 0. This leaves the equation v" YJ + (2y. By Theorem 2 of Section 2. Suppose that we know one solution y 1 of Eq. we would like to find the quotient v(x) = Y2(x). our problem is to find y 2 . (23). + V YI] + Qvy1 1 11 1 = 0. . (27) An integrating factor for Eq.1. which enables us to use one known solution y 1 of a secondorder homogeneous linear differential equation to find a second linearly independent solution y 2 . Consider the secondorder equation y" + P(x)y + Q(x)y = 1 0 (23) on an open interval I on which P and Q are continuous. (23). there exists a second linearly independent solution y2 . Equivalently. We discuss here the method of reduction of order. then Eq. (26) is linear in V 1 • Thus the substitution in (25) has reduced the secondorder linear equation in (23) to the firstorder (in v 1 ) linear equation in (26).238 Chapter 3 Power Series Methods Reduction of Order When only a single Frobenius series solution exists. + v"y1] + P 1 1 [vy. (24) YJ(X) Once we know v(x). (26) The key to the success of this method is that Eq. (23). p (x ) = y~ exp ( / P(x) d x ).
where we introduced the method of reduction of order in Problem 36 (though without deriving there the reductionoforder formula itself). so v' =u= ~ exp (. (23) with leading coefficient 1.3 we mentioned that the equidimensional equation x 2y" + p 0 xy' + q0 y = 0 has the power function y(x ) = xr as a solution if and only if r is a root of the quadratic equation r 2 + (p 0 .9x y' + 25 y = 0 by its leading coefficient x 2 to get the standard form II 9 I 25 y .1Or+ 25 = (r . (23) on any interval where Y1 (x) is never zero. so y 1 and Y2 are linearly independent solutions.J ~ dx) dx x . Another integration now gives Yz _ _ Jexp(jP(x)dx) .2. In Section 3.J P(x)dx). This gives the single power function solution y 1 (x) = x 5 of our differential equation. Thus our particular equidimensional equation has the two independent • solutions Y1 (x ) = x 5 and y 2 (x) = x 5 In x for x > 0.y = 0 x x2 in Eq.C dx 2 Y1 Yt + K. Here p 0 = 9 and q0 = 25.1)r + q 0 = 0.. (28) This formula provides a second solution y 2 (x) of Eq. Example 3 For an elementary application of the reductionoforder formula consider the differential equation x 2y" . Before we can apply the reductionoforder formula to find a second solution. so our quadratic equation is r 2 .9/x and Q(x) = 25jx 2. Thus we have P(x) = .y +.9xy' + 25y = 0.4 Method of Frobenius: The Exceptional Cases 239 We now integrate the equation in (27) to obtain uyfexp (! P(x)dx) = C.IO exp (9lnx) dx = x5 J = x5 J x .10 x 9 dx = x 5 lnx for x > 0. .. With the particular choices C = 1 and K = 0 we get the reductionoforder formula Yz = Yt J exp( J P(x)dx) Yt 2 dx. we must first divide the equation x 2y" . y 2 (x) is a nonconstant multiple of y 1(x ) .v. so the reductionoforder formula in (28) yields the second linearly independent solution Y2(x) = x 5 J (x~) 2 exp (. Similar applications of the reductionoforder formula can be found in Problems 3744 of Section 2.5) 2 = 0 and has the single (repeated) root r = 5. Note that because an exponential function never vanishes.3.
2rt = 1 . x x2 (1) under the assumption that its exponents r 1 and r 2 = r 1 .r1N) = 0. We assume that we have already found the Frobenius series solution Yl (x) = xr 1 L n= O 00 GnXn (ao =f. (r . We now substitute (29) and (31) in (28). 0) (29) for x > 0 corresponding to the larger exponent r 1 • Let us write P(x) for p(x)jx and Q(x) for q(x) j x 2 . Thus we can rewrite Eq.1)r + qo = + N) = r 2 + (N.2rl)r + (r~.N differ by the integer N ~ 0.r1)(r.N.N. Po . Because the indicia} equation has roots r 1 and r 2 = r 1 . with the choice a0 = 1 in (29). (30) In preparation for use of the reduction of order formula in (28).r1 so we see that that is.Pt X . so that exp (I P(x)dx) = x . this yields . we write P(x ) = Po+ PIX+ pzx X 2 + · ·· =  Po X + Pl + pzx + · · · .~pzx 2  • • ·). it can be factored easily: r2 + (po.240 Chapter 3 Power Series Methods The Logarithmic Cases We now investigate the general form of the second solution of the equation y" + p(x) y' + q(x) y = O. Then exp ( I P (x) dx) = exp (  I [:o + PI + pzx + · · · Jdx) · · ·) = exp (Po Inx. (23). (1) in the form y" + Py' + Qy = 0 ofEq.~pzx 2 = xPo exp (PIX. (31) In the last step we have used the fact that a composition of analytic functions is analytic and therefore has a power series representation. the initial coefficient of that series in (31) is 1 because e 0 = 1.Po (1 + A 1x + Azx 2 + · · ·) .
the general form of Y2 is Y2(x) = Yl (x) lnx xl+r 1 L bnxn. n=O r2 00 (33) Note the logarithmic term. (32) gives Y 2 = Y1 J(~ + + Yt e 1 + ezx 2 + · · ·) dx + ~e2 x 2 + · · ·) = Ytlnx = y 1 lnx = y. CASE 2: POSITIVE INTEGRAL DIFFERENCE (ri = + N).4 Method of Frobenius: The Exceptional Cases 241 We expand the denominator and simplify: Y2 = Y1 = Yt I I xPozr. then there is no logarithmic term. if so. (1 + A 1x + Azx 2 + · · ·) 1 + B. Eq. ·) N + 1 N ( r (L oo a x n) x n=O n 1 + N e.x + . CASE 1: EQUAL EXPONENTS (r1 = r2 ). 3 .x + e2x 2 + · · ·) dx (32) (Here we have substituted (30) and indicated the result of carrying out long division of series as illustrated in Fig.I. x + e2x 2 + · · · + eNxN + · · ·) d x 1 xN+I XN dx =y 1 e..x__N_+_' + .1.lnx N =eN y 1 lnx+x 2 + N + _e_. it is always present when r 1 = r2. 0.. Note the coefficient eN that appears in (34) but not in (33).lnx (e. If it happens that eN = 0.x xIN (1 + B2x 2 + · · · dx + e. ).) We now consider separately the cases N = 0 and N > 0. We want to ascertain the general form of y2 without keeping track of specific coefficients. + Consequently..x + ~e2x 2 + · · ·) + xr 1 (box+ b.. With N > 0.N eN X (1 + e._N = eN y. Eq. (32) gives Y2 = Yt I x .1. ) N + 1 ' so that Y2 (X) = e N Yl (x ) lnx + x r2 L n=O 00 bnx n' (34) where b0 = a0 j N =f=.x + xr' (1 + a 1x + · · ·) (e. in the case of equal exponents. With N = 0.3. (1) has a second Frobenius series solution (as in Example 1). ··· ) I(+++ + Yt (x_ . . noting in particular that the constant term of the quotient series is 1.. Eq. x 2 + b2x 3 + . This gives the general form of Y2 in the case of exponents differing by a positive integer.
The radii of convergence of the power series of this theorem are all at least p. Theorem 1 (next) is a summation of the preceding discussion and also tells where the series in (33) and (34) converge.r2 = N.242 Chapter 3 Power Series Methods In our derivation of Eqs. then Eq. of the indicia! equation r(r. a positive integer. with r 1 ~ r2. (4) Let p > 0 denote the minimum of the radii of convergence of the power series 00 00 p(x) = LPnXn n=O and q(x) = Lqnxn. respectivelywe have said nothing about the radii of convergence of the various power series that appear. so the logarithmic term may or may not actually be present in this case. THEOREM 1 The Exceptional Cases Suppose that x = 0 is a regular singular point of the equation x 2y" + xp(x)y' + q(x)y = 0. (33) and (34)which exhibit the general form of the second solution in the cases ri = r2 and ri .1) + por + qo = 0. then Eq. (4) has two solutions y 1 and yz of the forms 00 YI(x) = Xr Lanxn n=O 1 (ao =I 0) (35a) and 00 Y2(x) = YI(x )lnx +xq+l Lbnxn. (36b).(x)lnx +xr Lbnxn. . The coefficients in these series (and the constant C in Eq. n=O Let ri and r2 be the roots. (a) If ri = r2. we restrict our attention to solutions for x > 0. n=O 2 (36b) In Eq.r2 = N > 0.3. As in Theorem 1 of Section 3. (36b)) may be determined by direct substitution of the series in the differential equation in (4). n=O (35b) (b) If r1 . (4) has two solutions YI and Y2 of the forms 00 YI(X) = x rl L:anXn n=O (ao =I 0) (36a) and 00 Y2(X) = c y. bo =I 0 but C may be either zero or nonzero.
3.l)n (2n) 22n (n!)2 (42) .22n n' 2 +blx+22 b2x 2 n=l ( .Y1 X 2 I  Yl 2 X + ""' ~ n (n . and it follows that all the coefficients of odd subscript in Y2 vanish.3 we found the first solution (38) According to Eq.) oo n + ~(n 2 bn+bn. 22 . we read the recurrence relation (2)(.2 = 0 if n is odd. n=2 00 We substitute these in Eq. (1 !)2 For n ~  1 4 (41) 2. (37) for which r 1 = r2 = 0. First we see that 1)_( 2 b2 = 2 · =(=_)~ 22 . n=l n=l and it follows that oo ( l)n2 2n "'"' nx 0=2~ .l)bnX n2 . n=3 "'"' (40) The only term involving x in Eq. (40) is b 1x. so b 1 = 0. Now we examine the coefficients with even subscripts in Eq. (37) and use the fact that J0 (x) also satisfies this equation to obtain 0 = x 211 y2 + x y 2 + x 2 Y2 I I I = ( x 211 y 1 + xy 1 + x 2 Yt )I n x + 2xy 1 00 00 00 + Ln(nn=2 1)bnxn + Lnbnxn + Lbnxn+2. (35b) the second solution will have the form 00 Y2 = Ytlnx + Lbnxn. In Example 5 of Section 3. But n 2bn + bn.4 Method of Frobenius: The Exceptional Cases 243 Example 4 We will illustrate the case r 1 = r2 by deriving the second solution of Bessel's equation of order zero.2)X. n=l (39) The first two derivatives of y 2 are 00 1 1 I Y2 = Y1 nx Yl + ""' b n1 +~n nX X n=l and y 2 = y 1 In x 11 11 + . (40).
Tf . The power series in (45) converges for all x. 1 1 1 1 c6=c4+ =1++3 2 3' Cg = C6 + 4 = 1 + 2 + 3 + 4. Noting also the alternation of sign.4 x2 128 + . Finally. Substitution of (43) in (42) gives ( 2)( 2t(2n) 22n(n!)2 which boils down to the extremely simple recurrence relation C2n = C2n . Evidently. We note the presence of 22n (n !) 2 on the righthand side in (42). with n = 1 in (43)..2 + . and their solution often requires a bit of ingenuity. Note the "nonhomogeneous" term (not involving the unknown coefficients) on the righthand side in (42). 1 1 1 (44) where we denote by Hn the nth partial sum of the harmonic series L (1 I n). we get c2 = 1.. we make the substitution (43) in the expectation that the recurrence relation for c2n will be simpler than the one for bzn· We chose ( l)n+l rather than ( l)n because b2 = ~ > 0. The most commonly used linearly independent [of J0 (x)] second solution is Yo(x) = .··· 13824 11x 6 (45) of Bessel's equation of order zero. in conjunction with the coefficient (2n) 2 on the lefthand side. we substitute (43) and (44) in (39) to obtain the second solution y 2 (x) = J0 (x) lnx +"' n=l oo ( l)n+l Hnx2n ~ 22n(n!) 2 3x 4 = Jo(x) lnx +. Such nonhomogeneous recurrence relations are typical of the exceptional cases of the method of Frobenius. n 1 1 Thus C4 = C2 + 2 = 1 + 2. 1 1 1 1 and so on.244 Chapter 3 Power Series Methods from (40).(y .ln2)yi Tf 2 + 2 y2. we are induced to think of b 2n as something divided by 2 2n (n !) 2 .. c2n =1+++···+=H 2 3 n n. The usual strategy depends on detecting the most conspicuous dependence of b2n on n. keeping in mind that the coefficients of odd subscript are all zero.
+ .. According to With P (x) = 1I x from (48). 8 32 4608 x x3 llx 5 (50) . (48) is 1.+ ··· ) .+ .YI = YI I ~dx I I xyl 1 dx x(x/2.. the reduction of order formula in (28) yields Y2 = YI . (48) the associated indicia! equation has roots r1 = 1 and r2 Problem 39 of Section 3.3.+··· ) 4 192 4608 3 1 7x 19x ) ++ + +··· x 3 192 4608 2 4 2 4 6 dx by long) ( division 4x dx 1 = Ytlnx +4yl ( . • Example 5 As an alternative to the method of substitution..4 Method of Frobenius: The Exceptional Cases 245 that is.57722. (46) where y denotes Euler's constant: y = lim (Hn noo Inn)~ 0.r2 = N by employing the technique of reduction of order to derive a second solution of Bessel's equation of order 1. 7x 19x + +.++. we illustrate the case r 1 . one solution of Eq. 2x2 384 18432 Thus yz(x) = YJ(x ) lnx  1 X + .. (47) This particular combination Y0 (x) is chosen because of its nice behavior as x + +oo. Y0 (x) =  2 [ JT: X (l)n+IHnX2n] (y +In) lo(x) +"" 2 f=r 2 n(n!) oo 2 2 .x3j16 + x 5 j384.. it is called the Bessel function of order zero of the second kind..x 2/8 + x 4 /192  x 6 / 9216 + · · · )2 dx = 4yl =4y l f I ( 1 ( x3 1 7x 19x 1+ x .3.x 7 / 18432 + · · · )2 4 x\1 .
3y' + 2y = 0 12.2y = 0 13. n.1+ Loo (l)n(Hn + Hn_J)x 2n. x 2 y" + (2x 2 . The reader can verify that the terms (52) The most commonly used linearly independent [of JJ] solution of Bessel's equation of order 1 is the combination =  2[(y + ln X) J1 (x) . . 4 128 13284 x2 3x 4 llx 6 In Problems 9 through I 4.4 +. Then use the reduction of order technique (as in Example 4) to find the logarithmic term and the first three nonzero terms in a second linearly independent solution. xy" + (3 .(4+x)y' +3y = 0 6. or find one such solution and show (as in Example 2) that a second such solution does not exist. 64 2304 x4 x6 Using the method of reduction of order.2y = 0 8.l] .xy = 0 10. x 2 y" + x(l + x)y'. 5xy" + (30 + 3x)y' + 3y = 0 5. 16. x 2 y" + (2x + 3x 2 )y' . x 2 y" + x 2 y'.y = 0 2.x)y'... 1.3x)y' + 3y = 0 14..xy' + (x 2 + l)y 11. either apply the method of Example I to find two linearly independent Frobenius series solutions.(6 + 2x) y' + y = 0 7. 2 X n=i Tr 22nn! (n. first find the first four nonzero terms in a Frobenius series solution of the given differential equation.+ ··· . + (51) where Hn is defined in (44) for n shown in Eq. In Problems I through 8. xy" + (5 + 3x)y' + 3y = 0 4.direct substitution and reduction of order. (50) agree with ~ 1. xy" + (5..x(l + x)y' + y = 0. n .x)y' .1)! (53) • Examples 4 and 5 illustrate two methods of finding the solution in the logarithmic cases. Begin with xz Jo(x) = 1 . 2x y" . Find two linearly independent Frobenius series solutions of Bessel's equation of order ~. derive the second linearly independent solution Yz(x) = Jo(x)lnx ++. x 2 y" . xy" + y' . A third alternative is outlined in Problem 19. With a computation similar to that shown in Example 4 (but more complicated see Problem 21 ). 9.4y = 0 15. the method of substitution can be used to derive the solution 1 y3(x) = YI(X) lnxX +L oo n=i ( l)n(Hn 22n 1( HnJ)X2ni _ 1)' . x(l .x)y".246 Chapter 3 Power Series Methods Note that the technique of reduction of order readily yields the first several terms of the series. but does not provide a recurrence relation that can be used to determine the general term of the series. (a) Verify that y 1 (x) = xex is one solution of x 2 y" .y = 0 3. xy". Ho = 0. x 2 y" + (x 2  3x)y' + 4y = 0 =0 17. of Bessel's equation of order zero.
(b) Differentiate Eq.d Co.1)! + 2) 2(r + 4) 2 · · · (r + 2n) 2 (61) With r = r 1 = 0 in (58). Use the method of Problem 19 to derive both the solutions in (38) and (45) of Bessel's equation of order zero. ¢(r + n) . Hence you need to compute cn(O) and c~(O) only for n even. r)] = x'(r . this gives J0 (x). (54).l)bn+l n=2 + bn_IJxn (62) (a) Deduce from the discussion preceding Eq. r 1) is a second solution of Eq. CJ. show that Eq. Derive the logarithmic solution in (51) of Bessel's equation of order 1 by the method of substitution. (a) Substitute 19..for n :::0: 2.rJ) 2 = 0. which has exponents r1 = 1 and r2 = 0 at x = 0. (c) Deduce from (60) that Czn(r) Conclude from this result that a second solution is Y2(x) = Y1 (x) lnx = ( l)n (r + 1 oo Hn + Hn1 n L x . (55) reduces in this case to (b) Substitute Y2 = Cy1lnx + Lbnxn n=O Cn .r) = Lcn(r)xn+r. Suppose that the differential equation L[y] = x 2y" + xp(x)y' + q(x)y = 0 has equal exponents r 1 = r 2 at the regular singular point x = 0. ~ 2 2n(n + 1)! n! . (57) with respect to r to show that = xex lnx oo H xn+l L _n_ _ n=l Deduce that y 2 = y.y = 0 to derive the recurrence relation n(n + l)bn+l  bn =  2n + 1 C.bl +X+ L[(n 2 .3. (a) Take c0 = 1. . n=O (56) .4 Method of Frobenius: The Exceptional Cases (b) Note that r 1 = r 2 = 1. and then deduce from (60) that Cn(O) = c~(O) = 0 for n odd. (n + 1)!n! (b) Next show that c 1(0) = c~ (0) = 0. so that its indicia! equation is ¢(r) = (r .bn1 = .(n + r) 2 (60) forn ~ 2.Hn. (55) in Bessel's equation to obtain 00 Then define the function y(x. ~ Let co = 1 and define Cn (r) for n that is. Cn. rJ) = L Cn(rJ)Xn+r n=O 00 1 (58) in the differential equation to deduce that b 1 1 and that 1 nbn . in the equation xy" . (9) that L[y(x . (c) Differentiate Eq. (9). Ln(r. r) of x and r to be 00 y(x. Thus the second solution is Yz(x) is one solution of Eq..y = 0. (54). The following steps outline this computation. Substitute 00 247 Hence deduce that Y1 Yz = Y1lnx + Lhnxn+l n=l = y(x. (58) with respect to r to show that Yz = Y1 lnx n! 18. 21. (x. n=l 00 (59) ~ n! (n00 00 xn 1)!. (57) oo ( l)n (2n + l)x 2n+l ] +C [ x + ~ = 0. The following steps outline this computation. n! (c) Substitute bn = cnfn! in this recurrence relation and conclude from the result that Cn = . (a) Derive the Frobenius series solution Yl(x) = + x' 1 L c~ (rJ)xn. n=l n! (n.r 1) 2 .2(r) Cn(r) = . (d) Differentiate (61) to show that (54) Substitution of this result in (59) gives the second solution in (45). Consider the equation xy". . 20.. Cn(r) =  1 by using Eq.
If we substitute y = 2. The equation itself appears in a 1764 article by Euler on the vibrations of a circular drumhead. 2nd ed.248 Chapter 3 Power Series Methods in Eq. The Case r = p formula > 0 If we use r = p and write am in place of em. (2) is left to the reader (Problem 6). (d) Take b 2 = ~ and substitute ( 1)"Czn 22"(n. 26 . 3(p + l)(p + 2)(p + 3) . The first few even coefficients are a2  ao  2(2p + 2) . (3) Because a 1 = 0.1] b2n+2 + bzn = (~ 1)"(2n + 1) 2 "(n + l)!n! b _ 2"  (63) C2n+2.. 1. (2) yields the recursion am = am. The verification of Eq. Such functions first appeared in the 1730s in the work of Daniel Bernoulli and Euler on the oscillations of a vertically suspended chain. The standard source of information on Bessel functions is G.:. (63) to obtain 1 (b) Deduce from Eq.C2n =  n+1 + . then Eq.  ao 22 (p + 1) .2 m(2p + m) . and Fourier used Bessel functions in his classical treatise on heat (1822). (62) that C = 1 and that bn = 0 for n odd. (1) Its solutions are now called Bessel functions of order p. with roots r = ±p. then b 2n is determined for all n > 1.: cm x m+r in Eq. who was investigating the motion of planets. N. Its 36 pages of references. a2 ao a 4 = . which cover only the period up to 1922. Note that if b2 is chosen arbitrarily. (c) Next deduce the recurrence relation [ (2n + If. n 1 for n . Watson's A Treatise on the Theory of Bessel Functions. (Cambridge: Cambridge University Press.2(p_+_1_)(_p_+_2_)' 4 a6 = a4 6(2p + 6) =  ao .1)!n! (e) Note that c2 = 1 = H 1 + H 0 and deduce that liD Bessel's Equation We have already seen several cases of Bessel's equation of order p ~ 0.4(2p + 4) = 2 . (1). Bessel ( 17841846). it follows that am = 0 for all odd values of m. give some idea of the vast literature of this subject. But their general properties were first studied systematically in an 1824 memoir by the German astronomer and mathematician Friedrich W.p 2 = 0. 1944). Bessel's equation in (1) has indicia! equation r 2 . 2 . we find in the usual manner that Ct = 0 and that (2) form ~ 2.
· (p + m) = 00 ( l)m x2m+p (4) If p 0 this is the only Frobenius series solution. it is the The Case r = p 0 If we user = . we take bm = 0 form odd and define the coefficients of even subscript in terms of bo by means of the recursion formula bm. k(k. then no value of bm can satisfy this equation.. (5) is simply 0 · bm + bm_2 = 0. Hence if pis not a positive integer. Thus in this case we obtain the second solution 00 ( l)m x 2m. .p Y2(x ) = bo "\" £=o 22mm! (p + 1)(p + 2) .. (7) The series in (4) and (7) converge for all x > 0 because x = 0 is the only singular point of Bessel's equation.2p) bm =  m ::::: 2. Eq. For suppose that p = k/2 where k is an odd positive integer. · (p + m) . whereas the leading term in y2 is b0 x .5 Bessel's Equation 249 The general pattern is ( l)mao 2 2mm! (p a2m = + I)(p + 2) · · · (p + m) . but y 2 (x) + ±oo as x + 0. 22mm! (p + I)(p + 2) . Thus if bm2 =I= 0. whereas b 1 = 0. If p > 0. We see that there is a potential difficulty if it happens that 2 p is a positive integerthat is. i. if p is either a positive integer or an odd positive integral multiple of i· For then when m = 2p. except with p replaced with . < = 1 as well.k)bk + bk2 = 0.p. we see that (6) will lead to the same result as that in (4). Eq. and this equation will hold because bk = bk. The crucial step is the kth step.3. so it is clear that y 1 and y 2 are linearly independent solutions of Bessel's equation of order p > 0.2 ' m(m .2p)bm + bm2 = 0 (5) for m ~ 2. Then we need only choose bm = 0 for all odd values of m . so with the larger root r = p we get the solution Yt (x) = ao £=o "\" .P . (6) In comparing (6) with (3). (2) takes the form m(m. with a0 function J0 (x) we have seen before. Hence y 1 (0) = 0.p and write bm in place of Cm. then the leading term in y 1 is a 0 xP. But if p is an odd positive integral multiple of we can circumvent this difficulty.2 = 0.
(9) and (10). r(2) = 2! . for instance. we note first that f(1) = 1 0 00 et dt = lim b'>00 [er] b 0 = 1. X . [See. (11) An important special value of the gamma function is (12) where we have substituted u 2 fort in the first integral. r(x + 1) = xr(x) . Calculus: Early Transcendentals. (10) This is the most important property of the gamma function. To see the way in which r(x) is a generalization of n!. r(4) = 3. r(3) = 3!. (9) Then we integrate by parts with u = tx and dv = er dt: =X (lim rb et txl dt) . b+ oo lo that is. If we combine Eqs. and in general that r(n + 1) = n! for n ~ 0 an integer. which is defined only if n is a nonnegative integer. but is far from obvious. 2008). the fact that 1 o oo e . NJ: Prentice Hall.4 of Edwards and Penney. then r(x ) = r(x + 1).] Although r (x ) is defined in (8) only for x > 0.u2 d ~ u= 2 is known.250 Chapter 3 Power Series Methods The Gamma Function The formulas in (4) and (7) can be simplified by use of the gamma function r (x). If . 7th edition (Upper Saddle River. r(l) = 1!. Example 5 in Section 13. r(3) = 2. The gamma function is a generalization for x > 0 of the factorial function n!.1 < x < 0. we see that r(2) = 1 . which is defined for x > 0 by (8) It is not difficult to show that this improper integral converges for each x > 0. we can use the recursion formula in (10) to define r(x) whenever x is neither zero nor a negative integer.
The graph of the extended gamma function. If pis not an integer. respectively (see Problem 27). 2). (13) through (15) to get the correct solutions for x < 0. Bessel Functions of the First Kind FIGURE 3. (p + 2)(p + l)r(p + 1) by repeated application of Eq. this is one of the finest expositions in the entire literature of mathematics.1. if p > 0 is not an integer. Indeed. the zeros of the . As suggested by Fig.5. The same formula may then be used to extend the definition of r (x) to the open interval (2.2. where p (p > 0. xP must be replaced with lx IP in Eqs. 3. then to the open interval ( 3. Thus and The graphs of J0 (x) and 11 (x) are shown in Fig. (13) gives ln(x ) .3. (l)m (X )2m+n m=O m! (m + n)! (16) 2 for the Bessel functions of the first kind of integral order. 3. (10). The student who would like to pursue this fascinating topic further should consult Artin's The Gamma Function (New York: Holt.. If p = n . 1964)..5. If we choose a 0 r(p = 1/[2Pr(p + 1)] in (4). 3. and so on. In only 39 pages. In a general way they resemble damped cosine and sine oscillations. if you examine the series in (17). and note that + m + 1) = + m)(p + m.p + m + 1) 2 (X )2mp (14) of Bessel's equation of order p. 1). a nonnegative integer.~ _ f.5 Bessel's Equation 251 the righthand side is defined because 0 < x + 1 < 1. you can see part of the reason why J0 (x) and cos x might be similaronly minor changes in the denominators in (17) are needed to produce the Taylor series for cosx.1. Rinehart and Winston.5. then Eq. we have the general solution (15) for x > 0.5. we choose bo obtain the linearly independent second solution = 1/[2Pr(p + 1)] in (7) to J_p(x) = ~ oo ( l)m m! r(.2.1) . we can write the Bessel function of the first kind of order p very concisely with the aid of the gamma function: l)m lp(x) = ~ m!r(p + m 00 ( + 1) 2 (X )2m+p · (13) Similarly. The graph of the gamma function thus extended is shown in Fig.
(19) Bessel Functions of the Second Kind The methods of Section 3. It turns out that J P (x) is an elementary function if the order p is half an odd integer.. the nth zero of 1 1(x) is approximately (n + !} 7T.!} rr.1735 13.1)! m_!_x_n:2m with the notation used there.7915 14.9270 7. FIGURE 3. For n large.2(x) = {2 cosx. the nth zero of J0 (x) is approximately (n. You can see the way the accuracy of these approximations increases with increasing n by rounding the entries in the table in Fig.6537 11.4706 (n + D1r 3.3.3237 16. For instance.6537.5.2(x) = y. (13) and (14).6394 11. Yn (x) is called the Bessel function of the second kind of integral order n 2:: 0.252 Chapter 3 Power Series Methods y ·n. Here.9226 nth Zero of Jt(x) 3. 3.2. Thus the interval between consecutive zeros of either lo(x) or J 1(x) is approximately rranother similarity with cosx and sinx.4048 5. y. If n = 0 then the first sum in (20) is taken to be zero.5201 8. 8.. 2.. and 11. functions J0 (x) and 11(x) are interlacedbetween any two consecutive zeros of lo(x) there is precisely one zero of 1 1(x) (Problem 26) and vice versa. 5.8317 7. (21) It is important to note that Yn(x) + oo as x + 0 (Fig. A very complicated generalization of Example 3 in that section gives the formula Yn(X) = 2 ( 7T y +In 2 ln(X) X) 1 7T L __ m=O nl 2n . Thus if y(x) is a continuous solution of Bessel's equation of order n.4978 8.0156 10.7915.4 must be used to find linearly independent second solutions of integral order.5. Zeros of J0 (x) and J 1 (x).0686 10.5. The graphs of the Bessel functions J0 (x) and 1 1 (x) . The first four zeros of J0 (x) are approximately 2. the results can be recognized (Problem 2) as l1. it follows that y(x) = cln(x) .5201.3562 5. 3.3 to two decimal places..3518 16.2102 13. respectively.4934 X 1 2 3 4 5 2.m(n  2 m.9309 FIGURE 3.4048. {2 sinx and l1.':""'41r ( ·" 'i) . The general solution of Bessel's equation of integral order n is y(x) = c1ln(x) + c2Yn(x). on substitution of p = ~and p = ~in Eqs. Hence c2 = 0 in Eq. (21) if y(x) is continuous at x = 0.7810 14.4).5.
Differentiation of Jp(X) =]. The graphs of the Bessel functions Y0 (x) and Y1 (x). Figure 3. we obtain (Problem 8) the identities (24) .5.P.. and thus we have shown that l)m x2m+2pl 22m+p.1)!' f.. respectively.lm! (p + m.5.1)! oo ( l)mX2m+pl .. especially in the evaluation of integrals involving Bessel functions. .1(x).4. we see in addition that if n = 0.3.d[P x lp(x) ].xP ""' .Im! (p + m.[xP lp(x)] = . then c = y(O)..5 Bessel's Equation 253 for some constant c. FIGURE 3. Because 10 (0) = 1. m! d dx 00 oo ( l)m r(p + m (X )2m+p + 1) 2 (13) in the case that p is a nonnegative integer gives d . d ..xP lp+l (x).5 illustrates the fact that for n > 1 the graphs of ln(x) and Yn(x) look generally like those of 1 1(x) and Y 1(x). In (0) = 0 while Yn (x) + oo as x + o+. and both functions undergo damped oscillation as x + +oo.22m+p. In Section 9. (22) and (23) and then divide the resulting identities by xP and x . dx (23) If we carry out the differentiations in Eqs.5.:Q + 00 ( l)m x2m+2p 2 2m+Pm! (p ( m)! =].. Bessel Function Identities Bessel functions are analogous to trigonometric functions in that they satisfy a large number of standard identities of frequent utility. y y FIGURE 3.[xP J (x)] dx P =""' f. In particular.P dx (22) Similarly.4 we will see that this single fact regarding Bessel functions has numerous physical applications. The graphs of the Bessel functions ]z(x) and Y2 (x).:Q x l p.5.
Subtraction of Eq.lp1 (x). The identities in Eqs.lo(x)+C. (26). Using first p = 2 and then p = 1 in Eq. X 2p (27) it can be used to express Bessel functions of large negative order in terms of Bessel functions of numerically smaller negative orders. (22) gives J xJ0 (x)dx = x11 (x) +C. Eq. X 2p (26) which can be used to express Bessel functions of high order in terms of Bessel functions of lower orders. Example 3 To antidifferentiate x h (x ). In the form lp1 (x) = lp(x). (22) through (27) hold wherever they are meaningfulthat is. they hold for all nonintegral values of p. (25) from Eq. (23) with p = 1.lp+ l (x). • Jl1(x) .1h(x)dx = .lo(x) so that J3(x) = ~lo(x) + (:2  1) l1(x ). (23) gives J Example 2 J 1 (x)dx= .l1(x) = X X 4 4[2 X J1(x). we first note that J x .1 J 1 (x) +C by Eq. Eq. we get J3(x) = h(x). with p = 0. With similar manipulations every Bessel function of positive integral order can be • expressed in terms of J0 (x) and 1 1 (x).x . In particular. We therefore write . whenever no Bessel functions of negative integral order appear. Similarly.254 Chapter 3 Power Series Methods and (25) Thus we may express the derivatives of Bessel functions in terms of Bessel functions themselves. (24) gives the recursion formula lp+l (x) = Jp(x). Example 1 With p = 0.
(see Fig.2 and hence has an infinite sequence of positive zeros Yni· Yn2• Yn 3• . y(L) = 0 (31) on the interval [0. Because Yn(x)+ oo as x+ 0 but ln(O) is finite. du = 2xdx. xJI(x) and v = xI JI (x). A Stegun. 0 (29) with general solution y(t) = c 1 ln(t) Eq. (28). This gives J xh(x)dx = +2 J JI(x)dx = xJI(x). (32) for n ~ 8 and k ~ 20 are tabulated in Table 9. L] . 3. ln (x) oscillates rather like JI (x) in Fig.5 of M. 3.3. As we will see in Chapter 9. then the differential equation in (31) is that in Eq. . (28) is Hence the general solution of (30) Now consider the eigenvalue problem x 2y" + x y' + (Ax 2  n 2 ) = 0. L].2J0 (x) + C.. • (28) The Parametric Bessel Equation The parametric Bessel equation of order n is where a is a positive parameter. If we write A = a 2 ..5. this equation appears in the solution of Laplace's equation in polar coordinates. Abramowitz and I. It is easy to see (Problem 9) that the substitution t = ax transforms Eq. It follows that the kth positive eigenvalue of the problem in (31) is Ak = (ak) 2 = (y2~) 2 . with the aid of the second result of Example 1. (28) into the (standard) Bessel equation d 2y dy t 2+t dt 2 dt + (t 2 n 2)y = + c2Yn(t) . We seek the positive values of A for which there exists a nontrivial solution of (31) that is continuous on [0. (32) For n > 1. (30). (33) and that its associated eigenfunction is (34) The roots Ynk of Eq.5 Bessel's Equation 255 and integrate by parts with dv =xI h(x) dx. Handbook of Mathematical Functions (New York: Dover. so its general solution is given in Eq. The endpoint condition y(L) = 0 now implies that z = a L must be a (positive) root of the equation ln(Z) = 0. 1965). Thus y(x) = ciln(ax). the continuity of y (x) requires that c2 = 0.5.6).
17. (2) for Bessel's equation.. (22) and (23). of the Bessel function ln (x ). (26). 6.o · oo (l) m3mx2m J 4. J J x 3 lo(x ) dx 15. 7. 9.. 16. Yn3· .z(x) + 1. Use the identities in Eqs.yrr.z(x) + Ls. .:: = . j xlt(x )dx x 3 1 1(x) dx 5.1) r. Verify that the substitution t = ax transforms the parametric Bessel equation in (28) into the equation in (29). 20. (x ) = l p. ·(p+m) · (b) Use the result of part (a) to verify the formulas in Eq. 2" (x j 2)P f(p+1) 1 +~m!(p+l)(p+2) 00 (1)m(x / 2) 2m ] . 18.21p(x) + lp+z(x ). then Jp(X) = [ 2 1 · 3 · 5 · (2n .x cosx) 13. 2. 10. J3/Z (x) =  v~ ~ (cosx + xsin x ).3(x) = Use a computer algebra system to graph y (x) for x near 0. (22) and (23) to evaluate the integrals in Problems 13 through 21. (x/2) .6. 5 . Express J4 (x) in terms of J0 (x ) and J 1(x ). (24) and (25) from those in Eqs. (a) Suppose that m is a positive integer.. Verify the identity in (23) by termwise differentiation.1/3 ( 1 r (D +~22m m!. (13) that Lt. Use the series of Problem 11 to find y(O) = lim y(x) if x+0 r (m + 3) = 2 2 · 5 · 8 · · · (3m . (19) for J 112 (x) and }_ 112 (x). The positive zeros Ynl• Yn z . (10) and (12) that 1) r (n+ 11. Use the relation f(x + 1) = xr(x ) to deduce from Eqs. Does the graph corroborate your value of y(O)? ) Any integral of the form J xm ln (x) dx can be evaluated in terms of Bessel functions and the indefinite integral J 0 (x) dx. (3m. but the function J0 (t) dt is tabulated in Table 11. 12.5. Construct a figure showing the graphs of these two functions. and construct a figure showing the graphs of these functions. . J j x 2 1 1 (x) dx 4 jx 1 1 (x) dx j }z(x) dx h(x)dx 22. h. Deduce the identities in Eqs.. and (27) to show that J.256 Chapter 3 Power Series Methods y Y = Yn(X) X FIGURE 3. The latter integral cannot be simplified further. JJ. Show that This form is more convenient for the computation of JP (x) because only the single value f(p+ 1) of the gamma function is required. X (smx.1 of Abramowitz and Stegun. 2. Apply Eqs.2(x ) . 21. . Explain why this constitutes a proof.lf2 (x) (b) Conclude from part (a) and Eq. (a) Deduce from Eqs. (19).z(x ) = and {!. J jx x 2 l o(x ) dx 4 J0 (x)dx 14. Differentiate termwise the series for J0 (x) to show directly that J~(x) = 11(x) (another analogy with the cosine and sine functions). Prove that J0 (x) = 1 T( 17r cos(x sin e ) de 0 by showing that the righthand side satisfies Bessel's equation of order zero and has the value J 0 (0) when x = 0. Show that 4J. 2 y(x) = x 2 [ls. 1.1) 3m r (3) . 19. (13) and (14) that if pis not a negative integer.z(x)J. 3. 8. Derive the recursion formula in Eq.
~ y. 25.. and x large. 24. Use Eqs.~. {2 cos [x.3. liD Applications _?f B~ssel F~nctions The importance of Bessel functions stems not only from the frequent appearance of Bessel's equation in applications. It can be shown that ln (x) =  (b) If X is so large that (p 2 jx 2 is negligible./'I1i[ and a = (2n + 1)n I 4 in (35) yield the best approximation to ln (x) for x large: ln(x) = 1 12n 0 eixsine 2n dO. (22) and (23) and Rolle's theorem to prove that between any two consecutive zeros of ln (x) there is precisely one zero of ln+l (x). but also from the fact that the solutions of many other secondorder linear differential equations can be expressed in terms of Bessel functions.112 cos(x a) with C and a constants. Explain why this constitutes a proof. Yn(x) ~ y. {2 sin [x. Deduce from Problem 22 that lo(x) Asymptotic Approximations It is known that the choices C = . {2 cos (x. Prove that 11(x) 257 yields = 11n 0 cos(O . 12n 0 eixsin& de= 1n 0 (eixsin& + eixsine) de. Use a computer algebra system to construct a figure illustrating this fact with n = 10 (for instance). ][ = cx. These are asymptotic approximations in that the ratio of the two sides in each approximation approaches unity as x+ +oo. Explain why this does not suffice to prove the preceding assertion. J 0 (x) and Y0 (x) ~ {f sin (x .~(2n + l)n J.01 y. With n ~ 2. (a) Show that the substitution y = x.x sin B) dO z" + ( 1  p'x~ l) n z= 0 ][ by showing that the righthand side satisfies Bessel's equation of order 1 and that its derivative has the value J{ (0) when x = 0.112 z in Bessel's equation of order p.. (2) Then a routine but somewhat tedious transformation (Problem 14) of Eq.x sin 0) dO. (0). To see how this comes about.. we begin with Bessel's equation of order p in the form (1) and substitute w = x . (36) (Suggestion: Show first that Similarly. z= kx/3.. Explain why this suggests (without proving it) that if y(x) is a solution of Bessel's equation..112 (A cosx 11n 0 + B sinx) (35) cos(ne . ( 1) yields .rr) if x is large.. ~ y. 27.~(2n + l)n J. then y(x) ~ x. show that the righthand side satisfies Bessel's equation of order n and also agrees with the values ln (0) and 1.) 26..~n) (37) then use Euler's formula. then the latter equation reduces to z" + z ~ 0..6 Applications of Bessel Functions 23. In particular.
1/ 2 [ CJ J1. (4) It is a simple matter to solve the equations in (4) for a= 1A 2 ' q (3 = 2' j(l. This establishes the following result. i· = i. and p= Under the assumption that the square roots in (5) are real. C a= (3 = 2.A) 2 q (5) 4B k=2~. C = {Pk 2.~. C. (3) is Solutions in Bessel Functions 0.2 (z) = {2' sin z v~ and 1.312 (A cos~+ B sin : .258 Chapter 3 Power Series Methods that is. (7) can be written in the elementary form y(x) = x .5 that J1 . then J_p is to be replaced with Yp. v~ 2 ). (7) is y(x) = x. • If we recall from Eq. Then the equations in (5) give Thus the general solution in (6) ofEq. q and q = 2(3. (3) where the constants A. (3) is where w(z) = CJJp(Z) + c2Yp(Z) (assuming that p is not an integer) is the general solution of the Bessel equation in (1). and q = 4.2 {ix 2) + c2l.2a. (3. B = . If p is an integer.r. and q are given by A= 1. q Eq. Example 1 Solve the equation 4x 2 y" + 8xy' + (x 4  3)y = 0. (7) with Eq.A) 2 ~ 4B. (3). it follows that the general solution of Eq. then the general solution (for x > 0) of i= (6) where a. i. k = and p = i. B = a2  {Pp 2. {2' cos z. (19) of Section 3. we rewrite the former as x 2 y" + 2xy' + ( ~ + ix 4 ) y = 0 and see that A = 2. and p are given by the equations in (5). THEOREM 1 If C > 0. (7) Solution To compare Eq.1. and (1 . k.2 (Z) = we see that a general solution of Eq.2 {ix 2)]. B.
perhaps. and g is gravitational acceleration. The buckling column. and q = 3. being nudged laterally just a bit by a passing breeze).y .6. we now consider the problem of determining when a uniform vertical column will buckle under its own weight (after. (8) Solution First we rewrite the given equation in the form This is the special case of Eq. where E is the Young's modulus of the material of the column.6. we assume that the bottom is rigidly imbedded in the ground.1.6 Applications of Bessel Functions ~· ····· 259 Example 2 Solve the Airy equation y" + 9xy = 0. B(L) = 0. perhaps in concrete. (9) FIGURE 3. We take x = 0 at the free top end of the column and x = L > 0 at its bottom. B(L) = 0. fJ = ~. p is the linear density of the column. 3. and p = ~. fJ = ~. C = y 2 . The equations in (5) give a = ~.. k = ~ y. (3) with A = B = 0. (10) We will accept (9) and ( 10) as an appropriate statement of the problem and attempt to solve it in this form. Thus the general solution of Eq. and q = 3. C = 9. k = 2. Denote the angular deflection of the column at the point x by B(x).EI' (11) we have the eigenvalue problem 8" + y 2 xB = 0. It follows from the equations in (5) that a = ~. I is its crosssectional moment of inertia.3.gp . The differential equation in (12) is an Airy equation similar to the one in Example 2. see Fig. (12) The column can buckle only if there is a nontrivial solution of (12).ected vertical position. From the theory of elasticity it follows that EIdx 2 d2 B + gpxB = 0. 8'(0) = 0.1. So the general solution is (1 3) . It has the form of Eq. otherwise the column wiii remain in its undefl. (8) is • Buckling of a Vertical Column For a practical application. With A.2 . and p = ~. (3) with A = B = 0. For physical reasonsno bending at the free top of the column and no deflection at its imbedded bottomthe boundary conditions are B' (0) = 0.
so (14) The endpoint condition ()(L) = 0 now gives (15) Thus the column will buckle only if z = ~ y L 312 is a root of the equation lI f3 (Z) = 0.5) is shown in Fig. From this it is clear that the endpoint condition ()' (0) = 0 implies that c 1 = 0.. For instance.. The graph of Jl j3 (Z). (see Problem 3 of Section 3.x)=O.• 2) FindRoot[BesselJ[1/J. Most technical computing systems can find roots like this one. {x. 2) (Maple) (Mathematica) (MATLAB) yield the value z1 = 1. each of the computer system commands fsolve(BesselJ(1/J.2}] fzero ( 'besselj ( 1/3.2. we finally get EI ) goA (17) . ) 180 .6..86635 and p = oA. where 8 is the volumetric density of the material of the column and A is its crosssectional area. where we see that the smallest positive zero z 1 is a bit less than 2. we substitute p = ±~ in lp(x) =I.260 Chapter 3 Power Series Methods In order to apply the initial conditions.x]==O. oo ( l)m m! f(p + m + 1) 2 (X )2m+p ' and find after some simplifications that + y I/ 3 f cz31 /3 n) (1  y2x3 6 + y4x6 . The shortest length L 1 for which the column will buckle under its own weight is Ll ~ c~r ~ [3~~ L1 ~ (1 .6. x. The graph of (16) 5 10 15 z FIGURE 3. 3.86635 (rounded accurate to five decimal places). 1 .2.986) (  (!:rr 1 3 / If we substitute z 1 ~ 1. x)'.
Shortest Buckling Length L 1 30ft 6 in. Circular with r = 0. with a steel column or rod for which E = 2. 12. and router= 1. 7. 8. Problems 16.4 y (O) = 0 y (x) =x 0x (. 2 and g8 = 0. (a) Substitute the series of Problem 11 of Section 3.y' + 36x 3 y = 0 x 2 y" . Circular with r = 1.. 11.5 in the result of Problem 15 here to show that the solution of the initial value problem In Problems 1 through 12.l2xy' + (15 + 16x)y = 0 16x2 y". 3.l/ 4 2 x (b) Deduce similarly that the solution of the initial value problem is 2 ) + r G) 1 .28 lbjin. ·:. 15.!.5xy' + (8 +x)y = 0 36x 2 y" + 60xy' + (9x 3 .25 in. 6.= x 2 + y 2 . (b) Show that the general solution of dyjdx = x 2 + y 2 is X (Suggestion: Apply the identities in Eqs. 4. 13. h .5 in.. J .4. 1 du y=u dx transforms the Riccati equation d yjdx = x 2 + y 2 into u" + x 2 u = 0.(5.4 y (x )= x 2r (3) (' 4 ]_1/ 4 2x 2) .5)y = 0 l6x 2 y" + 24x y' + (1 + 144x 3 )y = 0 x 2 y" + 3x y' + (1 + x 2 )y = 0 4x 2 y" . 3.3xy' .x 5 )y = 0 y" + x 4 y = 0 y" + 4x 3 y = 0 Apply Theorem 1 to show that the general solution of xy" + 2y' dy dx = x 2 is + l. 63ft 5 in. 2.4. .3.:· . 3.xy' + (1 + x 2 )y = 0 xy" + 3y' + xy = 0 xy" . SolutiOn curves of . Annular with r. For example. in r for a circular 4 liD 1. We have used the familiar formulas A = n r 2 and I disk. x 2 y" .144x 3 )y = 0 2x 2 y".6 Applications of Bessel Functions 261 for the critical buckling length..5. the formula in (17) gives the results shown in the table in Fig.6.3/4 (~x 2 ) 2r G) h .6.1 (Acos x+ Bsinx) . 14. The data in the table show why flagpoles are hollow.6.6. FIGURE 3. 3 . 75ft 7 in.[' (') 4 Jl /4 (' 2x 2)· Gx + xy = 0 isy(x) = x .nner = 1. (3).) .5 in. 2) 2 ).5 in.. (a) Show that the substitution Some solution curves of the equation dyjdx = x 2 + y 2 are shown in Fig. Verify that the substitutions in (2) in Bessel's equation (Eq.3. .3.2(14. dx . (22) and (23) of Section 3. dy FIGURE 3. 5. 9. express the general solution of the given differential equation in terms of Bessel functions.8 x 107 lbj in. The location of the asymptotes where y (x ) + + oo can be found by using Newton's method to find the zeros of the denominators in the formulas for the solutions as listed here. 10. (1)) yield Eq.
18. and Hobson in the March 1985 issue of Mathematics Magazine (Vol.= A(x )l dx + B(x)y + C(x). Many Riccati equations like the ones listed next can be solved explicitly in terms of Bessel functions.6. Assume that its length is increasing linearly with time. y(a) = y(b) = 0. 3. + A.262 Chapter 3 Power Series Methods Note that if a = b. (28) of Section 2." see the article by Borrelli. 3. where A. about a foot in length from horn to horn. y(a) = y(b) = 0.6.6. = fJ} = P b4 j E I0 • (a) Apply the theorem of this section to show that the general solution of x 4 y" + JJ} y = 0 is y(x) =x (A cos~+ Bsin~).6. Here. this result reduces to Eq. (1) (2) . the moment of inertia I cross section at x is given by I(x) = I (x) of the = ~rr(kx) 4 = 10 · (~) 4 . and the under edge as keen as that of a razor .5. L (t) = a + bt. . . subject to an axial force P of compression. (b) Conclude that the nth eigenvalue is given by /Ln nrrabjL.6.8. As in Section 2. Figure 3.6 Application A Riccati equation is one of the form dy . down and still down it came") of Edgar Allan Poe's macabre classic "The Pit and the Pendulum.y = 0. where L = b a is the length of the rod. where I0 = I (b). Coleman. It can be shown that the oscillations of this pendulum satisfy the differential equation LB" 17. .6. the horns upward. and hence that the nth buckling force is m FIGURE 3. (l8) + 2L'()' + g() = 0 under the usual condition that () is so small that sin() is very well approximated by (): () ~ sin e. and the whole hissed as it swung through the air . 58. the value of I at x =b.5 shows a linearly tapered rod with circular cross section. however. The tapered rod of Problem 17. Consider a variablelength pendulum as indicated in Fig.8. pp. Substitute L = a + bt to derive the general solution X x=a x=b FIGURE 3. 7883). its deflection curve y = y(x) satisfies the endpoint value problem Ely"+ Py = 0. Substitution of I (x) in the differential equation in ( 18) yields the eigenvalue problem x 4 y" For the application of this solution to a discussion of the steadily descending pendulum ("its nether extremity was formed of a crescent of glittering steel. A variablelength pendulum. .
(7) are involved. y(x)) or the Mathematica command DSolve[ y'[x) == x~2 y[x]~2.y2. dx (4) (5) (6) For example. dx dy 2 . x) agrees with Eq. y[x].4. the general solution of Eq. dx ' dy .c/1 .c I 2/3 (2 3x 3/ 2) 112 I 2/3 ( 3x L1 . such as the Maple command dsolve(diff(y(x). (5) is given for x > 0 by y(x) =x 2 3/2) .=y X. Next. 3 3x 312) (2 ) (2 ' (9) . If Bessel functions other than those appearing in Eq.= X .6 Applications of Bessel Functions 263 (3) dy dx = lx2. Problem 15 in this section says that the general solution of Eq. 3.5 to transform the computer's "answer" to (7).6. Each has a general solution of the same general form in (7)a quotient of linear combinations of Bessel functions.3. you may need to apply the identities in (26) and (27) of Section 3. dy . 3 3x 3l 2 . Then see whether your system can take the limit as x ~ 0 in (7) to show that the arbitrary constant c is given in terms of the initial value y(O) by c=:::'''y(O)r (t) 2r (D (8) Now you should be able to use builtin Bessel functions to plot typical solution curves like those shown in Fig. (7).=x+y2. (1) is given by (7) See whether the symbolic DE solver command in your computer algebra system.x) = + x~2 + y(x)~2. In addition to lp(x) and Yp(x). these solutions may involve the modified Bessel functions and Kp(x) = li . For instance. investigate similarly one of the other equations in (2) through (6).p [ lp(ix) n + Yp(ix)] 0 that satisfy the modified Bessel equation x 2y" + xy'  (x 2 + p 2)y = of order p.
.+ . . G) (10) 4 2 >..6. = y2  x. (6)we see a funnel near y = +Jx and a spout near y = Jx. (1) through (6) have interesting solutions which. 1956)..7 shows some typical solution curves. . For instance. The secondorder differential equations of the form y" = f (x.X (11) with the same initial value y(O) = 0 but different slopes y'(O) = 3. an equation that arose historically in the classification of nonlinear secondorder differential equations in terms of their critical points (see Chapter 14 of E. L. Nevertheless..8 shows solution curves of the secondorder equation y" = y2.5.6. 3. .3.1.3. however. y'(O) = . . . y) with the same righthand sides as in Eqs. but the power series of the modified function In (x) is simply that of the unmodified function In (x) except without the alternating minus signs. Equation (11) is a form of the first Painleve transcendant.6. The first Painleve transcendant y" y(O) = 0. . . Solution curves dy of= x.3.8.. the interesting pattern in Fig. 3. New York: Dover Publications. 0.0 2 Figure 3.6.3. The Bessel functions with imaginary argument that appear in the definitions of IP (x) and K P (x) may look exotic.7. (17) and (18) in Section 3.31/3f (~). Figure 3. Ordinary Differential Equations.y 2 • dx and x x3 h (x) = 2 + 16 + 384 + 18432 + · · · · Check these power series expansions using your computer algebra systemlook at Bessell in either Maple or Mathematicaand compare them with Eqs. together with the parabola y 2 = x that appears to bear an interesting relation to Eq. Io(x) = 1 + 4 X 4 x2 + .1.7.264 Chapter 3 Power Series Methods where y(O)r c =. Ince.6. they can be investigated using an ODE plotter.7. 64 2304 x7 x5 x4 x6 FIGURE 3. 2 0 2 4 X FIGURE 3. cannot be expressed in terms of elementary functions and/or "known" special functions such as Bessel functions. 0. .+ . For instance.8 was suggested by an article by Anne Noonburg containing a similar figure in the Spring 1993 issue of the C• ODE•E Newsletter.
1. Can you explain where the rational function part comes from. more exotic special functions are involved. where r is an even integer. or at least verify it? For further examples of this sort. and/or replace the x on the righthand side with x 5 . here's a related example that was inspired by a Maple demonstration package. (13) Show that Theorem 1 in this section explains the "Bessel part" of the alleged solution in Eq. (13) with r 2 .1 (c1 lw(x) + c2Yw(x)) (12) + x. (With parities other than these.11 (1857945600 + 51609600x 2 + 806400x 4 + 9600x 6 + 100x8 + x 10 ) of the nonhomogeneous secondorder equation x 2y" + 3xy' + (x 2  99)y = x. (12).6 Applications of Bessel Functions 265 Finally. where s is an odd integer. The Maple dsol ve command yields the general solution y(x) = x.3.) . you can replace the coefficient 99 in Eq.
we will see in Section 4. In this case the methods of Chapter 2 can be quite awkward.1.2 that the Laplace transform converts a differential equation in the unknown function f(t) into an algebraic equation in F(s). when the voltage supplied to an electrical circuit is turned off and on periodically. and the Laplace transform method is more convenient..for example.1. corresponding to a massspringdashpot system and a series RLC circuit. has discontinuities. The Laplace transformation £ involves the operation of integration and yields the new function £{f(t)} = F(s) of a new independent variable s .1. It often happens in practice that the forcing term. The situation is diagrammed in Fig. recall the differential equations m x " +ex'+ kx = F(t) and LI" 1 + RI' + 1 = c E'(t) D {f(t)) = f'(t) f (t). yields the new function D{f(t)} = J'(t) . '. The differentiation operator D can be viewed as a transformation which. I n Chapter 2 we saw that linear differential equations with constant coefficients have numerous applications and can be solved systematically. 266 . when applied to the function f(t). After learning in this section how to compute the Laplace transform F (s) of a function f(t). this is one method that simplifies the problem of finding the solution f(t). Transformation of a function: £ in analogy with D.i'{f(t) ) = F (s ) FIGURE 4. 4.Laplace Transform Methods Laplace Transforms and Inverse Transforms f(t).. in which the alternative methods of this chapter are preferable. respectively. There are common situations. F(t) or E'(t).1. For example. Because algebraic equations are generally easier to solve than differential equations. however.
e (sa)rJoo o o s.1 Laplace Transforms and Inverse Transforms 267 DEFINITION The Laplace Transform ~ Given a function f(t) defined for all t function F defined as follows: F(s) 0. 1 a 00 g(t) dt = lim b'> oo 1b a g(t) dt.st dt = [~es 1 ] = lim [~ebs + ~]' }0 s b'>oo s s 0 and therefore d.in problems as well as in examples. the definition of the Laplace transform in (1) gives = 00 r)() e. Note that the integrand of the improper integral in (1) contains the parameters in addition to the variable of integration t. This is typical of Laplace transforms. Recall that an improper integral over an infinite interval is defined as a limit of integrals over bounded intervals.a r=o . Therefore. it diverges or fails to exist. otherwise. • Example 2 ··············· · With f(t) =eat fort oC{eat} ~ 0. it is typical for the improper integral in the definition of oC {f (t)} to converge for some values of s and diverge for others. b'>oo a (4) • Remark: The limit we computed in Example 1 would not exist if s < 0. then we say that the improper integral converges. the domain of a transform is normally of the forms >a for some number a. Also. but to afunction F of s. we obtain . in this computation we have used the common abbreviation [ g(t)J 00 a = lim [g(t)Jb.. the Laplace transform off is the = oC{f(t)} = 1 00 est f(t) dt (1) for all values of s for which the improper integral converges.4. when the integral in (1) converges. Hence oC{1} is defined only for s > 0. As in the following examples. p{ 1}  ~ s <: lOr s > 0. (3) As in (3). Example 1 With f(t) = 1 fort oC{1} ~ 0. it's good practice to specify the domain of the Laplace transform.st eat dt = 1 oo e<sa)t dt = [. = 1 oo e. it converges not merely to a number. that is. (2) If the limit in (2) exists. for then (1js)ebs would become unbounded as b+ +oo.
1)1(n . then e(sa)t + 0 as t + +oo. (5) Note here that the improper integral giving £{ea 1 } diverges if s ~ a .1) · (n. which is defined for x > 0 by the formula (6) For an elementary discussion of 1 (x) . and dt = dujs in this integral. agrees with the factorial function for x = n. so it follows that _1_ c 10r d. thus 1(n + 1) = n! (9) if n is a positive integer.268 Chapter 4 Laplace Transform Methods If s a > 0. e(sa)t = eifJt e(sa)t + 0 as t + +oo.5.st ta dt. the function 1(x + 1). Therefore. see the subsection on the gamma function in Section 3. which is defined and continuous for all x > 1 . For then.2) = n(n. It is worth noting also that the formula in (5) holds if a is a complex number. then 1(n + 1) = n1(n) (8) = n · (n . It then follows that if n is a positive integer.2)1(n . • The Laplace transform £{ta} of a power function is most conveniently expressed in terms of the gamma function 1(x). where it is shown that 1(1) = 1 (7) and that 1(x + 1) = x1(x) for x > 0.. Example 3 Suppose that f (t) = t a where a is real and a > 1. provided that s > a = Re[a]. Then £ {ta} = ·· 100 e .l)(n . with a =a+ if3.2) · · · 2 · 1(2) = n(n . a positive integer.1) = n · (n . recall that eifJt = cos f3t + i sin f3t. P{eat} _ sa s > a.1)(n.2) · · · 2 · 1 · 1(1). we get (10) . If we substitute u = st. t = ujs.
