*145*
*74*
*10MB*

*English*
*Pages 617*
*Year 2017*

- Author / Uploaded
- William E. Boyce
- Richard C. DiPrima
- Douglas B. Meade

Boyce 9131 FM 2

October 12, 2016

17:16 iii

Elementary Differential Equations and Boundary Value Problems Eleventh Edition

WILLIAM E. BOYCE Edward P. Hamilton Professor Emeritus Department of Mathematical Sciences Rensselaer Polytechnic Institute

RICHARD C. DIPRIMA formerly Eliza Ricketts Foundation Professor Department of Mathematical Sciences Rensselaer Polytechnic Institute

DOUGLAS B. MEADE Department of Mathematics University of South Carolina - Columbia

iii

Boyce 9131 FM 2

October 22, 2016

10:12 iv

VICE PRESIDENT AND DIRECTOR ACQUISITIONS EDITOR SPONSORING EDITOR MARKETING MANAGER EDITORIAL ASSISTANT SENIOR CONTENT MANAGER SENIOR PRODUCTION EDITOR SENIOR DESIGNER PRODUCT DESIGNER PRODUCT DESIGN MANAGER PRODUCTION SERVICES COVER ART

Laurie Rosatone Shannon Corliss Jennifer Brady John LaVacca III Ryann Dannelly Valerie Zaborski Ken Santor Wendy Lai David Dietz Tom Kulesa Jeanine Furino/Cenveo Publisher Services John Lund/Getty Images, Inc.

This book was set in STIX by Cenveo Publisher Services using LaTeX and printed and bound by Quad Graphics Versailles. The cover was printed by Quad Graphics Versailles. This book is printed on acid-free paper. ∞ c Copyright 2017, 2012, 2009 John Wiley & Sons, Inc. All rights reserved.

No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, scanning or otherwise, except as permitted under Sections 107 or 108 of the 1976 United States Copyright Act, without either the prior written permission of the Publisher, or authorization through payment of the appropriate percopy fee to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, website www.copyright.com. Requests to the Publisher for permission should be addressed to the Permissions Department, John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030-5774, (201) 748-6011, Fax: (201) 748-6008, website http://www.wiley.com/go/permissions. Evaluation copies are provided to qualified academics and professionals for review purposes only, for use in their courses during the next academic year. These copies are licensed and may not be sold or transferred to a third party. Upon completion of the review period, please return the evaluation copy to Wiley. Return instructions and a free of charge return shipping label are available at www.wiley.com/go/returnlabel. Outside of the United States, please contact your local representative. The inside back cover will contain printing identification and country of origin if omitted from this page. In addition, if the ISBN on the back cover differs from the ISBN on this page, the one on the back cover is correct.

ePub ISBN-13 978-1-119-37792-4 BRV ISBN-13 978-1-119-25600-7 EVAL ISBN-13 978-1-119-37575-3 Printed in the United States of America 10 9 8 7 6 5 4 3 2 1

iv

Preface As we have prepared an updated edition our first priorities are to preserve, and to enhance, the qualities that have made previous editions so successful. In particular, we adopt the viewpoint of an applied mathematician with diverse interests in differential equations, ranging from quite theoretical to intensely practical--and usually a combination of both. Three pillars of our presentation of the material are methods of solution, analysis of solutions, and approximations of solutions. Regardless of the specific viewpoint adopted, we have sought to ensure the exposition is simultaneously correct and complete, but not needlessly abstract. The intended audience is undergraduate STEM students whose degree program includes an introductory course in differential equations during the first two years. The essential prerequisite is a working knowledge of calculus, typically a two- or three-semester course sequence or an equivalent. While a basic familiarity with matrices is helpful, Sections 7.2 and 7.3 provide an overview of the essential linear algebra ideas needed for the parts of the book that deal with systems of differential equations (the remainder of Chapter 7, Section 8.5, and Chapter 9). A strength of this book is its appropriateness in a wide variety of instructional settings. In particular, it allows instructors flexibility in the selection of and the ordering of topics and in the use of technology. The essential core material is Chapter 1, Sections 2.1 through 2.5, and Sections 3.1 through 3.5. After completing these sections, the selection of additional topics, and the order and depth of coverage are generally at the discretion of the instructor. Chapters 4 through 11 are essentially independent of each other, except that Chapter 7 should precede Chapter 9, and Chapter 10 should precede Chapter 11. A particularly appealing aspect of differential equations is that even the simplest differential equations have a direct correspondence to realistic physical phenomena: exponential growth and decay, spring-mass systems, electrical circuits, competitive species, and wave propagation. More complex natural processes can often be understood by combining and building upon simpler and more basic models. A thorough knowledge of these basic models, the differential equations that describe them, and their solutions--either explicit solutions or qualitative properties of the solution--is the first and indispensable step toward analyzing the solutions of more complex and realistic problems. The modeling process is detailed in Chapter 1 and Section 2.3. Careful constructions of models appear also in Sections 2.5, 3.7, 9.4, 10.5, and 10.7 (and the appendices to ChapUer 10). Various problem sets throughout the book include problems that involve modeling to formulate an appropriate differential equation, and then to solve it or to determine some qualitative properties of its solution. The primary purposes of these applied problems are to provide students with hands-on experience in the derivation of differential equations, and to convince them that differential

equations arise naturally in a wide variety of real-world applications. Another important concept emphasized repeatedly throughout the book is the transportability of mathematical knowledge. While a specific solution method applies to only a particular class of differential equations, it can be used in any application in which that particular type of differential equation arises. Once this point is made in a convincing manner, we believe that it is unnecessary to provide specific applications of every method of solution or type of equation that we consider. This decision helps to keep this book to a reasonable size, and allows us to keep the primary emphasis on the development of more solution methods for additional types of differential equations. From a student’s point of view, the problems that are assigned as homework and that appear on examinations define the course. We believe that the most outstanding feature of this book is the number, and above all the variety and range, of the problems that it contains. Many problems are entirely straightforward, but many others are more challenging, and some are fairly open-ended and can even serve as the basis for independent student projects. The observant reader will notice that there are fewer problems in this edition than in previous editions; many of these problems remain available to instructors via the WileyPlus course. The remaining 1600 problems are still far more problems than any instructor can use in any given course, and this provides instructors with a multitude of choices in tailoring their course to meet their own goals and the needs of their students. The answers to almost all of these problems can be found in the pages at the back of the book; full solutions are in either the Student’s Solution Manual or the Instructor’s Solution Manual. While we make numerous references to the use of technology, we do so without limiting instructor freedom to use as much, or as little, technology as they desire. Appropriate technologies include advanced graphing calculators (TI Nspire), a spreadsheet (Excel), web-based resources (applets), computer algebra systems, (Maple, Mathematica, Sage), scientific computation systems (MATLAB), or traditional programming (FORTRAN, Javascript, Python). Problems marked with a G are ones we believe are best approached with a graphical tool; those marked with a N are best solved with the use of a numerical tool. Instructors should consider setting their own policies, consistent with their interests and intents about student use of technology when completing assigned problems. Many problems in this book are best solved through a combination of analytic, graphic, and numeric methods. Pencil-and-paper methods are used to develop a model that is best solved (or analyzed) using a symbolic or graphic tool. The quantitative results and graphs, frequently produced using computer-based resources, serve to illustrate and to clarify conclusions that might not be readily apparent from a complicated explicit solution formula. Conversely, the vii

Boyce 9131 FM 2

viii

October 12, 2016

17:16 viii

PREFACE

implementation of an efficient numerical method to obtain an approximate solution typically requires a good deal of preliminary analysis--to determine qualitative features of the solution as a guide to computation, to investigate limiting or special cases, or to discover ranges of the variables or parameters that require an appropriate combination of both analytic and numeric computation. Good judgment may well be required to determine the best choice of solution methods in each particular case. Within this context we point out that problems that request a “sketch” are generally intended to be completed without the use of any technology (except your writing device). We believe that it is important for students to understand that (except perhaps in courses on differential equations) the goal of solving a differential equation is seldom simply to obtain the solution. Rather, we seek the solution in order to obtain insight into the behavior of the process that the equation purports to model. In other words, the solution is not an end in itself. Thus, we have included in the text a great many problems, as well as some examples, that call for conclusions to be drawn about the solution. Sometimes this takes the form of finding the value of the independent variable at which the solution has a certain property, or determining the long-term behavior of the solution. Other problems ask for the effect of variations in a parameter, or for the determination of all values of a parameter at which the solution experiences a substantial change. Such problems are typical of those that arise in the applications of differential equations, and, depending on the goals of the course, an instructor has the option of assigning as few or as many of these problems as desired. Readers familiar with the preceding edition will observe that the general structure of the book is unchanged. The minor revisions that we have made in this edition are in many cases the result of suggestions from users of earlier editions. The goals are to improve the clarity and readability of our presentation of basic material about differential equations and their applications. More specifically, the most important revisions include the following: 1. Chapter 1 has been rewritten. Instead of a separate section on the History of Differential Equations, this material appears in three installments in the remaining three section. 2. Additional words of explanation and/or more explicit details in the steps in a derivation have been added throughout each chapter. These are too numerous and widespread to mention individually, but collectively they should help to make the book more readable for many students. 3. There are about forty new or revised problems scattered throughout the book. The total number of problems has been reduced by about 400 problems, which are still available through WileyPlus, leaving about 1600 problems in print. 4. There are new examples in Sections 2.1, 3.8, and 7.5. 5. The majority (is this correct?) of the figures have been redrawn, mainly by the use full color to allow for easier identification of critical properties of the solution. In

addition, numerous captions have been expanded to clarify the purpose of the figure without requiring a search of the surrounding text. 6. There are several new references, and some others have been updated. The authors have found differential equations to be a never-ending source of interesting, and sometimes surprising, results and phenomena. We hope that users of this book, both students and instructors, will share our enthusiasm for the subject. William E. Boyce and Douglas B. Meade Watervliet, New York and Columbia, SC 29 August 2016

Supplemental Resources for Instructors and Students An Instructor’s Solutions Manual, ISBN 978-1-119-16976-5, includes solutions for all problems not contained in the Student Solutions Manual. A Student Solutions Manual, ISBN 978-1-119-16975-8, includes solutions for selected problems in the text. A Book Companion Site, www.wiley.com/college/boyce, provides a wealth of resources for students and instructors, including

• PowerPoint slides of important definitions, examples, and •

• •

theorems from the book, as well as graphics for presentation in lectures or for study and note taking. Chapter Review Sheets, which enable students to test their knowledge of key concepts. For further review, diagnostic feedback is provided that refers to pertinent sections in the text. Mathematica, Maple, and MATLAB data files for selected problems in the text providing opportunities for further exploration of important concepts. Projects that deal with extended problems normally not included among traditional topics in differential equations, many involving applications from a variety of disciplines. These vary in length and complexity, and they can be assigned as individual homework or as group assignments.

A series of supplemental guidebooks, also published by John Wiley & Sons, can be used with Boyce/DiPrima/Meade in order to incorporate computing technologies into the course. These books emphasize numerical methods and graphical analysis, showing how these methods enable us to interpret solutions of ordinary differential equations (ODEs) in the real world. Separate guidebooks cover each of the three major mathematical software formats, but the ODE subject matter is the same in each.

• Hunt, Lipsman, Osborn, and Rosenberg, Differential Equations with MATLAB , 3rd ed., 2012, ISBN 978-1-11837680-5

Boyce 9131 FM 2

October 12, 2016

17:16 ix

PREFACE

• Hunt, Lardy, Lipsman, Osborn, and Rosenberg, Differential •

Equations with Maple, 3rd ed., 2008, ISBN 978-0-47177317-7 Hunt, Outing, Lipsman, Osborn, and Rosenberg, Differential Equations with Mathematica, 3rd ed., 2009, ISBN 978-0-471-77316-0

WileyPLUS, is loaded with all of the supplements above, and it also features

• The E-book, which is an exact version of the print text •

WileyPLUS

WileyPLUS is an innovative, research-based online environment for effective teaching and learning. WileyPLUS builds students’ confidence because it takes the guesswork out of studying by providing students with a clear roadmap: what to do, how to do it, if they did it right. Students will take more initiative so you’ll have greater impact on their achievement in the classroom and beyond.

ix

• •

but also features hyperlinks to questions, definitions, and supplements for quicker and easier support. Guided Online (GO) Exercises, which prompt students to build solutions step-by-step. Rather than simply grading an exercise answer as wrong, GO problems show students precisely where they are making a mistake. Homework management tools, which enable instructors easily to assign and grade questions, as well as to gauge student comprehension. QuickStart pre-designed reading and homework assign ments. Use them as is, or customize them to fit the needs of your classroom.

Acknowledgments It is a pleasure to express my appreciation to the many people who have generously assisted in various ways in the preparation of this book. To the individuals listed below, who reviewed the manuscript and/or provided valuable suggestions for its improvement: Irina Gheorghiciuc, Carnegie Mellon University Bernard Brooks, Rochester Institute of Technology James Moseley, West Virginia University D. Glenn Lasseigne, Old Dominion University Stephen Summers, University of Florida Fabio Milner, Arizona State University Mohamed Boudjelkha, Rensselaer Polytechnic Institute Yuval Flicker, The Ohio State University Y. Charles Li, University of Missouri, Columbia Will Murray, California State University, Long Beach Yue Zhao, University of Central Florida Vladimir Shtelen, Rutgers University Zhilan Feng, Purdue University Mathew Johnson, University of Kansas Bulent Tosun, University of Alabama Juha Pohjanpelto, Oregon State University Patricia Diute, Rochester Institute of Technology Ning Ju, Oklahoma State University Ian Christie, West Virginia University Jonathan Rosenberg, University of Maryland Irina Kogan, North Carolina State University

To our colleagues and students at Rensselaer and The University of South Carolina, whose suggestions and reactions through the years have done much to sharpen our knowledge of differential equations, as well as our ideas on how to present the subject. To those readers of the preceding edition who called errors or omissions to our attention. To Tom Polaski (Winthrop University), who is primarily responsible for the revision of the Instructor’s Solutions Manual and the Student Solutions Manual. To Mark McKibben (West Chester University), who checked the answers in the back of the text and the Instructor’s Solutions Manual for accuracy, and carefully checked the entire manuscript. To the editorial and production staff of John Wiley & Sons, who have always been ready to offer assistance and have displayed the highest standards of professionalism. The last, but most important, people we want to thank are our wives: Elsa, for discussing questions both mathematical and stylistic and above all for her unfailing support and encouragement, and Betsy, for her encouragement, patience and understanding.

WILLIAM E. BOYCE AND DOUGLAS B. MEADE

Brief Contents PREFACE

vii

1

Introduction 1

2

First-Order Differential Equations 24

3

Second-Order Linear Differential Equations 103

4

Higher-Order Linear Differential Equations 169

5

Series Solutions of Second-Order Linear Equations 189

6

The Laplace Transform 241

7

Systems of First-Order Linear Equations 281

8

Numerical Methods 354

9

Nonlinear Differential Equations and Stability 388

10

Partial Differential Equations and Fourier Series 463

11

Boundary Value Problems and Sturm-Liouville Theory 529

ANSWERS TO PROBLEMS INDEX

x

606

573

Boyce 9131 FM 2

October 12, 2016

17:16 xi

Contents PREFACE

vii

4.3 4.4

1 Introduction 1.1 1.2 1.3

1

Some Basic Mathematical Models; Direction Fields 1 Solutions of Some Differential Equations 9 Classification of Differential Equations 16

2 First-Order Differential Equations 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9

3 Second-Order Linear Differential Equations 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8

103

Homogeneous Differential Equations with Constant Coefficients 103 Solutions of Linear Homogeneous Equations; the Wronskian 110 Complex Roots of the Characteristic Equation 120 Repeated Roots; Reduction of Order 127 Nonhomogeneous Equations; Method of Undetermined Coefficients 133 Variation of Parameters 142 Mechanical and Electrical Vibrations 147 Forced Periodic Vibrations 159

4 Higher-Order Linear Differential Equations 4.1 4.2

5 Series Solutions of Second-Order Linear Equations 5.1 5.2 5.3

24

Linear Differential Equations; Method of Integrating Factors 24 Separable Differential Equations 33 Modeling with First-Order Differential Equations 39 Differences Between Linear and Nonlinear Differential Equations 51 Autonomous Differential Equations and Population Dynamics 58 Exact Differential Equations and Integrating Factors 70 Numerical Approximations: Euler’s Method 76 The Existence and Uniqueness Theorem 83 First-Order Difference Equations 91

169

The Method of Undetermined Coefficients 181 The Method of Variation of Parameters 185

5.4 5.5 5.6 5.7

189

Review of Power Series 189 Series Solutions Near an Ordinary Point, Part I 195 Series Solutions Near an Ordinary Point, Part II 205 Euler Equations; Regular Singular Points 211 Series Solutions Near a Regular Singular Point, Part I 219 Series Solutions Near a Regular Singular Point, Part II 224 Bessel’s Equation 230

6 The Laplace Transform 6.1 6.2 6.3 6.4 6.5 6.6

241

Definition of the Laplace Transform 241 Solution of Initial Value Problems 248 Step Functions 257 Differential Equations with Discontinuous Forcing Functions 264 Impulse Functions 270 The Convolution Integral 275

7 Systems of First-Order Linear Equations 7.1 7.2 7.3 7.4 7.5 7.6 7.7 7.8 7.9

281

Introduction 281 Matrices 286 Systems of Linear Algebraic Equations; Linear Independence, Eigenvalues, Eigenvectors 295 Basic Theory of Systems of First-Order Linear Equations 304 Homogeneous Linear Systems with Constant Coefficients 309 Complex-Valued Eigenvalues 319 Fundamental Matrices 329 Repeated Eigenvalues 337 Nonhomogeneous Linear Systems 345

8 Numerical Methods

354

th

General Theory of n Order Linear Differential Equations 169 Homogeneous Differential Equations with Constant Coefficients 174

8.1 8.2

The Euler or Tangent Line Method 354 Improvements on the Euler Method 363 xi

xii

8.3 8.4 8.5 8.6

CONTENTS

The Runge-Kutta Method 367 Multistep Methods 371 Systems of First-Order Equations 376 More on Errors; Stability 378

9 Nonlinear Differential Equations and Stability 9.1 9.2 9.3 9.4 9.5 9.6 9.7 9.8

388

The Phase Plane: Linear Systems 388 Autonomous Systems and Stability 398 Locally Linear Systems 407 Competing Species 417 Predator-Prey Equations 428 Liapunov’s Second Method 435 Periodic Solutions and Limit Cycles 444 Chaos and Strange Attractors: The Lorenz Equations 454

10 Partial Differential Equations and Fourier Series 10.1 10.2 10.3

10.4 10.5 10.6 10.7 10.8

11 Boundary Value Problems and Sturm-Liouville Theory 11.1 11.2 11.3 11.4 11.5 11.6

463

Two-Point Boundary Value Problems 463 Fourier Series 469 The Fourier Convergence Theorem 477

Even and Odd Functions 482 Separation of Variables; Heat Conduction in a Rod 488 Other Heat Conduction Problems 496 The Wave Equation: Vibrations of an Elastic String 504 Laplace's Equation 514

The Occurrence of Two-Point Boundary Value Problems 529 Sturm-Liouville Boundary Value Problems 535 Nonhomogeneous Boundary Value Problems 545 Singular Sturm-Liouville Problems 556 Further Remarks on the Method of Separation of Variables: A Bessel Series Expansion 562 Series of Orthogonal Functions: Mean Convergence 566

ANSWERS TO PROBLEMS INDEX

529

606

573

Boyce 9131 Ch01 2

September 29, 2016

17:13

1

CHAPTER 1 Introduction In this first chapter we provide a foundation for your study of differential equations in several different ways. First, we use two problems to illustrate some of the basic ideas that we will return to, and elaborate upon, frequently throughout the remainder of the book. Later, to provide organizational structure for the book, we indicate several ways of classifying differential equations. The study of differential equations has attracted the attention of many of the world’s greatest mathematicians during the past three centuries. On the other hand, it is important to recognize that differential equations remains a dynamic field of inquiry today, with many interesting open questions. We outline some of the major trends in the historical development of the subject and mention a few of the outstanding mathematicians who have contributed to it. Additional biographical information about some of these contributors will be highlighted at appropriate times in later chapters.

Some Basic Mathematical Models; Direction Fields 1.1

Before embarking on a serious study of differential equations (for example, by reading this book or major portions of it), you should have some idea of the possible benefits to be gained by doing so. For some students the intrinsic interest of the subject itself is enough motivation, but for most it is the likelihood of important applications to other fields that makes the undertaking worthwhile. Many of the principles, or laws, underlying the behavior of the natural world are statements or relations involving rates at which things happen. When expressed in mathematical terms, the relations are equations and the rates are derivatives. Equations containing derivatives are differential equations. Therefore, to understand and to investigate problems involving the motion of fluids, the flow of current in electric circuits, the dissipation of heat in solid objects, the propagation and detection of seismic waves, or the increase or decrease of populations, among many others, it is necessary to know something about differential equations. A differential equation that describes some physical process is often called a mathematical model of the process, and many such models are discussed throughout this book. In this section we begin with two models leading to equations that are easy to solve. It is noteworthy that even the simplest differential equations provide useful models of important physical processes.

EXAMPLE 1 | A Falling Object

▼

Suppose that an object is falling in the atmosphere near sea level. Formulate a differential equation that describes the motion. 1

Boyce 9131 Ch01 2

2

September 29, 2016

17:13

2

CHAPTER 1 Introduction

▼ Solution: We begin by introducing letters to represent various quantities that may be of interest in this problem. The motion takes place during a certain time interval, so let us use t to denote time. Also, let us use v to represent the velocity of the falling object. The velocity will presumably change with time, so we think of v as a function of t; in other words, t is the independent variable and v is the dependent variable. The choice of units of measurement is somewhat arbitrary, and there is nothing in the statement of the problem to suggest appropriate units, so we are free to make any choice that seems reasonable. To be specific, let us measure time t in seconds and velocity v in meters/second. Further, we will assume that v is positive in the downward direction---that is, when the object is falling. The physical law that governs the motion of objects is Newton’s second law, which states that the mass of the object times its acceleration is equal to the net force on the object. In mathematical terms this law is expressed by the equation F = ma,

(1)

where m is the mass of the object, a is its acceleration, and F is the net force exerted on the object. To keep our units consistent, we will measure m in kilograms, a in meters/second2 , and F in newtons. Of course, a is related to v by a = dv/dt, so we can rewrite equation (1) in the form F =m

dv . dt

(2)

Next, consider the forces that act on the object as it falls. Gravity exerts a force equal to the weight of the object, or mg, where g is the acceleration due to gravity. In the units we have chosen, g has been determined experimentally to be approximately equal to 9.8 m/s2 near the earth’s surface. There is also a force due to air resistance, or drag, that is more difficult to model. This is not the place for an extended discussion of the drag force; suffice it to say that it is often assumed that the drag is proportional to the velocity, and we will make that assumption here. Thus the drag force has the magnitude γ v, where γ is a constant called the drag coefficient. The numerical value of the drag coefficient varies widely from one object to another; smooth streamlined objects have much smaller drag coefficients than rough blunt ones. The physical units for γ are mass/time, or kg/s for this problem; if these units seem peculiar, remember that γ v must have the units of force, namely, kg·m/s2 . In writing an expression for the net force F, we need to remember that gravity always acts in the downward (positive) direction, whereas, for a falling object, drag acts in the upward (negative) direction, as shown in Figure 1.1.1. Thus F = mg − γ v

(3)

dv = mg − γ v. dt

(4)

and equation (2) then becomes m

Differential equation (4) is a mathematical model for the velocity v of an object falling in the atmosphere near sea level. Note that the model contains the three constants m, g, and γ . The constants m and γ depend very much on the particular object that is falling, and they are usually different for different objects. It is common to refer to them as parameters, since they may take on a range of values during the course of an experiment. On the other hand, g is a physical constant, whose value is the same for all objects.

γυ m mg

FIGURE 1.1.1 Free-body diagram of the forces on a falling object.

Boyce 9131 Ch01 2

September 29, 2016

17:13

3

1.1 Some Basic Mathematical Models; Direction Fields

To solve equation (4), we need to find a function v = v( t) that satisfies the equation. It is not hard to do this, and we will show you how in the next section. For the present, however, let us see what we can learn about solutions without actually finding any of them. Our task is simplified slightly if we assign numerical values to m and γ , but the procedure is the same regardless of which values we choose. So, let us suppose that m = 10 kg and γ = 2 kg/s. Then equation (4) can be rewritten as dv v = 9.8 − . dt 5

(5)

EXAMPLE 2 | A Falling Object (continued) Investigate the behavior of solutions of equation (5) without solving the differential equation. Solution: First let us consider what information can be obtained directly from the differential equation itself. Suppose that the velocity v has a certain given value. Then, by evaluating the right-hand side of differential equation (5), we can find the corresponding value of dv/dt. For instance, if v = 40, then dv/dt = 1.8. This means that the slope of a solution v = v( t) has the value 1.8 at any point where v = 40. We can display this information graphically in the tv-plane by drawing short line segments with slope 1.8 at several points on the line v = 40. (See Figure 1.1.2(a)). Similarly, when v = 50, then dv/dt = −0.2, and when v = 60, then dv/dt = −2.2, so we draw line segments with slope −0.2 at several points on the line v = 50 (see Figure 1.1.2(b)) and line segments with slope −2.2 at several points on the line v = 60 (see Figure 1.1.2(c)). Proceeding in the same way with other values of v we create what is called a direction field, or a slope field. The direction field for differential equation (5) is shown in Figure 1.1.3. Remember that a solution of equation (5) is a function v = v( t) whose graph is a curve in the tv-plane. The importance of Figure 1.1.3 is that each line segment is a tangent line to one of these solution curves. Thus, even though we have not found any solutions, and no graphs of solutions appear in the figure, we can nonetheless draw some qualitative conclusions about the behavior of solutions. For instance, if v is less than a certain critical value, then all the line segments have positive slopes, and the speed of the falling object increases as it falls. On the other hand, if v is greater than the critical value, then the line segments have negative slopes, and the falling object slows down as it falls. What is this critical value of v that separates objects whose speed is increasing from those whose speed is decreasing? Referring again to equation (5), we ask what value of v will cause dv/dt to be zero. The answer is v = ( 5) ( 9.8) = 49 m/s. In fact, the constant function v( t) = 49 is a solution of equation (5). To verify this statement, substitute v( t) = 49 into equation (5) and observe that each side of the equation is zero. Because it does not change with time, the solution v( t) = 49 is called an equilibrium solution. It is the solution that corresponds to a perfect balance between gravity and drag. In Figure 1.1.3 we show the equilibrium solution v( t) = 49 superimposed on the direction field. From this figure we can draw another conclusion, namely, that all other solutions seem to be converging to the equilibrium solution as t increases. Thus, in this context, the equilibrium solution is often called the terminal velocity.

All slopes 1.8

All slopes –0.2

All slopes –2.2

υ 60

υ 60

υ 60

55

55

55

50

50

50

45

45

45

40

40 2

4

6

(a)

8

10 t

40 2

4

6

(b)

8

10 t

2

4

6

8

10 t

(c)

FIGURE 1.1.2 Assembling a direction field for equation (5): dv/dt = 9.8−v/5. (a) when v = 40,

dv/dt = 1.8, (b) when v = 50, dv/dt = −0.2, and (c) when v = 60, dv/dt = −2.2.

▼

3

Boyce 9131 Ch01 2

4

September 29, 2016

17:13

4

CHAPTER 1 Introduction

▼

υ 60

55

50

45

40 2

4

6

8

10

t

FIGURE 1.1.3 Direction field and equilibrium solution for equation (5):

dv/dt = 9.8 − v/5.

The approach illustrated in Example 2 can be applied equally well to the more general differential equation (4), where the parameters m and γ are unspecified positive numbers. The results are essentially identical to those of Example 2. The equilibrium solution of equation (4) is the constant solution v( t) = mg/γ . Solutions below the equilibrium solution increase with time, and those above it decrease with time. As a result, we conclude that all solutions approach the equilibrium solution as t becomes large. Direction Fields. Direction fields are valuable tools in studying the solutions of differential equations of the form dy (6) = f ( t, y) , dt where f is a given function of the two variables t and y, sometimes referred to as the rate function. A direction field for equations of the form (6) can be constructed by evaluating f at each point of a rectangular grid. At each point of the grid, a short line segment is drawn whose slope is the value of f at that point. Thus each line segment is tangent to the graph of the solution passing through that point. A direction field drawn on a fairly fine grid gives a good picture of the overall behavior of solutions of a differential equation. Usually a grid consisting of a few hundred points is sufficient. The construction of a direction field is often a useful first step in the investigation of a differential equation. Two observations are worth particular mention. First, in constructing a direction field, we do not have to solve equation (6); we just have to evaluate the given function f ( t, y) many times. Thus direction fields can be readily constructed even for equations that may be quite difficult to solve. Second, repeated evaluation of a given function and drawing a direction field are tasks for which a computer or other computational or graphical aid are well suited. All the direction fields shown in this book, such as the one in Figures 1.1.2 and 1.1.3, were computer generated. Field Mice and Owls. Now let us look at another, quite different example. Consider a population of field mice that inhabit a certain rural area. In the absence of predators we assume that the mouse population increases at a rate proportional to the current population. This assumption is not a well-established physical law (as Newton’s law of motion is in Example 1), but it is a common initial hypothesis1 in a study of population growth. If we denote time by t and the mouse population at time t by p( t) , then the assumption about population growth can be expressed by the equation dp (7) = r p, dt ......................................................................................................................................................................... 1A

better model of population growth is discussed in Section 2.5.

Boyce 9131 Ch01 2

September 29, 2016

17:13

5

1.1 Some Basic Mathematical Models; Direction Fields

where the proportionality factor r is called the rate constant or growth rate. To be specific, suppose that time is measured in months and that the rate constant r has the value 0.5/month. Then the two terms in equation (7) have the units of mice/month. Now let us add to the problem by supposing that several owls live in the same neighborhood and that they kill 15 field mice per day. To incorporate this information into the model, we must add another term to the differential equation (7), so that it becomes dp p (8) = − 450. dt 2 Observe that the predation term is −450 rather than −15 because time is measured in months, so the monthly predation rate is needed.

EXAMPLE 3 Investigate the solutions of differential equation (8) graphically. Solution: A direction field for equation (8) is shown in Figure 1.1.4. For sufficiently large values of p it can be seen from the figure, or directly from equation (8) itself, that dp/dt is positive, so that solutions increase. On the other hand, if p is small, then dp/dt is negative and solutions decrease. Again, the critical value of p that separates solutions that increase from those that decrease is the value of p for which dp/dt is zero. By setting dp/dt equal to zero in equation (8) and then solving for p, we find the equilibrium solution p( t) = 900, for which the growth term and the predation term in equation (8) are exactly balanced. The equilibrium solution is also shown in Figure 1.1.4. p 1000

950

900

850

800 1

2

3

4

5

t

FIGURE 1.1.4 Direction field (red) and equilibrium solution (blue) for

equation (8): dp/dt = p/2 − 450.

Comparing Examples 2 and 3, we note that in both cases the equilibrium solution separates increasing from decreasing solutions. In Example 2 other solutions converge to, or are attracted by, the equilibrium solution, so that after the object falls long enough, an observer will see it moving at very nearly the equilibrium velocity. On the other hand, in Example 3 other solutions diverge from, or are repelled by, the equilibrium solution. Solutions behave very differently depending on whether they start above or below the equilibrium solution. As time passes, an observer might see populations either much larger or much smaller than the equilibrium population, but the equilibrium solution itself will not, in practice, be observed. In both problems, however, the equilibrium solution is very important in understanding how solutions of the given differential equation behave. A more general version of equation (8) is dp = r p − k, dt

(9)

5

Boyce 9131 Ch01 2

6

September 29, 2016

17:13

6

CHAPTER 1 Introduction

where the growth rate r and the predation rate k are positive constants that are otherwise unspecified. Solutions of this more general equation are very similar to those of equation (8). The equilibrium solution of equation (9) is p( t) = k/ r . Solutions above the equilibrium solution increase, while those below it decrease. You should keep in mind that both of the models discussed in this section have their limitations. The model (5) of the falling object is valid only as long as the object is falling freely, without encountering any obstacles. If the velocity is large enough, the assumption that the frictional resistance is linearly proportional to the velocity has to be replaced with a nonlinear approximation (see Problem 21). The population model (8) eventually predicts negative numbers of mice (if p < 900) or enormously large numbers (if p > 900). Both of these predictions are unrealistic, so this model becomes unacceptable after a fairly short time interval. Constructing Mathematical Models. In applying differential equations to any of the numerous fields in which they are useful, it is necessary first to formulate the appropriate differential equation that describes, or models, the problem being investigated. In this section we have looked at two examples of this modeling process, one drawn from physics and the other from ecology. In constructing future mathematical models yourself, you should recognize that each problem is different, and that successful modeling cannot be reduced to the observance of a set of prescribed rules. Indeed, constructing a satisfactory model is sometimes the most difficult part of the problem. Nevertheless, it may be helpful to list some steps that are often part of the process: 1. Identify the independent and dependent variables and assign letters to represent them. Often the independent variable is time. 2. Choose the units of measurement for each variable. In a sense the choice of units is arbitrary, but some choices may be much more convenient than others. For example, we chose to measure time in seconds for the falling-object problem and in months for the population problem. 3. Articulate the basic principle that underlies or governs the problem you are investigating. This may be a widely recognized physical law, such as Newton’s law of motion, or it may be a more speculative assumption that may be based on your own experience or observations. In any case, this step is likely not to be a purely mathematical one, but will require you to be familiar with the field in which the problem originates. 4. Express the principle or law in step 3 in terms of the variables you chose in step 1. This may be easier said than done. It may require the introduction of physical constants or parameters (such as the drag coefficient in Example 1) and the determination of appropriate values for them. Or it may involve the use of auxiliary or intermediate variables that must then be related to the primary variables. 5. If the units agree, then your equation at least is dimensionally consistent, although it may have other shortcomings that this test does not reveal. 6. In the problems considered here, the result of step 4 is a single differential equation, which constitutes the desired mathematical model. Keep in mind, though, that in more complex problems the resulting mathematical model may be much more complicated, perhaps involving a system of several differential equations, for example. Historical Background, Part I: Newton, Leibniz, and the Bernoullis. Without knowing something about differential equations and methods of solving them, it is difficult to appreciate the history of this important branch of mathematics. Further, the development of differential equations is intimately interwoven with the general development of mathematics and cannot be separated from it. Nevertheless, to provide some historical perspective, we indicate here some of the major trends in the history of the subject and identify the most prominent early contributors. The rest of the historical background in this section focuses on the earliest contributors from the seventeenth century. The story continues at the end of Section 1.2 with an overview of the contributions of Euler and other eighteenth-century (and early-nineteenthcentury) mathematicians. More recent advances, including the use of computers and other

Boyce 9131 Ch01 2

September 29, 2016

17:13

7

1.1 Some Basic Mathematical Models; Direction Fields

technologies, are summarized at the end of Section 1.3. Additional historical information is contained in footnotes scattered throughout the book and in the references listed at the end of the chapter. The subject of differential equations originated in the study of calculus by Isaac Newton (1643--1727) and Gottfried Wilhelm Leibniz (1646--1716) in the seventeenth century. Newton grew up in the English countryside, was educated at Trinity College, Cambridge, and became Lucasian Professor of Mathematics there in 1669. His epochal discoveries of calculus and of the fundamental laws of mechanics date to 1665. They were circulated privately among his friends, but Newton was extremely sensitive to criticism and did not begin to publish his results until 1687 with the appearance of his most famous book Philosophiae Naturalis Principia Mathematica. Although Newton did relatively little work in differential equations as such, his development of the calculus and elucidation of the basic principles of mechanics provided a basis for their applications in the eighteenth century, most notably by Euler (see Historical Background, Part II in Section 1.2). Newton identified three forms of first-order differential equations: dy/d x = f ( x) , dy/d x = f ( y) , and dy/d x = f ( x, y) . For the latter equation he developed a method of solution using infinite series when f ( x, y) is a polynomial in x and y. Newton’s active research in mathematics ended in the early 1690s, except for the solution of occasional “challenge problems” and the revision and publication of results obtained much earlier. He was appointed Warden of the British Mint in 1696 and resigned his professorship a few years later. He was knighted in 1705 and, upon his death in 1727, became the first scientist buried in Westminster Abbey. Leibniz was born in Leipzig, Germany, and completed his doctorate in philosophy at the age of 20 at the University of Altdorf. Throughout his life he engaged in scholarly work in several different fields. He was mainly self-taught in mathematics, since his interest in this subject developed when he was in his twenties. Leibniz arrived at the fundamental results of calculus independently, although a little later than Newton, but was the first to publish them, in 1684. Leibniz was very conscious of the power of good mathematical notation and was responsible for the notation dy/d x for the derivative and for the integral sign. He discovered the method of separation of variables (Section 2.2) in 1691, the reduction of homogeneous equations to separable ones (Section 2.2, Problem 30) in 1691, and the procedure for solving first-order linear equations (Section 2.1) in 1694. He spent his life as ambassador and adviser to several German royal families, which permitted him to travel widely and to carry on an extensive correspondence with other mathematicians, especially the Bernoulli brothers. In the course of this correspondence many problems in differential equations were solved during the latter part of the seventeenth century. The Bernoulli brothers, Jakob (1654--1705) and Johann (1667--1748), of Basel, Switzerland did much to develop methods of solving differential equations and to extend the range of their applications. Jakob became professor of mathematics at Basel in 1687, and Johann was appointed to the same position upon his brother’s death in 1705. Both men were quarrelsome, jealous, and frequently embroiled in disputes, especially with each other. Nevertheless, both also made significant contributions to several areas of mathematics. With the aid of calculus, they solved a number of problems in mechanics by formulating them as differential equations. For example, Jakob Bernoulli solved the differential equation 1/2 y = a 3 /( b2 y − a 3 ) (see Problem 9 in Section 2.2) in 1690 and, in the same paper, first used the term “integral” in the modern sense. In 1694 Johann Bernoulli was able to solve the equation dy/d x = y/( ax) (see Problem 10 in Section 2.2). One problem that both brothers solved, and that led to much friction between them, was the brachistochrone problem (see Problem 24 in Section 2.3). The brachistochrone problem was also solved by Leibniz, Newton, and the Marquis de l’Hôpital. It is said, perhaps apocryphally, that Newton learned of the problem late in the afternoon of a tiring day at the Mint and solved it that evening after dinner. He published the solution anonymously, but upon seeing it, Johann Bernoulli exclaimed, “Ah, I know the lion by his paw.” Daniel Bernoulli (1700--1782), son of Johann, migrated to St. Petersburg, Russia, as a young man to join the newly established St. Petersburg Academy, but returned to Basel in 1733 as professor of botany and, later, of physics. His interests were primarily in partial differential equations and their applications. For instance, it is his name that is associated with the Bernoulli equation in fluid mechanics. He was also the first to encounter the functions that a century later became known as Bessel functions (Section 5.7).