.........4... Once we know the Laplace transforms of several functions. For instance.. (11) For instance........st f(t) dt) c+oo lo r lo e... The computation of £{tnl 2 } is based on the known special value (13) _.. ....1 Laplace Transforms and Inverse Transforms 269 for all s > 0 (so that u = st > 0)....... Example 4 ·... .. .. . ....... The proof of Theorem 1 follows immediately from the linearity of the operations of taking limits and of integration: oC{af(t) + bg(t)} = 1 00 e ... ... THEOREM 1 Linearity of the Laplace Transform If a and b are constants.... ....... ~ r (~) = ~0T 2 2 2 2 2 2 4.. • Linearity of Transforms It is not necessary for us to proceed much further in the computation of Laplace transforms directly from the definition. we see that oC{tn} = n! n+I + 1) = n! if n is a nonnegative s for s > 0........st [af(t) + bg(t)] dt = lim { c est [af(t) c+oo lo + bg(t)] dt + b (lim c+ oo = a (lim re.. without the use of the gamma function.... ... .. ........st g(t) dt) = aoC{f(t)} + boC{g(t)} .. ..... 1 oC{t} = 2' s oC{t } = 3' s 2 2 and oC{t } = 4 . then oC{af(t) + bg(t)} = aoC{f(t)} + boC{g(t)} (12) for all s such that the Laplace transforms of the functions f and g both exist. The reason is that the Laplace transformation is a linear operation... ... ... .. .. . these formulas can be derived immediately from the definition. it follows that r (~) = ~r (~) = ~ . . Because r(n integer..... of the gamma function.. ...·········· ... s 3 6 As in Problems 1 and 2..... we can combine them to obtain transforms of other functions.......
1 s. If k > 0. (14) . (16) (The domain follows from s > Re[ik] = 0.) Similarly. £{smhkt} = Because cos kt k 2 s k 2 fors > k > 0. we get £{ 3e21 + 2sin2 3t} = £{3e 21 ···. first with Now the formulas in (10) through (12) yield X = ~ and then with X = ~· • Example 5 Recall that coshkt = (ekr together give + ekr) / 2..··«·  + 1.ik s ik = 2s s (ik):2 2. £{coshkt} = Similarly. and a familiar trigonometric identity.2)(s 2 + 36) 3s 3 • . then Theorem 1 and Example 2 1 1} £{coshkt} = £{ek 2 that is. £{smkt} = . 1 + £{ek 2 1} 1(1 =2 sk 1 ) . .2 +~ + 36 for s > 0. the formula in (16).~.~ Example 6  ~········ Applying linearity. £{coskt} = s s +k 2 2 fors > 0. + 144s .270 Chapter 4 Laplace Transform Methods using the formula f(X + 1) = xf(x) in (9).. the formula in (5) (with a = ik) yields 2 £{coskt} = and thus 1( 1 + +1) 2.k 2 2 for s > k > 0. (17) • . + __ s+k s s .72 s (s . k s +k 2 2 fors > 0.cos 6t} 1 s s2 3 = s . (15) = (eikt + eikt)/2.· ~~.
and 5 we see that 1 { £ s3 1} = 1 2 2t ' £1 {1 } s+2 = e2r ' and so on.1 Laplace Transforms and Inverse Transforms 271 Inverse Transforms According to Theorem 3 of this section. then we call f(t) the inverse Laplace transform of F(s) and write f(t) = £. f(t) has a finite limit as t approaches each endpoint of each subinterval from its interior.E) . This observation allows us to make the following definition: If F(s) = £{f(t)}. We say that f is piecewise continuous for t ~ 0 if it is piecewise continuous on every bounded subinterval of [0.k2 k s2. no two different functions that are both continuous for all t ~ 0 can have the same Laplace transform.2. (18) Example 7 Using the Laplace transforms derived in Examples 2.a) s s2 + kz k f is continuous in the interior of each of these subintervals. The table in Fig. A short table of Laplace transforms.1. sa coskt sin kt cosh kt sinh kt u(t.2 lists the transforms derived in this section.1. Throughout this chapter we denote functions oft by lowercase letters.4. A table of Laplace transforms serves a purpose similar to that of a table of integrals. many additional transforms can be derived from these few.4. whose graph appears in Fig. using various general properties of the Laplace transformation (which we will discuss in subsequent sections).1 . 4. fort ~ 0. The jump in f(t) at the point c is defined to be f(c+). then f(t) is uniquely determined. 3.as s and f(c) = lim f(c. 4. and sZ + k2 2. 4. Thus F(s) is the Laplace transform of f (t) and x (t) is the inverse Laplace transform of X (s).1 {F(s)}.3. E+0+ FIGURE 4. Thus a piecewise continuous function has only simple discontinuities (if any) and only at isolated points.k2 e . Perhaps the simplest piecewise continuous (but discontinuous) function is the unit step function. The transform of a function will al 1 s s2 tn (n ~ (s > 0) (s > 0) (s > 0) 0) n! 5 n+l ways be denoted by that same letter capitalized. as indicated in Fig.1. The function f(t) is said to be piecewise continuous on the bounded interval a ~ t ~ b provided that [a.f(c). (19) . +oo). where f(c+) = lim f(c +E) E+0+ s s z . It is defined as follows: u(t) = {1 0 fort < 0. At such points the value of the function experiences a finite jump. b] can be subdivided into finitely many abutting subintervals in such a way that 1. Thus if F(s) is the transform of some continuous function f(t). • f(t) NOTATION: FUNCTIONS AND THEIR TRANSFORMS. ta (a > 1) ear f(a + 1) (s > 0) 5 a+l (s > 0) (s > 0) (s > 0) (s > lkl) (s > lkl) (s > 0) Piecewise Continuous Functions As we remarked at the beginning of this section. we need to be able to handle certain types of discontinuous functions.
Because u(t) = 1 fort ~ 0 and because the Laplace transform involves only the values of a function fort ~ 0.1. (22) • General Properties of Transforms It is a familiar fact from calculus that the integral 1b g(t) dt exists if g is piecewise continuous on the bounded interval [a . Its jump occurs at t =a rather than at t = 0.st S t=a consequently.(t) = u(t a)={~ fort< a.__ Jb .the limit of this last integral as b + +oo.a) t =a FIGURE 4. u. equivalently. a > 0). u"(t) = u(t." . (20) s The graph of the unit step function ua(t) = u(t. (21) Example 8 Find £{ua (t)} if a > 0. The unit step function Ua (t) has a jump at t = a. But in order for F (s ).272 y Chapter 4 Laplace Transform Methods (0_.stdt= lim b ? oo [_:. We begin with the definition of the Laplace transform. We obtain £{ua(t)}= Solution 1 0 oo estua(t)dt= 1oo a e. the solid dots indicate values of the function at discontinuities.5. b ].1.1.3. fort 2': a.. .a) appears in Fig.5. The graph of a piecewise continuous function. Hence if piecewise continuous for t ~ 0. The graph of the unit step function. FIGURE 4.1. we need some condition to limit the rate of growth of f(t) as . it follow s that the integral f is exists for all b < +oo.. a b X 1)1u(t) t (a .. 1 ) . FIGURE 4. eas £{ua(t)} =  s (s > 0. 4...4.to exist. we see immediately that 1 £{u(t)} = (s > 0).
More precisely. we see that the improper integral f000 estet2 dt that would define £{e12 } does not exist (for any s). then its Laplace transform F(s) = £{f(t)} exists. consider the function f(t) = et 2 = exp(t 2 ). Proof: First we note that we can take T = 0 in (23). and T are not so important. c. this is true (with c = 0) if f(t) itself is bounded. The particular values of M. The function 273 f is said to be of exponential order as t + +oo if there exist nonnegative constants M. To do this. lf(t)l is bounded on [0. = t+oo hm e 1 2_ et = +oo because t 2 . For an example of an elementary function that is continuous and therefore bounded on every (finite) interval. What is important is that some such values exist so that the condition in (23) is satisfied. it then follows that IJ(t)l ~ Meet for all t ~ 0. and T such that lf(t)l :£Meet fort 2: T.ct + +oo as t + +oo. THEOREM 2 Existence of Laplace Transforms If the function f is piecewise continuous for t ~ 0 and is of exponential order as t + +oo. we see that (t) hm . The condition in (23) merely says that f(t)/eet lies between M and M and is therefore bounded in value fort sufficiently large. we can therefore assume that If (t) I :£ M if 0 :£ t :£ T. Because ec1 ~ 1 for t ~ 0. Thus every bounded function.t+ oo eet . f . so we conclude that the function f(t) = e12 is not of exponential order. Whatever the value of c.4. (23) Thus a function is of exponential order provided that it grows no more rapidly (as t + +oo) than a constant multiple of some exponential function with a linear exponent. By a standard theorem on convergence of improper integralsthe fact that absolute convergence implies convergenceit suffices for us to prove that the integral exists for s > c. because est e 12 + +oo as t + +oo. In particular. then the familiar fact that p(t)et + 0 as t + +oo implies that (23) holds (for T sufficiently large) with M = c = 1. c. if f is piecewise continuous and satisfies the condition in (23).such as coskt or sin ktis of exponential order. Hence the condition in (23) cannot hold for any (finite) value M. If p(t) is a polynomial. T]. Increasing Min (23) if necessary. and therefore that the function e 12 does not have a Laplace transform. Thus every polynomial function is of exponential order. Similarly. then F(s) exists for all s >c. it suffices in tum to show that the value of the integral . For by piecewise continuity. The following theorem guarantees that piecewise functions of exponential order do have Laplace transforms. et = t+oo hm eet 2 .1 Laplace Transforms and Inverse Transforms t + +oo. but nevertheless is not of exponential order.
274 Chapter 4 Laplace Transform Methods remains bounded as b + that +oo. the function f(t) = 1/. The remainder of this chapter is devoted largely to techniques for solving a differential equation by first finding the Laplace transform of its solution.) = £{t1 /2 } = _2_ sl/2 ~ s both exists and violates the condition in (24 ). that the function of s we have found has only one inverse Laplace transform that could be the desired solution. (New York: McGrawHill. we get the following result. COROLLARY F(s) for s Large If f (t) satisfies the hypotheses of Theorem 2. that IF(s)l ~ 1 0 lest f(t)l dt ~  M sc (24) if s > c. the function G (s) = s j (s + 1) cannot be the Laplace transform of any "reasonable" function because its limit as s + +oo is 1. then f(t) = g(t) wherever on [0. But the fact that If (t) I ~ M ect for all t ~ 0 implies ~ M 1 o oo oo e(sc)t dt = . We have shown. which would imply that sF (s) remains bounded as s + +oo. The following theorem is proved in Chapter 6 of Churchill's Operational Mathematics. as we shall see) a Laplace transform only if the degree of its numerator is less than that of its denominator. More generally. not 0. On the other hand. M sc if s > c. It is then vital for us to know that this uniquely determines the solution of the differential equation. (25) The condition in (25) severely limits the functions that can be Laplace transforms. +oo) both f and g are continuous. the hypotheses of Theorem 2 are sufficient. but nevertheless (Example 3 with a=~> 1) its Laplace transform r (1. THEOREM 3 Uniqueness of Inverse Laplace Transforms Suppose that the functions f(t) and g(t) satisfy the hypotheses of Theorem 2. conditions for existence of the Laplace transform of f(t). 1972). . that is. This proves Theorem 2. so that their Laplace transforms F(s) and G(s) both exist. If F(s) = G(s) for all s > c (for some c). For instance. 3rd ed. a rational functiona quotient of two polynomialscan be (and is. moreover. but not necessary. then lim F(s) S+00 = 0. When we take limits ass + +oo. For example./i fails to be piecewise continuous (at t = 0).
Use the transforms in Fig. f (t) = sinh2 3t .1. Historical Remark: Laplace transforms have an interesting history. 13. The socalled operational methods for solving differential equations. 0) ~(1. f(t) = t 2. and around the beginning of the twentieth century their validity was the subject of considerable controversy. It is customary in mathematics to name a technique or theorem for the next person after Euler to discover it (else there would be several hundred different examples of "Euler's theorem").) Apply the definition in (I) to find directly the Laplace transforms of the functions described (by formula or graph) in Problems I through 10. f(t) = sin 3t cos 3t 20.IOr = t 312  16.1. These techniques were successfully and widely applied before they had been rigorously justified. f(t) = Ji + 3t f(t) = t . 4. and hence must be the same solution if they have the same Laplace transform. 15. f(t) = e3r+I 5.4. 21. 10.6.1 Laplace Transforms and Inverse Transforms 275 Thus two piecewise continuous functions of exponential order with the same Laplace transform can differ only at their isolated points of discontinuity.1)~ (1. One reason is that Heaviside blithely assumed the existence of functions whose Laplace transforms contradict the condition that F (s) j> 0 ass j> 0.l) I FIGURE 4.8. f(t) = sin 2 t 3. f(t) =sin 2t +cos 2t 18. who employed such integrals in his work on probability theory. The integral in the definition of the Laplace transform probably appeared first in the work of Euler. 1) j I ~ • c 11. so we may regard inverse Laplace transforms as being essentially unique. f(t) = cost 6. were not exploited by Laplace. FIGURE 4. two solutions of a differential equation must both be continuous.7.2 to find the Laplace transforms of the functions in Problems II through 22. thereby raising questions as to the nature and role of numbers in mathematics. k:(l. f(t) = (1 + t) 3 = te 1 FIGURE 4. (0. the next person was the French mathematician Pierre Simon de Laplace (17491827). thereby raising questions as to the meaning and nature of functions in mathematics. f(t) = 3t 512 14. A preliminary integration by parts may be necessary. f(t) f(t) = cos2 2t 19.1. 17. they were discovered and popularized by practicing engineersnotably the English electrical engineer Oliver Heaviside (18501925). 9. (1.1) FIGURE 4.2e31 f(t) = 1 +cosh 5t 12. In this case.I. 1. which are based on Laplace transforms. In particular. Indeed. 8. f(t) = t cos 2t 22. f(t) = sinht 7. f(t)  4t 3 e . 1) (2. This is of no importance in most practical applications.9. f(t) = t 2 4.1. (This is reminiscent of the way Leibniz two centuries earlier had obtained correct results in calculus using "infinitely small" real numbers.
Then express f in terms of unit step functions to show that£{/(!)}= s. n=1. Express g in terms of the function f of Problem 40 and hence deduce that to obtain £{cos kt} directly from the definition of the Laplace transform. Show that f can be written in the form 00 27. First. 38. 50 FIGURE 4. First. 4.:.2 to find the inverse Laplace transforms of the functions in Problems 23 through 32. The graph of the function of Problem 41.1. Derive the transform of f(t) =sin kt by the method used in the text to derive the formula in ( 16).es) =s 1 2 s 512 26. ..__ .. Given that 0 < a < b. F(s) = s 2 + 4 9+s 30. 36. F(s) = 4 s 25.n). 4. F(s) 24.:. let f(t) = 1 if 0 . n=O 5.276 Chapter 4 Laplace Transform Methods Apply the geometric series to obtain the result £{f(t)} Use the transforms in Fig.. Derive the transform of f(t) =sinh kt by the method used in the text to derive the formula in (14). F(s) = s..1.:.1.1(1. F(s) = s4 3 3s + 1 28. Given constants a and b. f(t) = 0 if t ~ a. F(s) (b) Use the method of Problem 39 to show that = lOs. Given a > 0. 1 33.. 1f n . . Then express f in terms of unit step functions to show that£{!(!)}= s .1 .10.. 39. Show that the function f(t) = sin(e12 ) is of exponential order as t ~ +oo but that its derivative is not. sketch the graph of the function f.2. Use the tabulated integral 0 4 • 5 06  f eax eax cosbxdx= 2 2 (acosbx +b sinbx)+C a +b FIGURE 4. t < b.1 :::.312 1 = 1 s(l.11. 41. The graph of the squarewave function g(t) is shown in Fig. 42. F(s) = s+5 40. (c) Assume that the Laplace transform of the infinite series in part (b) can be taken term wise (it can). The unit staircase function is defined as follows: f(t)=n I L ~I . 3 23.e .1(e . t < a. F(s) = s 2 + 9 31. =L n=O 00 u(t...bs ). . f(t) = 0 if either t < a or t ~ b. t < n and n IS even. making clear its value at t = a.__ . = a + bes s(l + es) ..F(s)= 4 _s 2 f(t) = L(l)nu(t..3 25. (a) Sketch the graph off to see why its name is appropriate..1. making clear its values at t = a and t = b..10. (a) The graph of the function f is shown in Fig... define h(t) fort ~ 0 by if nI .3s 29..s2 £{f(t)} = s(l + es) .1. . let f(t) = 1 if a .a 4 .a 34. (b) Show that f(t) h(t) = I a b if n ..a 2 3 .. t < n and n is odd. sketch the graph of the function f.eas) . 37. 4. The graph of the function of Problem 40.3.. .n) Sketch the graph of h and apply one of the preceding problems to show that £{h(t)} for all t ~ 0.11. t < n .as ..:. 35.
We may assign arbitrary values to f(t) at the isolated points at which f is not differentiable. the integrated term e . (3) b X Then £{f'(t)} exists for s > c. and its value at the lower limit t = 0 contributes .1. 'b X a ~ Piecewise continuous derivative FIGURE 4. c. lo Because of (3). Then £{f' (t)} exists when s > c. and T such that lf(t)l ~Meet a I : I I I I I I I I I I I I I I I I fort~ T.1 indicates how "comers" on the graph of f correspond to discontinuities in its derivative f './(0). such as ax"(t) + bx'(t) + cx(t) = f(t). we get £{!' (t)} = r>O e .2 Transformation of Initial Value Problems 277 II!J Transformation of Initial Value Problems We now discuss the application of Laplace transforms to solve a linear differential equation with constant coefficients.st f(t)]oo t=O + s {00 e. y THEOREM 1 Transforms of Derivatives Suppose that the function f(t) is continuous and piecewise smooth fort ~ 0 and is of exponential order as t + +oo.4. The function f is called piecewise smooth on the bounded interval [a. We will defer the case in which f'(t) has isolated discontinuities to the end of this section.st!' (t) dt = lo [e . The main idea of the proof of Theorem 1 is exhibited best by the case in which f' (t) is continuous (not merely piecewise continuous) fort ~ 0.2. The integral that remains is simply £{f(t) }. (4) I I I I I I : ~ I . b] and differentiable except at finitely many points.f(O) to the evaluation of the preceding expression. . which tells us how to express the transform of the derivative of a function in terms of the transform of the function itself. beginning with the definition of £ {f' (t)} and integrating by parts. (1) with given initial conditions x (0) = x 0 and x' (0) = xb.2.1. Figure 4. We say that f is piecewise smooth for t ~ 0 if it is piecewise smooth on every bounded subinterval of [0.f(O) = sF(s). Then. b] if it is piecewise continuous on [a. with f'(t) being piecewise continuous on [a. by Theorem 2 of Section 4. so that there exist nonnegative constants M. b]. The transformed equation is a£{x"(t)} + b£{x'(t)} + c£{x(t)} = £{f(t)}. (1) by separately taking the Laplace transform of each term in the equation.st f(t) dt.st f(t) approaches zero (when s > c) as t + +oo. +oo). (4). we can transform Eq. The discontinuities of f' correspond to "corners" on the graph of f. and its value is that given in Eq. and £{/'(t)} I I y' N I I I = s£{f(t)}. (2) it involves the transforms of the derivatives x' and x" of the unknown function x (t). The key to the method is Theorem 1. the integral converges when s > c. By the linearity of the Laplace transformation.
(4) and (5) yield £{x'(t)} = s£{x(t)}  x(O) = sX(s). f ' .sj<n2\0)/ (n .2]  6 [X(s)] = 0.x' .sn2 J' (0). . Then £{/(n)(t)} exists when s > c.6)X(s). .1)(0). .2s + 1.sf(O).6x = 0. . we need the transform of the second derivative as well. x (0) = 2.3 s 2 .278 Chapter 4 Laplace Transform Methods Solution of Initial Value Problems In order to transform Eq.. (7) Example 1 Solve the initial value problem x " . then that theorem implies that £{f"(t)} = £{g' (t)} = s£{g(t)}./(0)]. f ".sn. .x'(O) = s 2X(s) .sx(O) . (6) After finitely many such steps we obtain the following extension of Theorem 1. which we quickly simplify to (s 2 . (1)..s./'(0).2 and £{x"(t)} = s 2£{x(t)}.2s + 1]. and that each of these functions satisfies the conditions in (3) with the same values of M and c.[sX(s).. Solution With the given initial values. where (according to our convention about notation) X (s) denotes the Laplace transform of the (unknown) function x(t) . and thus £{/"(t)} = s 2F(s). Thus X(s) = 2s ./'(0).s 2f(O).f(n1>(0) = sn F(s). Eqs.s. and £{f(n) (t)} = sn £{f(t)}. t<n. COROLLARY Transforms of Higher Derivatives Suppose that the functions f.3 (s .· · · .g(O) = s£{/' (t)}  !' (0) = s [s£{f(t)}.2s +3= 0.1 /(0). Hence the transformed equation is [s 2X(s). x' (0) = 1.6 = 2s .sf'(O)  / "(0). (5) A repetition of this calculation gives £{/"'(t)} = s£{/"(t)}  /"(0) = s 3 F(s).1) are continuous and piecewise smooth fort ~ 0..3)(s + 2) .sn1 /(0). If we assume that g (t) = f' (t) satisfies the hypotheses of Theorem 1.
substitution of s 3 7 2 shows that X(s) = £{x(t)} = ___l_ s3 + ___l_.3) (s + 2) s. it follows that is the solution of the original initial value problem. (s .4.a)}= eat.3).6 = (s . 4.2 Transformation of Initial Value Problems 279 By the method of partial fractions (of integral calculus). 2A . Then upon equating coefficients of terms of like degree. In lieu of any such shortcut. If we substitute s B =~ .3 = A(s + 2) + B(s. we find that A = ~. Note that we did not first find the general solution of the differential equation.3)(s + 2) into the equation 2s. 2s . x(O) = x' (0) = 0.3 = A(s + 2) + B(s.3B = .3) (s + 2) yields the identity 2s.3 = (A+ B)s + (2A. automatically taking into accountvia Theorem 1 and its corollarythe given initial conditions.s .3 which are readily solved for the same values A = ~ and B = ~ • Example 2 Solve the initial value problem x" + 4x = sin 3t.1{1/(s. we get the linear equations A+ B = 2.2. there exist constants A and B such that 2s3 A B =+.3).3) that resulted from clearing fractions. Such a problem arises in the motion of a massandspring system with external force.3 s+2 and multiplication of both sides of this equation by (s . • Remark: In Example 1 we found the values of the partialfraction coefficients A and B by the "trick" of separately substituting the roots s = 3 and s = 2 of the original denominator s 2 .Hence 3. The Laplace transform method directly yields the desired particular solution. the "surefire" method is to collect coefficients of powers of s on the righthand side. as shown in Fig. .2. s+2 Because £.
. A massandspring system satisfying the initial value problem in Example 2. and D. + 4) (s 2 + 9) · The method of partial fractions calls for 3 (s 2 + 4)(s 2 + 9) As+ B s2 + 4 Cs + D s2 + 9 · + The surefire approach would be to clear fractions by multiplying both sides by the common denominator. it follows that sin 2t ! sin 3t. 4. Figure 4.4. Because £{sin 2t} = 2/(s 2 + 4) and £{sin 3t} = x(t) = 3 10 3/(s 2 + 9) . Note that the Laplace transform method again gives the solution directly.=::(s 2 FIGURE 4. Equating coefficients of like powers on the two sides of the resulting equation would then yield four linear equations that we could solve for A. The mass is initially at rest in its equilibrium position. because neither the numerator nor the denominator on the left involves any odd powers of s. here we can anticipate that A = C = 0. Eq.1) and thereby get the transformed equation 3 s 2 X (s) + 4X (s) = 2 9 s + Therefore. When we equate coefficients of like powers of s we get the linear equations B+D =0. which are readily solved forB= ~and D =~. FIGURE 4. However. X sZ 2 1 +4. Thus nonhomogeneous equations • are solved in exactly the same manner as are homogeneous equations.2.Hence 3 X(s) = oC{x(t)} = 10. The position function x(t) in Example 2.3 shows the graph of this period 2n position function of the mass. 98 +4D = 3. 3 X (s) .3. 4.2.280 Chapter 4 Laplace Transform Methods Solution Because both initial values are zero.2.1.2 (Section 4.2. whereas nonzero values for A or C would lead to odddegree terms on the right.5. without the necessity of first finding the complementary function and a particular solution of the original nonhomogeneous differential equation. B. So we replace A and C with zero before clearing fractions. Examples 1 and 2 illustrate the solution procedure that is outlined in Fig. C.2. The result is the identity 3 = B(s 2 + 9) + D(s 2 + 4) = (B + D)s 2 + (9B + 4D) . (5) yields oC{x" (t)} = s 2 X (s). sZ 3 +9. We read the transform of sin 3t from the table in Fig. and then collect coefficients of powers of s on the righthand side.
Then the initial conditions in (9) and £ {y"(t)} £{x" (t)} = s 2 X (s) = s 2Y(s). ..and. y(t) . Using the Laplace transform to solve an initial value problem. A mass.2. ..2 Transformation of Initial Value Problems Differential equation in x(t) Solution x(t) of differential 281 Algebraic equation in X(s) Solution X(s) of algebraic equation FIGURE 4.2. y" = 2x.4.2. the Laplace transform reduces such a linear system of differential equations to a linear system of algebraic equations in which the unknowns are the transforms of the solution functions. Solution We write X(s ) imply that = £{x(t)} and Y(s) = £{y(t)} . of the independent variable t .4. As Example 3 illustrates.spring system satisfying the initial value problem in Example 3. FIGURE 4. assuming that the force f(t) = 40 sin 3t is suddenly applied to the second mass at the time t = 0 when both masses are at rest in their equilibrium positions.5. the technique for a system is essentially the same as for a single linear differential equation with constant coefficients. (8) (9) This initial value problem determines the indicated displacement functions x(t) and y (t) of the two masses shown in Fig. Linear Systems Laplace transforms are frequently used in engineering problems to solve a system of two or more constantcoefficient linear differential equations involving two or more unknown functions x(t). Both masses are initially at rest in their equilibrium positions. When initial conditions are specified. 4.5. Example 3 Solve the system 2x" = 6x + 2y . subject to the initial conditions x(O) = x'(O) = y(O) = y' (O) = 0.2y + 40sin3t.
(12) yields B = 8. Thus we obtain the partial fraction decomposition shown in Eq.6 sin 3t. for instancethe system in (10) for 120 5 8 3 X (s) = (s 2 + 1) (s 2 + 4) (s 2 + 9) = s 2 + 1 . the inverse Laplace transforms of the expressions in Eqs.2. substitution of s 2 = 4 in Eq. Figure 4.6. 2 s +9 (10) 2X(s) + (s 2 + 2)Y(s) = The determinant of this pair of linear equations in X (s) and Y (s) is and we readily solveusing Cramer's rule. At any rate. 5sint 4sin2t + sin3t. so A = 5.s 2 + 4 + s 2 + 9 and 120(s 2 + 3) 10 8 18 Y(s)= (s 2 +l)(s 2 +4)(s2+9) = s2+1 + s2+4. we can write ~~~= 120 (s 2 + 1)(s 2 + 4)(s 2 + 9) A s2 + 1 + s2 + B 4 + s2 + 9 ' C and it follows that 120 = A(s 2 + 4)(s 2 + 9) + B(s 2 + 1)(s 2 + 9) + C(s 2 + 1)(s 2 + 4). (11a) and (11b) are readily found using the method of Example 2. y (t) = 10 sin t + 4 sin 2t .s2 + 9' ° 1a) (llb) The partial fraction decompositions in Eqs. ( 11 a) and (11b) give the solution x(t) = FIGURE 4. 120 . Similarly. (12) y(t) Substitution of s 2 = 1 (that is. the transforms of the equations in (8) are the equations 2s 2X(s) = 6X(s) + 2Y(s). a zero of the factor s 2 + 1) in Eq.6 shows the graphs of these two period 2n position functions of the two masses. s = i. • . 120 s 2Y(s)=2X(s)2Y(s)+ 2 · s +9 Thus the transformed system is (s 2 + 3)X(s) Y(s)=O. noting that the denominator factors are linear in s 2. (12) gives 120 = A · 3 · 8. The position functions x(t) and y(t) in Example 3.282 Chapter 4 Laplace Transform Methods Because £{sin 3t} = 3/(s 2 + 9).2. For instance. (lla). and substitution of s 2 = 9 yields C = 3.
(13) for X (s). I (s) (14) where Z(s) = ms 2 + cs + k and I (s) = mx(O)s + mx'(O) Note that Z(s) depends only on the physical system itself. Much of the remainder of this chapter is devoted to finding Laplace transforms and inverse transforms.7. 4. m [s 2 X(s). a linear equationin the "unknown" X (s). It follows from the linearity of the transform that . (14). we get F(s) X (s) = Z(s) + Z(s)' + cx(O). we seek those methods that are sufficiently powerful to enable us to solve problems thatunlike those in Examples 1 and 2.x(O)] + kX(s) = F(s). (13) Note that Eq.2 Transformation of Initial Value Problems 283 The Transform Perspective Let us regard the general constantcoefficient secondorder equation as the equation of motion mx" +ex'+ kx = f(t) of the familiar massspringdashpot system (Fig. respectively.Z(s) of the steady periodic solution and the transient solution. Hence Theorem I gives cl{eat +ateat} = cl{f'(t)} = scl{f(t)} = scl{tea1 }. In particular. Additional Transform Techniques Example 4 Show that Solution If f(t) =teat. In the case of an underdamped system. then f(O) = 0 and J'(t) =eat+ at eat.x'(O)] + c [sX(s). (13) is an algebraic equationindeed. (14) presents X (s) = cl {x (t)} as the surn of a term depending only on the external force and one depending only on the initial conditions.7). these two terms are the transforms cl{X5p(t)} = F(s) Z(s) and I (s) cl{Xrr(t)} = . If we solve Eq. A massspringdashpot system with external force f(t).cannot be solved readily by the methods of Chapter 2.2.sx(O). This is the source of the power of the Laplace transform method: Linear differential equations are transformed into readily solved algebraic equations.2. The only potential difficulty in finding these solutions is in finding the inverse Laplace transform of the righthand side in Eq.4. Thus Eq. Then the transformed equation is FIGURE 4.
then oC Jo f(r) dr for s > c. then differentiation of f corresponds to multiplication of its transform by s.· Example 5 Solution Find oC{t sinkt}.284 Chapter 4 Laplace Transform Methods Hence (15) because .a).£{f(t)} by the formula in (5) for the transform of the second derivative.£{tsinkt}. It is reasonable to expect the inverse operation of integration (antidifferentiation) to correspond to division of the transform by s. Finally. Equivalently. (16) This procedure is considerably more pleasant than the alternative of evaluating the integral oC{tsinkt} = 1 00 test sinktdt.1 L{f(t)} = F(s) s (17) (18) . f' (0) = 0 and But oC{f"(t)} = s 2 .£{eat}= 1/(s. and oC{coskt} = s/(s 2 + k 2 ). Let f (t) = t sin kt. • • = 0 and v~. {t } = . The result is f"(t) = 2kcoskt k2 tsinkt. Then f (0) j'(t) = sinkt +ktcoskt. The derivative involves the new function t cos kt. we solve this equation for oC{t sinkt} = 2ks (s 2 2 2 + k) . so we have 2ks s +k 2 2 eoC{tsinkt} = s 2 . THEOREM 2 Transforms of Integrals If f(t) is a piecewise continuous function fort ~ 0 and satisfies the condition of exponential order lf(t)l ~Meet fort~ T. so we note that differentiate again. • Examples 4 and 5 exploit the fact that if f (0) = 0.
ecr dr = (ect1) < ec 0 c c so g(t) is of exponential order as this gives t + +oo. find the inverse transform of the resulting simpler expression.g(O). thus g is continuous and piecewise smooth for t ~ 0.4. lg(t)l ~ 1 t lf(r)l dr ~ M 1 0 M M 1.. L{f(t)} = L{g'(t)} = soC{g(t)}._ )]t = a1 (e 0 2 at 1).a) } = 1 1 L1 { 0 s(s a) T 1 } dr = 1/ at o a (ear .a) l } = t L 1{. Hence we can apply Theorem 1 tog. and finally integrate from 0 tot (to "correct" for the missing factors). Now g(O) = 0. We now repeat the technique to obtain L. Furthermore. so division by s yields L {1 1 f(r) dr} L{g(t)} = L{f(t)}' s which completes the proof. Thus L1{ s(s. This technique is often a more convenient way than the method of partial fractions • for finding an inverse transform of a fraction of the form P(s)j[sn Q(s)]. the fundamental theorem of calculus implies that g(t) = 1 1 1 f(r) dr is continuous and that g'(t) = f(t) where f is continuous.1 . Eq. Example 6 Find the inverse Laplace transform of 1 s 2 (s a)· G(s) Solution In effect.2 Transformation of Initial Value Problems 285 Proof: Because f is piecewise continuous. .1 Jo sa } dr = tear dr = Jo a ~(eat  1).1) dr 1 1 eaT = [ . (18) means that we can delete a factor of s from the denominator._ ( .1 { s 2 (s 1 .
! agrees with a function that is continuous on the whole closed interval and has "endpoint values" j (r:_1 ) = lim f (t) t + t+ n. t 3 .: ) = lim f(t) t+tn  that may not agree with the actual values f(tn ..stk.within each interval [tn .2) + e . .1 est j(t) dt.l f Uk. tn) where f' is continuous..f(O). and let t 1 . tki be the points interior to the interval [0. This yields Jo est f'(t) dt = L 0 {b k f tnest J'(t) dt 'n1 t n=i k = L[est J(t) Jn + L n=i tnJ k s itn ln . But from Eq.f(O) + esb f(b) + s 1b est f(t) dt. . Let to = 0 and tk = b.f(t 0 ) + estk f(tk) = . We need to prove that the limit lim {best J'(t)dt b+oo Jo exists and also need to find its value...i)] (20) + [ estk1 f(tkd + estk f(tk) J in (19) telescopes down to. let t 1. ~ +oo in the preceding Extension of Theorem 1 Now suppose that the function f is only piecewise continuous (instead of continuous).J) and f(tn). The value of an integral on an interval is not affected by changing the values of the integrand at the . and the second summation adds up to s times the integral from to = 0 to t k =b. (3) we get if s > c. Then we can integrate by parts on each interval (lni. t 2 . we get the desired result L{f'(t)} = soC{J(t)}. . Therefore (19) reduces to 1b est f' (t) dt = .1 and f(t. With b fixed.sb f(b) .286 Chapter 4 Laplace Transform Methods Proof of Theorem 1: We conclude this section with the proof of Theorem 1 in the general case in which f' is merely piecewise continuous.2 f (tk. Therefore. The fact that f is piecewise continuous includes the assumption that. finally taking limits (with s fixed) as b equation. be the points (fort > 0) where either for f' is discontinuous. (19) n= i Now the first summation + · · · + [ estk.f(O) + e. b] at which J' is discontinuous.i. tn] between successive points of discontinuity. t2 .
2.2 Transformation of Initial Value Problems 287 endpoints.es) I 2 3 4 5 6 t f:o FIGURE 4. Assuming that J:{f'(t)} exists. x(O) = 3. x '(O) = 4 . x " + x = cos 3t.3 5.Le.s < 1.x' . 4.. . x" + 4x 2. x' (O) = 0 B Problems Use Laplace transforms to solve the initial value problems in Problems 1 through 16. Hence Eq. x(O) = 1. Example 7 j(t) 6 5 4 Let f(t) = 1 + [t] be the unit staircase function.) J = . 1 1. x(O) = 2.. 2. x'(O) = . and }1 (n) = 1 for each integer n = 1. n=l k1 (20') where (21) denotes the (finite) jump in f(t) at t = tn. Then f(O) = 1.stk f(tk. x (0) = 0. x" + 9x = 0.1J(t:_I) + e .4.f(O+) L hUn)+ esbf(b). x (O) = 0 = x'(O) 7. The graph of the unit staircase function of Example 7.. x(O) = 0 = x' (O) 6. n=I 3 2 so the Laplace transform of f (t) is 1 00 1 F(s) ='"'ens=s s(l . J'(t) 0.2.ns. However. We therefore use the "continuous from within the interval" endpoint values above in evaluating (by parts) the integrals on the right in (19). if the fundamental theorem of calculus is applied to find the value of the integral. • 3. we therefore get the generalization 00 oC{f'(t)} = sF(s). x' (0) = 2 4. + [ eSlk2 J(t:_ 2 ) + eSlk1 f(tJ:_J)] + [ estk. its graph is shown in Fig. x " + 8x' + l5x = 0. 3. x " . 1.1.f(O+).f(O) when we now take the limit in (19) as b+ +oo.2x = 0.Lestnh(tn) n=l (22) of J:{f'(t)} = sF(s). x '(O) = 0 = 0. x " + 4x = cos t . (22) yields = 00 0 = sF(s)... then the antiderivative function must be continuous on the closed interval.X' with x = e. x " + x = sin2t. The result is + .8.8. x(O) = 5. In the last step we used the formula for the sum of a geometric series.
.. The graph of the squarewave function of Problem 34.) j(t) 1 ..k2 + k2) 2 2ks 2 2 2 2 3 4 5 6 29. If f(t) is the sawtoothfunction whose graph is shown in Fig.s) 2k Apply the extension of Theorem 1 in Eq.s (1 . x'(O) = y'(O) = 1 15.1e. x'. 27. z(O) = 0 Apply Theorem 2 to find the inverse Laplace transforms of the functions in Problems 17 through 24.s2(s2.9.y = 0.) 32.a ··· FIGURE 4.. x(O) = y(O) = 0 x' + 2y' + x = 0.. L{t coskt} = (s2 s2 . b] (where 0 < a < b) and f(t) = 0 otherwise. Apply the results in Example 5 and Problem 28 to show that L 1 { 1 L{j(t)} es (s 2 + k 2 )2 1 } ..k ) 30.k ) 31.. L{u(t. 28.2. x(O) = 1. y' = 6x + 3y. 4. 9. x(O) = 1.. y(O) = 2 x' = x + 2y. If g(t) is the triangular wave function whose graph is shown in Fig. = 2 s . ..12. . y(O) = 1 x" + 2x + 4y = 0.1) s(s + 1)(s + 2) 25. x'(O) = y'(O) = 0 16.10. 10. y(O) = 0.288 8..2. If f(t) = 1 on the interval [a. L{tsinhkt} = (s .a)n+I for n = 1. x' = x+z. y" + x' + y' + 4x .2. If f (t) = ( 1 )rr1n is the squarewave function whose graph is shown in Fig. 1 s L{f(t)} =tanh..a ··· 1 17..2. y' = x + e. Chapter 4 Laplace Transform Methods 34. 35.2y = 0... Apply Theorem 1 to derive L {sin kt} from the formula for L{coskt}. Apply Theorem 1 as in Example 5 to derive the Laplace transforms in Problems 28 through 30. y' = x+y. x'(O) = 2 x' = 2x + y...1 ..11.s 2 3 4 5 6 FIGURE 4. F(s) = s(s2 + 4) = 3 s(s+5) 2s + 1 20.2. 3..a)}= s .10. z' = 2x z. F(s) = 1 s(s2..a . 36. = . x(O) = y(O) = 0. 13.. 37.13 (smktktcoskt)..y' + y = 0. 2. F(s) = .9) = s(l + es) 1 1 1 24. F(s) = 1 s2(s2 + 1) 22. L{tcoshkt} = s2 2 + k2 2 2 FIGURE 4. The graph of the sawtooth function of Problem 37.. x(O) = y(O) = 1. then x" + 9x = 1.2. then 1 s L{g(t)} =tanh.. 12. Apply Theorem 1 to derive L {cosh kt} from the formula for L{sinhkt}... (22) to derive the Laplace transforms given in Problems 32 through 37. x(O) = 0. then eas . x" + x' + y' + 2x . 11.2.2. F(s) = 123456 FIGURE 4..12... (Suggestion: Note that f ' (t) j(t) = 1 where it is defined.a . If f(t) is the unit onofffunction whose graph is shown in Fig.. x (0) = 0 = x' (0) x" + 4x' + 3x = 1.ebs L{f(t)} = .11. F(s) 1 s(s3) 1 19. 2 s 2 g(t) (b) Deduce that L{tnea'} = n!j(s . 4. y" + x + 2y = 0. s 2 (Suggestion: Use the geometric series. (a) Apply Theorem 1 to show that 23. then L{f(t)} j(t ) 1 . F(s) = s(s 2 + 9) 18. then (s . 33.as fora> 0.e . 14. F(s) =  21. The graph of the onoff function of Problem 35. The graph of the triangular wave function of Problem 36.. x(O) = 0. 4. 4.. 26.9. x (0) = 0 = x' (0) x" + 3x' + 2x = t.
. . in terms of the factorization of the denominator Q(s) into linear factors and irreducible quadratic factors corresponding to the real and complex zeros. and Bn are constants. First we must find the partial fraction decomposition of R(s). then .1{R(s)} is based on the same method of partial fractions that we use in elementary calculus to integrate rational functions. THEOREM 1 Translation on the sAxis If F(s) = .4. The following two rules describe the partial fraction decomposition of R(s).£{/(t)} exists for s > c. Finding . B2..a in the transform corresponds to multiplication of the original function oft by eat..a ) 2 + b 2 of multiplicity n is a sum of n partial fractions.. of Q(s).. sa (s a) 2 where A1. An.. and .. . .. Thus the translation s + s .£{eat f(t)} exists for s >a+ c. The latter step is based on the following elementary property of Laplace transforms. . respectively.£I {R (s)} involves two steps..a).£.a of multiplicity n is a sum of n partial fractions. and An are constants. and then we must find the inverse Laplace transform of each of the individual partial fractions of the types that appear in (2) and (3). A2. having the form where A1. B 1. . (4) (5) Equivalently. The technique for finding . . .£{eat f(t)} = F(s. . RULE 2 Quadratic Factor Partial Fractions The portion of the partial fraction decomposition corresponding to the irreducible quadratic factor (s . RULE 1 Linear Factor Partial Fractions The portion of the partial fraction decomposition of R (s) corresponding to the linear factor s .3 Translation and Partial Fractions 289 IIIJ Translation and Partial Fractions As illustrated by Examples 1 and 2 of Section 4. A2.2. the solution of a linear differential equation with constant coefficients can often be reduced to the matter of finding the inverse Laplace transform of a rational function of the form R(s) = P(s) Q(s) (1) where the degree of P(s) is less than that of Q(s). having the form A1 Az An (2) + +···+ (s a)n .
2 shows the graph of this rapidly decaying damped oscillation.290 Chapter 4 Laplace Transform Methods Proof: If we simply replaces with sa in the definition of F(s) = L{f(t)}.a in the transformswe get the following additions to the table in Fig.3t (3 cos 5t + 2 sin 5t). X(s) = 3s + 19 s2 + 6s + 34 = 3 · (s +3 5 2 2 + 3) + 25 + .~.3] + 34X(s) = 0.~=.2. 4.a)= 1 00 e(sa)t f(t) dt = 1 00 est [eat j(t)] dt = L{eat j(t)}. FIGURE 4. let x(t) denote the displacement of the mass m from its equilibrium position.3. cos kt.1.~X~~p~i~g.3 in mks units (Fig. (5) is the same. and it is clear that Eq. X We take the Laplace transform of each term of the differential equation.3s which we solve for 2 1] + 6 [sX(s). x'(O) = 1. The position function x(t) in Example 1. we get the equation = (s 2X(s) .Co~sict~.3.1. all the Laplace transforms derived in this chapter are listed in the table of transforms that appears in the endpapers.dashpot system of Example 1. Because (obviously) L{O} 0. we obtain F(s. and· ~~·. If the mass is set in motion with x (0) = 3 and x' (0) = 1. = 0.a) 2 k + k2 +k2 eat sin kt (sa) 2 (s > a) (8) For ready reference. and sin kt that we already knowmultiplying each of these functions by eat and replacing s with s . This is Eq. The massspring.. FIGURE 4.1).y~~~~ith ~ ~·:·k = 17.3. If we apply the translation theorem to the formulas for the Laplace transforms of tn.a)n+J sa (s > a) (s >a) (6) (7) eat cos kt (s. Solution The differential equation is ~x" value problem x" + 3x' + 17x 0.3. we now see that x(t) = e . eattn n! (s.{~.·. 4. • . As usual. (4). Figure 4. (s + 3)2 + 25 · s Applying the formulas in (7) and (8) with a 2 1t = 3 and k = 5. Example 1 . find x (t) for the resulting damped free oscillations. so we need to solve the initial + 6x' + 34x = x(O) = 3.2.
8s = .8s s s+ 2 s. 2 s3 .2s. Hence A B C s2 + 1 === .4) + Cs(s + 2).4). Solution The transformed equation is s 2Y(s) Thus + 4sY(s) + 4Y(s) 2 = 3· s Y(s) = 2 A B C D E = .4) + Bs(s.+.. 3 2 s  2s . B. we multiply both sides by s 3 to obtain 2 2 (s + 2) =A+ Bs + Cs 2 + s 3F(s). B = ? 2. and 24C = 17.3 Translation and Partial Fractions 291 Example 2 illustrates a useful technique for finding the partial fraction coefficients in the case of nonrepeated linear factors.2s .+. 12B = 5. s denominator Q(s) in this equation. Example 2 Find the inverse Laplace transform of R(s) = s2 + 1 . When we successively substitute the three zeros s = 0. + s+2 + s4' 2 and therefore £ I I s2 + 1 } =+e 1 5 2t +17 4t e . and s = 4 of the 8A Thus A = 1.8s Solution Note that the denominator of R(s) factors as Q(s) = s(s + 2)(s. and C . s3 (s + 2)2 s3 s2 s (s + 2)2 s + 2 (9) To find A.+ . (10) . so s2 + 1 k 152 ~~ s3 .+ + . = k.8s 8 12 24 • Example 3 illustrates a differentiation technique for finding the partial fraction coefficients in the case of repeated linear factors.4 Multiplication of each term of this equation by Q(s) yields s 2 + 1 = A(s + 2)(s. y(O) = y' (O) = 0.+ .4. Example 3 Solve the initial value problem y" + 4y' + 4y = t2. s 3 . and C = ~~.2s 2 . we get the results = 2.
(11) yields B =!. Substitution of s = 0 in Eq. The transformed equation is s 2X(s) + 6sX(s) + 34X(s) = s2 +4 = As+ B s2 + 4 60 Hence X(s) = 60 (s2 + 4)[(s + 3) 2 + 25] + Cs + D (s + 3)2 + 25 · .292 Chapter 4 Laplace Transform Methods where F(s) = D(s + 2). The initial value problem we need to solve is x" Solution + 6x' + 34x = 30 sin 2t.3 + Bs2 + cs.2 + E(s + 2).. we differentiate Eq.1 is the sum of the two partial fractions corresponding to (s + 2) 2 . but with initial conditions x(O) = x'(O) = 0 and with the imposed external force F(t) = 15 sin 2t. = ~ and E (14) Substitution of s = 2 in Eqs.4 (s + 2) = 2C + 6sF(s) + 6s 2F ' (s) + s 3 F"(s). (13) and (14) now yields D Thus 1 1 = ~. Example 4 Consider the massspringdashpot system as in Example 1. 5. we multiply each side in Eq. and then differentiate to obtain s 6 4 = E + 2(s + 2)G(s) + (s + 2) 2G'(s). Find the resulting transient motion and steady periodic motion of the mass. (13) where G(s) = As. (10) yields A = !· To find Band C. (9) by (s + 2)2 to get 2 3 = D s + E(s + 2) + (s + 2) 2G(s). Y(s) = 2 _ 2 s3 s2 + ]_ _ s 3 1 3 __ 8_ (s + 2) 2 s + 2' 4 so the solution of the given initial value problem is • Examples 4. (10) twice to obtain . x(O) = x' (0) = 0. and 6 illustrate techniques for dealing with quadratic factors in partial fraction decompositions.1.=3 (s 4 + 2) = B + 2Cs + 3s 2F(s) + s 3 F ' (s) (11) and 12 .and substitution of s = 0 in Eq. (12) yields C = ~· To find D and E. (12) Now substitution of s = 0 in Eq.
To find A and B.3. FIGURE 4. which we simplify to 60 = ( 24A + 30B) + (60A + 12B)i.5 29 + 3) . (s + 3)2 + 25 After we compute the inverse Laplace transforms. and we readily find their solution to be C = D = ~~. the result is (15) = 2i of the quadratic factor s 2 + 4 60 = (2i A+ B)[(2i + 3) 2 + 25]. our partial fractions decomposition of X (s) is 1 ( 1 Os X(s) = 29 + 50 1Os + 10 ) s2 + 4 + (s + 3) 2 + 25 s2 + 4 = _I (lOs + 25 · 2 + lO(s 0. 4. C. Again we equate real parts and imaginary parts. To find C and D. and D.12D) + (30C. we get 60 = (As+ B)[(s + 3) 2 + 25] + (Cs + D)(s 2 + 4). we get the position function . (15). • .12D = 60 and 30C . B. which are readily solved for A = .4 · 5) .3).3. With these values of the coefficients A.3.4. this yields the two linear equations 186C. we substitute the zeros in Eq. We now equate real parts and imaginary parts on each side of this equation to obtain the two linear equations 24A + 30B = 60 and 60A + 12B = 0. Note that the transient motion is nonzero even though both initial conditions are zero.30D)i.0.3 Translation and Partial Fractions 293 When we multiply both sides by the common denominator. which we simplify to 60 = (186C. we substitute the zeros = 3 + 5i of the quadratic factor (s + 3) 2 + 25 in Eq. damped transient motion Xtr(t). whereas the exponentially damped terms of circular frequency 5 constitute its transient motion. The periodic forced oscillation Xsp(t).5 x(t) = { 9 ( . which disappears very rapidly (see Fig.30D = 0. and solution x(t) = Xsp(t) + Xtr(t) in Example 4.3' (5 cos 5t  2 sin 5t). (15) and get 60 = [C( 3 + 5i) + D][( 3 + 5i) 2 + 4].~~ and B = ~~. The terms of circular frequency 2 constitute the steady periodic forced oscillation of the mass.2 cos 2t + 5 sin 2t) + i9 e.
Remark: The solution curve defined in Eq.1 { (s 2 ~(sinktktcoskt).w0 wo w Fowo (s 2 2 2' + w0 ) Fow w 2 . kt. a repeated quadratic factor ordinarily signals the phenomenon of resonance in an undamped mechanical or electrical system.1 t sm 2k = (16) £. 4. together with its envelope curves x = ±C(t) .3. we have X (s) = 4 so Eq. = . ) . Because of the presence in Eqs. The resonance solution in (18) with w0 = ~ and F0 = 1. we find without difficulty that X(s) = so it follows that x(t) = Fow ( . This technique for constructing envelope curves of resonance solutions is illustrated further in the application material for this section.294 Chapter 4 Laplace Transform Methods Resonance and Repeated Quadratic Factors The following two inverse Laplace transforms are useful in inverting partial fractions that correspond to the case of repeated quadratic factors: £ 1 { (s 2 + k 2)2 + k 2)2 1 s I } . 2k (17) These follow from Example 5 and Problem 31 of Section 4. we get the equation s X(s)+w0 X(s) 2 2 = Fow s + w 2 2. 2w0 (18) • 4 FIGURE 4. Example 5 Use Laplace transforms to solve the initial value problem x" 2 + w0 x 17 • = ro smwt. Solution When we transform the differential equation. But if w = w 0 . so X(s) = Fow (s 2 + w 2 )(s 2 + w 0 ) 2 .2. x(t) = . respectively.2 (sm wot . If w =!= w 0 .w~ ( s 2 + w~ 1  s 2 + w2 1 ) .4. (16) and (17) of the terms t sin kt and t cos kt. In this case we find that x(t) = A(t) coswot C(t) = Fo2 2w0 Jw~t2 + 1. x(O) = 0 = x '(O) that determines the undamped forced oscillations of a mass on a spring.4) between the "envelope curves" x = ±C(t) that are obtained by writing ( 18) in the form + B(t) sinwot and then defining the usual "amplitude" C = J A2 + B 2 . (18) bounces back and forth (see Fig. (17) yields the resonance solution Fo . • .wot cos wot).1 sinwot1 smwt .3. 2 2 w .
we substitute in succession the values s = 0. and F=l.1) 2 + Es(s. C = 2. s = 2. and the familiar transforms of cost and sin t.1) 2(s 2 + 1) = 4.+ 1) 2 s .90E + 45F = 21. 9D .(s.1) 2 (s 4 + 2s 2 + l)Y(s) = 4 (s.1)2 s .===. and F . and thus obtain Y(s) = (s  1 .l)(s 2 + 1) 2 + Cs(s . the translation property. • . C.3 Translation and Partial Fractions 295 Example 6 Solve the initial value problem yC4 ) + 2y" + y = 4te1 . This yields the system B 8B25B + 75B 200B + + 4C + 2C+ 18C + 12C + 3. (16). y(O) = y' (0) = y" (0) = yC3) (0) = 0.96 (21) of five linear equations in B. D. With the aid of a calculator programmed to solve linear systems. E. s = 1. 4D.. (20) Upon substituting s = 1 we find that A = 1. we get the equation A(s 2 + 1) 2 + B(s. D = 0. E = 2.1) 2(s 2 + 1) 2. We now substitute in Eq. (20).8£ + SF = D+ 10£ + 5F = 21. and s = 3 in Eq.1) 2(s 2 + 1) + F(s. 4D + 120£ + 40F = . we see finally that the solution of the given initial value problem is y(t) = (t  2)e1 + (t + 1) sint + 2cost . Equation (20) is an identity that holds for all values of s.1) 2 . (18) the coefficients we have found.1) 2(s 2 + 1)2 = A B Cs + D Es + F ++ +2 2 s2 + 1 · (s . s = 2.1 (s + 1) (19) If we multiply by the common denominator (s .1 (s 2 + 2 2s + s2 + 1· 1)2 2s + 1 Recalling Eq. Solution First we observe that and Hence the transformed equation is L{te1 } = 1 (s.1) 2 + D(s. To find the values of the remaining coefficients. we find that B = 2.4. D F= + 0. Thus our problem is to find the inverse transform of 4 y (s) .
f(t) = 3. + 6x' + 25x = 0. F(s) = s2 + 3 21. F(s) = s2 + 7s + 10 15. and F(t) = 40.2s s 4 + 5s +4 2 20.4s + 5) 2 = te. 7. 25. c 6e . to derive the inverse Laplace transforms listed in Problems 23 through 26. 35. f(t) = t3f2e4t 4. 4.1 { s 4 : 4a 4 24. F(s) = s2 + = 4s + 4 8. x(O) = x'(O) = x"(O) = I 11. 2s. £ . c = 0.1 (s + 1) 3 s+2 s2 + 4s + 5 2s. F(s) = = = s.4 I 6.6s + 25 = 3t.296 Chapter 4 Laplace Transform Methods II!] Problems Use Laplace transforms to solve the initial value problems in Problems 27 through 38. The graph of the damped oscillation in Problem 40. f(t) = t4errr 2.I2s + 20 32. x(O) = x'(O) = 0 x<3> + x".04.. x ' (O) = x"(O) = x < (0) = 0 3> x<4> + x = 0. x'(O) = x " (O) = I 3> x<4>.s2 . Suppose that m = I.4 5. x(O) = 0. 2 3 } = cosh at cos at L 4 : { 4a 4 } = 2 } 2~ 2 sinh at sin at _!__ (cosh a t sin at + sinh a t cos at) 2a 4a 2 FIGURE 4. Suppose that m 1 17.1 s + 4a 4 = 1 4 4} s +4a =~ 3 (coshatsinatsinhatcosat) . Construct a figure that illustrates the resonance that occurs.. F(s) 9" F(s) 3s + 5 s2 . F(s) = .4x 30. x " .s 2 4 = 1.4.6 12. Derive the solution x(t) (4s2. F(s) = .112 cos2 (t 28.115 is A(5) = 5/e.x = 0. F (s) 36. 1. x< 3>(0) =0 13. x<4> + 8x" x<3 > (0) = + 16x = 0.x(O) = 0. 33. £ .5. x(O) = x ' (O) = x"(O) = 0. x(O) = 1. F(s) = 1 3 2 = s3 . = e21 . j(t) = e. 3 5. £ .1 Show that the maximum value of the amplitude function A(t) = t e . x'(O) = x" .1 .. 31. 39. x ' (O) = 1 s .(s2 + 2s + 2) 2 Use the factorization 22... F(s) =.115 cos 3t.. x(O) = 1..3 9s2 . 27.6) 2 18. 29.3. = 0. k = 9.2s + 4x' + 13x = te. x(O) = x"(O) = 0. Thus (as indicated in Fig. x < (0) = 1 x<4> + 13x" + 36x = 0. x' (O) = 2 x " + 6x' + 18x = cos2t. 0.5) the oscillations of the mass increase in amplitude during the first 5 s before being damped out as t ~ +oo. x'(O) = x(3> (0) = 13 2. x<4>+2x" +x 37.21 sin 3m e. and F(t) = 6cos3t. x(O) = x ' (O) = 0 x " + 4x' + 8x = e.115 sin3t. F(s) = s2.3s 5s 4 14. £ . F(s) =s2 . F(s) = s4 ___ 1_6 = s3 (s _ 4 )4 19. Use the inverse transform given in Eq.6x' + 8x = 2. x(O) = x'(O) = x" (O) = 1.1 { 4 s 26. 10" F(s) Use partial fractions to find the inverse Laplace transforms of the functions in Problems 11 through 22.6x' = 0. 1 8 2 6 s . (16) to derive the solution x(t) = t sin 3t.s +1 2s 3 .3. x" ~n) Apply the translation theorem to find the Laplace transforms of the functions in Problems 1 through 4. F(s) Problems 39 and 40 illustrate two types of resonance in a massspringdashpot system with given external force F(t) and with the initial conditions x (0) = x' (0) = 0. 23.2s 1 5s. x(O) = x' (O) = 0 3 Apply the translation theorem to find the inverse Laplace transforms of the functions in Problems 5 through 10. x " 38. k = 9. x(O) = 2. F(s) = s 2 ..5s I 16. 34. F(s) = .(s2 + s.
r ) d r = 1° f(t .r) dr (1) has the desired property that £{h(t)} = H(s) = F(s) · G(s). Thus the convolution is commutative: f *g = g *f .u)g(u)( .4. (3) We will also write f(t) * g (t) when convenient. In terms of the convolution product.du) g(u)f(t. we get · 2 1 = cl {cos t} · cl {sin t}. Theorem 1 of this section will tell us that the function h(t) = lr f(r)g(t . DEFINITION The Convolution of Two Functions The convolution f * g of the piecewise continuous functions for t ~ 0 as follows: (f f and g is defined * g)(t) = 1 1 f(r)g(t. when we transform the initial value problem x" + x =cost. But obviously x(t) is not simply the product of cost and sin t. It is denoted by f * g. x(O) = x ' (O) = 0. Thus cl {cos t sin t} =1= £ {cos t} · £ {sin t}. . because X (s) = (s 2 + 1)2 s = £ {cos t sin t} = £ Bsin 2t } = s 2 +4 1 =I= (s 2 + 1)2 s . we see that f (t) * g (t) = lt = lt f (r )g (t .r) dr. Integrals. For example. Theorem 1 of this section says that cl{f * g} = cl{f}. and Products of Transforms 297 The Laplace transform of the (initially unknown) solution of a differential equation is sometimes recognizable as the product of the transforms of two known functions. (2) The new function oft defined as the integral in (1) depends only on f and g and is called the convolution of f and g. the idea being that it is a new type of product of f and g.r in the integral in (3). £{g }. 2s s +1 s +1 This strongly suggests that there ought to be a way of combining the two functions sin t and cost to obtain a function x(t) whose transform is the product of their transforms. so tailored that its transform is the product of the transforms of f and g. If we make the substitution u = t .u) du = g(t) * f(t).4 Derivatives.
r) dr. (5') Example 2 illustrates the fact that convolution often provides a convenient alternative to the use of partial fractions for finding inverse transforms. convolution yields L.l)(s 2 + 4) } = (sin2t) * et = r et.···· Example 2 With j(t) =sin 2t and g(t) = e 1 . provided that we can evaluate the integral L. (cost) * (sin t) = 4t sin t.I 2 (s .l)(s 2 + 4) l 2 et ...298 Chapter 4 Laplace Transform Methods Example 1 The convolution of cost and sin t is (cost) * (sin t) = We apply the trigonometric identity cos A sin B = to obtain (cost)* (sin t) it cos r sin(t .2cos2r) t [ Jr .2 that the Laplace transform of is indeed sj(s 2 + 1)2 • Theorem 1 is proved at the end of this section.t)] dr 1 . THEOREM 1 The Convolution Property Suppose that f(t) and g(t) are piecewise continuous fort ~ 0 and that lf(t)l and !g(t)l are bounded by Meet as t+ +oo.{g(t)} (4) and L. L{j(t) * g(t)} = L{j(t)} · cf.1 { 2 (s . 4[sin(A +B). = 5 5 5 • . moreover.B)] 4[sin t sin(2r .1 sm . 2t . 4t sin t • And we recall from Example 5 of Section 4. =it = 4[r sin t + 4cos(2r  t)J r=O that is.1{F(s) · G(s)} =it lo f(r)g(t . ··· .r = e 1 Jo e .1{F(s) · G(s)} = f(t) * g(t).. 0 so cL.r)dr.2 cos 2t.r sin2r dr = et e 5 (. Then the Laplace transform of the convolution j(t) * g(t) exists for s > c. _ .sin(A. (5) Thus we can find the inverse transform of the product F(s) · G(s).r sin2rdr .sin2r.