7

Boyce 9131 Ch01 2

8

September 29, 2016

17:13

8

CHAPTER 1 Introduction

Problems In each of Problems 1 through 4, draw a direction field for the given differential equation. Based on the direction field, determine the behavior of y as t → ∞. If this behavior depends on the initial value of y at t = 0, describe the dependency.

12. The direction field of Figure 1.1.6. y 4

G

1. y = 3 − 2y

3

G

2. y = 2y − 3

2

G

3. y = −1 − 2y

G

4. y = 1 + 2y

1

In each of Problems 5 and 6, write down a differential equation of the form dy/dt = ay + b whose solutions have the required behavior as t → ∞.

5. All solutions approach y = 2/3. 6. All other solutions diverge from y = 2.

1

2

3

4 t

3

4

3

4t

3

4t

FIGURE 1.1.6 Problem 12.

13. The direction field of Figure 1.1.7.

In each of Problems 7 through 10, draw a direction field for the given differential equation. Based on the direction field, determine the behavior of y as t → ∞. If this behavior depends on the initial value of y at t = 0, describe this dependency. Note that in these problems the equations are not of the form y = ay + b, and the behavior of their solutions is somewhat more complicated than for the equations in the text.

y

1

2

t

–1 –2

G

7. y = y( 4 − y)

G

8. y = −y( 5 − y)

G

9. y = y 2

G

–3 –4

10. y = y( y − 2)

2

FIGURE 1.1.7 Problem 13.

Consider the following list of differential equations, some of which produced the direction fields shown in Figures 1.1.5 through 1.1.10. In each of Problems 11 through 16, identify the differential equation that corresponds to the given direction field.

a. b. c. d. e. f. g. h. i. j.

y y y y y y y y y y

= 2y − 1 =2+y = y−2 = y( y + 3) = y( y − 3) = 1 + 2y = −2 − y = y( 3 − y) = 1 − 2y =2−y

14. The direction field of Figure 1.1.8. y

1

2

–1 –2 –3 –4 FIGURE 1.1.8 Problem 14.

15. The direction field of Figure 1.1.9.

11. The direction field of Figure 1.1.5.

y 5 4

y 4 3

3 2

2

1

1 1

1

2

FIGURE 1.1.5 Problem 11.

3

4 t

2

–1 FIGURE 1.1.9 Problem 15.

Boyce 9131 Ch01 2

September 29, 2016

17:13

9

1.2 Solutions of Some Differential Equations

16. The direction field of Figure 1.1.10.

20. A certain drug is being administered intravenously to a hospital

y 5 4 3 2 1 1

2

3

4t

–1 FIGURE 1.1.10 Problem 16.

17. A pond initially contains 1,000,000 gal of water and an unknown amount of an undesirable chemical. Water containing 0.01 grams of this chemical per gallon flows into the pond at a rate of 300 gal/h. The mixture flows out at the same rate, so the amount of water in the pond remains constant. Assume that the chemical is uniformly distributed throughout the pond.

a. Write a differential equation for the amount of chemical in the pond at any time. b. How much of the chemical will be in the pond after a very long time? Does this limiting amount depend on the amount that was present initially? c. Write a differential equation for the concentration of the chemical in the pond at time t. Hint: The concentration is c = a/v = a( t) /106 . 18. A spherical raindrop evaporates at a rate proportional to its surface area. Write a differential equation for the volume of the raindrop as a function of time.

19. Newton’s law of cooling states that the temperature of an object changes at a rate proportional to the difference between the temperature of the object itself and the temperature of its surroundings (the ambient air temperature in most cases). Suppose that the ambient temperature is 70◦ F and that the rate constant is 0.05 (min)−1 . Write a differential equation for the temperature of the object at any time. Note that the differential equation is the same whether the temperature of the object is above or below the ambient temperature.

1.2

9

patient. Fluid containing 5 mg/cm3 of the drug enters the patient’s bloodstream at a rate of 100 cm3 /h. The drug is absorbed by body tissues or otherwise leaves the bloodstream at a rate proportional to the amount present, with a rate constant of 0.4/h. a. Assuming that the drug is always uniformly distributed throughout the bloodstream, write a differential equation for the amount of the drug that is present in the bloodstream at any time. b. How much of the drug is present in the bloodstream after a long time? N 21. For small, slowly falling objects, the assumption made in the text that the drag force is proportional to the velocity is a good one. For larger, more rapidly falling objects, it is more accurate to assume that the drag force is proportional to the square of the velocity.2 a. Write a differential equation for the velocity of a falling object of mass m if the magnitude of the drag force is proportional to the square of the velocity and its direction is opposite to that of the velocity. b. Determine the limiting velocity after a long time. c. If m = 10 kg, find the drag coefficient so that the limiting velocity is 49 m/s. N d. Using the data in part c, draw a direction field and compare it with Figure 1.1.3.

In each of Problems 22 through 25, draw a direction field for the given differential equation. Based on the direction field, determine the behavior of y as t → ∞. If this behavior depends on the initial value of y at t = 0, describe this dependency. Note that the right-hand sides of these equations depend on t as well as y; therefore, their solutions can exhibit more complicated behavior than those in the text. G

22. y = −2 + t − y

G

23. y = e−t + y

G

24. y = 3 sin t + 1 + y

G

25. y = −

2t + y 2y

............................................................................................................................. 2 See

Lyle N. Long and Howard Weiss, “The Velocity Dependence of Aerodynamic Drag: A Primer for Mathematicians,” American Mathematical Monthly 106 (1999), 2, pp. 127--135.

Solutions of Some Differential Equations

In the preceding section we derived the differential equations dv m (1) = mg − γ v dt and dp (2) = r p − k. dt Equation (1) models a falling object, and equation (2) models a population of field mice preyed on by owls. Both of these equations are of the general form dy = ay − b, (3) dt where a and b are given constants. We were able to draw some important qualitative conclusions about the behavior of solutions of equations (1) and (2) by considering the associated direction fields. To answer questions of a quantitative nature, however, we need to find the solutions themselves, and we now investigate how to do that.

Boyce 9131 Ch01 2

10

September 29, 2016

17:13

10

CHAPTER 1 Introduction

EXAMPLE 1 | Field Mice and Owls (continued) Consider the equation dp = 0.5 p − 450, dt

(4)

which describes the interaction of certain populations of field mice and owls (see equation (8) of Section 1.1). Find solutions of this equation. Solution: To solve equation (4), we need to find functions p( t) that, when substituted into the equation, reduce it to an obvious identity. Here is one way to proceed. First, rewrite equation (4) in the form dp p − 900 = , dt 2

(5)

1 dp/dt = . p − 900 2

(6)

or, if p = 900,

By the chain rule the left-hand side of equation (6) is the derivative of ln | p − 900| with respect to t, so we have 1 d ln | p − 900| = . dt 2

(7)

Then, by integrating both sides of equation (7), we obtain ln | p − 900| =

t + C, 2

(8)

where C is an arbitrary constant of integration. Therefore, by taking the exponential of both sides of equation (8), we find that | p − 900| = et/2+C = eC et/2 ,

(9)

p − 900 = ±eC et/2 ,

(10)

p = 900 + cet/2 ,

(11)

or

and finally

where c = ±e is also an arbitrary (nonzero) constant. Note that the constant function p = 900 is also a solution of equation (5) and that it is contained in the expression (11) if we allow c to take the value zero. Graphs of equation (11) for several values of c are shown in Figure 1.2.1. C

p 1200 1100 1000 900 800 700 600 1

2

3

4

5

t

FIGURE 1.2.1 Graphs of p = 900 + ce for several values of c. Each blue curve is a solution of dp/dt = 0.5 p − 450. t/2

▼

Boyce 9131 Ch01 2

September 29, 2016

17:13

11

1.2 Solutions of Some Differential Equations

▼

Note that they have the character inferred from the direction field in Figure 1.1.4. For instance, solutions lying on either side of the equilibrium solution p = 900 tend to diverge from that solution.

In Example 1 we found infinitely many solutions of the differential equation (4), corresponding to the infinitely many values that the arbitrary constant c in equation (11) might have. This is typical of what happens when you solve a differential equation. The solution process involves an integration, which brings with it an arbitrary constant, whose possible values generate an infinite family of solutions. Frequently, we want to focus our attention on a single member of the infinite family of solutions by specifying the value of the arbitrary constant. Most often, we do this indirectly by specifying instead a point that must lie on the graph of the solution. For example, to determine the constant c in equation (11), we could require that the population have a given value at a certain time, such as the value 850 at time t = 0. In other words, the graph of the solution must pass through the point ( 0, 850) . Symbolically, we can express this condition as p( 0) = 850.

(12)

Then, substituting t = 0 and p = 850 into equation (11), we obtain 850 = 900 + c. Hence c = −50, and by inserting this value into equation (11), we obtain the desired solution, namely, p = 900 − 50et/2 .

(13)

The additional condition (12) that we used to determine c is an example of an initial condition. The differential equation (4) together with the initial condition (12) forms an initial value problem. Now consider the more general problem consisting of the differential equation (3) dy = ay − b dt and the initial condition y( 0) = y0 ,

(14)

where y0 is an arbitrary initial value. We can solve this problem by the same method as in Example 1. If a = 0 and y = b/a, then we can rewrite equation (3) as dy/dt = a. y − ab By integrating both sides, we find that b ln y( t) − = at + C, a

(15)

(16)

where C is an arbitrary constant. Then, taking the exponential of both sides of equation (16) and solving for y, we obtain b y( t) = + ceat , (17) a where c = ±eC is also an arbitrary constant. Observe that c = 0 corresponds to the equilibrium solution y( t) = b/a. Finally, the initial condition (14) requires that c = y0 − ( b/a) , so the solution of the initial value problem (3), (14) is b b at y( t) = + y0 − (18) e . a a For a = 0 the expression (17) contains all possible solutions of equation (3) and is called the general solution.3 The geometric representation of the general solution (17) is an infinite family of curves called integral curves. Each integral curve is associated with a particular ......................................................................................................................................................................... = 0, then the solution of equation (3) is not given by equation (17). We leave it to you to find the general solution in this case.

3 If a

11

Boyce 9131 Ch01 2

12

September 29, 2016

17:13

12

CHAPTER 1 Introduction

value of c and is the graph of the solution corresponding to that value of c. Satisfying an initial condition amounts to identifying the integral curve that passes through the given initial point. To relate the solution (18) to equation (2), which models the field mouse population, we need only replace a by the growth rate r and replace b by the predation rate k; we assume that r > 0 and k > 0. Then the solution (18) becomes k k rt p( t) = + p0 − (19) e , r r where p0 is the initial population of field mice. The solution (19) confirms the conclusions reached on the basis of the direction field and Example 1. If p0 = k/ r , then from equation (19) it follows that p( t) = k/ r for all t; this is the constant, or equilibrium, solution. If p0 = k/ r , then the behavior of the solution depends on the sign of the coefficient p0 − k/ r of the exponential term in equation (19). If p0 > k/ r , then p grows exponentially with time t; if p0 < k/ r , then p decreases and becomes zero (at a finite time), corresponding to extinction of the field mouse population. Negative values of p, while possible for the expression (19), make no sense in the context of this particular problem. To put the falling-object equation (1) in the form (3), we must identify a with −γ / m and b with −g. Observe that assuming γ > 0 and m > 0 implies that a < 0 and b < 0. Making these substitutions in the solution (18), we obtain mg mg −γ t/ m v( t) = + v0 − e , (20) γ γ where v 0 is the initial velocity. Again, this solution confirms the conclusions reached in Section 1.1 on the basis of a direction field. There is an equilibrium, or constant, solution v( t) = mg/γ , and all other solutions tend to approach this equilibrium solution. The speed of convergence to the equilibrium solution is determined by the exponent −γ / m. Thus, for a given mass m, the velocity approaches the equilibrium value more rapidly as the drag coefficient γ increases.

EXAMPLE 2 | A Falling Object (continued) Suppose that, as in Example 2 of Section 1.1, we consider a falling object of mass m = 10 kg and drag coefficient γ = 2 kg/s. Then the equation of motion (1) becomes dv v = 9.8 − . dt 5

(21)

Suppose this object is dropped from a height of 300 m. Find its velocity at any time t. How long will it take to fall to the ground, and how fast will it be moving at the time of impact? Solution: The first step is to state an appropriate initial condition for equation (21). The word “dropped” in the statement of the problem suggests that the object starts from rest, that is, its initial velocity is zero, so we will use the initial condition v( 0) = 0.

(22)

The solution of equation (21) can be found by substituting the values of the coefficients into the solution (20), but we will proceed instead to solve equation (21) directly. First, rewrite the equation as

By integrating both sides, we obtain

dv/dt 1 =− . v − 49 5

(23)

t lnv( t) − 49 = − + C, 5

(24)

and then the general solution of equation (21) is v( t) = 49 + ce−t/5 ,

▼

(25)

where the constant c is arbitrary. To determine the particular value of c that corresponds to the initial condition (22), we substitute t = 0 and v = 0 into equation (25), with the result that c = −49. Then

Boyce 9131 Ch01 2

September 29, 2016

17:13

13

1.2 Solutions of Some Differential Equations

▼ the solution of the initial value problem (21), (22) is

v( t) = 49 1 − e−t/5 .

(26)

Equation (26) gives the velocity of the falling object at any positive time after being dropped---until it hits the ground, of course. Graphs of the solution (25) for several values of c are shown in Figure 1.2.2, with the solution (26) shown by the green curve. It is evident that, regardless of the initial velocity of the object, all solutions tend to approach the equilibrium solution v( t) = 49. This confirms the conclusions we reached in Section 1.1 on the basis of the direction fields in Figures 1.1.2 and 1.1.3. v 100 80 60 40

(10.51, 43.01) v = 49 (1 – e–t/5)

20

2

4

6

8

10

12

t

FIGURE 1.2.2 Graphs of the solution (25), v = 49 + ce−t/5 , for several values

of c. The green curve corresponds to the initial condition v( 0) = 0. The point ( 10.51, 43.01) shows the velocity when the object hits the ground.

To find the velocity of the object when it hits the ground, we need to know the time at which impact occurs. In other words, we need to determine how long it takes the object to fall 300 m. To do this, we note that the distance x the object has fallen is related to its velocity v by the differential equation v = d x/dt, or

dx = 49 1 − e−t/5 . dt Consequently, by integrating both sides of equation (27) with respect to t, we have x = 49t + 245e−t/5 + k,

(27)

(28)

where k is an arbitrary constant of integration. The object starts to fall when t = 0, so we know that x = 0 when t = 0. From equation (28) it follows that k = −245, so the distance the object has fallen at time t is given by x = 49t + 245e−t/5 − 245.

(29)

Let T be the time at which the object hits the ground; then x = 300 when t = T . By substituting these values in equation (29), we obtain the equation 49T + 245e−T /5 − 245 = 300. The value of T satisfying equation (30) can be approximated by a a calculator or other computational tool, with the result that T ∼ = corresponding velocity v T is found from equation (26) to be v T ( 10.51, 43.01) is also shown in Figure 1.2.2.

(30) numerical process4 using 10.51 s. At this time, the ∼ = 43.01m/s. The point

......................................................................................................................................................................... 4A

computer algebra system provides this capability; many calculators also have built-in routines for solving such equations.

13

Boyce 9131 Ch01 2

14

September 29, 2016

17:13

14

CHAPTER 1 Introduction

Further Remarks on Mathematical Modeling. Up to this point we have related our discussion of differential equations to mathematical models of a falling object and of a hypothetical relation between field mice and owls. The derivation of these models may have been plausible, and possibly even convincing, but you should remember that the ultimate test of any mathematical model is whether its predictions agree with observations or experimental results. We have no actual observations or experimental results to use for comparison purposes here, but there are several sources of possible discrepancies. In the case of the falling object, the underlying physical principle (Newton’s laws of motion) is well established and widely applicable. However, the assumption that the drag force is proportional to the velocity is less certain. Even if this assumption is correct, the determination of the drag coefficient γ by direct measurement presents difficulties. Indeed, sometimes one finds the drag coefficient indirectly---for example, by measuring the time of fall from a given height and then calculating the value of γ that predicts this observed time. The model of the field mouse population is subject to various uncertainties. The determination of the growth rate r and the predation rate k depends on observations of actual populations, which may be subject to considerable variation. The assumption that r and k are constants may also be questionable. For example, a constant predation rate becomes harder to sustain as the field mouse population becomes smaller. Further, the model predicts that a population above the equilibrium value will grow exponentially larger and larger. This seems at variance with the behavior of actual populations; see the further discussion of population dynamics in Section 2.5. If the differences between actual observations and a mathematical model’s predictions are too great, then you need to consider refining the model, making more careful observations, or perhaps both. There is almost always a tradeoff between accuracy and simplicity. Both are desirable, but a gain in one usually involves a loss in the other. However, even if a mathematical model is incomplete or somewhat inaccurate, it may nevertheless be useful in explaining qualitative features of the problem under investigation. It may also give satisfactory results under some circumstances but not others. Thus you should always use good judgment and common sense in constructing mathematical models and in using their predictions. Historical Background, Part II: Euler, Lagrange, and Laplace. The greatest mathematician of the eighteenth century, Leonhard Euler (1707--1783), grew up near Basel, Switzerland and was a student of Johann Bernoulli. He followed his friend Daniel Bernoulli to St. Petersburg in 1727. For the remainder of his life he was associated with the St. Petersburg Academy (1727--1741 and 1766--1783) and the Berlin Academy (1741--1766). Losing sight in his right eye in 1738, and in his left eye in 1766, did not stop Euler from being one of the most prolific mathematicians of all time. In addition to publishing more than 500 books and papers during his life, an additional 400 have appeared posthumously. Of particular interest here is Euler’s formulation of problems in mechanics in mathematical language and his development of methods of solving these mathematical problems. Lagrange said of Euler’s work in mechanics, “The first great work in which analysis is applied to the science of movement.” Among other things, Euler identified the condition for exactness of first-order differential equations (Section 2.6) in 1734--1735, developed the theory of integrating factors (Section 2.6) in the same paper, and gave the general solution of homogeneous linear differential equations with constant coefficients (Sections 3.1, 3.3, 3.4, and 4.2) in 1743. He extended the latter results to nonhomogeneous differential equations in 1750--1751. Beginning about 1750, Euler made frequent use of power series (Chapter 5) in solving differential equations. He also proposed a numerical procedure (Sections 2.7 and 8.1) in 1768--1769, made important contributions in partial differential equations, and gave the first systematic treatment of the calculus of variations. Joseph-Louis Lagrange (1736--1813) became professor of mathematics in his native Turin, Italy, at the age of 19. He succeeded Euler in the chair of mathematics at the Berlin Academy in 1766 and moved on to the Paris Academy in 1787. He is most famous for his monumental work Mécanique analytique, published in 1788, an elegant and comprehensive treatise of Newtonian mechanics. With respect to elementary differential equations, Lagrange showed in 1762--1765 that the general solution of a homogeneous nth order linear differential equation is a linear combination of n independent solutions (Sections 3.2 and 4.1). Later, in 1774--1775, he offered a complete development of the method of variation of parameters (Sections 3.6 and 4.4). Lagrange is also known for fundamental work in partial differential equations and the calculus of variations.

Boyce 9131 Ch01 2

September 29, 2016

17:13

15

1.2 Solutions of Some Differential Equations

15

Pierre-Simon de Laplace (1749--1827) lived in Normandy, France, as a boy but arrived in Paris in 1768 and quickly made his mark in scientific circles, winning election to the Académie des Sciences in 1773. He was preeminent in the field of celestial mechanics; his greatest work, Traité de mécanique céleste, was published in five volumes between 1799 and 1825. Laplace’s equation is fundamental in many branches of mathematical physics, and Laplace studied it extensively in connection with gravitational attraction. The Laplace transform (Chapter 6) is also named for him, although its usefulness in solving differential equations was not recognized until much later. By the end of the eighteenth century many elementary methods of solving ordinary differential equations had been discovered. In the nineteenth century interest turned more toward the investigation of theoretical questions of existence and uniqueness and to the development of less elementary methods such as those based on power series expansions (see Chapter 5). These methods find their natural setting in the complex plane. Consequently, they benefitted from, and to some extent stimulated, the more or less simultaneous development of the theory of complex analytic functions. Partial differential equations also began to be studied intensively, as their crucial role in mathematical physics became clear. In this connection a number of functions, arising as solutions of certain ordinary differential equations, occurred repeatedly and were studied exhaustively. Known collectively as higher transcendental functions, many of them are associated with the names of mathematicians, including Bessel (Section 5.7), Legendre (Section 5.3), Hermite (Section 5.2), Chebyshev (Section 5.3), Hankel, and many others.

Problems N 1. Solve each of the following initial value problems and plot the solutions for several values of y0 . Then describe in a few words how the solutions resemble, and differ from, each other. a. dy/dt = −y + 5, y( 0) = y0 b. dy/dt = −2y + 5, y( 0) = y0 c. dy/dt = −2y + 10, y( 0) = y0 G

2. Follow the instructions for Problem 1 for the following initial-value problems: a. dy/dt = y − 5, y( 0) = y0 G b. dy/dt = 2y − 5, y( 0) = y0 c. dy/dt = 2y − 10, y( 0) = y0 3. Consider the differential equation dy/dt = −ay + b, where both a and b are positive numbers. a. Find the general solution of the differential equation. G b. Sketch the solution for several different initial conditions. c. Describe how the solutions change under each of the following conditions: i. a increases. ii. b increases. iii. Both a and b increase, but the ratio b/a remains the same.

4. Consider the differential equation dy/dt = ay − b. a. Find the equilibrium solution ye . b. Let Y ( t) = y − ye ; thus Y ( t) is the deviation from the equilibrium solution. Find the differential equation satisfied by Y ( t) .

5. Undetermined Coefficients. Here is an alternative way to solve the equation dy = ay − b. dt a. Solve the simpler equation dy = ay. dt

(31)

(32)

Call the solution y1 ( t) .

b. Observe that the only difference between equations (31) and (32) is the constant −b in equation (31). Therefore, it may seem reasonable to assume that the solutions of these two equations also differ only by a constant. Test this assumption by trying to find a constant k such that y = y1 ( t) + k is a solution of equation (31). c. Compare your solution from part b with the solution given in the text in equation (17). Note: This method can also be used in some cases in which the constant b is replaced by a function g( t) . It depends on whether you can guess the general form that the solution is likely to take. This method is described in detail in Section 3.5 in connection with secondorder equations.

6. Use the method of Problem 5 to solve the equation dy = −ay + b. dt 7. The field mouse population in Example 1 satisfies the differential equation p dy = − 450. dt 2 a. Find the time at which the population becomes extinct if p( 0) = 850. b. Find the time of extinction if p( 0) = p0 , where 0 < p0 < 900. N c. Find the initial population p0 if the population is to become extinct in 1 year.

8. The falling object in Example 2 satisfies the initial value problem v dv = 9.8 − , v( 0) = 0. dt 5 a. Find the time that must elapse for the object to reach 98% of its limiting velocity. b. How far does the object fall in the time found in part a?

Boyce 9131 Ch01 2

16

September 29, 2016

17:13

16

CHAPTER 1 Introduction

9. Consider the falling object of mass 10 kg in Example 2, but assume now that the drag force is proportional to the square of the velocity. a. If the limiting velocity is 49 m/s (the same as in Example 2), show that the equation of motion can be written as dv 1 2 49 − v 2 . = dt 245 Also see Problem 21 of Section 1.1. b. If v( 0) = 0, find an expression for v( t) at any time. G c. Plot your solution from part b and the solution (26) from Example 2 on the same axes. d. Based on your plots in part c, compare the effect of a quadratic drag force with that of a linear drag force. e. Find the distance x( t) that the object falls in time t. N f. Find the time T it takes the object to fall 300 m. 10. A radioactive material, such as the isotope thorium-234, disintegrates at a rate proportional to the amount currently present. If Q( t) is the amount present at time t, then d Q/dt = −r Q, where r > 0 is the decay rate. a. If 100 mg of thorium-234 decays to 82.04 mg in 1 week, determine the decay rate r . b. Find an expression for the amount of thorium-234 present at any time t. c. Find the time required for the thorium-234 to decay to onehalf its original amount.

11. The half-life of a radioactive material is the time required for an amount of this material to decay to one-half its original value. Show that for any radioactive material that decays according to the equation Q = −r Q, the half-life τ and the decay rate r satisfy the equation r τ = ln 2.

12. According to Newton’s law of cooling (see Problem 19 of Section 1.1), the temperature u( t) of an object satisfies the differential equation du = −k( u − T ) , dt where T is the constant ambient temperature and k is a positive constant. Suppose that the initial temperature of the object is u( 0) = u 0 . a. Find the temperature of the object at any time. b. Let τ be the time at which the initial temperature difference u 0 − T has been reduced by half. Find the relation between k and τ .

13. Consider an electric circuit containing a capacitor, resistor, and

1.3

battery; see Figure 1.2.3. The charge Q( t) on the capacitor satisfies the equation5 Q dQ + = V, dt C where R is the resistance, C is the capacitance, and V is the constant voltage supplied by the battery. G a. If Q( 0) = 0, find Q( t) at any time t, and sketch the graph of Q versus t. b. Find the limiting value Q L that Q( t) approaches after a long time. G c. Suppose that Q( t1 ) = Q L and that at time t = t1 the battery is removed and the circuit is closed again. Find Q( t) for t > t1 and sketch its graph. R

R V C FIGURE 1.2.3 The electric circuit of Problem 13.

N 14. A pond containing 1,000,000 gal of water is initially free of a certain undesirable chemical (see Problem 17 of Section 1.1). Water containing 0.01 g/gal of the chemical flows into the pond at a rate of 300 gal/h, and water also flows out of the pond at the same rate. Assume that the chemical is uniformly distributed throughout the pond. a. Let Q( t) be the amount of the chemical in the pond at time t. Write down an initial value problem for Q( t) . b. Solve the problem in part a for Q( t) . How much chemical is in the pond after 1 year? c. At the end of 1 year the source of the chemical in the pond is removed; thereafter pure water flows into the pond, and the mixture flows out at the same rate as before. Write down the initial value problem that describes this new situation. d. Solve the initial value problem in part c. How much chemical remains in the pond after 1 additional year (2 years from the beginning of the problem)? e. How long does it take for Q( t) to be reduced to 10 g? G f. Plot Q( t) versus t for 3 years.

.............................................................................................................................. 5 This equation results from Kirchhoff’s laws, which are discussed in Section 3.7.

Classification of Differential Equations

The main purposes of this book are to discuss some of the properties of solutions of differential equations and to present some of the methods that have proved effective in finding solutions or, in some cases, in approximating them. To provide a framework for our presentation, we describe here several useful ways of classifying differential equations. Mastery of this vocabulary is essential to selecting appropriate solution methods and to describing properties of solutions of differential equations that you encounter later in this book---and in the real world. Ordinary and Partial Differential Equations. One important classification is based on whether the unknown function depends on a single independent variable or on several

Boyce 9131 Ch01 2

September 29, 2016

17:13

17

1.3 Classification of Differential Equations

independent variables. In the first case, only ordinary derivatives appear in the differential equation, and it is said to be an ordinary differential equation. In the second case, the derivatives are partial derivatives, and the equation is called a partial differential equation. All the differential equations discussed in the preceding two sections are ordinary differential equations. Another example of an ordinary differential equation is L

d 2 Q( t) d Q( t) 1 +R + Q( t) = E( t) , dt C dt 2

(1)

for the charge Q( t) on a capacitor in a circuit with capacitance C, resistance R, and inductance L; this equation is derived in Section 3.7. Typical examples of partial differential equations are the heat conduction equation α2

∂ 2 u( x, t) ∂ u( x, t) = 2 ∂t ∂x

(2)

and the wave equation ∂ 2 u( x, t) ∂ 2 u( x, t) = . (3) ∂ x2 ∂ t2 Here, α 2 and a 2 are certain physical constants. Note that in both equations (2) and (3) the dependent variable u depends on the two independent variables x and t. The heat conduction equation describes the conduction of heat in a solid body, and the wave equation arises in a variety of problems involving wave motion in solids or fluids. a2

Systems of Differential Equations. Another classification of differential equations depends on the number of unknown functions that are involved. If there is a single function to be determined, then one differential equation is sufficient. However, if there are two or more unknown functions, then a system of differential equations is required. For example, the Lotka-Volterra, or predator-prey, equations are important in ecological modeling. They have the form dx = ax − α x y dt dy = −cy + γ x y, dt

(4)

where x( t) and y( t) are the respective populations of the prey and predator species. The positive constants a, α, c, and γ are based on empirical observations and depend on the particular species being studied. Systems of equations are discussed in Chapters 7 and 9; in particular, the Lotka-Volterra equations are examined in Section 9.5. In some areas of application it is not unusual to encounter very large systems containing hundreds, or even many thousands, of differential equations. Order. The order of a differential equation is the order of the highest derivative that appears in the equation. The equations in the preceding sections are all first-order equations, whereas equation (1) is a second-order equation. Equations (2) and (3) are also second-order partial differential equations. More generally, the equation F t, u( t) , u ( t) , . . . , u ( n) ( t) = 0 (5) is an ordinary differential equation of the n th order. Equation (5) expresses a relation between the independent variable t and the values of the function u and its first n derivatives u , u , . . . , u ( n) . It is convenient and customary in differential equations to write y for u( t) , with y , y , . . . , y ( n) standing for u ( t) , u ( t) , . . . , u ( n) ( t) . Thus equation (5) is written as (6) F t, y, y , . . . , y ( n) = 0. For example, y + 2et y + yy = t 4

(7)

is a third-order differential equation for y = u( t) . Occasionally, other letters will be used instead of t and y for the independent and dependent variables; the meaning should be clear from the context.

17

Boyce 9131 Ch01 2

18

September 29, 2016

17:13

18

CHAPTER 1 Introduction

We assume that it is always possible to solve a given ordinary differential equation for the highest derivative, obtaining y ( n) = f t, y, y , y , . . . , y ( n−1) . (8) This is mainly to avoid the ambiguity that may arise because a single equation of the form (6) may correspond to several equations of the form (8). For example, the equation ( y ) 2 + t y + 4y = 0 leads to the two equations

y =

−t +

t 2 − 16y −t − or y = 2

(9)

t 2 − 16y . 2

(10)

Linear and Nonlinear Equations. A crucial classification of differential equations is whether they are linear or nonlinear. The ordinary differential equation F t, y, y , . . . , y ( n) = 0 is said to be linear if F is a linear function of the variables y, y , . . . , y ( n) ; a similar definition applies to partial differential equations. Thus the general linear ordinary differential equation of order n is a0 ( t) y ( n) + a1 ( t) y ( n−1) + · · · + an ( t) y = g( t) .

(11)

Most of the equations you have seen thus far in this book are linear; examples are the equations in Sections 1.1 and 1.2 describing the falling object and the field mouse population. Similarly, in this section, equation (1) is a linear ordinary differential equation and equations (2) and (3) are linear partial differential equations. An equation that is not of the form (11) is a nonlinear equation. Equation (7) is nonlinear because of the term yy . Similarly, each equation in the system (4) is nonlinear because of the terms that involve the product of the two unknown functions x y. A simple physical problem that leads to a nonlinear differential equation is the oscillating pendulum. The angle θ = θ ( t) that an oscillating pendulum of length L makes with the vertical direction (see Figure 1.3.1) satisfies the equation d 2θ g + sin θ = 0, (12) 2 L dt whose derivation is outlined in Problems 22 through 24. The presence of the term involving sin θ makes equation (12) nonlinear.

θ

L

m

mg FIGURE 1.3.1 An oscillating pendulum.