Equation (8) gives cl{t 2 sinkt} = (1) 2::2 Solution ( 52 ~ k2) J (9) d [ 2ks 6ks 2 .1(1 / s)}. Example 4 Find cl. (7): cl 1 { Solution tan 1 s 1} = cl 1 t 1 { d .. 2. Integrals. The derivative oftan..1tan ds s 1} .t. if f(O) = 0 then differentiation of f(t) corresponds to multiplication of its transform by s. • The form of the differentiation property in Eq.4.2. Example 3 Find i{ [2sin kt}. (7) is often helpful in finding an inverse transform when the derivative of the transform is easier to work with than the transform itself. . and Products of Transforms 299 Differentiation of Transforms According to Theorem 1 of Section 4..4 Derivatives. 3. Theorem 2. so we apply Eq.2k 3 = ds (s 2 + k2)2 = (s 2 + k2)3 . (6) gives (8) for n = 1. then cl{tf(t)} = F'(s) for s > c.1 (1/s) is a simple rational function. (6) (7) Repeated application of Eq. Equivalently. THEOREM 2 Differentiation of Transforms If j(t) is piecewise continuous fort ~ 0 and lf(t)l ~Meet as t ~ +oo.T{tan. tells us that differentiation of the transform F (s) corresponds to multiplication of the original function f (t) by . proved at the end of this section.
rather than constant.[s 2 X(s). tx" +x' +tx such that x(O) = 1 and x' (0) denoted by J0 (t). This equation is separableX' (s) X(s) s s2 + 1' its general solution is X(s) = c . sin t t • Equation (8) can be applied to transform a linear differential equation having polynomial. The result of differentiation and simplification is the differential equation (s 2 + 1) X' (s) + sX(s) = 0. and because x and x" are each multiplied by t. 1 tan _ 1  L _1 { s I= . = 0.1 and L{x"(t)} = s 2 X(s). application of Eq. This solution of Bessel's equation is customarily = sX(s). it follows that 1 L{Jo(t)} = Js 2 + 1 . coefficients. Example 5 Let x(t) be the solution of Bessel's equation of order zero.s.[X(s)] = ds . Because X (s) = L{Jo(t)}.. Because L{x '(t)} = 0.s] 0.300 Chapter 4 Laplace Transform Methods Therefore. (6) yields the transformed equation d ds d + [sX(s). of course. whether this procedure leads to success depends. (10) • . on whether we can solve the new equation more readily than the old one.1 ] . Js 2 + 1 In Problem 39 we outline the argument that C = 1. The result will be a differential equation involving the transform.
(13) is often helpful in finding an inverse transform when the indefinite integral of the transform is easier to handle than the transform itself.a 1 +1 1 ) da = 2 1[ In a (j  +1 1] 00 s Therefore.t)} = 1 00 F(a) da (12) (13) ············ ·········· Example 6 Solution Find £ {(sinh t )/t}.1 because ln 1 = 0. that is. that f(t) satisfies the condition in (11). with f(t) =sinh t. 0 from the right. (12). Integrals. Then £ { for s > c.. t (11) THEOREM 3 Integration of Transforms Suppose that f(t) is piecewise continuous fort ~ 0.. £ { sinh t } = t ~ ln s + 1 ' 2 s.1.= lim = lim = 1. We first verify that the condition in (11) holds: sinh t e1 . provided that the resulting quotient f(t)/t remains well behaved as t ___. Example 7 . yields £ { sinht} = t 1oo £{sinht}da = 1oo :: 1 s s (j = 1 2 1 s 00 ( a . Equivalently.e. and that lf(i)l ~Meet as t ___.1 e1 + e1 lim . provided that lim f(t) t+O+ exists and is finite.4. t + 0 t t + 0 2t t + 0 2 with the aid of l'H6pital's rule.4 Derivatives. proved at the end of this section. and Products of Transforms 301 Integration of Transforms Differentiation of F(s) corresponds to multiplication of j(t) by t (together with a change of sign).. confirms this. Then Eq. f.. It is therefore natural to expect that integration of F(s) will correspond to division of f (t) by t. Theorem 3. • The form of the integration property in Eq.. +oo.
(The proof of this requires a discussion of uniform convergence of improper integrals. (13).st [f(t) * g(t)] dt .r) = 0 whenever r > t.st g(t . but it is much simpler to apply Eq.:_1 { 2s } .1)2 00 } a2 . Now our hypotheses on f and g imply that the order of integration may be reversed. 00 e .1 . This gives . and can be found in Chapter 2 of Churchill's Operational Mathematics.st f(r)g(t.1.1)2 2s l t sinh t.r) dt (u = t .r) dr) dt 00 e .1 s t . because we may define f(t) and g(t) to be zero fort < 0.:_1 { {loo 2a da} 1{[ 1 ] = 1{ s (a2.st (lot f(r)g(t.r) dt. For any r > 0 the definition of the Laplace transform gives G(s) = 1 00 esu g(u) du = and therefore G(s) =eST 1 1 00 es(tr) g(t .1)2  = t . F(s)G(s) = L{f(t) * g(t)}. (New York: McGrawHill. 1} (s2. • .:_ and therefore .302 Chapter 4 Laplace Transform Methods Solution We could use partial fractions. and therefore. 3rd ed.r) dr) dt = = 1 1 00 e . 1972). Then F(s)G(s) = G(s) = 1 00 e ST f(r) dr = 1 00 eST j(r)G(s) dr fooo esr f(r) (eST fooo est g(t  r) dt) dr = 1 (1 00 00 e .r) dt) dr. This completes the proof of Theorem 1.r).tL1 (s2.:_ s2 . • *Proofs of Theorems Proof of Theorem 1: The transforms F (s) and G (s) exist when s > c by Theorem 2 of Section 4.) Hence F(s)G(s) = 1 (1 00 00 est j(r)g(t . We replace the upper limit of the inner integral with t because g (t.
. which is Eq. The validity of differentiation under the integral sign depends on uniform convergence of the resulting integral. f(t) = t.t.b) = t.A. f(t) Problen1s Find the convolution f(t) * g(t) in Problems 1 through 6. f(t) 3. F(a) = fooo e . Integrals. So integration of F(a) from s to +oo gives Under the hypotheses of the theorem.at f(t) dt. BJ 1. We obtain Eq. j(t) 5.1 and then multiplying by t. the order of integration may be reversed (see Churchill's book once again). it follows that at ]oo = oo [ e_ 0 1 f(t)dt t a=s This verifies Eq. g(t) = cost 6. . (7) by applying el .A. and Products of Transforms 303 Proof of Theorem 2: Because F(s) = fooo est f(t) dt. 2. g(t) = eb 1 (a f=. Proof of Theorem 3: By definition. and Eq.4 Derivatives. g(t) = 1 = g(t) = sin t = g(t) = eat Apply the convolution theorem to find the inverse Laplace transforms of the functions in Problems 7 through 14.1 and then dividing by . F'(s) = el{tf(t)}. g(t) =eat 4. (13) follows upon first applying el.4. f(t) =eat. . (6). this is discussed in Chapter 2 of the book by Churchill just mentioned. differentiation under the integral sign yields F(s) = d ds o 00 1 d ds 00 est f(t) dt = thus 1 o [est f(t)] dt = 100 est [ o tf(t)] dt. (12). f(t) = t 2 .
. Termwise Inverse Transformation of Series In Chapter 2 of Churchill's Operational Mathematics. f(t) = tsin3t 17. F(s) = .4)x = 0 Apply the convolution theorem to show that £I { 1 (s 1) Js} = 2~ r~~ s vn Jo = .. f(t) = te.304 7. x(t) f(t).t et.1{ eu2 du = elerf. F(s) Chapter 4 Laplace Transform Methods = s(s 3) 1 8.. f(t) = te 21 cos 3t sin t 19. Mathematical models of mechanical or electrical systems often involve functions with discontinuities corresponding to external forces that are turned abruptly on or off.s(s2 + 4s + 5) 14. F(s).1 sin 2 t f 1 .et 22. 35.r)e. 40. apply either Theorem 2 or Theorem 3 to find the Laplace transform of f(t). transfo rm the given differential equation to find a nontrivial solution such that x(O) = 0. F (s) + 4x' + 13x = =3 = .2)x' + x = 0 tx" + (3t.t F(s) = L oo a n+:+l n=O S Find the inverse transforms of the functions in Problems 23 through 28. that f(t) is of exponential order as t + +oo. F(s) = tan. One such simple on. (t) = .j1ri (Suggestion: Substitute u In Problems 36 through 38.1 21. 3 26. F(s) = (s2 + 4)2 s 13.off function is the unit step function that we introduced in ..1)x' . f(t) = t e3t. apply the convolution theorem to derive the indicated solution x (t) of the given differential equation with initial conditions x (0) = x' (0) = 0.x(t) = 2 1 9. f(t) = . Expand with the aid of the binomial series and then compute the inverse transformation term by term to obtain 29. F(s) = (s. f(t) = . x" + 2x' + x 38.Jt .2)x ' + (13t . 33. and that 15. Expand the function F(s) = s.. F(s) =In .1)x' + 3x = 0 tx". Then s2 23. F(s) = (s2 + 9)2 s2 11. x" = f(t).112 e . note that 10 (0) = 1 implies that C = 1.1 . 32. 34. 31. f(t) = t 2 cos2t 18.t 16.3) 27. t x" + (t.Jt.r) dr 1 12.(4t + 1)x' + 2(2t + 1)x = 0 tx" + 2(t . In Example 5 it was shown that £{Jo(t)} = Js2+T = s2 + 1 c c ( 1 + 21 )l/2 s s In Problems 29 through 34. the following theorem is proved. x" +4x = f(t). 41. F(s) =In ( 1 + s1 2) s2 + 1 24.cos 2t 20.2' sin 3r dr In Problems 15 through 22.s(s2 + 4) 1 1 36. where 0 ~ k < 1 and the series converges absolutely for s >c. F(s) = . F(s) =In (s + 2)(s .2 s +4 f(t) = ~ f(n + k + 1) · 00 antn+k Apply this result in Problems 39 through 4I. 11/ =1 0 1 f(t.s+2 s 28. Suppose that f(t) is continuous for t ~ 0.:s4 + 5s 2 + 4 x(t) 11 0 1 f(t.r)sin2rdr re• f(t. F(s) 10.) Finally.1 to show that £ .2x = 0 tx"2x' +tx=0 t x " + (4t . 30. F(s) =Ins +2 s2 + 1 25..I f s in powers of s .3)(s2 + 1) =s2(s2 + k2) s 37.(s2 + 1)3 39. Show that _1_elfs}= Js _ 1_ COS 2Jf.
. Theorem 1 tells us that this fact.1 we saw that if a .1 {e .a) dt. . when properly interpreted.st f(t . then £{u(t. Then u ( t . if t ~ a.5. so it does not start arriving until time t = a.a)}= . Translation of f(t) a units to the right.a) f (t . Note that u(t. (2) s Because £{u(t)} = ljs. (2) implies that multiplication of the transform ofu(t) byeas corresponds to the translation t + t a in the original independent variable.. 4.as F(s) = 1 00 e . Note that the part (if any) of the graph of f(t) to the left oft = 0 is "cut off" and is not translated (Fig. 0. then eas £{u(t. The notation Ua (t) indicates succinctly where the unit upward step in value takes place (Fig. In some applications the function f(t) describes an incoming signal that starts arriving at time t = 0.a) if t < a.1). a is the translation by a units to the right of the graph of f (t) for t . we get a FIGURE 4.2). Recall that the unit step function at t = a is defined by X Ua(t) = u(t a) = a 0 1 1 if t <a.a)f(t.5 Periodic and Piecewise Continuous Input Functions 305 Section 4.a)f(t . 0..1 {eas F(s)} is the function whose graph fort.1...a) connotes the sometimes useful idea of a "time delay" a before the step is made.a) (3b) for s > c +a.. (1) FIGURE 4.as F(s)} = u(t. > 1ft_ a.5.4.5.a) = IOf(t.5. (4) Thus Theorem 1 implies that £ . Proof of Theorem 1: From the definition of £{f(t)}.a) denotes a signal ofthe same "shape" but with a time delay of a.2.1. Eq. The substitution t = r + a then yields e .a)}= eas F(s) (3a) and £ . In Example 8 of Section 4.a)f(t.. The graph of the unit step function at t =a. is a general property of the Laplace transformation. . THEOREM 1 Translation on the tAxis If £{f(t)} exists for s > c. whereas u(t. 4.
c. 4.a)f(t.4). • Example 2 Find £{g(t)} if g(t) = {~2 if t < 3.a)(t.306 Chapter 4 Laplace Transform Methods From Eq. if t ~a (Fig.5. (4) we see that this is the same as eas F(s) = fooo estu(t.a)f(t .3).3). Example 1 With f(t) = < a. 4. The function f(t) whose translation 3 units to the right agrees (for t ~ 3) with g(t) = t 2 is f(t) = (t + 3) 2 because f(t .5). because u(t . 4.a)f(t.af if t <a. But then F(s) = £{t 2 + 6t + 9} = 2 3 s +2 + .3.5.3) = t 2 . s s 6 9 so now Theorem 1 yields £{g(t)} = £{u(t.. Solution Before applying Theorem 1. 4t 2. X X 20 15 3n: t 10 5 a 2 3 4 FIGURE 4.3)} = e.3 s F(s) = e.5. FIGURE 4.3s 9) . The graph of the function g(t) of Example 2.4. we must first write g(t) in the form u(t.5.5.3)f(t .5.l {  1 {0 2 !Ct. The graph of the inverse transform of Example 1.a) 2 = eas} .. • (3s2+ s6+s 2 Example 3 Find £{f(t)} if j(t) = {~OS 21 if 0 ~ t < 2rr . FIGURE 4. The function f (t) of Examples 3 and 4 .5.a) dt = 0 for t £{u(t.a)}. Theorem 1 gives s3 = u(t.3)f(t. if t ~ 2rr (Fig. if t ~ 3 (Fig. This completes the proof of ..a) = Theorem 1.
2t = 4 by Eq.2n)] sin 2t.4. 4.6. x(O) = x'(O) = 0. but at timet = 2n this force is turned off (abruptly discontinued) and the mass is allowed to continue its motion unimpeded. The graph of the function x(t) of Example 4. As indicated by the graph of x(t) shown in Fig. but we see that its effect ceases immediately at the moment it is turned off.5.2n) sin2(t.2n) because of the periodicity of the cosine function. the mass oscillates with circular frequency w = 2 and with linearly increasing amplitude until the force is removed at time t = 2n.2n)] cos 2t =cos 2t . Hence Theorem 1 gives cl(f(t)} = cl{cos2t} e.6.u(t. Beginning at time t = 0 (seconds). light spring that is stretched 1 ft by a force of 4 lb (k = 4lb/ft).2n) = ~ [t .5.2rrscl{cos2t} = s(l _ e2rrs) .5 Periodic and Piecewise Continuous Input Functions Solution 307 We note first that f(t) = [1 . s Because oC 1 { (s 2 + 4) 2 S } I t sm . The transformed equation is (s 2 + 4)X(s) so = F(s) = s(l _ e2rrs) . Find the resulting position function x(t) of the mass. The force F(t) = cos 2t would produce pure resonance if continued indefinitely.u(t. Solution where f(t) is the function of Example 3. The mass is initially at rest in its equilibrium position. the mass continues to oscillate with the same frequency but with constant amplitude n/2. we find that the position function may x(t) = { jt sin 2t ~n sin 2t if t < 2n.2n) cos 2(t. 2 s +4 s X(s) = (s2 + 4)2  e . FIGURE 4. If we separate the cases t < 2n and t be written in the form X=rr:/2 ~ 2n. • .u(t. (16) of Section 4. 2 s +4 • Example 4 A mass that weighs 32 lb (mass m = 1 slug) is attached to the free end of a long. if t 2:: 2n.2n) · j(t. Thereafter. it follows from Theorem 1 that x(t) = ~t sin2t. We need to solve the initial value problem x" + 4x = f(t).2n) · (t .u(t.3.2rrs (s2 + 4)2. an external force f(t) =cos 2t is applied to the mass.
Thus e'(t) appears to have an infinite discontinuity at t = 1.5. and a battery supplying Eo = 90 V. With the given circuit elements.u(t .7 our strategy was to differentiate both sides of Eq. I Here we do not use that method. corresponding to the opening and closing of the switch. 1 e(t). (8) We substitute Eq. (6) FIGURE 4.7 the basic series circuit equation di Ldt Solution + Rl + q = c . with R = 110 Q. For now. Example 5 Consider the RLC circuit shown in Fig. (5). + R+l = dt 2 dt c e (t). The Laplace transform method works well with such an equation. Eq. C = 0.001 F. where e(t) = 90[1 . (7) yield.£ t < 2n and then solve a new problem with different initial conditions for the interval t ~ 2n. 4. . q(t) =loti c 0 (r) dr . it involves both the integral and the derivative of the unknown function i (t). and voltage and reserve uppercase letters for their transforms. This phenomenon will be discussed in Section 4. At time t = 1 it is opened and left open thereafter. upon integration. The series RLC circuit of Example 5. whereas the jump from e(t) = 90 when t < 1 to e(t) = 0 when t > 1 would seem to require that e'(l) = oo. (5) we use lowercase letters for current. In Section 2. we observe that the initial value q(O) = 0 and Eq. = dq dt (7) to obtain the secondorder equation d2i Ldi 1. L = 1 H. At time t = 0 the switch is closed and left closed for 1 second.7. charge. (8) in Eq. we would need to solve one problem for the interval 0 . (5) is R + llOi + 1000q = dt di e(t). (9) This is the integrodifferential equation of a series RLC circuit.7. (5) to obtain di Ldt 1 + Ri + 1t i(r) dr = e(t). we will simply note that it is an odd situation and circumvent it rather than attempt to deal with it here.1)]. We recall from Section 2.308 Chapter 4 Laplace Transform Methods If we were to attack Example 4 with the methods of Chapter 2.5. Find the resulting current in the circuit. then apply the relation l .6. because e' (t) = 0 except at t = 1. In such a situation the Laplace transform method enjoys the distinct advantage of not requiring the solution of different problems on different intervals. Initially there is no current in the circuit and no charge on the capacitor. To avoid the possible problem at t = 1.
10t _ e . if t 2: 1.e5 ). The portion e . (10) Because £ r i(r) dr I {Jo =  I (s) 8 by Theorem 2 of Section 4.5 Periodic and Piecewise Continuous Input Functions 309 In the present example.2 on transforms of integrals.lOOt of the solution would describe the current if the switch • were left closed for all t rather than being open for t ~ 1.lOOt _ u(t _ 1) [ e .4.= s I (s) (1. .u(t.1) ] .5 ) I(s) = s 2 + llOs + 1000· But 90 1 s s 2 + 11 Os + 1000 so we have + 10 1 s + 100 ' I (s) = s + 10  1 s + 100  1 _5 e ( 1 s + 10 . the transformed equation is si(s) + llOI(s) + 1000. thus the inverse transform is e .IOO(t .1) _ e . After we separate the cases t < 1 and t given by ~ 1. Eq.e . 90 s We solve this equation for I (s) to obtain 90(1 . we find that the current in the circuit is i (t) = { ee lO t .s + 100 1 ) · We now apply Theorem 1 with f(t) = i(t) = eIOt  e.1)].1001 .lOt IO(t1 ) + e IOO(t 1) if t < 1.e.101 .ee lOOt e lOOt .lO(t . (9) is di + llOi + 1000 dt 1t 0 i(r)dr = 90[1.
·. Then the dt. .310 Chapter 4 Laplace Transform Methods Transforms of Periodic Functions Periodic forcing functions in practical mechanical or electrical systems often are more complicated than pure sines or cosines.eps 1p est f(t) O 0.5.8.st f(t) np dt. F(s) = 1 fP esr f(r) dr.= 1+x+x2 +x 3 + . n=O The substitution t (n+l)p l np est = r + np in the nth integral following the summation sign yields 1p 0 f(t) dt = es(r+np) f(r + np) dr = enps 1p 0 esr f(r) dr because f(r + np) = f(r) by periodicity.2ps + · · ·) 1P esr f(r) dr. (11) holds is called the period of f. Such a function is shown in Fig. The least positive value of p (if any) for which Eq. THEOREM 2 Transforms of Periodic Functions ~ Let f(t) be periodic with period p and piecewise continuous fort transform F(s) = £{f(t)} exists for s > 0 and is given by F(s) = 1 1.A. The graph of a function with period p. (12).ps < 1 (for s > 0) to sum the series in the final step. (e•P' f e" 1 eps f(r)dr) = (1 + eps + e.. 4. Thus we have derived Eq. Consequently. The principal advantage of Theorem 2 is that it enables us to find the Laplace transform of a periodic function without the necessity of an explicit evaluation of an improper integral. (12) Proof: The definition of the Laplace transform gives F(s) = 1 0 oo e .8. 1x with x = e . . Thus F(s) = t. The nonconstant function f (t) defined for t ~ 0 is said to be periodic if there is a number p > 0 such that f(t t=p=J I I I I + p) = f(t) (11) FIGURE 4. for all t ~ 0. Jo We use the geometric series 1.st f(t) dt = L oo l(n+l)p e .5. Theorem 2 simplifies the computation of the Laplace transform of a periodic function.
e2as) Therefore.9 shows the graph of the squarewave function f(t) = ( 1)ITt/aD of period p = 2a. eas j2. The transformed equation is s 2 X(s) +4sX(s) +20X(s) = F(s). 4.. Suppose that the system is initially at rest at equilibrium (x (0) = x' (0) = 0) and that the mass is acted on beginning at time t = 0 by the external force f (t) whose graph is shown in Fig. s(l + e .11: the square wave with amplitude 20 and period 2n. Because the derivative g'(t) is the square wave function of Example 6..~ .···" Figure 4. and k = 20 in appropriate units.spring. By Theorem 2 the Laplace transform of f (t) is F(s) = 1 2a I 1 e.s(l + e .as) (13a) = 1 as =tanhs(eas/ 2 + eas/2) s 2 .eas)2 1 .9.s Jo {la est f(t) dt FIGURE 4. The graph of the triangularwave function of Example 7.eas s(1 .. The graph of the squarewave function of Example 6. 1 ( {a est dt 1 . [x] denotes the greatest integer not exceeding x. The initial value problem is x" + 4x' + 20x = f(t). Example 7 g(t)  ~. • Example 8 Consider a mass.eas F(s).5.5. it follows from the formula in ( 13b) and Theorem 2 of Section 4.e2as lo + 12a ( l)est dt) a (1 .2 that the transform of this triangularwave function is F(s) 1 as G(s) ==tanh.5. Find the position function f (t).10.as) 1 .5 Periodic and Piecewise Continuous Input Functions 311 Example 6 /(t) Figure 4.5. (15) . s s2 2 a 2a 3a 4a 5a 6a (14) FIGURE 4.dashpot system with m 1.10 shows the graph of a triangularwave function g(t) of period p = 2a.4.5.easj2 (13b) • ~'. Solution x(O) = x'(O) = 0. c = 4.
11. (18) where h(t) = e.e3JT:s + · · ·) = n: I I I I (1  2ens + 2e. (17).2r (cos4t r Using a tabulated formula for J e ar sin bt d t. 00 (16) F1GURE 4.e. (8) of Section 4. S n=l Substitution ofEq.h(t).. 2n: 3n: 4n: 5n: 6n: I I I I I I I I I I I I I I I I I I I I so that F(s) = S 20 ~ ~ ~ 20 40 +L(1tenns.=_F_(_s) __ . The result is 00 x (t) = g (t) + 2 L(1tu(t.2 (t. (19) Now we apply Theorem 1 to find the inverse transform of the righthand term in Eq. 2 2) + 16] e. g(t. .nn ) [cos4(t.2 we have g(t) =£I f s[(s + 20 1= Jo se2' sin4r dr.nJT) + ~ sin4(t. ).n.3.n.nJT). Moreover. (16) in Eq.r)] = 1 e 2nne. (15) yields X (s) . we get g(t) = 1 + ~ sin4t) = 1.2r (cos4t +! sin4t).r)g(t. n=l (20) and we note that for any fixed value oft the sum in Eq.312 Chapter 4 Laplace Transform Methods From Example 6 with a = JT we see that the transform of f(t) is 20 1 ens 1 +ens F(s)=· s s s fit) =  20 20 (1 ens) (1 ens+ e2JT:s . we get £.5.21 (cos 4t + ~ sin4t).1 { (s + 2) 2 + 16 20 } = Se. (20) is finite.nJT) = 1.2r sin 4t ' so by Theorem 2 of Section 4..2ns  2e3JT:s + . (17) From the transform in Eq.s 2 +4s + 20 20 = s[(s + 2) 2 + 16] + 2 L(l)n n= l 00 20enns s[(s + 2) 2 + 16] . The graph of the externalforce function of Example 8.
mr) = 1 . Then 2e2rr Xsp(t) =(It [ 1.nrr fort in the interval nrr < t < (n + 1)rr.2r cos(4r. If2rr < t < 3rr. The general expression fox: nrr < t < (n + 1)rr is x(t) = h(t) + (l)n .2t cos(4t. (24) The last two terms in Eq. ( l)ne2rr _ e2(tnrr) (cos 4t + l sin 4t) e2rr + 1 2 e2rr  1 (23) for nrr < t < (n + 1)rr. (22) finally gives.0.21 (cos4t + ~ sin4t) ~ (1.2319)e. Its most interesting feature is the appearance of periodically damped oscillations with a frequency four times that of the imposed force f(t).e 2rr h(t)] = 1 + h(t).12 shows the graph of X sp(t).2t (cos 4t + 2 sin 4t) + ( 1 t 2 err+ 1 2 .e2rr h(t)] + 2 [1. The first term in Eq.e 2nrr h(t).2 [1. x (t) = 1 e . (23) is the transient solution Xtr(t) ~ (0.2r (cos4r + e rr+ 1 1sin4r) ] (25) ~ (l)n [1.0. then x(t) = [1 . then x(t) = 1 . Figure 4.2h(t) [1.5. (21) Hence if 0 < t < rr. • .h(t).e4 rr h(t)] = 1 + h(t).h(t)]. In Chapter 8 (Fourier Series) we will see why a periodic external force sometimes excites oscillations at a higher frequency than the imposed frequency.2h(t) [1.4636)]. g(t.1139)e. with the aid of Eq. then x(t) = [1. (19).e2rr].h(t)]. If rr < t < 2rr. (22) which we obtained with the aid of the familiar formula for the sum of a finite geometric progression.e2rr + · · · + (lte2nrr] = h(t) + (It  1 + ( 1te2(n+l)rr 2h(t) 2 1+err .(2.5 Periodic and Piecewise Continuous Input Functions 313 Therefore.4.4636).2 e. A rearrangement of Eq.2 [1.e 2rr + e4rr].9963)e. (23) give the steady periodic solution Xsp· To investigate it. we writer = t .2h(t) [1.
2JTs 7. . f(t) = tift .1 ses s+2 eJTs 5. liD Prohl~~s 23. f(t) = 0 if t < 1 or if t > 2 1 eas F(s) = .5.2s s. f(t) = sinrrt if2 ~ t ~ 3. 25. e3s 1.14 is 11.5. 4.5. f(t) = 0 if t < 3 or if t > 5 19. Apply Theorem 2 to show that the Laplace transform of the squarewave function of Fig.s _ e2. f(t) = sint ifO ~ t ~ 2rr . F(s) = s2 + 1 £{f(t)) = s(l + eas) ~ 1 s 2 + 7r 2 2 1 .3s s2 e. F(s) 4. F(s) 8. F(s) = . f(t) = Oift > 3rr 16. f(t) = 1 if 1 ~ t ~ 4. F(s) = .f(t)=2tif1 ~t~2. Then sketch the graph of f.23)e Z(I n) I I I / I I I I I . f(t) = sint ifO ~ t ~ 3rr. 26. The graph of the sawtooth function of Problem 26.e. f (t) = t if t ~ 1.14. 24. FIGURE 4.(2. f(t) = 0 if t > 2 15. f(t) = 0 if t < rr or if t > 2rr 17.s e . e . f(t) = t 3 if 1 ~ t ~ 2. note the "periodically damped" oscillations with frequency four times that of the imposed force. F(s) = s 2 + 1 s(l . f(t) = 0 if t ~ 3 12..e2JTs) s2 + 4 2a 3a 4a Sa 6a Find the Laplace transforms of the functions given in Problems 11 through 22.s 3.e . The graph of the steady periodic solution for Example 8. Apply Theorem 2 with p = 1 to verify that £ {1} = 1Is. f(t) = Oift < 1.13 is Find the inverse Laplace transform f (t) of each function given in Problems 1 through 10.5. 4. f(t) = 2 if 0 ~ t < 3.f(t)=Oif t > 2 22.314 Chapter 4 Laplace Transform Methods I \ \ / .5. Apply Theorem 2 to verify that £{coskt) = s/(s 2 + k 2). f (t) = 1 if t > 1 21. f(t)=tift ~ 1. F(s) = = = = = e. F(s) = s(l + e3s) + 7r 2 s + 7r 2 a 11 1 I 11 I I I s 2 10.2 s 2.as ) j(t) a 2a 3a 4a Sa 6a FIGURE 4. as 2 s(l . The graph of the squarewave function of Problem 25. F(s) 2s(eJTs . Apply Theorem 2 to show that the Laplace transform of the sawtooth function f(t) of Fig.13. f(t) = Oift < 2 orift > 3 18.. F(s) 6. f(t) = Oift > 2rr 14.j 1 I + (2.23)e 2(1 n) FIGURE 4. f(t) = 0 if t < 1 or if t > 4 13. f(t) = cosrrt ifO ~ t ~ 2. f(t) =cos ~m if 3 ~ t ~ 5. f(t) = sin2t if rr ~ t ~ 2rr. 1 20.e 2s ) 9.12.
Under the assumption that x (0) = x' (0) = 0. shown in Fig. The halfwave rectification of sin kt.J I I . . 42.6rr)). C = 104 . where f(t) is the function of Problem 29 and k > 0. R = 0. C = 4 x 104 .J . The fullwave rectification of sin kt.errsfk). C = 104 . e(t) = 100sin lOt if 0 ~ t < n.5. c = t ~ 2 33.5. C = 10.5. T 31t = 0. dashpot resistance c. c = 0.. m = 1. L 37.14. Solve the initial value problem mx" +ex'+ kx = f(t). 39.spring. 4..4. Note that h(t) = f(t) + g(t) is the fullwave rectification of sinkt shown in Fig. C = 2x 104 .n/k)f(t . In Problems 31 through 35.eas) and construct the graph of the position function x(t). Let g(t) = u(t .15. R = 100. x(O) = x'(O) = 0 cl{g(t)} g(t) = s(1. f(t) = 1 if 0 t~Jr ~ ~ t < n. 6n] with the squarewave function that has amplitude A and period 2rr. If you would like to check your graph using a numerical DE solver.. and so forth. Let g(t) be the staircase function of Fig. c = 2. m = 1. e(t) = 0 if t ~ 1 L = 1.2n)(t. e(t) =lOOt ifO ~ t < 1.5. Suppose that f(t) is the halfwave rectification of sinkt. f(t) 4. f(t) = 0 if .3n) · cl{h(t)} = k s +k 2 2 coth  J<S 2k . f(t) = sint if 0 ~ t ~ 2rr. e(t) = 0 ift ~ n L = 1.17. R = 100.16. m = 1. 38. Show that In Problems 36 through 40. i(O) =0 0 with the given impressed voltage e(t). R = 150. e(t) = 0 if t ~ 1 FIGURE 4.J 32.17. e(t) = 100 if 0 ~ t < 2rr. m = 1. k = 4. 36.3 . and hence deduce that eas 315 massspringdashpot system with external forcing function. m = 1. The graph of the staircase function of Problem 27. k = 4. e (t) = 0 if t ~ 2n L = 1. Hence deduce from Problem 29 that In Problems 41 and 42. c = t :2: 1 35. Show that g(t) = (tja).5. f(t) = 1 ifO = 0. 29.. 4. (See a lso the definition of a squarewave function in terms of sawtooth and triangularwave functions in the application mate rial for this section.. 4. c = 0. e(t) = 100 if 0 ~ t < 1. e (t) = 0 if t ~ 1 L = 1. the values of mass m..f(t). cl{f(t)} = (s2 + k2)(1. 28. f(t) = 0 if t if 0 ~ t ~ 2. 0. f(t) = 0 if a 2a FIGURE 4. the values of the elements of an RLC circuit are given.dashpot system with external force f(t) is described. k = 9. k = 10.5 Periodic and Piecewise Continuous Input Functions 27.) FIGURE 4.njk).. f(t) = t if 0 ~ t = < 1. use the method of Example 8 to find the transient and steady periodic motions of the mass. m = 1. k = 4. the value A ifn < t < 2rr.15.5. I I 3 I I ..Sn)(t. and force f (t) are given for a 41.4n)(t. e(t) =SOt ifO ~ t < 1. f(t) = 0 if t < 2. Then construct the graph of the position function x(t). Solve the initial value problem di Ldt 1 + Ri + 1 c 1 i(r) dr = e(t). a mass. 30.5. m = 1. 4. f(t) is a squarewave function with amplitude 10 and period 2n. k (\ T 211: 40. Find cl{f(t)}.n)(t. 31. c f (t) = 0 if t > 2rr 34.1] !E k ~ k ~ k has the value +A ifO < t < n. spring constant k. f( t) is a squarewave function with amplitude 4 and period 2n. . Suppose that f(t) is a periodic function of period 2a with f(t) = t if 0 ~ t < a and f(t) = 0 if a ~ t < 2a. k = 1. and hence agrees on the interval [0. (t. R = 0. it may be useful to note that the function f( t) = A[2u((t. c = t > 2 5. k = 4. where f is the sawtooth function of Fig.16.
This is a function of t. Our strategy for handling such a situation is to set up a reasonable mathematical model in which the unknown force f(t) is replaced with a simple and explicit force that has the same impulse. In the case of a force f(t) that acts on a particle of mass m in linear motion. then we see (Fig. This is fortunate. Suppose for simplicity that f(t) has impulse 1 and acts during some brief time interval beginning at time t = a ~ 0. with a and E being parameters that specify the time interval [a. we need know neither the precise function f (t) nor even the precise time interval during which it acts. (3) I 1 I I I I I I 1 ii I I I I I I otherwise. a + E]. (2) Thus the impulse of the force is equal to the change in momentum of the particle. The number pin Eq.E (f) df = la ra+E! df E = 1. b] is a a +e P = FIGURE 4.E(t)dt = 1. b]. with f(t) = 0 outside this interval. integration of Newton's law f(t) = mv' (t) =  d dt [mv(t)] yields p = 1a bd dt [mv(t)] dt = mv(b)  mv(a).6. Thus da .E has a unit impulse.1) that the impulse of da. 4. for instance) is an analogous electrical phenomenon. we are unlikely to have such detailed information about the impulsive force that acts on the ball. Then we can select a fixed number E > 0 that approximates the length of this time interval and replace f (t) with the specific function X f. ! (t).1. (4) .6. lb a da.. In such a situation it often happens that the principal effect of the force depends only on the value of the integral p = 1b f(t)dt (1) and does not depend otherwise on precisely how f(t) varies with timet. A quick surge of voltage (resulting from a lightning bolt. because in a situation such as that of a batted ball. (1) is called the impulse of the force j(t) over the interval [a.316 Chapter 4 Laplace Transform Methods liD !mpulses and Delta Functions Consider a force f (t) that acts only during a very short time interval a ~ t ~ b. (t) ={ ~ if a~ t < a+ E. whatever the number E may be. If b ~ a+ E. So if change in momentum is the only effect with which we are concerned. Essentially the same computation gives 1 00 da. The graph of the impulse function da. we need know only the impulse of the force.e j r I I I ITI I Area= I da . A typical example would be the impulsive force of a bat striking a ballthe impact is almost instantaneous.E over [a .
We might try to model such an instantaneous unit impulse by taking the limit as E ~ 0. a+ E). then the mean value theorem for integrals implies that 1a+Eg(t) dt = Eg (t) for some point tin [a. Although we call it the delta function. Obviously.when applied to a continuous function g(t). It follows that lim E~O 1 0 oo g(t)da. it is not a genuine function. then it would follow that 1 But the limit in Eq. we therefore could conclude that Function g(l)~ 1 00 g(t)8a(t) dt = g(a). instead. and if we could interchange the limit and the integral in Eq. (4). (8). (5) gives Oa(t) = 00 Oa(t) dt = 1. (6) +oo {O if t =a. (9) as the definition(!) of the symbol 8a(t). Note that we will use the symbol 8a(t) only in the context of integrals such as that in Eq. This idea is shown schematically in Fig. If g(t) is continuous function.2.6. (9). it specifies the operation FIGURE 4.4. (6) and (7). However interpreted. 4.M.E(t)dt =lim 1a+E a E~O 1 _ g(t) · dt = limg (t) = g(a) E E ~O (8) by continuity of g at t = a. then its integral is not 1 but zero. A.6. who in the early 1930s introduced a "function" allegedly enjoying the properties in Eqs.sifts out or selects the value g(a) of this function at the point a ~ 0. the symbol 8a(t) is very useful. A diagram illustrating how the delta function "sifts out" the value g(a). it is tempting to think of an instantaneous impulse that occurs precisely at the instant t =a.6 Impulses and Delta Functions 317 Because the precise time interval during which the force acts seems unimportant. . which. no actual function can satisfy both (6) and (7)if a function is zero except at a single point. Delta Functions as Operators The following computation motivates the meaning that we will attach here to the symbol 8a(t). it is called the Dirac delta function at a after the British theoretical physicist P.2. if t =/= (7) a. thereby defining (5) where a ~ 0. (9) We take Eq. Dirac (19021984). Nevertheless. If 8a(t) were a function in the strict sense of the definition. or when it will appear subsequently in such an integral. If we could also take the limit under the integral sign in Eq.
(13) Note that if 8 (t) were an actual function satisfying the usual conditions for existence of its Laplace transform. (17). we get the equation . 8(t) is not a function. E+ 0 (16) where x" (t) is a solution of Ax"+ Bx' + Cx = da. (15) provided that x(t) = lim xE(t). (9). Delta Function Inputs Now. (14) To investigate the response of this system to a unit impulse at the instant t = a. But there is no problem here. (17) makes sense.1. the result is (10) We therefore define the Laplace transform of the delta function to be £{8a(t)} = eas (a ~ 0). and Eq. When we transform Eq. then Eq. it seems reasonable to replace f(t) with 8a(t) and begin with the equation Ax"+ Bx' + Cx = 8a(t). (15) But what is meant by the solution of such an equation? We will call x (t) a solution of Eq. suppose that we are given a mechanical system whose response x(t) to the external force f(t) is determined by the differential equation Ax" + Bx' + Cx = f(t). (11) If we write 8(t) = 8o(t) and 8(t a) = 8a(t). (13) would contradict the corollary to Theorem 2 of Section 4.318 Chapter 4 Laplace Transform Methods For instance. For simplicity suppose the initial conditions to be x(O) = x'(O) = 0. writing XE = oC{xE}. (12) then (11) with a = 0 gives £{8(t)} = 1. Because (17) (18) is an ordinary function. finally. (13) is our definition of £ {8 (t)}. E(t). if we take g(t) = esr in Eq. Eq.
3. we consider it beyond the scope of the present discussion.2rrs.5). (19) Note that this is precisely the same result that we would obtain if we transformed Eq.3s + 4X(s) = 8e.2rr) = 3 cos 2t + 8e.2rrs 2 • + 4u2rr (t) sin 2t. Note that the impulse at t = 2rr results in a visible discontinuity in the velocity at t = 2rr.a) with a= tan.6 Impulses and Delta Functions 319 If we take the limit in the last equation as E 1 ~ 0.9273) if t ~ 2rr. The resulting motion is shown in Fig.. separation of the cases t < 2rr and t :.2rr) sin 2(t ..6. According to Problem 15. • . 4. It is important to verify that the solution so obtained agrees with the one defined in Eq. and note that = 1 esf lim E+0 SE by l'Hopital's rule.. we need to solve the initial value problem x" + 4x = 882rr (t). At the instant t = 2rr the mass is struck with a hammer. we see that the inverse transform is x(t) = 3 cos 2t + 4u(t . providing an impulse p = 8. we get the equation (As 2 + Bs +C) X (s) = if X(s) =lim X~'(x).. using the fact that £{8a(t)} = eas. (16). x'(O) = 0. (15) directly.: 2rr gives x (t ) ~ { 3 cos 2t 5 cos(2t . The formal method is valid in all the examples of this section and will produce correct results in the subsequent problem set. there is no dashpot. E+ 0 eas.?:: 2rr. On this basis it is reasonable to solve a differential equation involving a delta function by employing the Laplace transform method exactly as if 8a (t) were an ordinary function.. ~ Solution We apply the Laplace transform to get s 2X(s).  A mass m = 1 is attached to a spring with constant k = 4.·. Determine the motion of the mass. Because 3cos2t + 4sin2t = 5cos(2t.0. ~~~ Example 1 ··"· ·. but this depends on a highly technical analysis of the limiting procedures involved. x(O) = 3. so X(s) = 2 3s s +4 s +4 Recalling the transforms of sine and cosine. as it instantaneously increases the amplitude of the oscillations of the mass from 3 to 5.~···'·~~~. .1(4/3) ~ 0. if t . as well as the theorem on translations on the taxis (Theorem 1 of Section 4.4. The mass is released from rest with x(O) = 3.9273.
4.5. c 1 In this example we have e(t) = 90 . with R = 110 Q.90ui (t) . . df E>0 df ' E>0 . dt ' ' Because a a+E FIGURE 4. Find the resulting current i (t) in the circuit. d . Example 2 We return to the RLC circuit of Example 5 of Section 4. At time t = 0 the switch is closed and at timet = 1 it is opened and left open.6. L = 1 H.ua(t) dt d = 8a(t) = 8(t.3.< (t) .4. we may begin with the ordinary circuit equation Li" + Ri' Solution + i = e' (t).E (t) to Ua (t) shown in Fig. Delta Functions and Step Functions X It is useful to regard the delta function 8a (t) as the derivative of the unit step function ua (t). The motion of the mass of Example 1.90u(t . . 6     x=5 X= 3 X= 6 0 2n 5 4n t 6n FIGURE 4.6. C = 0. We readily verify that d Ua E(t) = da E(t).320 Chapter 4 Laplace Transform Methods t = 2Jr. consider the continuous approximation ua. Approximation ofu0 (t) by Ua. (20) We may regard this as the formal definition of the derivative of the unit step function.ua(t) = hm Ua E(t) = hm da E(t). and therefore . Suppose that the circuit is initially passiveno current and no charge.4.6.5 we circumvented the discontinuity in the voltage by employing the integrodifferential form of the circuit equation. and a battery supplying e0 = 90 V. although Ua (t) is not differentiable in the ordinary sense at t = a. Now that delta functions are available. an interchange of limits and derivatives yields d .a) . To see why this is reasonable.001 F.1) = 90 . In Section 4.
. then x (t) = sin t . nrr. i" + llOi' + 1000i = 908 (t  i (0) (21) Li'(t) + Ri(t) + cq(t) = 1 e(t) with the numerical values i(O) = q(O) = 0 and e(O) = 90. We compute the inverse Laplace transform term by term.90 + llOsi (s) Hence + IOOOI (s) = 90es. (20).nrr) sin(t . it remains motionless until the third hammer blow starts it moving again. • . n=O Because oC {Onn (t)} = enn s. if the hammer blows are not perfectly synchronized then the motion of the mass will be quite different. if n is odd. x(O) = 0 = x'(O). 4. i' (0) = 90. . • Example 3 Consider a mass on a spring with m = k = 1 and x (0) = x' (0) = 0. we get the equation s 2 I (s).nns.1t that is. Of course.nrr) = (l)n sint and u(t . 3rr.5 ) I (s) = s 2 + llOs + 1000 This is precisely the same transform I (s) we found in Example 5 of Section 4.. When we transform the problem in (21 ). the transformed equation is 00 s 2 X(s) + X(s) = Le. Hence x(t) is the halfwave rectification of sin t shown in Fig. The physical explanation is that the first hammer blow (at timet = 0) starts the mass moving to the right.5..6. the result is x(t) =L n=O 00 u(t . and so on. so inversion of I (s) yields the same solution i (t) recorded there.nrr). x(t) n 2n 3n 4n FIGURE 4.5. rr. At each of the instants t = 0. 2rr. x(t) = { sin t 0 if n is even.sin t = 0 fort < nrr. We need to solve the initial value problem x" Solution +X= L 00 OmrU).6 Impulses and Delta Functions 321 so e' (t) = 908(tl) by Eq. The halfwave rectification of sin t. . Determine the resulting motion.4. Hence we want to solve the initial value problem = 0.. .nrr) nrr < t < (n + l)rr. The fact that i' (0) = 90 comes from substitution oft = 0 in the equation 1).. + sin t  ···+ (.6. the second hammer blow stops it dead.e. we see that if sin t. n=O so X(s) = L n=O 00 enns 2 S +1 . 90(1.5. the mass is struck a hammer blow with a unit impulse. Because sin(t. just as it returns to the origin.
The massspringdashpot system and the series RLC circuit are familiar examples of this general situation. Once w (t) has been determined. (23) The function 1 W(s)=as2 + bs + c (24) is called the transfer function of the system. From Eq. via the convolution integralDuhamel's principle reduces the problem of finding a system's outputs for all possible inputs to calculation of the single inverse Laplace transform in (25) that is needed to find its weight function. (24) we see by convolution that x(t) = 1 1 w(r)f(t . For simplicity we assume that the system is initially passive: x(O) = x' (0) = 0.322 Chapter 4 Laplace Transform Methods Systems Analysis and Duhamel's Principle Consider a physical system in which the output or response x(t) to the input function f (t) is described by the differential equation ax"+ bx' +ex = j(t).1{W(s)} (25) is called the weight function of the system. for instance) the response x (t) given by (26) automatically whenever a desired force function f (t ) is input. Then the transform of Eq. the integral in (26) gives the response of the system to an arbitrary input function j(t). Thus the transform of the response to the input j(t) is the product of W(s) and the transform F(s). b. so X (s) = F(s) as 2 + bs + c = W(s)F(s).r) dr. (22) is as 2X(s) + bsX(s) + cX(s) = F(s).springdashpot system described by (22) can be constructed in the form of a "black box" that is hardwired to calculate (and then tabulate or graph. In engineering practice. (22) where the constant coefficients a. so their behaviors can be studied without need for expensive or timeconsuming experimentation. In principlethat is. (26) This formula is Duhamel's principle for the system. Hence. The function w(t) = £. a computational analogue for a physical mass. all manner of physical systems are "modeled" in this manner. What is important is that the weight function w(t) is determined completely by the parameters of the system. . and care determined by the physical parameters of the system and are independent off (t).
..r) dr .. (23) that the transform of h(t) is H(s) = W(s).. or unit impulse response. either numerically or graphically. ·. and measure the response h(t) with an ammeter..... 1 1 h'(r)e' (t ...···.. Then Duhamel's principle implies that the response x ( t) to the force f (t) is x(t) = 1' e._... • Note that W(s) = as 2 1 £{8(t)} + bs + c as 2 + bs + c · Consequently. APPLICATIONS: To describe a typical application of Eq... What if the coefficients a. Substitution of (27) in Duhamel's principle gives x(t) = 1 1 h' (t)f(t..__ _ ._ ___ _ Example 4 Consider a massspringdashpot system (initially passive) that responds to the external force f(t) in accord with the equation x" + 6x' + lOx = f(t)..·· 323 . . Because £{u(t)} = 1/s...... we see from Eq.4. h (t) is the unit step response.. ..6 Impulses and Delta Functions ~. b.··· ... _________ _..... (28).(s + 3) 2 + 1 ' 1 so the weight function is w(t) = e.. the output current i (t) corresponding to the input voltage e(t) will be given by i (t) = (using the fact that f(t) = e' (t )). but with i in place of x. For this reason w(t) is sometimes called the unit impulse response. Then according to Eq..10 .r) dr (28) for the response of the system to the input f (t).. resistors. Assume that its circuit equation is a linear equation of the form in (22). We then compute the derivative h'(t). Then W (s)  1 ..r) dr. .s 2 + 6s + ....3 ' (sin r)j(t.3t sin t. and c are unknown. it follows from Eq.·· .. .... and capacitors. s It follows from the formula for transforms of integrals that h(t) = 1 1 w(r) dr.. is the derivative of the unit step response... so that w(t) = h' (t) ... (23) that the weight function is simply the response of the system to the delta function input 8(t) . A response that is usually easier to measure in practice is the response h (t) to the unit step function u (t).. perhaps only because they are too difficult to compute? We would still want to know the current i(t) corresponding to any input j(t) = e'(t). so that f(t) = e'(t) = 1 = u(t) . . suppose that we are given a complex series circuit containing many inductors....... We connect the circuit to a linearly increasing voltage e(t) = t.....··¥ .... (28). (27) Thus the weight function._....
we remark that around 1950. This is a generalization of Problem 15.6. e 0 = 100 V. Verify that u'(t. x(O) = x'(O) = 0 x" + 6x' + 9x = f(t). x(O) = x ' (O) = 0 This problem deals with a mass m. indeed. x' (0) =0 . Show that i(t) > 0 if 1 < t < 2 and that i (t) < 0 if t > 2.rr). Consider an initially passive RC circuit (no inductance) with a battery supplying e0 volts. have the same solution. x(O) = x'(O) = 0 6. x(O) = x ' (O) = 0 x" + 4x' + 8x = j(t).2). (a) Find the solution xf (t) of the problem mx" 17. x(O) = x(O) = 0. 18.x' (O) = 1 5.a). C = 104 F. problem mx" = po(t). HISTORICAL REMARK: Solve the initial value problems in Problems I through 8. Thus the effect of p 0 8(t) is. (a) If the switch is closed at time t = 0 and opened at time t = a > 0. 10. x'(O) = 0 3. Every piecewise continuous ordinary function is a generalized function. 1 i(O) = 0. Consider an initially passive LC circuit (no resistance) with a battery supplying e0 volts. and graph each solution function x(t).a). x" + 4x 2. x" + 2x' +X = o(t).324 Chapter 4 Laplace Transform Methods In conclusion. x" + 4x = j(t). x" +2x' +x = t + 8(t).e0 8(t. x" + 4x' + 4x = 1 + o(t.3rr) +cos 3t. if t > rr.b). 11. i(O) = i' (O) = 0. Thus the effect of the term av0 8(t) is to supply the initial condition x'(O) = v0 • 12. = w. x" + 9x = o(t.a).2 F.rr).6). x(O) = x'(O) = 0 x" + 6x' + 8x = j(t). e0 = sin lOt 10 V. This problem deals with a mass mona spring (with constant k) that receives an impulse p 0 = mv0 at timet = 0. Show that the initial value problems mx" +Ci 1 = e0 8(t) .a) by solving the problem x' = o(t  a). x(O) = x'(O) = 0. after engineers and physicists had been using delta functions widely and fruitfully for about 20 years without rigorous justification. 12. Po8(t). x'(O) = vo and =2 ax" + bx' +ex= f(t) + av0 8(t). 9. x'(O) 8. 4. the French mathematician Laurent Schwartz developed a rigorous mathematical theory of generalized functions that supplied the missing logical foundation for delta function techniques.( I t) = = rr (s). x(O) = x'(O) = 2 1. x' (O) = vo (b) If L = 1 H. x(O) =0 Li" to obtain x(t) = u(t. x(O) = O. x(O) = x'(O) = 0 7. but the delta function is an example of a generalized function that is not an ordinary function. show that the current in the circuit satisfies the initial value problem Ri' + Ci = eoo(t.a)= o(t. Thus the current oscillates through five cycles and then stops abruptly when the switch is opened (Fig. 16. show that the current in the circuit satisfies the initial value problem (c) Show that mv p fort > 0 (v = dxjdt). = x(O) = x ' (O) = 0. 0 if t < rr. that receives an impulse p at time t = 0. x(O) = x'(O) = 0 = 8(t) + o(t. 14. and b = 2 (s). initially at rest at the origin. C show that l and mx" + kx = x(O) = 0. a= 1 (s).o(t.eoo(t. (a) If the switch to the battery is closed at time t = a and opened at time t = b > a (and left open thereafter). to impart to the particle an initial momentum p 0 . Show that the problems ax" + bx' +ex= f(t). = pdo. x(O) = x'(O) = 0 Apply Duhamel's principle to write an integral formula for the solution of each initial value problem in Problems 9 through have the same solution for t > 0.x(O) = 0. x" +4x = o(t). x(O) = x'(O) = 0 4.2). and a + kx = 0. x" + 2x' + 2x = 28(t. 13. x"+4x'+5x = 8(trr)+8(t2rr).(t) agrees with the solution of the f + 0 (b) Solve this problem if R = 100 Q. . 15. (b) Show that limx.f(t). x(O) = 0.
then x(t) = (n + 1) sint. Thus resonance occurs because the mass is struck each time it passes through the origin moving to the rightin contrast with Example 3. Thus a resonance phenomenon occurs (see Fig.6. 4rr. 21.. 2rr /10. Consider an RLC circuit in series with a battery. Show that the position function x(t) ofthe mass satisfies the initial value problem x" +X= L n=O 00 o(t. Show that i (t) satisfies the initial value problem i" + 60i' + 1000i . 2rr.8). . Thus the current in alternate cycles of length rr/5 first executes a sine oscillation during one cycle. (a) Show that i (t) satisfies the initial value problem i"+lOOi = 10 ~)lto(t.. n=O 10 = (n + 1) sin lOt if mr (n <t < + l)rr 10 10 i(O) = i'(O) = 0. . x(O) = x ' (O) = 0. in which the mass was struck each time it returned to the origin. Repeat Problem 19. n=O 10 00 i(O) = i'(O):::: 0. 2rr /10. nn/5. FIGURE 4.6. Suppose that each hammer blow imparts an impulse of+ 1.. rr/5. 10 FIGURE 4. Consider the LC circuit of Problem 18(b). construct a figure showing the graph of this position function.2nn). and so on (see Fig. i(t) (b) Solve this problem to show that if <t<nJT (n + 1)rr lO 10 10 then i(t) = e3mr +3rr e3rr  1 1 e. Consider a mass m = 1 on a spring with constant k = 1. initially at rest. but struck with a hammer at each of the instants t = 0. . except suppose that the switch is alternately closed and opened at times t = 0. (b) Solve this initial value problem to show that i(t) = oo ( 10 L(l)no t nn) . rr /10. C = 103 F. .6.8. (a) Suppose that the switch is alternately closed and opened at times t = 0. Finally.4..6. The current function of Problem 20.6. if n is odd. 19. . then is dormant during the next cycle. 22. Now show that if nrr (n + l)rr <t< Construct a figure showing the graph of this current function.. 2rr/5. .6 Impulses and Delta Functions i(t) i(t) 1\ 1\ 325 In 1\ A 1 v v v v v FIGURE 4. with L = 1 H. The current function of Problem 19. and e0 = 10 V..301 sin lOt. . except suppose that the switch is alternate! y closed and opened at times t = 0. Solve this problem to show that if 2nn < t < 2(n + 1)rr. 4.. 4.7. R = 60 Q.. . The current function of Problem 18. .6.. 20. . . rr/10. .7).nrr). 5 5 then if n is even. .
y. the position (x(t). y. x. y 1) = 0. y 1 (t). Applying Newton's law rna= F componentwise. require the use of two or more dependent variables. y 1) = 0.. each a function of a single independent variable (typically time). x. y I . II F2t. z. z . however. x 1 .z II 326 . We will usually denote the independent variable by t and the dependent variables (the unknown functions oft) by x 1. I I I) m z = F3t. Primes will indicate derivatives with respect to t. . x I . y. x . or by x. y. x 2. For an example of a secondorder system. A solution of this system is a pair x(t). y. a system of two firstorder equations in the dependent variables x and y has the general form f(t . .y . and its velocity (x 1(t) . Z1 (t)). Many applications. X 1 . x . consider a particle of mass m that moves in space under the influence of a force field F that depends on time t. x 3.zI ) ....Linear Systems of Differential Equations Ill FirstOrder ~ystems and Ap_plications n the preceding chapters we have discussed methods for solving an ordinary differential equation that involves only one dependent variable. we get the system mx II = F 1 (t. g(t .x. For instance. z I ) .y. (2) ( x . (1) where the functions f and g are given. y(t).z. Such a problem leads naturally to a system of simultaneous ordinary differential equations.x.. y. I We will restrict our attention to systems in which the number of equations is the same as the number of dependent variables (unknown functions). ( I I my= z . y (t) of functions of t that satisfy both equations identically over some interval of values oft. z(t)) of the particle.