The mathematical theory and methods for solving linear equations are highly developed. In contrast, for nonlinear equations the theory is more complicated, and methods of solution are less satisfactory. In view of this, it is fortunate that many significant problems lead to linear ordinary differential equations or can be approximated by linear equations. For example, for the pendulum, if the angle θ is small, then sin θ ∼ = θ and equation (12) can be approximated by the linear equation d 2θ g + θ = 0. (13) L dt 2 This process of approximating a nonlinear equation by a linear one is called linearization; it is an extremely valuable way to deal with nonlinear equations. Nevertheless, there are many

Boyce 9131 Ch01 2

September 29, 2016

17:13

19

1.3 Classification of Differential Equations

physical phenomena that simply cannot be represented adequately by linear equations. To study these phenomena, it is essential to deal with nonlinear equations. In an elementary text it is natural to emphasize the simpler and more straightforward parts of the subject. Therefore, the greater part of this book is devoted to linear equations and various methods for solving them. However, Chapters 8 and 9, as well as parts of Chapter 2, are concerned with nonlinear equations. Whenever it is appropriate, we point out why nonlinear equations are, in general, more difficult and why many of the techniques that are useful in solving linear equations cannot be applied to nonlinear equations. Solutions. A solution of the n th order ordinary differential equation (8) on the interval α < t < β is a function φ such that φ , φ , . . . , φ ( n) exist and satisfy φ ( n) ( t) = f t, φ ( t) , φ ( t) , . . . , φ ( n−1) ( t) (14) for every t in α < t < β . Unless stated otherwise, we assume that the function f of equation (8) is a real-valued function, and we are interested in obtaining real-valued solutions y = φ ( t) . Recall that in Section 1.2 we found solutions of certain equations by a process of direct integration. For instance, we found that the equation dp p = − 450 dt 2

(15)

p( t) = 900 + cet/2 ,

(16)

has the solution

where c is an arbitrary constant. It is often not so easy to find solutions of differential equations. However, if you find a function that you think may be a solution of a given equation, it is usually relatively easy to determine whether the function is actually a solution: just substitute the function into the equation. For example, in this way it is easy to show that the function y1 ( t) = cos t is a solution of y + y = 0

(17)

for all t. To confirm this, observe that y1 ( t) = −sin t and y1 ( t) = −cos t; then it follows that y1 ( t) + y1 ( t) = 0. In the same way you can easily show that y2 ( t) = sin t is also a solution of equation (17). Of course, this does not constitute a satisfactory way to solve most differential equations, because there are far too many possible functions for you to have a good chance of finding the correct one by a random choice. Nevertheless, you should realize that you can verify whether any proposed solution is correct by substituting it into the differential equation. This can be a very useful check; it is one that you should make a habit of considering. Some Important Questions. Although for the differential equations (15) and (17) we are able to verify that certain simple functions are solutions, in general we do not have such solutions readily available. Thus a fundamental question is the following: Does an equation of the form (8) always have a solution? The answer is “No.” Merely writing down an equation of the form (8) does not necessarily mean that there is a function y = φ ( t) that satisfies it. So, how can we tell whether some particular equation has a solution? This is the question of existence of a solution, and it is answered by theorems stating that under certain restrictions on the function f in equation (8), the equation always has solutions. This is not a purely theoretical concern for at least two reasons. If a problem has no solution, we would prefer to know that fact before investing time and effort in a vain attempt to solve the problem. Further, if a sensible physical problem is modeled mathematically as a differential equation, then the equation should have a solution. If it does not, then presumably there is something wrong with the formulation. In this sense an engineer or scientist has some check on the validity of the mathematical model. If we assume that a given differential equation has at least one solution, then we may need to consider how many solutions it has, and what additional conditions must be specified to single out a particular solution. This is the question of uniqueness. In general, solutions

19

Boyce 9131 Ch01 2

20

September 29, 2016

17:13

20

CHAPTER 1 Introduction

of differential equations contain one or more arbitrary constants of integration, as does the solution (16) of equation (15). Equation (16) represents an infinity of functions corresponding to the infinity of possible choices of the constant c. As we saw in Section 1.2, if p is specified at some time t, this condition will determine a specific value for c; even so, we have not yet ruled out the possibility that there may be other solutions of equation (15) that also have the prescribed value of p at the prescribed time t. As in the question of existence of solutions, the issue of uniqueness has practical as well as theoretical implications. If we are fortunate enough to find a solution of a given problem, and if we know that the problem has a unique solution, then we can be sure that we have completely solved the problem. If there may be other solutions, then perhaps we should continue to search for them. A third important question is: Given a differential equation of the form (8), can we actually determine a solution, and if so, how? Note that if we find a solution of the given equation, we have at the same time answered the question of the existence of a solution. However, without knowledge of existence theory we might, for example, use a computer to find a numerical approximation to a “solution” that does not exist. On the other hand, even though we may know that a solution exists, it may be that the solution is not expressible in terms of the usual elementary functions---polynomial, trigonometric, exponential, logarithmic, and hyperbolic functions. Unfortunately, this is the situation for most differential equations. Thus, we discuss both elementary methods that can be used to obtain exact solutions of certain relatively simple problems, and also methods of a more general nature that can be used to find approximations to solutions of more difficult problems. Technology Use in Differential Equations. Technology provides many extremely valuable tools for the study of differential equations. For many years computers have been used to execute numerical algorithms, such as those described in Chapter 8, to construct numerical approximations to solutions of differential equations. These algorithms have been refined to an extremely high level of generality and efficiency. A few lines of computer code, written in a high-level programming language and executed (often within a few seconds) on a relatively inexpensive computer, suffice to approximate to a high degree of accuracy the solutions of a wide range of differential equations. More sophisticated routines are also readily available. These routines combine the ability to handle very large and complicated systems with numerous diagnostic features that alert the user to possible problems as they are encountered. The usual output from a numerical algorithm is a table of numbers, listing selected values of the independent variable and the corresponding values of the dependent variable. With appropriate software it is easy to display the solution of a differential equation graphically, whether the solution has been obtained numerically or as the result of an analytical procedure of some kind. Such a graphical display is often much more illuminating and helpful in understanding and interpreting the solution of a differential equation than a table of numbers or a complicated analytical formula. There are on the market several well-crafted and relatively inexpensive special-purpose software packages for the graphical investigation of differential equations. The increased power and sophistication of modern smartphones, tablets, and other mobile devices have brought powerful computational and graphical capability within the reach of individual students. You should consider, in the light of your own circumstances, how best to take advantage of the available computing resources. You will surely find it enlightening to do so. Another aspect of computer use that is very relevant to the study of differential equations is the availability of extremely powerful and general software packages that can perform a wide variety of mathematical operations. Among these are Maple, Mathematica, and MATLAB, each of which can be used on various kinds of personal computers or workstations. All three of these packages can execute extensive numerical computations and have versatile graphical facilities. Maple and Mathematica also have very extensive analytical capabilities. For example, they can perform the analytical steps involved in solving many differential equations, often in response to a single command. Anyone who expects to deal with differential equations in more than a superficial way should become familiar with at least one of these products and explore the ways in which it can be used. For you, the student, these computing resources have an effect on how you should study differential equations. To become confident in using differential equations, it is essential to understand how the solution methods work, and this understanding is achieved, in part, by

Boyce 9131 Ch01 2

September 29, 2016

17:13

21

1.3 Classification of Differential Equations

working out a sufficient number of examples in detail. However, eventually you should plan to utilize appropriate computational tools to complete as many as possible of the routine (often repetitive) details, while you focus on the proper formulation of the problem and on the interpretation of the solution. Our viewpoint is that you should always try to use the best methods and tools available for each task. In particular, you should strive to combine numerical, graphical, and analytical methods so as to attain maximum understanding of the behavior of the solution and of the underlying process that the problem models. You should also remember that some tasks can best be done with pencil and paper, while others require the use of some sort of computational technology. Good judgment is often needed in selecting an effective combination. Historical Background, Part III: Recent and Ongoing Advances. The numerous differential equations that resisted solution by analytical means led to the investigation of methods of numerical approximation (see Chapter 8). By 1900 fairly effective numerical integration methods had been devised, but their implementation was severely restricted by the need to execute the computations by hand or with very primitive computing equipment. Since World War II the development of increasingly powerful and versatile computers has vastly enlarged the range of problems that can be investigated effectively by numerical methods. Extremely refined and robust numerical integrators were developed during the same period and now are readily available, even on smartphones and other mobile devices. These technological advances have brought the ability to solve a great many significant problems within the reach of individual students. Another characteristic of modern differential equations is the creation of geometric or topological methods, especially for nonlinear equations. The goal is to understand at least the qualitative behavior of solutions from a geometrical, as well as from an analytical, point of view. If more detailed information is needed, it can usually be obtained by using numerical approximations. An introduction to geometric methods appears in Chapter 9. We conclude this brief historical review with two examples that illustrate how computational and real-world experiences have motivated important analytical and theoretical discoveries. In 1834 John Scott Russell (1808--1882), a Scottish civil engineer, was conducting experiments to determine the most efficient design for canal boats when he noticed that “when the boat suddenly stopped” the water being pushed by the boat “accumulated round the prow of the vessel in a state of violent agitation, then suddenly leaving it [the boat] behind, [the water] rolled forward with great velocity, assuming the form of a large solitary elevation, a rounded, smooth and well-defined heap of water.” 6 Many mathematicians did not believe that the solitary traveling waves reported by Russell existed. These objections were silenced when the doctoral dissertation of Dutch mathematician Gustav de Vries (1866--1934) included a nonlinear partial differential equation model for water waves in a shallow canal. Today these equations are known as the Korteweg-de Vries (KdV) equations. Diederik Johannes Korteweg (1848--1941) was de Vries’s thesis advisor. Unknown to either Korteweg or de Vries, their Korteweg-de Vries model appeared as a footnote ten years earlier in French mathematician Joseph Valentin Boussinesq’s (1842--1929) 680-page treatise Essai sur la théorie des eaux courantes. The work of Boussinesq and of Korteweg and deVries remained largely unnoticed until two Americans, physicist Norman J. Zabusky (1929--) and mathematician Martin David Kruskal (1925--2006), used computer simulations to discover, in 1965, that all solutions of the KdV equations eventually consist of a finite set of localized traveling waves. Today, nearly 200 years after Russell’s observations and 50 years after the computational experiments of Zabusky and Kruskal, the study of “solitons” remains an active area of differential equations research. Other early contributors to nonlinear wave propagation include David Hilbert (German, 1862--1943), Richard Courant (GermanAmerican, 1888--1972), and John von Neumann (Hungarian-American, 1903--1957); we will encounter some of these ideas again in Chapter 9. Computational results were also an essential element in the discovery of “chaos theory.” In 1961, Edward Lorenz (1917--2008), an American mathematician and meteorologist at the Massachusetts Institute of Technology, was developing weather prediction models when he observed different results upon restarting a simulation in the middle of the time period using ......................................................................................................................................................................... 6 “Report

on Waves,” in Proceedings of the Fourteenth Meeting of the British Association for the Advancement of Science, 1845, pp. 311--390, plus plates 47--57, http://www.macs.hw.ac.uk/∼chris/Scott-Russell/SR44.pdf.

21

Boyce 9131 Ch01 2

22

September 29, 2016

17:13

22

CHAPTER 1 Introduction

previously computed results. (Lorenz restarted the computation with three-digit approximate solutions, not the six-digit approximations that were stored in the computer.) In 1976 the Australian mathematician Sir Robert M. May (1938--) introduced and analyzed the logistic map, showing that there are special values of the problem’s parameter where the solutions undergo drastic changes. The common trait that small changes in the problem produce large changes in the solution is one of the defining characteristics of chaos. May’s logistic map is discussed in more detail in Section 2.9. Other classical examples of what we now recognize as “chaos” include the work by French mathematician Henri Poincaré (1854--1912) on planetary motion and the studies of turbulent fluid flow by Soviet mathematician Andrey Nikolaevich Kolmogorov (1903--1987), American mathematician Mitchell Feigenbaum (1944--), and many others. In addition to these and other classical examples of chaos, new examples continue to be found. Solitons and chaos are just two of many examples where computers, and especially computer graphics, have given a new impetus to the study of systems of nonlinear differential equations. Other unexpected phenomena (Section 9.8), such as strange attractors (David Ruelle, Belgium, 1935--) and fractals (Benoit Mandelbrot, Poland, 1924--2010), have been discovered, are being intensively studied, and are leading to important new insights in a variety of applications. Although it is an old subject about which much is known, the study of differential equations in the twenty-first century remains a fertile source of fascinating and important unsolved problems.

Problems In each of Problems 1 through 4, determine the order of the given differential equation; also state whether the equation is linear or nonlinear. d2 y dy + 2y = sin t 1. t 2 2 + t dt dt d2 y dy + y = et 2. 1 + y 2 +t dt dt 2 d3 y d2 y dy d4 y +y=1 3. + + + dt dt 4 dt 3 dt 2 d2 y 4. + sin( t + y) = sin t dt 2 In each of Problems 5 through 10, verify that each given function is a solution of the differential equation.

5. y − y = 0;

y1 ( t) = et , y2 ( t) = cosh t

6. y + 2y − 3y = 0; 7. t y − y = t 2 ;

y1 ( t) = e−3t , y2 ( t) = et

y = 3t + t 2

8. y + 4y + 3y = t;

y1 ( t) = t/3, y2 ( t) = e−t + t/3

9. t 2 y + 5t y + 4y = 0, t > 0; y1 ( t) = t −2 , y2 ( t) = t −2 ln t

t 2 2 2 10. y − 2t y = 1; y = et e−s ds + et 0

In each of Problems 11 through 13, determine the values of r for which the given differential equation has solutions of the form y = er t .

11. y + 2y = 0 12. y + y − 6y = 0 13. y − 3y + 2y = 0 In each of Problems 14 and 15, determine the values of r for which the given differential equation has solutions of the form y = t r for t > 0.

14. t 2 y + 4t y + 2y = 0 15. t 2 y − 4t y + 4y = 0

In each of Problems 16 through 18, determine the order of the given partial differential equation; also state whether the equation is linear or nonlinear. Partial derivatives are denoted by subscripts. 16. u x x + u yy + u zz = 0

17. u x x x x + 2u x x yy + u yyyy = 0 18. u t + uu x = 1 + u x x In each of Problems 19 through 21, verify that each given function is a solution of the given partial differential equation. 19. u x x + u yy = 0; u 1 ( x, y) = cos x cosh y, u 2 ( x, y) = ln( x 2 + y 2 )

20. α 2 u x x = u t ; u 2 ( x, t) = e

−α 2 λ 2 t

2

u 1 ( x, t) = e−α t sin x, sin λ x, λ a real constant

21. a 2 u x x = u tt ;

u 1 ( x, t) = sin( λ x) sin( λat) , u 2 ( x, t) = sin( x − at) , λ a real constant

22. Follow the steps indicated here to derive the equation of motion of a pendulum, equation (12) in the text. Assume that the rod is rigid and weightless, that the mass is a point mass, and that there is no friction or drag anywhere in the system. a. Assume that the mass is in an arbitrary displaced position, indicated by the angle θ . Draw a free-body diagram showing the forces acting on the mass. b. Apply Newton’s law of motion in the direction tangential to the circular arc on which the mass moves. Then the tensile force in the rod does not enter the equation. Observe that you need to find the component of the gravitational force in the tangential direction. Observe also that the linear acceleration, as opposed to the angular acceleration, is Ld 2 θ /dt 2 , where L is the length of the rod. c. Simplify the result from part b to obtain equation (12) in the text.

Boyce 9131 Ch01 2

September 29, 2016

17:13

23

References

23. Another way to derive the pendulum equation (12) is based on the principle of conservation of energy. a. Show that the kinetic energy T of the pendulum in motion is 1 T = m L2 2

dθ dt

2 .

b. Show that the potential energy V of the pendulum, relative to its rest position, is V = mgL( 1 − cos θ ) .

c. By the principle of conservation of energy, the total energy

23

E = T + V is constant. Calculate d E/dt, set it equal to zero, and show that the resulting equation reduces to equation (12).

24. A third derivation of the pendulum equation depends on the principle of angular momentum: The rate of change of angular momentum about any point is equal to the net external moment about the same point. a. Show that the angular momentum M, or moment of momentum, about the point of support is given by M = m L 2 dθ /dt. b. Set d M/dt equal to the moment of the gravitational force, and show that the resulting equation reduces to equation (12). Note that positive moments are counterclockwise.

References Computer software for differential equations changes too fast for particulars to be given in a book such as this. A Google search for Maple, Mathematica, Sage, or MATLAB is a good way to begin if you need information about one of these computer algebra and numerical systems. There are many instructional books on computer algebra systems, such as the following: Cheung, C.-K., Keough, G. E., Gross, R. H., and Landraitis, C., Getting Started with Mathematica (3rd ed.) (New York: Wiley, 2009). Meade, D. B., May, M., Cheung, C.-K., and Keough, G. E., Getting Started with Maple (3rd ed.) (New York: Wiley, 2009). For further reading in the history of mathematics, see books such as those listed below: Boyer, C. B., and Merzbach, U. C., A History of Mathematics (2nd ed.) (New York: Wiley, 1989). Kline, M., Mathematical Thought from Ancient to Modern Times (3 vols.) (New York: Oxford University Press, 1990).

A useful historical appendix on the early development of differential equations appears in Ince, E. L., Ordinary Differential Equations (London: Longmans, Green, 1927; New York: Dover, 1956). Encyclopedic sources of information about the lives and achievements of mathematicians of the past are Gillespie, C. C., ed., Dictionary of Scientific Biography (15 vols.) (New York: Scribner’s, 1971). Koertge, N., ed., New Dictionary of Scientific Biography (8 vols.) (New York: Scribner’s, 2007). Koertge, N., ed., Complete Dictionary of Scientific Biography (New York: Scribner’s, 2007 [e-book]). Much historical information can be found on the Internet. One excellent site is the MacTutor History of Mathematics archive http://www-history.mcs.st-and.ac.uk/history/ created by O’Connor, J. J., and Robertson, E. F., Department of Mathematics and Statistics, University of St. Andrews, Scotland.

Boyce 9131 Ch02 2

September 29, 2016

17:16

24

CHAPTER 2 First-Order Differential Equations This chapter deals with differential equations of first order dy = f ( t, y) , (1) dt where f is a given function of two variables. Any differentiable function y = φ ( t) that satisfies this equation for all t in some interval is called a solution, and our objective is to determine whether such functions exist and, if so, to develop methods for finding them. Unfortunately, for an arbitrary function f , there is no general method for solving the equation in terms of elementary functions. Instead, we will describe several methods, each of which is applicable to a certain subclass of first-order equations. The most important of these are linear equations (Section 2.1), separable equations (Section 2.2), and exact equations (Section 2.6). Other sections of this chapter describe some of the important applications of first-order differential equations, introduce the idea of approximating a solution by numerical computation, and discuss some theoretical questions related to the existence and uniqueness of solutions. The final section includes an example of chaotic solutions in the context of first-order difference equations, which have some important points of similarity with differential equations and are simpler to investigate.

Linear Differential Equations; Method of Integrating Factors 2.1

24

If the function f in equation (1) depends linearly on the dependent variable y, then equation (1) is a first-order linear differential equation. In Sections 1.1 and 1.2 we discussed a restricted type of first-order linear differential equation in which the coefficients are constants. A typical example is dy = −ay + b, (2) dt where a and b are given constants. Recall that an equation of this form describes the motion of an object falling in the atmosphere. Now we want to consider the most general first-order linear differential equation, which is obtained by replacing the coefficients a and b in equation (2) by arbitrary functions of t. We will usually write the general first-order linear differential equation in the standard form dy (3) + p( t) y = g( t) , dt where p and g are given functions of the independent variable t. Sometimes it is more convenient to write the equation in the form dy (4) P( t) + Q( t) y = G( t) , dt where P, Q, and G are given. Of course, as long as P( t) = 0, you can convert equation (4) to equation (3) by dividing both sides of equation (4) by P( t) . In some cases it is possible to solve a first-order linear differential equation immediately by integrating the equation, as in the next example.

Boyce 9131 Ch02 2

September 29, 2016

17:16

25

2.1 Linear Differential Equations; Method of Integrating Factors

EXAMPLE 1 Solve the differential equation ( 4 + t 2)

dy + 2t y = 4t. dt

(5)

Solution: The left-hand side of equation (5) is a linear combination of dy/dt and y, a combination that also appears in the rule from calculus for differentiating a product. In fact, ( 4 + t 2)

dy d ( 4 + t 2) y ; + 2t y = dt dt

it follows that equation (5) can be rewritten as

d ( 4 + t 2 ) y = 4t. dt

(6)

Thus, even though y is unknown, we can integrate both sides of equation (6) with respect to t, thereby obtaining ( 4 + t 2 ) y = 2t 2 + c,

(7)

where c is an arbitrary constant of integration. Solving for y, we find that y=

c 2t 2 + . 2 4+t 4 + t2

(8)

This is the general solution of equation (5).

Unfortunately, most first-order linear differential equations cannot be solved as illustrated in Example 1 because their left-hand sides are not the derivative of the product of y and some other function. However, Leibniz discovered that if the differential equation is multiplied by a certain function μ ( t) , then the equation is converted into one that is immediately integrable by using the product rule for derivatives, just as in Example 1. The function μ ( t) is called an integrating factor and our main task in this section is to determine how to find it for a given equation. We will show how this method works first for an example and then for the general first-order linear differential equation in the standard form (3).

EXAMPLE 2 Find the general solution of the differential equation 1 1 dy + y = et/3 . dt 2 2

(9)

Draw some representative integral curves; that is, plot solutions corresponding to several values of the arbitrary constant c. Also find the particular solution whose graph contains the point ( 0, 1) . Solution: The first step is to multiply equation (9) by a function μ ( t) , as yet undetermined; thus μ ( t)

dy 1 1 + μ ( t) y = μ ( t) et/3 . dt 2 2

(10)

The question now is whether we can choose μ ( t) so that the left-hand side of equation (10) is the derivative of the product μ ( t) y. For any differentiable function μ ( t) we have dy dμ ( t) d ( μ ( t) y) = μ ( t) + y. dt dt dt

(11)

Thus the left-hand side of equation (10) and the right-hand side of equation (11) are identical, provided that we choose μ ( t) to satisfy dμ ( t) 1 = μ ( t) . dt 2

▼

(12)

25

Boyce 9131 Ch02 2

26

September 29, 2016

17:16

26

CHAPTER 2 First-Order Differential Equations

▼

Our search for an integrating factor will be successful if we can find a solution of equation (12). Perhaps you can readily identify a function that satisfies equation (12): What well-known function from calculus has a derivative that is equal to one-half times the original function? More systematically, rewrite equation (12) as 1 dμ ( t) 1 = , μ ( t) dt 2 which is equivalent to 1 d ln |μ ( t) | = . dt 2

(13)

Then it follows that ln |μ ( t) | =

1 t + C, 2

or μ ( t) = cet/2 .

(14)

The function μ ( t) given by equation (14) is an integrating factor for equation (9). Since we do not need the most general integrating factor, we will choose c to be 1 in equation (14) and use μ ( t) = et/2 . Now we return to equation (9), multiply it by the integrating factor et/2 , and obtain et/2

dy 1 1 + et/2 y = e5t/6 . dt 2 2

(15)

By the choice we have made of the integrating factor, the left-hand side of equation (15) is the derivative of et/2 y, so that equation (15) becomes 1 d t/2 ( e y) = e5t/6 . (16) dt 2 By integrating both sides of equation (16), we obtain 3 et/2 y = e5t/6 + c, 5

(17)

where c is an arbitrary constant. Finally, on solving equation (17) for y, we have the general solution of equation (9), namely, 3 y = et/3 + ce−t/2 . (18) 5 To find the solution passing through the point ( 0, 1) , we set t = 0 and y = 1 in equation (18), obtaining 1 = 3/5 + c. Thus c = 2/5, and the desired solution is 3 2 y = et/3 + e−t/2 . (19) 5 5 Figure 2.1.1 includes the graphs of equation (18) for several values of c with a direction field in the background. The solution satisfying y( 0) = 1 is shown by the green curve. y 4 3 2 1

1

2

3

4

5

6 t

–1 –2 1 1 FIGURE 2.1.1 Direction field and integral curves of y + 2 y = 2 et/3 ; the

green curve passes through the point ( 0, 1) .

Boyce 9131 Ch02 2

September 29, 2016

17:16

27

2.1 Linear Differential Equations; Method of Integrating Factors

Let us now extend the method of integrating factors to equations of the form dy + ay = g( t) , dt

(20)

where a is a given constant and g( t) is a given function. Proceeding as in Example 2, we find that the integrating factor μ ( t) must satisfy dμ = aμ , dt

(21)

rather than equation (12). Thus the integrating factor is μ ( t) = eat . Multiplying equation (20) by μ ( t) , we obtain eat

dy + aeat y = eat g( t) , dt

or d at ( e y) = eat g( t) . dt By integrating both sides of equation (22), we find that at e y = eat g( t) dt + c,

(22)

(23)

where c is an arbitrary constant. For many simple functions g( t) , we can evaluate the integral in equation (23) and express the solution y in terms of elementary functions, as in Example 2. However, for more complicated functions g( t) , it is necessary to leave the solution in integral form. In this case t y = e−at eas g( s) ds + ce−at . (24) t0

Note that in equation (24) we have used s to denote the integration variable to distinguish it from the independent variable t, and we have chosen some convenient value t0 as the lower limit of integration. (See Theorem 2.4.1.) The choice of t0 determines the specific value of the constant c but does not change the solution. For example, plugging t = t0 into the solution formula (24) shows that c = y( t0 ) eat0 .

EXAMPLE 3 Find the general solution of the differential equation dy − 2y = 4 − t dt

(25)

and plot the graphs of several solutions. Discuss the behavior of solutions as t → ∞. Solution: Equation (25) is of the form (20) with a = −2; therefore, the integrating factor is μ ( t) = e−2t . Multiplying the differential equation (25) by μ ( t) , we obtain e−2t

dy − 2e−2t y = 4e−2t − te−2t , dt

or d −2t e y = 4e−2t − te−2t . dt Then, by integrating both sides of this equation, we have

(26)

1 1 e−2t y = −2e−2t + te−2t + e−2t + c, 2 4 where we have used integration by parts on the last term in equation (26). Thus the general solution of equation (25) is

▼

7 1 y = − + t + ce2t . 4 2

(27)

27

Boyce 9131 Ch02 2

28

September 29, 2016

17:16

28

CHAPTER 2 First-Order Differential Equations

▼

Figure 2.1.2 shows the direction field and graphs of the solution (27) for several values of c. The behavior of the solution for large values of t is determined by the term ce2t . If c = 0, then the solution grows exponentially large in magnitude, with the same sign as c itself. Thus the solutions diverge as t becomes large. The boundary between solutions that ultimately grow positively and those that ultimately grow negatively occurs when c = 0. If we substitute c = 0 into equation (27) and then set t = 0, we find that y = −7/4 is the separation point on the y-axis. Note that for this initial value, 7 1 the solution is y = − + t; it grows positively, but linearly rather than exponentially. 4 2 y

0.5

1

1.5

2 t

–1

–2

–3

–4 FIGURE 2.1.2 Direction field and integral curves of y − 2y = 4 − t.

Now we return to the general first-order linear differential equation (3) dy + p( t) y = g( t) , dt where p and g are given functions. To determine an appropriate integrating factor, we multiply equation (3) by an as yet undetermined function μ ( t) , obtaining μ ( t)

dy + p( t) μ ( t) y = μ ( t) g( t) . dt

(28)

Following the same line of development as in Example 2, we see that the left-hand side of equation (28) is the derivative of the product μ ( t) y, provided that μ ( t) satisfies the equation dμ ( t) (29) = p( t) μ ( t) . dt If we assume temporarily that μ ( t) is positive, then we have 1 dμ ( t) = p( t) , μ ( t) dt and consequently

ln |μ ( t) | =

p( t) dt + k.

By choosing the arbitrary constant k to be zero, we obtain the simplest possible function for μ , namely, μ ( t) = exp p( t) dt. (30) Note that μ ( t) is positive for all t, as we assumed. Returning to equation (28), we have d ( μ ( t) y) = μ ( t) g( t) . dt Hence

(31)

μ ( t) y =

μ ( t) g( t) dt + c,

(32)

Boyce 9131 Ch02 2

September 29, 2016

17:16

29

2.1 Linear Differential Equations; Method of Integrating Factors

where c is an arbitrary constant. Sometimes the integral in equation (32) can be evaluated in terms of elementary functions. However, in general this is not possible, so the general solution of equation (3) is t 1 y= μ ( s) g( s) ds + c , (33) μ ( t) t0 where again t0 is some convenient lower limit of integration. Observe that equation (33) involves two integrations, one to obtain μ ( t) from equation (30) and the other to determine y from equation (33).

EXAMPLE 4 Solve the initial value problem t y + 2y = 4t 2 , y( 1) = 2.

(34) (35)

Solution: In order to determine p( t) and g( t) correctly, we must first rewrite equation (34) in the standard form (3). Thus we have 2 (36) y + y = 4t, t so p( t) = 2/ t and g( t) = 4t. To solve equation (36), we first compute the integrating factor μ ( t) :

2 dt t

μ ( t) = exp

= e2 ln |t| = t 2 .

On multiplying equation (36) by μ ( t) = t 2 , we obtain t 2 y + 2t y = ( t 2 y) = 4t 3 , and therefore

t y=

4t 3 dt = t 4 + c,

2

where c is an arbitrary constant. It follows that, for t > 0, c (37) y = t2 + 2 t is the general solution of equation (34). Integral curves of equation (34) for several values of c are shown in Figure 2.1.3. To satisfy initial condition (35), set t = 1 and y = 2 in equation (37): 2 = 1 + c, so c = 1; thus y = t2 +

1 , t2

t>0

(38)

is the solution of the initial value problem (24), (25). This solution is shown by the green curve in Figure 2.1.3. Note that it becomes unbounded and is asymptotic to the positive y-axis as t → 0 from the right. This is the effect of the infinite discontinuity in the coefficient p( t) at the origin. It is important to note that while the function y = t 2 + 1/ t 2 for t < 0 is part of the general solution of equation (34), it is not part of the solution of this initial value problem. This is the first example in which the solution fails to exist for some values of t. Again, this is due to the infinite discontinuity in p( t) at t = 0, which restricts the solution to the interval 0 < t < ∞. Looking again at Figure 2.1.3, we see that some solutions (those for which c > 0) are asymptotic to the positive y-axis as t → 0 from the right, while other solutions (for which c < 0) are asymptotic to the negative y-axis. If we generalize the initial condition (35) to y( 1) = y0 ,

(39)

then c = y0 − 1 and the solution (38) becomes y = t2 +

y0 − 1 t2

,

t>0

(40)

Note that when y0 = 1, so c = 0, the solution is y = t 2 , which remains bounded and differentiable even at t = 0. (This is the red curve in Figure 2.1.3.)

▼

29

Boyce 9131 Ch02 2

30

September 29, 2016

17:16

30

CHAPTER 2 First-Order Differential Equations

▼ As in Example 3, this is another instance where there is a critical initial value, namely, y0 = 1, that separates solutions that behave in one way from others that behave quite differently. y 3 2

(1, 2)

1

–1

1

t

–1 FIGURE 2.1.3 Integral curves of the differential equation t y + 2y = 4t 2 ;

the green curve is the particular solution with y( 1) = 2. The red curve is the particular solution with y( 1) = 1.

EXAMPLE 5 Solve the initial value problem 2y + t y = 2,

(41)

y( 0) = 1.

(42)

Solution: To convert the differential equation (41) to the standard form (3), we must divide equation (41) by 2, obtaining t y + y = 1. (43) 2 Thus p( t) = t/2, and the integrating factor is μ ( t) = exp( t 2 /4) . Then multiply equation (43) by μ ( t) , so that t 2 2 2 et /4 y + et /4 y = et /4 . (44) 2 The left-hand side of equation (44) is the derivative of et equation (44), we obtain et

2 /4

y=

et

2 /4

2 /4

y, so by integrating both sides of

dt + c.

(45)

The integral on the right-hand side of equation (45) cannot be evaluated in terms of the usual elementary functions, so we leave the integral unevaluated. By choosing the lower limit of integration as the initial point t = 0, we can replace equation (45) by et

2 /4

t

y=

es

2 /4

ds + c,

(46)

0

where c is an arbitrary constant. It then follows that the general solution y of equation (41) is given by y=e

−t 2 /4

t

es

2 /4

ds + ce−t

2 /4

.

(47)

0

To determine the particular solution that satisfies the initial condition (42), set t = 0 and y = 1 in equation (47):

1=e

0

0 0

= 0 + c,

▼

so c = 1.

e−s

2 /4

ds + ce0

Boyce 9131 Ch02 2

September 29, 2016

17:16

31

2.1 Linear Differential Equations; Method of Integrating Factors

▼

31

The main purpose of this example is to illustrate that sometimes the solution must be left in terms of an integral. This is usually at most a slight inconvenience, rather than a serious obstacle. For a given value of t, the integral in equation (47) is a definite integral and can be approximated to any desired degree of accuracy by using readily available numerical integrators. By repeating this process for many values of t and plotting the results, you can obtain a graph of a solution. Alternatively, you can use a numerical approximation method, such as those discussed in Chapter 8, that proceed directly from the differential equation and need no expression for the solution. Software packages such as Maple, Mathematica, MATLAB and Sage readily execute such procedures and produce graphs of solutions of differential equations. Figure 2.1.4 displays graphs of the solution (47) for several values of c. The particular solution satisfying the initial condition y( 0) = 1 is shown in black. From the figure it may be plausible to conjecture that all solutions approach a limit as t → ∞. The limit can also be found analytically (see Problem 22). y 3 2 1

1

2

3

5

4

6 t

–1 –2 –3 FIGURE 2.1.4 Integral curves of 2y + t y = 2; the green curve is the particular

solution satisfying the initial condition y( 0) = 1.

Problems In each of Problems 1 through 8: G a. Draw a direction field for the given differential equation. b. Based on an inspection of the direction field, describe how solutions behave for large t. c. Find the general solution of the given differential equation, and use it to determine how solutions behave as t → ∞.

1. y + 3y = t + e−2t 2. y − 2y = t 2 e2t 3. y + y = te−t + 1 1 t y − 2y = 3et

4. y + y = 3 cos( 2t) ,

t>0

5. 6. t y − y = t 2 e−t , t > 0 7. y + y = 5 sin( 2t) 8. 2y + y = 3t 2 In each of Problems 9 through 12, find the solution of the given initial value problem.

9. y − y = 2te2t , y( 0) = 1 10. y + 2y = te−2t , y( 1) = 0 2 cos t , y( π ) = 0, t > 0 t t2 t y + ( t + 1) y = t, y( ln 2) = 1, t > 0

11. y + y = 12.

In each of Problems 13 and 14: G a. Draw a direction field for the given differential equation. How do solutions appear to behave as t becomes large? Does the behavior depend on the choice of the initial value a? Let a0 be the value of a for which the transition from one type of behavior to another occurs. Estimate the value of a0 . b. Solve the initial value problem and find the critical value a0 exactly. c. Describe the behavior of the solution corresponding to the initial value a0 . 1 13. y − y = 2 cos t, y( 0) = a 2 14. 3y − 2y = e−π t/2 , y( 0) = a

Boyce 9131 Ch02 2

32

September 29, 2016

17:16

32

CHAPTER 2 First-Order Differential Equations

In each of Problems 15 and 16: G a. Draw a direction field for the given differential equation. How do solutions appear to behave as t → 0? Does the behavior depend on the choice of the initial value a? Let a0 be the critical value of a, that is, the initial value such that the solutions for a < a0 and the solutions for a > a0 have different behaviors as t → ∞. Estimate the value of a0 . b. Solve the initial value problem and find the critical value a0 exactly. c. Describe the behavior of the solution corresponding to the initial value a0 .

15. t y + ( t + 1) y = 2te−t , y( 1) = a, t > 0 16. ( sin t) y + ( cos t) y = et , y( 1) = a, 0 < t < π G 17. Consider the initial value problem 1 y = 2 cos t, y( 0) = −1. 2 Find the coordinates of the first local maximum point of the solution for t > 0. y +

N

18. Consider the initial value problem y +

2 1 y = 1 − t, 3 2

term in the solution (47) is indeterminate with form 0 · ∞. Then, use l’Hôpital’s rule to compute the limit as t → ∞.

23. Show that if a and λ are positive constants, and b is any real number, then every solution of the equation y + ay = be−λt has the property that y → 0 as t → ∞. Hint: Consider the cases a = λ and a = λ separately. In each of Problems 24 through 27, construct a first-order linear differential equation whose solutions have the required behavior as t → ∞. Then solve your equation and confirm that the solutions do indeed have the specified property.

24. 25. 26. 27. 28.

All solutions have the limit 3 as t → ∞. All solutions are asymptotic to the line y = 3 − t as t → ∞. All solutions are asymptotic to the line y = 2t − 5 as t → ∞. All solutions approach the curve y = 4 − t 2 as t → ∞.

Variation of Parameters. Consider the following method of solving the general linear equation of first order: y + p( t) y = g( t) .

y( 0) = y0 .

Find the value of y0 for which the solution touches, but does not cross, the t-axis.

(48)

a. If g( t) = 0 for all t, show that the solution is y = A exp −

p( t) dt ,

(49)

19. Consider the initial value problem y +

1 y = 3 + 2 cos( 2t) , 4

y( 0) = 0.

where A is a constant. b. If g( t) is not everywhere zero, assume that the solution of equation (48) is of the form

a. Find the solution of this initial value problem and describe its behavior for large t. N b. Determine the value of t for which the solution first intersects the line y = 12.

20. Find the value of y0 for which the solution of the initial value problem

y − y = 1 + 3 sin t, remains finite as t → ∞. 3 y = 3t + 2et , 2

y( 0) = y0 .

Find the value of y0 that separates solutions that grow positively as t → ∞ from those that grow negatively. How does the solution that corresponds to this critical value of y0 behave as t → ∞?

22. Show that all solutions of 2y + t y = 2 [equation (41) of the

text] approach a limit as t → ∞, and find the limiting value. Hint: Consider the general solution, equation (47). Show that the first

p( t) dt ,

(50)

where A is now a function of t. By substituting for y in the given differential equation, show that A( t) must satisfy the condition

A ( t) = g( t) exp

y( 0) = y0

21. Consider the initial value problem y −

y = A( t) exp −

p( t) dt .

(51)

c. Find A( t) from equation (51). Then substitute for A( t) in equation (50) and determine y. Verify that the solution obtained in this manner agrees with that of equation (33) in the text. This technique is known as the method of variation of parameters; it is discussed in detail in Section 3.6 in connection with secondorder linear equations. In each of Problems 29 and 30, use the method of Problem 28 to solve the given differential equation.

29. y − 2y = t 2 e2t 1 t

30. y + y = cos( 2t) ,

t>0

Boyce 9131 Ch02 2

September 29, 2016

17:16

33

2.2 Separable Differential Equations

2.2

Separable Differential Equations

In Section 1.2 we used a process of direct integration to solve first-order linear differential equations of the form dy (1) = ay + b, dt where a and b are constants. We will now show that this process is actually applicable to a much larger class of nonlinear differential equations. We will use x, rather than t, to denote the independent variable in this section for two reasons. In the first place, different letters are frequently used for the variables in a differential equation, and you should not become too accustomed to using a single pair. In particular, x often occurs as the independent variable. Further, we want to reserve t for another purpose later in the section. The general first-order differential equation is dy = f ( x, y) . dx

(2)

Linear differential equations were considered in the preceding section, but if equation (2) is nonlinear, then there is no universally applicable method for solving the equation. Here, we consider a subclass of first-order equations that can be solved by direct integration. To identify this class of equations, we first rewrite equation (2) in the form M( x, y) + N ( x, y)

dy = 0. dx

(3)

It is always possible to do this by setting M( x, y) = − f ( x, y) and N ( x, y) = 1, but there may be other ways as well. When M is a function of x only and N is a function of y only, then equation (3) becomes dy = 0. (4) dx Such an equation is said to be separable, because if it is written in the differential form M( x) + N ( y)

M( x) d x + N ( y) dy = 0,

(5)

then, if you wish, terms involving each variable may be placed on opposite sides of the equation. The differential form (5) is also more symmetric and tends to suppress the distinction between independent and dependent variables. A separable equation can be solved by integrating the functions M and N . We illustrate the process by an example and then discuss it in general for equation (4).

EXAMPLE 1 Show that the equation x2 dy = dx 1 − y2

(6)

is separable, and then find an equation for its integral curves. Solution: If we write equation (6) as dy = 0, (7) dx then it has the form (4) and is therefore separable. Recall from calculus that if y is a function of x, then by the chain rule, −x 2 + ( 1 − y 2 )

▼

d dy d dy f ( y) = f ( y) = f ( y) . dx dy dx dx

33

Boyce 9131 Ch02 2

34

September 29, 2016

17:16

34

CHAPTER 2 First-Order Differential Equations

▼ For example, if f ( y) = y − y 3 /3, then d dx

y−

y3 3

= ( 1 − y2)

dy . dx

Thus the second term in equation (7) is the derivative with respect to x of y − y 3 /3, and the first term is the derivative of −x 3 /3. Thus equation (7) can be written as d dx

−

or d dx

x3 3

+

d dx

y−

x3 y3 − +y− 3 3

y3 3

= 0,

= 0.

Therefore, by integrating (and multiplying the result by 3), we obtain −x 3 + 3y − y 3 = c,

(8)

where c is an arbitrary constant. Equation (8) is an equation for the integral curves of equation (6). A direction field and several integral curves are shown in Figure 2.2.1. Any differentiable function y = φ ( x) that satisfies equation (8) is a solution of equation (6). An equation of the integral curve passing through a particular point (x0 , y0 ) can be found by substituting x0 and y0 for x and y, respectively, in equation (8) and determining the corresponding value of c. y 4

2

–4

–2

2

4 x

–2

–4 FIGURE 2.2.1 Direction field and integral curves of y = x 2 /( 1 − y 2 ) .

Essentially the same procedure can be followed for any separable equation. Returning to equation (4), let H1 and H2 be any antiderivatives of M and N , respectively. Thus H1 ( x) = M( x) ,

H2 ( y) = N ( y) ,

(9)

and equation (4) becomes dy = 0. dx If y is regarded as a function of x, then according to the chain rule, H1 ( x) + H2 ( y)

H2 ( y)

dy dy d d = H2 ( y) = H2 ( y) . dx dy dx dx

Consequently, we can write equation (10) as d H1 ( x) + H2 ( y) = 0. dx

(10)

(11)

(12)

Boyce 9131 Ch02 2

September 29, 2016

17:16

35

2.2 Separable Differential Equations

By integrating equation (12) with respect to x, we obtain H1 ( x) + H2 ( y) = c,

(13)

where c is an arbitrary constant. Any differentiable function y = φ ( x) that satisfies equation (13) is a solution of equation (4); in other words, equation (13) defines the solution implicitly rather than explicitly. In practice, equation (13) is usually obtained from equation (5) by integrating the first term with respect to x and the second term with respect to y. The justification for this is the argument that we have just given. The differential equation (4), together with an initial condition y( x0 ) = y0 ,

(14)

forms an initial value problem. To solve this initial value problem, we must determine the appropriate value for the constant c in equation (13). We do this by setting x = x0 and y = y0 in equation (13) with the result that c = H1 ( x0 ) + H2 ( y0 ) . Substituting this value of c in equation (13) and noting that x H1 ( x) − H1 ( x0 ) = M( s) ds, H2 ( y) − H2 ( y0 ) = x0

we obtain

x

(15)

y

N ( s) ds, y0

M( s) ds +

x0

y

N ( s) ds = 0.

(16)

y0

Equation (16) is an implicit representation of the solution of the differential equation (4) that also satisfies the initial condition (14). Bear in mind that to determine an explicit formula for the solution, you need to solve equation (16) for y as a function of x. Unfortunately, it is often impossible to do this analytically; in such cases you can resort to numerical methods to find approximate values of y for given values of x.

EXAMPLE 2 Solve the initial value problem 3x 2 + 4x + 2 dy = , dx 2( y − 1)

y( 0) = −1,

(17)

and determine the interval in which the solution exists. Solution: The differential equation can be written as 2( y − 1) dy = ( 3x 2 + 4x + 2) d x. Integrating the left-hand side with respect to y and the right-hand side with respect to x gives y 2 − 2y = x 3 + 2x 2 + 2x + c,

(18)

where c is an arbitrary constant. To determine the solution satisfying the prescribed initial condition, we substitute x = 0 and y = −1 in equation (18), obtaining c = 3. Hence the solution of the initial value problem is given implicitly by y 2 − 2y = x 3 + 2x 2 + 2x + 3.

(19)

To obtain the solution explicitly, we must solve equation (19) for y in terms of x. That is a simple matter in this case, since equation (19) is quadratic in y, and we obtain y =1±

x 3 + 2x 2 + 2x + 4.

(20)

Equation (20) gives two solutions of the differential equation, only one of which, however, satisfies the given initial condition. This is the solution corresponding to the minus sign in equation (20), so we finally obtain y = φ ( x) = 1 −

▼

x 3 + 2x 2 + 2x + 4

(21)

35

Boyce 9131 Ch02 2

36

September 29, 2016

17:16

36

CHAPTER 2 First-Order Differential Equations

▼ as the solution of the initial value problem (15). Note that if we choose the plus sign by mistake in

equation (20), then we obtain the solution of the same differential equation that satisfies the initial condition y( 0) = 3. Finally, to determine the interval in which the solution (21) is valid, we must find the interval in which the quantity under the radical is positive. The only real zero of this expression is x = −2, so the desired interval is x > −2. Some integral curves of the differential equation are shown in Figure 2.2.2. The green curve passes through the point ( 0, −1) and thus is the solution of the initial value problem (15). Observe that the boundary of the interval of validity of the solution (21) is determined by the point ( −2, 1) at which the tangent line is vertical. y 3 2 1

(–2, 1)

–2

–1

1

2

x

(0, –1) –1 –2

FIGURE 2.2.2 Integral curves of y = ( 3x 2 + 4x + 2) /2( y − 1) ; the solution

satisfying y( 0) = − 1 is shown in green and is valid for x > − 2.

EXAMPLE 3 Solve the separable differential equation 4x − x 3 dy = dx 4 + y3

(22)

and draw graphs of several integral curves. Also find the solution passing through the point ( 0, 1) and determine its interval of validity. Solution: Rewriting equation (22) as ( 4 + y 3 ) dy = ( 4x − x 3 ) d x, integrating each side, multiplying by 4, and rearranging the terms, we obtain y 4 + 16y + x 4 − 8x 2 = c,

(23)

where c is an arbitrary constant. Any differentiable function y = φ ( x) that satisfies equation (23) is a solution of the differential equation (22). Graphs of equation (23) for several values of c are shown in Figure 2.2.3. To find the particular solution passing through ( 0, 1) , we set x = 0 and y = 1 in equation (23) with the result that c = 17. Thus the solution in question is given implicitly by y 4 + 16y + x 4 − 8x 2 = 17.

▼

(24)

It is shown by the green curve in Figure 2.2.3. The interval of validity of this solution extends on either side of the initial point as long as the function remains differentiable. From the figure we see that the interval ends when we reach points where the tangent line is vertical. It follows from the differential equation (22) that these are points where 4 + y 3 = 0, or y = ( −4) 1/3 ∼ = −1.5874. From equation (24) the corresponding values of x are x ∼ = ±3.3488. These points are marked on the graph in Figure 2.2.3.

Boyce 9131 Ch02 2

September 29, 2016

17:16

37

2.2 Separable Differential Equations

▼ y 2

1 –3

–2

–1

1

2

3 x

(–3.3488, –1.5874)

–1

(3.3488, –1.5874)

–2

–3 FIGURE 2.2.3 Integral curves of y = ( 4x − x 3 ) /( 4+ y 3 ) . The solution passing through ( 0, 1) is shown by the green curve.

Note 1: Sometimes a differential equation of the form (2): dy = f ( x, y) dx has a constant solution y = y0 . Such a solution is usually easy to find because if f ( x, y0 ) = 0 for some value y0 and for all x, then the constant function y = y0 is a solution of the differential equation (2). For example, the equation dy ( y − 3) cos x (25) = dx 1 + 2y 2 has the constant solution y = 3. Other solutions of this equation can be found by separating the variables and integrating. Note 2: The investigation of a first-order nonlinear differential equation can sometimes be facilitated by regarding both x and y as functions of a third variable t. Then dy dy/dt (26) = . dx d x/dt If the differential equation is dy F( x, y) = , dx G( x, y)

(27)

then, by comparing numerators and denominators in equations (26) and (27), we obtain the system dx dy (28) = G( x, y) , = F( x, y) . dt dt At first sight it may seem unlikely that a problem will be simplified by replacing a single equation by a pair of equations, but in fact, the system (28) may well be more amenable to investigation than the single equation (27). Chapter 9 is devoted to nonlinear systems of the form (28). Note 3: In Example 2 it was not difficult to solve explicitly for y as a function of x. However, this situation is exceptional, and often it will be better to leave the solution in implicit form, as in Examples 1 and 3. Thus, in the problems below and in other sections where nonlinear equations appear, the words “solve the following differential equation” mean to find the solution explicitly if it is convenient to do so, but otherwise to find an equation defining the solution implicitly.

37

Boyce 9131 Ch02 2

38

September 29, 2016

17:16

38

CHAPTER 2 First-Order Differential Equations

Problems In each of Problems 1 through 8, solve the given differential equation. x2 1. y = y 2. y + y 2 sin x = 0

G

y = 2y 2 + x y 2 ,

20. Solve the initial value problem y =

−x

x −e dy = 5. dx y + ey x2 dy = 6. dx 1 + y2 y dy = 7. dx x dy −x 8. = dx y In each of Problems 9 through 16: a. Find the solution of the given initial value problem in explicit form. G b. Plot the graph of the solution. c. Determine (at least approximately) the interval in which the solution is defined.

9. 10. 11. 12. 13. 14. 15. 16.

y = ( 1 − 2x) y 2 ,

y( 0) = −1/6

y = ( 1 − 2x) / y,

y( 1) = −2

x d x + ye−x dy = 0, dr/dθ = r 2 /θ ,

r ( 1) = 2

y = 2x/( 1 + 2y) ,

y( 0) = 1

y( 2) = 0

y = ( 3x 2 − e x ) /( 2y − 5) , sin( 2x) d x + cos( 3y) dy = 0,

y( 0) = 1 y( π/2) = π/3

Some of the results requested in Problems 17 through 22 can be obtained either by solving the given equations analytically or by plotting numerically generated approximations to the solutions. Try to form an opinion about the advantages and disadvantages of each approach. G

17. Solve the initial value problem y =

1 + 3x 2 , 3y 2 − 6y

y( 0) = 1

and determine the interval in which the solution is valid. Hint: To find the interval of definition, look for points where the integral curve has a vertical tangent. G

2 − ex , 3 + 2y

y( 0) = 0

and determine where the solution attains its maximum value. G

21. Consider the initial value problem y =

t y( 4 − y) , 3

y( 0) = y0 .

a. Determine how the behavior of the solution as t increases depends on the initial value y0 . b. Suppose that y0 = 0.5. Find the time T at which the solution first reaches the value 3.98. G

22. Consider the initial value problem y =

t y( 4 − y) , 1+t

y( 0) = y0 > 0.

a. Determine how the solution behaves as t → ∞. b. If y0 = 2, find the time T at which the solution first reaches the value 3.99. c. Find the range of initial values for which the solution lies in the interval 3.99 < y < 4.01 by the time t = 2.

y( 0) = 1

y = x y 3 ( 1 + x 2 ) −1/2 ,

y( 0) = 1

and determine where the solution attains its minimum value. G

3. y = cos2 ( x) cos2 ( 2y) 4. x y = ( 1 − y 2 ) 1/2

19. Solve the initial value problem

23. Solve the equation dy ay + b = , dx cy + d where a, b, c, and d are constants.

24. Use separation of variables to solve the differential equation dQ = r ( a + bQ) , dt

Q( 0) = Q 0 ,

where a, b, r , and Q 0 are constants. Determine how the solution behaves as t → ∞ Homogeneous Equations. If the right-hand side of the equation dy/d x = f ( x, y) can be expressed as a function of the ratio y/ x only, then the equation is said to be homogeneous.1 Such equations can always be transformed into separable equations by a change of the dependent variable. Problem 25 illustrates how to solve first-order homogeneous equations.

18. Solve the initial value problem y =

3x 2 , 3y 2 − 4

.............................................................................................................................. y( 1) = 0

and determine the interval in which the solution is valid. Hint: To find the interval of definition, look for points where the integral curve has a vertical tangent.

1 The word “homogeneous” has different meanings in different mathematical contexts. The homogeneous equations considered here have nothing to do with the homogeneous equations that will occur in Chapter 3 and elsewhere.

Boyce 9131 Ch02 2

September 29, 2016

17:16

39

2.3 Modeling with First-Order Differential Equations N

25. Consider the equation dy y − 4x = . dx x−y

(29)

a. Show that equation (29) can be rewritten as ( y/ x) − 4 dy = ; dx 1 − ( y/ x)

(30)

thus equation (29) is homogeneous. b. Introduce a new dependent variable v so that v = y/ x, or y = xv( x) . Express dy/d x in terms of x, v, and dv/d x. c. Replace y and dy/d x in equation (30) by the expressions from part b that involve v and dv/d x. Show that the resulting differential equation is v+x

dv v −4 = , dx 1−v

or x

v2 − 4 dv = . dx 1−v

(31)

Observe that equation (31) is separable. d. Solve equation (31), obtaining v implicitly in terms of x. e. Find the solution of equation (29) by replacing v by y/ x in the solution in part d. f. Draw a direction field and some integral curves for equation (29). Recall that the right-hand side of equation (29) actually depends only on the ratio y/ x. This means that integral curves have the same slope at all points on any given straight line

39

through the origin, although the slope changes from one line to another. Therefore, the direction field and the integral curves are symmetric with respect to the origin. Is this symmetry property evident from your plot? The method outlined in Problem 25 can be used for any homogeneous equation. That is, the substitution y = xv( x) transforms a homogeneous equation into a separable equation. The latter equation can be solved by direct integration, and then replacing v by y/ x gives the solution to the original equation. In each of Problems 26 through 31: a. Show that the given equation is homogeneous. b. Solve the differential equation. G c. Draw a direction field and some integral curves. Are they symmetric with respect to the origin? dy x 2 + x y + y2 26. = dx x2 2 dy x + 3y 2 27. = dx 2x y dy 4y − 3x = 28. dx 2x − y dy 4x + 3y 29. =− dx 2x + y dy x 2 − 3y 2 30. = dx 2x y dy 3y 2 − x 2 = 31. dx 2x y

Modeling with First-Order Differential Equations 2.3

Differential equations are of interest to nonmathematicians primarily because of the possibility of using them to investigate a wide variety of problems in the physical, biological, and social sciences. One reason for this is that mathematical models and their solutions lead to equations relating the variables and parameters in the problem. These equations often enable you to make predictions about how the natural process will behave in various circumstances. It is often easy to vary parameters in the mathematical model over wide ranges, whereas this may be very time-consuming or expensive, if not impossible, in an experimental setting. Nevertheless, mathematical modeling and experiment or observation are both critically important and have somewhat complementary roles in scientific investigations. Mathematical models are validated by comparison of their predictions with experimental results. On the other hand, mathematical analyses may suggest the most promising directions to explore experimentally, and they may indicate fairly precisely what experimental data will be most helpful. In Sections 1.1 and 1.2 we formulated and investigated a few simple mathematical models. We begin by recapitulating and expanding on some of the conclusions reached in those sections. Regardless of the specific field of application, there are three identifiable steps that are always present in the process of mathematical modeling. Step 1: Construction of the Model. In this step the physical situation is translated into mathematical terms, often using the steps listed at the end of Section 1.1. Perhaps most critical at this stage is to state clearly the physical principle(s) that are believed to govern the process. For example, it has been observed that in some circumstances heat passes from a warmer to a cooler body at a rate proportional to the temperature difference, that objects move about in accordance with Newton’s laws of motion, and that isolated insect populations grow at a rate proportional to the current population. Each of these statements involves a rate of

Boyce 9131 Ch02 2

40

September 29, 2016

17:16

40

CHAPTER 2 First-Order Differential Equations

change (derivative) and consequently, when expressed mathematically, leads to a differential equation. The differential equation is a mathematical model of the process. It is important to realize that the mathematical equations are almost always only an approximate description of the actual process. For example, bodies moving at speeds comparable to the speed of light are not governed by Newton’s laws, insect populations do not grow indefinitely as stated because of eventual lack of food or space, and heat transfer is affected by factors other than the temperature difference. Thus you should always be aware of the limitations of the model so that you will use it only when it is reasonable to believe that it is accurate. Alternatively, you can adopt the point of view that the mathematical equations exactly describe the operation of a simplified physical model, which has been constructed (or conceived of) so as to embody the most important features of the actual process. Sometimes, the process of mathematical modeling involves the conceptual replacement of a discrete process by a continuous one. For instance, the number of members in an insect population changes by discrete amounts; however, if the population is large, it seems reasonable to consider it as a continuous variable and even to speak of its derivative. Step 2: Analysis of the Model. Once the problem has been formulated mathematically, you are often faced with the problem of solving one or more differential equations or, failing that, of finding out as much as possible about the properties of the solution. It may happen that this mathematical problem is quite difficult, and if so, further approximations may be indicated at this stage to make the problem mathematically tractable. For example, a nonlinear equation may be approximated by a linear one, or a slowly varying coefficient may be replaced by a constant. Naturally, any such approximations must also be examined from the physical point of view to make sure that the simplified mathematical problem still reflects the essential features of the physical process under investigation. At the same time, an intimate knowledge of the physics of the problem may suggest reasonable mathematical approximations that will make the mathematical problem more amenable to analysis. This interplay of understanding of physical phenomena and knowledge of mathematical techniques and their limitations is characteristic of applied mathematics at its best, and it is indispensable in successfully constructing useful mathematical models of intricate physical processes. Step 3: Comparison with Experiment or Observation. Finally, having obtained the solution (or at least some information about it), you must interpret this information in the context in which the problem arose. In particular, you should always check that the mathematical solution appears physically reasonable. If possible, calculate the values of the solution at selected points and compare them with experimentally observed values. Or ask whether the behavior of the solution after a long time is consistent with observations. Or examine the solutions corresponding to certain special values of parameters in the problem. Of course, the fact that the mathematical solution appears to be reasonable does not guarantee that it is correct. However, if the predictions of the mathematical model are seriously inconsistent with observations of the physical system it purports to describe, this suggests that errors have been made in solving the mathematical problem, that the mathematical model itself needs refinement, or that observations must be made with greater care. The examples in this section are typical of applications in which first-order differential equations arise.

EXAMPLE 1 | Mixing

▼

At time t = 0 a tank contains Q 0 lb of salt dissolved in 100 gal of water; see Figure 2.3.1. Assume 1 that water containing lb of salt per gallon is entering the tank at a rate of r gal/min and that the 4 well-stirred mixture is draining from the tank at the same rate. Set up the initial value problem that describes this flow process. Find the amount of salt Q( t) in the tank at any time, and also find the limiting amount Q L that is present after a very long time. If r = 3 and Q 0 = 2Q L , find the time T after which the salt level is within 2% of Q L . Also find the flow rate that is required if the value of T is not to exceed 45 min.

Boyce 9131 Ch02 2

September 29, 2016

17:16

41

2.3 Modeling with First-Order Differential Equations

▼ Solution: 1

r gal/min, 4 lb/gal

r gal/min

FIGURE 2.3.1 The water tank in Example 1.

We assume that salt is neither created nor destroyed in the tank. Therefore, variations in the amount of salt are due solely to the flows in and out of the tank. More precisely, the rate of change of salt in the tank, d Q/dt, is equal to the rate at which salt is flowing in minus the rate at which it is flowing out. In symbols, dQ = rate in − rate out. (1) dt The rate at which salt enters the tank is the concentration 14 lb/gal times the flow rate r gal/min, or r/4 lb/min. To find the rate at which salt leaves the tank, we need to multiply the concentration of salt in the tank by the rate of outflow, r gal/min. Since the rates of flow in and out are equal, the volume of water in the tank remains constant at 100 gal, and since the mixture is “well-stirred,” the concentration throughout the tank is the same, namely, Q( t) /100 lb/gal. Therefore, the rate at which salt leaves the tank is r Q( t) /100 lb/min. Thus the differential equation governing this process is dQ r rQ = − . (2) dt 4 100 The initial condition is Q( 0) = Q 0 .

(3)

Upon thinking about the problem physically, we might anticipate that eventually the mixture originally in the tank will be essentially replaced by the mixture flowing in, whose concentration is 1 lb/gal. Consequently, we might expect that ultimately the amount of salt in the tank would be very 4 close to 25 lb. We can also find the limiting amount Q L = 25 by setting d Q/dt equal to zero in equation (2) and solving the resulting algebraic equation for Q. To solve the initial value problem (2), (3) analytically, note that equation (2) is linear. (It is also separable, see Problem 24 in Section 2.2.) Rewriting the differential equation (2) in the standard form for a linear differential equation, we have dQ rQ r + = . dt 100 4

(4)

Thus the integrating factor is er t/100 and the general solution is Q( t) = 25 + ce−r t/100 ,

(5)

where c is an arbitrary constant. To satisfy the initial condition (3), we must choose c = Q 0 − 25. Therefore, the solution of the initial value problem (2), (3) is Q( t) = 25 + ( Q 0 − 25) e−r t/100 ,

(6)

Q( t) = 25( 1 − e−r t/100 ) + Q 0 e−r t/100 .

(7)

or

From either form of the solution, (6) or (7), you can see that Q( t) → 25 (lb) as t → ∞, so the limiting value Q L is 25, confirming our physical intuition.

▼

41

Boyce 9131 Ch02 2

42

September 29, 2016

17:16

42

CHAPTER 2 First-Order Differential Equations

▼

Further, Q( t) approaches the limit more rapidly as r increases. In interpreting the solution (7), note that the second term on the right-hand side is the portion of the original salt that remains at time t, while the first term gives the amount of salt in the tank as a consequence of the flow processes. Plots of the solution for r = 3 and for several values of Q 0 are shown in Figure 2.3.2. Q 50

40

30

20

10

20

40

60

80

100

t

FIGURE 2.3.2 Solutions of the initial value problem (2): d Q/dt = r/4 − r Q/100, Q( 0) = Q 0 for r = 3 and several values of Q 0 .

Now suppose that r = 3 and Q 0 = 2Q L = 50; then equation (6) becomes Q( t) = 25 + 25e−0.03t .

(8)

Since 2% of 25 is 0.5, we wish to find the time T at which Q( t) has the value 25.5. Substituting t = T and Q = 25.5 in equation (8) and solving for T , we obtain T =

ln( 50) ∼ = 130.4 (min). 0.03

(9)

To determine r so that T = 45, return to equation (6), set t = 45, Q 0 = 50, Q( t) = 25.5, and solve for r . The result is r=

100 ln 50 ∼ = 8.69 gal/min. 45

(10)

Since this example is hypothetical, the validity of the model is not in question. If the flow rates are as stated, and if the concentration of salt in the tank is uniform, then the differential equation (1) is an accurate description of the flow process. Although this particular example has no special significance, models of this kind are often used in problems involving a pollutant in a lake, or a drug in an organ of the body, for example, rather than a tank of salt water. In such cases the flow rates may not be easy to determine or may vary with time. Similarly, the concentration may be far from uniform in some cases. Finally, the rates of inflow and outflow may be different, which means that the variation of the amount of liquid in the problem must also be taken into account.

EXAMPLE 2 | Compound Interest

▼

Suppose that a sum of money, S0 , is deposited in a bank or money fund that pays interest at an annual rate r . The value S( t) of the investment at any time t depends on the frequency with which interest is compounded as well as on the interest rate. Financial institutions have various policies concerning compounding: some compound monthly, some weekly, and some even daily. Assume that compounding takes place continuously. Set up an initial value problem that describes the growth of the investment.

Boyce 9131 Ch02 2

September 29, 2016

17:16

43

2.3 Modeling with First-Order Differential Equations

▼ Solution: The rate of change of the value of the investment is d S/dt, and this quantity is equal to the rate at which interest accrues, which is the interest rate r times the current value of the investment S( t) . Thus dS = rS (11) dt is the differential equation that governs the process. If we let t denote the time, in years, since the original deposit, the corresponding initial condition is S( 0) = S0 .

(12)

Then the solution of the initial value problem (8) gives the balance S( t) in the account at any time t. This initial value problem is readily solved, since the differential equation (11) is both linear and separable. Consequently, by solving equations (11) and (12), we find that S( t) = S0 er t .

(13)

Thus a bank account with continuously compounding interest grows exponentially.

The model in Example 2 is easily extended to situations involving deposits or withdrawals in addition to the accrual of interest, dividends, or annual capital gains. If we assume that the deposits or withdrawals take place at a constant rate k, then equation (11) is replaced by dS = r S + k, dt or, in standard form, dS (14) − r S = k, dt where k is positive for deposits and negative for withdrawals. Equation (14) is linear with the integrating factor e−r t , so its general solution is k S( t) = cer t − , r where c is an arbitrary constant. To satisfy the initial condition (12), we must choose c = S0 + k/ r . Thus the solution of the initial value problem (10), (8) is k S( t) = S0 er t + ( er t − 1) . (15) r The first term in expression (15) is the part of S( t) that is due to the return accumulated on the initial amount S0 , and the second term is the part that is due to the deposit or withdrawal rate k. The advantage of stating the problem in this general way without specific values for S0 , r , or k lies in the generality of the resulting formula (15) for S( t) . With this formula we can readily compare the results of different investment programs or different rates of return. For instance, suppose that one opens an individual retirement account (IRA) at age 25 and makes annual investments of $2000 thereafter in a continuous manner. Assuming a rate of return of 8%, what will be the balance in the IRA at age 65? We have S0 = 0, r = 0.08, and k = $2000, and we wish to determine S( 40) . From equation (15) we have S( 40) = 25,000( e3.2 − 1) = $588,313.

(16)

It is interesting to note that the total amount invested is $80,000, so the remaining amount of $508,313 results from the accumulated return on the investment. The balance after 40 years is also fairly sensitive to the assumed rate. For instance, S( 40) = $508,948 if r = 0.075 and S( 40) = $681,508 if r = 0.085. Let us now examine the assumptions that have gone into the model. First, we have assumed that the return is compounded continuously and that additional capital is invested continuously. Neither of these is true in an actual financial situation. We have also assumed that the return rate r is constant for the entire period involved, whereas in fact it is likely to fluctuate considerably. Although we cannot reliably predict future rates, we can use solution (15) to determine the approximate effect of different rate projections. It is also possible to consider r and k in equation (14) to be functions of t rather than constants; in that case, of course, the solution may be much more complicated than equation (15). The initial value problem (10), (8) and the solution (15) can also be used to analyze a number of other financial situations, including annuities, mortgages, and automobile loans.

43

Boyce 9131 Ch02 2

44

September 29, 2016

17:16

44

CHAPTER 2 First-Order Differential Equations

Let us now compare the results from the model with continuously compounded interest (and no other deposits or withdrawals) with the corresponding situation in which compounding occurs at finite time intervals. If interest is compounded once a year, then after t years S( t) = S0 ( 1 + r ) t . If interest is compounded twice a year, then at the end of 6 months the value of the investment is S0 ( 1 + (r/2) ) , and at the end of 1 year it is S0 ( 1 + r/2) 2 . Thus, after t years, we have r 2t S( t) = S0 1 + . 2 In general, if interest is compounded m times per year, then

r mt S( t) = S0 1 + . m The relation between formulas (13) and (17) is clarified if we recall from calculus that

r mt lim S0 1 + = S0 er t . m m→∞

(17)

The same model applies equally well to more general investments in which dividends and perhaps capital gains can also accumulate, as well as interest. In recognition of this fact, we will from now on refer to r as the rate of return. Table 2.3.1 shows the effect of changing the frequency of compounding for a return rate r of 8%. The second and third columns are calculated from equation (17) for quarterly and daily compounding, respectively, and the fourth column is calculated from equation (13) for continuous compounding. The results show that the frequency of compounding is not particularly important in most cases. For example, during a 10-year period the difference between quarterly and continuous compounding is $17.50 per $1000 invested, or less than $2/year. The difference would be somewhat greater for higher rates of return and less for lower rates. From the first row in the table, we see that for the return rate r = 8%, the annual yield for quarterly compounding is 8.24% and for daily or continuous compounding it is 8.33%. T A B L E 2.3.1

Years

1 2 5 10 20 30 40

Growth of Capital at a Return Rate r = 8% for Several Modes of Compounding

S(t)/S(t0 ) From Equation (17) m=4 m = 365

1.0824 1.1717 1.4859 2.2080 4.8754 10.7652 23.7699

S(t)/S(t0 ) From Equation (13)

1.0833 1.1735 1.4918 2.2253 4.9522 11.0203 24.5239

1.0833 1.1735 1.4918 2.2255 4.9530 11.0232 24.5325

EXAMPLE 3 | Chemicals in a Pond Consider a pond that initially contains 10 million gallons of fresh water. Water containing an undesirable chemical flows into the pond at the rate of 5 million gallons per year, and the mixture in the pond flows out at the same rate. The concentration γ ( t) of chemical in the incoming water varies periodically with time according to the expression γ ( t) = 2 + sin( 2t) g/gal. Construct a mathematical model of this flow process and determine the amount of chemical in the pond at any time. Plot the solution and describe in words the effect of the variation in the incoming concentration. Solution: Since the incoming and outgoing flows of water are the same, the amount of water in the pond remains constant at 107 gal. Let us denote time by t, measured in years, and the chemical by Q( t) , measured in grams. This example is similar to Example 1, and the same inflow/outflow principle applies. Thus dQ = rate in − rate out, dt

▼

Boyce 9131 Ch02 2

September 29, 2016

17:16

45

2.3 Modeling with First-Order Differential Equations

▼ where “rate in” and “rate out” refer to the rates at which the chemical flows into and out of the pond, respectively. The rate at which the chemical flows in is given by

rate in = ( 5 × 106 ) gal/yr ( 2 + sin( 2t) ) g/gal.

(18)

7

The concentration of chemical in the pond is Q( t) /10 g/gal, so the rate of flow out is rate out = ( 5 × 106 ) gal/year ( Q( t) /107 ) g/gal = Q( t) /2 g/yr.

(19)

Thus we obtain the differential equation dQ Q( t) = ( 5 × 106 ) ( 2 + sin( 2t) ) − , (20) dt 2 where each term has the units of g/yr. To make the coefficients more manageable, it is convenient to introduce a new dependent variable defined by q( t) = Q( t) /106 , or Q( t) = 106 q( t) . This means that q( t) is measured in millions of grams, or megagrams (metric tons). If we make this substitution in equation (20), then each term contains the factor 106 , which can be canceled. If we also transpose the term involving q( t) to the left-hand side of the equation, we finally have dq 1 + q = 10 + 5 sin( 2t) . (21) dt 2 Originally, there is no chemical in the pond, so the initial condition is q( 0) = 0.

(22)

Equation (21) is linear, and although the right-hand side is a function of time, the coefficient of q is a constant. Thus the integrating factor is et/2 . Multiplying equation (21) by this factor and integrating the resulting equation, we obtain the general solution 40 10 q( t) = 20 − (23) cos( 2t) + sin( 2t) + ce−t/2 . 17 17 The initial condition (22) requires that c = −300/17, so the solution of the initial value problem (17), (18) is 10 300 −t/2 40 cos( 2t) + sin( 2t) − e . (24) q( t) = 20 − 17 17 17 A plot of the solution (24) is shown in Figure 2.3.3, along with the line q = 20 (shown in black). The exponential term in the solution is important for small t, but it diminishes rapidly as t increases. Later, the solution consists of an oscillation, due to the sin( 2t) and cos( 2t) terms, about the constant level q = 20. Note that if the sin( 2t) term were not present in equation (21), then q = 20 would be the equilibrium solution of that equation. q 24 20 16 12 8 4 2

4

6

8

10

12

14

16

18

20

t

FIGURE 2.3.3 Solution of the initial value problem (17), (18):

dq/dt + q/2 = 10 + 5 sin( 2t) , q( 0) = 0. Let us now consider the adequacy of the mathematical model itself for this problem. The model rests on several assumptions that have not yet been stated explicitly. In the first place, the amount of water in the pond is controlled entirely by the rates of flow in and out---none is lost by evaporation or by seepage into the ground, and none is gained by rainfall. The same is true of the chemical; it flows into and out of the pond, but none is absorbed by fish or other organisms living in the pond. In addition, we assume that the concentration of chemical in the pond is uniform throughout the pond. Whether the results obtained from the model are accurate depends strongly on the validity of these simplifying assumptions.

45

Boyce 9131 Ch02 2

46

September 29, 2016

17:16

46

CHAPTER 2 First-Order Differential Equations

EXAMPLE 4 | Escape Velocity A body of constant mass m is projected away from the earth in a direction perpendicular to the earth’s surface with an initial velocity v 0 . Assuming that there is no air resistance, but taking into account the variation of the earth’s gravitational field with distance, find an expression for the velocity during the ensuing motion. Also find the initial velocity that is required to lift the body to a given maximum altitude Amax above the surface of the earth, and find the least initial velocity for which the body will not return to the earth; the latter is the escape velocity. mgR 2 (R + x)2

R

m

x

FIGURE 2.3.4 A body in the earth’s gravitational field is pulled towards the center of the earth.