Consider the system of two masses and two springs shown in Fig. . The two brine tanks of Example 2. . .. ..1.30 · + 10 · = ... ..1.. .. m 2y" = k2(y.1. 20y' = 6x..1. In addition. fresh water flows into tank 1 at 20 galjmin..x units... The "free of differential equations that the position functions x(t) and y(t) must satisfy...x + . F3 are the components of the vectorvalued function F. In the configuration in Fig..5.1.. We denote by x(t) the displacement (to the right) of the mass m 1 from its static equilibrium position (when the system is motionless and in equilibrium and f(t) = 0) and by y(t) the displacement of the mass m 2 from its static position.. and brine is pumped from each tank to the other at the rates indicated in Fig. 5. + 2y. ifm 1 = 2.. .... k 2 = 2. and the brine in tank 2 flows out at 20 gal/min (so the total volume of brine in the two tanks remains constant). . we therefore get the system of differential equations that x(t) and y(t) must satisfy: 1 X Y 3 1 x = ..y 100 200 10 20 ' 1 X Y Y 3 3 y = 30 · .. 5. .3.1...x). Initial Applications Examples 1 through 3 further illustrate how systems of differential equations arise naturally in scientific problems.3. . The brine in each tank is kept uniform by stirring. respectively._t) FIGURE 5. we thereby obtain the system m1x" = k1x + k2(Y..1. 20x' = 6x + y... The salt concentrations in the two tanks are x 1100 pounds per gallon and y /200 pounds per gallon.1 FirstOrder Systems and Applications 327 x.. . m 2 = 1.. ' ~ ... then the system in (3) reduces to 2x" = . k 1 = 4. (4) • Example 2 20 gal/min Fresh water 20 gal/min FIGURE 5. 5.. the three righthandside functions F 1. The mass andspring system of Example I.3.2. " "'" ' '' '" .1.. Thus the two springs are neither stretched nor compressed when x and y are zero. Consider two brine tanks connected as shown in Fig... For instance... . Tank 1 contains x(t) pounds of salt in 100 gal of brine and tank 2 contains y(t) pounds of salt in 200 gal of brine. and f(t) = 40sin3t in appropriate physical units. 5.x) + f(t) (3) k~(y  x) [ii:'t__ f~.... the first spring is stretched x units and the second by y ... F2 .2.1.6x y" = 2x  body diagrams" for the system of Example 1. z. 5 . When we compute the rates of change of the amount of salt in the two tanks.. . (5) • .1.......10 · .. with a given external force f(t) acting on the righthand mass m 2 .. We apply Newton's law of motion to the two "free body diagrams" shown in Fig.x .. ' ~··· Example 1 Equilibrium positions FIGURE 5.3y..20 · = . of three secondorder equations with independent variable t and dependent variables y.1. 2y + 40 sin 3t.y· 100 200 200 10 20 ' that is.
. (6) FIGURE 5.. 7.=. differentiation of each side of Eq.h in the direction indicated.· ·· .5. where / 1(t) denotes the current in the indicated direction through the inductor L and h (t) denotes the current through the resistor R 2 .¥. 5.1. because the voltage drop from the negative to the positive pole of the battery is 100. Voltage drops across common circuit elements.4. We recall Kirchhoff's voltage law to the effect that the (algebraic) sum of the voltage drops around any closed loop of such a network is zero. (7) yields 125/2 dlz d/1 + 7550. respectively..008 farads £0 : .1. Element :~~rcui't >.~ "' "··· _ _. ~ Example 3  · . 2 .100 = 0.. 5.= dt dt 0.~ ~·.1.·.. 100 volts _ +G 11 R1 : 50 ohms FIGURE 5. we get the system dh + 25h dt d/1 25/z = 50. The current through the resistor R 1 is I = It . the voltage drops across the three types of circuit elements are those shown in Fig..y ··¥~·~¥.4.1. We apply Kirchhoff's law to the lefthand loop of the network to obtain CQ 2dt dh + 50(/t  /2) .35/z =0 dt dt of differential equations that the currents It (t) and / 2 (t) must satisfy.328 Chapter 5 Linear Systems of Differential Equations L: 2 henries C: 0. (6) and (8) by the factors 2 and 25. The righthand loop yields the equation (7) where Q 2 (t) is the charge on the capacitor. (8) After dividing Eqs. The electrical network of Example 3.. 'Voltage DtQP Ldt RI 1 Inductor Resistor Capacitor dl Consider the electrical network shown in Fig. dh (9) • . As in Section 2. Because dQ 2 jdt = ]z .5.~ ~ · ..
as explicit functions of t and lowerorder derivatives of the dependent variables.5.x. .. . .. Example 4 The thirdorder equation xC 3l + 3x " + 2x 1  Sx = sin 2t is of the form in ( 11) with f (t ... • . X3 = X = X2 t yield the system X I = X2 . II . I X3 =X . . Evidently. x . X].. Hence the substitution of (l2). To describe how such a transformation is accomplished. x 2 . (11) We introduce the dependent variables x 1. (11) yields the system = x" = x 3 .2x 1  3x" + sin 2t. in the case of a system of two secondorder equations. X 1 . X2 =X = t I X I.3x 3 + sin2t of three firstorder equations. .in (13) I X n1 = Xn. our assumption is that it can be written in the form (10) It is of both practical and theoretical importance that any such higherorder system can be transformed into an equivalent system of firstorder equations.2x2 .. For instance. Xn defined as follows: Xn =X. X2. X2 =X . Xn (t) defined in (12) satisfy the system of equations in (13). I I x2 = X 3.1 FirstOrder Systems and Applications 329 FirstOrder Systems Consider a system of differential equations that can be solved for the highestorder derivatives of the dependent variables that appear. (12) Note that x ~ = X 1 = x 2 . If Hence the substitutions Xi = X. Xn) X~= j(t. x " ) = Sx .. of n firstorder equations. we consider first the "system" consisting of the single nthorder equation _ j( 1 x (n) t. and so on.x.. ' = X(n. . .. in the sense that x(t) is a solution of Eq. X] . (11) if and only if the functions x 1 (t). x 2 (t). x (n 1)) . this system is equivalent to the original nthorder equation in (11). . .1). x~ = Sx1 . x~ Eq.
Solution Motivated by the equations in (12). + 40sin3t 2x. large systems of higherorder differential equations typically are solved numerically with the aid of the computer. The corresponding firstorder system is (14) and we will see in Section 6. I (17) .py. From a practical viewpoint. Transform this system into an equivalent firstorder system. x2=x =x1 . But suppose that we were confronted with the nonlinear equation x" = x 3 + (x 1 ) 3 . x 2.4 that there exist effective numerical techniques for approximating the solution of essentially any firstorder system.qx . y~ = 2xi .330 Chapter 5 Linear Systems of Differential Equations It may appear that the firstorder system obtained in Example 4 offers little advantage because we could use the methods of Chapter 2 to solve the original (linear) thirdorder equation. we define XJ=X. y 1. and the first step is to transform such a system into a firstorder system for which a standard computer program is available. So in this case the transformation to a firstorder system is advantageous. and Y2· Simple TwoDimensional Systems The linear secondorder differential equation x" + px + qx = 1 0 (16) (with constant coefficients and independent variable t) transforms via the substitutions x 1 = y. to which none of our earlier methods can be applied. Example 5 The system 2x" = 6x y" = + 2y.2y1 I (15) + 40sin 3t • of four firstorder equations in the dependent variables x 1. Y2=Y =Y1· I I Then the system in (4) yields the system Y1 = Y2. y = . I I Y1=y.2y (4) of secondorder equations was derived in Example 1. x" = y 1 into the twodimensional linear system xl = y.
.. Because the moving point (x(t)._. Thus the trajectories of the system in (18) are the ellipses of Fig.1. Example 6 To solve the twodimensional system x' = 2y. y(t)) has velocity vector (x'(t). Additional information can be shown in the separate graphs of x(t) and y(t) as functions oft. for each value oft.L.1 FirstOrder Systems and Applications 331 Conversely. A solution (x(t). the direction field plotted in Fig.can be plotted. The choice of an initial point (x(O). Direction field and solution curves for the system x' = 2y. this direction field indicates the point's direction of motion along its trajectory.1. with semiaxes C and C /2. Figure 5.. This gives the single secondorder equation x" x (t) = A cos t 5 4 3 2 I "' 0 1 1't__. The picture showing a system's trajectories in the xyplaneits socalled phase plane portraitfails to reveal precisely how the point (x(t)..a).6 shows several such ellipses in the xyplane .a) where A= C cos a and B = C sin a. I ! (18) we begin with the observation that x" = 2y' = 2 Gx) = x. we can solve this system in ( 17) by solving the familiar single equation in (16).. the point (x(t).:.1.5. x. y(O)) determines which one of these trajectories a particular solution parametrizes.._"'t". y(t)) lies on the ellipse FIGURE 5.6 indicates that each such point moves counterclockwise around its elliptical trajectory.y) • may be regarded as a parametrization of a solution curve or trajectory of the system in the xyplane. y(t)) of a twodimensional system x' = f(t.tr. 5...x. y'=g(t. y) and y' = g (x . 2 3 4 5 X The identity cos 2 () + sin2 () = 1 therefore implies that. y = 2X. then a direction fieldshowing typical arrows representing vectors with components (proportional to) the derivatives x' = f (x.. y'(t)). y ).6. y' = ~x of Example 6.. If the functions f and g do not involve the independent variable t. . 5..1.. y(t)) moves along its trajectory. Then y(t) = ~x'(t) = ~(Asint + Bcost) = !C sin(t..H +x = 0 with general solution + B sin t = C cos (t . For instance._~7:'..6. y)..
(20) and (21) are shown in Fig. y (t) = sin t. • 4 3 2 :: 0 "' 1 2 3 4 0 5 10 15 FIGURE 5. It follows that. y(O)  = ~B = 0.1.7. The graphs of the two functions are shown in Fig.1. Typical phase plane trajectories of the system in (19) parametrized by Eqs. y' = 2x + y of Example 7. x.8. x' (19) we begin with the observation that x" = y' = 2x +y = + 2x. We see that x(t) initially decreases while y(t) increases.332 Chapter 5 Linear Systems of Differential Equations " "" ¥" ~ · ··· ···· · . the general solution in Example 6 yields x(O) =A= 2. y(t)) traverses the trajectory ~x 2 + y 2 = 1 in the counterclockwise direction.7.x' .2 = (r + l)(r  2) =0 (20) and general solution x(t) = Ae.1.t (21) FIGURE 5. • . y' = 2x + y.6. 5. Direction field and solution curves for the system x' = y. but Problem 23 shows that their actual form is somewhat more complicated.8. These trajectories may resemble hyperbolas sharing common asymptotes. y(O) = 0.2x = 0 with characteristic equation r2  r. 5. ··~· ·· Example 6 Continued With initial values x(O) = 2. as indicated by the direction field vectors in Fig. Example 7 To find a general solution of the system x' = y. y' = !x.andysolution curves for the initial value problem x' = 2y. as t increases. This gives the single linear secondorder equation x" . the solution point (x(t).1. x(O) = 2. The resulting particular solution is given by x (t) = 2 cos t.1. y(O) = 0.t + Be21 • + 2Be21 • Then X y(t) = x ' (t) = Ae. 5.
y' = (1.1. y' = (l.2)y of Example 8. y(O) = B = 1. + P22(t)x2 + · · · + P2nXn + f2(t).1 ± i. characteristic roots 0.1.01)x. and general solution x(t) = et/lO(A cost+ B sin t). we begin with the observation that (22) y(0)=1.01 = (r + 0. consider a linear firstorder system of the form xi FIGURE 5.2)y] = ( 1.lOcost).1.and ysolution curves for the initial value problem of Example 8. so the desired solution of the system in (22) is x(t) = et. x" = y' = [(l. = Pll (t)XJ + P12(t)x2 + · · · + PinXn + !1 (t) . Io sint.4. the general theory of systems and systematic solution techniques are more easily and more concisely described for firstorder systems than for higherorder systems. the trajectory approaches the origin as t+ + oo.2)y.1 FirstOrder Systems and Applications 333 Example 8 To solve the initial value problem x' = y.9.1. + Pn2(t)X2 + · · · + PnnXn + fn(t). Finally.1) 2 + 1 = 0.y. (23) y(t) = /0et/IO(sint.(0. x.10 shows the x. 5.2)r + 1.1.01)x. Then x(O) =A= 0.1.9.6. 5.5. x(0)=0.(0. and 5.(0. These equations parametrize the spiral trajectory in Fig.Ol)x.01)x = 0 with characteristic equation r2 + (0. we will learn why the superficially similar systems in Examples 6 through 8 have the markedly different trajectories shown in Figs.t/IO cost. so x(t) = Bet!IO sint. For instance.Ol)x.(0.2)x'.8.2)x' + (1.10. Figure 5.9.1.andysolution curves given in (23). Direction field and solution curve for the system x' = . This gives the single linear secondorder equation x" + (0. (24) X~= P21 (t)XJ X~= Pni(t)XJ .Be . Linear Systems In addition to practical advantages for numerical computation. 5. X FIGURE 5. y(t) = x' (t) = /0 Bet flO sin t . • When we study linear systems using the eigenvalue method of Section 5.
. then the system has a unique solution satisfying given initial conditions. the system in (24) has a unique solution on the entire interval I that satisfies the n initial conditions . . Xn (t) that (on some interval) identically satisfy each of the equations in (24 ).. given then numbers b 1. Typical initial values would be the initial positions x (0) and y (0) and the initial velocities x' (0) and y' (0). fn are continuous on the open interval I containing the point a.. It tells us that if the coefficient functions PiJ and /j in (24) are continuous. fn are all identically zero. x2(t). .... Thus the linear system in (5) is homogeneous. we found that the amounts x (t) and y (t) of salt in the two tanks of Example 2 are described by the system 20x' = . h .6x + y. y"= 2x2y+40sin3t. We will see that the general theory of a system of n linear firstorder equations shares many similarities with the general theory of a single nthorder linear differential equation. . we saw in Example 5 that the secondorder linear system 2x" = . . otherwise. The system in ( 14) is nonlinear because the right hand side of the second equation is not a linear function of the dependent variables x 1 and x2. Then. Theorem 1 (proved in the Appendix) is analogous to Theorem 2 of Section 2. .3y of two firstorder linear equations. (25) Thus n initial conditions are needed to determine a solution of a system of n linear firstorder equations. . it is nonhomogeneous. ... .. bn. and we therefore expect a general solution of such a system to involve n arbitrary constants. h. whereas the linear system in (15) is nonhomogeneous.334 Chapter 5 Linear Systems of Differential Equations We say that this system is homogeneous if the functions f 1. . we often must transform it into an equivalent firstorder system to discover how many initial conditions are needed to determine a unique solution. is equivalent to the system of four firstorder linear equations in (15). For instance. THEOREM 1 Existence and Uniqueness for Linear Systems Suppose that the functions Pll· PI2· . which describes the position functions x(t) and y(t) of Example 1.6x + 2y. On the other hand.2. Pnn and the functions ft. 20y' = 6x. A solution of the system in (24) is anntuple of functions x 1 (t).. Hence the two initial values x(O) and y(O) should suffice to determine the solution. Hence four initial conditions would be needed to determine the subsequent motions of the two masses in Example 1. Theorem 1 tells us that the number of such conditions is precisely the same as the number of equations in the equivalent firstorder system.. ••• . Given a higherorder system. b2 .
1 FirstOrder Systems and Applications 335 Bl Problems 24. 5. 22.l)x = 0 4. y(O) = 7 19. x' = y. x' = lOy.x)y 11 + k2)x1 + k2x2.[y(t)] 2 to show that the trajectories of the system x' = y.1) 2 = 1 to show that the trajectories of the system x' = y. Derive the equations m1x. y' = x 13. calculate x 2 + y 2 to show that the trajectories are circles.1.x 12. 25. 1. x< 3) = (x') 2 +cos x 6. z'' = 5x .13. If initial conditions are given.' = 2y l + Y2. ~ ~ ~ ~ t ~. y' = . y' = 8x 16. x" + 3x' + 7x = t 2 2.1. x<4 ) + 6x" . y' = 13x + 4y. y' = 2x. use a computer system or graphing calculator to construct a direction field and typical solution curves for the given system. x" = 3x .1 and e 2 1 in terms of x(t).y + 2z. y' = x of Problem 12 are hyperbolas.2t 2x 11 + 3tx' + 5x = lnt 5. at timet (i = I. The system of Problem 24. and the constants A and B. X 11 = (1.1. 5. y' = 2x of Problem 13. = x 1 lOx~= X1 Xz Xz XJ . transform the given differential equation or system into an equivalent system offirstorder differential equations.1.5y = 0 7 II kx II ky • x = . (a) Beginning with the general solution of the system x' = .x and y = 2x that are visible in Fig.(k2 + k3)x2 for the displacements (from equilibrium) of the two masses shown in Fig.5x + 4y = 0.3xi + / cated in Fig. y' = 6x. For each problem. x' = y. y m X FIGURE 5. y' = 9x + 6y 21.2y = 0. 26. x' =By. x(O) = 0. Suppose that the mixture circulates between the tanks at the rate of 10 gal/min. Y + 2y' .11. (a) Calculate [x(t)f + [y(t)f to show that the trajectories of the system x' = y.2y. t 3x< 3). The mechanical system of Problem 25. x' = y. x(O) = l. 23. (20) and (21) for e. Show that the displacements y 1 and y2 satisfy the equations ky.5. 5. Assume Use the method of Examples 6. (b) Show similarly that the trajectories of the system x' = ~ y. 1.z 10. (t) the amount (in pounds) of alcohol in tank T. 11. y" = (1 . 2. Two particles each of mass m are attached to a string under (constant) tension T. (b) Calculate [x(t)] 2 . y" + 4x.12. FIGURE 5.(x2 + y2)3f2' Y = . y' = lOx. The fermentation tanks Then show that C = 0 yields the straight lines y = .12.2y2 where k = mLjT. t 2x" + tx' + (t 2 . x" + 3x' + 4x . parallel to the yaxis) with amplitudes so small that the sines of the angles shown are accurately approximated by their tangents. as indicated in Fig.3x + y = cost 9.' = (k1 m2x~ == In Problems I through IO. y(O) = 3 20.y. Denote by x. and 8 to find general solutions of the systems in Problems II through 20. Then substitute the results in (e 21)(e.4z.y . 7. y' = .7y. y' = 2x 17. x' = y. x' = y. x". x(O) = 2. y" = x + y . ~'· XI X2 FIGURE 5. 3). y(O) = 4 15.x of Problem 11 are that the particles oscillate vertically (that is. y(t). 5.3x' + x = cos 3t 3. First solve Eqs. ky~ = Y1 . k2x1 .8x of Problem 15 are ellipses with equations of the form l6x 2 + / = C 2 . and the mixtures in each tank are kept uniform by stirring. x(O) = 1. + X3 lOx~ = = C (constant). x(O) = 3. x' = 2y. y' = . y(O) = 2 18.11.1. find the corresponding particular solution. of Problem 26.13. y' = lOx. y' = 2x + y in Example 7 satisfy an equation of the form 4x 3 . y(O) = 0 14. x' = y.1.8. .y)x. Derive the equations lOx.(x2 + y2)3f2 8. Three 100gal fermentation vats are connected as indi circles. x' = ~y.
r3 X FIGURE 5. Kepler concluded that the motion of the planets around the sun is described by the following three propositions. Repeat Problem 27. which shows an inductor. Show that the equations of motion of the projectile are 27. A particle of mass m moves in the plane with coordinates (x(t). The square of the planet's period of revolution is proportional to the cube of the major semiaxis of its elliptical orbit. In his Principia Mathematica ( 1687) Isaac Newton deduced the inversesquare law of gravitation from Kepler's laws. 2.14.1. now known as Kepler's laws of planetary motion: 1.14. Suppose that a projectile of mass m moves in a vertical plane in the atmosphere near the surface of the earth under the influence of two forces: a downward gravitational r Jx mx" = +qBy'.15. 3. 5. The orbit of each planet is an ellipse with the sun at one focus. my"= kvy' .336 Chapter 5 Linear Systems of Differential Equations force of magnitude mg. Johannes Kepler analyzed a lifetime of planetary observations by the astronomer Tycho Brahe.15). 28. The radius vector from the sun to each planet sweeps out area at a constant rate. 29. Then the inversesquare law of gravitation implies (Problem 29) that the . y v '' FIGURE 5. (1) where i = (1. The trajectory of the projectile of Problem 30. y(t)) = x i+ yj.andydirections. except with the generator replaced with a battery supplying an emf of 100 V and with the inductor replaced with a 1millifarad (mF) capacitor. so the force on the particle is F = qv x B if its velocity is v. see Fig. In this application we lead you (in the opposite direction) through a derivation of Kepler's first two laws from Newton's law of gravitation. Suppose that a particle with mass m and electrical charge q moves in the xyplane under the influence of the magnetic field B = Bk (thus a uniform field parallel to the zaxis).1. = J (x')2 + (y')2. 5. where = 30.1. Set up a system of firstorder differential equations for the indicated currents / 1 and [z in the electrical circuit of Fig. 31. and write the position vector of the planet in the form r(t) = (x(t). 5. Assume that the sun is located at the origin in the plane of motion of a planet. y(t)) under the influence of a force that is directed toward the origin and has magnitude kj(x 2 + y 2)an inversesquare central force field. Show that II kx mx = . and a resistive force F R that is directed opposite to the velocity vector v and has magnitude kv 2 (where v = lvl is the speed of the projectile.mg. Show that the equations of motion of the particle are and II ky my = . 0) and j = (0. 2 + y2 .1 Application Gravitation and Kepler's Laws of Planetary Motion Around the tum of the 17th century. and a generator which supplies an alternating voltage drop of E(t) = l00sin60t V in the direction of the current / 1• mx" where v = kvx'. two resistors. my"= qBx' .r3 .1. 1) denote the unit vectors in the positive x . The electrical circuit of Problem 27.
Equating the transverse componentsthat is.= h dt ' FIGURE 5. Because the polarcoordinate area elementfor computation of the area A(t) in Fig. (8) implies that the derivative A' (t) is constant.1. Eq. 5. Area swept out by the radius vector. so Ur = r j r. STEP 1: Differentiate the equations in (3) componentwise to show that dt du.1.5.+rue. (8) where h is a constant. which is a statement of Kepler's second law.16 are given by u.1. the coefficients ofuewe get (7) so it follows that 2d8 r .= u. 8 (t)).1. then the radial and transverse unit vectors shown in Fig. If the polar coordinates of the planet at time t are (r (t). The radial and transverse unit vectors Ur and u 11 • The radial unit vector Ur (when located at the planet's position) always points directly away from the origin.17. (3) FIGURE 5. de = ue dt and =Ur· due dt dB dt (4) STEP 2: Use the equations in (4) to differentiate the planet's position vector r = ru. dt dt dt (5) Differentiate again to show that the planet's acceleration vector a dvjdt is given by STEP 3: a= [d r _ r (d8) ] dt 2 dt 2 2 Ur + [~~ (r2d8)]ue.17is given by dA = ~r 2 d8.1 FirstOrder Systems and Applications 337 acceleration vector r" (t) of the planet is given by y kr r" = r 3' (2) where r = J x 2 + y 2 is the distance from the sun to the planet. 5: Equate radial components in (2) and (6) and then use the result in (8) to show that the planet's radial coordinate function r(t) satisfies the secondorder differential equation STEP (9) . and thereby show that its velocity vector is given by dr dr dB v = . and the transverse unit vector ue is obtained from Ur by a 90° counterclockwise rotation..16. X = icose +jsin8 and ue = isin8 +jcose. (2) and (6) must agree. 5. r dt dt (6) STEP 4: The radial and transverse components on the righthand sides in Eqs.
(8) to show that if r = ljz. and first use the chain rule and Eq.1.. it can be transformed to a linear equation by means of a simple substitution. semilatus rectum L (Fig. In rectangular coordinates you can write x(t) = r(t)cost. a parabola if e = 1. (9) that the function z(()) = 1/r(()) satisfies the secondorder equation (10) STEP 7: Show that the general solution ofEq.a) FIGURE 5. so close to zero that the orbit looks nearly circular (though with the sun off center). and the eccentricities of the other planetary orbits range from 0.e).. As indicated in Fig. The object of this procedure is to eliminate dependent variables in succession until there remains only a single equation containing only one dependent variable.2486 for Pluto. C cos a= A. After its solution has been found.1.1. The elliptical orbit r=  L 1 + ecos(B.1 ecos((). + Bcos() + h2 .0068 for Venus and 0. BJ The Method of Elimination The most elementary approach to linear systems of differential equations involves the elimination of dependent variables by appropriately combining pairs of equations.1. sizes. Planetary orbits are bounded and therefore are ellipses with eccentricity e < 1. C sin a = B. y(t) = r(t)sint . (10) is z(()) = Asm() . the other dependent variables can be found in tum. 5.2056 for Mercury and 0. and L = h 2 jk.1. 0 ::::. The shape of the orbit of Halley's comet. using either the original differential equations or those that have appeared in the elimination process. the major axis of the ellipse lies along the radial line() = a. This remaining equation will usually be a linear equation of high order and can frequently be solved by the methods of Chapter 2.338 Chapter 5 Linear Systems of Differential Equations STEP 6: Although the differential equation in (9) is nonlinear.18.18. and rotation angle a.97 (Fig. k (l1) STEP 8: Finally. (12) is a conic section of eccentricity ean ellipse if 0 ~ e < 1. like Halley's comet withe ~ 0. withe= Ch 2 jk. deduce from Eq. 5 . For this purpose.0167. STEP 9: Plot some typical elliptical orbits as described by (12) with different eccentricities. 5.a)  L + (12) with perihelion distance r 1 = Lj(l +e) and aphelion distance r 2 = Lj(l . The eccentricity of the earth's orbit is e ~ 0. assume that the orbit can be written in the polarcoordinate form r = r(()). (11) that r(()) = 1/z(()) is given by r(()) = . But many comets have highly eccentric orbits. The polarcoordinate graph of Eq. 2n FIGURE 5. and a hyperbola if e > 1with focus at the origin. .19). then dr dt y hd() dz Differentiate again to deduce from Eq.18). to plot an elliptical orbit with eccentricity e.0933 for Mars to 0.19. and orientations. t::::.
as well as for theoretical discussion. that satisfies the initial conditions x(O) Solution y 1 = 6x. . this yields 1 If 7 I 4([ 7 ) 6Y + 6Y = 6Y I + 6Y . (4) and (5) constitute the general solution of the system in (1).3y.10 = (r.2 The Method of Elimination 339 The method of elimination for linear differential systems is similar to the solution of a linear system of algebraic equations by a process of eliminating the unknowns one at a time until only a single equation with a single unknown remains. (5) Thus Eqs. For such systems the method of elimination provides a simple and concrete approach that requires little preliminary theory or formal machinery. . But for larger systems of differential equations. (3) We then substitute these expressions for x and x 1 in the first equation of the system in (1).. the matrix methods of later sections in this chapter are preferable.2)(r + 5) = 0. we get (2) so that X  1 _1 6Y"+7 6Y I . It is most convenient in the case of manageably small systems: those containing no more than two or three equations. which we simplify to y" + 3Y 1  lOy = 0.3y. If we solve the second equation in ( 1) for x.5. The given initial conditions imply that and that y (O) = CJ + cz = . substitution of (4) in (2) gives that is. This secondorder linear equation has characteristic equation r 2 + 3r. so its general solution is (4) Next. .7y (1) = 2.1. y(O) = 1.~ ·· Example 1 Find the particular solution of the system x 1 = 4x.
• Remark: The general solution defined by Eqs. if L 1 = D + a and L 2 = D + b. 3 = c I (le2t e2t) 2 ' This expression presents the general solution of the system in ( 1) as a linear combination of the two particular solutions • Polynomial Differential Operators In Example 1 we used an ad hoc procedure to eliminate one of the independent variables by expressing it in terms of the other. y(t)). Direction field and solution curves for the system x' = 4x. By contrast. We now describe a systematic elimination procedure. Operator notation is most convenient for these purposes.St) . e.340 Chapter 5 Linear Systems of Differential Equations these equations are readily solved for c 1 = 2 and c2 = 3. (x(t). Recalling the componentwise addition of vectors (and multiplication of vectors by scalars).2. we can write the general solution in (4) and (5) in the form FIGURE 5. Recall from Section 2.3eSt. D(Dx This illustrates the fact that two polynomial operators with constant coefficients can be multiplied as if they were ordinary polynomials in the "variable" D. . Figure 5.St .1 shows this and other typical solution curves parametrized by the equations x(t) = ~c 1 e 2 t + ~c2 esr.7y of Example 1. then = L 1 L2[x] = (D + a)[(D + b)x] = + bx) + a(Dx + bx) [D 2 +(a+ b)D + ab]x.3 that a polynomial differential operator is one of the form (6) where D denotes differentiation with respect to the independent variable t. it follows that (8) if the necessary derivatives of x(t) exist. (4) and (5) may be regarded as the pair or vector (x(t). y' = 6x.1. Because the multiplication of such polynomials is commutative. y(t)) = (~cie2t + ~c2esr. this property of commutativity generally fails for polynomial operators with variable coefficientssee Problems 21 and 22. y(t) = c 1e 2t + c2esr with different values of the arbitrary constants c 1 and c2 • We see two families of curves resembling hyperbolas sharing the same pair of (oblique) asymptotes. If L 1 and L 2 are two such operators.3y. y(t) = 2e2t . cle2t + c2est) + c2 (le . Hence the desired solution is x(t) = 3e2t eSt.2. then their product L 1 L 2 is defined this way: (7) For instance.
Either process is especially simple if the system is homogeneous (f1(t) 0 and fz(t) 0). After solving for y = y(t) we can substitute the result into either of the original equations in (9) and then solve for x = x(t). L 2. = = . and (15) are zero. (11) Subtraction of the first from the second of these equations yields the single equation (12) in the single dependent variable y. Note that the same operator L 1 L 4 . we would get the equation (1 3) which can now be solved for x = x(t). For instance. (12) and (13) can be rewritten as Lz L 3 L4 L1 L1 L3 lx =I Lz L4 ly= I L3 L1 f1 (t) Lz fz(t) L4 f1 (t) fz(t) I. we operate with L 3 on the first equation and with L 1 on the second. 0. This is the operational determinant (14) of the system in (9). you can solve a system of two linear differential equations either by carrying out the systematic elimination procedure described here or by directly employing the determinant notation in ( 15). L 3.4)x 3y = + 6x + (D + 7)y = 0. (9) where L 1 . (13). because in this case the righthand sides of the equations in (12). (6).2 The Method of Elimination 341 Any system of two linear differential equations with constant coefficients can be written in the form L 1x + Lzy = !1 (t). L 2 = 3. and L 4 = D + 7. L 3 = 6. If so. and f 1 (t) and fz(t) are given functions. (12) and Eq. Thus we obtain the system L3L1x + L 3L2y = L 3f1 (t). Indeed.5. we could eliminate in like manner the dependent variable y from the original system in (9). In determinant notation Eqs. L1L3x + L1L4y = Lifz(t).4. (15) I· It is important to note that the determinants on the righthand side in (15) are evaluated by means of the operators operating on the functions. The equations in ( 15) are strongly reminiscent of Cramer's rule for the solution of two linear equations in two (algebraic) variables and are thereby easy to remember. (10) with L 1 = D. and L 4 are polynomial differential operators (perhaps of different orders) as in Eq. (1 3). the system in (1) (Example 1) can be written in the form (D. Alternatively. To eliminate the dependent variable x from the system in (9). L3x + L4y = fz(t).L 2L 3 appears on the lefthand side in both Eq.
lOx = 0.4)x + 3y = 0. (13) and (12) are x" + 3x'.342 Chapter 5 Linear Systems of Differential Equations Example 2 Find a general solution of the system (D.5a2e5t). • As illustrated by Example 2. but actually are not independent.10 = (r. (16) Hence Eqs. The "extra" constants must then be eliminated by substitution of the proposed general solution into one or more of the original differential equations. Solution (10) The operational determinant of this system is (D. This apparent difficulty demands a resolution. the elimination procedure used to solve a linear system frequently will introduce a number of interdependent constants that may appear to be arbitrary. and b2.2)(r + 5) = 0. so it follows that a 1 = ~b 1 and a2 = ~b2 . 6x+(D+7)y=0. Therefore. we get 0 = x ' . We can discover them by substituting the solutions in (17) into either of the original equations in (10). so their (separate) general solutions are x(t) = a1e 2t y(t) = + a 2e. the desired general solution is given by Note that this result is in accord with the general solution (Eqs.1 that the general solution of a system of two firstorder equations involves only two arbitrary constants. b1e 2t + b2eSt.st are linearly independent functions.5t. (17) At this point we appear to have four arbitrary constants a 1. (4) and (5)) that we obtained by a different method in Example 1.lOy = 0. a 2. But e 21 and e.4x + 3y = (2ale2t.4)(D + 7)  3 · ( 6) = D 2 + 3D. The explanation is simple: There must be some hidden relations among our four constants. On substitution in the first equation. that is. b 1. 0 = (2al + 3b!)e2t + (9a2 + 3b2)est. The characteristic equation of each is r2 + 3r.4(ale2t + a2e5t) + 3(b!e2t + b2e5t).10. The appropriate number of arbitrary constants in a general solution of a linear system is determined by the following proposition: . But it follows from Theorem 1 in Section 5. y" + 3y'.
On the other hand. A degenerate system may have either no solution or infinitely many independent solutions. its degree as a polynomial in D. 2Dx. For the system L11x L21x L 31X + L12y + L13Z = f1 (t). however. the equations Dx 2Dx + Dy = + 2Dy = t.2Dy = 1 with operational determinant zero are obviously inconsistent and thus have no solutions. then the system is said to be degenerate. Although the aforementioned procedures and results are described for the case of a system of two equations.L. For instance. because its operational determinant D 2 +3D. If the operational determinant is identically zero. we can substitute any (continuously differentiable) function for x (t) and then integrate to obtain y (t). see pages 144150 of E. every degenerate system is equivalent to either an inconsistent system or a redundant system. the equations Dx. the dependent variable x (t) satisfies the single linear equation L11 L12 L13 L21 L22 L23 X = L 31 L32 L 33 !1 (t) L12 L13 h(t) L22 L23 !J(t) L 32 L33 (19) with analogous equations for y = y (t) and z = z (t). Roughly speaking.Dy = 0.10 is of order 2. they can be generalized readily to systems of three or more equations.) Thus the general solution of the system in (1 0) of Example 2 involves two arbitrary constants. + L22y + L23Z = h(t).5. Example 3 illustrates this approach. The methods of this section often can be applied to analyze the "natural modes of oscillation" of a given system.2 The Method of Elimination If the operational determinant in (15) is not identically zero. then the number of independent arbitrary constants in a general solution of the system in (9) is equal to the order of its operational determinantthat is. 1956). Ince's Ordinary Differential Equations (New York: Dover. 2t with operational determinant zero are obviously redundant. 343 (For a proof of this fact. For most systems of more than three equations. the method of operational determinants is too tedious to be practical. Mechanical Vibrations A mechanical system typically vibrates or oscillates periodically in one or more specific ways. + L 32Y + L 33Z = j 3(t) (18) of three linear equations. .
y(t) = c 1 cost + c2 sin t + d 1 cos 2t + d2 sin 2t.2. So the general solutions of the equations in (21) are x(t) = a 1 cost+ a2 sint + b 1 cos2t + b2 sin2t. thus 0 = (2at . (D2 + 1)(D2 + 4)y = 0.spring system of Example 3.a2 sin t . The massand.2.4b2 sin 2t) + 3(a 1 cost + a2 sin t + b 1 cos 2t + b2 sin 2t) . (21) The characteristic equation (r 2 + 1) (r 2 + 4) = 0 has roots i. we get 0 = x" + 3x.(c1 sin t + c2 sin t + d 1 cos 2t + d2 sin 2t) . Hence the equations for x(t) and y(t) are (D 2 + 1)(D2 + 4)x = 0. (23) • .(1)(2) = D 4 + 5D2 +4 = (D 2 + l)(D 2 +4). 5. it follows that their coefficients in the last equation are zero.y = ( .b1 .dt) cos 2t + ( b2.2. 0 (20) for the displacements of the two masses in Fig. Here f(t) 0 because we assume that there is no external force. Find the general solution of the system in (20).b2 sin 2t is the desired general solution of the system in (20). and . Because the operational determinant is of order 4. cos 2t.b1 cos 2t . the general solution should contain four (rather than eight) arbitrary constants. Because cost. we derived the equations ( D 2 + 3)x 2x + (1) y = + (D 2 + 2)y = 0. y (t) = 2a 1 cost + 2a2 sin t .1. 2i . Thus Therefore x(t) = a 1 cost+ a2 sint + b1 cos 2t + b2 sin2t.i.d2) sin 2t.c2) sin t + ( .Ct) cost + (2a2. sin t.4b 1 cos 2t . When we substitute x(t) and y(t) from (22) in the first equation in (20).2i. and sin 2t are linearly independent.a 1 cost . .344 Chapter 5 Linear Systems of Differential Equations Example 3 In Example 1 of Section 5.2. = Solution The operational determinant of the system in (20) is (D 2 + 3)(D2 + 2). Equilibrium positions (22) FIGURE 5.
in which the masses move in synchrony in opposite directions. B = Jby + b~. 2 y. as do the latter two. The two masses move in the same direction. 2 cos(t.cos(2t. x ' =x + 3y. 2a 1 cost + 2a2 sin t = 2A cos(t. The expression (x(t).. .a). and tan f3 = b2/b 1• Then (25) FIGURE 5. (26).. y(t)) = a 1 (cost.{3)..2. use a computer system or graphing calculator to construct a direc tion field and typical solution curves for the given system.{3).{3) 1 2 3~~~~ 0 21t 41t with A = Jar+ ai. YI (t)) = (cos(t. b 1 cos 2t . with the constants A. Four initial conditions (typically. yJ) of Eq. a.o >( I 2 3 0 1t FIGURE 5. In Problems I through 6.3 (where a = 0) illustrates the natural mode (x 1 . sin2t) (24) then presents the general solution of the system in (20) as a linear combination of four particular solutions. but with the amplitude of m 2 twice that of m 1 (because y1 = 2x 1). 2 cost)+ a 2 (sin t. y' = 2x  3y . they exhibit its two (circular) natural frequencies w 1 = 1 and w2 = 2. If initial conditions are given. (27). y2(t)) = (cos(2t. describe the two natural modes of oscillation of the massandspring system. with the same frequency w2 = 2 and with equal amplitudes of oscillation (because y2 = . (25) represents an arbitrary free oscillation of the massandspring system as a superposition of its two natural modes of oscillation. in which the two masses move in synchrony in the same direction with the same frequency of oscillation w1 = 1. Indeed. find the particuLar solution that satisfies them.2motion subject to no external forces.4 (where f3 = 0) illustrates the natural mode (x2 . 1. the first two of these particular solutions represent physically similar oscillations of the masses.2. cos2t) +b2 (sin2t. Figure 5.B cos(2t .4.3. and b2 .5. The two masses move in opposite directions with frequency w2 = 2. 2 sin t) +b 1 (cos2t.2.b2 sin 2t = . bJ. B. (24) takes the form = a2fa 1. = 2 cos (t) a 1 cost + a2 sin t = A cos(t a)..a) and b 1 cos 2t + b2 sin 2t = B cos(2t . Find general solutions of the linear systems in Problems I through 20.x 2 ). and f3 determined by the initial conditions.2. The linear combination in Eq. where the particular solutions (x1 (t).2. y' = 2y 2y. we can (by the usual trigonometric machinations) write 3. each with frequency w 1 = 1. y2) of Eq. initial displacements and velocities) would be required to determine the values of a 1.2 The Method of Elimination 345 The equations in (23) describe free oscillations of the massandspring system of Fig.a)) (26) and (x2(t). tan a Eq. a2 . Moreover.{3)) (27) >. x' = x  2. 5. Moreover. Figure 5.
22. 23.2y. z' = x + y (Suggestion: Solve the characteristic equation by inspection. Thus linear operators with variable coefficients generally do not commute.4x = 6sint. 2x" + 3y" . Suppose that L 1 = a 1 D 2 + b 1 D + c 1 and L 2 = a 2 D 2 + b2 D + c2 .2. for a particle of mass m and electrical charge q under the influence of the uniform magnetic field B = Bk.2y. x' (O) = 0. the charged particle of Problem 35 moves with velocity v under the influence of a uniform electric field E = Ei. x(O) = 0.8y 14. x' = 3x. 33. and k2 = 25. x' = y + z + e.2y = 0 16. in addition to the magnetic field B = Bk.5. Suppose that L 1x = t Dx + x and that L 2 x = Dx + tx. z' = 4y +4z 20. x" = 4x + sint.1.y. The graph of such a cycloid is shown in Fig.1.l)x + D 2 y = 0 FIGURE 5. recall the equations of motion mx" = qBy' . x" + 13y'. x' = 3x .2. 25.2x + 2y = 0. Suppose that the initial conditions are x(O) = r 0 .z 19. x' = x + 2y + z. and y'(O) = wr0 where w = qBjm . 2x'. so that at timet = 0 there is no charge on the capacitor.4x + 6y = 0 18. y' = 4x +4y. Show that its trajectory is the cycloid Show that the systems in Problems 23 through 25 are degenerate.J3. + (D + 2) y = e. (a) Find the general solution of the equations of motion of the system.3y + 2sin2t. y' = 6x. (5) there to find the amounts x (t) and y(t) of salt in the two tanks at timet.9y = 0 17. (D 2 + l)x (D 2 .8y 15. whereas the other two are filled with fresh water. y(O) = 2 4. Then attempt to solve the system explicitly so as to find such a general solution. m 2 = 0. x' = 2x. y(O) = 0. y' = 2x + y 8.e 21 9.D 2 y = 2e.3y.D)y = 0 (D 2 + l)x .y' .5.x' = x + 3y + e 1 . y" = 2x. and that x(t) is a twice differentiable function.1. 31. Find the amounts of salt in each of the three tanks at time t . y" + 3x'. x' = 2x + y. x(O) = 3.2x = 0.1 (D 2 . y" = 4x. Show that the trajectory of the particle is a circle of radius r0 . Assume that it is closed at time t = 0. 26. y(O) = 1 11.cos2t 10. y = . 5. If. (b) Describe the natural modes of oscillation of the system.4y' = x. (9) there to find I 1 (t) and I 2 (t).21 (D + 2)x + (D + 2)y = t (D + 3)x + (D + 3)y = t 2 (D 2 + 5D + 6) x + D(D + 2)y = (D + 3)x + Dy = 0 x 0 = a(l . Repeat Problem 31.l)x 27. Assume that I 1 (0) = 2 and Q(O) = 0. 32. x " + y" . In Problems 26 through 29. except use the electrical network of Problem 27 of Section 5.3y'. Repeat Problem 31. 29.31 + (D + 3)y = e. except use the electrical network of Problem 28 of Section 5. suppose instead that m 1 = 2.2z. my" = qBx' 3. x' = 4x + y + 2t. Suppose that the electrical network of Example 3 of Section 5.1 initially (t = 0) is 0. then the force acting on it is F = q(E + v x B). x".l)x + D 2 y = 0 (D 2 + D)x + D 2 y = 2e.1 (D 2 . 28. y' = x + z . 5 . y' = 3x + 4y. z' = x. x' + 2y' = 4x + 5y.1 12. first calculate the operational determinant of the given system in order to determine how many arbitrary constants should appear in a general solution. x' = 3x + 2y. (D + 2)x (D + 3)x 24.1 is initially openno currents are flowing. 37.y' = 3x.3x' .9x' . y' = 2x + y 6.15y + e. . (Suggestion: Examine the equations to be derived in Problem 26 of Section 5. x(O) = 1.4y.) 21. 2y'.cos wt) .1. x" = 6x + 2y. In the massandspring system of Example 3.1.y. Three 100gal brine tanks are connected as indicated in Fig.1.) 35.1. Verify that L 1 L2x = L 2 L 1x. + D 2y = 2e. y(O) = 2 7. In each problem determineby attempting to solve the systemwhether it has infinitely many solutions or no solutions.sin wt) where a = Ej(wB) and w = q Bjm.5. 36.2y' .a(wt . x(O) = 1. From Problem 31 of Section 5. y' = 2x.5y. x" = 5x + 2y. Assume that the particle starts from rest at the origin. where the coefficients are all constants. The cycloidal path of the particle of Problem 36. y' = x. y' = 5x. In particular.13 of Section 5. Suppose that the salt concentration in each of the two brine tanks of Example 2 of Section 5. Show that L 1 L 2 x =/= L 2 L 1x. y(O) = 1 5.1 ( D 2 . x' =x +9y.346 Chapter 5 Linear Systems of Differential Equations 30. x' = 4x.5 lbjgal.2x' . show that its natural frequencies are w 1 = 5 and Wz = 5. Then solve the system in Eq. solve the system in Eq.1 + D2y = 0 (D 2 + l)x + (D 2 + 2)y = 2e.2y. y' = x + 2y .l)x + (D 2 . y" = 3x + 7 y 13. 3x' . k 1 = 75. Assume that the first tank initially contains 100 lb of salt. y". 34.
The mechanical system of Problem 38.2 suffice for the solution of small linear systems containing only two or three equations with constant coefficients. m 2 = 1. ~ X y Z + k3)y. mx" = 2kx my" m z" + ky.are introduced as needed in subsequent sections of this chapter. ~ In>· 111 > j. k 3 = 2 m. = 1. k. k2 = 4. = 4.2y' + 3x = 0. this section begins with a complete and selfcontained account of the matrix notation and terminology that is needed.m2 = 1.2ky + kz. = 1. w2 = J2 .Matrices and L~~ea~_ ~y~~~~s lthough the simple elimination techniques of Section 5. 0) X = = kx . BJ.k2 = 2. Show that the natural fre quencies of oscillation of the system are w. In Problems 39 through 46. If P begins at A(a. k2 = 4. k2 = 6. = .are most easily and concisely described using the language and notation of vectors and matrices. x(O) = 4.'.2. 5. those associated with eigenvalues and eigenvectors. 0) when t = 0. = 2. 43. Consider the system of two masses and three springs shown in Fig. Verify that these equations describe the hypocycloid traced by a point P (x. derive the equations of motion m.x" = (k..3 Matrices and Linear Systems 38. = 1.k2 =4.k 1 = 1.as well as solution methods suitable for larger systems. Suppose that the trajectory (x(t). k. m 2 = 2. y" + 2x' + 3 y = 0. m2 = 2. 47. 42. k 3 = 0 m.2. Solve this problem.k3 = 1 m. (b) Assume that m = k = 1.6.3 and 5.m 2 = 1.2.8. 46. k3 = 0 m 1 = 2. y(t) = 3 sin t sin 3t. ky . 41.2.5. 44. = 4. k. k 1 = 100. m2y" = 347 + k2)x + k2x . y(t)) of a particle moving in the plane satisfies the initial value problem Equilibrium positions FIGURE 5. For ready reference and review. 40. k 2 =50. y ) fixed on the circumference of a circle of radius b = 1 that rolls around inside a circle of radius a = 4.7. The mechanical system of Problem 47.k3 = 2 m. = 1. y m. Special techniques of linear algebraspecifically. Derive the equations of motion m.4 ). = 8. find the general solution of the system in Problem 38 with the given masses and spring constants.J2. and w 3 = J2 + . 5.J2. x(t) y(t) 48. kz = 2. 5. the general properties of linear systems. Find the natural frequencies of the massandspring system and describe its natural modes of oscillation.k2 = l. 45. A(a.6.2.8.2kz . A . s FIGURE 5. FIGURE 5.k.2. You should obtain x(t) = 3cost +cos3t .m2 = 1. The hypocycloid of Problem 48.2.k 1 = 2. 39. k. y (O) = x'(O) = y'(O) = 0. Use a computer system or graphing calculator to illustrate the two natural modes graphically (as in Figs.7. = 1.2. m2 = 2. x".J2.(k2 k2y.k3 = 1 m 1 = 1. = 1. then the parameter t represents the angle AOC shown in Fig. k3 = 4 (a) For the system shown in Fig. m 2 =I. 5. k3 = 4 = 1.
Two m x n matrices A = [ a ij ] and B = [ bij ] are said to be equal if corresponding elements are equal~ that is. To multiply the matrix A by the number c. OJ • . but the single symbol 0 will suffice for all these zero matrices. (3) Thus the element in row i and column j of C = A+ B is Cij = aij + bij. ~l 5  11 ~] . by (2) Actually for each pair of positive integers m and n there is an m x n zero matrix. (4) Example 1 If A=[~ then 3J 7 ' B= A + B = [~ and [13 ~~l c 3J [13 10 J. we simply multiply each of its elements by c: cA = Ac = [caij]. Sometimes we use the abbreviation A = [ aij ] for the matrix with the element aij in the ith row and jth column.7 OJ . if aij = bij for 1 ~ i ~ m and 1 ~ j ~ n. each entry of which is zero. (1).[11 7 and 7 + 7 = [. We add A and B by adding corresponding entries: A + B = [ a ij ] + [ b ij ] = [ aij + bij ] . as in Eq. We denote the zero matrix.[18 30 42 .348 Chapter 5 Linear Systems of Differential Equations Review of Matrix Notation and Terminology An m x n matrix A is a rectangular array of mn numbers (or elements) arranged in m (horizontal) rows and n (vertical) columns: all a21 a31 a12 azz a32 a13 a23 a33 Gtj azj a3j Gtn azn a3n A= ail aiz ai3 aij Gin (1) Gmt Gm2 Gm3 Gmj Gmn We will ordinarily denote matrices by boldface capital letters.
.5. bn ] ' it is understood that a 1 . aij + bij = bij + aij for all i and j because addition of real numbers is commutative. such as c = [ 5 17 0 3].3 Matrices and Linear Systems 349 We denote ( 1)A by A and define subtraction of matrices as follows: A.•• . (9) Each of these properties is readily verified by elementwise application of a corresponding property of the real numbers.one having only a single row. we will frequently write a column vector as the transpose of a row vector. The transpose AT of them x n matrix A = [aiJ] is then x m (note!) matrix whose jth column is the jth row of A (and consequently. Consequently. and am are the row vectors of the matrix A and that b1. For aesthetic and typographical reasons. (5) The matrix operations just defined have the following properties. . (c+d)A =cA+dA. a row vector is a 1 x n matrix. a 2 . Thus AT = [ aJi although this is not notationally perfect. you must remember that AT will not have the same shape as A unless A is a square matrixthat is. . or simply a vector. A + B = [ aij + biJ J = [ biJ + aij J= B + A. each of which is analogous to a familiar algebraic property of the real number system: A+O=O+A=A.B =A+ (B). Thus if we write and B = [ b1 . . (distributivity) (6) (7) (8) A + (B + C) = (A + B) + C c(A +B) = cA + cB. (commutativity). . (associativity). An m x 1 matrixone having only a single columnis called a column vector. for example. and bn are the column vectors of the matrix B. whose ith row is the ith column of A). unless m = n. as in J. . b2.. For example. A+B=B+A AA=O. We often denote column vectors by boldface lowercase letters. . or x= Similarly. the two preceding column vectors may be written in the forms b= [3 7 0 r and X = [X] X2 x mr Sometimes it is convenient to describe an m x n matrix in terms of either its m row vectors or its n column vectors.
The product AB of two matrices is defined only if the number of columns of A is equal to the number of rows of B. It may help to think of "pouring the rows of A down the columns of B. (12) For purposes of hand computation.350 Chapter 5 Linear Systems of Differential Equations Matrix Multiplication The properties listed in Eqs. The first surprises in the realm of matrix arithmetic come with multiplication. Example 2 Check your understanding of the definition of matrix multiplication by verifying that if . (6) through (9) are quite natural and expected. ai + bpi bp2 am! a m2 a mp f?pJ bpn t b· J which shows that one forms the dot product of the row vector ai with the column vector b 1 to obtain the element ciJ in the ith row and the jth column of AB. a 21 a12 a22 a1p a2p bll b2! b12 b 22 btj bi n b2n hz · ·. (11) can be ciJ = L k= I aikbkJ. If A is an m x p matrix and B is a p x n matrix. each having the same number p of elements. (11) and (12) is easy to remember by visualizing the picture a. then their product AB is the m x n matrix C = [ ciJ ] . the definition in Eqs. We define first the scalar product of a row vector a and a column vector b. Thus C = AB = [ ai · b1 ] . and B (11) In terms of the individual entries of A recast in the form = [ aiJ] p = [ biJ ]. (10) exactly as in the scalar or dot product of two vectorsa familiar topic from elementary calculus. where ciJ is the scalar product of the ith row vector ai of A and the jth column vector b1 of B.1." This also reminds us that the number of columns of A must be equal to the number of rows of B. If then a · b is defined as follows: p a· b = Lakbk = a 1b 1 k=I + a2b2 + · · · + apbp. .. Eq.c: .
Inverse Matrices A square n x n matrix is said to have order n. if A and B are both n x n matrices (so that both the products AB and BA are defined and have the same dimensionsn x n ). it can happen that AB = 0 even though A =f. AB Moreover. 0.3 Matrices and Linear Systems 351 then AB = [ Similarly.2z 6x 7y 15 23 3J 71] 1 3 5 .5. (15) =f. 0 and B =f. But matrix multiplication is not commutative. that is. although you can easily construct your own examples using 2 x 2 matrices with small integral elements. then. The identity matrix of order n is the square matrix I= 0 0 1 0 0 0 1 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 1 (17) 0 0 . verify that i . (16) Examples illustrating the phenomena in (15) and (16) may be found in the problems. A(BC) = (AB)C and A(B + C) = AB + AC. • It can be shown by direct (though lengthy) computation based on its definition that matrix multiplication is associative and is also distributive with respect to matrix addition. ] [ ~ ~ J 1 = [ 1 ~ ~~ l ] 3 5 7 and that 2 0 1] [x] [ y = z 2x+z 4x + 53y y . in general. That is. BA. (14) (13) provided that the matrices are of such sizes that the indicated multiplications and additions are possible.
1• It is clear that some square matrices do not have inversesconsider any square zero matrix. In linear algebra it is proved that A . If A = [ aij J is ann x n matrix. then this inverse is unique. It is also easy to show that if A I exists. the results are the same in all 2n cases.1 exists and (A I ). let Aij denote the (n .1 exists if and only the determinant det(A) of the square matrix A is nonzero. then an inverse of A is a square matrix B of the same order as A such that both AB = I and BA = I . The expansion of the determinant IAI along its ith row is given by n IAI = L(l)i+jaijiAijl j=1 (i fixed). we may speak of the inverse of A. (20a) and its expansion along its jth column is given by n IAI = L(l)i+jaijiAijl i= 1 (j fixed). (20b) It is shown in linear algebra that whichever row we use in Eq.1 =A. and we will denote it by A.1) x (n . Determinants We assume that the student has computed 2 x 2 and 3 x 3 determinants in earlier courses. the matrix A is said to be nonsingular. Consequently. If A [ aij J is a 2 x 2 matrix.1 • Thus AA. then A is called a singular matrix.352 Chapter 5 Linear Systems of Differential Equations for which each entry on the principal diagonal is 1 and all offdiagonal entries are zero. It is not difficult to show that if the matrix A has an inverse. If A is a square matrix.1A . . Hence IAI is well defined by these formulas. (20a) and whichever column we use in Eq.1 =I= A.1). if det(A) = 0. If so. then its determinant det(A) = IAI is defined as Determinants of higher order may be defined by induction. (20b). It is quite easy to verify that AI= A= lA (18) for every square matrix A of the same order as I. as follows. (19) given the existence of A . then (A .1) matrix obtained from A by deleting its ith row and its jth column.
11. 11 = And the expansion of IAI along its third column is IAI = . (21) is obtained from A by interchanging the diagonal elements and changing the signs of the offdiagonal elements. 6 2 3 • . 11 + 5. 1 1 ·I ~ j I+ 5 ·I ~ ~ I = 2. ·I 2 3 1 3 = 4. For instance.2 ·1i .be =I= 0. Example 4 If A=[~~].5 · 8 = 2. then its inverse matrix is A 1 = 1 [ !AI c d b] a . 11 + 2. but determinants and inverses of 2 x 2 matrices are easy to compute by hand. then lA I = 6 · 7. • Calculators and computers are convenient for the calculation of higherdimensional determinants and inverse matrices.3 Matrices and Linear Systems 353 Example 3 If A= [2~ then the expansion of IAI along its second row is lA I = 4 ·11 3 21 5 +2 ·I 2 3 211 5 33. if the 2 x 2 matrix has nonzero determinant IAI =ad. (21) gives At=~ [7 2 5 You should pause to verify that 8] [ ! 4].1 . Hence Eq. (21) Note that the matrix on the righthand side of Eq.5. 16. 2 = 33.1 .
it is important not to reverse the order of the factors in Eqs. J. • (24) The differentiation rules (A+B) = .(cA) = c . that is. (23) Example 5 If x(t) = [ ~'] and A(t) = [ sint t cost 1 then dx dt [~·] d dt d dt and A1 (t) = [COS f 1 dB dt smt 0 . is a matrix such as Xi (t) x(t) = X2(t) (22a) Xn(t) or au (t) a12(t) a22(t) ain(t) a2n (t) A(t) = a21 (t) (22b) ami (t) am2(t) amn (t) in which each entry is a function of t. dt dt and d dA (AC)= C. We say that the matrix function A(t) is continuous (or differentiable) at a point (or on an interval) if each of its elements has the same property.+dt dA and (AB) =A+ . A' (t) = dA dt = [ daij dt J.. or simply matrix function.. .354 Chapter 5 Linear Systems of Differential Equations MatrixValued Functions A matrixvalued function. If c is a (constant) real number and C is a constant matrix. J. (25) and (26). The derivative of a differentiable matrix function is defined by elementwise differentiation. then . dt dt d dA d dA (CA) =C.B dt dt dB dA (25) follow readily by elementwise application of the analogous differentiation rules of elementary calculus for realvalued functions. dt dt (26) Because of the noncommutativity of matrix multiplication.