Solution: Let the positive x-axis point away from the center of the earth along the line of motion with x = 0 lying on the earth’s surface; see Figure 2.3.4. The figure is drawn horizontally to remind you that gravity is directed toward the center of the earth, which is not necessarily downward from a perspective away from the earth’s surface. The gravitational force acting on the body (that is, its weight) is inversely proportional to the square of the distance from the center of the earth and is given by w( x) = −k/( x + R) 2 , where k is a constant, R is the radius of the earth, and the minus sign signifies that w( x) is directed in the negative x direction. We know that on the earth’s surface w( 0) is given by −mg, where g is the acceleration due to gravity at sea level. Therefore, k = mg R 2 and w( x) = −

mg R 2 . ( R + x) 2

(25)

Since there are no other forces acting on the body, the equation of motion is m

dv mg R 2 =− , dt ( R + x) 2

(26)

and the initial condition is v( 0) = v 0 .

(27)

Unfortunately, equation (26) involves too many variables since it depends on t, x, and v. To remedy this situation, we can eliminate t from equation (26) by thinking of x, rather than t, as the independent variable. Then we can express dv/dt in terms of dv/d x by using the chain rule; hence dv d x dv dv = =v , dt d x dt dx and equation (26) is replaced by v

dv g R2 . =− dx ( R + x) 2

(28)

Equation (28) is separable but not linear, so by separating the variables and integrating, we obtain v2 g R2 = + c. 2 R+x

(29)

Since x = 0 when t = 0, the initial condition (27) at t = 0 can be replaced by the condition that 2 v = v 0 when x = 0. Hence c = ( v 0 /2) − g R and

v=±

▼

2

v 0 − 2g R +

2g R 2 . R+x

(30)

Note that equation (30) gives the velocity as a function of altitude rather than as a function of time. The plus sign must be chosen if the body is rising, and the minus sign must be chosen if it is falling back to earth.

Boyce 9131 Ch02 2

September 29, 2016

17:16

47

2.3 Modeling with First-Order Differential Equations

47

▼ To determine the maximum altitude Amax that the body reaches, we set v = 0 and x = Amax in equation (30) and then solve for Amax , obtaining

2

Amax =

v0 R 2

2g R − v 0

.

(31)

Solving equation (31) for v 0 , we find the initial velocity required to lift the body to the altitude Amax , namely,

v0 =

2g R

Amax . R + Amax

(32)

The escape velocity v e is then found by letting Amax → ∞. Consequently, ve =

2g R.

(33)

The numerical value of v e is approximately 6.9 mi/s, or 11.1 km/s. The preceding calculation of the escape velocity neglects the effect of air resistance, so the actual escape velocity (including the effect of air resistance) is somewhat higher. On the other hand, the effective escape velocity can be significantly reduced if the body is transported a considerable distance above sea level before being launched. Both gravitational and frictional forces are thereby reduced; air resistance, in particular, diminishes quite rapidly with increasing altitude. You should keep in mind also that it may well be impractical to impart too large an initial velocity instantaneously; space vehicles, for instance, receive their initial acceleration during a period of a few minutes.

Problems 1. Consider a tank used in certain hydrodynamic experiments. After one experiment the tank contains 200 L of a dye solution with a concentration of 1 g/L. To prepare for the next experiment, the tank is to be rinsed with fresh water flowing in at a rate of 2 L/min, the well-stirred solution flowing out at the same rate. Find the time that will elapse before the concentration of dye in the tank reaches 1% of its original value.

2. A tank initially contains 120 L of pure water. A mixture containing a concentration of γ g/L of salt enters the tank at a rate of 2 L/min, and the well-stirred mixture leaves the tank at the same rate. Find an expression in terms of γ for the amount of salt in the tank at any time t. Also find the limiting amount of salt in the tank as t → ∞. 3. A tank contains 100 gal of water and 50oz of salt. Water

1 1 1 + sin t oz/gal flows into 4 2 the tank at a rate of 2 gal/min, and the mixture in the tank flows out at the same rate. a. Find the amount of salt in the tank at any time. G b. Plot the solution for a time period long enough so that you see the ultimate behavior of the graph. c. The long-time behavior of the solution is an oscillation about a certain constant level. What is this level? What is the amplitude of the oscillation?

containing a salt concentration of

4. Suppose that a tank containing a certain liquid has an outlet near the bottom. Let h( t) be the height of the liquid surface above the outlet at time t. Torricelli’s2 principle states that the outflow velocity v at the outlet is equal to the velocity of a particle falling freely (with no drag) from the height h. ............................................................................................................................. 2 Evangelista

Torricelli (1608--1647), successor to Galileo as court mathematician in Florence, published this result in 1644. In addition to this work in fluid dynamics, he is also known for constructing the first mercury barometer and for making important contributions to geometry.

a. Show that v =

2gh, where g is the acceleration due to gravity. b. By equating the rate of outflow to the rate of change of liquid in the tank, show that h( t) satisfies the equation A( h)

dh = −α a 2gh, dt

(34)

where A( h) is the area of the cross section of the tank at height h and a is the area of the outlet. The constant α is a contraction coefficient that accounts for the observed fact that the cross section of the (smooth) outflow stream is smaller than a. The value of α for water is about 0.6. c. Consider a water tank in the form of a right circular cylinder that is 3m high above the outlet. The radius of the tank is 1m, and the radius of the circular outlet is 0.1m. If the tank is initially full of water, determine how long it takes to drain the tank down to the level of the outlet.

5. Suppose that a sum S0 is invested at an annual rate of return r compounded continuously. a. Find the time T required for the original sum to double in value as a function of r . b. Determine T if r = 7%. c. Find the return rate that must be achieved if the initial investment is to double in 8 years. 6. A young person with no initial capital invests k dollars per year at an annual rate of return r . Assume that investments are made continuously and that the return is compounded continuously. a. Determine the sum S( t) accumulated at any time t. b. If r = 7.5%, determine k so that $1 million will be available for retirement in 40 years. c. If k = $2000/year, determine the return rate r that must be obtained to have $1 million available in 40 years.

Boyce 9131 Ch02 2

48

September 29, 2016

17:16

48

CHAPTER 2 First-Order Differential Equations

7. A certain college graduate borrows $8000 to buy a car. The lender charges interest at an annual rate of 10%. Assuming that interest is compounded continuously and that the borrower makes payments continuously at a constant annual rate k, determine the payment rate k that is required to pay off the loan in 3 years. Also determine how much interest is paid during the 3-year period. N 8. A recent college graduate borrows $150,000 at an interest rate of 6% to purchase a condominium. Anticipating steady salary increases, the buyer expects to make payments at a monthly rate of 800 + 10t, where t is the number of months since the loan was made. a. Assuming that this payment schedule can be maintained, when will the loan be fully paid? b. Assuming the same payment schedule, how large a loan could be paid off in exactly 20 years?

9. An important tool in archeological research is radiocarbon dating, developed by the American chemist Willard F. Libby.3 This is a means of determining the age of certain wood and plant remains, and hence of animal or human bones or artifacts found buried at the same levels. Radiocarbon dating is based on the fact that some wood or plant remains contain residual amounts of carbon-14, a radioactive isotope of carbon. This isotope is accumulated during the lifetime of the plant and begins to decay at its death. Since the half-life of carbon-14 is long (approximately 5730 years),4 measurable amounts of carbon-14 remain after many thousands of years. If even a tiny fraction of the original amount of carbon-14 is still present, then by appropriate laboratory measurements the proportion of the original amount of carbon-14 that remains can be accurately determined. In other words, if Q( t) is the amount of carbon-14 at time t and Q 0 is the original amount, then the ratio Q( t) / Q 0 can be determined, as long as this quantity is not too small. Present measurement techniques permit the use of this method for time periods of 50,000 years or more. a. Assuming that Q satisfies the differential equation Q = −r Q, determine the decay constant r for carbon-14. b. Find an expression for Q( t) at any time t, if Q( 0) = Q 0 . c. Suppose that certain remains are discovered in which the current residual amount of carbon-14 is 20% of the original amount. Determine the age of these remains. N 10. Suppose that a certain population has a growth rate that varies with time and that this population satisfies the differential equation

dy y = ( 0.5 + sin t) . dt 5

a. If y( 0) = 1, find (or estimate) the time τ at which the population has doubled. Choose other initial conditions and determine whether the doubling time τ depends on the initial population. b. Suppose that the growth rate is replaced by its average value 1/10. Determine the doubling time τ in this case. c. Suppose that the term sin t in the differential equation is replaced by sin 2π t; that is, the variation in the growth rate has a substantially higher frequency. What effect does this have on the doubling time τ ? d. Plot the solutions obtained in parts a, b, and c on a single set of axes. .............................................................................................................................. 3 Willard

F. Libby (1908--1980) was born in rural Colorado and received his education at the University of California at Berkeley. He developed the method of radiocarbon dating beginning in 1947 while he was at the University of Chicago. For this work he was awarded the Nobel Prize in Chemistry in 1960.

4 McGraw-Hill

Encyclopedia of Science and Technology (8th ed.) (New York: McGraw-Hill, 1997), Vol. 5, p. 48.

N

11. Suppose that a certain population satisfies the initial value

problem dy/dt = r ( t) y − k,

y( 0) = y0 ,

where the growth rate r ( t) is given by r ( t) = ( 1 + sin t) /5, and k represents the rate of predation. G a. Suppose that k = 1/5. Plot y versus t for several values of y0 between 1/2 and 1. b. Estimate the critical initial population yc below which the population will become extinct. c. Choose other values of k and find the corresponding yc for each one. G d. Use the data you have found in parts b and c to plot yc versus k.

12. Newton’s law of cooling states that the temperature of an object changes at a rate proportional to the difference between its temperature and that of its surroundings. Suppose that the temperature of a cup of coffee obeys Newton’s law of cooling. If the coffee has a temperature of 200◦ F when freshly poured, and 1 min later has cooled to 190◦ F in a room at 70◦ F, determine when the coffee reaches a temperature of 150◦ F.

13. Heat transfer from a body to its surroundings by radiation, based on the Stefan--Boltzmann5 law, is described by the differential equation du (35) = −α ( u 4 − T 4 ) , dt where u( t) is the absolute temperature of the body at time t, T is the absolute temperature of the surroundings, and α is a constant depending on the physical parameters of the body. However, if u is much larger than T , then solutions of equation (35) are well approximated by solutions of the simpler equation du = −α u 4 . dt

(36)

Suppose that a body with initial temperature 2000 K is surrounded by a medium with temperature 300 K and that α = 2.0 × 10−12 K−3 /s. a. Determine the temperature of the body at any time by solving equation (36). G b. Plot the graph of u versus t. N c. Find the time τ at which u( τ ) = 600---that is, twice the ambient temperature. Up to this time the error in using equation (36) to approximate the solutions of equation (35) is no more than 1%. N 14. Consider an insulated box (a building, perhaps) with internal temperature u( t) . According to Newton’s law of cooling, u satisfies the differential equation

du = −k( u − T ( t) ) , dt

(37)

where T ( t) is the ambient (external) temperature. Suppose that T ( t) varies sinusoidally; for example, assume that T ( t) = T0 + T1 cos( ω t) . .............................................................................................................................. 5 Jozef Stefan (1835--1893), professor of physics at Vienna, stated the radiation law on empirical grounds in 1879. His student Ludwig Boltzmann (1844--1906) derived it theoretically from the principles of thermodynamics in 1884. Boltzmann is best known for his pioneering work in statistical mechanics.

Boyce 9131 Ch02 2

September 29, 2016

17:16

49

2.3 Modeling with First-Order Differential Equations

a. Solve equation (37) and express u( t) in terms of t, k, T0 , T1 , and ω . Observe that part of your solution approaches zero as t becomes large; this is called the transient part. The remainder of the solution is called the steady state; denote it by S( t) . G b. Suppose that t is measured in hours and that ω = π/12, corresponding to a period of 24 h for T ( t) . Further, let T0 = 60◦ F, T1 = 15◦ F, and k = 0.2/h. Draw graphs of S( t) and T ( t) versus t on the same axes. From your graph estimate the amplitude R of the oscillatory part of S( t) . Also estimate the time lag τ between corresponding maxima of T ( t) and S( t) . c. Let k, T0 , T1 , and ω now be unspecified. Write the oscillatory part of S( t) in the form R cos( ω ( t − τ ) ) . Use trigonometric identities to find expressions for R and τ . Let T1 and ω have the values given in part b, and plot graphs of R and τ versus k.

15. Consider a lake of constant volume V containing at time t an amount Q( t) of pollutant, evenly distributed throughout the lake with a concentration c( t) , where c( t) = Q( t) / V . Assume that water containing a concentration k of pollutant enters the lake at a rate r , and that water leaves the lake at the same rate. Suppose that pollutants are also added directly to the lake at a constant rate P. Note that the given assumptions neglect a number of factors that may, in some cases, be important---for example, the water added or lost by precipitation, absorption, and evaporation; the stratifying effect of temperature differences in a deep lake; the tendency of irregularities in the coastline to produce sheltered bays; and the fact that pollutants are deposited unevenly throughout the lake but (usually) at isolated points around its periphery. The results below must be interpreted in light of the neglect of such factors as these. a. If at time t = 0 the concentration of pollutant is c0 , find an expression for the concentration c( t) at any time. What is the limiting concentration as t → ∞? b. If the addition of pollutants to the lake is terminated ( k = 0 and P = 0 for t > 0), determine the time interval T that must elapse before the concentration of pollutants is reduced to 50% of its original value; to 10% of its original value. c. Table 2.3.2 contains data6 for several of the Great Lakes. Using these data, determine from part b the time T that is needed to reduce the contamination of each of these lakes to 10% of the original value.

T A B L E 2.3.2

Volume and Flow Data for the Great Lakes

Lake

103 × V (km3 )

r (km3 /year)

Superior Michigan Erie Ontario

12.2 4.9 0.46 1.6

65.2 158 175 209

N 16. A ball with mass 0.15 kg is thrown upward with initial velocity 20 m/s from the roof of a building 30 m high. Neglect air resistance. a. Find the maximum height above the ground that the ball reaches. b. Assuming that the ball misses the building on the way down, find the time that it hits the ground. G c. Plot the graphs of velocity and position versus time.

............................................................................................................................. 6

This problem is based on R. H. Rainey, “Natural Displacement of Pollution from the Great Lakes,” Science 155 (1967), pp. 1242--1243; the information in the table was taken from that source.

49

N 17. Assume that the conditions are as in Problem 16 except that there is a force due to air resistance of magnitude |v|/30 directed opposite to the velocity, where the velocity v is measured in m/s. a. Find the maximum height above the ground that the ball reaches. b. Find the time that the ball hits the ground. G c. Plot the graphs of velocity and position versus time. Compare these graphs with the corresponding ones in Problem 16. N 18. Assume that the conditions are as in Problem 16 except that there is a force due to air resistance of magnitude v 2 /1325 directed opposite to the velocity, where the velocity v is measured in m/s. a. Find the maximum height above the ground that the ball reaches. b. Find the time that the ball hits the ground. G c. Plot the graphs of velocity and position versus time. Compare these graphs with the corresponding ones in Problems 16 and 17.

19. A body of constant mass m is projected vertically upward with an initial velocity v 0 in a medium offering a resistance k|v|, where k is a constant. Neglect changes in the gravitational force.

a. Find the maximum height xm attained by the body and the time tm at which this maximum height is reached. b. Show that if kv 0 / mg < 1, then tm and xm can be expressed as

v0 tm = g 2

xm =

v0

2g

1 1 kv 0 + 1− 2 mg 3

1 2 kv 0 + 1− 3 mg 2

kv 0 mg kv 0 mg

2

− ··· ,

2

− ··· .

c. Show that the quantity kv 0 / mg is dimensionless. 20. A body of mass m is projected vertically upward with an initial velocity v 0 in a medium offering a resistance k|v|, where k is a constant. Assume that the gravitational attraction of the earth is constant.

a. Find the velocity v( t) of the body at any time. b. Use the result of part a to calculate the limit of v( t) as k → 0---that is, as the resistance approaches zero. Does this result agree with the velocity of a mass m projected upward with an initial velocity v 0 in a vacuum?

c. Use the result of part a to calculate the limit of v( t) as m → 0---that is, as the mass approaches zero. 21. A body falling in a relatively dense fluid, oil for example, is acted on by three forces (see Figure 2.3.5): a resistive force R, a buoyant force B, and its weight w due to gravity. The buoyant force is equal to the weight of the fluid displaced by the object. For a slowly moving spherical body of radius a, the resistive force is given by Stokes’s law, R = 6π μ a|v|, where v is the velocity of the body, and μ is the coefficient of viscosity of the surrounding fluid.7 ............................................................................................................................. 7 Sir George Gabriel Stokes (1819--1903) was born in Ireland but spent most of

his life at Cambridge University, first as a student and later as a professor. Stokes was one of the foremost applied mathematicians of the nineteenth century, best known for his work in fluid dynamics and the wave theory of light. The basic equations of fluid mechanics (the Navier--Stokes equations) are named partly in his honor, and one of the fundamental theorems of vector calculus bears his name. He was also one of the pioneers in the use of divergent (asymptotic) series.

Boyce 9131 Ch02 2

50

September 29, 2016

17:16

50

CHAPTER 2 First-Order Differential Equations

a. Determine v( t) and w( t) in terms of initial speed u and initial

a. Find the limiting velocity of a solid sphere of radius a and density ρ falling freely in a medium of density ρ and coefficient of viscosity μ . b. In 1910 R. A. Millikan8 studied the motion of tiny droplets of oil falling in an electric field. A field of strength E exerts a force Ee on a droplet with charge e. Assume that E has been adjusted so the droplet is held stationary ( v = 0) and that w and B are as given above. Find an expression for e. Millikan repeated this experiment many times, and from the data that he gathered he was able to deduce the charge on an electron. R

B

a

w FIGURE 2.3.5 A body falling in a dense fluid (see

Problem 21).

22. Let v( t) and w( t) be the horizontal and vertical components, respectively, of the velocity of a batted (or thrown) baseball. In the absence of air resistance, v and w satisfy the equations dv dw = 0, = −g. dt dt a. Show that v = u cos A, w = −gt + u sin A,

angle of elevation A.

b. Find x( t) and y( t) if x( 0) = 0 and y( 0) = h. G c. Plot the trajectory of the ball for r = 1/5, u = 125, h = 3, and for several values of A. How do the trajectories differ from those in Problem 22 with r = 0? d. Assuming that r = 1/5 and h = 3, find the minimum initial velocity u and the optimal angle A for which the ball will clear a wall that is 350 ft distant and 10 ft high. Compare this result with that in Problem 22f.

24. Brachistochrone Problem. One of the famous problems in the history of mathematics is the brachistochrone9 problem: to find the curve along which a particle will slide without friction in the minimum time from one given point P to another Q, the second point being lower than the first but not directly beneath it (see Figure 2.3.6). This problem was posed by Johann Bernoulli in 1696 as a challenge problem to the mathematicians of his day. Correct solutions were found by Johann Bernoulli and his brother Jakob Bernoulli and by Isaac Newton, Gottfried Leibniz, and the Marquis de L’Hôpital. The brachistochrone problem is important in the development of mathematics as one of the forerunners of the calculus of variations. In solving this problem, it is convenient to take the origin as the upper point P and to orient the axes as shown in Figure 2.3.6. The lower point Q has coordinates ( x0 , y0 ) . It is then possible to show that the curve of minimum time is given by a function y = φ ( x) that satisfies the differential equation (38) ( 1 + y 2 ) y = k 2 , where k is a certain positive constant to be determined later. a. Solve equation (38) for y . Why is it necessary to choose the positive square root? b. Introduce the new variable t by the relation 2

y = k 2 sin2 t. Show that the equation found in part a then takes the form

where u is the initial speed of the ball and A is its initial angle of elevation. b. Let x( t) and y( t) be the horizontal and vertical coordinates, respectively, of the ball at time t. If x( 0) = 0 and y( 0) = h, find x( t) and y( t) at any time t. G c. Let g = 32 ft/s2 , u = 125 ft/s, and h = 3 ft. Plot the trajectory of the ball for several values of the angle A; that is, plot x( t) and y( t) parametrically. d. Suppose the outfield wall is at a distance L and has height H . Find a relation between u and A that must be satisfied if the ball is to clear the wall. e. Suppose that L = 350 ft and H = 10 ft. Using the relation in part (d), find (or estimate from a plot) the range of values of A that correspond to an initial velocity of u = 110 ft/s. f. For L = 350 and H = 10, find the minimum initial velocity u and the corresponding optimal angle A for which the ball will clear the wall. N 23. A more realistic model (than that in Problem 22) of a baseball in flight includes the effect of air resistance. In this case the equations of motion are dv dw = −r v, = −g − r w, dt dt where r is the coefficient of resistance. .............................................................................................................................. 8 Robert

A. Millikan (1868--1953) was educated at Oberlin College and Columbia University. Later he was a professor at the University of Chicago and California Institute of Technology. His determination of the charge on an electron was published in 1910. For this work, and for other studies of the photoelectric effect, he was awarded the Nobel Prize for Physics in 1923.

(39)

2k 2 sin2 t dt = d x.

(40)

c. Letting θ = 2t, show that the solution of equation (40) for which x = 0 when y = 0 is given by x = k 2 ( θ − sin θ ) /2,

y = k 2 ( 1 − cos θ ) /2.

(41)

Equations (41) are parametric equations of the solution of equation (38) that passes through ( 0, 0) . The graph of equations (41) is called a cycloid. d. If we make a proper choice of the constant k, then the cycloid also passes through the point ( x0 , y0 ) and is the solution of the brachistochrone problem. Find k if x0 = 1 and y0 = 2. P

x

Q(x 0, y0)

y FIGURE 2.3.6 The brachistochrone (see Problem 24). .............................................................................................................................. 9 The

word “brachistochrone” comes from the Greek words brachistos, meaning shortest, and chronos, meaning time.

Boyce 9131 Ch02 2

September 29, 2016

17:16

51

2.4 Differences Between Linear and Nonlinear Differential Equations

Differences Between Linear and Nonlinear Differential Equations 2.4

Up to now, we have been primarily concerned with showing that first-order differential equations can be used to investigate many different kinds of problems in the natural sciences, and with presenting methods of solving such equations if they are either linear or separable. Now it is time to turn our attention to some more general questions about differential equations and to explore in more detail some important ways in which nonlinear equations differ from linear ones. Existence and Uniqueness of Solutions. So far, we have discussed a number of initial value problems, each of which had a solution and apparently only one solution. That raises the question of whether this is true of all initial value problems for first-order equations. In other words, does every initial value problem have exactly one solution? This may be an important question even for nonmathematicians. If you encounter an initial value problem in the course of investigating some physical problem, you might want to know that it has a solution before spending very much time and effort in trying to find it. Further, if you are successful in finding one solution, you might be interested in knowing whether you should continue a search for other possible solutions or whether you can be sure that there are no other solutions. For linear equations, the answers to these questions are given by the following fundamental theorem.

Theorem 2.4.1 | Existence and Uniqueness Theorem for First-Order Linear Equations If the functions p and g are continuous on an open interval I : α < t < β containing the point t = t0 , then there exists a unique function y = φ ( t) that satisfies the differential equation y + p( t) y = g( t)

(1)

for each t in I , and that also satisfies the initial condition y( t0 ) = y0 ,

(2)

where y0 is an arbitrary prescribed initial value.

Observe that Theorem 2.4.1 states that the given initial value problem has a solution and also that the problem has only one solution. In other words, the theorem asserts both the existence and the uniqueness of the solution of the initial value problem (1). In addition, it states that the solution exists throughout any interval I containing the initial point t0 in which the coefficients p and g are continuous. That is, the solution can be discontinuous or fail to exist only at points where at least one of p and g is discontinuous. Such points can often be identified at a glance. The proof of this theorem is partly contained in the discussion in Section 2.1 leading to the formula (see equation (32) in Section 2.1) μ ( t) y = μ ( t) g( t) dt + c, (3) where [equation (30) in Section 2.1] μ ( t) = exp

p( t) dt.

(4)

The derivation in Section 2.1 shows that if equation (1) has a solution, then it must be given by equation (3). By looking slightly more closely at that derivation, we can also conclude that the differential equation (1) must indeed have a solution. Since p is continuous for α < t < β , it follows that on the interval α < t < β , μ is defined, is a differentiable function, and is

51

Boyce 9131 Ch02 2

52

September 29, 2016

17:16

52

CHAPTER 2 First-Order Differential Equations

nonzero. Upon multiplying equation (1) by μ ( t) , we obtain ( μ ( t) y) = μ ( t) g( t) .

(5)

Since both μ and g are continuous, the function μg is integrable, and equation (3) follows from equation (5). Further, the integral of μg is differentiable, so y as given by equation (3) exists and is differentiable throughout the interval α < t < β . By substituting the expression for y from equation (3) into either equation (1) or equation (5), you can verify that this expression satisfies the differential equation throughout the interval α < t < β . Finally, the initial condition (2) determines the constant c uniquely, so there is only one solution of the initial value problem; this completes the proof. Equation (4) determines the integrating factor μ ( t) only up to a multiplicative factor that depends on the lower limit of integration. If we choose this lower limit to be t0 , then t μ ( t) = exp p( s) ds, (6) t0

and it follows that μ ( t0 ) = 1. Using the integrating factor given by equation (6), and choosing the lower limit of integration in equation (3) also to be t0 , we obtain the general solution of equation (1) in the form t 1 y= μ ( s) g( s) ds + c . (7) μ ( t) t0 To satisfy the initial condition (2), we must choose c = y0 . Thus the solution of the initial value problem (1) is t 1 y= μ ( s) g( s) ds + y0 , (8) μ ( t) t0 where μ ( t) is given by equation (6). Turning now to nonlinear differential equations, we must replace Theorem 2.4.1 by a more general theorem, such as the one that follows.

Theorem 2.4.2 | Existence and Uniqueness Theorem for First-Order Nonlinear Equations Let the functions f and ∂ f /∂ y be continuous in some rectangle α < t < β , γ < y < δ containing the point ( t0 , y0 ) . Then, in some interval t0 − h < t < t0 + h contained in α < t < β , there is a unique solution y = φ ( t) of the initial value problem y = f ( t, y) ,

y( t0 ) = y0 .

(9)

Observe that the hypotheses in Theorem 2.4.2 reduce to those in Theorem 2.4.1 if the differential equation is linear. In this case ∂ f ( t, y) f ( t, y) = − p( t) y + g( t) and = − p( t) , ∂y ∂f so the continuity of f and is equivalent to the continuity of p and g. ∂y The proof of Theorem 2.4.1 was comparatively simple because it could be based on the expression (3) that gives the solution of an arbitrary linear equation. There is no corresponding expression for the solution of the differential equation (9), so the proof of Theorem 2.4.2 is much more difficult. It is discussed to some extent in Section 2.8 and in greater depth in more advanced books on differential equations. We note that the conditions stated in Theorem 2.4.2 are sufficient to guarantee the existence of a unique solution of the initial value problem (6) in some interval ( t0 − h, t0 + h) , but they are not necessary. That is, the conclusion remains true under slightly weaker hypotheses about the function f . In fact, the existence of a solution (but not its uniqueness) can be established on the basis of the continuity of f alone. An important geometrical consequence of the uniqueness parts of Theorems 2.4.1 and 2.4.2 is that the graphs of two solutions cannot intersect each other. Otherwise, there would

Boyce 9131 Ch02 2

September 29, 2016

17:16

53

2.4 Differences Between Linear and Nonlinear Differential Equations

be two solutions that satisfy the initial condition corresponding to the point of intersection, in contradiction to Theorem 2.4.1 or 2.4.2. We now consider some examples.

EXAMPLE 1 Use Theorem 2.4.1 to find an interval in which the initial value problem t y + 2y = 4t 2 ,

(10)

y( 1) = 2

(11)

has a unique solution. Then do the same when the initial condition is changed to y( −1) = 2. Solution: Rewriting equation (10) in the standard form (1), we have y + ( 2/ t) y = 4t, so p( t) = 2/ t and g( t) = 4t. Thus, for this equation, g is continuous for all t, while p is continuous only for t < 0 or for t > 0. The interval t > 0 contains the initial point; consequently, Theorem 2.4.1 guarantees that the problem (7), (8) has a unique solution on the interval 0 < t < ∞. In Example 4 of Section 2.1 we found the solution of this initial value problem to be y = t2 +

1 , t2

t > 0.

(12)

Now suppose that the initial condition (11) is changed to y( −1) = 2. Then Theorem 2.4.1 asserts the existence of a unique solution for t < 0. As you can readily verify, the solution is again given by equation (12), but now on the interval t < 0.

EXAMPLE 2 Apply Theorem 2.4.2 to the initial value problem 3x 2 + 4x + 2 dy = , dx 2( y − 1)

y( 0) = −1.

(13)

Repeat this analysis when the initial condition is changed to y( 0) = 1. Solution: Note that Theorem 2.4.1 is not applicable to this problem since the differential equation is nonlinear. To apply Theorem 2.4.2, observe that f ( x, y) =

3x 2 + 4x + 2 , 2( y − 1)

∂f 3x 2 + 4x + 2 ( x, y) = − . ∂y 2( y − 1) 2

Thus each of these functions is continuous everywhere except on the line y = 1. Consequently, a rectangle can be drawn about the initial point ( 0, −1) in which both f and ∂ f /∂ y are continuous. Therefore, Theorem 2.4.2 guarantees that the initial value problem has a unique solution in some interval about x = 0. However, even though the rectangle can be stretched infinitely far in both the positive and the negative x directions, this does not necessarily mean that the solution exists for all x. Indeed, the initial value problem (9) was solved in Example 2 of Section 2.2, and the solution exists only for x > −2. Now suppose we change the initial condition to y( 0) = 1. The initial point now lies on the line y = 1, so no rectangle can be drawn about it within which f and ∂ f /∂ y are continuous. Consequently, Theorem 2.4.2 says nothing about possible solutions of this modified problem. However, if we separate the variables and integrate, as in Section 2.2, we find that y 2 − 2y = x 3 + 2x 2 + 2x + c. Further, if x = 0 and y = 1, then c = −1. Finally, by solving for y, we obtain y =1±

▼

x 3 + 2x 2 + 2x.

(14)

53

Boyce 9131 Ch02 2

54

September 29, 2016

17:16

54

CHAPTER 2 First-Order Differential Equations

▼ Equation (14) provides two functions that satisfy the given differential equation for x > 0 and also satisfy the initial condition y( 0) = 1. The fact that there are two solutions to this initial value problem reinforces the conclusion that Theorem 2.4.2 does not apply to this initial value problem.

EXAMPLE 3 Consider the initial value problem y = y 1/3 , y( 0) = 0 for t ≥ 0. Apply Theorem 2.4.2 to this initial value problem and then solve the problem.

(15)

Solution: 1 ∂f = y −2/3 does not exist when ∂y 3 y = 0, and hence it is not continuous there. Thus Theorem 2.4.2 does not apply to this problem, and no conclusion can be drawn from it. However, by the remark following Theorem 2.4.2, the continuity of f does ensure the existence of solutions, though not their uniqueness. To understand the situation more clearly, we must actually solve the problem, which is easy to do since the differential equation is separable. Thus we have y −1/3 dy = dt, The function f ( t, y) = y 1/3 is continuous everywhere, but

so 3 2/3 =t +c y 2 and

y=

2 ( t + c) 3

The initial condition is satisfied if c = 0, so y = φ 1 ( t) =

2 t 3

3/2 .

3/2 t ≥0

,

(16)

satisfies both of equations (15). On the other hand, the function

2 t y = φ 2 ( t) = − 3

3/2

,

t ≥0

(17)

is also a solution of the initial value problem. Moreover, the function y = ψ ( t) = 0, t ≥ 0

(18)

is yet another solution. Indeed, for an arbitrary positive t0 , the functions y = χ( t) =

⎧ ⎨0,

⎩±

2 (t 3

− t0 )

if 0 ≤ t < t0 ,

3/2 ,

(19)

if t ≥ t0

are continuous, are differentiable (in particular at t = t0 ), and are solutions of the initial value problem (11). Hence this problem has an infinite family of solutions; see Figure 2.4.1, where a few of these solutions are shown. y

χ (t) 1

φ1(t) ψ (t)

1

–1

2

φ 2(t)

FIGURE 2.4.1 Several solutions of the initial value problem

▼

y = y 1/3, y( 0) = 0.

t

χ (t)

Boyce 9131 Ch02 2

September 29, 2016

17:16

55

2.4 Differences Between Linear and Nonlinear Differential Equations

▼

As already noted, the nonuniqueness of the solutions of the problem (11) does not contradict the existence and uniqueness theorem, since Theorem 2.4.2 is not applicable if the initial point lies on the t-axis. If ( t0 , y0 ) is any point not on the t-axis, however, then the theorem guarantees that there is a unique solution of the differential equation y = y 1/3 passing through ( t0 , y0 ) .

Interval of Existence. According to Theorem 2.4.1, the solution of a linear equation (1) y + p( t) y = g( t) , subject to the initial condition y( t0 ) = y0 , exists throughout any interval about t = t0 in which the functions p and g are continuous. Thus vertical asymptotes or other discontinuities in the solution can occur only at points of discontinuity of p or g. For instance, the solutions in Example 1 (with one exception) are asymptotic to the y-axis, corresponding to the discontinuity at t = 0 in the coefficient p( t) = 2/ t, but none of the solutions has any other point where it fails to exist and to be differentiable. The one exceptional solution shows that solutions may sometimes remain continuous even at points of discontinuity of the coefficients. On the other hand, for a nonlinear initial value problem satisfying the hypotheses of Theorem 2.4.2, the interval in which a solution exists may be difficult to determine. The solution y = φ ( t) is certain to exist as long as the point ( t, φ ( t) ) remains within a region in which the hypotheses of Theorem 2.4.2 are satisfied. This is what determines the value of h in that theorem. However, since φ ( t) is usually not known, it may be impossible to locate the point ( t, φ ( t) ) with respect to this region. In any case, the interval in which a solution exists may have no simple relationship to the function f in the differential equation y = f ( t, y) . This is illustrated by the following example.

EXAMPLE 4 Solve the initial value problem y = y2,

y( 0) = 1,

(20)

and determine the interval in which the solution exists. Solution: ∂f = 2y ∂y are continuous everywhere. To find the solution, we separate the variables and integrate with the result that Theorem 2.4.2 guarantees that this problem has a unique solution since f ( t, y) = y 2 and

y −2 dy = dt

(21)

and −y −1 = t + c. Then, solving for y, we have y=−

1 . t +c

(22)

To satisfy the initial condition, we must choose c = −1, so y=

1 1−t

(23)

is the solution of the given initial value problem. Clearly, the solution becomes unbounded as t → 1; therefore, the solution exists only in the interval −∞ < t < 1. There is no indication from the differential equation itself, however, that the point t = 1 is in any way remarkable. Moreover, if the initial condition is replaced by y( 0) = y0 ,

▼

(24)

then the constant c in equation (22) must be chosen to be c = −1/ y0 ( y0 = 0) , and it follows that y0 (25) y= 1 − y0 t

55

Boyce 9131 Ch02 2

56

September 29, 2016

17:16

56

CHAPTER 2 First-Order Differential Equations

▼ is the solution of the initial value problem with the initial condition (24). Observe that the solution

(25) becomes unbounded as t → 1/ y0 , so the interval of existence of the solution is −∞ < t < 1/ y0 if y0 > 0, and is 1/ y0 < t < ∞ if y0 < 0. This example illustrates another feature of initial value problems for nonlinear equations: the singularities of the solution may depend in an essential way on the initial conditions as well as on the differential equation.

General Solution. Another way in which linear and nonlinear equations differ concerns the concept of a general solution. For a first-order linear differential equation it is possible to obtain a solution containing one arbitrary constant, from which all possible solutions follow by specifying values for this constant. For nonlinear equations this may not be the case; even though a solution containing an arbitrary constant may be found, there may be other solutions that cannot be obtained by giving values to this constant. For instance, for the differential equation y = y 2 in Example 4, the expression in equation (22) contains an arbitrary constant but does not include all solutions of the differential equation. To show this, observe that the function y = 0 for all t is certainly a solution of the differential equation, but it cannot be obtained from equation (22) by assigning a value to c. In this example we might anticipate that something of this sort might happen, because to rewrite the original differential equation in the form (21), we must require that y is not zero. However, the existence of “additional” solutions is not uncommon for nonlinear equations; a less obvious example is given in Problem 18. Thus we will use the term “general solution” only when discussing linear equations. Implicit Solutions. Recall again that for an initial value problem for a first-order linear differential equation, equation (8) provides an explicit formula for the solution y = φ ( t) . As long as the necessary antiderivatives can be found, the value of the solution at any point can be determined merely by substituting the appropriate value of t into the equation. The situation for nonlinear equations is much less satisfactory. Usually, the best that we can hope for is to find an equation F( t, y) = 0

(26)

involving t and y that is satisfied by the solution y = φ ( t) . Even this can be done only for differential equations of certain particular types, of which separable equations are the most important. The equation (26) is called an integral, or first integral, of the differential equation, and (as we have already noted) its graph is an integral curve, or perhaps a family of integral curves. Equation (26), assuming it can be found, defines the solution implicitly; that is, for each value of t we must solve equation (26) to find the corresponding value of y. If equation (26) is simple enough, it may be possible to solve it for y by analytical means and thereby obtain an explicit formula for the solution. However, more frequently this will not be possible, and you will have to resort to a numerical calculation to determine (approximately) the value of y for a given value of t. Once several pairs of values of t and y have been calculated, it is often helpful to plot them and then to sketch the integral curve that passes through them. You should take advantage of the wide range of computational and graphical utilities available to carry out these calculations and to create the graph of one or more integral curves. Examples 2, 3, and 4 involve nonlinear problems in which it is easy to solve for an explicit formula for the solution y = φ ( t) . On the other hand, Examples 1 and 3 in Section 2.2 are cases in which it is better to leave the solution in implicit form and to use numerical means to evaluate it for particular values of the independent variable. The latter situation is more typical; unless the implicit relation is quadratic in y or has some other particularly simple form, it is unlikely that it can be solved exactly by analytical methods. Indeed, more often than not, it is impossible even to find an implicit expression for the solution of a first-order nonlinear equation. Graphical or Numerical Construction of Integral Curves. Because of the difficulty in obtaining exact analytical solutions of nonlinear differential equations, methods that yield approximate solutions or other qualitative information about solutions are of correspondingly greater importance. We have already described, in Section 1.1, how the direction field of a differential equation can be constructed. The direction field can often show the qualitative form of solutions and can also be helpful in identifying regions of the t y-plane where solutions exhibit interesting features that merit more detailed analytical or numerical investigation. Graphical methods for first-order differential equations are discussed further in Section 2.5.

Boyce 9131 Ch02 2

September 29, 2016

17:16

57

2.4 Differences Between Linear and Nonlinear Differential Equations

57

An introduction to numerical methods for first-order equations is given in Section 2.7, and a systematic discussion of numerical methods appears in Chapter 8. However, it is not necessary to study the numerical algorithms themselves in order to use effectively one of the many software packages that generate and plot numerical approximations to solutions of initial value problems. Summary. The linear equation y + p( t) y = g( t) has several nice properties that can be summarized in the following statements: 1. Assuming that the coefficients are continuous, there is a general solution, containing an arbitrary constant, that includes all solutions of the differential equation. A particular solution that satisfies a given initial condition can be picked out by choosing the proper value for the arbitrary constant. 2. There is an expression for the solution, namely, equation (7) or equation (8). Moreover, although it involves two integrations, the expression is an explicit one for the solution y = φ ( t) rather than an equation that defines φ implicitly. 3. The possible points of discontinuity, or singularities, of the solution can be identified (without solving the problem) merely by finding the points of discontinuity of the coefficients. Thus, if the coefficients are continuous for all t, then the solution also exists and is differentiable for all t. None of these statements are true, in general, of nonlinear equations. Although a nonlinear equation may well have a solution involving an arbitrary constant, there may also be other solutions. There is no general formula for solutions of nonlinear equations. If you are able to integrate a nonlinear equation, you are likely to obtain an equation defining solutions implicitly rather than explicitly. Finally, the singularities of solutions of nonlinear equations can usually be found only by solving the equation and examining the solution. It is likely that the singularities will depend on the initial condition as well as on the differential equation.

Problems In each of Problems 1 through 4, determine (without solving the problem) an interval in which the solution of the given initial value problem is certain to exist.

1. 2. 3. 4.

( t − 3) y + ( ln t) y = 2t,

y( 1) = 2

In each of Problems 13 through 16, draw a direction field and plot (or sketch) several solutions of the given differential equation. Describe how solutions appear to behave as t increases and how their behavior depends on the initial value y0 when t = 0. G

13. y = t y( 3 − y)

G

14. y = y( 3 − t y)

y( 2) = 3

G

15. y = −y( 3 − t y)

In each of Problems 5 through 8, state where in the t y-plane the hypotheses of Theorem 2.4.2 are satisfied.

G

16. y = t − 1 − y 2

y + ( tan t) y = sin t,

y( π ) = 0

( 4 − t 2 ) y + 2t y = 3t 2 , ( ln t) y + y = cot t,

y( −3) = 1

5. y = ( 1 − t 2 − y 2 ) 1/2 6. 7.

ln |t y| y = 1 − t 2 + y2 y = ( t 2 + y 2 ) 3/2

1 + t2 3y − y 2 In each of Problems 9 through 12, solve the given initial value problem and determine how the interval in which the solution exists depends on the initial value y0 .

8. y =

9. y = −4t/ y, y( 0) = y0 10. y = 2t y 2 , y( 0) = y0 11. y + y 3 = 0, y( 0) = y0 12. y =

t2 , y( 1 + t 3 )

y( 0) = y0

17. Consider the initial value problem y = y 1/3 , y( 0) = 0 from Example 3 in the text. a. Is there a solution that passes through the point ( 1, 1) ? If so, find it. b. Is there a solution that passes through the point ( 2, 1) ? If so, find it. c. Consider all possible solutions of the given initial value problem. Determine the set of values that these solutions have at t = 2.

18. a. Verify that both y1 ( t) = 1 − t and y2 ( t) = −t 2 /4 are solutions of the initial value problem

y =

−t +

t 2 + 4y , 2

Where are these solutions valid?

y( 2) = −1.

Boyce 9131 Ch02 2

58

September 29, 2016

17:16

58

CHAPTER 2 First-Order Differential Equations

b. Explain why the existence of two solutions of the given problem does not contradict the uniqueness part of Theorem 2.4.2. c. Show that y = ct + c2 , where c is an arbitrary constant, satisfies the differential equation in part a for t ≥ −2c. If c = −1, the initial condition is also satisfied, and the solution y = y1 ( t) is obtained. Show that there is no choice of c that gives the second solution y = y2 ( t) . 19. a. Show that φ ( t) = e2t is a solution of y − 2y = 0 and that y = cφ ( t) is also a solution of this equation for any value of the constant c. b. Show that φ ( t) = 1/ t is a solution of y + y 2 = 0 for t > 0, but that y = cφ ( t) is not a solution of this equation unless c = 0 or c = 1. Note that the equation of part b is nonlinear, while that of part a is linear.

20. Show that if y = φ ( t) is a solution of y + p( t) y = 0, then y = cφ ( t) is also a solution for any value of the constant c.

21. Let y = y1 ( t) be a solution of y + p( t) y = 0,

(27)

and let y = y2 ( t) be a solution of y + p( t) y = g( t) .

(28)

Show that y = y1 ( t) + y2 ( t) is also a solution of equation (28).

and is called a Bernoulli equation after Jakob Bernoulli. Problems 23 and 25 deal with equations of this type.

23. a. Solve Bernoulli’s equation when n = 0; when n = 1. b. Show that if n = 0, 1, then the substitution v = y 1−n reduces Bernoulli’s equation to a linear equation. This method of solution was formulated by Leibniz in 1696. In each of Problems 24 through 25, the given equation is a Bernoulli equation. In each case solve it by using the substitution mentioned in Problem 23b.

24. y = r y − ky 2 , r > 0 and k > 0. This equation is important in population dynamics and is discussed in detail in Section 2.5.

25. y = y − σ y 3 , > 0 and σ > 0. This equation occurs in the study of the stability of fluid flow. Discontinuous Coefficients. Linear differential equations sometimes occur in which one or both of the functions p and g have jump discontinuities. If t0 is such a point of discontinuity, then it is necessary to solve the equation separately for t < t0 and t > t0 . Afterward, the two solutions are matched so that y is continuous at t0 ; this is accomplished by a proper choice of the arbitrary constants. The following two problems illustrate this situation. Note in each case that it is impossible also to make y continuous at t0 .

26. Solve the initial value problem y + 2y = g( t) ,

22. a. Show that the solution (7) of the general linear equation (1)

y( 0) = 0,

can be written in the form y = cy1 ( t) + y2 ( t) ,

(29)

where

where c is an arbitrary constant. b. Show that y1 is a solution of the differential equation

y + p( t) y = 0,

g( t) =

1,

0 ≤ t ≤ 1,

0,

t > 1.

(30)

27. Solve the initial value problem

corresponding to g( t) = 0. c. Show that y2 is a solution of the full linear equation (1). We see later (for example, in Section 3.5) that solutions of higherorder linear equations have a pattern similar to equation (29).

y + p( t) y = 0,

Bernoulli Equations. Sometimes it is possible to solve a nonlinear equation by making a change of the dependent variable that converts it into a linear equation. The most important such equation has the form

where

p( t) =

y( 0) = 1,

2,

0 ≤ t ≤ 1,

1,

t > 1.

y + p( t) y = q( t) y n ,

Autonomous Differential Equations and Population Dynamics 2.5

An important class of first-order equations consists of those in which the independent variable does not appear explicitly. Such equations are called autonomous and have the form dy/dt = f ( y) .

(1)

We will discuss these equations in the context of the growth or decline of the population of a given species, an important issue in fields ranging from medicine to ecology to global economics. A number of other applications are mentioned in some of the problems. Recall that in Sections 1.1 and 1.2 we considered the special case of equation (1) in which f ( y) = ay +b. Equation (1) is separable, so the discussion in Section 2.2 is applicable to it, but the main purpose of this section is to show how geometric methods can be used to obtain important qualitative information directly from the differential equation without solving the equation. Of

Boyce 9131 Ch02 2

September 29, 2016

17:16

59

2.5 Autonomous Differential Equations and Population Dynamics

fundamental importance in this effort are the concepts of stability and instability of solutions of differential equations. These ideas were introduced informally in Chapter 1, but without using this terminology. They are discussed further here and will be examined in greater depth and in a more general setting in Chapter 9. Exponential Growth. Let y = φ ( t) be the population of the given species at time t. The simplest hypothesis concerning the variation of population is that the rate of change of y is proportional10 to the current value of y; that is, dy = r y, dt

(2)

where the constant of proportionality r is called the rate of growth or decline, depending on whether r is positive or negative. Here, we assume that the population is growing, so r > 0. Solving equation (2) subject to the initial condition11 y( 0) = y0 ,

(3)

y = y0 er t .

(4)

we obtain

Thus the mathematical model consisting of the initial value problem (1), (2) with r > 0 predicts that the population will grow exponentially for all time, as shown in Figure 2.5.1 for several values of y0 . Under ideal conditions, equation (4) has been observed to be reasonably accurate for many populations, at least for limited periods of time. However, it is clear that such ideal conditions cannot continue indefinitely; eventually, limitations on space, food supply, or other resources will reduce the growth rate and bring an end to uninhibited exponential growth. y 10 8 6 4 2 1/r

2/r

3/r

4/r t

FIGURE 2.5.1 Exponential growth: y versus t for dy/dt = r y (r > 0) .

Logistic Growth. To take account of the fact that the growth rate actually depends on the population, we replace the constant r in equation (2) by a function h( y) and thereby obtain the modified equation dy = h( y) y. (5) dt We now want to choose h( y) so that h( y) ∼ = r > 0 when y is small, h( y) decreases as y grows larger, and h( y) < 0 when y is sufficiently large. The simplest function that has these properties is h( y) = r − ay, where a is also a positive constant. Using this function in equation (5), we obtain dy = (r − ay) y. dt

(6)

......................................................................................................................................................................... 10 It was apparently the British economist Thomas Malthus (1766--1834) who first observed that many biological populations increase at a rate proportional to the population. His first paper on populations appeared in 1798. 11 In

this section, because the unknown function is a population, we assume y0 > 0.

59

Boyce 9131 Ch02 2

60

September 29, 2016

17:16

60

CHAPTER 2 First-Order Differential Equations

Equation (6) is known as the Verhulst12 equation or the logistic equation. It is often convenient to write the logistic equation in the equivalent form

dy y (7) =r 1− y, dt K where K = r/a. In this form, the constant r is called the intrinsic growth rate---that is, the growth rate in the absence of any limiting factors. The interpretation of K will become clear shortly. We will investigate the solutions of equation (7) in some detail later in this section. Before doing that, however, we will show how you can easily draw a qualitatively correct sketch of the solutions. The same methods also apply to the more general equation (1). We first seek solutions of equation (7) of the simplest possible type---that is, constant functions. For such a solution dy/dt = 0 for all t, so any constant solution of equation (7) must satisfy the algebraic equation

y r 1− y = 0. K Thus the constant solutions are y = φ 1 ( t) = 0 and y = φ 2 ( t) = K . These solutions are called equilibrium solutions of equation (7) because they correspond to no change or variation in the value of y as t increases. In the same way, any equilibrium solutions of the more general equation (1) can be found by locating the roots of f ( y) = 0. The zeros of f ( y) are also called critical points. To visualize other solutions of equation (7) and to sketch their graphs quickly, we start by drawing the graph of f ( y) versus y. In the case of equation (7), f ( y) = r ( 1 − y/ K ) y, so the graph is the parabola shown in Figure 2.5.2. The intercepts are ( 0, 0) and ( K , 0) , corresponding to the critical points of equation (7), and the vertex of the parabola is ( K /2, r K /4) . Observe that dy/dt > 0 for 0 < y < K . Therefore, y is an increasing function of t when y is in this interval; this is indicated by the rightward-pointing arrows near the y-axis in Figure 2.5.2. Similarly, if y > K , then dy/dt < 0; hence y is decreasing, as indicated by the leftward-pointing arrow in Figure 2.5.2.

f ( y) rK /4

(K/2, rK /4)

K/2

FIGURE 2.5.2

K

y

f ( y) versus y for dy/dt = r ( 1−y/ K ) y.

In this context the y-axis is often called the phase line, and it is reproduced in its more customary vertical orientation in Figure 2.5.3a. The dots at y = 0 and y = K are the critical points, or equilibrium solutions. The arrows again indicate that y is increasing whenever 0 < y < K and that y is decreasing whenever y > K . ......................................................................................................................................................................... 12 Pierre F. Verhulst (1804--1849) was a Belgian mathematician who introduced equation (6) as a model for human population growth in 1838. He referred to it as logistic growth, so equation (6) is often called the logistic equation. He was unable to test the accuracy of his model because of inadequate census data, and it did not receive much attention until many years later. Reasonable agreement with experimental data was demonstrated by R. Pearl (1930) for Drosophila melanogaster (fruit fly) populations and by G. F. Gause (1935) for Paramecium and Tribolium (flour beetle) populations.

Boyce 9131 Ch02 2

September 29, 2016

17:16

61

2.5 Autonomous Differential Equations and Population Dynamics

Further, from Figure 2.5.2, note that if y is near zero or K , then the slope f ( y) is near zero, so the solution curves are relatively flat. They become steeper as the value of y leaves the neighborhood of zero or K . To sketch the graphs of solutions of equation (7) in the t y-plane, we start with the equilibrium solutions y = φ 1 ( t) = 0 and y = φ 2 ( t) = K ; then we draw other curves that are increasing when 0 < y < K , decreasing when y > K , and flatten out as y approaches either of the values 0 or K . Thus the graphs of solutions of equation (7) must have the general shape shown in Figure 2.5.3b, regardless of the values of r and K . Figure 2.5.3b may seem to show that other solutions intersect the equilibrium solution y = K , but is this really possible? No, the uniqueness part of Theorem 2.4.2, the fundamental existence and uniqueness theorem, states that only one solution can pass through a given point in the t y-plane. Thus, although other solutions may be asymptotic to the equilibrium solution as t → ∞, they cannot intersect it at any finite time. Consequently, a solution that starts in the interval 0 < y < K remains in this interval for all time, and similarly for a solution that starts in K < y < ∞. y

y

y = φ 2(t) = K K

K

K/2 y = φ 1(t) = 0 0

t

(a)

(b)

FIGURE 2.5.3 Logistic growth: dy/dt = r ( 1 − y/ K ) y. (a) The phase line. (b) Plots

of y versus t.

To carry the investigation one step further, we can determine the concavity of the solution curves and the location of inflection points by finding d 2 y/dt 2 . From the differential equation (1), we obtain (using the chain rule) d2 y d dy dy d = = f ( y) = f ( y) = f ( y) f ( y) . 2 dt dt dt dt dt

(8)

The graph of y versus t is concave up when y > 0---that is, when f and f have the same sign. Similarly, it is concave down when y < 0, which occurs when f and f have opposite signs. The signs of f and f can be easily identified from the graph of f ( y) versus y. Inflection points may occur when f ( y) = 0. In the case of equation (7), solutions are concave up for 0 < y < K /2 where f is positive and increasing (see Figure 2.5.2), so that both f and f are positive. Solutions are also concave up for y > K where f is negative and decreasing (both f and f are negative). For K /2 < y < K , solutions are concave down since here f is positive and decreasing, so f is positive but f is negative. There is an inflection point whenever the graph of y versus t crosses the line y = K /2. The graphs in Figure 2.5.3b exhibit these properties. Finally, observe that K is the upper bound that is approached, but not exceeded, by growing populations starting below this value. Thus it is natural to refer to K as the saturation level, or the environmental carrying capacity, for the given species. A comparison of Figures 2.5.1 and 2.5.3b reveals that solutions of the nonlinear equation (7) are strikingly different from those of the linear equation (1), at least for large values of t. Regardless of the value of K ---that is, no matter how small the nonlinear term in

61

Boyce 9131 Ch02 2

62

September 29, 2016

17:16

62

CHAPTER 2 First-Order Differential Equations

equation (7)---solutions of that equation approach a finite value as t → ∞, whereas solutions of equation (1) grow (exponentially) without bound as t → ∞. Thus even a tiny nonlinear term in the differential equation (7) has a decisive effect on the solution for large t. In many situations it is sufficient to have the qualitative information about a solution y = φ ( t) of equation (7) that is shown in Figure 2.5.3b. This information was obtained entirely from the graph of f ( y) versus y and without solving the differential equation (7). However, if we wish to have a more detailed description of logistic growth---for example, if we wish to know the value of the population at some particular time---then we must solve equation (7) subject to the initial condition (3). Provided that y = 0 and y = K , we can write equation (7) in the form dy = r dt. ( 1 − y/ K ) y Using a partial fraction expansion on the left-hand side, we have 1 1/ K + dy = r dt. y 1 − y/ K Then, by integrating both sides, we obtain y ln |y| − ln1 − = r t + c, K

(9)

where c is an arbitrary constant of integration to be determined from the initial condition y( 0) = y0 . We have already noted that if 0 < y0 < K , then y remains in this interval for all time. Thus in this case we can remove the absolute value bars in equation (9), and by taking the exponential of both sides, we find that y = Cer t , 1 − ( y/ K )

(10)

where C = ec . In order to satisfy the initial condition y( 0) = y0 , we must choose C = y0 /( 1 − ( y0 / K ) ) . Using this value for C in equation (10) and solving for y (see Problem 10), we obtain y=

y0 K . y0 + ( K − y0 ) e−r t

(11)

We have derived the solution (11) under the assumption that 0 < y0 < K . If y0 > K , then the details of dealing with equation (9) are only slightly different, and we leave it to you to show that equation (11) is also valid in this case. Finally, note that equation (11) also contains the equilibrium solutions y = φ 1 ( t) = 0 and y = φ 2 ( t) = K corresponding to the initial conditions y0 = 0 and y0 = K , respectively. All the qualitative conclusions that we reached earlier by geometrical reasoning can be confirmed by examining the solution (11). In particular, if y0 = 0, then equation (11) requires that y( t) = 0 for all t. If y0 > 0, and if we let t → ∞ in equation (11), then we obtain lim y( t) =

t→∞

y0 K = K. y0

Thus, for each y0 > 0, the solution approaches the equilibrium solution y = φ 2 ( t) = K asymptotically as t → ∞. Therefore, we say that the constant solution φ 2 ( t) = K is an asymptotically stable solution of equation (7) or that the point y = K is an asymptotically stable equilibrium or critical point. After a long time, the population is close to the saturation level K regardless of the initial population size, as long as it is positive. Other solutions approach the equilibrium solution more rapidly as r increases. On the other hand, the situation for the equilibrium solution y = φ 1 ( t) = 0 is quite different. Even solutions that start very near zero grow as t increases and, as we have seen, approach K as t → ∞. We say that φ 1 ( t) = 0 is an unstable equilibrium solution or that y = 0 is an unstable equilibrium or critical point. This means that the only way to guarantee that the solution remains near zero is to make sure its initial value is exactly equal to zero.

Boyce 9131 Ch02 2

September 29, 2016

17:16

63

2.5 Autonomous Differential Equations and Population Dynamics

EXAMPLE 1 The logistic model has been applied to the natural growth of the halibut population in certain areas of the Pacific Ocean.13 Let y, measured in kilograms, be the biomass, that is, the total mass, of the halibut population, at time t. The parameters in the logistic equation are estimated to have the values r = 0.71/year and K = 80.5 × 106 kg. If the initial biomass is y0 = 0.25K , find the biomass 2 years later. Also find the time τ for which y( τ ) = 0.75K . Solution: It is convenient to scale the solution (11) to the carrying capacity K ; thus we write equation (11) in the form y0 / K y . = K ( y0 / K ) + ( 1 − y0 / K ) e−r t

(12)

Using the data given in the problem, we find that 0.25 y( 2) ∼ = = 0.5797. K 0.25 + 0.75e−1.42 Consequently, y( 2) ∼ = 46.7 × 106 kg. To find τ , the time when y0 / K = 0.75 we first solve equation (12) for t, obtaining ( y0 / K ) ( 1 − y/ K )

e−r t = hence

( y/ K ) 1 − y0 / K

1 t = − ln r

;

( y0 / K ) ( 1 − y/ K )

( y/ K ) 1 − y0 / K

.

(13)

Using the given values of r and y0 / K and setting y/ K = 0.75, we find that τ =−

( 0.25) ( 0.25) 1 1 ln = ln 9 ∼ = 3.095 years. 0.71 ( 0.75) ( 0.75) 0.71

The graphs of y/ K versus t for the given parameter values and for several initial conditions are shown in Figure 2.5.4. The green curve corresponds to the initial condition y0 = 0.25K . y/ K 1.75 1.50 1.25 1.00 0.75 0.50 0.25 1

2

3 τ ≅ 3.095

4

5

6

t

FIGURE 2.5.4 y/ K versus t for population model of halibut in the Pacific Ocean. The green curve satisfies the initial condition y( 0) / K = 0.25. The solution with y( 0) = 0.25 reaches 75% of the carrying capacity at time t = τ 3.095 years.

..................................................................................................................................................................................... 13 A good source of information on the population dynamics and economics involved in making efficient use of a renewable resource, with particular emphasis on fisheries, is the book by Clark listed in the references at the end of this chapter. The parameter values used here are given on page 53 of this book and were obtained from a study by H. S. Mohring.

63

Boyce 9131 Ch02 2

64

September 29, 2016

17:16

64

CHAPTER 2 First-Order Differential Equations

A Critical Threshold. We now turn to a consideration of the equation

dy y = −r 1 − y, dt T

(14)

where r and T are given positive constants. Observe that (except for replacing the parameter K by T ) this equation differs from the logistic equation (7) only in the presence of the minus sign on the right-hand side. However, as we will see, the solutions of equation (14) behave very differently from those of equation (7). For equation (14) the graph of f ( y) versus y is the parabola shown in Figure 2.5.5. The intercepts on the y-axis are the critical points y = 0 and y = T , corresponding to the equilibrium solutions y = φ 1 ( t) = 0 and y = φ 2 ( t) = T . If 0 < y < T , then dy/dt < 0, and y is positive and decreases as t increases. Thus φ 1 ( t) = 0 is an asymptotically stable equilibrium solution. On the other hand, if y > T , then dy/dt > 0, so that y is positive and increasing as t increases; thus φ 2 ( t) = T is an unstable equilibrium solution. Furthermore, the concavity of solutions can be determined by looking at the sign of y = f ( y) f ( y) ; see equation (8). Figure 2.5.5 clearly shows that f ( y) is negative for 0 < y < T /2 and positive for T /2 < y < T , so the graph of y versus t is concave up and concave down, respectively, in these intervals. Also, f ( y) and f ( y) are both positive for y > T , so the graph of y versus t is also concave up there. f ( y)

T /2

–rT /4 FIGURE 2.5.5

T

y

(T/2, –rT /4)

f ( y) versus y for dy/dt = −r ( 1 − y/ T ) y.

Figure 2.5.6(a) shows the phase line (the y-axis) for equation (14). The dots at y = 0 and y = T are the critical points, or equilibrium solutions, and the arrows indicate where solutions are either increasing or decreasing. Solution curves of equation (14) can now be sketched quickly, as follows. First draw the equilibrium solutions y = φ 1 ( t) = 0 and y = φ 2 ( t) = T . Then sketch curves in the strip 0 < y < T that are decreasing as t increases and change concavity as they cross the line y = T /2. Next draw some curves above y = T that increase more and more steeply as t and y increase. Make sure that all curves become flatter as y approaches either zero or T . The result is Figure 2.5.6(b), which is a qualitatively accurate sketch of solutions of equation (14) for any r and T . From this figure it appears that as time increases, y either approaches zero or grows without bound, depending on whether the initial value y0 is less than or greater than T . Thus T is a threshold level, below which growth does not occur. We can confirm the conclusions that we have reached through geometrical reasoning by solving the differential equation (14). This can be done by separating the variables and integrating, just as we did for equation (7). However, if we note that equation (14) can be obtained from equation (7) by replacing K by T and r by −r , then we can make the same substitutions in the solution (11) and thereby obtain y=

y0 T , y0 + ( T − y0 ) er t

(15)

which is the solution of equation (14) subject to the initial condition y( 0) = y0 . If 0 < y0 < T , then it follows from equation (15) that y → 0 as t → ∞. This agrees with our qualitative geometric analysis. If y0 > T , then the denominator on the right-hand

Boyce 9131 Ch02 2

September 29, 2016

17:16

65

2.5 Autonomous Differential Equations and Population Dynamics

y

y y = φ 2(t) = T

T

T

T /2 y = φ 1(t) = 0

0

t (a)

(b)

FIGURE 2.5.6 Growth with a threshold: dy/dt = −r ( 1 − y/ T ) y; y = T is an asymptotically unstable equilibrium, while y = 0 is asymptotically stable. (a) The phase line. (b) Plots of y versus t.

side of equation (15) is zero for a certain finite value of t. We denote this value by t ∗ and calculate it from ∗

y0 − ( y0 − T ) er t = 0, which gives (see Problem 12) t∗ =

y0 1 ln . r y0 − T

(16)

Thus, if the initial population y0 is above the threshold T , the threshold model predicts that the graph of y versus t has a vertical asymptote at t = t ∗ ; in other words, the population becomes unbounded in a finite time, whose value depends on y0 , T , and r . The existence and location of this asymptote were not apparent from the geometric analysis, so in this case the explicit solution yields additional important qualitative, as well as quantitative, information. The populations of some species exhibit the threshold phenomenon. If too few are present, then the species cannot propagate itself successfully and the population becomes extinct. However, if the population is larger than the threshold level, then further growth occurs. Of course, the population cannot become unbounded, so eventually equation (14) must be modified to take this into account. Critical thresholds also occur in other circumstances. For example, in fluid mechanics, equations of the form (7) or (14) often govern the evolution of a small disturbance y in a laminar (or smooth) fluid flow. For instance, if equation (14) holds and y < T , then the disturbance is damped out and the laminar flow persists. However, if y > T , then the disturbance grows larger and the laminar flow breaks up into a turbulent one. In this case T is referred to as the critical amplitude. Experimenters speak of keeping the disturbance level in a wind tunnel low enough so that they can study laminar flow over an airfoil, for example. Logistic Growth with a Threshold. As we mentioned in the last subsection, the threshold model (14) may need to be modified so that unbounded growth does not occur when y is above the threshold T . The simplest way to do this is to introduce another factor that will have the effect of making dy/dt negative when y is large. Thus we consider

dy y

y = −r 1 − 1− y, (17) dt T K where r > 0 and 0 < T < K . The graph of f ( y) versus y is shown in Figure 2.5.7. In this problem there are three critical points, y = 0, y = T , and y = K , corresponding to the equilibrium solutions y = φ 1 ( t) = 0, y = φ 2 ( t) = T , and y = φ 3 ( t) = K , respectively. From Figure 2.5.7 we observe that dy/dt > 0 for T < y < K , and consequently y is increasing there. Further, dy/dt < 0 for y < T and for y > K , so y is decreasing in these intervals. Consequently, the equilibrium solutions y = φ 1 ( t) = 0 and y = φ 3 ( t) = K are asymptotically stable, and the solution y = φ 2 ( t) = T is unstable.

65

Boyce 9131 Ch02 2

66

September 29, 2016

17:16

66

CHAPTER 2 First-Order Differential Equations

f ( y)

0

FIGURE 2.5.7

y2

T

y1

K

y

f ( y) versus y for dy/dt = −r ( 1 − y/ T ) ( 1 − y/ K ) y.

The phase line for equation (17) is shown in Figure 2.5.8a, and the graphs of some solutions are sketched in Figure 2.5.8b. You should make sure that you understand the relation between these two figures, as well as the relation between Figures 2.5.7 and 2.5.8a. From Figure 2.5.8b we see that if y starts below the threshold T , then y declines to ultimate extinction. On the other hand, if y starts above T , then y eventually approaches the carrying capacity K . The inflection points on the graphs of y versus t in Figure 2.5.8b correspond to the maximum and minimum points, y1 and y2 , respectively, on the graph of f ( y) versus y in Figure 2.5.7. These values can be obtained by differentiating the right-hand side of equation (17) with respect to y, setting the result equal to zero, and solving for y. We obtain y1, 2 = ( K + T ± K 2 − K T + T 2 ) /3, (18) where the plus sign yields y1 and the minus sign y2 . y

y

y = φ 3(t) = K K

K

y1

y = φ 2(t) = T T

T

y2

y = φ 1(t) = 0

0

t (a)

(b)

FIGURE 2.5.8 Logistic growth with a threshold: dy/dt = −r ( 1 − y/ T ) ( 1 − y/ K ) y; y = φ 1 ( t) = 0 and y = φ 3 ( t) = K are asymptotically stable equilibria and y = φ 2 ( t) = T is an asymptotically unstable equilibrium. (a) The phase line. (b) Plots of y versus t.

A model of this general sort apparently describes the population of the passenger pigeon,14 which was present in the United States in vast numbers until the late nineteenth century. It was heavily hunted for food and for sport, and consequently its numbers were drastically reduced by the 1880s. Unfortunately, the passenger pigeon could apparently breed successfully only when present in a large concentration, corresponding to a relatively high threshold T . Although a reasonably large number of individual birds remained alive in the late 1880s, there were not enough in any one place to permit successful breeding, and the population rapidly declined to extinction. The last passenger pigeon died in 1914. The precipitous decline in the passenger pigeon population from huge numbers to extinction in a few decades was one of the early factors contributing to a concern for conservation in this country.

......................................................................................................................................................................... 14 See,

for example, Oliver L. Austin, Jr., Birds of the World (New York: Golden Press, 1983), pp. 143--145.

Boyce 9131 Ch02 2

September 29, 2016

17:16

67

2.5 Autonomous Differential Equations and Population Dynamics

67

Problems Problems 1 through 4 involve equations of the form dy/dt = f ( y) . In each problem sketch the graph of f ( y) versus y, determine the critical (equilibrium) points, and classify each one as asymptotically stable or unstable. Draw the phase line, and sketch several graphs of solutions in the t y-plane. G

1. 2. 3. 4.

G G G

dy/dt = ay + by 2 ,

a > 0, b > 0, −∞ < y0 < ∞

dy/dt = y( y − 1) ( y − 2) , dy/dt = e y − 1,

y0 ≥ 0

−∞ < y0 < ∞

dy/dt = e−y − 1,

(11) of the logistic model by solving equation (10) for y.

11. In Example 1, complete the manipulations needed to arrive at equation (13). That is, solve the solution (11) for t.

12. Complete the derivation of the location of the vertical asymptote in the solution (15) when y0 > T . That is, derive formula (16) by finding the value of t when the denominator of the right-hand side of equation (15) is zero.

13. Complete the derivation of formula (18) for the locations of the

−∞ < y0 < ∞

5. Semistable Equilibrium Solutions. Sometimes a constant equilibrium solution has the property that solutions lying on one side of the equilibrium solution tend to approach it, whereas solutions lying on the other side depart from it (see Figure 2.5.9). In this case the equilibrium solution is said to be semistable. a. Consider the equation dy/dt = k( 1 − y) 2 ,

(19)

where k is a positive constant. Show that y = 1 is the only critical point, with the corresponding equilibrium solution φ ( t) = 1. G b. Sketch f ( y) versus y. Show that y is increasing as a function of t for y < 1 and also for y > 1. The phase line has upward-pointing arrows both below and above y = 1. Thus solutions below the equilibrium solution approach it, and those above it grow farther away. Therefore, φ ( t) = 1 is semistable. c. Solve equation (19) subject to the initial condition y( 0) = y0 and confirm the conclusions reached in part b. y

10. Complete the derivation of the explicit formula for the solution

inflection points of the solution of the logistic growth model with a threshold (17). Hint: Follow the steps outlined on p. 66.

14. Consider the equation dy/dt = f ( y) and suppose that y1 is a

critical point---that is, f ( y1 ) = 0. Show that the constant equilibrium solution φ ( t) = y1 is asymptotically stable if f ( y1 ) < 0 and unstable if f ( y1 ) > 0.

15. Suppose that a certain population obeys the logistic equation dy/dt = r y( 1 − ( y/ K ) ) . a. If y0 = K /3, find the time τ at which the initial population has doubled. Find the value of τ corresponding to r = 0.025 per year. b. If y0 / K = α , find the time T at which y( T ) / K = β , where 0 < α , β < 1. Observe that T → ∞ as α → 0 or as β → 1. Find the value of T for r = 0.025 per year, α = 0.1, and β = 0.9. G 16. Another equation that has been used to model population growth is the Gompertz15 equation

φ (t) = k k

k

φ (t) = k

t

(a)

(b)

t

FIGURE 2.5.9 In both cases the equilibrium solution φ ( t) = k

Problems 6 through 9 involve equations of the form dy/dt = f ( y) . In each problem sketch the graph of f ( y) versus y, determine the critical (equilibrium) points, and classify each one as asymptotically stable, unstable, or semistable (see Problem 5). Draw the phase line, and sketch several graphs of solutions in the t y-plane.

6. dy/dt = y 2 ( y 2 − 1) ,

G

7. dy/dt = y( 1 − y 2 ) ,

G

8. dy/dt = y ( 4 − y ) , 9. dy/dt = y 2 ( 1 − y) 2 ,

G

2

2

−∞ < y0 < ∞ −∞ < y0 < ∞

,

where r and K are positive constants. a. Sketch the graph of f ( y) versus y, find the critical points, and determine whether each is asymptotically stable or unstable. b. For 0 ≤ y ≤ K , determine where the graph of y versus t is concave up and where it is concave down. c. For each y in 0 < y ≤ K , show that dy/dt as given by the Gompertz equation is never less than dy/dt as given by the logistic equation.

17. a. Solve the Gompertz equation

is semistable. (a) dy/dt ≤ 0; (b) dy/dt ≥ 0.