A solution of Eq. x~ = 6x 1 . especially those computations that would be burdensome in scalar notation. The matrix notation used in Eq.7xz can be written as the single matrix equation To verify that the vector functions Xt (t) 3e2t = [ 2e2t J and x2 (t) = [ 5t 3: _ 51 J . X~ = P2I (t)Xt + P22(t)xz + · · · + P2n (t)Xn + /z(t). (28) on the open interval I is a column vector function x(t) = [Xi (t) J such that the component functions of x satisfy the system in (27) identically on I.5. We discuss here the general system of n firstorder linear equations + Ptz(t)xz + · · · + Pin(t)Xn + /J(t). Example 6 The first order system x~ = 4xt .3xz. (28) not only emphasizes this analogy. then the system in (27) takes the form of a single matrix equation  dx dt = P(t)x + f(t). but also saves a great deal of space. Our main use for matrix notation will be the simplification of computations with systems of differential equations. X~= P3t(t)Xt + P3z(t)xz + · · · + P3n(t)xn + /3(t).1 guarantees the existence on I of a unique solution x(t) satisfying preassigned initial conditions x(a) = b. but it is readily assimilated with practice. X~= Pll(t)Xt (27) X~= Pni(t)XI + Pnz(t)xz + · · · + Pnn(t)Xn + fn(t). then Theorem 1 of Section 5. (28) We will see that the general theory of the linear system in (27) closely parallels that of a single nthorder equation. P(t) = ( Pij(t)] If we introduce the coefficient matrix and the column vectors x= [ Xi ] and f(t) = [ fi (t) ] . If the functions Pij (t) and /i (t) are all continuous on I.3 Matrices and Linear Systems 355 FirstOrder Linear Systems The notation and terminology of matrices and vectors may seem rather elaborate when first encountered.
. c11 are constants. then the linear combination (31) is also a solution of Eq. and such that every solution of Eq.2. x2 . (29) on I.. so it follows = = + c2x. We expect it to have n solutions x 1 . (28). x' = P(t)x. . Theorem 1 is analogous to Theorem 1 of Section 2. THEOREM 1 Principle of Superposition Let x 1. but with f(t) 0. + · · · + CnX~ CJP(t)Xt + c2P(t)x2 + · · · + CnP(t)X P(t)(CJXJ + C2X2 + · · · + C X 11 11 ). Xn that are independent in some appropriate sense. The remarkable simplicity of this proof demonstrates clearly one advantage of matrix notation. dt (29) which has the form shown in Eq. let us write Xtj(t) = (30) Thus Xij (t) denotes the ith component of the vector xj (t). .356 Chapter 5 Linear Systems of Differential Equations are both solutions of the matrix differential equation with coefficient matrix P. c2 .. . we consider first the associated homogeneous eq nation dx . . (29). x 2 .= P(t)x. If c 1 . . x2. as desired. . (28). 11 That is. so the second subscript refers to the vector function Xj (t). . Xn of Eq.A. . . we need only calculate Px I =[4 6 3 ] [ 3e21 7 2e2t ] =[ 6e 21 4e2r ] = x1 1 and • To investigate the general nature of the solutions of Eq. whereas the first subscript refers to a component of this function.. Proof: We know that x. Xn ben solutions of the homogeneous linear equation in (29) on the open interval I. •. (29) is a linear combination of these n particular solutions.. •. Given n solutions x 1. immediately that x' = CJX~ P(t)xi for each i (1 < < n).
If x 1.2.. its proof is essentially the same. en not all zero such that C[XJ (t) + CzXz(t) + · · · + CnXn (t) = 0 (32) for all t in I.51. Xz. We may write either W(t) or W(x 1 ..2).5. . Xz(t) + 3c2e. x 2 . . they are linearly independent.. In scalar form with x = [x 1 X! V.51 ] is also a solution. Xn.•• . there is a Wronskian determinant that tells us whether or not n given solutions of the homogeneous equation in (29) are lir:tearly dependent. For instance. .e 21 + c 2e. x 2 . with the definition of W (x. [ 2ez1 x2 3e21 ] + cz [ 3e51 e . Just as in the case of a single nthorder equation. Xn) in Eq.. neither is a scalar multiple of the other. . . . .2. Equivalently. Otherwise.51. e21 = 2c. The vectorvalued functions x 1. x2 . . (33) substituted for the definition of the Wronskian of n solutions of a single nthorder equation (see Problems 42 through 44). Xn are such solutions. . •• • .. then the linear combination x(t) = c 1x 1 (t) + czxz(t) = c. . this gives the solution (t) = 3c. Xn are linearly dependent on the interval I provided that there exist constants c 1. then their Wronskian is the n x n determinant X11 (t) XJ2(t) X in (t) x21(t) xzz(t) Xzn(t) W(t) = X n! (t) Xnz(t) (33) using the notation in (30) for the components of the solutions. . x2 .3 Matrices and Linear Systems 357 Example 6 Continued If x 1 and x2 are the two solutions of dx _ [ 4 dt 6 3] 7 X discussed in Example 6. c2 . Moreover. Xn).. the two solutions x 1 and x 2 of Example 6 are linearly independent because. . they are linearly independent provided that no one of them is a linear combination of the others. Note that W is the determinant of the matrix that has as its column vectors the solutions x 1 . Independence and General Solutions Linear independence is defined in the same way for vectorvalued functions as for realvalued functions (Section 2. which is equivalent to the general solution we found by the method of elimination • in Example 2 of Section 5. clearly.. Theorem 2 is analogous to Theorem 3 of Section 2.
. en such that (35) for all t in I. Xn Suppose that XI. and x3 are • Theorem 3 is analogous to Theorem 4 of Section 2. • If XI. . c2. . .2. x 2 .358 Chapter 5 Linear Systems of Differential Equations THEOREM 2 Wronskians of Solutions . which is never zero. . or W = 0 at no point of I. Xn be n linearly independent solutions of the homogeneous linear equation x' = P(t)x on an open interval I . .•.. Xn. . Thus there are only two possibilities for solutions of homogeneous systems: Either W = 0 at every point of I. are n solutions of the homogeneous linear equation x' = P(t)x on an open interval I. then W = 0 at every point of I. It says that a general solution of the homogeneous n x n system x' = P(t)x is a linear combination (35) of any n given linearly independent solutions XI. XI .. e3r are solutions of the equation (34) The Wronskian of these solutions is W= 2et 2e 31 2e1 0 et e3r 2 2 2 2 0 2 1 1 1 = 16e91 .. . then there exist numbers CI. x 2 . Suppose also that P(t) is continuous on I. x 2 . x2. . Hence Theorem 2 implies that the solutions linearly independent (on any open interval). x 2 .. THEOREM 3 General Solutions of Homogeneous Systems Let xI.••" " . where P(t) is continuous. Xn are linearly dependent on I . . <  N Oo <<•. .~''" "'''"''' '"' W Example 7 It is readily verified (as in Example 6) that X2(t) =[ 2e 03r ] . If x(t) is any solution whatsoever of the equation x' = P(t )x on I . x2.•• .. then W f= 0 at each point of I.' "' """ ' ' " " ' '' '" o . Xn are linearly independent on I. .• . Let Then: • If XI.
note that the given solution x(t) and the solution y(t) of Eq. This establishes Eq.1 that x(t) = y(t) for all t in I. x2 . It suffices to choose for Xj (t) the unique solution such that 0 0 0 Xj(a) = 0 1 +.. Xn.& Remark: Every n x n system x' = P(t )x with continuous coefficient matrix does have a set of n linearly independent solutions x 1 . Finally.1x(a) satisfies Eq. as desired. that is. . x2 .. We show first that there exist numbers c 1. . . . .5.. .1 • Therefore the vector c = X(a). c2.. such that (37) Let X(t) be then x n matrix with column vectors x 1. . It follows from the existenceuniqueness theorem of Section 5. (36)with the values of ci determined by the equation c = X(a). Xn are linearly independent. x2 . . . • . and let c be the column vector with components c 1. . en. (37) may be written in the form (38) X(a)c = x(a). (In other words. Xj(a) is merely the jth column of the identity matrix.. Cn such that the solution (36) has the same initial values at t =a as does the given solution x(t). (38). •.. . Xn are linearly independent by Theorem 2.3 Matrices and Linear Systems 359 Proof: Let a be a fixed point of I.. Then Eq.. c2 . (35).1x(a)have the same initial values (at t = a). Xn as in the hypotheses of Theorem 3.position j 0 0 that is. Hence the matrix X(a) has an inverse matrix X(a) . How actually to find these solutions explicitly is another matterone that we address in Section 5. (35) of the homogeneous linear system x' = P(t)x can be written in the form x(t) = X(t)c. . The Wronskian determinant W (a) = IX (a) I is nonzero because the solutions x1.. (39) . the column vector with all elements zero except for a 1 in row j. • Initial Value Problems and Elementary Row Operations The general solution in Eq. x2 .) Then so the solutions x 1 .4 (for the case of constant coefficient matrices).
it suffices to solve the system · · · bn r is given. until the first equation is finally solved for x r. constant vector b = [bi ]. First the last equation in (44) is solved for Xn. 2. n). The transformed system is then easily solved by the process of back substitution. (44) in which only the unknowns Xj. Xn appear explicitly in the jth equation (j = 1. Xj+I · . .. arrxr (43) with nonsingular coefficient matrix A = [aij ]. The transformation of the system in (43) to upper triangular form is most easily described in terms of elementary row operations on the augmented coefficient matrix all ar2 (A!h] = a2r a22 an2 arn br a2n b2 ann bn (45) ani that is obtained by adjoining the vector b to the matrix A as an additional column.I. The admissible elementary row operations are of the following three types: . . . a2IXI + a22X2 + · · · + a2nXn = b2. .•• . x 2 . Xn. x(a) = b. Cn in Eq. The basic idea is to transform the system in (43) into the simpler upper triangular form aiiXI + a 12X2 + · · · + arnXn = br. c2 . (39). according to (42) X(a)c = b to find the coefficients c 1 . and unknowns x 1 . then the nexttolast is solved for Xn. We therefore review briefly the elementary technique of row reduction to solve an n x n algebraic linear system + ar2x2 + · · · + arnXn = br. Xn. and so forth.360 Chapter 5 Linear Systems of Differential Equations where X(t) = [ x 1 (t) X2(t) Xn (t)] (40) is then x n matrix whose column vectors are the linearly independent solutions x 1. a22X2 + '' ' + a2nXn = b2.. (41) where the initial vector b = [ br b2 Eq.• .... . . (35).•• . and C = [ Ct C2 Cn is the vector of coefficients in the linear combination r (35) Suppose now that we wish to solve the initial value problem dx dt = Px. X2. Then.
Multiply any (single) row of the matrix by a nonzero constant. 3.1 1 1 : 6 I : i OJ 1 . The process of transforming [ A ! b Jis carried out one column at a time. In scalar form.C 2 + C3 = 6 with augmented coefficient matrix 2 2 2 ! 0 2 2 [ 2 1 . This upper triangular augmented coefficient matrix then corresponds to an upper triangular system as in (44).5. Example 8 Use the solution vectors given in Example 7 to solve the initial value problem dx dt Solution = [ ~ 0 1 2 OJ 3 2 3 X. from left to right. Ct .1 1 : 6 Multiplication of each of the first two rows by i OJ .2c3 = 2. X3(0) = 6. = X3(t) = Cte1  We seek the particular solution satisfying the initial conditions X! (0) = 0. we get the algebraic linear system 2ct + 2c2 + 2c3 = 0. . this gives the general solution Xt(t) = 2cte 1 x 2(t) 2cte 1 + 2c2e31 + 2c 3e51 .2c3e51 . ! gives [i 1 1 0 . as in the next example. Interchange any two rows of the matrix. x(O) = [H (46) It follows from Theorem 3 that the linear combination is a general solution of the 3 x 3 linear system in (46). . in tum) to transform [A ! b] into an upper triangular matrix. one that has only zeros beneath its principal diagonal. The goal is to use a sequence of such operations (one by one. C2e 31 + c3e51 . X2(0) = 2. 2ct . 361 2. When we substitute these values in the three preceding scalar equations. Subtract a constant multiple of one row from any other row.3 Matrices and Linear Systems 1.
Now we multiply the second row by 1.. . (47) The following theorem is analogous to Theorem 5 of Section 2. If x(t) is any solution whatsoever of Eq. We finally solve in tum for c3 = 1. (47) has the form x(t) = Xc(t) + Xp(t). 4c3 = 4. (47) and the complementary function Xc(t) is a general solution of the associated homogeneous equation x' = P(t)x.. and c 1 = 2. x2. (48) where xp(t) is a single particular solution ofEq. Theorem 4 means that the general solution of Eq. . In brief. . Let x 1. . c2.2. Xn be linearly independent solutions of the associated homogeneous equation on I. . en such that (49) for all tin I. . (t)  3x2(t) + x 3(t) = 4e1 [ 4e 1 2er  6e 31 + 2e 5' 3e3r + esr  + 2e 51 ] • • Nonhomogeneous Solutions We finally tum our attention to a nonhomogeneous linear system of the form dx dt = P(t)x + f(t). then add twice the result to the third row.. then there exist numbers c1. THEOREM 4 Solutions of Nonhomogeneous Systems Let Xp be a particular solution of the nonhomogeneous linear equation in (47) on an open interval I on which the functions P(t) and f(t) are continuous. c 2 = 3.2 and is proved in precisely the same way. Thereby we get the upper triangular augmented coefficient matrix [~ CI 1 1 0 2 1: OJ !1 I 4 : 4 that corresponds to the transformed system + C2 + C3 = c2 + 2c3 = Q. 1. Thus the desired particular solution is given by x(t) = 2x.362 Chapter 5 Linear Systems of Differential Equations then subtraction of the first row both from the second row and from the third row gives the matrix 1 1 1 [ 0 1 2 0 2 0 The first column of this matrix now has the desired form. substituting the preceding theorems in this section for the analogous theorems of Section 2. (4 7) on I.
In Example 7 we saw that a general solution of the associated homogeneous linear system dx dt is given by [! 23 2 OJ 1 3 X and we can verify by substitution that the function (found using a computer algebra system. xz(t) = 2c.15.e1 + 2c2 e 31 + 2c3e 51 + 3t. 2. Finding the general solution Xc(t) of the associated homogeneous system. by (t) = 2c.e'  Cze 3' + c3e 51 + 2t.5.8) is a particular solution of the original nonhomogeneous system. XJ x3(t) = c. Finding a single particular solution Xp (t) of the nonhomogeneous system.e1 . The sum x(t) = Xc(t)+xp(t) will then be a general solution of the nonhomogeneous system.2c 3 e 51 + 5. . Example 9 The nonhomogeneous linear system xi = 3x. Consequently.   xz + 3x3  6t + 7 is of the form in (47) with OJ P(t)=[ 2 1 3 3 2 ! ' f(t) = [ 9t + 13 7t.15 6t + 7 J. Theorem 4 implies that a general solution of the nonhomogeneous system is given by x(t) = Xc(t) + Xp(t). that is. • . 2x3 + 7t .2x2 x~ = x 1 + 3x2 x~ =  9t + 13 . or perhaps by a human being using a method discussed in Section 5.3 Matrices and Linear Systems 363 Thus finding a general solution of a nonhomogeneous linear system involves two separate steps: 1.
sint x' = y + z. ~e5' = det(A) · det(B) n for any two square matrices A and B of the same order? 8. 15. z' = e. (c) AB. x~ = 3x4. 25. 11. x' = [ ~ _ ~ Jx. 16.2B. z' = 5y .3y. y' = z + x.e1 y +cost. c= 3. 18. = X2 + X3 + 1. y' = x + y + 2z. if A 1B = A2B and B ::1 0.5B. (b) Let A = A 1 . 14. write the given system in the form x' = P(t)x + f(t). . Then use the Wronskian to show that they are linearly independent. X1 1 = e 31 [ ~ J. Suppose that A and B are the matrices of Problem 5. ~ [ . first verify that the given vectors are solutions of the given system. A(t) . 17.=[ ~ .A 2 and use part (a) to show that AB = 0.364 Chapter 5 Linear Systems of Differential Equations BJ 1.y . ~ e'' [ n. = [. x' = [ ~ ~ Jx. x2 J = e .J. Let 3 22. (b) 3A. 7.J. 2. 12.1x + t 2y. (c) AB. 19. 21. 6. 5. y' = x. X~ = X3 + X4 + t. (d) BA. . = x 2. (e) A. Are your results consistent with the theorem to the effect that det(AB) 24. x~ = 4x.tl. x' = [: = [ . verify the product law for differentiation. A2= [ ~ _ . X2 = [ 3 :=~: ] 26••• ~ [ ! =~ n . Verify that det(AB) = det(BA). Let Problems and B = [. z' = x + y x' = 2x. = [ ~:: l x2 = [ _:~: J Find (a) 7A + 4B. Thus the product of two nonzero matrices may be the zero matrix.2y. Finally. y' = e. write the general solution of the system.7z x' = 3x. x' = 3y. that is.3z + t 2. Let A and B be the matrices given in Problem 3 and let and y= [ s:~ t ] .. A(t) = t 8t In Problems I 1 through 20. Find AB and BA given 0 [ ~ ~ and l 4 B ~ [ ~ 2 ~]cost 4. z' = 6y 7z + t 3 x' = txy+e1 z. x2 = e [ _. (b) 3A . x~ = x 1 + x2 + t 3 Find Ay and Bx. 20.21 [ 21 1 5 J J B=[~ ~l (a) Show that A 1B = A2B and note that A 1 =I A2. it does not follow that A 1 = A2. x' = [ _ 3 23. 1 _ 3 x. x~ = x 1 + x 4 + t 2. Let A~ [ J ~ n and B ~ [! ~ = H In Problems 21 through 30. y' = 3x x' = 3x.:~: ]. 4] 1 9. Are the products Ax and By defined? Explain your answer. Verify that (a) A(BC) = (AB)C and that (b) A(B+C) = AB + AC.. 13. Thus the cancellation law does not hold for matrices.t 2 x' = tx. x. (d) BA. Compute the determinants of the matrices A and B in Problem 6. (AB)' = A'B + AB'. x. Find (a) 2A + 3B.4y + z + t. x~ = 2x3. where A and B are the matrices given in Problem 1 and et [ 10. x 1 = e 2t J [ 1 1 . x =~ Jx. x' A. ~ e· [H [ x. y' = 5x .. y' = 2x + y x' = 2x + 4 y + 3e1 .1x+3ty+t 3z x. y' = 2x+t 2yz. In Problems 9 and 10.
x 2 (0) = 2. x 2 (0) = 12. x' = [ 1 ! i 2 x. where the 2 x 2 matrix P(t) is continuous on the open interval I. righthand side the 3 x 1 column vector B = [ 0 2 6 ] T. Once the matrices A and B have been entered. ~ e' [  lJ ~ i] x3 e" [  6 2 are linearly independent on the real line.. is a constant multiple of the other on the open interval /. 2. and unknown column vector c = [ Cz C3 Figure 5.] 1 = [ t~ J and Xz = [ ~~ J = e. Suppose that the vectors x 1 (t) and x 2 (t) of Problem 42 are solutions of the equation x' = P(t)x. Solution of Linear Systems 5. (a) Show that the vector functions x. TI86 solution of the system AC = Bin (1).. that can be written in the form AC = B with 3 x 3 coefficient matrix A. For instance. x' ~ 6 ! . x 1 = [ x. x.3. Xn(t) be vector functions whose ith components (for some fixed i) Xi! (t).. that x 1 and x2 are linearly dependent. with the result that c 1 = 2. . Generalize Problems 42 and 43 to prove Theorem 2 for n an arbitrary positive integer. Xin(t) are linearly independent realvalued functions. 45. XJ(O) = 4 36. x 2 (0) = 7. The system of Problem 30: x 1 (0) = x 2 (0) = x 3 (0) x4(0) = 1 40. . Suppose that one of the vector functions and x 2 (t) = [ Xtz(t) xn(t) 0 2J 12 4 1 4 _0 1 0 _0 6 x.. .. that is. ~ e_.21 [ _. 39. x 2 (0) = 3.]. (b) Why does it follow from Theorem 2 that there is no continuous matrix P(t) such that x 1 and x 2 are both solutions ofx' = P(t)x? 42.x 3 = e [ ~ . Let x 1(t). x 4 = e [lJ ~ [6 t t In Problems 31 through 40.3.~ ~~~ . The system of Problem 22: x 1 (0) = 0. then there exist numbers c 1 and c2 not both zero such that c 1x 1 (a)+ c2 x 2 (a) = 0. Then conclude from the uniqueness of solutions of the equation x' = P(t)x that for all t in I. X3(0) = 3 38. 2] [ 1 • 1 . ~e = [ 3' [ jl 1 1 6] x. x 2 (0) = 5 The system of Problem 23: x 1 (0) = 5. xi 2 (t). 33. x 1 n 13 ! ]. and c3 = 1.3 Matrices and Linear Systems 365 28. ~ . X4(0) = 7 41.·. x2 (t). 2] [ 2. The system of Problem 27: x 1 (0) = 10. C3 = 6 (1) [ [ 01 [2] Cz [ 6] ] + [ [2 ] [ 3] [1 ] ] FIGURE 5. Conclude that the vector functions are themselves linearly independent... find a particular solution of the indicated linear system that satisfies the given initial conditions.5. 35. x 2 (0) = 0 The system of Problem 26: x 1 (0) = 0. 34. . . . c . x 2 (0) = 7 The system of Problem 25: x 1 (0) = 8. 0." .1 shows a TI calculator solution for = A I B.1 l H·A [ [2 2 2 ] [ £01 £21 £61 H·B [2 0 2] [1 1 1 ]] Linear systems with more than two or three equations are most frequently solved with the aid of calculators or computers. Show that if there exists a point a of I at which their Wronskian W(a) is zero. This proves part (a) of Theorem 2 in the case n = 2. r. (t) 29. 1 Xz =e 1 0 1 0 OJ OJ . The system of Problem 30: x 1(0) = 1. c2 = 3.1. [ x. recall that in Example 8 we needed to solve the linear system 2c 1 + 2cz 2c 1 CJ + 2c3 =  0. x 1 = e 1 [1J 0 J 1 0 . 44. x 2 (0) = 3 The system of Problem 24: x 1 (0) = 11. 43. 32. XJ(O) = 1 37. the same result can be found using the Maple command c. [ [ 2. The system of Problem 29: x 1 (0) = 1. 2c3 = 2. This proves part (b) of Theorem 2 in the case n = 2. XJ(O) = 11 31. X3(0) = 4. The system of Problem 29: x 1 (0) = 5. Show that their Wronskian W(t) = l[xij(t)]l must vanish identically on I. x 2 (0) = 0.3 Application Automatic ..
the linear combination (2) with arbitrary coefficients will then be a general solution of the system in (1). the Mathematica command C = Inverse[A]. . . VI. Vn in Eq.B).t V3eAt VI vz V3 eAt x(t) = X3 = veAt (3) Xn VneAI Vn where A. . (1) By Theorem 3 of Section 5. we know that it suffices to find n linearly independent solution vectors XI. . we proceed by analogy with the characteristic root method for solving a single homogeneous equation with constant coefficients (Section 2. To search for the n needed linearly independent solution vectors. indeed.366 Chapter 5 Linear Systems of Differential Equations C := multiply(inverse(A). It is reasonable to anticipate solution vectors of the form XI xz VI e AI v 2 eA. n) in (1 ).B or the MATLAB commands C = inv(A)*B Use your own calculator or available computer algebra system to solve "automatically" Problems 31 through 40 in this section. v 2 ... Vn are appropriate scalar constants. .. 2.. For if we substitute (i = 1.we can hope to solve for values of the coefficients v 1. . (3) so that x(t) = ve At is.. This will leave us with n linear equations which.3).• . We now introduce a powerful alternative to the method of elimination for constructing the general solution of a homogeneous firstorder linear system with constant coefficients. •. a solution of the system in (1). vz.. then each term in the resulting equations will have the factor eAt. Xz. x~ = x~ = a1IXJ aziXI + aizXz + · · · + a1nXn.for appropriate values of A. To investigate this possibility. so we can cancel it throughout. xn... it is more efficient to write the system in ( 1) in the matrix form x' = Ax (4) .3. + azzXz + · · · + a znXn.
that is.A [A.4 The Eigenvalue Method for Homogeneous Systems where A = [ aij ]. For this reason. •. (5) in the form (A. (6) by c =!= 0.AI[= 0. (5) holds.. then so is any nonzero constant scalar multiple cv of vthis follows upon multiplication of each side in Eq. When we substitute the trial solution x x' = AVeM in Eq. The prefix eigen is a German word with the approximate translation characteristic in this context.1 with derivative (5) This means that x = veM will be a nontrivial solution of Eq. (7) An eigenvector associated with the eigenvalue A is a nonzero vector v such that Av = AV. 367 = ve". (7) In its simplest formulation. the eigenvalue method for solving the system x' = Ax consists of finding A so that Eq. v2 . Vn. . By a standard theorem of linear algebra. the terms characteristic value and characteristic vector are in common use. (6) Given A. (6) Note that if vis an eigenvector associated with the eigenvalue A.5. The question now is this: How do we find v and A? To answer this question.Al)v = 0. so that (A. it has a nontrivial solution if and only if the determinant of its coefficient matrix vanishes. DEFINITION Eigenvalues and Eigenvectors The number A (either zero or nonzero) is called an eigenvalue of the n x n matrix A provided that [A . (4).A (8) .. The name of the method comes from the following definition. the matrix product Avis a scalar multiple of the vector v. Vn· Then x =veAl will be a solution vector. this is a system of n homogeneous linear equations in the unknowns v 1. v2 . the result is AveM = Ave"1 • We cancel the nonzero scalar factor e"1 to get Av = AV. the equation a11a 21 A a12 a in a2n a22. . (6) with this value of A to obtain v 1 .Al)v = 0.AI) = 0. if and only if [A . that is.AI[= ani an2 =0 ann . (7) holds and next solving Eq. . (4) provided that vis a nonzero vector and A is a constant such that Eq. we rewrite Eq..AI[ = det(A .
The case of repeated eigenvalues. (4) through (7) provides a proof of the following theorem. We will discuss separately the various cases that can occur. Our discussion of Eqs. if any).will be deferred to Section 5. possibly some are repeatedand thus an n x n matrix has n eigenvalues (counting repetitions. . . . V11 associated with these eigenvalues.. we get n linearly independent solutions (10) In this case the general solution of x' = Ax is a linear combination of these n solutions.6.368 Chapter 5 Linear Systems of Differential Equations is called the characteristic equation of the matrix A.. dt If v is an eigenvector a~sociated with A. 3. An of the matrix A. A2 . THEOREM 1 Eigenvalue Solutions of x' =Ax Let A be an eigenvalue of th~ [constant] coefficient matrix A of the firstorder linear system dx =Ax. 2. Next we attempt to find n linearly independent eigenvectors v 1. we allow the possibility of complex eigenvalues and complexvalued eigenvectors. which is the basis for the eigenvalue method of solving a firstorder linear system with constant coefficients. depending on whether the eigenvalues are distinct or repeated. real or complex. . its roots are the eigenvalues of A. but when it is. this method for solving then x n homogeneous constantcoefficient system x' = Ax proceeds as follows: 1. .multiple roots of the characteristic equation. Although we assume that the elements of A are real numbers. this equation has n rootspossibly some are complex. The Eigenvalue Method In outline. Step 2 is not always possible.. Upon expanding the determinant in (8). then x(t) = ve'At is a nontrivial solution of the system. We first solve the characteristic equation in (8) for the eigenvalues A1. v2 . we evidently get an nthdegree polynomial of the form (9) By the fundamental theorem of algebra. .
.. (11) Solution The matrix form of the system in ( 11) is (12) The characteristic equation of the coefficient matrix is 2 1=(4A)(1A)6 1. the two scalar equations 6a 3a + 2b = + b= 0. 1988).5. (13) yields that is. Example 1 Find a general solution of the system xi = 4xl + 2x 2 . For the coefficient matrix A in Eq.. NJ: Prentice Hall. .2 in Eq. for the associated eigenvector v = [ a CASE 1: AJ the system = 2.Al)v = 0 takes the form (13) b ]7 .4 The Eigenvalue Method for Homogeneous Systems 369 Distinct Real Eigenvalues If the eigenvalues A1 .we can choose a arbitrarily (but nonzero) and then solve for b.) In any particular example such linear independence can always be verified by using the Wronskian determinant of Section 5. Elementary Linear Algebra (Englewood Cliffs. (For instance. then we substitute each of them in turn in Eq. see Section 6. Therefore. x2 = 3xl. (14) has infinitely many nonzero solutions.5) = 0.A = A2  3A .10 =(A + 2)(A .3..2 of Edwards and Penney. . A2 . .x2.. In this case it can be proved that the particular solution vectors given in (10) are always linearly independent.3. An are real and distinct. Vn . v2 . . The following example illustrates the procedure. Substitution of the first eigenvalue AJ = . so we have the distinct real eigenvalues A1 = 2 and A2 = 5. (6) and solve for the associated eigenvectors v 1 . (14) In contrast with the nonsingular (algebraic) linear systems whose solutions we discussed in Section 5. the homogeneous linear system in (14) is singularthe two scalar equations obviously are equivalent (each being a multiple of the other). Eq. ( 12) the eigenvector equation (A. 0.
370 Chapter 5 Linear Systems of Differential Equations Substitution of an eigenvalue A in the eigenvector equation (A . Hence a general solution of the system in (11) is X(t ) 1 = CtXt (t) + C2X2(t) = Ct X t (t) 1 ] e[ _3 21 + C2 [ 2 l Je 51. x2(t) = 3ct e. They are linearly independent because their Wronskian 2est 7 3t e sr . 3a.Al)v = 0 always yields a singular homogeneous linear system. so is an eigenvector associated with A2 = 5.3. we had made another choice a = c =I= 0. b = c =I= 0 would merely give a [constant] multiple of v 2 . = Cte 21 + 2c2 e51 . the choice a = 1 yields b = 3. Looking at the second equation in (14). A different choice a = 2c. and thus is an eigenvector associated with At = 2 (as is any nonzero constant multiple of vi). b = . These two eigenvalues and associated eigenvectors yield the two solutions Xt(t) = [ ~ Je21 and x2 (t) =[ i] e5 r. any choice we make leads to (a constant multiple of) the same solution CASE 2: A2 = 5. and among its infinity of solutions we generally seek a "simple" solution with small integer values (if possible). in scalar form. b = 3c. we would have obtained the eigenvector = [ Vt 3~ J= C [ ~ J· Because this is a constant multiple of our previous result. With b 0.6b = 0 (15) = 1 in the first equation we get a = 2. Substitution of the second eigenvalue A = 5 in (13) yields the pair a+ 2b = of equivalent scalar equations.21 + c2 e 51• . Remark: If instead of the "simplest" choice a = 1.e is nonzero.
. . We see two families of hyperbolas sharing the same pair of asymptotes: the line x 1 = 2x2 obtained from the general solution with c 1 = 0. 2. .. ..1.. If each flow rate is r gallons per minute..·····""""""" ..x 2 of Example 1. . then a simple accounting of salt concentrations... Xn denote the components of a single vectorvalued solution x.· . .. and out of tank 3.5. and then the whole physical system is modeled by a system of differential equations. . yields the firstorder system x. ____ ~. (17) Example 2 ·... If VI = 20. • "" r (gal/min) Compartmental Analysis Frequently a complex process or system can be broken down into simpler subsystems or "compartments" that can be analyzed separately...3x 1. x~ = 3x 1 . xz(O) = x3(0) = 0.~vi· i = 1. . x 2 .1 shows some typical solution curves of the system in (11). • FIGURE 5..4.4. . and the line x 2 = 3xi obtained with cz = 0.. 3. it is convenient when discussing a linear system =Ax to use vectors x 1..•. and 3. . X1 Remark: As in Example 1.. The whole system can then be modeled by describing the interactions between the various compartments.4. then Xi (t) and xz(t) both tend to +oo as t + +oo.1.. . Given initial values x 1 (0) = b 1 .~~ ·· ·· ... then xi (t) and x 2 (t) both tend to oo as t + +oo. = k 1x 1. . .  _. while mixed brine flows from tank 1 into tank 2. 5. and the initial amounts of salt in the three brine tanks. The three brine tanks of Example 2.4. b 2 ) lies to the right of the line x 2 = 3x 1. ~ find the amount of salt in each tank at time t 0. in pounds..2 shows three brine tanks containing VI. . Thus a chemical plant may consist of a succession of separate stages (or even physical compartments) in which various reactants and products combine or are mixed.. = 4x 1 + 2x2... kiXJ .2. whereas the scalars x 1 ... x 2 . 2. xz(O) = hz.. Fig. Let xi (t) denote the amount (in pounds) of salt in tank i at time t for i = 1. respectively. Fresh water flows into tank 1.. and V3 gallons of brine..k3X3.. x2 = I I Tink{ · . :..kzX2. : Vr (gal) ~. b 2) lies to the left of the line x 2 = . It may happen that a single differential equation describes each compartment of the system.... • If (b 1 . . V3 = 50. V2 .r· tl ~ ~v FIGURE 5. are Xi (0) = 15...·· ·····~.._. Xn to denote different vectorvalued solutions of the system. As a simple example of a threestage system. Direction field and solution curves for the linear system x. (16) x3 = where r k· ... it is apparent from the figure that • If (b 1 .. kzx2 . . r = 10 (galjmin).4 The Eigenvalue Method for Homogeneous Systems 371 Figure 5. V2 = 40. as in Example 2 of Section 5. from tank 2 into tank 3..
HI= (0.6 5 = [3 r is associated with the eigenvalue A.0 0. Each of the first two rows implies that a = 0. 1 CASE = 0. 0.25 in (19). Thus the eigenvector 1 5 r is associated with the eigenvalue A. and division of the third row by 0.5 0.0.2 = .25 0.5.25. 0 0 0. and then the first equation . 3 = 0. which is satisfied by b = 1.5 0.2. = 0.3 c [ 0] 0 0 for the associated eigenvector v = [ a b c JT.0 ] 0.25 .A.2 X2(t) X3(t) (18) for the vector x(t) = [X[ (t) r. yield the scalar equations 2a + b 5b = 0.5 in (19).5) · I J v = [ 0.372 Chapter 5 Linear Systems of Differential Equations Solution Substituting the given numerical values in ( 16) and (17).A. Thus the eigenvector Vt = 6 and c = 5. (19) A . 2 = 0. we get the initial value problem x' (t) = [ 0. .25 leads readily to the characteristic equation lA.25. we get the equation [ A + (0.25.0 0. respectively.25 0.0 [ a b 0.2.2 0.25 0.5. 0.A.0 0.A. The second equation is satisfied by b gives a = 3.5 0. Thus the coefficient matrix A in (18) has the distinct eigenvalues A. Substituting A. after division by 0.05 gives the equation 5b + c = 0.05.0 0.)= 0.)(0.0] 0.25 and 0.AI = [ 0. CASE = 1: A. c = 5.25. 1 = 0.25) · I Jv = [ 0. The last two rows. A. 0.0 0.5.0] ] [A+ (0.2.A.25 2: A.5. The simple form of the matrix 0. Substituting A.0 0. + 6c = 0. 1 = 0.)(0.A.5 0.0 0. we get the equation = 0. and A.0 x.0 0.25 ~5 for the associated eigenvector v = [ a b c JT.5.25 0. = 0.
373 Substituting A.05 0. x 2 (t). we get the = 15. 6CJ + C2 0. + c3 [ ~] el02)< The resulting scalar equations are x 1(t) 3c 1e(0. X2(t) = 6cle(0.25)1' 25e< .5)r.0·2)t.5)t .3 0. As we would expect.25)r. the amounts of salt at time t in the three brine tanks are given by x 1(t) = X3 (t) = 15e<. we get the equation [A+ (0.3.2.2) ·I J v = [ 0.0·5)1.4.Sc2 + c3 = 0 3cl that are readily solved (in tum) for c 1 = 5. Thus = [ V3 o o 1 r is an eigenvector associated with A.25 for the eigenvector v. and x 3 (t). but the allzero third column leaves c arbitrary (but nonzero).150e<. respectively.2)1 .5)t .5c2e<0.0. 3 = 0. and b 0. and c3 = 125.4 The Eigenvalue Method for Homogeneous Systems CASE 3: A.3 shows the graphs of x 1 (t). • . When we impose the initial conditions x 1 (0) equations = 15.2. c 2 = 30.0 0. FIGURE 5.= 0. x 2 (0) = x 3 (0) = 0. The salt content functions of Example 2.2 in (19). Thus.4. tank 1 is rapidly "flushed" by the incoming fresh water. 0. 3 = 0.0.5. Sc1 .5)t x3(t) = + 5cle<0.25)t + c3e<.5)1 X2 (t) = . Figure 5.0·25 )1 + 125e< . and x 1 (t) + 0 as t + +oo. The first and third rows imply that a = 0.30e<  + 30e<  0.0 0.C= 051• + c2 [ _!] eH~l• c2e(0.5 0. finally. The amounts x2(t) and x 3(t) of salt in tanks 2 and 3 peak in tum and then approach zero as the whole threetank system is purged of salt as t + + oo. The general solution therefore takes the form x(t) = c 1 [ ~] .
if v= [a! +.AI)v = 0.. The complexvalued solution associated with A and vis then x(t) = veA. The only complication is that the eigenvectors associated with complex eigenvalues are ordinarily complex valued. ] a2 b2z .AI)v = o since A =A and I= I (these matrices being real) and the conjugate of a complex product is the product of the conjugates of the factors. x(t) = eP1 (acosqt .. Thus the conjugate v of v is an eigenvector associated with I. (21) Because the real and imaginary parts of a complexvalued solution are also solutions. we thus get the two realvalued solutions x 1 (t) = Re[x(t)] = eP1 (acosqt. so long as they are distinct the method described previously still yields n linearly independent solutions. If v is an eigenvector associated with A. we note thatbecause we are assuming that the matrix A has only real entriesthe coefficients in the characteristic equation in (8) will all be real. it is preferable in a specific example to proceed as follows: • First find explicitly a single complexvalued solution x(t) associated with the complex eigenvalue A. ] [bl. so that (A. Rather than memorizing the formulas in (22).bi. so we will have complexvalued solutions. i = a + bi. . then taking complex conjugates in this equation yields (A .bsinqt) + i eP1 (bcosqt + asinqt). that is. Of course the conjugate of a vector is defined componentwise. Consequently any complex eigenvalues must appear in complex conjugate pairs. x2(t) = lm[x(t)] = eP1 (bcosqt + asinqt) (22) associated with the complex conjugate eigenvalues p ± qi.. bl~] [a!.t = ve<p+qi)r =(a+ bi)eP1 (cos qt + i sinqt). Suppose then that A = p + qi and I= p . and I. + b2 .qi are such a pair of eigenvalues. It is easy to check that the same two realvalued solutions result from taking real and imaginary parts of veir. = a2 . To obtain realvalued solutions.bsinqt). (20) an+ bni an bn then v =a.374 Chapter 5 Linear Systems of Differential Equations Complex Eigenvalues Even if some of the eigenvalues are complex. • Then find the real and imaginary parts x 1 (t) and x2 (t) to get two independent realvalued solutions corresponding to the two complex conjugate eigenvalues A.
= 4x 1 . ] e 4r ( cos 3t . we get the equation [A. sm t [ t 3t + .e 4t [cos Xt ( t ) .c2 sin 3t] · 3t + c2 cos 3t · c 1 sm Finally. Division of each row by 3 yields the two scalar equations ia. (23) dx2 = 3xt +4x2.3i in the eigenvector equation (A . x~ = 3x 1 + 4x2 of Example 3. 3t] . I]v = [ 3~ .i sin .3i.(4.4. Substituting A = 4.3x2.4. The corresponding complexvalued solution x(t) = ve. x 2 (t) = e41 (ct sin 3t + c2 cos 3t). each of which is satisfied by a = 1 and b = i.b = 0. Thus v = [ 1 i is a complex eigenvector associated with the complex eigenvalue A = 4 . Each appears to spiral counterclockwise as it emanates from the origin in the x 1x 2 plane. we see that . 3t ] sm 3 t and xz(t) = e 41 [ sin3t] . cos 3t A realvalued general solution of x' = Ax is then given by X(t) = CtXJ (t) + C2X2(t) = e 4r [c1 cos 3t .Al)v = 0. cos t sm 3t The real and imaginary parts of x(t) are the realvalued solutions .c2 sin 3t). 3t ) _ e 4t .3i) .4. a general solution of the system in (23) in scalar form is x 1(t) = e41 (c1 cos 3t .5. dt Solution The coefficient matrix A= has characteristic equation [4 3] 3 4 and hence has the complex conjugate eigenvalues A = 4 . cos 3 . Figure 5. ] e (43i)t t [ r 1 .3i and A = 4 + 3i. FIGURE 5. a+ ib = 0. Direction field and solution curves for the linear system x. Actually.~] [:] = [ ~] for an associated eigenvalue v = [ a b JT .4 shows some typical solution curves of the system in (23)._1 of x' =Ax is then _ [ 1 _ x (t ) _ . because of the factor e41 in the general solution.t.4 The Eigenvalue Method for Homogeneous Systems 375 Example 3 Find a general solution of the system dxt dt = 4xt ..3x2 .
and r = 10 galjmin.2 ] 0 0.0.2 X (25) x.A. the appropriate modification of Eq.2 0.8) · A.2.2.0.2) ·A.4.A. V2 . I = 0. = 0 in Eq.4 ± (0.+ 0.)(0. Thus A has the zero eigenvalue A.2 0.2 . X2 X3 r as usual. With the same notation as in Example 2. (26) gives the eigenvector equation 0.A.4 0. · I = [ (26) along its first row.2 ] 0.0 · l)v =[ [a ]= [ b c 0] 0 0 .4.4 .0 0. 0.2 ] 0. and x3 (t) of salt at timet in the three brine tanks of Fig.4. we find that the characteristic equation of A is (0.2 is that now the inflow to tank 1 is the outflow from tank 3.5. Example 4 Find the amounts x 1 (t) .4 0. = . the point (x 1 (t). x 2 (t)) approaches the origin as t + oo.2 . • Flow rate: r Figure 5.k 1x 1 dt FIGURE 5.2 0. With the given numerical values.4. 0 = 0 and the complex conjugate eigenvalues A. V2 = 25 gal. [(A.0.376 Chapter 5 Linear Systems of Differential Equations • Along each solution curve.2 (A . 3  (0. dx2 dt (24) where ki = r1 vi as in (17).0 0.0 0. 5.4) 2 + (0. A .5 if V.4 0.0 . (16) is dx.0 0. CASE 1: A.A.A.A. 2  (0. 5. When we expand the determinant of the matrix 0.)(0. = A. =50 gal.= dt 0 with X= [ Solution 0 0. V3 =50 gal.)+ (0.2 .4) = .4. 0 = 0.0 ..5 shows a "closed" system of three brine tanks with volumes V1.A. whereas • The absolute values of x 1 (t) and x 2 (t) both increase without bound as t + +oo. The three brine tanks of Example 4. and V3 .4 0.4 0. (24) takes the form dx [ . The difference between this system and the "open" system of Fig.2) 2 ] = 0.A.2)(0.2)i. 0.2)(0. x2 (t). Substitution of A.2 0.
4) 1 (cos0. Then the first equation [0.2 + (0.2t] .(0.0.2t + i cos 0.OA)t ] cos 0.0 0.(0.2t .2t .2t . cos 0.i) JT is a complex eigenvector associated with the complex eigenvalue A = .2t +sin 0.2)i]a + (0.2t .4.2 0.OA)t (c 1 cos 0.2t sin 0.4 The Eigenvalue Method for Homogeneous Systems 377 = 2b.402i)l e( .0 0. The first row gives a = c and the second row gives a so v 0 = [ 2 1 2 JT is an eigenvector associated with the eigenvalue Ao = 0.2)i. Thus v = [ 1 i ( 1 . (c 1 sin 0.2t).2 + (0.2)i 0.4.2)c = 0 gives c = 1 .2)i)l] v = [ [a] b c The second equation (0.4 . the eigenvector equation Dl (27) Substitution of A = 0.OA)r e< .(0.2t The general solution has scalar components x 1 (t) = 2c0 + e< . (25) is the constant solution ~}(I)= CASE 2: A= 0.i sin 0.2t + i sin 0.2)ib = 0 is satisfied by a = l and b = i.2t).0 (0.2t) = e< .sin 0. The corresponding solution x0 (t) = v 0 eAor of Eq.2t .2t .2t + ( . c1  x 2 (t) X 3 (t) = c0 + (29) = 2co + e <OA) r [ (  cz) cos 0.4 0.2t i sin0.2)i in Eq. The corresponding complexvalued solution x(t) = veA' of (25) is x(t) = [ 1 = [ 1 1 .0.i [ r r e(0.5.sin 0.2)i.c 1 + cz) sin 0.i 1. cos 0.i.2)i 0.2t ] sin 0.(0.2t .2t (28) x2 (t) = e <O A)I [ .2)i [A .2t .2t The real and imaginary parts of x(t) are the realvalued solutions x 1 (t) = e<OA)I [ cos0.c2 sin 0.2t + c2 cos 0. (26) gives 0. for v = [a b cr.2)a + (0.sin 0.2 ] 0.( 0. cos 0.4.2t ] cos 0.2 + (0.i cos 0.
7 satisfy the differential equations 17. . = 4x. x. + Xz + 4x3 18. The amounts x 1(t) and x 2 (t) of salt in the two brine tanks of Fig. assuming that r = 10 (gal/min). x. x. 3.5x 2 .Sx. x~ = 6x 1 . x. find also the corresponding particular solution.9x2 . x~ = 4x.2xz4x3 22. 13. x~ = 4x 1 + 4xz + 2x3 25.Xz = 2x 1 .2x3.5x2 . x~ = 4x 1 + 3x2 = 7x 1 . 5. x. x 1 (0) = 2. the constant co in (30) is onefifth the total amount of salt. The salt content functions of Example 4. 14. 15. In Problems 17 through 25. • In Problems 1 through 16. = 2x 1 + xz . + Sxz. x~ = Sx. x. x~ = 2x 1 + xz. x. + Xz + X3. = r/V. x. use a computer system or graphing calculator to construct a direction field and typical solution curves for the given system. x~ = . . x. x. 7. x.4xz . +4xz dt that satisfies the initial conditions x 1 (0) X3(0)=17. Figure 5. x. x. + 20xz. 5. + Sxz + 3x3 23. 6. and x 2 (0) = 0. x~ = 2x 1 xz2x3.2x2 .2xz.X3.Sx2.2xz.3xz . = 3x. x~ = 4x. x~=x. x 2 (0) = 4 = x 1 .9x. x 1 (0) = 15 (lb). x~ = 2x 1 + 7x 2 + x 3. x~ =x. 10. X3 = 0. x 2 (0) = 3 = 3x 1 . x~ = x 1 . = 3x 1 + 2xz + 2x3. x~ = 100x 1 . +xz +4x3. x 1 (0) = x 2 (0) = 1 Xz = 6x 1 . + 3xz = 5x 1 . x~ = 6x 1 . x. in which case xr (0) = 50 and xz (0) = x 3 (0) = 0. t+00 lim xz(t) =co. x.Sxz.x 2 = 3x 1 . x~ = x. x 1 (0) = 1.6x 1 . x = x 2 (t) 15 20 FIGURE 5. apply the eigenvalue method of this section to find a general solution of the given system. . x. 11. x~ = 5x 1 + Sxz + 3x3 24.X3. +4xz+x3 . x~ = x. x~ = 9x 1 + 3x2 = x. 9. x. x. x. and 1+00 lim x 3 (t) = 2co. 16. + Xz + Sx3 21. = Sx.4. = x 1 + 2x2. x. the limiting distribution is one of uniform concentration throughout the system.2x2 = 9x. x 2 (0) = 0 = 3x 1 + 4x2 . x~ = 2x. . x. x. x. First solve for x 1(t) and Xz(t).+7xz+x3. x~ = 2x 1 + Xz = 3x 1 + 2xz. x. = 4x. we see that t+00 lim x 1 (t) = 2c0 . x. So whatever the initial distribution of salt among the three tanks. . In Problems 27 and 28 the volumes V 1 and V2 are given. 12. = x 1 + 2xz + 2x3. + x z+4x3 where k. x~ = . . x~ = x. x.5x2 = x.6. x. 1. and c2 = 10.6 shows the graphs of the three solution functions with c0 = 10.3xz . For each problem. 4. 8. Find the particular solution of the system dx 1 dt dx 3 = . = 5x 1 + Sxz + 2x3. Thus as t + +oo the salt in the system approaches a steadystate distribution with 40% of the salt in each of the two 50gallon tanks and 20% in the 25gallon tank.Sx3.4x2 . Because of the factors of e<OA)t in (29). x~ = 2x 1 + x 2 = 2x 1 + 3x2 .7x2 .6x3.=Sx 1 +xz+3x3. c 1 = 30. x 2 (0) = 0. x 1 (0) = 0. x~ = 2x 1 + Xz + 7x3 19. the eigenvalues of the coefficient matrix can be found by inspection and factoring. x~ = 6x 1 + 6xz + Sx3 26. = 4x 1 + 3x2 = SOx. 2.60x2 = 4x 1 + x 2 . Apply the eigenvalue method to find a general solution of each system. Observe that (30) 50 45 35 ><25 20 15 10 Of course the total amount of salt in the closed system is constant. x~ = 3x. + 4xz.378 Chapter 5 Linear Systems of Differential Equations giving the amounts of salt in the three tanks at timet.4. If initial values are given. +7xz +x3. 2.6xz .2xz. x~ = x. . x~ = 4x 1 . +xz+x3. = 5x 1 . fori = 1.4. x~ = 3x.X3. x~ = 6x 1  20. x~ = 4x.
x 1 (0) = 15 (lb). and from tank 3 into tank 1. X2 + X3 + 7x4. 42. Vz = 12. =4x 1 + has eigenvalues }q = . = 30. First solve for the amounts of salt in the three tanks at timet. 3 = 10. dt where k.3. 41. the zeros in the matrix make its characteristic polynomial easy to calculate. First solve for the amounts of salt in the three tanks at timet.4 Application below) in order to find a general solution of the linear system x' = Ax with the given coefficient matrix A. V2 . 5. and V3 (in gallons). = 20. v. Xz(O) = 0. v. Vz = 20.4. xz(t). The initial amounts x 1 (0) = x 0 (pounds). V2 = 25 (gal) 28. 27. from tank 2 into tank 3. 0 0 3 4 0 0 2 0 A~ ~ 0 1 0 0 0 0 21 5 27 [ 2 A= ~ 0 5 0 21 2 The coefficient matrix A of the 4 x 4 system A~ u [2 0 2 3 0 ~] I~] ~] FIGURE 5. V2 = 40 (gal) Problems 31 through 34 deal with the open threetank system of Fig.6. = 15. V2 = 25 (gal) 30. v. V1 = 50 (gal). Xo = 33. V2. and X3 (0) = 0 of salt in the three tanks are given. Vz = 10. r = 60. and x3(t). x 2 (0) = x3(0) = 1. = 20. 40. Xo = 18.4. A= [ 40 35 25 12 13 7 11 . 37. Xo = 27. mixed brine flows from tank 1 into tank 2. from tank 2 into tank 3. V3 = 30 43. Finally. 4 = 15.7. which is described by the equations in (24). Find the general solution ofx' = Ax. x 2 (t). and x 3(0) = 0 of salt in the three tanks are given.54] 46 34 7 31 31. solve for x1(t) and xz(t). A = [ 20 12 .4. 39. 5. 379 34.x. v. = 60. For each matrix A given in Problems 38 through 40. then determine the limiting amount (as t ~ +oo) of salt in each tank. r = 60. 36. 38. V. Xz (0) = 0. construct a figure showing the graphs of x 1 (t) and x 2 (t ). Xo = 40. v3 = 60 Problems 35 through 37 deal with the closed threetank system of Fig. Find the particular solution of this system that satisfies the initial conditions x1(0) = 3.8. assuming that r = 10 (gal/min).5. 2 = .4 The Eigenvalue Method for Homogeneous Systems Then find the maximum amount of salt ever in tank 2. and A. 35. = 20.kzxz. = r /V. v. A. x~ = 7x 1 + X2 + x 3 + 4x4 x. r = 60. and X3 (t). construct a figure showing the graphs of x.xo = 45. A. Vz = 50. v3 = 10 32.8 satisfy the differential equations dx2 = k. x4(0) = 3. 29. = 20. Finally. FIGURE 5. v. V2 = 40 (gal) The amounts x 1(t) and x 2 (t) of salt in the two brine tanks of Fig. (t). r = 120. . all at the given flow rater gallons per minute. The two brine tanks of Problems 29 and 30. then determine the maximal amount of salt that tank 3 ever contains. V1 = 25 (gal). Fresh water flows into tank 1. V1 =50 (gal). as usual. all at the given flow rate r gallons per minute. r = 10. Finally. V 1 = 25 (gal). r = 60.4.5. and out of tank 3. + 10x2 + 4x3 + x4. construct a figure showing the graphs of x 1 (t). 5. and V3 (in gallons). as are their volumes V 1. x~ = x. The two brine tanks of v3 = 40 v3 = 20 v3 = 30 Problems 27 and 28. Xo = 55. Then construct a figure showing the graphs of x 1 (t) and x 2 (t).1 21 .2.4. as are their volumes V1. X~= Xi + 4x2 + 10x3 + X4.(O) = x 0 (lb). Mixed brine flows from tank 1 into tank 2. Vz = 15. The initial amounts x. r = 30. In Problems 42 through 50. Vz = 6. In Problems 29 and 30.48 13] . v3 = 60 33. Xo = 45. use a calculator or computer system to calculate the eigenvalues and eigenvectors (as illustrated in the 5. and Xz(O) = 0. Vz = 30.
5. choose a positive integer n < 10 (n = 5. You can use these commands to find the eigenvalues and eigenvectors needed for any of the problems in this section.25 0. except with n rather than three successive tanks having volumes Vi = 10qi (i = 1. then the salt amounts x1 (t). A= i .5 0.0 ] 0. 3.. ~ Most computational systems offer the capability to find eigenvalues and eigenvectors readily. .. Once the matrix A has been entered. for instance) and let q 1. ······ . .4 Application Automatic Calculation . Note that with results presented in decimal form. q2.0 0.. A= [ 8 23 34 26 129 123 5 11 19 17 106 52 139] 70 20 31 22 33 16 7 26 25 12 202] ~] J] 8 5 2 32 18 48. n) in gallons. .1 . the Mathematica command Eigensystem[A] or the MATLAB command [V.. . We see each eigenvector displayed as a column vector beneath its eigenvalue.. qn denote the first n nonzero digits in your student ID number.. k 1 x1. ____ _ __ ..5 0. .JXi . . the Maple command eigenvects (A) . . (where D will be a diagonal matrix displaying the eigenvalues of A and the column vectors of v are the corresponding eigenvectors) produce similar results. Fig.25 0. If each flow rate is r = 10 gallons per minute.0 0. Now consider an open system of brine tanks as in Fig. .9 shows a graphing calculator computation of the eigenvalues and eigenvectors of the matrix ~ A= [ 0.0 0. .. A= 12 10 10 9 10 2 6 0 6 5 15 9 14 23 10 20 10 0 [ 10 47 5] of Eigenvalues and Eigenvectors 5.D] = eig(A) FIGURE 5. A= 370 152 16 59 13 35 38 7 23 95 10 9 13 0 0 0 13 14 19 10 20 10 4 7 30 12 30 12 18 SO. For a more substantial investigation.9. A= Chapter 5 Linear Systems of Differential Equations [ 147 23 9 15 7 7 17 13 42 16 6 6 18 6 27 21 90 90 12 9 45. For instance. = . A= [ 24 18 [ 13 46.4.1 47. TI86 calculation of the eigenvalues and eigenvectors of the matrix A.380 44.. 5. A= 139 40 167 121 232 248 64 360 14 52 14 [ 139 22 7 5 7 8 28] 38 139 38 76 49.~ is v = [ 1 2 ~ ] T..4. . .ki x i = 2. n) .2. it is up to us to guess (and verify by matrix multiplication) that the exact eigenvector associated with the third eigenvalue ). Xn(t) satisfy the linear system xi = x. (i = ki . 2.2 of Example 2. x2(t).4.
4.'. 5. _. It provides a sample of the more technical applications of eigenvalues to physics and engineering problems. 0 (2) • This optional section may be omitted without loss of continuity. A plot like Fig.3 and 5. In this section we apply the matrix methods of Sections 5. ?: · . suppose that the system of n tanks is closed as in Fig. X2(0) = X3(0) = · · · = Xn(O) = 0.k 1x 1..~ ~ SecondOrder Systems and Mechanical Applications* ~' .. _.__. "' k3 .~. We assume that the masses slide without friction and that each spring obeys Hooke's law. •M ¥0N O. Graph the solution functions and estimate graphically the maximum amount of salt that each tank ever contains.5.5. they actually follow similarly from Hooke's and Newton's laws. and x 3 of the three masses (from their respective equilibrium positions) are all positive.4.. whatever the signs of these displacements. The fourth spring is compressed the distance x 3 ..i k4. then • • • • The first spring is stretched the distance x 1 ..o¥> > N N  . Although we assumed in writing these equations that the displacements of the masses are all positive.X2) .. in this closed system.5. (1) m 3 X~ = .~~N>"~"'''>. For an alternative investigation.···. .5. ¥ •••• •••• •¥ ~ ..1 shows three masses connected to each other and to two walls by the four indicated springs..·i k 2 .'!!: . N • ¥ " ' "'''"'"'"'"·""  · .k3(X3.x 1 .>~• ~ VO~·. Then the first equation should be replaced with x.~ x1 x2 x3 FIGURE 5. as t + +oo the salt originally in tank 1 distributes itself with constant density throughout the various tanks. .6 should make this fairly obvious.Xt) + k3(X3  X2)._. = knxn.x 2 . x 2 . Apply the eigenvalue method to solve this system with initial Xt (0) = 10.4 to investigate the oscillations of typical massandspring systems having two or more degrees of freedom.· A' .__ . application of Newton's law F = ma to the three masses (as in Example 1 of Section 5... so that tank 1 receives as inflow the outflow from tank n (rather than fresh water). The third spring is stretched the distance x 3 .. ~· ~. 5.k4X3.its extension or compression x and force F of reaction are related by the formula F = kx.1. '? z._ • ~~ ~'""''''"'~···• ' >N m _ _. Three springcoupled masses. _ _ _ _ > < ' .. If the rightward displacements x 1. Therefore.1) yields their equations of motion: m2X~ = k2(X2._. Figure 5. Now show that. The second spring is stretched the distance x 2 .5 SecondOrder Systems and Mechanical Applications 381 where ki conditions r /Vi.~.hN·. the mass matrix In terms of the displacement vector X = [ XJ X2 X3 r. Our examples are chosen to illustrate phenomena that are generally characteristic of complex mechanical systems. .
(4).4 for a firstorder system) a trial solution of the form x(t) = v eal' (8) where v is a constant vector. Hence multiplication of each side in Eq..5.5.: The notation in Eqs.2. M= [~' . and a stiffness matrix K satisfying Eq.1 yields the homogeneous secondorder system x" =Ax. Then x" = a 2 vear. (7) where A = M . There is a wide variety of frictionless mechanical systems for which a displacement or position vector x.j> x. 0 mz 0 and (kt 0 IJ 0 0 0 0 (kn. so substitution of Eq. !i //:( . (1) through (4) generalizes in a natural way to the system of n springcoupled masses shown in Fig.1 +kn) kn (kn kn (5) + kz) 0 kz (kz 0 k3 (k3 kz + k3) k3 + k4) K= 0 0 k4 (6) 0 0 0 0 + kn+d for the mass and stiffness matrices in Eq. .382 Chapter 5 Linear Systems of Differential Equations and the stiffness matrix (3) the system in ( 1) takes the matrix form Mx" = Kx. (4) ~~ ( !. '•• ':. x2 xnl xn FIGURE 5. 5. . (7). .~1/~': .. . a nonsingular mass matrix M. We need only write 'h:c•s:pyr:ry:!0p1fr>.·~·:"' ._____ . A system of n springcoupled masses. (4) by M. Solution of SecondOrder Systems To seek a solution ofEq.. to get its inverse M.2. we substitute (as in Section 5.1 we need only replace each diagonal element with its reciprocal.1 K. (4) can be defined. (8) in (7) gives . The diagonal matrix M is obviously nonsingular.