G

K y

dy = r y ln dt

y

dy = r y ln dt

K y

,

subject to the initial condition y( 0) = y0 . Hint: You may wish to let u = ln( y/ K ) . b. For the data given in Example 1 in the text (r = 0.71 per year, K = 80.5 × 106 kg, y0 / K = 0.25), use the Gompertz model to find the predicted value of y( 2) . c. For the same data as in part b, use the Gompertz model to find the time τ at which y( τ ) = 0.75K .

−∞ < y0 < ∞

.............................................................................................................................

−∞ < y0 < ∞

15 Benjamin

Gompertz (1779--1865) was an English actuary. He developed his model for population growth, published in 1825, in the course of constructing mortality tables for his insurance company.

Boyce 9131 Ch02 2

68

September 29, 2016

17:16

68

CHAPTER 2 First-Order Differential Equations

18. A pond forms as water collects in a conical depression of radius a and depth h. Suppose that water flows in at a constant rate k and is lost through evaporation at a rate proportional to the surface area. a. Show that the volume V ( t) of water in the pond at time t satisfies the differential equation dV = k − α π ( 3a/π h) 2/3 V 2/3 , dt where α is the coefficient of evaporation. b. Find the equilibrium depth of water in the pond. Is the equilibrium asymptotically stable? c. Find a condition that must be satisfied if the pond is not to overflow. Harvesting a Renewable Resource. Suppose that the population y of a certain species of fish (for example, tuna or halibut) in a given area of the ocean is described by the logistic equation

dy y =r 1− y. dt K Although it is desirable to utilize this source of food, it is intuitively clear that if too many fish are caught, then the fish population may be reduced below a useful level and possibly even driven to extinction. Problems 19 and 20 explore some of the questions involved in formulating a rational strategy for managing the fishery.16

19. At a given level of effort, it is reasonable to assume that the rate at which fish are caught depends on the population y: the more fish there are, the easier it is to catch them. Thus we assume that the rate at which fish are caught is given by E y, where E is a positive constant, with units of 1/time, that measures the total effort made to harvest the given species of fish. To include this effect, the logistic equation is replaced by

y

dy =r 1− y − E y. (20) dt K This equation is known as the Schaefer model after the biologist M. B. Schaefer, who applied it to fish populations. a. Show that if E < r , then there are two equilibrium points, y1 = 0 and y2 = K ( 1 − E/ r ) > 0. b. Show that y = y1 is unstable and y = y2 is asymptotically stable. c. A sustainable yield Y of the fishery is a rate at which fish can be caught indefinitely. It is the product of the effort E and the asymptotically stable population y2 . Find Y as a function of the effort E; the graph of this function is known as the yield--effort curve. d. Determine E so as to maximize Y and thereby find the maximum sustainable yield Ym .

20. In this problem we assume that fish are caught at a constant rate h independent of the size of the fish population. Then y satisfies

dy y y − h. =r 1− dt K

(21)

The assumption of a constant catch rate h may be reasonable when y is large but becomes less so when y is small. a. If h < r K /4, show that equation (21) has two equilibrium points y1 and y2 with y1 < y2 ; determine these points. b. Show that y1 is unstable and y2 is asymptotically stable. c. From a plot of f ( y) versus y, show that if the initial population y0 > y1 , then y → y2 as t → ∞, but that if .............................................................................................................................. 16 An

excellent treatment of this kind of problem, which goes far beyond what is outlined here, may be found in the book by Clark mentioned previously, especially in the first two chapters. Numerous additional references are given there.

y0 < y1 , then y decreases as t increases. Note that y = 0 is not an equilibrium point, so if y0 < y1 , then extinction will be reached in a finite time. d. If h > r K /4, show that y decreases to zero as t increases, regardless of the value of y0 . e. If h = r K /4, show that there is a single equilibrium point y = K /2 and that this point is semistable (see Problem 5). Thus the maximum sustainable yield is h m = r K /4, corresponding to the equilibrium value y = K /2. Observe that h m has the same value as Ym in Problem 19d. The fishery is considered to be overexploited if y is reduced to a level below K /2. Epidemics. The use of mathematical methods to study the spread of contagious diseases goes back at least to some work by Daniel Bernoulli in 1760 on smallpox. In more recent years many mathematical models have been proposed and studied for many different diseases.17 Problems 21 through 23 deal with a few of the simpler models and the conclusions that can be drawn from them. Similar models have also been used to describe the spread of rumors and of consumer products.

21. Suppose that a given population can be divided into two parts: those who have a given disease and can infect others, and those who do not have it but are susceptible. Let x be the proportion of susceptible individuals and y the proportion of infectious individuals; then x + y = 1. Assume that the disease spreads by contact between sick and well members of the population and that the rate of spread dy/dt is proportional to the number of such contacts. Further, assume that members of both groups move about freely among each other, so the number of contacts is proportional to the product of x and y. Since x = 1 − y, we obtain the initial value problem dy = α y( 1 − y) , dt

y( 0) = y0 ,

(22)

where α is a positive proportionality factor, and y0 is the initial proportion of infectious individuals. a. Find the equilibrium points for the differential equation (22) and determine whether each is asymptotically stable, semistable, or unstable. b. Solve the initial value problem 22 and verify that the conclusions you reached in part a are correct. Show that y( t) → 1 as t → ∞, which means that ultimately the disease spreads through the entire population.

22. Some diseases (such as typhoid fever) are spread largely by carriers, individuals who can transmit the disease but who exhibit no overt symptoms. Let x and y denote the proportions of susceptibles and carriers, respectively, in the population. Suppose that carriers are identified and removed from the population at a rate β , so dy = −β y. dt

(23)

Suppose also that the disease spreads at a rate proportional to the product of x and y; thus dx = −α x y. dt

(24)

a. Determine y at any time t by solving equation (23) subject to the initial condition y( 0) = y0 .

b. Use the result of part a to find x at any time t by solving equation (24) subject to the initial condition x( 0) = x0 . c. Find the proportion of the population that escapes the epidemic by finding the limiting value of x as t → ∞. .............................................................................................................................. 17 A standard source is the book by Bailey listed in the references. The models in Problems 21, 22, and 23 are discussed by Bailey in Chapters 5, 10, and 20, respectively.

Boyce 9131 Ch02 2

September 29, 2016

17:16

69

69

2.5 Autonomous Differential Equations and Population Dynamics

23. Daniel Bernoulli’s work in 1760 had the goal of appraising the

a. Find all of the critical points for equation (29). Observe that

effectiveness of a controversial inoculation program against smallpox, which at that time was a major threat to public health. His model applies equally well to any other disease that, once contracted and survived, confers a lifetime immunity. Consider the cohort of individuals born in a given year ( t = 0) , and let n( t) be the number of these individuals surviving t years later. Let x( t) be the number of members of this cohort who have not had smallpox by year t and who are therefore still susceptible. Let β be the rate at which susceptibles contract smallpox, and let ν be the rate at which people who contract smallpox die from the disease. Finally, let μ ( t) be the death rate from all causes other than smallpox. Then d x/dt, the rate at which the number of susceptibles declines, is given by

there are no critical points if a < 0, one critical point if a = 0, and two critical points if a > 0.

dx = −( β + μ ( t) ) x. dt

G b. Draw the phase line in each case and determine whether each critical point is asymptotically stable, semistable, or unstable. G c. In each case sketch several solutions of equation (29) in the t y-plane.

Note: If we plot the location of the critical points as a function of a in the ay-plane, we obtain Figure 2.5.10. This is called the bifurcation diagram for equation (29). The bifurcation at a = 0 is called a saddle -- node bifurcation. This name is more natural in the context of second-order systems, which are discussed in Chapter 9.

(25)

y 2

The first term on the right-hand side of equation (25) is the rate at which susceptibles contract smallpox, and the second term is the rate at which they die from all other causes. Also

1

dn = −νβ x − μ ( t) n, dt

(26)

where dn/dt is the death rate of the entire cohort, and the two terms on the right-hand side are the death rates due to smallpox and to all other causes, respectively. a. Let z = x/ n, and show that z satisfies the initial value problem dz = −β z( 1 − ν z) , z( 0) = 1. (27) dt Observe that the initial value problem (27) does not depend on μ ( t) . b. Find z( t) by solving equation (27). c. Bernoulli estimated that ν = β = 1/8. Using these values, determine the proportion of 20-year-olds who have not had smallpox. Note: On the basis of the model just described and the best mortality data available at the time, Bernoulli calculated that if deaths due to smallpox could be eliminated ( ν = 0) , then approximately 3 years could be added to the average life expectancy (in 1760) of 26 years, 7 months. He therefore supported the inoculation program. Bifurcation Points. For an equation of the form dy = f ( a, y) , dt

(28)

where a is a real parameter, the critical points (equilibrium solutions) usually depend on the value of a. As a steadily increases or decreases, it often happens that at a certain value of a, called a bifurcation point, critical points come together, or separate, and equilibrium solutions may be either lost or gained. Bifurcation points are of great interest in many applications, because near them the nature of the solution of the underlying differential equation is undergoing an abrupt change. For example, in fluid mechanics a smooth (laminar) flow may break up and become turbulent. Or an axially loaded column may suddenly buckle and exhibit a large lateral displacement. Or, as the amount of one of the chemicals in a certain mixture is increased, spiral wave patterns of varying color may suddenly emerge in an originally quiescent fluid. Problems 24 through 26 describe three types of bifurcations that can occur in simple equations of the form (28).

24. Consider the equation dy = a − y2. dt

(29)

–2

–1

Asymptotically stable

1 –1

2

3

4 a

Unstable

–2

FIGURE 2.5.10 Bifurcation diagram for y = a − y 2 .

25. Consider the equation dy = ay − y 3 = y( a − y 2 ) . dt

(30)

G a. Again consider the cases a < 0, a = 0, and a > 0. In each case find the critical points, draw the phase line, and determine whether each critical point is asymptotically stable, semistable, or unstable. G b. In each case sketch several solutions of equation (30) in the t y-plane. G c. Draw the bifurcation diagram for equation (30)---that is, plot the location of the critical points versus a. Note: For equation (30) the bifurcation point at a = 0 is called a pitchfork bifurcation. Your diagram may suggest why this name is appropriate.

26. Consider the equation dy = ay − y 2 = y( a − y) . dt

(31)

a. Again consider the cases a < 0, a = 0, and a > 0. In each case find the critical points, draw the phase line, and determine whether each critical point is asymptotically stable, semistable, or unstable. b. In each case sketch several solutions of equation (31) in the t y-plane. c. Draw the bifurcation diagram for equation (31). Note: Observe that for equation (31) there are the same number of critical points for a < 0 and a > 0 but that their stability has changed. For a < 0 the equilibrium solution y = 0 is asymptotically stable and y = a is unstable, while for a > 0 the situation is reversed. Thus there has been an exchange of stability as a passes through the bifurcation point a = 0. This type of bifurcation is called a transcritical bifurcation.

Boyce 9131 Ch02 2

70

September 29, 2016

17:16

70

CHAPTER 2 First-Order Differential Equations

a. If x( 0) = 0, determine the limiting value of x( t) as t → ∞ without solving the differential equation. Then solve the initial value problem and find x( t) for any t. b. If the substances P and Q are the same, then p = q and equation (32) is replaced by

27. Chemical Reactions. A second-order chemical reaction involves the interaction (collision) of one molecule of a substance P with one molecule of a substance Q to produce one molecule of a new substance X ; this is denoted by P + Q → X . Suppose that p and q, where p = q, are the initial concentrations of P and Q, respectively, and let x( t) be the concentration of X at time t. Then p − x( t) and q − x( t) are the concentrations of P and Q at time t, and the rate at which the reaction occurs is given by the equation dx = α ( p − x) ( q − x) , dt

dx = α ( p − x) 2 . dt

(33)

If x( 0) = 0, determine the limiting value of x( t) as t → ∞ without solving the differential equation. Then solve the initial value problem and determine x( t) for any t.

(32)

where α is a positive constant.

Exact Differential Equations and Integrating Factors 2.6

For first-order differential equations there are a number of integration methods that are applicable to various classes of problems. The most important of these are linear equations and separable equations, which we have discussed previously. Here, we consider a class of equations known as exact differential equations for which there is also a well-defined method of solution. Keep in mind, however, that the first-order differential equations that can be solved by elementary integration methods are rather special; most first-order equations cannot be solved in this way.

EXAMPLE 1 Solve the differential equation 2x + y 2 + 2x yy = 0.

(1)

Solution: The equation is neither linear nor separable, so the methods suitable for those types of equations are not applicable here. However, observe that the function ψ ( x, y) = x 2 + x y 2 has the property that 2x + y 2 =

∂ψ , ∂x

2x y =

∂ψ . ∂y

(2)

Therefore, the differential equation can be written as ∂ψ ∂ ψ dy + = 0. ∂x ∂ y dx

(3)

Assuming that y is a function of x, we can use the chain rule to write the left-hand side of equation (3) as dψ ( x, y) /d x. Then equation (3) has the form dψ d 2 ( x, y) = ( x + x y 2 ) = 0. dx dx

(4)

Integrating equation (4) we obtain ψ ( x, y) = x 2 + x y 2 = c,

(5)

where c is an arbitrary constant. The level curves of ψ ( x, y) are the integral curves of equation (1). Solutions of equation (1) are defined implicitly by equation (5).

In solving equation (1) the key step was the recognition that there is a function ψ that satisfies equations (2). More generally, let the differential equation M( x, y) + N ( x, y) y = 0

(6)

Boyce 9131 Ch02 2

September 29, 2016

17:16

71

2.6 Exact Differential Equations and Integrating Factors

be given. Suppose that we can identify a function ψ ( x, y) such that ∂ψ ( x, y) = M( x, y) , ∂x

∂ψ ( x, y) = N ( x, y) , ∂y

(7)

and such that ψ ( x, y) = c defines y = φ ( x) implicitly as a differentiable function of x.18 When there is a function ψ ( x, y) such that ψ x = M and ψ y = N , we can write M( x, y) + N ( x, y) y =

∂ψ ∂ ψ dy d + = ψ ( x, φ ( x) ) ∂x ∂ y dx dx

and the differential equation (6) becomes d ψ ( x, φ ( x) ) = 0. dx

(8)

In this case equation (6) is said to be an exact differential equation because it can be expressed exactly as the derivative of a specific function. Solutions of equation (6), or the equivalent equation (8), are given implicitly by ψ ( x, y) = c,

(9)

where c is an arbitrary constant. In Example 1 it was relatively easy to see that the differential equation was exact and, in fact, easy to find its solution, at least implicitly, by recognizing the required function ψ . For more complicated equations it may not be possible to do this so easily. How can we tell whether a given equation is exact, and if it is, how can we find the function ψ ( x, y) ? The following theorem answers the first question, and its proof provides a way of answering the second.

Theorem 2.6.1 Let the functions M, N , M y , and N x , where subscripts denote partial derivatives, be continuous in the rectangular19 region R: α < x < β , γ < y < δ . Then equation (6) M( x, y) + N ( x, y) y = 0 is an exact differential equation in R if and only if M y ( x, y) = N x ( x, y)

(10)

at each point of R. That is, there exists a function ψ satisfying equations (7), ψ x ( x, y) = M( x, y) ,

ψ y ( x, y) = N ( x, y) ,

if and only if M and N satisfy equation (10).

The proof of this theorem has two parts. First, we show that if there is a function ψ such that equations (7) are true, then it follows that equation (10) is satisfied. Computing M y and N x from equations (7), we obtain M y ( x, y) = ψ x y ( x, y) ,

N x ( x, y) = ψ yx ( x, y) .

(11)

Since M y and N x are continuous, it follows that ψ x y and ψ yx are also continuous. This guarantees their equality, and equation (10) is valid. We now show that if M and N satisfy equation (10), then equation (6) is exact. The proof involves the construction of a function ψ satisfying equations (7) ψ x ( x, y) = M( x, y) ,

ψ y ( x, y) = N ( x, y) .

......................................................................................................................................................................... 18 While a complete discussion of when ψ ( x, y) = c defines y = φ ( x) implicitly as a differentiable function of x is beyond the scope and focus of this course, in general terms this condition is satisfied, locally, at points ( x, y) , where ∂ ψ/∂ y( x, y) = 0. More details can be found in most books on advanced calculus. 19 It is not essential that the region be rectangular, only that it be simply connected. In two dimensions this means that the region has no holes in its interior. Thus, for example, rectangular or circular regions are simply connected, but an annular region is not. More details can be found in most books on advanced calculus.

71

Boyce 9131 Ch02 2

72

September 29, 2016

17:16

72

CHAPTER 2 First-Order Differential Equations

We begin by integrating the first of equations (7) with respect to x, holding y constant. We obtain ψ ( x, y) = Q( x, y) + h( y) ,

(12)

where Q( x, y) is any differentiable function such that Q x = M. For example, we might choose x Q( x, y) = M( s, y) ds, (13) x0

where x0 is some specified constant with α < x0 < β . The function h in equation (12) is an arbitrary differentiable function of y, playing the role of the arbitrary constant (with respect to x). Now we must show that it is always possible to choose h( y) so that the second of equations (7) is satisfied---that is, ψ y = N . By differentiating equation (12) with respect to y and setting the result equal to N ( x, y) , we obtain ψ y ( x, y) =

∂Q ( x, y) + h ( y) = N ( x, y) . ∂y

Then, solving for h ( y) , we have h ( y) = N ( x, y) −

∂Q ( x, y) . ∂y

(14)

In order for us to determine h( y) from equation (14), the right-hand side of equation (14), despite its appearance, must be a function of y only. One way to show that this is true is to show that its derivative with respect to x is zero. Thus we differentiate the right-hand side of equation (14) with respect to x, obtaining the expression ∂N ∂ ∂Q ( x, y) − ( x, y) . ∂x ∂x ∂y

(15)

By interchanging the order of differentiation in the second term of equation (15), we have ∂N ∂ ∂Q ( x, y) − ( x, y) , ∂x ∂y ∂x or, since Q x = M, ∂N ∂M ( x, y) − ( x, y) , ∂x ∂y which is zero on account of equation (10). Hence, despite its apparent form, the right-hand side of equation (14) does not, in fact, depend on x. Then we find h( y) by integrating equation (14) and, upon substituting this function in equation (12), we obtain the required function ψ ( x, y) . This completes the proof of Theorem 2.6.1. It is possible to obtain an explicit expression for ψ ( x, y) in terms of integrals (see Problem 13), but in solving specific exact equations, it is usually simpler and easier just to repeat the procedure used in the preceding proof. That is, after showing that M y = N x , integrate ψ x = M with respect to x, including an arbitrary function of h( y) instead of an arbitrary constant, and then differentiate the result with respect to y and set it equal to N . Finally, use this last equation to solve for h( y) . The next example illustrates this procedure.

EXAMPLE 2 Solve the differential equation ( y cos x + 2xe y ) + ( sin x + x 2 e y − 1) y = 0. Solution: By calculating M y and N x , we find that M y ( x, y) = cos x + 2xe y = N x ( x, y) ,

▼

(16)

Boyce 9131 Ch02 2

September 29, 2016

17:16

73

2.6 Exact Differential Equations and Integrating Factors

▼ so the given equation is exact. Thus there is a ψ ( x, y) such that ψ x ( x, y) = y cos x + 2xe y , ψ y ( x, y) = sin x + x 2 e y − 1. Integrating the first of these equations with respect to x, we obtain ψ ( x, y) = y sin x + x 2 e y + h( y) .

(17)

Next, computing ψ y from equation (17) and setting ψ y = N gives ψ y ( x, y) = sin x + x 2 e y + h ( y) = sin x + x 2 e y − 1. Thus h ( y) = −1 and h( y) = −y. The constant of integration can be omitted since any solution of the preceding differential equation is satisfactory; we do not require the most general one. Substituting for h( y) in equation (17) gives ψ ( x, y) = y sin x + x 2 e y − y. Hence solutions of equation (16) are given implicitly by y sin x + x 2 e y − y = c.

(18)

( 3x y + y 2 ) + ( x 2 + x y) y = 0.

(19)

EXAMPLE 3 Solve the differential equation

Solution: We have M y ( x, y) = 3x + 2y,

N x ( x, y) = 2x + y;

since M y = N x , the given equation is not exact. To see that it cannot be solved by the procedure described above, let us seek a function ψ such that ψ x ( x, y) = 3x y + y 2 ,

ψ y ( x, y) = x 2 + x y.

(20)

Integrating the first of equations (20) with respect to x gives ψ ( x, y) =

3 2 x y + x y 2 + h( y) , 2

(21)

where h is an arbitrary function of y only. To try to satisfy the second of equations (20), we compute ψ y from equation (21) and set it equal to N , obtaining 3 2 x + 2x y + h ( y) = x 2 + x y 2 or 1 h ( y) = − x 2 − x y. (22) 2 Since the right-hand side of equation (22) depends on x as well as y, it is impossible to solve equation (22) for h( y) . Thus there is no ψ ( x, y) satisfying both of equations (20).

Integrating Factors. It is sometimes possible to convert a differential equation that is not exact into an exact differential equation by multiplying the equation by a suitable integrating factor. Recall that this is the procedure that we used in solving linear differential equations in Section 2.1. To investigate the possibility of implementing this idea more generally, let us multiply the equation M( x, y) + N ( x, y) y = 0

(23)

by a function μ and then try to choose μ so that the resulting equation μ ( x, y) M( x, y) + μ ( x, y) N ( x, y) y = 0

(24)

is exact. By Theorem 2.6.1, equation (24) is exact if and only if ( μ M) y = ( μ N ) x .

(25)

73

Boyce 9131 Ch02 2

74

September 29, 2016

17:16

74

CHAPTER 2 First-Order Differential Equations

Since M and N are given functions, equation (25) states that the integrating factor μ must satisfy the first-order partial differential equation Mμ y − N μ x + ( M y − N x ) μ = 0.

(26)

If a function μ satisfying equation (26) can be found, then equation (24) will be exact. The solution of equation (24) can then be obtained by the method described in the first part of this section. The solution found in this way also satisfies equation (23), since the integrating factor μ can be canceled out of equation (24). A partial differential equation of the form (26) may have more than one solution; if this is the case, any such solution may be used as an integrating factor of equation (23). This possible nonuniqueness of the integrating factor is illustrated in Example 4. Unfortunately, equation (26), which determines the integrating factor μ , is ordinarily at least as hard to solve as the original equation (23). Therefore, although in principle integrating factors are powerful tools for solving differential equations, in practice they can be found only in special cases. The most important situations in which simple integrating factors can be found occur when μ is a function of only one of the variables x or y, instead of both. Let us determine conditions on M and N so that equation (23) has an integrating factor μ that depends on x only. If we assume that μ is a function of x only, then the partial derivative μ x reduces to the ordinary derivative dμ /d x and μ y = 0. Making these substitutions in equation (26), we find that dμ M y − Nx = μ. (27) dx N If ( M y − N x ) / N is a function of x only, then there is an integrating factor μ that also depends only on x; further, μ ( x) can be found by solving differential equation (27), which is both linear and separable. A similar procedure can be used to determine a condition under which equation (23) has an integrating factor depending only on y; see Problem 17.

EXAMPLE 4 Find an integrating factor for the equation ( 3x y + y 2 ) + ( x 2 + x y) y = 0

(19)

and then solve the equation. Solution: In Example 3 we showed that this equation is not exact. Let us determine whether it has an integrating factor that depends on x only. On computing the quantity ( M y − N x ) / N , we find that M y ( x, y) − N x ( x, y) 1 3x + 2y − ( 2x + y) = . = 2 N ( x, y) x x + xy

(28)

Thus there is an integrating factor μ that is a function of x only, and it satisfies the differential equation dμ μ = . dx x

(29)

μ ( x) = x.

(30)

Hence (see Problem 7 in Section 2.2)

Multiplying equation (19) by this integrating factor, we obtain ( 3x 2 y + x y 2 ) + ( x 3 + x 2 y) y = 0.

(31)

Equation (31) is exact, since ∂ ∂ ( 3x 2 y + x y 2 ) = 3x 2 + 2x y = ( x 3 + x 2 y) . ∂y ∂x Thus there is a function ψ such that ψ x ( x, y) = 3x 2 y + x y 2 ,

▼

ψ y ( x, y) = x 3 + x 2 y.

(32)

Boyce 9131 Ch02 2

September 29, 2016

17:16

75

2.6 Exact Differential Equations and Integrating Factors

75

▼ Integrating the first of equations (32) with respect to x, we obtain ψ ( x, y) = x 3 y +

1 2 2 x y + h( y) . 2

Substituting this expression for ψ ( x, y) in the second of equations (32), we find that x 3 + x 2 y + h ( y) = x 3 + x 2 y, so h ( y) = 0 and h( y) is a constant. Thus the solutions of equation (31), and hence of equation (19), are given implicitly by x3 y +

1 2 2 x y = c. 2

(33)

Solutions may also be found in explicit form since equation (33) is quadratic in y. You may also verify that a second integrating factor for equation (19) is μ ( x, y) =

1 x y( 2x + y)

and that the same solution is obtained, though with much greater difficulty, if this integrating factor is used (see Problem 22).

Problems Determine whether each of the equations in Problems 1 through 8 is exact. If it is exact, find the solution.

14. Show that any separable equation M( x) + N ( y) y = 0

1. ( 2x + 3) + ( 2y − 2) y = 0 is also exact.

2. ( 2x + 4y) + ( 2x − 2y) y = 0 3. ( 3x 2 − 2x y + 2) + ( 6y 2 − x 2 + 3) y = 0 4. 5.

ax + by dy =− dx bx + cy ax − by dy =− dx bx − cy

15. x 2 y 3 + x( 1 + y 2 ) y = 0,

6. ( ye x y cos( 2x) −2e x y sin( 2x) +2x) +( xe x y cos( 2x) −3) y = 0 7. ( y/ x + 6x) + ( ln x − 2) y = 0, x > 0 y x dy =0 + 2 ( x 2 + y 2 ) 3/2 ( x + y 2 ) 3/2 d x In each of Problems 9 and 10, solve the given initial value problem and determine at least approximately where the solution is valid.

8.

9. ( 2x − y) + ( 2y − x) y = 0,

y( 1) = 3

10. ( 9x + y − 1) − ( 4y − x) y = 0, 2

y( 1) = 0

In each of Problems 11 and 12, find the value of b for which the given equation is exact, and then solve it using that value of b. 2

11. ( x y + bx y) + ( x + y) x y = 0 12. ( ye2x y + x) + bxe2x y y = 0 13. Assume that equation (6) meets the requirements of Theorem 2

2

2.6.1 in a rectangle R and is therefore exact. Show that a possible function ψ ( x, y) is

In each of Problems 15 and 16, show that the given equation is not exact but becomes exact when multiplied by the given integrating factor. Then solve the equation.

x

ψ ( x, y) =

y

M( s, y0 ) ds + x0

where ( x0 , y0 ) is a point in R.

N ( x, t) dt, y0

μ ( x, y) = 1/ x y 3

16. ( x + 2) sin y + ( x cos y) y = 0, μ ( x, y) = xe x 17. Show that if ( N x − M y ) / M = Q, where Q is a function of y only, then the differential equation M + N y = 0 has an integrating factor of the form

μ ( y) = exp

Q( y) dy.

In each of Problems 18 through 21, find an integrating factor and solve the given equation.

18. ( 3x 2 y + 2x y + y 3 ) + ( x 2 + y 2 ) y = 0 19. y = e2x + y − 1 20. 1 + ( x/ y − sin y) y = 0 21. y + ( 2x y − e−2y ) y = 0 22. Solve the differential equation ( 3x y + y 2 ) + ( x 2 + x y) y = 0 using the integrating factor μ ( x, y) = ( x y( 2x + y) ) −1 . Verify that the solution is the same as that obtained in Example 4 with a different integrating factor.

Boyce 9131 Ch02 2

76

September 29, 2016

17:16

76

CHAPTER 2 First-Order Differential Equations

Numerical Approximations: Euler’s Method 2.7

Recall two important facts about the first-order initial value problem dy = f ( t, y) , dt

y( t0 ) = y0 .

(1)

First, if f and ∂ f /∂ y are continuous, then the initial value problem (1) has a unique solution y = φ ( t) in some interval surrounding the initial point t = t0 . Second, it is usually not possible to find the solution φ by symbolic manipulations of the differential equation. Up to now we have considered the main exceptions to the latter statement: differential equations that are linear, separable, or exact, or that can be transformed into one of these types. Nevertheless, it remains true that solutions of the vast majority of first-order initial value problems cannot be found by analytical means, such as those considered in the first part of this chapter. Therefore, it is important to be able to approach the problem in other ways. As we have already seen, one of these ways is to draw a direction field for the differential equation (which does not involve solving the equation) and then to visualize the behavior of solutions from the direction field. This has the advantage of being a relatively simple process, even for complicated differential equations. However, it does not lend itself to quantitative computations or comparisons, and this is often a critical shortcoming. For example, Figure 2.7.1 shows a direction field for the differential equation dy = 3 − 2t − 0.5y. dt

(2)

From the direction field you can visualize the behavior of solutions on the rectangle shown in the figure. On this rectangle a solution starting at a point on the y-axis initially increases with t, but it soon reaches a maximum value and then begins to decrease as t increases further.

y 3

2

1

1

2

3 t

–1 FIGURE 2.7.1 A direction field for equation (2): dy/dt = 3 − 2t − 0.5y.

You may also observe that in Figure 2.7.1 many tangent line segments at successive values of t almost touch each other. It takes only a bit of imagination to consider starting at a point on the y-axis and linking line segments for successive values of t in the grid, thereby producing a piecewise linear graph. Such a graph would apparently be an approximation to a solution of

Boyce 9131 Ch02 2

September 29, 2016

17:16

77

2.7 Numerical Approximations: Euler's Method

the differential equation. To convert this idea into a useful method for generating approximate solutions, we must answer several questions, including the following: 1. Can we carry out the linking of tangent lines in a systematic and straightforward manner? 2. If so, does the resulting piecewise linear function provide an approximation to an actual solution of the differential equation? 3. If so, can we assess the accuracy of the approximation? That is, can we estimate how far the approximation deviates from the solution itself? It turns out that the answer to each of these questions is affirmative. The resulting method was originated by Euler about 1768 and is referred to as the tangent line method or the Euler method. We will deal with the first two questions in this section, but will defer a systematic discussion of the third question until Chapter 8. To see how the Euler method works, let us consider how the tangent lines might be used to approximate the solution y = φ ( t) of initial value problem (1) near t = t0 . We know that the solution passes through the initial point ( t0 , y0 ) , and from the differential equation, we also know that its slope at this point is f ( t0 , y0 ) . Thus we can write down an equation for the line tangent to the solution curve at ( t0 , y0 ) , namely, y = y0 + f ( t0 , y0 ) ( t − t0 ) .

(3)

The tangent line is a good approximation to the actual solution curve on an interval short enough so that the slope of the solution does not change appreciably from its value at the initial point; see Figure 2.7.2. Thus, if t1 is close enough to t0 , we can approximate φ ( t1 ) by the value y1 determined by substituting t = t1 into the tangent line approximation at t = t0 ; thus y1 = y0 + f ( t0 , y0 ) ( t1 − t0 ) .

(4)

y y = y0 + f (t0, y0) (t – t0) y1

φ (t1)

y = φ (t)

y0

t0

t1

t

FIGURE 2.7.2 A tangent line approximation of y = f ( t, y) at ( t0 , y0 ) .

To proceed further, we can try to repeat the process. Unfortunately, we do not know the value φ ( t1 ) of the solution at t1 . The best we can do is to use the approximate value y1 instead. Thus we construct the line through ( t1 , y1 ) with the slope f ( t1 , y1 ) , y = y1 + f ( t1 , y1 ) ( t − t1 ) .

(5)

To approximate the value of φ ( t) at a nearby point t2 , we use equation (5) instead of equation (3), obtaining y2 = y1 + f ( t1 , y1 ) ( t2 − t1 ) .

(6)

Continuing in this manner, we use the value of y calculated at each step to determine the slope of the approximation for the next step. The general expression for the tangent line starting at ( tn , yn ) is y = yn + f ( tn , yn ) ( t − tn ) ;

(7)

hence the approximate value yn+1 at tn+1 in terms of tn , tn+1 , and yn is yn+1 = yn + f ( tn , yn ) ( tn+1 − tn ) ,

n = 0, 1, 2, . . . .

(8)

77

Boyce 9131 Ch02 2

78

September 29, 2016

17:16

78

CHAPTER 2 First-Order Differential Equations

If we introduce the notation f n = f ( tn , yn ) , then we can rewrite equation (8) as yn+1 = yn + f n · ( tn+1 − tn ) ,

n = 0, 1, 2, . . . .

(9)

Finally, if we assume that there is a uniform step size h between the points t0 , t1 , t2 , . . . , then tn+1 = tn + h for each n, and we obtain Euler’s formula in the form yn+1 = yn + f n h,

n = 0, 1, 2, . . . .

(10)

To use Euler’s method, you repeatedly evaluate equation (9) or equation (10), depending on whether or not the step size is constant, using the result of each step to execute the next step. In this way you generate a sequence of values y1 , y2 , y3 , . . . that approximate the values of the solution φ ( t) at the points t1 , t2 , t3 , . . . . If, instead of a sequence of points, you need a function to approximate the solution φ ( t) , then you can use the piecewise linear function constructed from the collection of tangent line segments. That is, let y be given in [t0 , t1 ] by equation (7) with n = 0, in [t1 , t2 ] by equation (7) with n = 1, and so on.

EXAMPLE 1 Consider the initial value problem dy = 3 − 2t − 0.5y, dt

y( 0) = 1.

(11)

Use Euler’s method with step size h = 0.2 to find approximate values of the solution of initial value problem (9) at t = 0.2, 0.4, 0.6, 0.8, and 1. Compare them with the corresponding values of the actual solution of the initial value problem. Solution: Note that the differential equation in the given initial value problem is the same as in equation (2); its direction field is shown in Figure 2.7.1. Before applying Euler’s method, observe that this differential equation is linear, so it can be solved as in Section 2.1, using the integrating factor et/2 . The resulting solution of the initial value problem (9) is y = φ ( t) = 14 − 4t − 13e−t/2 .

(12)

We will use this information to assess how the approximate solution obtained by Euler’s method compares with the exact solution. To approximate this solution by Euler’s method, note that f ( t, y) = 3 − 2t − 0.5y. Using the initial values t0 = 0 and y0 = 1, we find that f 0 = f ( t0 , y0 ) = f ( 0, 1) = 3 − 0 − 0.5 = 2.5 and then, from equation (3), the tangent line approximation near t = 0 is y = 1 + 2.5( t − 0) = 1 + 2.5t.

(13)

Setting t = 0.2 in equation (13), we find the approximate value y1 of the solution at t = 0.2, namely, y1 = 1 + ( 2.5) ( 0.2) = 1.5. At the next step we have f 1 = f ( t1 , y1 ) = f ( 0.2, 1.5) = 3 − 2( 0.2) − ( 0.5) ( 1.5) = 3 − 0.4 − 0.75 = 1.85. Then the tangent line approximation near t = 0.2 is y = 1.5 + 1.85( t − 0.2) = 1.13 + 1.85t.

(14)

Evaluating the expression in equation (14) for t = 0.4, we obtain y2 = 1.13 + 1.85( 0.4) = 1.87.

▼

Repeating this computational procedure three more times, we obtain the results shown in Table 2.7.1.

Boyce 9131 Ch02 2

September 29, 2016

17:16

79

2.7 Numerical Approximations: Euler's Method

▼

T A B L E 2.7.1 n

tn

yn

0 1 2 3 4 5

0.0 0.2 0.4 0.6 0.8 1.0

1.00000 1.50000 1.87000 2.12300 2.27070 2.32363

Results of Euler’s Method with h = 0.2 for y = 3 − 2t − 0.5y, y(0) = 1 fn = f (tn , yn )

2.5 1.85 1.265 0.7385 0.26465

Tangent Line

Exact y(tn )

y = 1 + 2.5( t − 0) y = 1.5 + 1.85( t − 0.2) y = 1.87 + 1.265( t − 0.4) y = 2.123 + 0.7385( t − 0.6) y = 2.2707 + 0.26465( t − 0.8)

1.00000 1.43711 1.75650 1.96936 2.08584 2.11510

The second column contains the t-values separated by the step size h = 0.2. The third column shows the corresponding y-values computed from Euler’s formula (10). Column four contains the slopes f n of the tangent line at the current point, ( tn , yn ) . In the fifth column are the tangent line approximations found from equation (7). The sixth column contains values of the solution (12) of the initial value problem (9), correct to five decimal places. The solution (12) and the tangent line approximation are also plotted in Figure 2.7.3.

y 2.4 Tangent line approximation 2

Solution

1.6

1.2

0.2

0.4

0.6

0.8

1

t

FIGURE 2.7.3 Plots of the solution and a tangent line approximation with h = 0.2 for the initial value problem (9): dy/dt = 3 − 2t − 0.5y, y( 0) = 1.