. (11) with ai and bi arbitrary constants. then a general solution of x" =Ax is given by n x(t) = L(ai cos w. w~ with associated (real] eigenvectors v 1 . If x" = Ax models a mechanical system.5. xo(t) = (ao + bot)vo (12) is the corresponding part of the general solution. then x" = 0 · vo = (ao + bot) · 0 = (ao + bot)· (Avo) =Ax. . . If x(t) = (ao + bot)vo. v2 . ( 12).o with associated eigenvector v0 . In this case the solution given by Eq. THEOREM 1 SecondOrder Homogeneous Linear Systems If the n x n matrix A has distinct negative eigenvalues wf..t i= ! + bi sin w.. thus verifying the form in Eq. (8) is x(t) = The real and imaginary parts x 1(t) = vcoswt and x2 (t) = vsinwt (10) v e i wr = v( cos wt + i sin wt). Remark: The nonzero vector v0 is an eigenvector corresponding to Ao = 0 provided that Av0 = 0. wi. . . In the special case of a nonrepeated zero eigenvalue A.t)v.. • . .5 SecondOrder Systems and Mechanical Applications 383 which implies that (9) Therefore x(t) = vear is a solution of x" =Ax if and only if a 2 =A. of x(t) are then linearly independent realvalued solutions of the system. and v is an associated eigenvector. then it is typical that the eigenvalues of A are negative real numbers. This analysis leads to the following theorem. an eigenvalue of the matrix A. Vn. If then a = ±wi.
. The eigenvector equation (A . we set k3 = 0. m 2 = 1. 5. x2(t) = 2c 1 cos(5t . the system in (13) therefore has solutions with [circular] frequencies w 1 = 5 and W2 = 10. They describe the physical system's two natural modes of oscillation at its two [circular] natural frequencies w 1 = 5 and w 2 = 10. Because there is no third spring connected to a righthand wall.a 1). so A has the negative eigenvalues A1 = 25 and A2 = 100. The eigenvector equation (A . · ~ Example 1 Consider the massandspring system with n = 2 shown in Fig.3. By Theorem 1. By Eq.50· 25 =A 2 + 125A + 2500 =(A+ 25)(A + 100) = 0. (11) it follows that a general solution of the system in (13) is given by x(t) = (a 1 cos5t + b 1 sin5t)v 1 + (a 2 cos lOt+ b2 sin 10t)v2. and k 2 = 50. the two terms on the right in Eq. = 100. CASE 1: A1 = 25. The massandspring system of Example 1.a 1) . (14) represent free oscillations of the massandspring system.a 1) (with c 1 =Jar+ b[.2. The natural mode x1(t) = (a 1 cos5t + b 1 sin5t)v 1 = c 1 cos(5t .U)v = 0 is [ 50 50 25] 25 [a]_ [OJ b  0 ' so an eigenvector associated with At = 25 is Vt = [ 1 2 CASE r. (14) As in the discussion of Example 3 of Section 5. (t) = c 1 cos(5t. and sina 1 = b 1/c 1) has the scalar component x. 2: A2 = 100. k.A).5.5. The characteristic equation of A is ( 75.3. then the equation Mx" = Kx is (13) which reduces to x" = Ax with FIGURE 5.A)( 50.U)v = 0 is so an eigenvector associated with A2 = 100 is v2 = [ 1 1 ] 7 .384 Chapter 5 Linear Systems of Differential Equations *. (15) . If m 1 = 2. cos a 1 = equations [ ~J aifc1.
Oscillations in opposite directions with frequency w2 = 10. " " " " " " " I v '{ ~ v v 0 v I v xx 1(t) Jt v 2n 0 n/2 3n/2 t FIGURE 5. The three railway cars of Example 2. ll%i•h. k 2 = k3 = k .5. but disengage instead of stretching.. 5. FIGURE 5.a2).5. x2(t) = . the amplitudes of motion of the two masses are the same. we get the system [1' which is equivalent to FIGURE 5. • X= Xz(t) " " fl. so that c 1 = c3 . (19) If we assume further that m 1 = m 3 .6.5.5. (16) and therefore describes a free oscillation in which the two masses move in synchrony in opposite directions with the same frequency ~ 10 and with equal amplitudes of oscillation (see Fig.J•15J . Oscillations in the same direction with frequency u)J = 5. the amplitude of motion of mass 2 is twice that of mass 1.C3 c~] x (18) k mi (i=1.··~~r~ connect~d bybuffer springs that react when compressed.5 SecondOrder Systems and Mechanical Applications 385 and therefore describes a free oscillation in which the two masses move in synchrony in the same direction and with the same frequency w 1 = 5.a2). (2) through (4).5.. With n = 3. 5. 3).5).4.5.2.5. 0 m2 0] [ 0 x" = m 3 0 k k 0 k 2k k k ~]X. but with the amplitude of motion of m 2 twice that of m 1 (see Fig.4). The natural mode has the scalar component equations XI (t) = c2 cos(lOt. (17) x" [c. then a brief computation gives (20) .Fig~r~5:5~6·~h~~s three rail~. = C2 0 with Ci=  Cj 2c2 C3 . and k1 = k4 = 0 in Eqs.c2 cos(lOt .
w3 = 4.e. w 1 = 0. According to Theorem 1.=6. Then. For a numerical example. using fps units with mass measured in slugs (a weight of 32 pounds has a mass of 1 slug). Hence the matrix A has eigenvalues (2la) corresponding to the natural frequencies (21b) of the physical system. 750 C2= . we have and CJ 3000 = . (18). and A3 = 16. that the middle car weighs 8 tons. i.=4. the corresponding part of a general solution of x" = Ax is r CASE 2: A2 = 4. the corresponding part of a general solution of x" = Ax is . The eigenvector equation (A.. k = 3000 lb/ft.5 tonsjft. According to Theorem 1.. = 0 is (22) and the eigenvaluefrequency pairs given by (21a) and (21b) are AJ = 0. CASE 1: AJ = 0. w2 = 2. w2 = 2. A2 = 4. 3000 500 Hence the coefficient matrix A is 4 A= [ ~ 4 12 4 4 ~].386 Chapter 5 Linear Systems of Differential Equations for the characteristic equation of the coefficient matrix A in Eq. The eigenvector equation (A. w1 = 0..H)v 0 4 = 0 is (A + 4I)v = [ 6 0 8 4 so it is clear that v2 = [ 1 0 1 ] T is an eigenvector associated with A2 = 4.Al)v 4 Av= [ ~ so it is clear that V) = [ 1 1 1 is an eigenvector associated with A) = 0. suppose that the first and third railway cars weigh 12 tons each. and that the spring constant is k = 1.
2b2 cos2t + 4b3cos4t. (24a) (24b) Then substitution of (24a) in (23) gives the scalar equations a1 + a2 + a3 = 0.U)v 12 [ ~ = 0 is (A+ 16I)v = so it is clear that V3 = [ 1 3 1 is an eigenvector associated with A3 According to Theorem 1. let us suppose that the leftmost car is moving to the right with velocity v0 and at time t = 0 strikes the other two cars. (0) = X2(0) = X3(0) = 0. x2(t) = b1 t .5. x~(O) = x~ (O) = 0. x~(t) = b1 . (26) Substitution of (24b) in (26) gives the equations b1 bl b1 + 2b2 +  4b3 = Vo. (t) = b1 + 2b2 cos 2t + 4b3 cos 4t.3 = 16. which are together but at rest. x~(t) = b1 . . Hence the position functions of the three cars are XI (t) = b1t + b2 sin 2t + b3 sin4t.5 SecondOrder Systems and Mechanical Applications CASE 387 3: A. The corresponding initial conditions are XI (0) x. = vo . The eigenvector equation (A . (25) and their velocity functions are x. the corresponding part of a general solution of x" x3(t) = (a3 cos4t + b3 sin4t)v3. X3 (t) = b1 t . = Ax is The general solution x = x 1 + x2 + x3 of x" =Ax is therefore given by To determine a particular solution. which readily yield a 1 = a2 = a3 = 0.3b3 sin4t. a 1 . r = 16. w 3 = 4.12b3 cos4t. a1 .a2 + a3 = 0.b2 sin 2t + b3 sin 4t . 2b2 + 4b3 = 0 .3a3 = 0.12b3 = 0.
x3(t).. then the three cars travel a distance of 9n . .57 seconds of engagement. x3(t) = 1 3 2 v0 (12t + sin4t). and Fig. 5. 2.x 2 < 0 until t = n j2 . But these equations hold only so long as the two buffer springs remain compressed. while car 3 continues to the right with speed v0 • If. .8 shows the graphs of the functions x 1(t) .~ v0 (sin 2t) (1 + cos 2t) and. vo = 48 feet per second (about 33 miles per hour). X3(t) = 48t . 28. and x 3(t) in Eqs. the position functions + 8 sin 2t + 8sin2t sin 4t).x2(t) = ~vo(sin2t)(l. we compute x2(t) . 5. Finally. (27) ..X2 < 0. and b3 = in (25) are x1 (t) = 31 2 v0(12t 1 32 vo. Then the homogeneous equation Mx" = Kx is replaced with the nonhomogeneous equation Mx" = Kx + F (28) . (27) and (27').2 is subject to an external force Fi (i = 1. 1. while both X2 .57 (seconds). x2(t). similarly.7 illustrates the "before" and "after" situations. Figure 5.5. at which time the equations in (26) and (27) give the values We conclude that the three railway cars remain engaged and moving to the right until disengagement occurs at time t = n j2. Thereafter. for instance.4 sin 4t) = .5 .::::.::::.X 1 < 0 and X3 . To discover what this implies about t. n) in addition to the forces exerted by the springs attached to it.x 1(t) = 1 32 v0( 8 sin 2t .3~ vo(8 sin 2t + 8 sin 2t cos 2t) = . and X) (t) = X2(t) = 9n.. It follows that x 2 .5. that is.x 1 < 0 and x 3 . cars 1 and 2 remain at rest(!). b2 = ~ v0 .388 Chapter 5 Linear Systems of Differential Equations that readily yield b 1 = ~ v0 .15Jr (27') fort > n j2.3 sin 4t). .27 (ft) during their 1. • Forced Oscillations and Resonance Suppose now that the i th mass of the massandspring system in Fig.cos2t).
5. k 2 =50. The case in which w is a natural frequency corresponds to the phenomenon of resonance discussed in Section 2.5 (b) 1. W 11 of the system. Position functions of the three railway cars of Example 2. Then with m 1 = 2. m 2 = 1. an eigenvalue of A. Example 3 Suppose that the second mass in Example I is subjected to the external periodic force 50coswt.5.9. k 1 = 100. (31) exists provided that the external forcing frequency does not equal one of the natural frequencies w 1.. (29) takes the form 11 X = [ 75 50 (33) FIGURE 5. We then anticipate a periodic particular solution Xp(t) = ccoswt (31) with the known external frequency w and with a coefficient vector c yet to be determined. 5. Thus a periodic particular solution of the form in Eq. F2 tiplication by M . followed by cancellation of the common factor cos wt.Fo (32) to be solved for c. FIGURE 5.5 FIGURE 5.5 SecondOrder Systems and Mechanical Applications 389 50 (a) 25 0.5. Because x~ = w 2 c cos wt.5 2. (b) after. w2 .8. substitution of (30) and (31) in (29).7. The forced massandspring system of Example 3.9.5..0 2.1 yields Fn r is the external force vector for the system.5. Mul x" =Ax+ f (29) where f is the external force vector per unit mass. and the substitution x = c cos wt leads to the equation (34) .6. (32) can be solved for cunless w2 = /. Observe that the matrix A + w2 1 is nonsingularin which case Eq. (a) Before. We are especially interested in the case of a periodic external force f(t) = F 0 coswt (30) (where F 0 is a constant vector).0 1. where F = [ F. gives the linear system (A + w 2 1)c = . ••• . Eq. and F0 =50 in Fig.
(w2.25)(w2 .. (t) = ~cos wt.E "0 . The resulting forced periodic oscillation is described by X! 0. (38) Hence Xc(t) is a transient solution that depends only on the initial conditions. then (35) yields c 1 = ~.100)' . c 2 = 1. leaving the steady periodic solution Xp(t) resulting from the external driving force: x(t) ~ Xp(t) as t ~ +oo.m. Periodic and Transient Solutions It follows from Theorem 4 of Section 5. and now the two masses oscillate in synchrony in opposite directions. It is typical for the effects of frictional resistance in mechanical systems to damp out the complementary function solution Xc(t)... The peaks • at w 2 = 5 and w 2 = 10 exhibit visually the phenomenon of resonance. xz(t) = coswt. Figure 5.5.. The resulting forced periodic oscillation is described by XJ(t) = coswt. This system is readily solved for cz =50(w2 . (39) As a practical matter. every physical system includes frictional resistance (however small) that damps out transient solutions in this manner. If the external squared frequency is w 2 = 125. then (35) yields 1. but with the amplitude of motion of m 2 twice that of m 1 • It is evident from the denominators in (35) that c 1 and c2 approach +oo as w approaches either of the two natural frequencies w 1 = 5 and w2 = 10 (found in Example 1). 15. c1 = For instance.. Frequencyamplitude plot for Example 3..10.390 Chapter 5 Linear Systems of Differential Equations for the coefficient vector c = [ c 1 c2 JT.. c2 = 1. (37) where Xp(t) is a particular solution of the nonhomogeneous system and Xc (t) is a solution of the corresponding homogeneous system.1 0 shows a plot of the amplitude J cf + ci of the forced periodic solution x(t) = c cos wt as a function of the forced frequency w. . <t: 5 OOL~~~~ Forced frequency 5 10 15 FIGURE 5. (w2. so that Xc(t) ~ 0 as t ~ +oo. if the external squared frequency is w 2 = 50. xz(t) = cos wt... .:... Io s Thus the two masses oscillate in synchrony with equal amplitudes and in the same direction.25)(w2.3 that a particular solution of the forced system x" = Ax + F o cos wt will be of the form x(t) = Xc(t) (36) + Xp (t).100) (35) CJ  1250 . it dies out with time..5 .75) .
FIGURE 5. The mechanical system of Problem 14.5.5.11.k 1 =50. k2 = 1. x 3 = a3 cos wt.1.) 14.40x + 8y. Then find m 2 so that in the resulting steady periodic oscillations. k2 = k 3 = 4 mt = m2 = 1. and that w = 10. the mass m 1 will remain at rest(!). (18) in the text is 4 . =6 and are acted on by the same force. k3 = 1 m1 = m2 = 1. 3. The massandspring system of Problem 3. 7. using the given masses and spring constants. y" = 12x . k 3 = 0 (no walls) m1 = m2 = 1. and with the given mks values for the masses and spring constants. 16. F2 (t) = 60 cos t 11. with associated eigenvectors v 1 = ( 1 1 ] T and v2 = [ Ct .:. F2 (t) = 0 9. The massandspring system of Problem 2. (b) Assume that the two masses start in motion with the initial conditions x(O) with c.5. k2 = 6. and F0 = 5 in rnks units. k3 = 1 m1 = 1. 5. The two railway cars of Problems 16 through 19.' FIGURE 5. F2(t) = 120cos 3t 10. kt = 2. with F 1 (t) = 30 cos t. . k3 = 2 m1 = 1. m1 = m2 = 1. with F1 (t) = 96cos5t. The massandspring system of Problem 7. 2000 lb j ft). The massandspring system for Problems 1 through 6. x 1'(0) = 0 x 2'(0) = 0 = 19. m2 = 2. fori = 1. kt = k2 = k3 = k4 = 1 1.5. k 3 = 4 1. Figure 5.5. and w = 10 (all in rnks units) in the forced massandspring system of Fig. F0 = 100. This is an example of a dynamic damper. Suppose that m 1 = 2.assumethatm 1 = l.~'. 4. k2 = k3 = 2 mt = m2 = 1. InthesystemofFig. k2 = 2.13 shows two railway cars with a buffer spring. kt = 1. 17. 6. find the natural frequencies of the threemass system of Fig. with F 1 (t) 0. 2.195 cos 7t. kt = 1. k2 = 2. k2 = 25. show that the cars separate after 1r / 2 seconds. respectively. m 2 = k 1 = 75. Consider a massandspring system containing two masses m1 = 1 and m 2 = 1 whose displacement functions x(t) and y(t) satisfy the differential equations = x" = . k 2 = 10. If the two cars of Problem 16 both weigh 16 tons (so that m 1 = m 2 = 1000 (slugs)) and k = 1 tonj ft (that is. 9. k 1 = 1. kt = 0. It has an electrical analogy that some cable companies use to prevent your reception of certain cable channels.5. 15. Thus the original momentum of car 1 is completely transferred to car 2. k2 = 4.5. 12. F 1 (t) = F2(t) = . Describe the resulting motion as a superposition of oscillations at three different frequencies.11 with stiffness matrix In Problems 12 and I 3. Show that the eigenvalues of the coefficient matrix A are A1 = 0 and A2 = c 1 . 2. kt = 2. Find the resulting motion of the system and describe it as a superposition of oscillations at three different frequencies. kt = k2 = k3 = k4 = 2 (Hint: One eigenvalue is A = 4. We want to investigate the transfer of momentum that occurs after car 1 with initial velocity v0 impacts car 2 at rest. kt = 4. For each natural frequency w. m2 = 2. (a) Describe the two fundamental modes of free oscillation of the system. (0) = x~ (0) = 0) in its equilibrium position (x 1(0) = x 2 (0) = 0) with the given external forces Ft (t) and F2(t) acting on the masses m 1 and m 2.5. mt 13. (t) = 0 and x~ (t) = v0 thereafter. x 2 = a2 cos wt. and that x.C2 r.5. x'(O) = 12 and y(O) = 3. give the ratio a 1:a2:a3 of amplitudes for a corresponding natural mode x 1 = a 1cos wt. 5.13.c2 .60y. 1. 5. 5. ~ XI X2 ::. 8. y'(O) FIGURE 5.5. Find the two natural frequencies of the system and describe its two natural modes of oscillation. Thus the effect of the second massandspring pair will be to neutralize the effect of the force on the first mass.5 SecondOrder Systems and Mechanical Applications 391 B_ Problems Problems 1 through 7 deal with the massandspring system shown in Fig. = kjm. Find the solution of the system Mx" = Kx + F that satisfies the initial conditions x(O) = x (0) = 0. m 1 = m2 = m2 = m3 = = m3 = In Problems 8 through 10 the indicated massandspring system is set in motion from rest (x. The analog of Eq.12.12.
let x(t) denote the vertical displacement of the center of mass of the car from equilibrium. x3(0) = . 27. 10008". k 1 = k2 = 2000 = = 100. and that x. We assume that the car body acts as would a solid bar of mass m and length L = L 1 + L2. x.5.6 and discussed in Example 2. (a) Find the two natural frequencies of oscillation (in hertz). x~(O) = vo. (t ). I= 800. L 2 = 3 ft (it's a rearengine car). 19. x~(O) = 2vo x. Then Newton's laws of and an angular oscillation with frequency W2 = )k£2j(21). (b) Assume that this car is driven along a sinusoidal washboard surface with a wavelength of 40ft.8000x + 116.14 (the symmetric situation). 23. show that the two springs are compressed until t = 1r /2 and that FIGURE 5. respectively.(t) = ~v0 and x~(t) = x~(t) = +~v0 thereafter. x~ (t ). k 1 = 1000. and x~ (t) of the three cars for t > 1r /2.5. I= 800. 20.5. Resonance occurs when with w = w 1 or w = w2 . and k = 1500 lbjft. In Problems 27 through 29. The car has front and back suspension springs with Hooke's constants k 1 and k 2 . k 1 = k 2 = 2000 lbjft. that car 2 weighs 8 tons.5. Suppose that m = 75 slugs (the car weighs 2400 lb). L 1 = 6. Thus both cars continue in the original direction of motion. x~(O) = 0.vo x. Determine the values of these constant final velocities x. If cars 1 and 2 weigh 8 and 16 tons.14 is taken as a model for an undamped car with the given parameters in fps units. L 2 = 4. 26. If cars 1 and 2 weigh 24 and 8 tons. x~(O) = 2vo. Find the two critical speeds. the system of Fig.392 Chapter 5 Linear Systems of Differential Equations motion for linear and angular acceleration can be used to derive the equations 18. k2 = 2000 . (0) = vo. m 28. and that each spring constant is 4 tons/ft.6. Model of the twoaxle automobile. 24. v x. which is at distance L 1 from the front of the car.14. (t) = +~vo and x~(t) = +~vo Equilibrium position thereafter. respectively. but cars 2 and 3 continue with the same velocity.5. (0) = 3vo. Problems 20 through 23 deal with the same system of three railway cars (same masses) and two buffer springs (same spring constants) as shown in Fig. x3(0) = 2vo In the threerailwaycar system of Fig. k 1 = k2 = 2000 100. respectively. show that the cars separate after 1r /2 seconds. Show that the railway cars remain engaged until t = 1r /2 (s ). Then show that every free oscillation is a combination of a vertical oscillation with frequency The TwoAxle Automobile In Example 4 of Section 2. I= 1000. and I = 1000 ft·lb·s 2 • Then the equations in (40) take the form 75x" + 4000x  8oooe = o. 5. Figure 5. (0) = 2v0 . m = 100.5. after which time they proceed in their respective ways with constant velocities.(O) = v0 . (a) Find the two natural frequencies w 1 and w 2 of the car. Find the corresponding two critical speeds of the car (in feet per second and in miles per hour). L 1 = L 2 = 5. Suppose that k 1 = k 2 = k and L 1 = L 2 = ~L in Fig. but with different velocities.vo x. 5. The cars engage at time t = 0 with x 1 (0) = x 2 (0) = x 3 (0) = 0 and with the given initial velocities (where v0 = 48ftls). let e(t) denote its angular displacement (in radians) from the horizontal. and k = 3000 lb/ft.6 we investigated the vertical oscillations of a oneaxle caractually a unicycle. 5. L 1 = 7 ft. Thus the two cars rebound in opposite directions.oooe = o. Thus car 1 rebounds. 22.(t) = ~vo and x~(t) = +~v0 (40) thereafter. per second along a washboard surface shaped like a sine curve with a wavelength of 40 ft. (b) Now suppose that the car is driven at a speed of v feet x. If (0) = 0 and x~ (0) = x3 (0) = 0. When the car is in motion. £ 1 = L2 = 5. It has moment of inertia I about its center of mass C. The result is a periodic force on the car with frequency w = 2nvj40 = nv/20. 21. show that the two cars separate after 1r /3 seconds. x~ (0) = 0. In each problem you should find (as in Example 2) that the first and third railway cars exchange behaviors in some appropriate sense. 25. m 29. 5.14 represents the suspension system of such a car. and that x. Now we can analyze a more realistic model: a car with two axles and with separate front and rear suspension systems. x3 (0) = . suppose that cars 1 and 3 each weigh 32 tons.
)(.. Example 1 Find a general solution of the system 4 1 (5) 4 Solution The characteristic equation of the coefficient matrix A in Eq. then a general solution of the system =Ax dt is given by dx (l ) (2) with arbitrary constants c 1.8A +A 2 ) = (5. as needed to form the general solution in (2). An with respective associated eigenvectors v 1. If every eigenvalue of the matrix A is complete..)+ 24] = (3. so there is at least one eigenvector associated with A.A. A2 . Vn associated with the eigenvalues AJ..6 Multiple Eigenvalue Solutions 393 Multiple Eigenvalue Solutions ·· m••••«•~ In Section 5. the eigenvector equation (A. c2 . Let us call an eigenvalue of multiplicity k complete if it has k linearly independent associated eigenvectors. In this section we discuss the situation when lA.All= 0 (3) does not have n distinct roots.Al)v = 0 (4) has at least one nonzero solution v. Vn . and thus has at least one repeated root. For each eigenvalue A.All= 6 6 0 0 3A 4 = (3 . But an eigenvalue of multiplicity k > 1 may have fewer than k linearly independent associated eigenvectors.A)(15. In this case a general solution of x' = Ax is still given by the usual combination in (2). . . . .4 we saw that if the n x n matrix A has n distinct (real or complex) eigenvalues AJ. . An eigenvalue is of multiplicity k if it is a kfold root of Eq. . .5. en. . . (3).. (5) is 9A 4 1A lA .. .A)[(9 . the characteristic equation •. v2. A2. v2. . Thus A has the distinct eigenvalue AJ = 5 and the repeated eigenvalue A2 = 3 of multiplicity k = 2...A)(3 . thenbecause eigenvectors associated with different eigenvalues are linearly independentit follows that A does have a complete set of n linearly independent eigenvectors v 1 .A. An (each repeated with its multiplicity).A) 2 = 0..1 . In this case we are unable to find a "complete set" of n linearly independent eigenvectors of A.
then we must choose a to be nonzero. The fact that Eq. Thus we have found a complete set v 1 .A2 = 3. v2 .3I)v = [ 6 6 so the nonzero vector v = 4 4 4 [a b c r is an eigenvector if and only if 6a + 4b = 0. if a = 2 (to avoid fractions). The choice a = 1 then yields the eigenvector associated with the eigenvalue A. yields b = a. For instance. If c = 1. 3.wherev=[a (A .50v =[ ~ ~ J] [n ~ l =[ b cf. b = ~a. 3.3c3e 3t. Each of the first two equations.394 Chapter 5 Linear Systems of Differential Equations CASEl: )q=5.6b = 0.3 or is a second linearly independent eigenvector associated with the multiplicity 2 eigenvalue A. v3 of three eigenvectors associated with the eigenvalues 5. If c = 0. 4a + 4b = 0 and 6a .2 = 3. then we may choose a = b = 0. (6) does not involve c means that cis arbitrary. CASE 2: A. (5) is (7) with scalar component functions given by x 1 (t) = x2 (t) c 1est c1 est + = c1 est 2c3e 3t. Then the third equation reduces to 2a . so that c = a. . subject to the condition v =I= 0. (6) that is. 1 = 5. Now the eigenvector equation is 6 (A . • . X3 (t) = + c2e3t. this gives the eigenvector V2 = [ o o 1 r associated with . then b = 3.2 = 3.2c = 0. so V3 = [2 . IS Theeigenvectorequation(AU)v=O. The corresponding general solution of Eq.
A. and the new general solution X(t) = CJ VJ eSt+ C2V2e 3t + c 3 ve 3t would be equivalent to the one in Eq. 3a + 3b = 0.. Thus we need not worry about making the "right" choice of independent eigenvectors associated with a multiple eigenvalue.5. and hence is incomplete. we could choose v rather than v3 as our third eigenvector. we generally make the simplest one we can. The fact that b = ~a for any eigenvector associated with .4) 2 = 0.A 1 = 4 of multiplicity 2. • r. given a and c not both zero.A 2 = 3 means that any such eigenvector can be written as and thus is a linear combination of v2 and v3 . (7). The eigenvector equation (A.A)+ 9 = .All = 11 3 . Defective Eigenvalues The following example shows thatunfortunatelynot all multiple eigenvalues are complete.3 7 .. Therefore any eigen vector associated with AI = 4 is a nonzero multiple of v = [ I 1 Thus the multiplicity 2 eigenvalue .6 Multiple Eigenvalue Solutions 395 Remark: Our choice in Example I of the two eigenvectors V2 = [0 0 ] r and V3 = [2 3 0 r associated with the repeated eigenvalue .3a . Example 2 The matrix (8) has characteristic equation lA . Therefore.A 1 = 4 has only one independent eigenvector.A)(7.41)v = [ 3 3 then amounts to the equivalent scalar equations .A . .3b Hence b = = 0.A = (1 .A 2  8.A + 16 =(. • Any choice will do..A 2 = 3 bears comment. Thus A has the single eigenvalue . a if v = [a b JT is to be an eigenvector of A.
we might hope to find a second solution of the form (11) When we substitute x = v 2 teJ.t in x' =Ax. and hence that x2 (t) = 0. Thus the defective eigenvalue AJ = 4 in Example 2 has multiplicity k = 2 and defect d = 1. then the eigenvalue method as yet described will produce fewer than the needed n linearly independent solutions of the system x' = Ax.. we get the equation But because the coefficients of both eJ. The Case of Multiplicity k = 2 Let us begin with the case k = 2. and suppose that we have found (as in Example 2) that there is only a single eigenvector v 1 associated with the defective eigenvalue A.396 Chapter 5 Linear Systems of Differential Equations An eigenvalue A of multiplicity k > 1 is called defective if it is not complete. we get the equation v 1te>. This means thatcontrary to our hope.t (13) We equate coefficients of eJ.t here.t must balance. We therefore need to discover how to find the "missing solutions" corresponding to a defective eigenvalue A of multiplicity k > 1. .. let us extend it slightly and replace v2 t with v 1t + v2 • Thus we explore the possibility of a second solution of the form (12) where v 1 and v2 are nonzero constant vectors. By analogy with the case of a repeated characteristic root for a single linear differential equation (Section 2.the system x' =Ax does not have a nontrivial solution of the form assumed in (11). If the eigenvalues of the n x n matrix A are not all complete. then the number d=kp (9) of "missing" eigenvectors is called the defect of the defective eigenvalue A... it follows that v2 = 0.3). When we substitute x = in x' = Ax.H)vi =0 (14) and (15) that the vectors v 1 and v2 must satisfy in order for (12) to give a solution of x' = Ax. If A has only p < k linearly independent eigenvectors.t and teJ. because we saw that it has only p = 1 associated eigenvector..t and teJ. and thereby obtain the two equations (A. (11). Then at this point we have found only the single solution (10) of x' = Ax. Instead of simply giving up on the idea behind Eq.r +v2 e>.
Al)v2 is nonzero. If . the procedure described in the following algorithm always succeeds in finding two independent solutions associated with such an eigenvalue. in order to solve simultaneously the two equations in (14) and (15). X1 = [ 1 3 3] 7 X. (15) says that the vector v2 satisfies the equation It follows that. Consequently. it could happen that (A .5. it suffices to find a solution v2 of the single equation (A . (16) is [~ ~ J v2 = 0.6 Multiple Eigenvalue Solutions 397 Note that Eq. and therefore is satisfied by any choice of v2 . (14) merely confirms that v 1 is an eigenvector of A associated with the eigenvalue A. It turns out that this is always possible if the defective eigenvalue A of A is of multiplicity 2.41)v2 is nonzero (as desired) for some choices of v 2 though not for others.U) 2 v2 = 0 such that the resulting vector v 1 = (A. Then form the two independent solutions (18) and (19) of x' =Ax corresponding to A. 2. We therefore begin by calculating Hence Eq. and therefore is an eigenvector v 1 associated with A. First find a nonzero solution of the equation (16) such that (17) is nonzero. (20) has the defective eigenvalue A = 4 of multiplicity 2. ALGORITHM Defective Multiplicity 2 Eigenvalues v2 1. (20) Solution In Example 2 we found that the coefficient matrix A in Eq. In principle. Then Eq.
x 2(t) = 3c 1 e 41 .3x2 . The multiplicity 2 method described earlier boils down to finding a pair {v 1.. 0 1 2 3 4 FIGURE 5. and therefore is an eigenvector associated with A = 4. 5. then a rank r generalized eigenvector associated with A is a vector v such that (A.1 v i= 0.H)'v =0 but (A.H)vk = vkJ.j I xr (t) = ( 3c2t x2 (t) = 2 3 4 + c2.1. [  x 2( t) 4 3 = (vrt + v2 ) e 4t = 3t t+ 1 e 41 .4I)v2 = [ .] [bJ= [ . v2 } of generalized eigenvectors.. each solution curve with c2 i= 0 is tangent to the line x 1 = .J. Generalized Eigenvectors The vector v 2 in Eq. x2(t)) then recedes along this line away from the origin as t ___.6. which parametrize the line x 1 = x2 in the x 1x 2 plane. Direction field and solution curves for the linear system x.. such that (A.H)'. (It is 3 times the eigenvector found in Example 2.. 3 J The resulting general solution has scalar component functions ~ 0 1"f"'+'+f~~~.1. the point (x 1 (t). (3c2t + 3c 1 )e 41 • x. = x 1 . As indicated in Fig.6..x 2 at the origin. The vector v 2 in ( 16) is a rank 2 generalized eigenvector (and not an ordinary eigenvector).) Therefore the two solutions of Eq. one of rank 1 and one of rank 2. (16) is an example of a generalized eigenvector. (20) given by Eqs. A length k chain of generalized eigenvectors based on the eigenvector v 1 is a set {v 1 . ••• . then (21) simply means that v is an eigenvector associated with A (recalling the convention that the Oth power of a square matrix is the identity matrix). oo and approaches +oo along the • solution curve as t ___.] = Vt is nonzero. (A.H)vz = v 1 • . +oo.3c 1)e4r.H)vk. The point (x 1 (t). (22) (A. x~ = 3x 1 + 7x2 of Example 3. (21) If r = 1. With c2 = 0 these solution equations reduce to the equations x 1(t) = .1. to the northwest if c 1 > 0 and to the southeast if c 1 < 0.I = Vk2.398 Chapter 5 Linear Systems of Differential Equations we try v 2 = [ 1 0 ]T we find that 3 3 (A.3c 1e41. v 2 . (18) and (19) are x 1 ( t ) = v 1e 4t = [ 3] 3 e4t . vk} of k generalized eigenvectors such that (A. If A is an eigenvalue of the matrix A.H)v2 = v 1 • Higher multiplicity methods involve longer "chains" of generalized eigenvectors. x 2 (t)) approaches the origin as t ___. Thus a rank 1 generalized eigenvector is an ordinary eigenvector. +oo.
A3  3.2 · (3.5.A = 1 · [7.1 .. indeed.A of the matrix A.A 2  3. define a solution ofx' =Ax.A)+ 5] = . (25) is lA .(A + 1) 3 = 0. . (23). this is not always possible).3 . as we will see. x2(t) = (v. (A. (24) For instance. v2 . v 3 } is a length 3 chain of generalized eigenvectors associated with the multiple eigenvalue .6 Multiple Eigenvalue Solutions 399 Because v 1 is an ordinary eigenvector. Looking at Eq.1 = .5 1 A . v2 . x 3 (t) in (24) does.A.A)]+ (A)[(. the equations in (22) give so Therefore..A .AI)v 1 = 0. v3 } of generalized eigenvalues associated with A.. we see that we need only find a solution v3 of such that the vectors are both nonzero (although.A)(3.. it follows from (22) that (23) If {v 1 . Therefore.t + v2)eAI. in order to "handle" a multiplicity 3 eigenvalue . it suffices to find a length 3 chain {v 1 . Example 4 Find three linearly independent solutions of the system (25) Solution The characteristic equation of the coefficient matrix in Eq.A 0 1 . Consequently.All = [ . then it is easy to verify that three linearly independent solutions of x' = Ax are given by Xt(t)=VjeA.7 2 ] .
v. 0. = 1 of multiplicity 3. we calculate + l) 3 v3 0. = (A+I)v 3 =[~ 2 0 1 n[~J=[n n [n =[ VI= (A+ I)v 2 =[~ 2 0 =n Note that v 1 is the previously found eigenvector v with a = 2. To apply the method described here for triple eigenvalues. we first calculate r (A+ 1) 2 and = [~ [~ 1 2 0 n [~ 2][ 2 i . Substitution in (24) now yields the linearly independent solutions of the system x' = Ax. 1 1 2 0 i2] [2 = .= 1 is 2. this agreement serves as a check of the accuracy of our matrix computations. The eigenvector equation (A.1. and so the defect single associated eigenvector v = [ a of A. v3 } of generalized eigenvectors associated with the triple eigenvalue A. then the first row a + b + 2c = 0 gives b = a. to within a constant multiple. Thus.400 Chapter 5 Linear Systems of Differential Equations and thus A has the eigenvalue A.H)v = 0 for an eigenvector v = [a b c is r The third row a + c = 0 gives c = a. • . the eigenvalue A. r. Thus we have found a length 3 chain {v 1 . = . 1 1 3 1 3 3] 1 (A+ 1) 3 = 2 0 1 1 3 1 3 3] = [0 0 0 0 0 0 n Thus any nonzero vector v 3 will be a solution of the equation (A Beginning with V3 = [ 1 0 0 for instance. v2 . = 1 has only the a a with a f.
These n generalized eigenvectors may be arranged in chains.t. (27) ) A. vd of generalized eigenvectors (with v 1 an ordinary eigenvector associated with A) determines a set of k independent solutions of x' = Ax corresponding to the eigenvalue A: Xt(t) = VJeAt. Note that. • Two length 1 chains and a length 2 chain (defect 1). ALGORITHM Chains of Generalized Eigenvectors Begin with a nonzero solution u 1 of Eq. or • A length 4 chain (defect 3).Jt + vk e . . we can begin with the equation > (A. For instance.1)! + ··· + Vk.l)uk1 = Uk :j:. x 3(t) = (~v1t 2 + v2t + v 3) eA. But the structure of these chains depends on the defect of A. • A length 1 chain and a length 3 chain (defect 2). • Two length 2 chains (defect 2). . where d is the defect of the eigenvalue. X2(t) = (Vtf + V2) eA. If (A . Consequently..t + vk. (26) and successively multiply by the matrix A . a multiplicity 4 eigenvalue can correspond to • Four length 1 chains (defect 0).t.Al)uk = 0.6 Multiple Eigenvalue Solutions 401 The General Case A fundamental theorem of linear algebra states that every n x n matrix A has n linearly independent generalized eigenvectors. but (A .5. 0.Al)d+lu = 0 (26) to start building the chains of generalized eigenvectors associated with A.. in each of these cases. xk(t)= ( VJ t k 1 (k . and can be quite complicated.. once we have found all the ordinary eigenvectors associated with a multiple eigenvalue A.2t 2 2! . then the vectors (listed in reverse order of their appearance) form a length k chain of generalized eigenvectors based on the (ordinary) eigenvector v 1• Each length k chain {v 1./. the length of the longest chain is at most d + 1. v2 . and therefore know the defect d of A. with the sum of the lengths of the chains associated with a given eigenvalue A equal to the multiplicity of A.AI until the zero vector is obtained.
= k(XJ .21 .2.c(x. of magnitude c(x. • If two chains of generalized eigenvectors are based on linearly independent eigenvectors.x. Then A. then the union of these two chains is a linearly independent set of vectors (whether the two base eigenvectors are associated with different eigenvalues or with the same eigenvalue). whereas A. The railway cars of Example 6. x1) . respectively.C(X~ .402 Chapter 5 Linear Systems of Differential Equations Note that (27) reduces to Eqs. X2(t) (28) In terms of the position vector x(t) = [ X!(t) written in the matrix form r. . x3(t) = (v1t + v2)e21. 1 = 2 and A. • An Application Figure 5. and therefore produce a complete set of n linearly independent solutions of x' = Ax when we amalgamate all the "chains of solutions" corresponding to different chains of generalized eigenvectors.c1xi . X2(t) = v 1e. Example 5 Suppose that the 6 x 6 matrix A has two multiplicity 3 eigenvalues A. v 2 } of generalized eigenvectors (with the eigenvectors u1 and VJ being linearly independent).2 shows two railway cars that are connected with a spring (permanently attached to both cars) and with a damper that exerts opposite forces on the two cars.C2X~ . To ensure that we obtain n generalized eigenvectors of the n x n matrix A that are actually linearly independent. and w 3 are then linearly independent and yield the following six independent solutions of x' = Ax: XJ (t) = u 1e21 .x~ ) .X4. The six generalized eigenvectors u 1. these equations can be (29) Mx" = Kx + Rx' . (iw1t 2 + w2t + w 3 ) e 31• As Example 5 illustrates. . (18) through (19) and (24) in the cases k = 2 and k = 3. w2.). v 1 . 1 must have an associated eigenvector 01 and a length 2 chain {v 1 . xs(t) = (w1t x6(t) = + w2)e 31 . we may rely on the following two facts: • Any chain of generalized eigenvectors constitutes a linearly independent set of vectors. The determination of the chain structure associated with a given multiple eigenvalue can be more interesting (as in Example 6).6. . the computation of independent solutions corresponding to different eigenvalues and chains of generalized eigenvalues is a routine matter. respectively.x~) proportional to their relative velocity. . w 1. The two cars are also subject to frictional resistance forces c 1xi and c 2x~ proportional to their respective velocities.6.X2 ) . An application of Newton's law rna = F (as in Example 1 of Section 5.1) yields the equations of motion m1x~' = k(x2 m 2X~ FIGURE 5. w3} of generalized eigenvectors based on its single eigenvector w 1.(t) = w 1e 31 .2 = 3 with defects 1 and 2. v2 .2 must have a length 3 chain {w 1. w2 .
A. x3(t) =xi (t). the system in Eq. If m1 = m2 = 1 we get x' =Ax (30) where nOW X = [ Xt X2 X3 X4 r 0 0 k and A= [ ~ 1 0 (c+ct) (31) k c Example 6 With m 1 = m 2 = c = 1 and k = c 1 = c2 = 2. (32) 1 3 It is not too tedious to calculate manuallyalthough a computer algebra system such as Maple. 0 = 0. 0 = 0. Instead. or MATLAB is useful herethe characteristic equa tion of the coefficient matrix A in Eq.5 cannot be used. and x4(t) = x~(t).5. Thus A has the distinct eigenvalue . and R = [ (c c+ cJ) (c c + c2) ] is the resistance matrix.6 Multiple Eigenvalue Solutions 403 where M and K are mass and stiffness matrices (as in Eqs.A. 1 = 2. (30) is 0 0 0 0 1 f X= [ 2 2 2 2 0 3 : ] X.A. Thus Vo = [ 1 r is an eigenvector associated with .5). 0 = 0 and the triple eigenvalue A. CASE 1: . then the last two rows yield a = b. Mathematica. (2) and (3) of Section 5. . x2(t). Unfortunately. the methods of Section 5. (32). The eigenvalue equation (A . we write (28) as a firstorder system in the four unknown functions x1 (t).Al)v = 0 for the eigenvector v = [ a b c d ] T is Av= [ ~ 0 1 0 0 2 3 2 1 JJUJ =UJ· 1 0 0 The first two rows give c = d = 0. because of the presence of the term involving x'.
Hence v is determined by the first two equations. we need a generalized eigenvector of rank 2. . However. The eigenvector v 1 just found is neither of the two eigenvectors o 1 and 0 2 found previously. is nonzero. and therefore are redundant. then solve for c and d. we could choose either w 1 = o 1 or w 1 = 0 2 .404 Chapter 5 Linear Systems of Differential Equations CASE 2: AI = 2. b = 0 yields c = 2. and therefore is an eigenvector associated with A 1 = 2. 2a + c = 0 and 2b + d = 0.o 2 . For a length 1 chain w1 to complete the picture we can choose any linear combination of o 1 and o 2 that is independent of v 1• For instance.2 =v. v2} is the length 2 chain we need. and we find that (A + 21)v2 = [ _ 2 2 0 2 0 2 2 1 0 .H)v = 0 is (A+ 21)v = [ ~ 0 2 2 1 0 1 1 2 The third and fourth scalar equations here are the differences of the first and second equations. we will see momentarily that the particular choice W1 = 01 + 0 2 = [ 1 r yields a solution of the system that is of physical interest. Then {v 1. d = 1 2 and thereby the eigenvector = [0 Because A 1 = 2 has defect 1. d = 0 and thereby the eigenvector 01 = [ 1 0 2 The choice a = 0. b = 1 yields c 02 = 0.1 1 2 J] [J] [=lJ 1 2 . The choice a = 1. but we observe that v 1 = o 1 . The eigenvalue equation (A. We can choose a and b independently. Thereby we obtain two eigenvectors associated with the triple eigenvalue A 1 = 2. and hence a nonzero solution v2 of the equation 2 2 (A + 21) 2 V2 2 = [ 0 0 2 0 0 Obviously. V2 = [0 0 is such a vector.
e. (33) e. It is of interest to interpret physically the individual generalized eigenvector solutions given in (33). The degenerate (. x. The four scalar components of the general solution are described by the equations x 1(t) = X2(t) = x3(t) = x4(t) = + e .21 corresponding to the carefully chosen eigenvector w 1 describes damped motions = e. so x.c4 + 2c4t).C3 = 0. r r r e2t.21 ( 2c2 + 2c3.21 = [ 1 1 x4(t) = (v.6 Multiple Eigenvalue Solutions 405 Finally.t + v2)e.C4 = vo are readily solved for c 1 = ~v0 . Finally. (t) = [ 1 1 o describes the two masses at rest with position functions x 1 (t) .o = 0) solution x. (t) 0 2 2 2 2 = [t t 2t + 1 2t.21 of the system x' = Ax in (32). In this case the two railway cars continue in the same direction with equal but exponentially damped velocities. c1 (34) Recall that x 1 (t) and x 2 (t) are the position functions of the two masses. approaching the displacements x 1 = x 2 = ~ v0 as t+ +oo. Then the equations X! (Q) = Ci + C2 + C3 = 0. x~ (0) = .21 (c2 + c3 + C4t).C3.5. v2 } yield the four independent solutions = v 0e 0·1 = [ 1 1 1 x2(t) = w. For instance.2c2 + 2c3 . x~ (0) = . e21.e21 = [ 1 x3(t) = v.A.e. suppose that x 1 (0) = x 2 (0) = 0 and that x~ (0) = x~(O) = v0.21 ). whereas x 3(t) = x~ (t) and x 4(t) = x~(t) are their respective velocity functions. and c3 = c4 = 0.1 and x 2 (t) The solution 1 . and {v 1 . {wt}. the chains {v0 }.~v0 .2c2 .C4t).2c4t).21 of the two masses. X2(Q) = Ci + C2. v2 } both describe damped motion with the two masses moving in opposite • directions.1 or. the solutions x 3 (t) and x4 (t) resulting from the length 2 chain {v 1 .21 x.2c3 + c4. (t) = x2(t) = iva (1  e. .2c3 + C4 = Vo.21 and x 2 (t) = e. e21 ( 2c2. c2 = . e. (t) = x~(t) = voe.2 x 1 (t) 2 or r = 1.21 . with equal velocities in the same direction. Ci + e2t (C2.
x' = [ .2 28 15 15 ~ 100] x. =a. x' = 5 1 5] 3 10 x. 2 15.3 ~]X . 3 25. A = 2. 17. x' = 16. x' = i i 1] 1 X 1] 5 X 4] 9 X 18. 2 8 .f3i and I= a+ f3i each of multiplicity k. x' = [13 [ 3 ~ ~ 11.1 3 2 16] 16 . x' ~[ i 3 [ 3 ~ A= 3. 0 Problems ••P••••· . x' ~ 21.f3i) as if it were real to find k independent complexvalued solutions.1 ~ :]x 39 23. 12. 2. A= 1 . x' = 25 n·~]· 13 0 33 2~ 20. 3 In Problems 23 through 32 the eigenvalues of the coefficient matrix A are given. A = 2. A= 2. 3. ~} ~] 4 1 12 4 6 1 X ~]x 22. x' ~[2 : 19. 3 1 1 6 ~]· x. Especially in Problems 29 through 32. x' = [ ~ 4. x' = [ 2 !]x.1 0 1 1 0 .29 60 57 13. x' ~ ~Jx 0 9 0 12 5 6 12 5 4 40 23 0 0 1 0 0 1 . In Problems 1 through 6. use a computer system or graphing calculator to construct a direction field and typical solution curves for the given system. use of a computer algebra system may be useful.!]X 2] 5 X 2. 9. x' = 84] X 48] X 4] X ~ .19 0 8 10. Given a complex conjugate pair a ± f3i of eigenvalues of multiplicity k. x' ~ [~ [ 1~ 8. 2. x' = [ 1 ~ 14. x' = [ 36 72 24. Find a general solution of the indicated system x' = Ax. x' = [ .. x' = [ 8 5 16 50 33 30 17 6 1 . x' = [ 6. 3.1 x ' ~ [1i x' ~ [~ u i} 1 2 0 0 4 3 2 1 0 1 2 0 0 0 1 0 7 4 3 14 u 1 0 0 1 0 q. x' = [ 5. 3. 3 27.406 Chapter 5 Linear Systems of Differential Equations The methods of this section apply to complex multiple eigenvalues just as to real multiple eigenvalues (although the necessary computations tend to be somewhat lengthy).• • •••• • Find general solutions of the systems in Problems 1 through 22. x' = [ .X ~ [ ~ [ . See Problems 33 and 34.2 ~ !Jx . x' ~ [ I~ 27 0 7 9 0 3 4 5 1.X ~[ i n· n· 26. x' = [ 2 _1 3. The real and imaginary parts of these complexvalued solutions then provide 2k realvalued solutions associated with the two eigenvalues A.2 3 ~ [! 7. a .1 9 4 3 0 . we work with one of them (say. ~]x.
2 ~ ~!]x.)= (A. = 2 + 3i. Show that. . = 1. before stopping. The solution vectors of an n x n homogeneous linear system x' =Ax ~(t) (1) can be used to construct a square matrix X = tial equation that satisfies the matrix differen X' = AX (1 ') . x.1 3 0 27 9 3 23 1 26 6 3 0 0 [ ll . 6 45 25 A.2 parts of the complexvalued solutions VIeAI and (VI(+ Vz)eAI 1 [ 29. First show that the complex vectors VJ = [ i 3 + 3i 0 . I X= is [3 4 ~ 4 3 0 0  I 0 3 4 IJ = [9 X form a length 2 chain {vJ. x' = 1 ~ 4 1 ~··=U 2 2 1 2 3 5 13 22 27 to find four independent realvalued solutions of X1 = Ax.5.= 2. + 13) 2 = 0.1.2. x' = 0 9 5 1 48 3 138 30 18 A.1. A has the repeated complex conjugate pair 3 ± 4i of eigenvalues. + 25) 2 = 0.3.24 0 6 x.7 Matrix Exponentials and Linear Systems 407 28. Then calculate the real and imaginary How far do the cars travel before stopping? 36. Repeat Problem 35 under the assumption that car 1 is shielded from air resistance by car 2. 32. v 2 } associated with the eigenvalue A. 2 X= 1 [ ~~ 9 33 0 1 8 0 25 90 3 10 9 ~] 32 X 12 4 8 3 19 1 31. 34. X = i~ 9 x.)= (A. and the initial conditions are XI Therefore.1. First show that the complex vectors VI= [ (0) = Xz(O) = 0.1 if the physical parameters are given by ¢(A. 2. 2  4A. X1 = [ 15 34 17 7 16 7 l!]x. The characteristic equation of the coefficient matrix A of the system I~] x.=1. 1 0 0 r and v2 0 1 i r form a length 2 chain {vi. The characteristic equation of the coefficient matrix A of the system [ 35 30] is ¢(A.3.2. (0) = x~(O) = Vo. v 2 } associated with the eigenvalue A. Find the position functions XI (t) and x 2 (t) of the railway cars of Fig. A. Then calculate (as in Problem 33) four independent realvalued solutions of X 1 =Ax. A has the repeated complex conjugate pair 2 ± 3i of eigenvalues. 5.6.l. A=2. ~ 3] ~ Therefore. 2 6A.=1. the cars travel twice as far as those of Problem 35.1.4i. ]T . = 3 .2. A. so now ci = 0. 3 33. 35.
Then then x n matrix (2) having these solution vectors as its column vectors. that is. (4 ). Then the [unique] solution of the initial value problem x' = Ax.. (1). x2 (t). (1). so W(t) is a fundamental matrix for the system in (1 ). If lJI(t) is any other fundamental matrix for (1). Xn(t) are n linearly independent solutions of Eq. the general solution (3) of the system x' = Ax can be written in the form x(t) = ci>(t)c (4) where c = [c 1 c2 cn]T is an arbitrary constant vector. Fundamental Matrix Solutions Because the column vector x = x j (t) of the fundamental matrix ci> (t) in (2) satisfies the differential equation x' = Ax. any nonsingular matrix solution 'II (t) of Eq. In order that the solution x(t) in (3) satisfy a given initial condition x(O) = xo. THEOREM 1 Fundamental Matrix Solutions Let ci>(t) be a fundamental matrix for the homogeneous linear system x' = Ax. . (5) it suffices that the coefficient vector c in (4) be such that ci>(O)c = x0 . it follows (from the definition of matrix multiplication) that the matrix X = cl>(t) itself satisfies the matrix differential equation X' = AX. is called a fundamental matrix for the system in (1). . x(O) = Xo (7) is given by (8) . and therefore has an inverse matrix ci>(t).1 • Conversely. we get the conclusion of the following theorem. (1). (4) that 'll(t) = ci>(t)C (4' ) for some n x n matrix C of constants.408 Chapter 5 Linear Systems of Differential Equations associated with Eq. In terms of the fundamental matrix ci>(t) in (2). it also follows that the fundamental matrix cl»(t) is nonsingular. Suppose that x 1 (t). (1') has linearly independent column vectors that satisfy Eq. then each column vector of W(t) is a linear combination of the column vectors of ci>(t). that (6) When we substitute (6) in Eq. Because its column vectors are linearly independent.. so it follows from Eq.
A. . Therefore.) The formula in (12) is also valid upon generalization to n x n matrices.4 tells us how to find a fundamental matrix for the system x' =Ax (9) with constant n x n coefficient matrix A.y . Xn as column vectors is a fundamental matrix for the system x' = Ax. 0. . Vn associated with the (not necessarily distinct) eigenvalues )q. (12) where ~ = det(A) f. then x n matrix (10) having the solutions x 1. (8).•. The inverse of the nonsingular 3 x 3 matrix A 1 = +A11 1 ~ [ Azi + A31 A12 +A13] +A22 . respectively.!. y' = 3x. n. we must be able to compute the inverse matrix «Jl(0). . Example 1 Find a fundamental matrix for the system x' = 4x + 2y . 0 and AiJ denotes the determinant of the 2 x 2 submatrix of A obtained by deleting the ith row and jth column of A... 2. v2 . In order to apply Eq. In this event the corresponding solution vectors of Eq. . An. . but in practice inverses of larger matrices are usually computed instead by row reduction methods (see any linear algebra text) or by using a calculator or computer algebra system. (12). 2 .5. ..7 Matrix Exponentials and Linear Systems 409 Section 5. (Do not overlook the symbol T for transpose in Eq.. . ••• ._ [ ~ c d b a J' (11) f.A32 A23 + A 33 T . x2 . y (O) = .1. at least in the case where A has a complete set of n linearly independent eigenvectors v1 . The inverse of the nonsingular 2 x 2 matrix is A I where ~ = det(A) = ad be A = [aiJ] is given by = _. (9) are given by fori = 1. (13) then use it to find the solution of (13) that satisfies the initial conditions x(O) = 1.1.
21 2es 1 3e. y(O) = 49.1.21 9e.21 + 80es1 ] + 40es 1 • • . and so x(t) = 1 [ 3e2t + 4esr ] 7 _9e . we can calculate rapidly by matrix multiplication the solutions corresponding to different initial conditions.2t 3e2t 2es1 est ] [ 3] 2 .2r + 2 esr · Thus the solution of the original initial value problem is given by • Remark: An advantage of the fundamental matrix approach is this: Once we know the fundamental matrix c)(t) and the inverse matrix c)(Q) .410 Chapter 5 Linear Systems of Differential Equations Solution The linearly independent solutions 2t [ 3:2t x.4 yield the fundamental matrix (14) Then and the formula in ( 11) gives the inverse matrix (15) Hence the formula in (8) gives the solution 21 x(t) = [ 3:. For example.21 eSt ] [ 21] _[ 280  3e. suppose that we seek the solution of the system in (13) satisfying the new initial conditions x(O) = 77. Then substitution of (14) and (15) in (8) gives the new particular solution = 7 1[ e.2t 2est] est (~) 7 [ 3 1 2] [ 1] 1 1 = (~) 7 [ e . (t) = J found in Example 1 of Section 5.
k An) n! (18) where A0 =I. A2 = AA. and so on.k ? oo ?. 2! 3! n! z2 z3 zn (16) Similarly.7 Matrix Exponentials and Linear Systems 411 Exponential Matrices We now discuss the possibility of constructing a fundamental matrix for the constantcoefficient linear system x' = Ax directly from the coefficient matrix Athat is. The exponential ez of the complex number z may be defined (as in Section 2. ranging from the scalar equation x' = kx with solution x(t) = x 0ekt to the vector solution x(t) = veJ..5. A3 = AA 2 . Example 2 Consider the 2 x 2 diagonal matrix Then it is apparent that . An+! = AAn ifn ~ 0.+ · · · + .+ . if A is an n x n matrix. the exponential matrix eA is defined (by Eq. 00 An ( lim n! . That is. with associated eigenvector v.3) by means of the exponential series ez = 1+ Z + . The meaning of the infinite series on the right in (17) is given by ?. then the exponential matrix eA is the n x n matrix defined by the series eA=I+A++···++··· 2! n! A2 An (17) where I is the identity matrix. We now define exponentials of matrices in such a way that X(t) =eAr is a matrix solution of the matrix differential equation X'=AX with n x n coefficient matrix Ain analogy with the fact that the ordinary exponential function x(t) =eat is a scalar solution of the firstorder differential equation x' =ax. (17)) for every square matrix A. We have seen that exponential functions play a central role in the solution of linear differential equations and systems.+ · · · . without first applying the methods of earlier sections to find a linearly independent set of solution vectors. inductively.r of the linear system x' = Ax whose coefficient matrix A has eigenvalue A. It can be shown that the limit in ( 18) exists for every n x n square matrix A.
+ ··· 2! = [ 1 0] A2 01 + [ a Ob 0] + [ a /2! 0 2 Thus eA = [ e. At is obtained simply by multiplying each element of A by t . (22) In Problem 32 we ask you to conclude that (eArl= eA. (23) In particular. The exponential matrix eA satisfies most of the exponential relations that are familiar in the case of scalar exponents. (17) gives e At =I+At+A . 0 0 (19) 0 (20) 0 obtained by exponentiating each diagonal element of D. ~ J. then Eq. 1. (17) yields (21) the n x n identity matrix.) . It therefore follows that eA = I+A+. then eA+B = eA e8 . If t is a scalar variable. 2! n! 2 t2 n tn (24) (Of course. In Problem 31 we ask you to show that a useful law of exponents holds for n x n matrices that commute: If AB = BA. For instance.+···+A .:.412 Chapter 5 Linear Systems of Differential Equations for each integer n . if 0 is the n x n zero matrix. The exponential of the n x n diagonal matrix D= is the n x n diagonal matrix 6 [.+ ···. • The n x n analog of the 2 x 2 result in Example 2 is established in the same way. then substitution of At for A in Eq. so the exponential of the diagonal2 x 2 matrix A is obtained simply by exponentiating each diagonal element of A. the matrix eA is nonsingular for every n x n matrix A (reminiscent of the fact that ez =I= 0 for all z). It follows from elementary linear algebra that the column vectors of eA are always linearly independent.
0 thus • . (20) and (22) give 3t 1 4t + 9t2 ] 6t 1 . 1 0 OJ [0 0 1 0 + 0 [0 0 0 1 1 3t = [ 0 1 4t eAt + 9t 2 6t 1 J . Such a matrixwith a vanishing poweris said to be nilpotent. It therefore follows from Eq.7 Matrix Exponentials and Linear Systems 413 Example 3 If A=[~ then A2 so An 3 0 0 n A3 = [0 0 18] 0 0 0 0 0 0 and = [0 0 0 0 0 0 = 0 for n ~ 3.5. • Example 4 If A= then [2 3 0 2 0 0 ~l 0 0 A=U 3 4J 0 2 OJ 0 + [0 0 0 6 =D+B 0 2 0 where D = 21 is a diagonal matrix and B is the nilpotent matrix of Example 3. 0 0 Remark: If An = 0 for some positive integer n. so the exponential matrix eA (or eAt) is readily calculated as in Example 3. then the exponential series in (24) terminates after a finite number of terms. Therefore. (24) that n • = that is.
Example 5 In Example 1 we found that the system x' = Ax with . is given by x(O) = xo (26) (27) and this solution is unique. Conversely. In particular. it follows that the matrix exponential eAt is a fundamental matrix for the linear system x' = Ax. if we already know a fundamental matrix <P(t) for the linear system x' = Ax. (4')) and eA·O = e0 = I (the identity matrix) yield (28) So we can find the matrix exponential eAt by solving the linear system x' = Ax. then the solution of the initial value problem x' =Ax. it is the fundamental matrix X(t) such that X(O) =I. dt (25) in analogy to the formula D 1 (ekt) = kekt from elementary calculus. Because the matrix eAr is nonsingular. THEOREM 2 Matrix Exponential Solutions If A is an n x n matrix. Thus the matrixvalued function X(t) =eAt satisfies the matrix differential equation X' = AX.414 Chapter 5 Linear Systems of Differential Equations Matrix Exponential Solutions It happens that termbyterm differentiation of the series in (24) is valid. Theorem 1 implies the following result. that is. Therefore. with the result d t2 (e At ) = A+ A2 t +A3 dt 2! + ··· t2 = A ( I+ At +A2 2! + ··· ) · . d _(eAr)= AeAt. Thus the solution of homogeneous linear systems reduces to the task of computing exponential matrices. then the facts that eAt = <P(t)C (by Eq.
2. Hence.5. 2.7 Matrix Exponentials and Linear Systems has fundamental matrix 21 [ c)(t) = 3:21 Hence Eq. using Theorem 2. the solution of the initial value problem in (29) is given by 0 (19 + 243t [ (29 + 234t)e 21 + 351t2)e21 ] • 39e21 • . (28) gives with 415 • Example 6 Use an exponential matrix to solve the initial value problem x' = 2 3 4] [0 0 2 0 2 6 x. and so we do not yet have the three linearly independent solutions needed for a fundamental matrix. (29) Solution The coefficient matrix A in (29) evidently has characteristic equation (2. x(O)= un r. But we note that A is the same matrix whose matrix exponential 0 was calculated in Example 4.A) 3 = 0 and thus the triple eigenvalue A = 2. It is easy to see that the eigenvector equation Thus there has (to within a constant multiple) the single solution v = [ 1 0 0 is only a single eigenvector associated with the eigenvalue A = 2.
Then one wouldusing Eqs. In this way. an explicit computation similar to that in Eq. .6assemble the linearly independent solutions of the differential equation x' = Ax in (29).) Thus the matrix A. (31).21 is nilpotent: (A . (31) (This particular result is a special case of the Cay leyHamilton theorem of advanced linear algebra.21) 3 = 3 4] [ 00 0 OJ 0 0 [0 0 0 0 0 0 0 0 0 6 = 0.416 Chapter 5 Linear Systems of Differential Equations Remark: The same particular solution x(t) as in Example 6 could be found using the generalized eigenvector method of Section 5. For such a matrix. c 3 so that the particular solution x(t) = c 1x 1 (t) + c2 x2 (t) + c3 x 3 (t) satisfies the initial condition in (29).6. (30) A similar result holds for any 3 x 3 matrix A having a triple eigenvalue r. The final step would be to determine values of the coefficients c 1 . General Matrix Exponentials The relatively simple calculation of eAt carried out in Example 4 (and used in Example 6) was based on the observation that if A= 3 4] [2 0 0 2 0 2 6 3 . then A .r) 3 = 0. (30) will show that (A. according to which every matrix satisfies its own characteristic equation. One would start by finding the chain of generalized eigenvectors corresponding to the triple eigenvalue A = 2 of the matrix A.rl is nilpotent.rl) 3 = 0. in which case its characteristic equation reduces to (A . (27) in Section 5. and it follows that the exponential series here terminating because of Eq. c2 . At this point it should be apparent thatespecially if the matrix exponential eAt is readily available (for instance. from a computer algebra system)the method illustrated in Example 6 can well be more "computationally routine" than the generalized eigenvector method. we can rather easily calculate the matrix exponential e At for any square matrix having only a single eigenvalue.
then the n x n matrix <l>(t) = [ x. A has n linearly independent generalized eigenvectors u 1.1 satisfies the initial condition X(O) = I. and r: eAt u. 2! (34) using (33) and the fact that eu1 = eM I.) Even if we do not yet know eAt explicitly.1)! J. As we saw in Section 5. We have therefore outlined a proof of the following theorem..l)ut +(A. THEOREM 3 Computation of eAt Let u 1 . .(t) x2(t) · · · Xn(t) J (35) is a fundamental matrix for the system x' = Ax.. .6. of A and has a rank r ~ 1 such that (AUYu =0 but (33) (If r = 1. u 2 .... Un. (32) motivates a method of calculating eAt for any n x n matrix A whatsoever. . we can calculate x explicitly in terms of A. u. .'u (r. . then u is an ordinary eigenvector such that Au= 'Au. the specific fundamental matrix X(t) = <l>(t)ci>(0).. we can consider the function x(t) = which is a linear combination of the column vectors of eAt and is therefore a solution of the linear system x' = Ax with x(O) = u. Indeed. If the linearly independent solutions x 1(t). Un.5. Un be n linearly independent generalized eigenvectors of the n x n matrix A . let xi (t) be the solution of x' = Ax given by (34). For each i. Each generalized eigenvector u is associated with an eigenvalue A. Xn(t) ofx' =Ax are calculated using (34) with the linearly independent generalized eigenvectors u 1. x2(t). and rank r of the generalized eigenvector ui. substituting u = ui and the associated eigenvalue A. If the fundamental matrix <l>(t) is defined by (35). u2 .7 Matrix Exponentials and Linear Systems 417 The calculation in Eq.. and thus is the desired matrix exponential eAt.A. . Finally. u2.AI) 2 u+ ··· tr1 2 + (Auy.. so x(t) = e At [ u t +(A. . .. then (36) Example 7 Find e At if (37) . A. 1 ~ i ~ n.
With a = 0 we get the generalized eigenvector u3 = (0 2 .3I)'u = u~ I~] n n [ The first two equations 8b + 16c = 0 and 4b + 8c = 0 are satisfied by b = 2 and c = . (33). The eigenvector equation (A.418 Chapter 5 Linear Systems of Differential Equations Solution Theorem 3 would apply even if the matrix A were not upper triangular.1 of rank r = 2 associated with the eigenvalue A = 3.2a + 4b = 1 is satisfied by a = 2 and b = 1. Eq. but leave a arbitrary.31) 2 u = 0.3I)u =[ ~ ~ ~ ][ n ~l =[ = [ = 0 for u = (a b c r The first two equations 4b + 5c = 0 and 2b + 4c = 0 imply that b = c = 0. Thus A has the distinct eigenvalue AJ CASE = 5 and the repeated eigenvalue A2 = 3. The eigenvector equation (A. the fundamental matrix .Al)u = 0 for u = (a is r 2 (A . but leave a arbitrary.Al)u (A .1. (34) yields the third solution r With the solutions listed in Eqs. But because A is upper triangular. (38) CASE 2: A2 lS = 3.A) 2 = 0. Because (A. this fact enables us to see quickly that its characteristic equation is (5. b c 1: AJ = 5. we consider the equation (A.51)u = [ ~ The last two scalar equations 4c = 0 and 2c = 0 give c = 0. Thus the eigenvalue A1 = 5 The corresponding solution of has the (ordinary) eigenvector UJ = [ 2 1 0 the system x' = Ax is r. Then the first equation . The corresponding solution of the system x' = Ax is (39) To look for a generalized eigenvector of rank r = 2 in Eq. (39) and (40).A)(3. Thus the eigenvalue A2 = 3 has the single (ordinary) eigenvector u2 = [ 1 0 0 ] T.
x.5. 4x 2 . x~ = 4x.6xz. x. x~ = 2xt + 4xz x. J . 1. x' 2. 20. x' 3. = 13xt + 4xz. 14. x~ = 6x. 13. x. x(O) = [ j] = Ax Each coefficient matrix A in Problems 25 through 30 is the sum of a nilpotent matrix and a multiple of the identity matrix. x' = [ 5.15xz. A= u Jx. x' 27. x~ = 5x 1 . x . = 6x 1 . 19. (35) is 1 419 with 2 0 1 ~]. show that the matrix A is nilpotent and then use this fact to find (as in Example 3) the matrix exponential eAt. .4x2 . Theorem 3 suffices for the computation of eAt provided that a basis consisting of generalized eigenvectors of A can be found.8xz x. Jx. Jx. 15.x 2 4x. x(O) = [ ~J 7. 17. = 9x 1 + 2xz. Use this fact (as in Example 6) to solve the given initial value problem.X~=x r +3xz = 4xt + 2xz. . Hence Theorem 3 finally yields eAt = c)(t)c)(O)I [2e est ' 0 5 e3t 0 0 3te 2e3' t] e3t 3 [ 0 1 2 0 0  1 1 ~] = [ e'' 0 0 2eSt . . x~ = 2x 1 . x~ = 6x 1 . = [ 1~ = [~J = [ _ 1~ J 9. x~ = 6x. +xz . . 16. x(O) = [ _. e3t • • Remark: As in Example 7. + 7xz x. x(O) 21.7x2 x. x(O) = u] 23. ~ Jx. x.3xz x. x' =[ i . x' = [~ . x' 26. x~ = 12x 1 .2e3t est 4e5' 0 (4 + 3t)e3' ] 2est . = [~J = [ _~ In Problems 21 through 24. Jx. BJ Problems Find a fundamental matrix of each of the systems in Problems 1 through 8. 11.8xz. = [~ x(O) x(O) x(O) iJ J =. 12. 18. A= [ 24. = 10x 1 . 4. = 11x 1 . 25.7 Matrix Exponentials and Linear Systems defined by Eq. = 9x 1 .Sxz x. = 5x 1 2x. . x. X= [ n = = = = x. =[_ = [~J J = [ ~ . Compute the matrix exponential eAt for each system x' given in Problems 9 through 20. ~ Jx. x' = [ i ~ J ~ x. (8) to find a solution satisfying the given initial conditions. then apply Eq. ~ 22. x. x~ = 2x 1 + 6xz x. x' = [ . 10. x~ = 5x 1 .4xz 2x 1 3x.10x2 .3x2 . x(O) x(O) x. A= [ 6.X = u= ! =n ~ ~  8.2e3t . A = u03] 0 0 7 3 ~ _: J x. = 3x.6x 2 .2x2 = u~ n x(O) =[ i] .
the matrix eA is nonsingular with (eArl =eA. that AB = BA. 37. Apply Theorem 3 to calculate the matrix exponential eAt for each of the matrices in Problems 35 through 40. we know from Theorem 4 of Section 5.) 32. (1) has the form x(t) = Xc(t) + Xp(t). x' = [ 12 9 6 3 31. A= = I cosh t + A sinh t. A= [ ~ 33. Apply this fact to find a general solution of x' = Ax. so our task now is to find Xp(t). 34. (Suggestion: Group the terms in the product of the two series on the righthand side to obtain the series on the left. and Xp (t) is a single particular solution of the original nonhomogeneous system in (1). Suppose that A=[~ eAt 6]. Conclude that 39. Prove that eA+B = eAe8 . A= [ ~ ~ l x. Preceding sections have dealt with Xc(t). 29. 35. ~] ! ~ ~] 0 1 and apply this fact to find a general solution of x' = Ax. Each of these may be generalized to nonhomogeneous linear systems. ! . A= 20 10 0 20 5 3 3 2 0 40. such as inflow of liquid to a cascade of brine tanks or an external force acting on a massandspring system. Deduce from the result of Problem 31 that. Suppose that the n x n matrices A and B commute. for every square matrix A. u2 3] u n u 30] u n A=[~ n ~] 3 1 0 3 1 0 0 36. and verify that it is equivalent to the solution found by the eigenvalue method. x' = [ g 0 : 30. 4 2 0 0 4 4 2 0 IIIJ Nonhomogeneous Linear Systems In Section 2.420 28. A= 1 0 4 1 38. A= Show that A 2n = I and that A 2n+I = A if n is a positive integer. nonhomogeneous terms typically correspond to external influences. Suppose that x. (2) where • • Xc(t) = c 1x 1 (t) + c2 xz(t) + · · · + CnXn(t) is a general solution of the associated homogeneous system x' = Ax. Verify that it is equivalent to the general solution found by the eigenvalue method. Given the nonhomogeneous firstorder linear system x'=Ax+f(t) (1) where A is ann x n constant matrix and the "nonhomogeneous term" f(t) is a given continuous vectorvalued function.5 we exhibited two techniques for finding a single particular solution of a single nonhomogeneous nthorder linear differential equationthe method of undetermined coefficients and the method of variation of parameters. In a linear system modeling a physical situation. x' Chapter 5 Linear Systems of Differential Equations =[ 20 1~ ~ 0 30 g] 5 x. . Show that eAr = I cos 2t + ~A sin 2t.3 that a general solution of Eq. that is.
~  •··• • Example 1 Find a particular solution of the nonhomogeneous system (3) Solution The nonhomogeneous term f = [ 3 2t linear trial particular solution of the form f [ is linear. (4) gives the particular solution X = [ Xj X2 of (3) described in scalar form by r Xt(t) = 4t + 17. Substitution of these coefficients in Eq.• •• ··m• • • .•.• ·  •. a2.25. xz (t) = 6t . (5) We solve the first two equations in (5) for a 1 = 4 and a 2 = .A ~ • •. . (3). With these values we can then solve the last two equations in (5) for b1 = 17 and b2 = ...8 Nonhomogeneous Linear Systems 421 Undetermined Coefficients First we suppose that the nonhomogeneous term f(t) in (1) is a linear combination (with constant vector coefficients) of products of polynomials. exponential functions. Moreover. the choice of this general form is essentially the same as in the case of a single equation (discussed in Section 2. 0.. (1). We will therefore confine the present discussion to illustrative examples..6.5).• "''"'"' •• ••· . we modify it only by using undetermined vector coefficients rather than undetermined scalars. Then the method of undetermined coefficients for systems is essentially the same as for a single linear differential equation. then attempt to determine the coefficients in Xp by substitution in Eq.. and sines and cosines.· • ••~···~··. . (4) Upon substitution of x = Xp in Eq.· . we get We equate the coefficients oft and the constant terms (in both x 1. • . We make an intelligent guess as to the general form of a particular solution Xp.• • • • ••.5.· . at.and x 2components) and thereby obtain the equations 3at + 2a2 = 7at + 5a2 + 2 = 3bt + 2b2 + 3 = 7bt + 5b2 = 0. so it is reasonable to select a Xp (t) = at +b = :~ Jt + [ :~ J.25.
4. and c3 = .~0 . Thus our particular solution is Xp (t) = [ 40 80 100 In Example 2 of Section 5.5 [ OJ ~ = ~5 0 0. Then substitution of x = Xp in (6) yields the system [ 0. a 2 = 80. V2 = 40. Referring to Eq.tf2 . Substituting that are readily solved for c 1 = . 0. with no (external) inflow of salt into tanks 2 and 3. we find that the amounts of salt in the three tanks at time t are given by XI (t) = 40. so that 20 pounds of salt flow into tank 1 per minute.2 a3 + [ 20 0] 0 that we readily solve for a 1 = 40. The three brine tanks of Example 2. c2 = .5 0. we get the scalar equations 6c1 Sc1 + 40 = + c2 + 80 = . [~] .Sc2 + c3 + 100 = 0.5 dt 0 0 0.8.114  25e.25 0. Suppose that all three tanks contain fresh water initially.(l) = c 1 [ ~] . 0 00 .4. (7).•I' + c 2[ J] .25 0. The nonhomogeneous term f = [ 20 0 0 JT here corresponds to the 20 lb/min inflow of salt to tank I. and V3 = 50 (gal).2e.160. we see that the vector x(t) = [X[ (t) X2(t) X3(t) of amounts of salt (in pounds) in the three tanks at time t satisfies the nonhomogeneous initial value problem r r!l dx.253 these coefficients in Eq. so a general solution x = nonhomogeneous system in (6) is given by + Xp of the When we apply the zero initial conditions in (6).[ 0.112 (8) + 24e. Because the nonhomogeneous term is constant.422 Chapter 5 Linear Systems of Differential Equations ·. .1. .4 we found the general solution r. and the common flow rate is r = 10 (gal/min). but that the inflow to tank 1 is brine containing 2 pounds of salt per gallon.8.2 ~ ] X+ [ 2~]' 0 = x(O) = [H (6) lr FIGURE 5. we naturally select a constant trial function Xp = [ a 1 a2 a 3 JT. X2(t) = 80 + 80e.1 shows the system of three brine tanks investigated in Example 2 of Section 5.25 0 ] [ a2 a1 ] 0 0.160et f4 . The volumes of the three tanks are V1 = 20. and a 3 = 100 in turn.•1 ' Xc of the associated homogeneous system.•/4 + c.• · ·~ ~·• • •• v ~  ·"" •• •~·· Example 2 ~ r (gal/min) ·< Tank 1 ~t_S!a2 ~ Figure 5. x.tl 5 ). for which x~ 0.25 0. (18) in Section 5.40et/2 . x3(t) = 100 + 1 ~ ( .
5 that the method of variation of parameters may be applied to a linear differential equation with variable coefficients and is not restricted to nonhomogeneous terms involving only polynomials. as t + +oo.21 + be2t exhibits duplication with the complementary function in (10).8 Nonhomogeneous Linear Systems 423 As illustrated in Fig. For a system.5). (9) In Example 1 of Section 5. but also by all lower (nonnegative integral) powers of t as well. In the case of duplicate expressions in the complementary function and the nonhomogeneous terms. and we would then have six scalar coefficients to determine.21 as our trial solution. • Variation of Parameters Recall from Section 2.5.4 we found the solution Xc(t) = CJ [ _ 1 ] 2t 3 e + c2 [ 2 ] e 5t 1 (10) of the associated homogeneous system. and all the resulting terms must be included in the trial solution.21 + ce. the usual first choice for a trial solution must be multiplied not only by the smallest integral power oft that will eliminate duplication. there is one difference between the method of undetermined coefficients for systems and for single equations (Rule 2 in Section 2.8. It is simpler to use the method of variation of parameters. We would therefore select Xp(t) = at 2 e.2. a uniform density of 2lbjgalthe same as the salt density in the inflow to tank 1. our next topic. The method of variation of parameters for systems enjoys the same flexibility and has a concise matrix formulation that is convenient for both practical and theoretical purposes.21 + bte. • 120 100 80 ~ x 3 (t) ~ 100 80 x 2(t) ~ 60 40 20 00 10 x 1(t) ~ 40 20 30 40 50 60 FIGURE 5. Example 3 Consider the nonhomogeneous system X1 = [ 4 2 3 l J [ 15 te 4J X 2t . and sinusoidal functions.2. exponentials. 5. The salt amount solution curves defined in (8). we see the salt in each of the three tanks approaching. .8. A preliminary trial solution Xp (t) = ate.
(15) and (16) in (11) yields • ' (t)u(t) + •(t)u' (t) = P(t). indeed. . (17) reduces to •(t)u'(t) = f(t). (13). satisfy Eq.... so that (21) Upon substitution of (21) in (15). . (17) But (18) because each column vector of •(t) satisfies Eq. Therefore. Eq. Thus we seek a particular solution of the form Xp(t) = •(t)u(t). The derivative of xp(t) is (by the product rule) x~(t) = c)' (t)u(t) + •(t)u' (t). (15) We must determine u(t) so that Xp does.. Xn to rewrite the complementary function in ( 12) as (14) where c denotes the column vector whose entries are the coefficients c 1 . as stated in the following theorem. we finally obtain the desired particular solution. Our idea is to replace the vector "parameter" c with a variable vector u(t). x2. c2 . . (19) Thus it suffices to choose u(t) so that (20) that is. en. given that we have already found a general solution (11) (12) of the associated homogeneous system x' = P(t)x.424 Chapter 5 Linear Systems of Differential Equations We want to find a particular solution Xp of the nonhomogeneous linear system x = P(t)x + f(t).(t)u(t) + f(t). (13) We first use the fundamental matrix •(t) with column vectors x 1. . . ( 11 ). (16) Hence substitution of Eqs.
we get the solution (25) of the nonhomogeneous initial value problem x' = P(t)x + f(t). for we need only a single particular solution. x(a) = Xa.8 Nonhomogeneous Linear Systems 425 THEOREM 1 Variation of Parameters If c)(t) is a fundamental matrix for the homogeneous system x' = P(t)x on some interval where P(t) and f(t) are continuous. then a particular solution of the nonhomogeneous system x' = P(t)x + f(t) is given by (22) This is the variation of parameters formula for firstorder linear systems.1 = e . (26) Equations (22) and (25) hold for any fundamental matrix 4>(t) of the homogeneous system x' = P(t)x. In the constantcoefficient case P(t) = A we can use for 4>(t) the exponential matrix eA1. In solving initial value problems it often is convenient to choose the constant of integration so that Xp (a) = 0.5. If we add this particular solution and the complementary function in (14 ). (22) is immaterial. x(a) = Xa. x(a) =0 in (24) to the solution X c(t) = 4>(t)4>(a).At.1f(t) dt (23) of the nonhomogeneous system in (11). The choice of the constant of integration in Eq. substitution of 4>(t) = eAt in (22) yields the particular solution (27) . Then. because (eAt) .1xa of the associated homogeneous problem x' = P(t)x. and thus integrate from a tot: (24) If we add the particular solution of the nonhomogeneous problem x' = P(t)x + f(t). we get the general solution x(t) = 4>(t)c + 4>(t) J 4>(t). the particular fundamental matrix such that 4>(0) = I.that is.
5s 4se2s Jd s 3 1 J+ 1 [ . (29) Remark: If we retain t as the independent variable but use s for the variable of integration.2 7t 2 +4e .2e2s + 2e5s ] [ . (28) in Section 5. substitution of 4>(t) eAt in Eq.2r e .21 .71 + 28te.5s .s 3s  14se=7s 7se ?s J ds • = [7] + [4+ 3 14 .I = [ _ 3e .15se. x(O) = xo.71 . 1 7 3 .3e2s + 6e.5s + 3e.2 1] [1 Then the variation of parameters formula in Eq. x(O) = [ . (25) with a = 0 yields the solution = (28) of the initial value problem x' = P(t)x + f(t).426 Chapter 5 Linear Systems of Differential Equations of the nonhomogeneous system x' = P(t)x + f(t). (28) gives = [ 3 = [ 7 7 J+ Jo (7 0 1 I [ e2s .2s 6e2s + e . then the solutions in (27) and (28) can be rewritten in the forms ·m ~~ ~ ' ~v ~ ~ · •• • • Example 4 Solve the initial value problem X' = [4 3 _2 1 J [ 15 4 Jte X . It gives the fundamental matrix It follows by Eq.21 2e 51 e5t ] . (1 0).7 that the matrix exponential for the coefficient matrix A in (30) is e At = 4>(t)4>(0) .71 ] 2 1t 2 + 2e.71 + 14te. ] . (30) Solution The solution of the associated homogeneous system is displayed in Eq. Similarly.
8 Nonhomogeneous Linear Systems Therefore. At e 1 [ 427 x(t) = 14 94.5. x~ = Qx 1 .7t + 28te7t] + 2e7 r + 14te7 t · Upon multiplication of the righthand side here by eAt. y II = x 1 = x2 e equation m (31) 1s equivalent to the linear system x~ = x 2 .1fdt in (22) yields = [Y1 y~] Yi Y2 f ~ [y f] 2 dt. we find that the solution of the initial value problem in (30) is given by X t ( )  . X1 = P(t)x + f(t).Px2 + f(t). let us investigate how the variation of parameters formula in (22) "reconciles" with the variation of parameters formula in Theorem 1 of Section 2.! w Therefore the variation of parameters formula Xp = f{) J • . W Y1f .7t 2 40 + 21t 2 + 4e.7t 2 40 + 21t 2 + 4e7t + 28te71 J + 2e7t + 14te7t • In conclusion. that is. t h en th e smg = x 2 . y I = x 1 I II I • 1 • • • If we wnte . so the inverse fundamental matrix is <) .5 for the secondorder linear differential equation y" + Py + Qy = 1 f(t). (31) • y = x 1. (32) where Now two linearly independent solutions y 1 and y 2 of the homogeneous system y 11 + Py 1 + Qy = 0 associated with (31) provide two linearly independent solutions x1 = [ ~~ ] and x2 = [ ~~ ] of the homogeneous system x1 = P(t)x associated with (32). [ 7 _ e2t 3e2r + 6 est + 3esr 2e2t + 2e5t] 6e2t + e 5 t · 1 [ 14 94. Observe that the determinant of the fundamental matrix f{) = [ x 1 x2] is simply the Wronskian W = Y1 Yi = _1 Y2 Y~ of the solutions y 1 and y2 .!.
x' = 2x + 4y + 2. X 1 = X + y + 2t. l and x(l) = t 1 .Sy + COS 2t. W If. Primes denote derivatives with respect tot. y(O) = 1 12. A= [ ~ 4 =.t f(t) = [ ~~:~: x(O) = [~ l eAt=.!. BJ Problems 18. A= [ Problems 15 and 16 are similar to Example 2. 5. x(O) = [ ~ In each problem we provide the matrix exponential eAr as provided by a computer algebra system. 23.2y. x' = 3x +4y +sint.e1 . but with two brine tanks (having volumes V1 and V2 gallons as in Fig. x' = x . (33) of Section 2. X 1 = X . finally. ] . y' = 2x + y . but with f(t) = In Problems 17 through 34. x' = 9x + y + 2e 1 .5 (where. VI = 100.2 2. Co = 3 e At = [I + ~ =~ 9t 3t l f(t) = [~ l l x(O) = [ .S sin t. y' = 8x . but with f(t) replaced with [ 21. Repeat Problem 21. l 27.2t 13. Repeat Problem 23. i =~ t l f(t) 2 = [ 3~: ].3 cost 9. y' = 2x .y 10.2e. x' = 2x + y + 2e'. VI = 200. but with f(t) = [ [~ l 25. however. we supply the independent variable t throughout. but with f(t) replaced with [ Apply the method of undetermined coefficients to find a particular solution of each of the systems in Problems 1 through 14. r = 10. eAz = [ 1 + 2t 17. Co = 2 16. x' = 4x + y + e1 .1 6.3e1 14. y' = x + 2y . and the inflow to tank 1 at the rate of r gallons per minute has a salt concentration of co pounds per gallon. y' = x .3t J t~2 ] 24. (b) Find the limiting (longterm) amount of salt in each tank. 15.428 Chapter 5 Linear Systems of Differential Equations The first component of this column vector is Yp = [ YI Y2 ]f .2t 3. A = [ = Xa. Repeat Problem 17.~:~r J. find the particular solution that satisfies these conditions. x(O) = y(O) = 0 4. y' = X . x' = x.2y + te 1 7. l er se.7y + 10. use the method of variation of parameters (and perhaps a computer algebra system) to solve the initial value problem x' =Ax+ f(t).y + e 1 sin t 11. If initial conditions are given. x' = 2x + y + 1. the final result on the righthand side here is simply the variation of parameters formula in Eq. V2 = 100. (c) Find how long it takes for each tank to reach a salt concentration of 1 lb j gal.Sy + 2 sin t. y' = 2x + y.y . x' = 6x . but with f(t) replaced with [ . y' = 4x + 2y + e 41 ~~~ 19. x' = 3x + 4y. y' = 3x + 2y + t 2 . A = [ i =. x' = x + 2y + 3. r = 10.2 Sill t 26.1 [ . At e ~ l f(t) = [ 18g~ l l x(O) = [ ~ l 7 Se~ l l = ' [ 5 e3z _ 2e3r + 4e2r + 2e2r 2e3t + 2e2r ] 4e3r + e2r 20. A= [ ~ =~ l f(t) = [ ~~ l x(O) = [ ~ l 4t ] 1 . Each tank initially contains fresh water. the independent variable is denoted by x ). x(a) [~~~~:]and x(O) = [. but with f(t) = [ 4t~n/ Jand x(l) = [~l . Repeat Problem 19. y' = x + 2y + 3. A= [. x (0) = y(O) = 1 5. x' = 2x + 3y + S.Sy. (a) Find the amounts x 1 (t) and x2 (t) of salt in the two tanks after t minutes. 1.2) instead of three tanks. Repeat Problem 2S. y(O) = 0 8. x(O) = 1. y' = X + y . y' = 6x. [ + Se3r + Se3t et _ e3t ] set . y' = x . Repeat Problem 27.Y2 f W Y1f ] dt = yi f Y2 d f t W + Y2 f YId f t. y' = 6x .8. l f(t) = [ 4~ x(O) = [~ l l eAt = [ cos t ~ 2 sin t Sill t COS t . x(O) = 1. v2= 200.2t 28.e3t 22.2y.y .
x(O) = [OJ O . e 21 0 32. t . A= [ 0 1 1] O . f(t) = [ ~ 2 ~~ ] . 0 1 et 2tet et (3t = [ 0 0 0 0 1 0 32t 2 + 8t 3 8t + 6t 2 4t 1 ] + 2t 2)et ] 2tet et [ 0 34.5. A = [ = 0 eAt ~ i ~].sin t cost J 0 0 0 4 8 3 0 0 4t 1 8t 30. et eAt = [ 0 3te1 e' _ [ cos e At . f(t) = [ ~ ~~: ~: c~s2t sm2t . A 0 1 =[ 0 ~ 0 ~ ~]. x(O) = [ ~]. x(O) = [ 0 ~].sin2t cos 2t l x(O) = [ ~ l [ 0 33. x(O) = [ 0 ~]. A= [ ~ ~]. 1 4t At e + e2t) 0 =[ Q 0 0 .8 Nonhomogeneous Linear Systems 29. f(t) smt 429 = [sect] O . A= 0 0 1 At e Q 0 0 0 eAt= [ J 6et + 6t 2 3t 31. 2 4( 1 0 e2t f(t) = [ 6~]. A = [ gg~ 0 0 0 1 0 0 !]. f(t) = [ ~ ] .
consider the simple equation =e dy dx .Numerical Methods IIIIJ _N~~!:!_~cal A. y ). y 0 ) and attempts to thread its way through the slope field of a given differential equation y ' = f (x . Y2 ).x2 • But it is known that every antiderivative of f (x) = ex2 is a nonelementary functionone that cannot be expressed as a finite combination of the familiar functions of elementary calculus. Any attempt to use the symbolic techniques of Chapter 1 to find a simple explicit formula for a solution of ( 1) is therefore doomed to failure. y!). YI). and now moves a tiny distance along the slope segment through this new starting point (xi.x2 (1) A solution of Eq.PP_~~~~_mation: Euler's Method It is the exception rather than the rule when a differential equation of the general form dy dx = f(x. • At (xi. As a possible alternative. (1) is finitely expressible in terms of elementary functions. y) can be solved exactly and explicitly by elementary methods like those discussed in Chapter 1. Yo) and moves a tiny distance along the slope segment though (x0 . (1) is simply an antiderivative of e. Hence no particular solution of Eq. 430 . • The plotter pen starts at the initial point (xo. This takes it to the point (xi . y 1) the pen changes direction. an oldfashioned computer plotterone that uses an ink pen to draw curves mechanicallycan be programmed to draw a solution curve that starts at the initial point (x0 . The procedure the plotter carries out can be described as follows. For example. This takes it to the next starting point (x2. y 0 ).
Hence a horizontal change of h from X11 to Xn+I corresponds to a vertical change of m · h = h · f(xn. Yn+l = Yn + h · j(Xrz. Yn) from Yn to Yn+l. Y2). Then the resulting polygonal curve looks like a smooth. y3) .. Yn) ism = j(x11 . suppose that each "tiny distance" the pen travels along a slope segmentbefore the midcourse correction that sends it along a fresh new slope segmentis so very small that the naked eye cannot distinguish the individual line segments constituting the polygonal curve. y 2 ). 6. yi). However. Yn+1) are given in terms of the old coordinates by Xn+l = Xn FIGURE 6. Yn). x 3. y Solution curve FIGURE 6. Leonhard Eulerthe great 18thcentury mathematician for whom so many mathematical concepts.. we ordinarily do not sketch the corresponding polygonal approximation. and his idea was to do all this numerically rather than graphically. In order to approximate the solution of the initial value problem dy dx = f(x.. y(x3). Y2 · y3 . y0 ) and applying the formulas x. y 1). (2) we first choose a fixed (horizontal) step size h to use in making each step from one point to the next. . this is (in essence) how most of the solution curves shown in the figures of Chapter 1 were computer generated. y (x2 ). Xn. Indeed. . y3 ).2. · · · .1... .. the numerical result of applying Euler's method is the sequence of approximations Y1.. . on an approximate solution curve. However. . Instead.1. y). Given the initial value problem in (2). (xJ.. Yn) to (Xn+l. (x2. . x 2. of the exact (though unknown) solution y (x ) of the initial value problem.. The step from (Xn.. Yo) Y1 + h · f(x. The slope of the direction segment through (x11 . Y2) the pen changes direction again. . (x3. Then the step from (xn.1 Numerical Approximation: Euler's Method 431 • At (x2. methods. Y2) . y0 ) and after n steps have reached the point (xn. (x3..6. Figure 6. Yn) to the next point (xn+l. at the points x 1 . Yn).. Yn+l ) . y3). y(xo) =Yo. Euler's method with step size h consists of starting with the initial point (x 0 . continuously turning solution curve of the differential equation. Y2) to calculate successive points (x 1. Yo). This takes it to the next starting point (x3.. Yn). These results typically are presented in the form of a table of approximate values of the desired solution. formulas. Yn+J) is illustrated in Fig. Suppose we've started at the initial point (x 0 . In this figure we see a polygonal curve consisting of line segments that connect the successive points (xo.1. and results are nameddid not have a computer plotter. Yn. = xo X3 = X2 +h +h Y1 = Yo Y2 = Y3 = X2 = XJ +h + h · f (xo. . and now moves a tiny distance along the slope segment through (x2 . The first few steps in approximating a solution curve. · · · to the true values y(xi). y (x n). . (x2.1 . Therefore the coordinates of the new point (xn+ 1 .1 illustrates the result of continuing in this fashionby a sequence of discrete straightline steps from one starting point to the next.2.1.. Yt) Y2 + h · j(x2. + h.
x3.. Although the most important applications of Euler's method are to nonlinear equations.5.7430 Figure 6. 1].. We see that decreasing the step size increases the accuracy.6 2 3.3 shows the graph of this approximation.32 3 1..32)] = 1. and x 5 = 5. _ _. y(x2).6912 5 4.. . .. . • .. we first illustrate the method with a simple initial value problem whose exact solution is available. Y2 + h · [x2 + ~Y2J = (3. • • ·• ·• • u . Yn) (n ::= 0) (3) to calculate successive approximations y 1 .. and h = 1 the iterative formula in (3) yields the approximate values Y1 =Yo+ h · [xo +~Yo]= (3) Y2 = Y3 = Y4 = Ys = + (1)[0 + ~(3)] = 3._w • Example 1 Apply Euler's method to approximate the solution of the initial value problem dy 1 dx =X+ sY· y(O) = 3. Y3 + h · [x3 + ~y3] = (1. dx y(xo) =Yo.32) + (1)[2 + ~(3.5x . . x 2 = 2.7430 and at the points x 1 = 1. f( x ..984)] = 0. • • • . y 0 3 1 3. Note how the result of each calculation feeds into the next one. together with the graphs of the Euler approximations obtained with step sizes h = 0. Solution (a) With x 0 = 0. The resulting table of approximate values is X Approx. x 4 = 4. y(x3).= f(x. 5]. (4) (a) first with step size h = 1 on the interval [0.6912) + (1)[4 + ~(0. the accuracy decreases with distance from the initial point. .3.05. y2.6. y0 = 3. Yl + h · [X J + ~yJ] = (3.2 on the interval [0. just for the purpose of comparison of approximate and actual solutions.• • • _ _.6192.. .984 4 0. of the [exact] solution y = y(x) at the points x 1. as well as the graph of the exact solution y(x) = 22exfS ...25 that is readily found using the linearequation technique of Section 1. (b) then with step size h = 0.432 Chapter 6 Numerical Methods ALGORITHM The Euler Method Given the initial value problem dy . x2. y). y3.984. respectively.32.. "~'"' '"'''''""' '' '' ' •••••• •••• •·· " "•• • · . (2) Euler's method with step size h consists of applying the iterative formula Yn+l = Yn + h · f(xn. x 3 = 3._.6)] = .984) + (1)[3 + ~(1. . " " ' " "• ·• • ' '" '' • • • '" • . . to the [true] values y(xJ).6912)] ~ 4.6) + (1)[1 + ~(3. Y4 + h · [x4 + ~y4] = (0. .. y) = x + ~y . .1. but with any single approximation.2 and 0.. The iterative formula in (3) tells us how to make the typical step from Yn to Yn+l and is the heart of Euler's method.•« • •··' ~ • • •" • • _ _.
0.205) + (0. j(x.234 • High accuracy with Euler's method usually requires a very small step size and hence a larger number of steps than can reasonably be carried out by hand.2 + ~(3.~ ·  •~ · •• Example 2 Suppose the baseball of Example 3 in Section 1.8.2 3.2)[0 + ~(3)] = 3. h = 0.. Graphs of Euler approximations with step sizes h = 1.3 is simply dropped (instead of being thrown downward) from the helicopter. y) = x + ~y. Y5 = Y4 + h · [x4 + ~y4] ~ (3. The application material for this section contains calculator and computer programs for automating Euler's method.2)[0. Y4 = Y3 + h · [x3 + ~y3 ] ~ (3.253)] ~ 3.2. y 0 3 0.2)[0.2)[0. x4 = 0. v0 = 0. v) = 32 .1.12)] ~ 3.4 3.2.6 3.253.1 Numerical Approximation: Euler's Method 433 5 y 0 3 5 0 2 X 3 4 5 FIGURE 6.6 + ~(3. and h = 0. Yz = YI + h · [x1 + ~yd = (3.16v. One of these programs was used to calculate the table entries shown in Fig. F (t .253) + (0. and h = 0.3.002) from x = 0 to x = 1 yield values that are accurate to within 0.2)[0. x 3 table of approximate values is X = 0.263. (b) Starting afresh with xo = 0.12 0. and x 5 = 1.12. Yo = 3.263 1 3.205 0.. (5) We use Euler's method with h = 1 to track the ball's increasing velocity at 1second intervals for the first 10 seconds of fall. Then its velocity v(t) after t seconds satisfies the initial value problem dv dt = 32.16v. The resulting Approx.205.8 3.4. ~ . v(O) = 0.205)] ~ 3. .263)] ~ 3.6.4.2.263) + (0.234 YI =Yo+ h · [xo +~yo]= (3) at the points x 1 = 0. With to = 0. x 2 = 0.1.001.8 + ~(3.4 + ~(3. 6.~ .~ · ' ~•·•~· ·nm.6. We see that 500 Euler steps (with step size h = 0.05..0. Y3 = Yz + h · [xz + ~yz] ~ (3..12) + (0. we get the approximate values + (0.253 0.
= (81.16(32)] = 58.88.195 3.129 FIGURE 6.16(58.0.16(0)] = 32.172 3 .184 3. Actual value of v 32 59 81 100 116 130 141 150 158 165 30 55 77 • 2 3 4 5 6 7 8 9 10 95 111 124 135 145 153 160 30 55 76 95 110 123 135 144 153 160 FIGURE 6.000 3.43.253 3.46)]:::::: 100. h = 0.140 3.168 3.16v3] vs = v4 + h · [32.0.16v4 ] + (1)[32.205 3.1.88) + (1)[32. = (0) Continuing in this fashion. 6.000 3.201 3 .46) + (1)[32.43) + (1)[32. ·Approx y · ··. we complete the h = 1 column of vvalues shown in the table of Fig. The error in the linear approximation formula (6) .263 3.0.43)] ~ 116.8 1 3.1.1.0.0.5where we have rounded off velocity entries to the nearest foot per second.02.88)]:::::: 81. Note also that after 10 seconds the falling ball has attained about 80% of its limiting velocity of 200 ft/s .16(81. with h = 0. Euler approximations in Example 2 with step sizes h = 1 and h = 0.4 0.234 3.16v2 ] v4 = v3 + h · [32.0.2. = (32) + (1)[32 .1.16vo] v2 = VJ + h · [32 .4.196 3. and h = 1 the iterative formula in (3) yields the approximate values vi= vo + h · [32.000 3 .183 3.104 3 .130 Actual value of y 3.0.16(100.46.1 were calculated using a computer. = (58. and we see that they are accurate to within about 1 ft/s.5.102 3.2 0.168 3.16vJ] V3 = V2 + h · (32.191 3.002 0 0.0. and = (100.0.120 3. The values corresponding to h = 0. Euler approximations with step sizes h = 0.6 0.434 Chapter 6 Numerical Methods '· · :.000 3. those for which Xn is not sufficiently close to x 0 .36.002.102 3.0. Local and Cumulative Errors There are several sources of error in Euler's method that may make the approximation Yn to y(xn) unreliable for large values of n. and h = 0.
This error. Thus the smaller the step size.7. Thus the tangent line in Fig. is called the local error in Euler's method.001. rather than merely an approximation to the actual value y(x 11 ). The percentage errors at the final point x = 1 range from 7.1.1 down to only 0. introduced at each step in the process.6 would be the total error in Yn+l if the starting point Yn in (6) were an exact value.1 Numerical Approximation: Euler's Method y 435 I (xn+l• Yn+ .1.6 is tangent to the "wrong" solution curvethe one through (x11 . h = 0. The local error indicated in Fig. 6. The local error in Euler's method. 7 illustrates this cumulative error in Euler's method. Figure 6. Yn) rather than the actual solution curve through the initial point (x0 .1. Yn) departs from the solution curve through (x11 . y ! C"m"l"i" '""' (xn . y(O) = 1.8 we observe that.1.8 shows the results obtained in approximating the exact solution y (x) = 2ex .6. the more slowly does the error grow with increasing distance from the starting point. For instance. so that x 11 is an integral multiple of 0.1. with h = 0. but the value Yn is shown only when n is a multiple of 100. We show computed values only at intervals of L\x = 0. 6. 6. The cumulative error in Euler's method. By scanning the columns in Fig. for each fixed step size h. it is the amount by which the polygonal stepwise path from (xo.) I I I I I I X FIGURE 6. Yo). Yo) departs from the actual solution curve through (x0 .08% with h = 0.1 of the initial value problem dy dx = x + y. But by scanning the rows of the table we see that for each fixed x. 6. .6. using the successively smaller step sizes h = 0. Yn) l Approximate values Xz X FIGURE 6. is the amount by which the tangent line at (x 11 . the computation required 1000 Euler steps.1. the error decreases as the step size h is reduced. and h = 0.005. the error Yactual .1. as illustrated in Fig.6.25% with h = 0.1.1. h = 0. The usual way of attempting to reduce the cumulative error in Euler's method is to decrease the step size h.Yapprox increases as x gets farther from the starting point xo = 0.02. But Yn itself suffers from the accumulated effects of all the local errors introduced at the previous steps. y 11 ).1.x . y0 ). The table in Fig.1.001. 6.001.
1 . .0192 3.5282 1.Ywith y with h =0. the computer itself will contribute roundoff error at each stage because only finitely many significant digits can be used in each calculation.7812 2. Once an appropriate computer program has been written. For example.6493 3. 6.4238 Actual y 1.5807 1.4872 2. whereas if h is too small.0442 2.1.2380 1.··y \fi'fh . because the cumulative effect of roundoff error in the latter case might exceed combined cumulative and roundoff error in the case h = 0.3275 2.9461 2. y) in the initial value problem in (2).4 0. and on the specific computer used.8. In addition to the local and cumulative errors discussed previously. do we not simply choose an exceedingly small step size (such as h = 10.3832 FIGURE 6.3 0.12).1 in Fig.2 0. 1.1 and about 1 min 50s with h = 0. 6.1000 1. An Euler's method computation with h = 0.7 0. the data in Fig.3205 2.5831 1.1102 1.3620 1.7966 2.1082 1.1.3977 1. But 50 steps are required to reach x = 1 with h = 0.0001 will introduce roundoff errors 1000 times as often as one with h = 0. The subject of error propagation in numerical algorithms is treated in numerical analysis courses and textbooks. with the expectation that very great accuracy will result? There are two reasons for not doing so.3993 1.1103 1.3997 1.001.0 1. the approximations inherent in Euler's method may not be sufficiently accurate. . <··· n··· <no!· = h.1098 1.5 0. the computer hardly cares how many steps it is asked to carry out.1.001 1. and 1000 steps with h = 0.:::::moos 1.4230 . then roundoff errors may accumulate to an unacceptable degree or the program may require too much time to be practical. Why.5836 1.12 it would require over 3000 years! The second reason is more subtle.2200 1. on the exact code in which the program is written.2416 1.4266 0. one step size isin principlejust as convenient as another. 200 steps with h = 0.1 might actually produce more accurate results than those obtained with h = 0.2426 1.3917 1. Approximating the solution of dyjdx = x successively smaller step sizes.1.3261 2.0388 2. + y. then.5719 1.005.8159 3.2428 1.6422 3.8 requires only 10 steps.02.1 0.7974 2.436 Chapter 6 Numerical Methods X h ¥:: Oi'l .001.0082 3.0431 2.6511 3.9757 3.0227 2. after all. It depends on the nature of the function f (x .7210 1. But with h = 10.1974 2. The "best" choice of h is difficult to determine in practice as well as in theory. A computer is almost always used to implement Euler's method when more than 10 or 20 steps are required. The first is obvious: the time required for the computation. y(O) = 1 with The column of data for h = 0.6161 2.1875 >' wi~Ji ~"'~ . Thus it required slightly over one second to approximate y(l) with h = 0.7933 2.0170 3.6 0. With a step size that is too large.0001.9 1. so Euler's method can be carried out with a handheld calculator. Hence with certain differential equations. h = 0.8 0.8 were obtained using a handheld calculator that carried out nine Euler steps per second.2998 2.
1. ________ __ ~ ·  Example 4 The exact solution of the initial value problem dy .~ 8 .£ 6n with n = 50. dx y(O) = 1 is the periodic function y (x ) = esin x .8 illustrate the common strategy of applying a numerical algorithm.= y cosx. • . In the next two examples we present graphically the results of successive applications of Euler's method. and 400 subintervals. Approximating a logistic solution using Euler's method with n = 5.£ 5 with n = 5. Euler's method evidently has considerable difficulty keeping up with the oscillations i. 6 4 2 0 X 5 10 X 15 FIGURE 6.:>.£ x .9. Visual comparison of successive results often can provide an "intuitive feel" for their accuracy. .1. Each of these "curves" actually consists of line segments joining successive points (x n.1.. Figure 6... 12. n = 100.1. 200.. n = 10. y(O) = 1 is y(x) = 8/(1 + 7e.£ x .10. Approximating the exact solution y = e sinx using Euler's method with 50. such as Euler's method.1. beginning with a selected number n of subintervals for the first application.y). 6. the more accurate methods discussed in succeeding sections are needed for serious numerical investigations. n = 10. .6. n = 200.n the actual solution. several times in succession. and n = 20 subintervals. Example 3 The exact solution of the logistic initial value problem dy dx I = 3y(8 .8xf3 ). and the approximation with 10 subintervals also overshoots the limiting value y = 8 of the solution before leveling off. and n = 400 subintervals..1 0 shows both the exact solution curve and approximate solution curves obtained by applying Euler's method on the interval 0 .  FIGURE 6... Consequently. and n = 20 subintervals. then doubling n for each succeeding application of the method. Yn +d· The Euler approximation with 5 subintervals is poor. Yn) and (x n+ t.1 Numerical Approximation: Euler's Method 437 The computations in Fig. Even with this many subintervals.9 shows both the exact solution curve and approximate solution curves obtained by applying Euler's method on the interval 0 . but with 20 subintervals we obtain fairly good qualitative agreement with the • actual behavior of the solution. Figure 6. I 00.
Rounded to four decimal places. h = 0.1995 = 2.5078 = 4. so the iterative formula of Euler's method is (8) With step size h = 0. shows that some initial value problems are not so well behaved.5) ~ 1.969811 (approximately).1. Y2 = 1. Solution Here f(x.. 1]. Observe that now the "stability" of the procedure in Example 1 is missing. The table in Fig.1.80.97.~ ·•~·• •• • • • ··••• •• . Indeed. • . • ··•w"''''·' ''' ''• •"•'~. Figure 6. y) = x 2 + y 2 . y(O) = 1 on the interval [0. an exact solution using Bessel functions (see Problem 16 in Section 3.222) 2 ] ~ 1. That is.• • • • •··•.7193 = 3.005.222.·• ·"'" ' " •• ••• Example 5 Use Euler's method to approximate the solution of the initial value problem (7) on the interval [0.6) can be used to show that y(x) + +oo as x + 0. together with a solution curve through (0. Moreover. the actual solution does not exist on the entire interval [0. it seems obvious that something is going wrong near x = 1. Example 5.8371 Y6 Y7 Ys Y9 YIO = 2. · . 6. in accord with the actual values shown in the final column of the table. Euler's method is unable to "keep up" with the rapid changes in y (x) that occur as x approaches the infinite discontinuity near 0.·  ••• ·.• •• • •• •• · ~~·~··"·w.1 Y3 = + (0. 1].~.··· ·· . 1.1895 But instead of naively accepting these results as accurate approximations.2) 2 + (1.1 we obtain Y1 = 1 + (0. 1].40 and y(0.438 Chapter 6 Numerical Methods A Word of Caution The data shown in Fig. Indeed.~.5735 1.1000 1.5 suggest that y(0.3 and x = 0.3753.1) · [(0) 2 + (1) 2 ] = 1.2220 1. 1) plotted using one of the more accurate approximation methods of the following two sections.1..··•. It appears that this solution curve may have a vertical asymptote near x = 0.1) 2 ] = 1.1. in contrast.3753 1.8 indicate that Euler's method works well in approximating the solution of dyjdx = x + y.1) 2 + (1.969811. 6.1. the approximate values in the rows corresponding to x = 0.02. and so forth.11 shows the results obtained with step sizes h = 0. for each fixed x it appears that the approximate values approach the actual value of y(x) as the step size h is decreased.12 provides a graphical clue to the difficulty. and h = 0. we decided to use a computer to repeat the computations with smaller values of h.1) · [(0. It shows a slope field for dyjdx = x 2 + y 2 . Although Euler's method gives values (albeit spurious ones) at x = 1.8023 = 7. the first ten values obtained in this manner are Yl Y2 Y3 Y4 Ys = = = = = 1.222 + (0.1) · [(0. For instance.3) ~ 1.
y(O) = y(x ) = ! e2x = y + 1. ' : : : : :( : : : : 2.. then with step size h = 0...1 Numerical Approximation: Euler's Method 439 0. y 2y' = 2x 5 . y' 2. y' 8. then with step size h = 0. y(O) = 1. or it mayas in Example 5reveal the presence of some hidden difficulty in the problem. together with the percentage error in the more accurate approximation. '. first with step size h = 0. . One should never accept as accurate the results of applying Euler's method with a single fixed step size h.0512 2. y(x ) = ln (x + 1) = ~ (1 + yl )..9601 9..1. y (O) = 1.2. y(x) = 2ex.02. • y(x). 0.8 0.. ' ' '' '. y (1) = 1..5763 12.5078 4. y(2) = 3.01.y. ' '' ''' ' .ex = . . primes denote derivatives with respect to x.x3 = ey.6882 2.1. .5 0.. first with step size h = 0. y' = y./ .97 . y(O) = 2.. '• .0000 30. y(O) = 3. and 0.0 2.004.1000 1. Then apply Euler's method twice to approximate (to four decimal places) this solution on the given interval. y' 4..25. '• '. y(O) = 1. A computer with a printer is required for Problems 17 through 24.1 =X . y(x) = 2ex + x.2512 1.1. .. 2 ~ x ~ 3 2xy 2 'y( 0 ) .005.X . y' 6.8371 2. y' 3.2xy. use Euler's method with step sizes h = 0.1995 2. The moral of Example 5 is that there are pitfalls in the numerical solution of certain initial value problems.. 4 . ' '' '''' .3x 2y .. I X = 0.. FIGURE 6. Compare the threedecimalplace values of the two approximations at x = 4with the value y (!) of the actual solution.2y. y' = y.2.1' . y(x) = 3e. thereby suggesting their accuracy.7 0.6I04 3.x = 2y. y(O) = 1.'· 1 . A programmable calculator or a computer will be useful for Problems II through I6. . Attempting to approximate the solution of dyjdx = x 2 + y 2 . y (O) = 2.:.6658 2.. xy' = 3x .2458 1..0 1.x2 = . y(l) = 3.1. 0. 0 ~ X ~ I y' = 4 (y .7193 3.1895 1. y' 9. ' . 12.0074 2.'I' .11x 2 10• y I  Note: The application following this problem set lists illustrative calculator/computer programs that can be used in the remaining problems. y' 5.6. y(O) = 1. Solution of dyjdx = x 2 + y 2 .2061 1502. Throughout.2220 1. y(x) = 2e.1 0.0008 to approximate to four decimal places the values of the solution at ten equally .1108 1. 13. Certainly it's pointless to attempt to approximate a solution on an interval where it doesn't even exist (or where it is not unique.3612 4. in which case there's no general way to predict which way the numerical approximations will branch at a point of nonuniqueness).12. 11.11.I ) 2 ' y(O) = 2.. I·. y(O) = 1.2090 8 6 4 . . 1 ~ x ~ 2 15.5706 5.. 1 ~ x ~ 2 xy' = y 2. say. Many problems simply require the more accurate and powerful methods that are discussed in the final two sections of this chapter.2 0 .4 0.0 X FIGURE 6. 2 I.9167 1. . for x an integral multiple of 0.4357 1.4243 1. y(x) = 2e. In these initial value problems.' ••.3 0.1..1 = y..2 0.1.8023 7. Make a table showing the approximate values and the actual value. y' 7.' .. Apply Euler's method twice to approximate to this solution on the interval [0. . 1..9 1. 0 ~ X ~ 1 yy' = 2x 3 . '. an initial value problem and its exact solution y(x) are given. 2 ~ x ~ 3 16. A second "run" with smaller step size (h/2. y(x) = 2 + x. y (O) = 0. y(2) = 3. In Problems I through I 0. y(x ) = tan ~(x + rr) !. y(O) = 1.5201 3.1088 1. ' ' ' " ' I . 41. 14. or h/10) may give seemingly consistent results.6 0...3753 1. In each problem find the exact solution of the given initial value problem. or h/5. .5735 1.
. y(2) = 0.13 lists TI85 and BASIC programs implementing Euler's method to approximate the solution of the initial value problem dy dx = x + y. 23.20.006. so your downward velocity satisfies the initial value problem dv dt = 32. in which y(O) = 1.1 Application Construction of a calculator or computer program to implement a numerical algorithm can sharpen one's understanding of the algorithm.y(O)=O.x'/7. 2] to verify empirically that the solution of the initial value problem 25. 0 ~ x ~ 2 2 2. Would you now suspect a discontinuity in the exact solution? 30. Print the results in tabular form with appropriate headings to make it easy to gauge the effect of varying the step size h. y' = . y(O) = 1. (b) Apply Euler's method with step size h = 0. 6. v(O) = 0 (with t in seconds and v in ft/sec).15 to approximate this solution on the interval 1 ~ x ~ 0.1. 20. y(O) =0 dP = 0.5.5. Use Euler's method with a programmable calculator or computer to approximate the solution for 0 ~ t ~ 2. (Contrast this with Example 2. 22. 19.  X ::.0225 P dt 0. Apply Euler's method with successively smaller step sizes on the interval [0. (c) Finally. . Note that. (5). 1 ::. y' =X+ . 0. y' =X+ 4y 2 . Figure 6. . 0 ~X ~ 2 y' =X+~. Now k = 1. 1 + y2' . first with step size h = 1 and then with h = 0.1. The reason is that the numerical approximation "jumps across the discontinuity" to another solution of 7xy' + y = 0 for x > 0.50. Consider the initial value problem spaced points of the given interval. from these data alone.6 in Eq. 0. The general solution of the equation dy dx = (1 + /)COS X Use Euler's method with a computer system to find the desired solution values in Problems 27 and 28. and 1. Use Euler's method to verify this fact empirically. 0 ~ x ~ 1 18.005.35. which has an infinite discontinuity at x = 0. 0 ~ x ~ 2 1 X dy 7x+y=0. y(2) =? 29.90334. apply Euler's method with step sizes h = 0. but still printing results only at the original points x = 1.00. Start with step size is y (x) = tan(C + sinx). y'=sinx+cosy.03 and h = 0. y(O) = 0. Throughout.6v.  1 1 y(x) =. and then use successively smaller step sizes until successive approximate solution values at x = 2 agree rounded off to two decimal places.440 Chapter 6 Numerical Methods h = 0.' (rr/4) ~ 0. y' = x 2 . With the initial condition y(O) = 0 the solution y(x) = tan(sinx) is well behaved. But with y(O) = 1 the solution y(x) =tan {±rr + sinx) has a vertical asymptote at x = sin. 17. What percentage of the limiting population of 75 deer has been attained after 5 years? After 10 years? has a vertical asymptote near x = 2. 21. 27.01 and then with h = 0. first with step size h = 0. 1 ~ x ~ 2 y' = x 213 + y 213 . What percentage of the limiting velocity 20 ft/sec has been attained after 1 second? After 2 seconds? 26.70. rounding off approximate ?values to integral numbers of deer.0003 P 2 (with t in months). you might not suspect any difficulty near x = 0. 0 ~ x ~ 2 y' = lny. y(O) = 1.. y(2) =? 28.O ~ x ~ (a) Solve this problem for the exact solution 24. Suppose the deer population P (t) in a small forest initially numbers 25 and satisfies the logistic equation dy dx = x 2 + l. y' = x + y y(O) = 0.y ( 1) = 1. Use Euler's method with a programmable calculator or computer to approximate the solution for 10 years. y(O) = 1.y 2 . y(O) = 1.1.. rounding off approximate vvalues to one decimal place. You bail out of the helicopter of Example 2 and immediately pull the ripcord of your parachute. The comments provided in the final column should make these programs intelligible even if you have little familiarity with the BASIC and .85.JY. y (O) = 1 considered in this section.) 31. dx y(1)=1.003147. primes denote derivatives with respect to x. y' = x 2 + y 2 .1. y(1) = 2. 1. 1.
N) :X+Y~F :Y+H*F~Y :X+H~X :Disp X. 6.14. % Euler iteration X = X + h.6. % initial x X = y = y. TI calculator programming languages. the final value xl of x.1. the name BASIC is an acronym describing the Beginner's Allpurpose Symbolic Instruction Code introduced in 1963. YP function yp = f(x. and the desired number n of subintervals.) To increase the number of steps (and thereby decrease the step size) you need only change the value ofN specified in the first line ofthe program.1.x)/n. MATLAB implementation of Euler's method. .y.• · BASIC PROGRAM:EULER :10~N 441 Comment Program title Number of steps Initial x Initial y Final x Step size Begin loop Function value Euler iteration New x Display results End loop Program EULER N :o~x :1~Y :1~X1 X y = = = 10 0 1 : (X1X) /N~H :For(I. TI85 and BASIC Euler's method programs. 6.1.Y :End X1 = 1 H = (X1X)/N FOR I=1 TO N F = X + y y = y + H*F X = X + H PRINT X. % new x [X. y ). Some of the modem functional programming languages mirror standard mathematical notation even more closely. The euler function takes as input the initial value x.x1. x.1. you need change only the single line that calculates the function value F.Y) = euler(x.x).13 and subsequent ones in this text) for brief description of mathematical algorithms in a transparent form intermediate between English and higher programming languages. To apply Euler's method to a different equation dyjdx = f(x . Figure 6. % initial y for i = 1:n % begin loop y = y + h*f(x. the initial value y. (Appropriately.13.Y NEXT I FIGURE 6. the BASIC language is no longer widely used for programming computers but is still useful (as in Fig.1.y]. X = % update xcolumn y = [Y. % YP = Y' function [X.y) = X + y.n) % step size h = (x1 .y).1. % update ycolumn end % end loop FIGURE 6.14 shows a MATLAB implementation of Euler's method. intially for instructional use at Dartmouth College. Any other procedural programming language (such as FORTRAN or Pascal) would follow the pattern illustrated by the parallel lines ofTI85 and BASIC code in Fig. Indeed.13 .1 Numerical Approximation: Euler's Method · ~·~· .
10) then generates the Xn. y (xo) = Yo (1) has a unique solution y(x) on the closed interval [a . The number e = y(l). and a careful study of this method yields insights into the workings of more accurate methods. b]. 6. 1.:::. because many of the latter are extensions or refinements of the Euler method. You should begin this project by implementing Euler's method with your own calculator or computer system.1 is not often used in practice.:.:. y(O) = 1. Theorem 1 tells what degree of accuracy we can expect when we use Euler's method. y ) . y(l) = 0. Y] = eu1er(O. f x. where y(x) is the solution of the initial value problem dyjdx = 4/(1 + x 2 ). and assume that y(x) has a continuous second derivative on [a. then to some of the problems for this section. The number JT = y(l).442 Chapter 6 Numerical Methods For instance.1.14159 as specific values of solutions of certain initial value problems. But Euler's method has the advantage of simplicity. ..8. we need some way to measure the accuracy of each. y(O) = 0. ln2.:. THEOREM 1 The Error in the Euler Method Suppose that the initial value problem dy d x = f (x. (This would follow from the assumption that f. 2. 2. How many subintervals are needed to obtaintwice in successionthe correct value of the target number rounded to three decimal places? 1.. mainly because more accurate methods are available. 3. 200. where y(x) is the solution of the initial value problem dyjdx = y.) Then there exists a constant C such that . b].69315. subintervals (doubling n each time). 3.71828.:::. and JT . Test your program by first applying it to the initial value problem in Example 1. b] with a= x 0 .and Yndata shown in the first two columns of the table of Fig. apply Euler's method with n = 50. IIIJ A Closer Look at the Euler Method The Euler method as presented in Section 6. 1. The number ln 2 = y(2). where y(x) is the solution of the initial value problem dyjdx = 1jx . where c ~ y(x) ~ d for all x in [a. Also explain in each problem what the point iswhy the indicated famous number is the expected numerical result. 0. 100. Famous Numbers Investigation The following problems describe the numbers e. In each case. and j y are all continuous for a ~ x ~ b and c ~ y ~ d. the MATLAB command [X.:::. To compare two different methods of numerical approximation.
Birkhoff and G.. .1. Repeat with h/2.1.. 1.C. the error is bounded by a [predetermined] constant C multiplied by the step size h. Because C tends to increase as the maximum value of Iy'' (x) I on [a . In practice. then (2) for each n = 1. it follows that C must depend in a fairly complicated way on y . The constant C deserves some comment. 6. 4th ed. Continue until the results obtained at one stage agree.2 A Closer Look at the Euler Method 443 the following is true: If the approximations y 1 . 1989). . 2.Yn in (2) denotes the [cumulative] error in Euler's method after n steps in the approximation. Then the approximate values obtained at this stage are considered likely to be accurate to the indicated number of significant digits.. we begin with a step size h = 0. 3. with step size hjlO we get 10 times the accuracy (that is. The table in Fig. 6. y(xk) at points of [a. b] are computed using Euler's method with step size h > 0. (New York: John Wiley. Yk to the actual values y(xt). y(x3 ). 1/10 the maximum error) as with step size h. similarly. Indeed..1 shows the approximate values of y(l) obtained with successively smaller values of h. k. exclusive of roundoff error (as though we were using a perfect machine that made no roundoff errors). y(x 2 ). Ordinary Differential Equations.14. y3. perhaps one of those listed in Figs. that (on a given closed interval) halving the step size cuts the maximum error in half.6. Example 1 Carry out this procedure with the initial value problem dy dx 2xy 1 +x 2 ' y(O) = 1 (3) to approximate accurately the value y(l) of the solution at x = 1. 2. The data suggest that the true value of y (l) is exactly 0. Consequently. but one can be found in Chapter 7 of G. Remark: The error Yactual Yapprox = y(Xn). It follows. • We will omit the proof of this theorem. . y 2 . Solution Using an Euler method program. hj4.13 and 6.5. at each stage halving the step size for the next application of Euler's method. Rota. we canin principleget any degree of accuracy we want by choosing h sufficiently small. and actual computation of a value of C such that the inequality in (2) holds is usually impractical.to an appropriate number of significant digitswith those obtained at the previous stage. . . Apply Euler's method to the initial value problem in (1) with a reasonable value of h. The theorem can be summarized by saying that the error in Euler's method is of order h.to reach x = 1. that is.. the exact solution of the initial value problem in (3) is y (x) = 1/(1 + x 2 ) . b] increases. . the following type of procedure is commonly employed. and so forth.04 requiring n = 25 steps. so the true value of y ( 1) is exactly 4.2.. for instance. • . 3.
Figure 6.04 0. (4) x x+h X FIGURE 6. It uses the predicted slope k = f(xn .0003125 0.10 0. True and predicted values in Euler's method. y (x + h.2. · .50000 0.Yapprox 1/h.11 0.50007 0.50003 Actual y(l) 0.Appro~iinat~y(l) 0.000625 0.50000 0.50000 0.444 Chapter 6 Numerical Methods . suppose that after carrying out n steps with step size h we have computed the approximation Yn to the actual value y(xn) of the solution at Xn = x 0 + nh. Yn) of the graph of the solution at the lefthand endpoint of the interval [xn. y(x +h)) The final column of the table in Fig. 1 .50054 0.10 FIGURE 6.2.50000 IErrorl/h 0.005 0. the error bound in (2) appears to hold with a value of C slightly larger than 0.50220 0. yvalue i} Error An Improvement in Euler's Method As Fig.11 0. 6.2.1. Yn) at x = Xn has already been calculated.50013 0. Thus Now that Un+l ~ y(xn+t) has been computed.01 0.1. Why not average these two slopes to obtain a more accurate estimate of the average slope of the solution curve over the entire subinterval [xn.in this computation.1 displays the ratio of the magnitude of the error to h. Given the initial value problem dy dx = f(x.50000 0. . Euler's method is rather unsymmetrical.50109 0.0025 0. y).50000 0. it is known as the improved Euler method. that is.3 shows the geometry behind this method. Xn+tl? This idea is the essence of the improved Euler method.2.. Table of values in Example 1.11 0. We now turn our attention to a way in which increased accuracy can easily be obtained. Xn + h] as if it were the actual slope of the solution over that entire interval.00125 0.50027 0.2 shows.11 0.11 0. / I I I I /Slope y'(x) I I I I I I y(xo) =Yo. Observe how the data in this column substantiate Theorem l . .50451 0. 6. the approximate slope k 1 = f(xn.2. we can take as a second estimate of the slope of the solution curve y = y(x) at x = Xn +I· Of course.50000 0..02 0.50000 0.2. We can use the Euler method to obtain a first estimatewhich we now call Un+l rather than Yn+ 1of the value of the solution at Xn+I = Xn +h.Predicted . IYactual .11 0.
3. y(x2 ). . xz. kz = f(xn+l• Un+I). Remark: The final formula in (5) takes the "Euler form" Yn+l = Yn +h ·k + kz if we write k = kt 2 for the approximate average slope on the interval [xn.2 A Closer Look at the Euler Method y 445 (xn . Un+l = Yn + h · k1. respectively. The improved Euler method: Average the slopes of the tangent lines at (Xn. ••• of the [exact] solution y = y(x) at the points x 1. Yn) I I I I I I I I I n+l X FIGURE 6. to the [true] values y(xJ).. Yn+I = Yn +h · !Ckt +kz) (5) to compute successive approximations Yl· yz. y3. x3. . then it is used to correct itself. y(x 3 ). First a predictor un+I of the next yvalue is computed. Thus the improved Euler method with step ... the improved Euler method with step size h consists in applying the iterative formulas kt = f(xn.. y(xo) =Yo. ALGORITHM The Improved Euler Method dy dx Given the initial value problem  = f(x. Xn+Il· • The improved Euler method is one of a class of numerical techniques known as predictorcorrector methods.6.2. y). Yn). . Yn) and (Xn+l· Un+f).
Answer: Under the assumption that the exact solution y = y(x) of the initial value problem in (4) has a continuous third derivative. to the values y(x 1). as compared with the single function evaluation required for an ordinary Euler step.11) + (0.2. and with step size h /10 we get 100 times the accuracy (that is. U2 = 1.2 + 1. .1 + 1. this means that the improved Euler method is more accurate than Euler's method itself.11. . YI = 1 + (0. 1/100 the maximum error) as with step size h.1 we calculate UJ = 1 + (0. y2. the error in the Euler approximation to y(l) is 7 . but the error in the improved Euler approximation is only 0.11 + (0. The table in Fig.11 + (0. (6) and (7).1 is used.05) · [(0 + 1) + (0. y) = x +yin Eqs. y2. But the factor h2 in (8) means that halving the step size results in 1/4 the maximum error.446 Chapter 6 Numerical Methods size h consists of using the predictor Un+l = Yn + h · j(Xn. When the same step size h = 0.05) · [(0. 6. the predictorcorrector formulas for the improved Euler method are Un+l = Yn + h · (Xn + Yn). and so forth.25%.24%. it can be provedsee Chapter 7 of Birkhoff and Rotathat the error in the improved Euler method is of order h 2 • This means that on a given bounded interval [a. Yn+l = Yn + h · ~ [Cxn + Yn) + (Xn +l + Un+l)] · With step size h = 0.1)] = 1. of the actual solution of the initial value problem in (4).24205.11) = 1..1.x . This advantage is offset by the fact that about twice as many computations are required. y(x3).1 + 1. We naturally wonder whether this doubled computational labor is worth the trouble. . • Example 2 Figure 6. Remark: Each improved Euler step requires two evaluations of the function f (x. b ]. y). each approximate value Yn satisfies the inequality (8) where the constant C does not depend on h.231)] = 1.. Because h 2 is much smaller than h if h itself is small. y(x2).1) · (0.1.231..1. With f(x. y (0) = 1 (9) with exact solution y(x) = 2ex.1) · (0+ 1) = 1.. Yn) (6) and the corrector (7) iteratively to compute successive approximations y 1. Y 2 = 1.4 compares the results obtained using the improved Euler method with those obtained previously using the "unimproved" Euler method.8 shows results of applying Euler's method to the initial value problem dy dx = x + y.1 + 1.
but the former requires only 20 such evaluations.• ' ' •::" ·y ' A .2200 1.6. Example 3 . and thus doubling the number of subintervals of a fixed interval on which we are approximating a solution.1 we applied Euler's method to the logistic initial value problem dy I dx = 3y (8 .6 0. so in this case the improved Euler method yields greater accuracy with only about onetenth the work.1 447 Actual y 1.11034 1.005.3 0.58364 1. Euler and improved Euler approximations to the solution of + y. In Example 3 of Section 6.0 0.24280 1. Accuracy of five significant figures is apparent in the table.39972 1.3997 1.4230 1.4872 2.0192 3. · Euler.2 A Closer Look at the Euler Method lniproved Euler.·].1 is more accurate (in this example) than the original Euler method with h = 0.5 0. The results obtained by successively halving the step size appear in the table in Fig.0388 2.Y 1.7 0.5282 1.1875 1.32751 2.6. The latter requires 200 evaluations of the function f (x.·.79744 2.1100 1.3977 1. in contrast with the original Euler method.1098 1.7949 2.0124 3..65107 3.4 0.6456 3.1974 2.4.1 0.y ). 6.0082 3.7974 2. and that each halving of the step size reduces the error by a factor of almost exactly 4.04424 2.5807 1.4282 dyjdx = x FIGURE 6.9431 2. Note that the final column of this table impressively corroborates the form of the error bound in (8).005.1 0.3205 2.58365 1.7210 1.5.1000 1.79744 2.6422 3.2421 1. of Example 1.> · Appro. • An improved Euler program (similar to the ones listed in the project material for this section) was used to compute approximations to the exact value y(l) = 0.0 1. the improved Euler method with h = 0.5818 1.3 0.2 0.8159 3. the improved Euler method is sufficiently accurate for certain practical applicationssuch as plotting solution curves.3231 2. h =0.2.005.7 0.6 0.4366 Values of y 0.43654 Indeed.5 shows the results obtained when the improved Euler method is applied to the initial value problem in (9) using step size h = 0.1103 1.43656 0.2 0.0442 2.Xini~te X •· •• •. y (O) = 1.0409 2. as should happen if the error is proportional to h2 • In the following two examples we exhibit graphical results obtained by employing this strategy of successively halving the step size.9 1.9 1.65108 3.2428 1.3620 1. > . Improved Euler approximation to the solution of Eq.01919 3. ·<.24281 1.2.8 0.39971 1. Figure 6.• . This suggests that.00000 1.32749 2. (9) with step size h = 0.11034 1.4 0.~~·.2.2. y ).5 0. Improve4: · .01921 3.8 0.7933 2.5 of the solution y (x) = 1/(1 + x 2 ) of the initial value problem dy 2xy dx 1 +x 2 ' y(O) = 1 (3) FIGURE 6.6511 3.00000 1. y(O) = 1.3275 2.2416 1.04423 2.3985 1.5836 1.0 1.
Approximating a logistic solution using the improved Euler method with n = 5. x . In contrast.12 0. The approximation with five subintervals is still badperhaps worse! It appears to level off considerably short of the actual limiting population M = 8. Figure 6.0003125 0.6.7385 7. n = 20.000003117 0.01 0. Improved Euler approximation to y(l) for dyjdx = .2.500195903 0.9 shows an obvious difference between the exact solution y(x) = 8/(1 + 7eSxf3) and the Euler approximation on 0 .9987 7..000625 0. 12 10 8 .12 0.2. four significant digits) on the interval 0 .12 0.000049494 0.:. x .7.:. y(O) = 1.500003117 0. after increasing appropriately during the first step. Figure 6. In the project for this section we ask you to show empirically that the improved Euler approximate solution with step size h = 1 levels off at y ~ 4. and with n = 40 subintervals the exact and approximate solution curves are indistinguishable in Fig. the approximate solution decreases in the second step rather than continuing to increase (as it should).500049494 0.2.00125 0..2.1.2.12 0.02 0.2xy I (1 + x 2 ). and n = 40 subintervals. the approximate solution curve with n = 20 subintervals tracks the exact solution curve rather closely.12 FIGURE 6.500000012 0.005 0.3809 7.000000195 0. 6.:. the improved Euler method (using several hundred subintervals) is considered adequate for many graphical purposes.9999 FIGURE 6.2.. .9999 1.448 Chapter 6 Numerical Methods 1Errorl/h2 0.7.8 indicates that the improved Euler approximation with n = 200 subintervals is accurate rounded to three decimal places (that is.500000049 0.12 0.04 0.000012437 0.000000049 0.6 4 2 00 2 X 3 4 5 FIGURE 6. Because discrepancies in the fourth significant digit are not visually apparent at the resolution of an ordinary computer screen.500000780 0.12 0.000195903 0.9813 7.0000 5.:. The table in Fig.500000195 0.3822 7.7379 7. n = 10.000000012 0.9987 7. • 0 1 2 3 4 5 1.0000 5.8.3542.9812 7.500012437 0. 6.12 0. You should carry out at least the first two improved Euler steps manually to see for yourself how it happens that. 5 using n = 20 subintervals.000000780 0.7 shows approximate solution curves plotted using the improved Euler's method.0025 0. 5. Using the improved Euler method to approximate the actual solution of the initial value problem in Example 3.
2. y ' = 2xy 2 . 6.2. the roundoff error resulting from so many successive steps might well overwhelm the cumulative error predicted by theory (which assumes exact computations in each separate step).2 .2. 6. y(x) = 2ex + x . y' = eY. 3. In each problem find the exact solution of the given initial value problem.y.5] with step size h = 0. 2.X . 0.:. Although Figs. y' = + y 2 ). The approximation obtained with n = 200 is indistinguishable from the exact solution curve.898. Then apply the im . y(O) = 1. then the resulting error E in the final approximation Yn ~ y(l) is given by 2 0. x .5.12)n. 6rr with as many as n = 400 subintervals. y(O) = y(x) = 4e 2x y + 1. and n = 200 subintervals. y(O) = 2. y' y' y' y' 5.:.2. Aside from the possible impracticality of this many steps (using available computational resources). • 0 5 10 X 01i 15 FIGURE 6.:.7 and 6. y' 1.6.. Approximating the exact solution y = esinx using the improved Euler method with n = 50.3x 2 y. 0. Thus. y(O) = 2. 0. 4. consider again the initial value problem dy dx 2xy 1 +x 2 ' y(O) = 1 of Example 1. y(O) = 3. y(O) = 1.. The final column of the table in Fig. y(x) = 2ex.2. A handheld calculator will suffice for Problems 1 through 10. Note: The application following this problem set lists illustrative calculator/computer programs that can be used in Problems 11 through 24. and 200 subintervals. y(O) = 1.9 indicate that the improved Euler method can provide accuracy that suffices for many graphical purposes. y' 7. then 12place accuracy (for instance) in the value y(l) would require that (0.y.4. and the approximation with n = 100 is only barely distinguishable from it.:.1 x . it does not provide the higherprecision numerical accuracy that sometimes is needed for more careful investigations.3.1 0 shows obvious visual differences between the periodic exact solution y(x) = esinx and the Euler approximations on 0. y(O) = 1.3. Such a method is presented in Section 6. 100. 0. y(x) = 2e . 6.12)h = . if the improved Euler method is used on the interval 0 .9. where an initial value problem and its exact solution are given. Consequently.2 A Closer Look at the Euler Method 449 Example 4 In Example 4 of Section 6. 8. roughly half a million steps of length h ~ 0. y(x) = 2 +X .1. For instance. still more accurate methods than the improved Euler method are needed for such highprecision computations.1 y . x . y(x) = 2e. y(O) = 1.6 suggests that.ex 4. 0. n = 100.1. y(O) = 0. y(x) = 1 2 1 x = = = = = . y(x) =tan ±<x +n) = 2xy. 1 with n subintervals and step size h = 1/n.1.9 shows the exact solution curve and approximate solution curves plotted using the improved Euler method with n = 50.Ynl ~ (0.x2 y' = .x 2y.1.13 .. which means that n ~ 489.. 3 2 Figure 6. Apply the improved Euler method to approximate this solution on the interval [0. n If so. Construct a table showing fourdecimalplace values of the approximate solution and actual solution at the points x = 0.000002 would be required.12 E = ly(l) .2 < 5 x 10. y(O) = 1. y(x) = ln(x + 1) 9. y(x) = 3ex 3 iO 10. A programmable calculator or a computer will be useful for Problems 11 through 16.1 we applied Euler's method to the initial value problem dy dx = y cosx. Figure 6.2.
v(O) =0 (with t in seconds and v in ft/s).. x.. 2. y(l) = 3. x. first with step size h = 0. y' =X+ ~y 2 . 0. proved Euler method twice to approximate (to jive decimal places) this solution on the given interval. 22. Throughout.. for x an integral multiple of0. 0 . y (O) = 0 .0.6v. 0.:.:.01 and then with h = 0.:. primes denote derivatives with respect to x. 0 . rounding off approximate vvalues to three decimal places.1.:. explain how you could use the improved Euler method to approximate closely (a) the bolt's time of ascent to its apex (given in Section 1.41 sin the air. 10 using both n = I 00 and n = 200 subintervals. y(O) = 0. .JY. Throughout. its velocity function v(t) satisfies the initial value problem dv . y(2) = 3. Print the results in tabular form with appropriate headings to make it easy to gauge the effect of varying the step size h. y(2) = 3. 10 using both n = 50 and n = 100 subintervals. 2 y' = Iny.8.61 s) and (b) its impact velocity after 9.8 required separate investigations of the bolt's ascent and its descent.= 32. explain how you could use the improved Euler method to approximate closely (a) the bolt's time of ascent to its apex (given in Section 1. It still is shot straight upward from the ground with an initial velocity of 49 m j s. y(O) = 1.8.1 . x . But the improved Euler method requires no such distinction.. y' = x 2 + y 2 . Do the two approximationseach rounded to two decimal placesagree with each other? If an exact solution were unavailable.= (0. Use the improved Euler method with a programmable calculator or computer to approximate the solution for 0 ..2y. y' = y .:...005.:. 0. 0.. Use the improved Euler method with a programmable calculator or computer to approximate the solution for 10 years.1) = 1. As in Problem 26 of Section 6.. 15.. rounding off approximate Pvalues to three decimal places.. Display the results at intervals of I second. but because of air resistance proportional to the square of its velocity. As in Problem 25 of Section 6.:.41 sin the air. 11. x . suppose the deer population P(t) in a small forest initially numbers 25 and satisfies the logistic equation dP = 0. together with the percentage error in the more accurate approximations. and then use successively smaller step sizes until successive approximate solution values at x = 2 agree rounded off to four decimal places. X~ 1 yy' = 2x 3 .:.. v(O) = 49 17.. its velocity function v(t) satisfies the initial value problem dv dt = (0.1. y(2) =? 29.:.:. What percentage of the limiting velocity 20 ft/s has been attained after 1 second? After 2 seconds? 26. y(2) = ? 28. y' = 1 : yZ . you bail out of a helicopter and immediately open your parachute. 3 A computer with a printer is required for Problems 17 through 24.56 s) and (b) its impact velocity after 9. 20. dt v(O) = 49. 1. x . y (0) = 1.0008 to approximate to jive decimal places the values of the solution at ten equally spaced points of the given interval.. y(O) = 0. 18.. y(O) = 1.:.:. with v(t) given by a tangent function during ascent and by a hyperbolic tangent function during descent..y 2 .450 Chapter 6 Numerical Methods with h = 0.:.:. .:.8.004. 0 .:.245. shot straight upward from the ground with an initial velocity of 49 mjs. y (l) = 1. and 0.8 as 4..04)v.:.:. I y' = x 2 .:. 3 y 2 y' = 2x 5 . y(O) = 1. y' = x 2 + y 2 .1.:. Use a calculator or computer implementation of the improved Euler method to approximate v(t) for 0 . 23. 1 . 0 . y(l) = 2. x . 0 .0003P 2 dt (with tin months). 2 xy' = y 2 . use the improved Euler method with step sizes h = 0.9. 30. t .. so your downward velocity satisfies the initial value problem .9.8 as 4.2. Use a calculator or computer implementation of the improved Euler method to approximate v(t) for 0 .. dt dv with exact solution v(t ) = 294e1125 .0011)vlvl...01.If. Do the two approximationseach rounded to two decimal placesagree both with each other and with the exact solution? If the exact solution were unavailable.1.:.. 13. 2 y' =X+ 0'· y(O) = 1. x . 0 . Consider now the crossbow bolt of Example 3 in Section 1. y(O) = 2. 21. x. Start with step size h = 0. x ... x ~ 1 X 24. first with step size h = 0.:. first with step size h = 1 and then The symbolic solution discussed in Section 1. 2 y' = X + .:..:. What percentage of the limiting population of 75 deer has been attained after 5 years? After 10 years? Use the improved Euler method with a computer system to find the desired solution values in Problems 27 and 28. X .005. t .0225P . 2. 12.02.:. primes denote derivatives with respect to x. 2 y' = x 213 + y 2!3. 2 y' = sinx +cosy.:.. 2 xy' = 3x.1. y(2) = 0..8. 14.. y ( .:. Display the results at intervals of I second. x .. 16.:.. 1 y' = ~(y. 19..2. Consider the crossbow bolt of Example 2 in Section 1.:. Because of linear air resistance.:. 27. In these initial value problems. ~ 1 25.:. I . t . then with step size h = 0. 2.5.1. Make a table showing the approximate values and the actual value.
2 A Closer Look at the Euler Method 451 6. The impeuler function takes as input the initial value x. Figure 6. y ).Y) Y = YO + H*K1 X = X + H K2 = FNF(X. then to some of the problems for this section. 6. As output it produces the resulting column vectors x andY of x . the MATLAB command [X.andyvalues. To increase the number of steps (and thereby decrease the step size) one need only change the value of N specified in the second line of the program. of steps Initial x Initial y Final x Step size Begin loop Save previous y First slope Predictor New x Second slope Average slope Corrector Display results End loop FIGURE 6.2.6.Y NEXT I Program title Define function f No. and the desired number n of subintervals.11 exhibits one MATLAB implementation of the improved Euler method. The comments provided in the final column should make these programs intelligible even if you have little familiarity with the BASIC and TI programming languages. Comment PROGRAM:IMPEULER :F=X+Y : 10+N : 0+X :1+Y : 1+ X1 : (X1X) /N+B :For(I.Y) K = (K1 + K2)/2 Y = YO + H*K PRINT X.Y :End Program IMPEULER DEF FN F(X. To apply the improved Euler method to a differential equationdyjdx = f(x. Y] = impeuler(O. .2.10 lists TI85 and BASIC programs implementing the improved Euler method to approximate the solution of the initial value problem dy dx = x + y.2 Application Figure 6. y (0) = 1 considered in Example 2 of this section. Test your program by applying it first to the initial value problem of Example 1. For instance. the final value x1 of x.2. in which the function f is defined.1.2.10.4. TI85 and BASIC improved Euler programs.Y) = X + Y N = 10 X = 0 y = 1 X1 = 1 H = (X1X)/N FOR I=1 TO N YO = Y K1 = FNF(X. 1. You should begin this project by implementing the improved Euler method with your own calculator or computer system. 1.N) :Y+YO : F+K1 : YO+B*K1+ Y :X+B+X : F+K2 : (K1+K2) /2+K : YO+B*K+ Y :Disp X. 10) then generates the first and fourth columns of data shown in Fig. the initial value y. one need only change the initial line of the program.
Perhaps a table of values for 0 ~ x ~ 100 will make this apparent. 20. consider the initial value probIem dy 1 dx = . % first slope k2 = f(x+h. Does the improved Euler approximation with step size h = 1 level off at the "correct" limiting value of the exact solution? If not.Y] = impeuler(x. . where y (x) is the solution of the initial value problem dyjdx = y. Logistic Population Investigation Apply your improved Euler program to the initial value problem d yjdx = ~ y(8. How many subintervals are needed to obtaintwice in successionthe correct value of the target number rounded to five decimal places? 1.1415927 as specific values of certain initial value problems. where y (x) is the solution of the initial value problem dyjdx = 4/(1 + x 2 ).y ).3542 rather than at the limiting value y = 8 of the exact solution. MATLAB implementation of improved Euler method.6931472. where m and n are (for instance) the largest and smallest nonzero digits in your student ID number. y(O) = 0.y+h*kl).y. y (O) = 1 of Example 3. In 2 ~ 0. % average slope % new x X = X + h.7182818. For your own logistic population to investigate. % begin loop kl = f(x. % update xcolumn y = [Y. find a smaller value of h so that it does. y (O) = 1. and n ~ 3.. % new y X = [X.y).. % second slope k (kl + k2)/2.11. 3.2. x].y ). y = y + h*k. apply the improved Euler method with n = 10.xl.452 Chapter 6 Numerical Methods function yp YP = X + Yi = f(x. y(O) = 1. 2. In particular. 40.y) % yp = Y' function [X. y(m . subintervals (doubling n each time). The number In 2 = y(2). .y]. % update ycolumn end % end loop  FIGURE 6. where y(x) is the solution of the initial value problem dyjdx = 1/x . Famous Numbers Revisited The following problems describe the numbers e ~ 2.. The number n = y(l) . % step size X = x· % initial X ' y = Yi % initial y for i = l:n.. The number e = y (1). y(l) = 0. verify (as claimed) that the approximate solution with step size h = 1 levels off at y ~ 4.n) h = (xl x)/n. In each case.
.0 1. and P). y(x3). y(x2) . y(xn) and now want to compute Yn+! ~ y(xn+d· Then y(xn+!). ..y(xn) = 1 Xn 1 Xn+l y 1 (x) dx = l xn +h y 1 (x) dx Xn (2) by the fundamental theorem of calculus. Next. Yn to the actual values y(xJ). +2 1 (3) Hence we want to define Yn+ l so that Yn+l ~ Yn h[Y (Xn) + 2y +6 1 Xn h) + 2y +2 1 ( Xn h) + Y (Xn+ !)J. Y2. y3. 6. It is called the RungeKutta method.sin2.y). This figure suggestsalthough it does not provethe existence of a threshold initial population such that • Beginning with an initial population above this threshold. the population oscillates (perhaps with period P?) around the (unharvested) stable limiting population y (t) = M. whereas • The population dies out if it begins with an initial population below this threshold.y(xn) ~ h[Y (Xn) + 4y 6 1 ( 1 ( Xn h) + Y (Xn+ !) J.3 The RungeKutta Method 2. suppose that we have computed the approximations YJ.. y). ... A numerical approximation program was used to plot the typical solution curves for the case k = M = h = P = 1 that are shown in Fig.5 1.6.0 ~~~~~~~~ 2 4 5 models a logistic population that is periodically harvested and restocked with period P and maximal harvesting/restocking rate h. .5 (2nt) 0. Solution curves = y(l. Do the observations indicated here appear to hold for your population? ofdyfdt FIGURE 6. We now discuss a method for approximating the solution y = y(x) of the initial value problem dy dx = f(x. after the German mathematicians who developed it.12. y(xo) =Yo (1) that is considerably more accurate than the improved Euler method and is more widely used in practice than any of the numerical methods discussed in Sections 6. 0.y). M. With the usual notation. Carl Runge (18561927) and Wilhelm Kutta (18671944).5 453 Periodic Harvesting and Restocking The differential equation dy = ky(M.12.1 and 6.0 .2.nt.2.>. Simpson's rule for numerical integration yields y(Xn+!).2. +2 1 (4) . h. .h sin dt p 1. Use an appropriate plotting utility to investigate your own logistic population with periodic harvesting and restocking (selecting typical values of the parameters k.
Xn+ Il using the Euler method to predict the ordinate there. b] with a = x 0 is of order h 4 .. Xn+Il· The RungeKutta method is a fourthorder methodit can be proved that the cumulative error on a bounded interval [a.. but does not depend on the step size h. the result is the iterative formula (6) The use of this formula to compute the approximations y 1.454 Chapter 6 Numerical Methods we have split 4y' (xn +~h) into a sum of two terms because we intend to approximate the slope y' (Xn + ~h) at the midpoint Xn + ~ h of the interval [Xn . with the following estimates. (5d) • This is the Euler method slope at Xn+I. (Sa) • This is the Euler method slope at Xn. (6) takes the "Euler form" Yn+I = Yn +h ·k if we write (7) for the approximate average slope on the interval [xn . successively constitutes the RungeKutta method. (8) where the constant C depends on the function f(x. Note that Eq. The following example illustrates this high accuracy in comparison with the lowerorder accuracy of our previous numerical methods. (Thus the iteration in (6) is sometimes called the fourthorder RungeKutta method because it is possible to develop RungeKutta methods of other orders. using the improved slope k3 at the midpoint to step to Xn+I· When these substitutions are made in (4 ). y3. b]. (5b) • This is an estimate of the slope at the midpoint of the interval [xn. we replace the [true] slope values y'(xn). y' (xn + ~h). andy' (xn+I). y' (xn + ~h). . On the righthand side in (4).) That is. . y) and the interval [a. respectively. (5c) • This is an improved Euler value for the slope at the midpoint. Xn+ 1] in two different ways. Y2.
In the first step we use the formulas in (5) and (6) to calculate kt = 0 + 1 = 1. + (1 + (0.2. so only two steps are required to go from x = 0 to x = 1. k2 = (0 + 0. . y(O) = 1 with the same step size h = 0. It is customary to measure the computational labor involved in solving dyjdx = f(x. We see that even with the larger step size.25). (1. .5) + (0. . + (0.3. 6. the Runge.5 YI = 1 + 6[1 + 2 · (1. RungeKutta and improved Euler results for the initial value problem dyjdx = x + y.0 1.6.1 and again in Example 2 of Section 6.03% 0.4366 FIGURE 6. Thus the RungeKutta method gave over four times the accuracy with only 40% of the labor. y(0)=1 (9) that we considered in Fig.x . Figure 6.25).05% 1.4347. 6.24% 1.7974 3. y) that are required.0 0. Figure 6.1.5) + 2 · (1.5)) = 1.8 of Section 6.3125.1.7949 3. In this comparison the Runge. 2. In Example 1. a larger step size than in any previous example. . y) numerically by counting the number of evaluations of the function f (x. whereas the improved Euler method required 20 such evaluations (two for each of 10 steps). The relative error in the improved Euler value at x = 1 is about 0.625)) = and then 0. y) = x + y (four at each step). To make a point we use h = 0.7969 3.2 shows the results obtained by applying the improved Euler and RungeKutta methods to the problem dyjdx = x + y.1.7969.1.2.0000 1.Kutta method is about 2000 times as accurate.5 1.00012%.0000 1. y(O) = 1.25) + (1 k4 = (0.00% 0.1 presents these results together with the results (from Fig.5).25) + (1 k3 = (0 + 0. (1)) = 1.3 The RungeKutta Method Example 1 455 We first apply the RungeKutta method to the illustrative initial value problem dy dx =x+y. but for the RungeKutta value it is 0.625.5. RungeKutta 0. • .4282 0.Kutta method required eight evaluations of f(x. Computer programs implementing the RungeKutta method are listed in the project material for this section.3125] ~ 1. but requires only twice as many function evaluations. Similarly. The exact solution of this problem is y (x) = 2ex . (1.3.4347 0. as the improved Euler method. the second step yields y 2 ~ 3.14% 0.3.0000 1.1.00% 0.625) + 2.5.24%. the RungeKutta method gives (for this problem) four to five times the accuracy (in terms of relative percentage errors) of the improved Euler method.4) of applying the improved Euler method with step size h = 0.
1 0.0 X 2.05.9 1.436564 FIGURE 6.583649 1. .7 0.3. .0 1. .1115 1.6529 14.0218 1.2. RungeKutta and improved Euler results for the initial value problem dyjdx = x + y. y(O) = L 0. Figure 6.399717 1.0670 3.651079 3. .3).8 0. There is still some difficulty near x = 0. .0124 3.242806 1.1 we saw that Euler's method is not adequate to approximate the solution y(x) of the initial value problem .5) ~ 2.1.6522 14. . FIGURE 6. . . . .2 0.5818 1.3. . y(0)=1 (10) .2712 1.3.0670.4282 1. . computed with step sizes h = 0.1100 1.0 as x approaches the infinite discontinuity near x = 0.1.4.9. .0670 3. .044238 2. with the same step size h = 0. .110342 1.4397 2.5 0.0670 3. and h = 0.4 0. .327505 2.2 0  ~~ ~ ~ ~ ~_r_ ~ __+ 1 x=~0~.5 0. __ : : ::::::::::: : :1:: ::: 1.9]. . h = 0.4397 2.3 0. .1115 1.019203 3.969811 (see Fig.3. . . . . . .797443 2. Solutions of dyjdx = x 2 + y 2 .583648 1. 'I' .= x2 6 dy dx +l. Approximating the solution of the initial value problem in Eq. It follows from the inequality in (8) that (on a fixed bounded interval) halving the step size decreases the absolute error by 4 = 1~.651082 3. .025.1 0.0409 2.327503 2.3985 1.1115 1.019206 3.' ' ' ' ' 1 .3021 FIGURE 6.3.242805 1.9~7 ' ' ' ' ' ' ' .3 0.110342 1.2421 1.4 shows RungeKutta results on the interval [0.399718 1. .3231 2.0.456 Chapter 6 Numerical Methods 0. the common practice of successively halving a factor of the step size until the computed results "stabilize" is particularly effective with the RungeKutta method.436559 1. . (10). y(O) = 1. .4397 2. .797441 2. 0.6456 3. I· . Consequently. . . The error bound (8) for the RungeKutta method results in a rapid decrease in the magnitude of errors when the step size h is reduced (except for the possibility that very small step sizes may result in unacceptable roundoff errors).9 1. G) Example 2 In Example 5 of Section 6. .6 0. 6.7 0.:. .3.' ' ' ' 2 ' '. .044236 2.6529 14.7949 2. Now we apply the RungeKutta method to this initial value problem. but it seems safe to conclude from these data that y(0.
0670.9 2.95 is y(0.3.9].1073 33.90 0.0670 2.3.7 0. 0. •m N Example 3 A skydiver with a mass of 60 kg jumps from a helicopter hovering at an initial altitude of 5 kilometers.3049 16.92 0.91 0.ooos 0.6440 3.6529 5.6440 3.0670 2. • y with h~ o. ·~ ~·• o• ~ ·~o.0670 2.95] for the initial value problem dx = x 2 + l. Fig.3049.3049 FIGURE 6.5 0. We now conclude that y(0.93 0.005.8486 14.0025 2. dy y(0.3 The RungeKutta Method 457 We therefore begin anew and apply the RungeKutta method to the initial value problem dx = x 2 + l.5) = 2.90. 6.3049 16.5 shows results on the interval [0.304864. dy y(0. and h = 0. Our slight overestimate results mainly from the fact that the fourplace initial value in (12) is (in effect) the result of rounding up the actual value y(0.7024 20.9) ~ 14.4723 FIGURE 6.3049 16.6529 5. 0. 005 0.1073 33.8486 14. h = 0. Finally.3049. what will be her terminal velocity? How fast will she be falling after 5 shave elapsed? After 10 s? After 20 s? .6529 5.6 0. (11). The actual value of the solution at x = 0.3048 2.95 14.3.95) ~ 50. Our final approximate result is y(0.3.0617 25.002.6 shows results on the interval [0.0096)(100v + 10v2 + v3) (in newtons. (11) Figure 6. (12) obtained using step sizes h = 0. Approximating the solution of the initial value problem in Eq.5363 50.7024 20. Assume that she falls vertically with initial velocity zero and experiences an upward force F R of air resistance given in terms of her velocity v (in meters per second) by FR = (0.0617 25.5. such errors are magnified considerably as we approach the vertical asymptote.5. obtained with step sizes h 0. and h = 0.8 0.6440 3.001.5363 50.94 0. If she does not open her parachute.01.0005.6.95) ~ 50. (12).0617 25.9) ~ 14.8486 14. Approximating the solution of the initial value problem in Eq. .6.4723.0025.7024 20.9) = 14.471867.3049 = y with h = 0.4723 14.5363 50. and with the coordinate axis directed downward so that v > 0 during her descent to the ground). h = 0.1073 33.4722 14.
We can therefore calculate her terminal velocity immediately by solving the equation f(v) = 9.9 shows the results obtained by applying the RungeKutta method on the interval [0..578 mjs.Kutta approximations to the solution of the initial value problem in (14).239 35.8. But any departure. Although y(x) = ex ~ 0 as x ~ +oo. the numerical methods we have discussed are not nearly so successful as in the previous examples.574 35.5780 (found graphically or by using a calculator or computer Solve procedure).00016)(100v 0 2 3 4 5 6 7 8 9 10 0 9. v(O) = 0. the step sizes h = 0.(0.00016)(100v + 10v2 + v 3 ).415 35. Observe that the terminal velocity is effectively attained in only 15 s. • The final example of this section contains a warning: For certain types of initial value problems.875 35. Thus the velocity function v(t) satisfies the initial dv dt = f(v).386 25. And as indicated in Fig.3. The explanation lies in the fact that the general solution of the equation dyjdx = 5y . (18) The particular solution of (1 7) satisfying the initial condition y (O) = 1 is obtained with C = 0.6e.05.9604020 0 20 40 60 80 100 v because m = 60 and g value problem = 9.xeven if due only to roundoff errorintroduces [in effect] a nonzero value of C in Eq. 6.1.949 32.577 35. however small.3. Thus the skydiver's terminal speed is approximately 35. (14) where FIGURE 6.636 18.3.85 % of her terminal velocity after only 5 s.00016)(100v+ 10v2 +v 3 ). 4] with step sizes h = 0.578 + 10v2 + v3 ) = 0. • . all solution curves of the form in (18) with C i= 0 diverge rapidly away from the one with C = 0.0096)(100v + l0v 2 + v3) (13) 20 40 6 . Graph of j(v) = 9. and 99.8 shows the results of Runge. 6.541 35.678 34.3.569 35.578 35. Figure 6.7 shows the graph of the function f(v) and exhibits the single real solution v ~ 35.2 and h = 0.2.137 34. even if their initial values are close to 1.x is y (x) =ex+ Ce 5 x. X y(O) = 1 (17) whose exact solution is y (x) = e. so f (v) = 0. (15) The skydiver reaches her terminal velocity when the forces of gravity and air resistance balance.(0.3. (16) Figure 6.8.8. Consider the seemingly innocuous initial value problem dy dx = 5y FIGURE 6. 60~~ = (60)(9.500 ll 12 13 14 15 16 17 18 19 20 35.. Obviously these attempts are spectacularly unsuccessful.560 35. The skydiver's velocity data.1 0.299 29. h = 0. The table in Fig.578 35.8). from the exact solution y(x) = e. But the skydiver's velocity is 91.x.576 35.(0. (18).578 35. it appears that our numerical approximations are headed toward oo rather than zero.3.7. about 128 km/h (almost 80 mi/h). and h = 0.78% after 10 s. Example 4 6e.1 yield the same results (to three decimal places).8.(0.458 Chapter 6 Numerical Methods Solution 40 20 Newton's law F =rna gives that is. f(v) = 9.8.
6 4.3.2 1.29376 0.0 L~L.4 2.69395 14059. y' = y.. (17)..x .12102 1 .44933 0. y (O) = 1.20190 0. y(O) = 1.x + x. y(x ) = ln(x + 1) I.67032 0.25 and 0.r. y(x ) = 2ex2 7.3x 2 y.5 1.35494 103. Direction field and solution curves for dy jdx = 5y .5.J_L.5 1. y' 5.14697 0..90419 22.04076 0.6 2. y(O) = 3.44926 0.9.1. I26. y(x) = ~ e2x = y + 1.09072 0.19802 0. 0.1 y .8 1. = 0. y' 3. Apply the Runge.o Ul~W~~~~~ 0. y' = ..25023 1.5I868 85.67020 0.5 o.30067 0.. y (x ) = 2 + x .I . 2.05352 I63. y' + y 2 ) .25. 2.Y. y' = 2xy . y(O) = 2.0 2. RungeKutta attempts to solve numerically the initial value problem in Eq..72142 35. y' = iO 10.10668 0.2 3. y(O) = ~...4 0. y(x) = 2ex = = 2x y 2 .12866 = o.67031 0.0 0.03934 FIGURE 6.ex Note: The application following this problem set lists illustrative calculator/computer programs that can be used in the remaining problems. Construct a table showing fivedecimalplace values of the approx imate solution and actual solution at the points x = 0. Approximate values whose order of magnitude varies with changing step size are a common indicator of such instability. Difficulties of the sort illustrated by Example 4 sometimes are unavoidable.Kutta method to approximate this solution on the interval [0.43713 0. y (x ) = 2e.J.02732 0.44833 0. y' 2.8 3. but one can at least hope to recognize such a problem when it appears.I X .46019 4.6ex.21099 0. y (O) = 1.10.U.3. where an initial value problem and its exact solution are given.30199 0.5270 0. y' 4.x = . y(O) = 2.0 1.__1.3 The RungeKutta Method 459 RungeKutta y withh 0. .x = 2y. These difficulties are discussed in numerical analysis textbooks and are the subject of current research in the field. 11:11 Problems 6.25077 1205. y(x) = 2e ..38156 631.27026 2..50367 Il.53415 261. y(x) = 3ex3 8.5 . y(O) = 1.66880 0.5] with step size h = 0.os Actual y 0.01832 0.y.71249 8903.' 0 X FIGURE 6..13534 0. y (x ) = 2 1.0608I 0.0 0.rr. y(x) = tan i <x + n) I 9. y(O) = 1.6. y(O) A handheld calculator will suffice f or Problems 1 through 10. y' = e.916.
£ 1 X X = x 2 + y 2 . together with the percentage error in the more accurate approximation. y' 18. .2.£ x . y(O) = 1.£ X . y 2 y' = 2x 5 .£ 2 = X + . y(l) = 2. Thus.. Use the RungeKutta method with a programmable calculator or computer to approximate the solution for 10 years. lm (with tn+1 = tn +h)..£ X .6v. 11.£ 2. 1 . ).. Because of linear air resistance.1. y' = . t2. y(O) = 1. shot straight upward from the ground with an initial velocity of 49 m js. Consider again the crossbow bolt of Example 2 in Section 1. 3.2. As in Problem 25 of Section 6. you bail out of a helicopter and immediately open your parachute.245. What percentage of the limiting velocity 20 ft/s has been attained after 1 second? After 2 seconds? 26.9. y (0) = 0.£ x . y(l) = 3. t3. first with step size h = 0. Use the RungeKutta method with a programmable calculator or computer to approximate the solution for 0 . 2.£ X . by beginning with the velocity data in Fig.y . VJ.2y. then with step size h = 0. 0 ~ x That is.04)v.£ 2 = In y. the linear acceleration a = dvjdt of a moving particle is given by a formula dvjdt = f(t. As in Problem 26 of Section 6.1. Suppose that the velocity v(t) is approximated using the RungeKutta method to solve numerically the initial value problem dv dt imate to six decimal places the values of the solution at five equally spaced points of the given interval. y(O) = 0. Make a table showing the approximate values and the actual value. the formulas in Eqs. 0 ~ x . y' 19. rounding off approximate Pvalues to four decimal places.. In these initial value problems.£X . (a) Use a calculator or computer implementation of the Runge. tn+I ].£ 2 = sinx +cosy.£ x ~ 2 = x 213 + y 213. v(O) = 49 dP = 0. Start with step size h = 1.3. 1 . y(2) = 3. . = f(t. primes denote derivatives with respect to x. once a table of approximate velocities has been calculated. Vm at the successive times t 1. The formula in (20) would give the correct increment (from Yn to Yn+I) if the acceleration an remained constant during the time interval [tn. y(O) = 1. 0 . Vn) ~ v'(tn) is the particle's approximate acceleration at time tn. 2 ~ x ~ 3 16. 0. 0 . We can do this by beginning with the initial position y(O) = Yo and calculating (20) 24. where the velocity v = dyjdt is the derivative of the function y = y(t) giving the position of the particle at timet.  S 1  25. y' 21.Kutta method to approximate v(t) for 0 ~ t ~ 10 using both .£ 1 = 4(y 1)2 .1 and then with h = 0. Eq.460 Chapter 6 Numerical Methods (with t in months). y' 12. first with step size h = 0.1125 .£ 3 A computer with a printer is required for Problems 17 through 24. y' = x 2 + y 2 .8. 0 . v(O) = 0 (with t in seconds and v in ft/s).£ 2 = X + ~. . y' 22. In each problem find the exact solution of the given initial value problem. and 0. y' 23. What percentage of the limiting population of 75 deer has been attained after 5 years? After 10 years? A programmable calculator or a computer will be useful for Problems 11 through 16.£ x . Then apply the RungeKutta method twice to approximate (to five decimal places) this solution on the given interval. its velocity function v = dyjdt satisfies the initial value problem dv dt = (0. v(O) = vo. y' = y. y(2) =? 28.. y(2) = 3.£ t .JY. xy'=y 2 .0225 P . 0 . 6. for x an integral multiple of 0. y(2) =? 13. (20) provides a simple way to calculate a table of corresponding successive positions. yy' = 2x 3 .0003P 2 dt with exact solution v(t) = 294e. 2 .1. and then use successively smaller step sizes until successive approximate solution values at x = 2 agree rounded off to five decimal places. 0 .1~x~2 15.£ X . Now suppose that we also want to approximate the distance y(t) traveled by the particle. Throughout. xy' = 3x. 0. . .05. where an = f(tn. (19) 17. y(O) = 0. Print the results in tabular form with appropriate headings to make it easy to gauge the effect of varying the step size h.2.2. starting with t0 = 0 and v0 .. so your downward velocity satisfies the initial value problem dv dt = 32. y(2) = 0.8.£ 1 Use the RungeKutta method with a computer system to find the desired solution values in Problems 27 and 28. (5) and (6) are appliedwith t and v in place of x and yto calculate the successive approximate velocity values v 1... y (0) = 2. y' 20.2. y' =X+ 4y 2 .y(1)=1. y' .1.05. 8 (Example 3) and proceeding to follow the skydiver's position during her descent to the ground. rounding off approximate vvalues to three decimal places.0. 0 .025 to approx VelocityAcceleration Problems In Problems 29 and 30. y(1) = 1· 1 S 1 + y2 . v). y(O) = 1. v). primes denote derivatives with respect to x.2.£ 2 14. 29. v2 . This process is illustrated in the project for this section. y(O) = 1. Throughout. 27.£ I 2 2 = x . use the RungeKutta method with step sizes h = 0. first with step size h = 6 and then with h = 3. suppose the deer population P(t) in a small forest initially numbers 25 and satisfies the logistic equation (n = I.
4. the bolt's velocity and position functions during ascent and descent are given by the following formulas.245t? (c) If the exact solution were unavailable. explain how you could use the RungeKutta method to approximate closely the bolt's times of ascent and descent and the maximum height it attains. According to the results of Problems 17 and 18 in Section 1. Now consider again the crossbow bolt of Example 3 in Section 1. v(O) = 49.465 + (909.1/ 25 ) . As output it produces the resulting column vectors x andY of x. one need only change the initial line of the program. y). but because of air resistance proportional to the square of its velocity. Do these approximate position valueseach rounded to two decimal placesagree with the exact solution y(t) = 7350 (1 461 Beginning with this initial value problem. the initial value y. its velocity function v(t) satisfies the initial value problem dv .11 lists TI85 and BASIC programs implementing the RungeKutta method to approximate the solution of the initial value problem dy dx = x + y. Ascent: v(t)  e.103827]t). 12 exhibits a MATLAB implementation of the RungeKutta method.091) In (cos(0.388) tan(0.3.OOll)vlvl.8.8. 30. dt y(t) = = (94.3 The RungeKutta Method n = 100 and n = 200 subintervals. You should begin this project by implementing the RungeKutta method with your own calculator or computer system. 108.9. Y] = rk(O. 6.3. Display the results at intervals of 1 second. Descent: v(t) = (94. and the desired number n of subintervals.(909.091) In (cosh(O.= (O.3. then to some of the problems for this section.103827]t)). repeat parts (a) through (c) of Problem 25 (except that you may need n = 200 subintervals to get fourplace accuracy in part (a) and n = 400 subintervals for twoplace accuracy in part (b)). 10) then generates the first and third columns of data shown in the table in Fig. Then the rk function takes as input the initial value x.465 . y) has been defined.2. y(t) = 108. 6.6119])).4. Do the two approximationseach rounded to four decimal placesagree both with each other and with the exact solution? (b) Now use the velocity data from part (a) to approximate y(t) for 0 ~ t ~ 10 using n = 200 subintervals.andyvalues.8. To apply the RungeKutta method to a different equation dyjdx = f(x. .6119]). the MATLAB command [X. 1. It still is shot straight upward from the ground with an initial velocity of 49 mjs.478837.l03827[t. Suppose that the function f describing the differential equation y' = f (x. one need only change the value of N specified in the second line of the program. For instance. Test your program by applying it first to the initial value problem in Example 1. Figure 6. y(O) =1 considered in Example 1 of this section.388) tanh(0. Display the results at intervals of 1 second. To increase the number of steps (and thereby decrease the step size). the final value x1 of x.3 Application RungeKutta Implementation Figure 6.478837. 1. in which the function f is defined.[0.[0.103827[t.6. The comments provided in the final column should make these programs intelligible even if you have little familiarity with the BASIC and TI programming languages.
for i = l:n k1 = f(x.3.y+h*k2/2).Y) = X + Y N = 10 X 0 y = 1 : 1~x1 : (X1X) /N~H :For(I.y+h*k1/2). end % yp = y' % % % % % % % % % % % % % % step size initial X initial y begin loop first slope second slope third slope fourth slope average slope new x new y update xcolumn update ycolumn end loop FIGURE 6.y).11.y+h*k3). M ATLAB implementation of the Runge. y = (Y. TI85 and BASIC RungeKutta programs. function yp = f(x.n) h = (x1 .1.Y] = rk(x.Y :End X1 = 1 H = (X1X)/N FOR I=1 TO N XO = X YO = Y K1 = FNF(X.3.Kutta method.y. X = x· ' y = y. X = X + h.12. k = (k1+2*k2+2*k3+k4)/6.Y) X = XO + H/2 Y = YO + H*K1/2 K2 = FNF(X.N) :x~xo :Y~YO :F~K1 :XO+H/2~X :YO+H*K1/2~Y :F~K2 :YO+H*K2/2~Y :F~K3 :XO+H~X :YO+H*K3~Y :F~K4 : (K1+2*K2+2*K3 +K4) /6~K :YO+H*K~Y :Disp X. of steps Initial x Initial y Final x Step size Begin loop Save previous x Save previous y First slope Midpoint Midpt predictor Second slope Midpt predictor Third slope New x Endpt predictor Fourth slope Average slope Corrector Display results End loop FIGURE 6. y = y + h*k.Y) X = XO + H Y = YO + H*K3 K4 = FNF(X. k4 f(x+h. X = [X. k3 = f(x+h/2.y).Y) K = (K1+2*K2+2*K3 +K4)/6 Y = YO + K*K PRINT X.x).Y NEXT I Program title Define function f No.x)/n.462 Chapter 6 Numerical Methods PROGRAM:RK :F=X+Y :10~N :o~x :1~Y Program RK DEF FN F(X. .Y) Y = YO + H*K2/2 K3 = FNF(X.y) yp = X + Yi function [X.x1. k2 = f(x+h/2.
.v) vp = 9. (20) in the instructions for Problems 29 and 30. or about 2 min 23 s.866 % 200 subintervals % RungeKutta approximation % Display every lOth entry 0 18. k).14159265359 as specific values of certain initial value problems. The skydiver's velocity and position data. apply the RungeKutta method with n = 10. Hence her total time of descent is 20 + (4370. 0.192 345.: k = 200 [t.13.v] = rk(O.763 203. v(l:lO:k+l)] 0 18. 2.6. Finally.s*a*h~2.c1. The results of these calculations are shown in the table in Fig. The number ln2 = y(2).866 m during her first 20 s of descent.833 s.13. y(O) = 0..71828182846. It appears that the skydiver falls 629.00016*(100*v + 10*v~2 + v~3). end [t(l:20:k+l).3. function vp = f(t. 20. y(1) = 0. t_{s) · 0 2 4 6 8 10 12 14 16 18 20 1t ~gll~):i:. . using your own mass m and a plausible airresistance force of the form FR = a v + bv 2 + cv 3 • . 6.3.y(1:20:k+l)] % initialize y % step size % for n = 1 to k % acceleration % Equation (20) % end loop % each 20th entry FIGURE 6. The number n = y ( 1).8. 40.500 35.239 35.825 133. Then the commands .137 35. ln2 ~ 0.949 34. where y(x) is the solution of the initial value problem dyjdx = y.v(n)): y(n+l) = y(n) + v(n)*h + o.3 The RungeKutta Method 463 Famous Numbers Revisited.392 274. How many subintervals are needed to obtaintwice in successionthe correct value of the target number rounded to nine decimal places? 1. for n = l:k a= f(t(n).386 29.266 416.3.577 35.134/35. and then free falls the remaining 4370.710 629. 6. y(O) = 1. 20. analyze your own skydive (perhaps from a different height). where y (x) is the solution of the initial value problem dyjdx = 4/(1 + x 2 ). In each case. where y(x) is the solution of the initial value problem dyjdx = 1jx.1..578 35.560 35. [t(l:lO:k+l).v(l:20:k+l). One Last Time The following problems describe the numbers e ~ 2.578 produce the table of approximate velocities shown in Fig. carry out the position function calculations described in Eq.578 mjs...0.134 meters to the ground at her terminal speed of 35. and n ~ 3.l): h = 0.403 487.578) ~ 142.574 35.555 558.69314718056. The Skydiver's Descent The following MATLAB function describes the skydiver's acceleration function in Example 3. subintervals (doubling n each time).8. The number e = y(1).984 68. 3. For an individual problem to solve after implementing these methods using an available computer system. the commands y = zeros(k+l.
let us write x = [~ J and f =[~ J. Yn ).. . Our goal is to apply the methods of Sections 6. Euler Methods for Systems For example.. With this assurance we can proceed to discuss the numerical approximation of this solution. We can then make the step from Xn to the next approximation Xn+I ~ x(tn+I) by any one of the methods of Sections 6. the iterative formula of Euler's method for systems is Xn+l = Xn + hf(t. ' of the exact solution of the system in (1). (4) . y' = g(t . and are vectorvalued functions. x). where tn+I = tn + h for n ~ 0.. (3) Then the initial value problem in ( 1) is x' = f(t. Yn + h g (tn . x0 ).. tz. Xn. Yn ).. x(to) = xo (1) for a system of m firstorder differential equations.3 to the initial value problem x' = f(t.3. y (to) = Yo