From Table 2.7.1 and Figure 2.7.3 we see that the approximations given by Euler’s method for this problem are greater than the corresponding values of the actual solution. This is because the graph of the solution is concave down and therefore the tangent line approximations lie above the graph. The accuracy of the approximations in this example is not good enough to be satisfactory in a typical scientific or engineering application. For example, at t = 1 the error in the approximation is 2.32363 − 2.11510 = 0.20853, which is a percentage error of about 9.86% relative to the exact solution. One way to achieve more accurate results is to use a smaller step size, with a corresponding increase in the number of computational steps. We explore this possibility in the next example.

Of course, computations such as those in Example 1 and in the other examples in this section are usually done on a computer. Some software packages include code for the Euler method, while others do not. In any case, it is straightforward to write a computer program that will carry out the calculations required to produce results such as those in Table 2.7.1.

79

Boyce 9131 Ch02 2

80

September 29, 2016

17:16

80

CHAPTER 2 First-Order Differential Equations

Basically, what is required is a loop that will evaluate equation (10) repetitively, along with suitable instructions for input and output. The output can be a list of numbers, as in Table 2.7.1, or a plot, as in Figure 2.7.3. The specific instructions can be written in any highlevel programming language with which you are familiar.

EXAMPLE 2 Consider again the initial value problem (9) dy = 3 − 2t − 0.5y, dt

y( 0) = 1.

Use Euler’s method with various step sizes to calculate approximate values of the solution for 0 ≤ t ≤ 5. Compare the calculated results with the corresponding values of the exact solution (12) y = 14 − 4t − 13e−t/2 . Solution: We used step sizes h = 0.1, 0.05, 0.025, and 0.01, corresponding to 50, 100, 200, and 500 steps, respectively, to go from t = 0 to t = 5. The results of these calculations, along with the values of the exact solution, are summarized in Table 2.7.2. All computed entries are rounded to four decimal places, although more digits were retained in the intermediate calculations.

T A B L E 2.7.2

Comparison of the Exact Solution with Euler’s Method for Several Step Sizes h for y = 3 − 2t − 0.5y, y(0) = 1

t

h = 0.1

h = 0.05

h = 0.025

h = 0.01

Exact

0.0 1.0 2.0 3.0 4.0 5.0

1.0000 2.2164 1.3397 −0.7903 −3.6707 −7.0003

1.0000 2.1651 1.2780 −0.8459 −3.7152 −7.0337

1.0000 2.1399 1.2476 −0.8734 −3.7373 −7.0504

1.0000 2.1250 1.2295 −0.8898 −3.7506 −7.0604

1.0000 2.1151 1.2176 −0.9007 −3.7594 −7.0671

What conclusions can we draw from the data in Table 2.7.2? The most important observation is that, for a fixed value of t, the computed approximate values become more accurate as the step size h decreases. You can see this by reading across a particular row in the table from left to right. This is what we would expect, of course, but it is encouraging that the data confirm our expectations. For example, for t = 2 the approximate value with h = 0.1 is too large by 0.1221 (about 10%), whereas the value with h = 0.01 is too large by only 0.0119 (about 1%). In this case, reducing the step size by a factor of 10 (and performing 10 times as many computations) also reduces the error by a factor of about 10. Comparing the errors for other pairs of values in the table confirms that this relation between step size and error holds for them also: reducing the step size by a given factor also reduces the error by approximately the same factor. Does this mean that for the Euler method the error is approximately proportional to the step size? Of course, one example does not establish such a general result, but it is at least an interesting conjecture.20 A second observation from Table 2.7.2 is that, for a fixed step size h, the approximations become more accurate as t increases, at least for t > 2. For instance, for h = 0.1 the error for t = 5 is only 0.0668, which is a little more than one-half of the error at t = 2. We will return to this matter later in this section. All in all, Euler’s method seems to work rather well for this problem. Reasonably good results are obtained even for a moderately large step size h = 0.1, and the approximation can be improved by decreasing h.

......................................................................................................................................................................... 20 A

more detailed discussion of the errors in using the Euler method appears in Chapter 8.

Boyce 9131 Ch02 2

September 29, 2016

17:16

81

2.7 Numerical Approximations: Euler's Method

Let us now look at another example.

EXAMPLE 3 Consider the initial value problem dy = 4 − t + 2y, dt

y( 0) = 1.

(15)

The general solution of this differential equation was found in Example 2 of Section 2.1, and the solution of the initial value problem (11) is 7 1 11 y = − + t + e2t . 4 2 4

(16)

Use Euler’s method with several step sizes to find approximate values of the solution on the interval 0 ≤ t ≤ 5. Compare the results with the corresponding values of the solution (16). Solution: Using the same range of step sizes as in Example 2, we obtain the results presented in Table 2.7.3. T A B L E 2.7.3

Comparison of the Exact Solution with Euler’s Method for Several Step Sizes h for y = 4 − t + 2y, y(0) = 1

t

h = 0.1

h = 0.05

h = 0.025

h = 0.01

Exact

0.0 1.0 2.0 3.0 4.0 5.0

1.000000 15.77728 104.6784 652.5349 4042.122 25026.95

1.000000 17.25062 123.7130 837.0745 5633.351 37897.43

1.000000 18.10997 135.5440 959.2580 6755.175 47555.35

1.000000 18.67278 143.5835 1045.395 7575.577 54881.32

1.000000 19.06990 149.3949 1109.179 8197.884 60573.53

The data in Table 2.7.3 again confirm our expectation that, for a given value of t, accuracy improves as the step size h is reduced. For example, for t = 1 the percentage error diminishes from 17.3% when h = 0.1 to 2.1% when h = 0.01. However, the error increases fairly rapidly as t increases for a fixed h. Even for h = 0.01, the error at t = 5 is 9.4%, and it is much greater for larger step sizes. Of course, the accuracy that is needed depends on the purpose for which the results are intended, but the errors in Table 2.7.3 are too large for most scientific or engineering applications. To improve the situation, we might either try even smaller step sizes or else restrict the computations to a rather short interval away from the initial point. Nevertheless, it is clear that Euler’s method is much less effective in this example than in Example 2.

To understand better what is happening in these examples, let us look again at Euler’s method for the general initial value problem (1) dy = f ( t, y) , dt

y( t0 ) = y0 ,

whose exact solution we denote by φ ( t) . Recall that a first-order differential equation has an infinite family of solutions, indexed by an arbitrary constant c, and that the initial condition picks out one member of this infinite family by determining the value of c. Thus in the infinite family of solutions, φ ( t) is the one solution that satisfies the initial condition φ ( t0 ) = y0 . At the first step Euler’s method uses the tangent line approximation to the graph of y = φ ( t) passing through the initial point ( t0 , y0 ) , and this produces the approximate value φ ( t1 ) , so at the second step Euler’s method uses the tangent line y1 at t1 . Usually, y1 = approximation not to y = φ ( t) , but to a nearby solution y = φ 1 ( t) that passes through the point ( t1 , y1 ) . So it is at each subsequent step. Euler’s method uses a succession of tangent line approximations to a sequence of different solutions φ ( t) , φ 1 ( t) , φ 2 ( t) , . . . of the differential equation. At each step the tangent line is constructed to the solution passing through the point determined by the result of the preceding step, as shown in Figure 2.7.4. The quality of the approximation after many steps depends strongly on the behavior of the set of solutions that pass through the points ( tn , yn ) for n = 1, 2, 3, . . . .

81

Boyce 9131 Ch02 2

82

September 29, 2016

17:16

82

CHAPTER 2 First-Order Differential Equations

y y = φ 2 (t) (t2, y2) (t3, y3)

(t1, y1)

y = φ (t)

y = φ 1 (t)

y0

t0

t1

t2

t3

t

FIGURE 2.7.4 The Euler method.

In Example 2 the general solution of the differential equation is y = 14 − 4t + ce−t/2

(17)

and the solution of the initial value problem (9) corresponds to c = −13. The family of solutions (17) is a converging family since the term involving the arbitrary constant c approaches zero as t → ∞. It does not matter very much which solutions we are approximating by tangent lines in the implementation of Euler’s method, since all the solutions are getting closer and closer to each other as t increases. On the other hand, in Example 3 the general solution of the differential equation is 7 1 y = − + t + ce2t , 4 2

(18)

and, because the term involving the arbitrary constant c grows without bound as t → ∞, this is a diverging family. Note that solutions corresponding to two nearby values of c become arbitrarily far apart as t increases. In Example 3 we are trying approximate the solution for c = 11/4, but in the use of Euler’s method we are actually at each step following another solution that separates from the desired one faster and faster as t increases. This explains why the errors in Example 3 are so much larger than those in Example 2. In using a numerical procedure such as the Euler method, you must always keep in mind the question of whether the results are accurate enough to be useful. In the preceding examples, the accuracy of the numerical results could be determined directly by a comparison with the solution obtained analytically. Of course, usually the analytical solution is not available if a numerical procedure is to be employed, so what we usually need are bounds for, or at least estimates of, the error that do not require a knowledge of the exact solution. You should also keep in mind that the best that we can expect, or hope for, from a numerical approximation is that it reflects the behavior of the actual solution. Thus a member of a diverging family of solutions will always be harder to approximate than a member of a converging family. If you wish to read more about numerical approximations to solutions of initial value problems, you may go directly to Chapter 8 at this point. There, we present some information on the analysis of errors and also discuss several algorithms that are computationally much more efficient than the Euler method.

Problems Note about Variations of Computed Results. Most of the problems in this section call for fairly extensive numerical computations. To handle these problems you need suitable computing hardware and software. Keep in mind that numerical results may vary somewhat, depending on how your program is constructed and on how your computer executes arithmetic steps, rounds off, and so forth. Minor variations in the last decimal place may be due to such causes and do not necessarily indicate that something is amiss. Answers in the back

of the book are recorded to six digits in most cases, although more digits were retained in the intermediate calculations. In each of Problems 1 through 4: N a. Find approximate values of the solution of the given initial value problem at t = 0.1, 0.2, 0.3, and 0.4 using the Euler method with h = 0.1. N b. Repeat part (a) with h = 0.05. Compare the results with those found in a.

Boyce 9131 Ch02 2

September 29, 2016

17:16

83

2.8 The Existence and Uniqueness Theorem N c. Repeat part a with h = 0.025. Compare the results with those found in a and b. N d. Find the solution y = φ ( t) of the given problem and evaluate φ ( t) at t = 0.1, 0.2, 0.3, and 0.4. Compare these values with the results of a, b, and c.

1. 2. 3. 4.

y = 3 + t − y,

y = 2y − 1,

y( 0) = 1 y( 0) = 1

y = 0.5 − t + 2y,

y = y2 − t 2,

y( 0) = 0

In each of Problems 5 through 8, draw a direction field for the given differential equation and state whether you think that the solutions are converging or diverging. √ G 5. y = 5 − 3 y G G G

6. y = y( 3 − t y) 7. y = −t y + 0.1y 3 8. y = t 2 + y 2

10. y = y( 3 − t y) ,

N

under suitable conditions on f , the numerical approximation generated by the Euler method for the initial value problem y = f ( t, y) , y( t0 ) = y0 converges to the exact solution as the step size h decreases. This is illustrated by the following example. Consider the initial value problem y = 1 − t + y,

2

3t , 3y 2 − 4

y( 1) = 0.

yk = ( 1 + h) yk−1 + h − htk−1 ,

a. Use Euler’s method with h = 0.1 to obtain approximate

12. Consider the initial value problem y = t 2 + y2,

y( 0) = 1.

13. Consider the initial value problem y = −t y + 0.1y 3 ,

c. Noting that y1 = ( 1 + h) ( y0 − t0 ) + t1 , show by induction that yn = ( 1 + h) n ( y0 − t0 ) + tn

(19)

for each positive integer n.

d. Consider a fixed point t > t0 and for a given n choose

h = ( t − t0 ) / n. Then tn = t for every n. Note also that h → 0 as n → ∞. By substituting for h in equation (19) and letting n → ∞, show that yn → φ ( t) as n → ∞. Hint: lim ( 1 + a/ n) n = ea .

In each of Problems 16 and 17, use the technique discussed in Problem 15 to show that the approximation obtained by the Euler method converges to the exact solution at any fixed point as h → 0.

16. y = y,

y( 0) = 1

17. y = 2y − 1,

y( 0) = 1

y( 0) = α,

where α is a given number.

The Existence and Uniqueness Theorem

In this section we discuss the proof of Theorem 2.4.2, the fundamental existence and uniqueness theorem for first-order initial value problems. Recall that this theorem states that under certain conditions on f ( t, y) , the initial value problem y = f ( t, y) ,

k = 1, 2, . . . .

n→∞

Use Euler’s method with h = 0.1, 0.05, 0.025, and 0.01 to explore the solution of this problem for 0 ≤ t ≤ 1. What is your best estimate of the value of the solution at t = 0.8? At t = 1? Are your results consistent with the direction field in Problem 8?

2.8

y( t0 ) = y0 .

a. Show that the exact solution is y = φ ( t) = ( y0 −t0 ) et−t0 +t. N b. Using the Euler formula, show that

values of the solution at t = 1.2, 1.4, 1.6, and 1.8. N b. Repeat part a with h = 0.05. c. Compare the results of parts a and b. Note that they are reasonably close for t = 1.2, 1.4, and 1.6 but are quite different for t = 1.8. Also note (from the differential equation) that the line tangent to the solution is parallel to the y-axis when y = ±2/ 3 ∼ = ±1.155. Explain how this might cause such a difference in the calculated values. N

where α is a given number. G a. Draw a direction field for the differential equation. Note that there is a critical value of α in the interval 0 ≤ α ≤ 1 that separates converging solutions from diverging ones. Call this critical value α 0 . N b. Use Euler’s method with h = 0.01 to estimate α 0 . Do this by restricting α 0 to an interval [a, b], where b − a = 0.01.

y( 0) = 0.5

11. Consider the initial value problem y =

y( 0) = α,

15. Convergence of Euler’s Method. It can be shown that

In each of Problems 9 and 10, use Euler’s method to find approximate values of the solution of the given initial value problem at t = 0.5, 1, 1.5, 2, 2.5, and 3: (a) With h = 0.1, (b) With h = 0.05, (c) With h = 0.025, (d) With h = 0.01. √ N 9. y = 5 − 3 y, y( 0) = 2 N

G a. Draw a direction field for the differential equation (or reexamine the one from Problem 7). Observe that there is a critical value of α in the interval 2 ≤ α ≤ 3 that separates converging solutions from diverging ones. Call this critical value α 0 . N b. Use Euler’s method with h = 0.01 to estimate α 0 . Do this by restricting α 0 to an interval [a, b], where b − a = 0.01.

14. Consider the initial value problem

y( 0) = 1

y = 3 cos t − 2y,

83

y( t0 ) = y0

has a unique solution in some interval containing the point t0 .

(1)

Hint: y1 = ( 1 + 2h) /2 + 1/2

Boyce 9131 Ch02 2

84

September 29, 2016

17:16

84

CHAPTER 2 First-Order Differential Equations

In some cases (for example, if the differential equation is linear), the existence of a solution of the initial value problem (1) can be established directly by actually solving the problem and exhibiting a formula for the solution. However, in general, this approach is not feasible because there is no method of solving the differential equation that applies in all cases. Therefore, for the general case, it is necessary to adopt an indirect approach that demonstrates the existence of a solution of initial value problem (1) but usually does not provide a practical means of finding it. The heart of this method is the construction of a sequence of functions that converges to a limit function satisfying the initial value problem, although the members of the sequence individually do not. As a rule, it is impossible to compute explicitly more than a few members of the sequence; therefore, the limit function can be determined only in rare cases. Nevertheless, under the restrictions on f ( t, y) stated in Theorem 2.4.2, it is possible to show that the sequence in question converges and that the limit function has the desired properties. The argument is fairly intricate and depends, in part, on techniques and results that are usually encountered for the first time in a course on advanced calculus. Consequently, we do not go into all the details of the proof here; we do, however, indicate its main features and point out some of the difficulties that must be overcome. First of all, we note that it is sufficient to consider the problem in which the initial point ( t0 , y0 ) is the origin; that is, we consider the problem y = f ( t, y) , y( 0) = 0. (2) If some other initial point is given, then we can always make a preliminary change of variables, corresponding to a translation of the coordinate axes, that will take the given point ( t0 , y0 ) into the origin. The existence and uniqueness theorem can now be stated in the following way.

Theorem 2.8.1 | Existence and Uniqueness of Solutions of y = f(t, y), y(0) = 0 If f and ∂ f /∂ y are continuous in a rectangle R: |t| ≤ a, |y| ≤ b, then there is some interval |t| ≤ h ≤ a in which there exists a unique solution y = φ ( t) of the initial value problem (2).

For the method of proof discussed here it is necessary to transform initial value problem (2) into a more convenient form. If we suppose temporarily that there is a differentiable function y = φ ( t) that satisfies the initial value problem, then f ( t, φ ( t) ) is a continuous function of t only. Hence we can integrate y = f ( t, y) from the initial point t = 0 to an arbitrary value of t, obtaining t φ ( t) = f ( s, φ ( s) ) ds, (3) 0

where we have made use of the initial condition φ ( 0) = 0. We also denote the dummy variable of integration by s. Since equation (3) contains an integral of the unknown function φ , it is called an integral equation. This integral equation is not a formula for the solution of the initial value problem, but it does provide another relation satisfied by any solution of equations (2). Conversely, suppose that there is a continuous function y = φ ( t) that satisfies the integral equation (3); then this function also satisfies the initial value problem (2). To show this, we first substitute zero for t in equation (3), which shows that the initial condition is satisfied. Further, since the integrand in equation (3) is continuous, it follows from the fundamental theorem of calculus that φ is differentiable and that φ ( t) = f ( t, φ ( t) ) . Therefore, the initial value problem and the integral equation are equivalent in the sense that any solution of one is also a solution of the other. It is more convenient to show that there is a unique solution of the integral equation in a certain interval |t| ≤ h. The same conclusion also holds for the initial value problem (2). One method of showing that the integral equation (3) has a unique solution is known as the method of successive approximations or Picard’s21 iteration method. In using this method, ......................................................................................................................................................................... 21 Charles-Émile Picard (1856--1914) was appointed professor at the Sorbonne before the age of 30. Except for Henri Poincaré, he is perhaps the most distinguished French mathematician of his generation. He is known for important theorems in complex variables and algebraic geometry as well as differential equations. A special case of the method of successive approximations was first published by Liouville in 1838. However, the method is usually credited to Picard, who established it in a general and widely applicable form in a series of papers beginning in 1890.

Boyce 9131 Ch02 2

September 29, 2016

17:16

85

2.8 The Existence and Uniqueness Theorem

we start by choosing an initial function φ 0 , either arbitrarily or to approximate in some way the solution of the initial value problem. The simplest choice is φ 0 ( t) = 0;

(4)

then φ 0 at least satisfies the initial condition in equations (2), although presumably not the differential equation. The next approximation φ 1 is obtained by substituting φ 0 ( s) for φ ( s) in the right-hand side of equation (3) and calling the result of this operation φ 1 ( t) . Thus t φ 1 ( t) = f s, φ 0 ( s) ds. (5) 0

Similarly, φ 2 is obtained from φ 1 :

f s, φ 1 ( s) ds,

t

φ 2 ( t) =

(6)

0

and, in general,

t

φ n+1 ( t) =

f ( s, φ n ( s) ) ds.

(7)

0

In this manner we generate the sequence of functions {φ n } = {φ 0 , φ 1 , φ 2 , . . . , φ n , . . . }. Each member of the sequence satisfies the initial condition, but in general none satisfies the differential equation. However, if at some stage, say, for n = k, we find that φ k+1 ( t) = φ k ( t) , then it follows that φ k is a solution of the integral equation (3). Hence φ k is also a solution of the initial value problem (2), and the sequence is terminated at this point. In general, this does not occur, and it is necessary to consider the entire infinite sequence. To establish Theorem 2.8.1, we must answer four principal questions: 1. Do all members of the sequence {φ n } exist, or may the process break down at some stage? 2. Does the sequence converge? 3. What are the properties of the limit function? In particular, does it satisfy the integral equation (3) and hence the initial value problem (2)? 4. Is this the only solution, or may there be others? We first show how these questions can be answered in a specific and relatively simple example and then comment on some of the difficulties that may be encountered in the general case.

EXAMPLE 1 Solve the initial value problem y = 2t ( 1 + y) ,

y( 0) = 0

(8)

by the method of successive approximations. Solution: Note first that if y = φ ( t) , then the corresponding integral equation is

t

φ ( t) =

2s( 1 + φ ( s) ) ds.

(9)

0

If the initial approximation is φ 0 ( t) = 0, it follows that

φ 1 ( t) =

t

0

Similarly,

φ 2 ( t) =

▼

t

t

2sds = t 2 .

(10)

0

2s 1 + φ 1 ( s) ds = 0

2s 1 + φ 0 ( s) ds =

t

2s( 1 + s 2 ) ds = t 2 + 0

t4 2

(11)

85

Boyce 9131 Ch02 2

86

September 29, 2016

17:16

86

CHAPTER 2 First-Order Differential Equations

▼ and

t

φ 3 ( t) =

2s( 1 + φ 2 ( s) ) ds = 0

t

s4 2s 1 + s + 2

ds = t 2 +

2

0

t4 t6 + . 2 2·3

(12)

Equations (10), (11), and (12) suggest that φ n ( t) = t 2 +

t6 t 2n t4 + + ··· + 2! 3! n!

(13)

for each n ≥ 1, and this result can be established by mathematical induction, as follows. Equation (13) is certainly true for n = 1; see equation (10). We must show that if it is true for n = k, then it also holds for n = k + 1. We have

t

φ k+1 ( t) =

2s( 1 + φ k ( s) ) ds 0

t

=

s4 s 2k 2s 1 + s + + ··· + 2! k!

2

0

=

t

2s + 2s 3 + 0

= t2 +

ds

2s 5 2s 2k+1 + ··· + ds 2! k!

(14)

t6 t 2k+2 t4 + + ··· + , 2! 3! ( k + 1) !

and the inductive proof is complete. A plot of the first four iterates, φ 1 ( t) , φ 2 ( t) , φ 3 ( t) , and φ 4 ( t) , is shown in Figure 2.8.1. As k increases, the iterates seem to remain close over a gradually increasing interval, suggesting eventual convergence to a limit function. y 3

y = φ 4(t)

y = φ 3(t) y = φ 2(t)

2.5 2 1.5

y = φ 1(t)

1 0.5

–1.5

–1

–0.5

0.5

1

1.5 t

FIGURE 2.8.1 Plots of the first four Picard iterates y = φ 1 ( t) , . . . , y = φ 4 ( t) for Example 1: dy/dt = 2t ( 1 + y) , y( 0) = 0.

It follows from equation (13) that φ n ( t) is the n th partial sum of the infinite series ∞ t 2k k=1

k!

;

(15)

hence lim φ n ( t) exists if and only if the series (15) converges. Applying the ratio test, we see that, n→∞

for each t,

2k+2 t k! t2 ( k + 1) ! t 2k = k + 1 → 0 as k → ∞.

(16)

Thus the interval of convergence for series (15) is the entire t-axis. This means its sum φ ( t) is the limit of the sequence {φ n ( t) } for every value of t. Further, since the series (15) is a Taylor series, it can be differentiated or integrated term-by-term for all values of t. Therefore, we can verify by direct computation that φ ( t) =

▼

∞

k=1

t 2k / k! is a solution of the integral equation (9). Alternatively,

Boyce 9131 Ch02 2

September 29, 2016

17:16

87

2.8 The Existence and Uniqueness Theorem

▼ by substituting φ ( t) for y in equations (8), we can verify that this function satisfies the initial value

problem (6). In this example it is also possible, from the series (15), to identify the solution φ ( t) 2 in terms of elementary functions, namely, φ ( t) = et − 1. (See Problem 13.) However, this is not necessary for the discussion of existence and uniqueness. Explicit knowledge of φ ( t) does make it possible to visualize the convergence of the sequence of iterates more clearly by plotting the difference ek ( t) = φ ( t) − φ k ( t) for various values of k. Figure 2.8.2 shows this difference for k = 1, 2, 3, 4. This figure clearly illustrates the gradually increasing interval over which successive iterates provide a good approximation to the solution of the initial value problem. y 1

k=1 k=2

k=3

0.8

k=4 0.6 0.4 0.2

–1.5

–1

–0.5

0.5

1

1.5

t

FIGURE 2.8.2 Plots of y = ek ( t) = φ ( t) − φ k ( t) for Example 1 for k = 1, . . . , 4.

Finally, to deal with the question of uniqueness, let us suppose that the initial value problem has two different solutions φ and ψ . The assumption that φ and ψ are different means there is at least one value of t for which φ ( t) − ψ ( t) = 0. Also, since φ and ψ both satisfy the integral equation (9), we have by subtraction (and the linearity of integration) that

t

2s( φ ( s) − ψ ( s) ) ds.

φ ( t) − ψ ( t) = 0

Taking absolute values of both sides, we have, if t > 0,

|φ ( t) − ψ ( t) | =

t

0

2s( φ ( s) − ψ ( s) ) ds ≤

t

2s|φ ( s) − ψ ( s) |ds. 0

If we restrict t to lie in the interval 0 ≤ t ≤ A/2, where A is arbitrary, then 2t ≤ A and

t

|φ ( s) − ψ ( s) |ds for 0 ≤ t ≤ A/2.

|φ ( t) − ψ ( t) | ≤ A

(17)

0

It is now convenient to introduce the function U defined by

t

|φ ( s) − ψ ( s) |ds.

U ( t) =

(18)

0

Then it follows at once that U ( 0) = 0,

(19)

U ( t) ≥ 0, for t ≥ 0.

(20)

Further, U is differentiable, and U ( t) = |φ ( t) − ψ ( t) |. Hence, by equation (17), U ( t) − AU ( t) ≤ 0 for 0 ≤ t ≤ A/2.

(21)

Multiplying equation (21) by the positive quantity e−At gives

▼

e−At U ( t)

≤ 0 for 0 ≤ t ≤ A/2.

(22)

87

Boyce 9131 Ch02 2

88

September 29, 2016

17:16

88

CHAPTER 2 First-Order Differential Equations

▼ Then, upon integrating equation (22) from zero to t and using equation (19), we obtain e−At U ( t) ≤ 0 for 0 ≤ t ≤ A/2.

Hence U ( t) ≤ 0 for 0 ≤ t ≤ A/2. However, since A is arbitrary, we conclude that U ( t) ≤ 0 for all nonnegative t. This result and equation (20) are compatible only if U ( t) = 0 for each t ≥ 0. Thus U ( t) = 0 and therefore ψ ( t) = φ ( t) for all t ≥ 0. This contradicts the hypothesis that φ and ψ are two different solutions. Consequently, there cannot be two different solutions of the initial value problem for t ≥ 0. A slight modification of this argument leads to the same conclusion for t ≤ 0.

Returning now to the general problem of solving the integral equation (3), let us consider briefly each of the questions raised earlier: 1. Do all members of the sequence {φ n } exist? In the example, f and ∂ f /∂ y were continuous in the whole t y-plane, and each member of the sequence could be explicitly calculated. In contrast, in the general case, f and ∂ f /∂ y are assumed to be continuous only in the rectangle R: |t| ≤ a, |y| ≤ b (see Figure 2.8.3). Furthermore, the members of the sequence cannot as a rule be explicitly determined. The danger is that at some stage, say, for n = k, the graph of y = φ k ( t) may contain points that lie outside the rectangle R. More precisely, in the computation of φ k+1 ( t) it would be necessary to evaluate f ( t, y) at points where it is not known to be continuous or even to exist. Thus the calculation of φ k+1 ( t) might be impossible. To avoid this danger, it may be necessary to restrict t to a smaller interval than |t| ≤ a. To find such an interval, we make use of the fact that a continuous function on a closed bounded region is bounded. Hence f is bounded on R; thus there exists a positive number M such that | f ( t, y) | ≤ M,

( t, y) in R.

(23)

We have mentioned before that φ n ( 0) = 0 for each n. Since f ( t, φ k ( t) ) is equal to φ k+1 ( t) , the maximum absolute slope of the graph of the equation y = φ k+1 ( t) is M. Since this graph contains the point ( 0, 0) , it must lie in a bow tie-shaped shaded region as shown in Figure 2.8.4. Hence the point ( t, φ k+1 ( t) ) remains in R at least as long as R contains the bow tie-shaped region, which is for |t| ≤ b/ M. We hereafter consider only the rectangle D: |t| ≤ h, |y| ≤ b, where h is equal either to a or to b/ M, whichever is smaller. With this restriction, all members of the sequence {φ n ( t) } exist. Note that whenever b/ M < a, you can try to obtain a larger value of h by finding a better (that is, smaller) bound M for | f ( t, y) |, if this is possible. y (–a, b)

(a, b) R t

(–a, –b)

(a, –b)

FIGURE 2.8.3 Region of

definition for Theorem 2.8.1. y = φ n(t) y

y = φ n(t) y y=b

y=b

t

t y = –b t = –a

t=

b M

(a)

t=

b M

y = –b t = –a

t=a

t=a (b)

FIGURE 2.8.4 Bow-tie regions in which successive iterates lie. (a) if b/ M < a then h = b/ M; (b) if b/ M > a then h = a.

Boyce 9131 Ch02 2

September 29, 2016

17:16

89

2.8 The Existence and Uniqueness Theorem

2. Does the sequence {φ n ( t) } converge? We can identify φ n ( t) = φ 1 ( t) + ( φ 2 ( t) − φ 1 ( t) ) + · · · + ( φ n ( t) − φ n−1 ( t) ) as the n th partial sum of the series φ 1 ( t) +

∞

( φ k+1 ( t) − φ k ( t) ) .

(24)

k=1

The convergence of the sequence {φ n ( t) } is established by showing that the series (24) converges. To do this, it is necessary to estimate the magnitude |φ k+1 ( t) − φ k ( t) | of the general term. The argument by which this is done is indicated in Problems 14 through 17 and will be omitted here. Assuming that the sequence converges, denote the limit function by φ , and so φ ( t) = lim φ n ( t) .

(25)

n→∞

3. What are the properties of the limit function φ ? In the first place, we would like to know that φ is continuous. This is not, however, a necessary consequence of the convergence of the sequence {φ n ( t) }, even though each member of the sequence is itself continuous. Sometimes a sequence of continuous functions converges to a limit function that is discontinuous. A simple example of this phenomenon is given in Problem 11. One way to show that φ is continuous is to show not only that the sequence {φ n } converges, but also that it converges in a certain manner, known as uniform convergence. We do not take up this matter here, but note only that the argument referred to in the discussion of question 2 is sufficient to establish the uniform convergence of the sequence {φ n } and, hence, the continuity of the limit function φ in the interval |t| ≤ h. Now let us return to equation (7) t φ n+1 ( t) = f ( s, φ n ( s) ) ds. 0

Allowing n to approach ∞ on both sides, we obtain t φ ( t) = lim f ( s, φ n ( s) ) ds. n→∞

(26)

0

We would like to interchange the operations of integrating and taking the limit on the right-hand side of equation (26) so as to obtain t φ ( t) = lim f ( s, φ n ( s) ) ds. (27) 0 n→∞

In general, such an interchange is not permissible (see Problem 12, for example), but once again, the fact that the sequence {φ n ( t) } converges uniformly is sufficient to allow us to take the limiting operation inside the integral sign. Next, we wish to take the limit inside the function f , which would give t φ ( t) = f s, lim φ n ( s) ds (28) n→∞

0

and hence

φ ( t) =

t

f ( s, φ ( s) ) ds.

(29)

0

The statement that

lim f ( s, φ n ( s) ) = f s, lim φ n ( s)

n→∞

n→∞

is equivalent to the statement that f is continuous in its second variable, which is known by hypothesis. Hence equation (29) is valid, and the function φ satisfies the integral equation (3). Thus y = φ ( t) is also a solution of the initial value problem (2).

89

Boyce 9131 Ch02 2

90

September 29, 2016

17:16

90

CHAPTER 2 First-Order Differential Equations

4. Are there other solutions of the integral equation (3) besides y = φ ( t) ? To show the uniqueness of the solution y = φ ( t) , we can proceed much as in the example. First, assume the existence of another solution y = ψ ( t) . It is then possible to show (see Problem 18) that the difference φ ( t) − ψ ( t) satisfies the inequality t |φ ( t) − ψ ( t) | ≤ A |φ ( s) − ψ ( s) | ds (30) 0

for 0 ≤ t ≤ h and a suitable positive number A. From this point the argument is identical to that given in the example, and we conclude that there is no solution of the initial value problem (2) other than the one generated by the method of successive approximations.

Problems In each of Problems 1 and 2, transform the given initial value problem into an equivalent problem with the initial point at the origin.

1. dy/dt = t 2 + y 2 , y( 1) = 2 2. dy/dt = 1 − y 3 , y( −1) = 3 In each of Problems 3 through 4, let φ 0 ( t) = 0 and define {φ n ( t) } by the method of successive approximations. a. Determine φ n ( t) for an arbitrary value of n. G b. Plot φ n ( t) for n = 1, . . . , 4. Observe whether the iterates appear to be converging. c. Express lim φ n ( t) = φ ( t) in terms of elementary

11. Let φ n ( x) = x n for 0 ≤ x ≤ 1 and show that lim φ n ( x) = n→∞

N N

y( 0) = 0

In each of Problems 7 and 8, let φ 0 ( t) = 0 and use the method of successive approximations to approximate the solution of the given initial value problem. a. Calculate φ 1 ( t) , . . . , φ 3 ( t) . G b. Plot φ 1 ( t) , . . . , φ 3 ( t) . Observe whether the iterates appear to be converging.

7. y = t 2 + y 2 , y( 0) = 0 8. y = 1 − y 3 , y( 0) = 0 In each of Problems 9 and 10, let φ 0 ( t) = 0 and use the method of successive approximations to approximate the solution of the given initial value problem. a. Calculate φ 1 ( t) , . . . , φ 4 ( t) , or (if necessary) Taylor approximations to these iterates. Keep terms up to order six. G b. Plot the functions you found in part a and observe whether they appear to be converging.

9. y = − sin y + 1, 10.

3t 2 + 4t + 2 , y = 2( y − 1)

y( 0) = 0 y( 0) = 0

x = 1.

2

n→∞

1

lim φ n ( x) d x = 0. 0 n→∞

1

b. Show that

2

2nxe−nx d x = 1 − e−n ; hence

0

3. y = 2( y + 1) , y( 0) = 0 4. y = −y/2 + t, y( 0) = 0

6. y = t 2 y − t,

1,

12. Consider the sequence φ n ( x) = 2nxe−nx , 0 ≤ x ≤ 1. a. Show that lim φ n ( x) = 0 for 0 ≤ x ≤ 1; hence

In each of Problems 5 and 6, let φ 0 ( t) = 0 and use the method of successive approximations to solve the given initial value problem. a. Determine φ n ( t) for an arbitrary value of n. G b. Plot φ n ( t) for n = 1, . . . , 4. Observe whether the iterates appear to be converging. c. Show that the sequence {φ n ( t) } converges. 5. y = t y + 1, y( 0) = 0

0 ≤ x < 1,

This example shows that a sequence of continuous functions may converge to a limit function that is discontinuous.

n→∞

functions; that is, solve the given initial value problem. G d. Plot |φ ( t) − φ n ( t) | for n = 1, . . . , 4. For each of φ 1 ( t) , . . . , φ 4 ( t) , estimate the interval in which it is a reasonably good approximation to the actual solution.

0,

1

φ n ( x) d x = 1.

lim n→∞

0

Thus, in this example,

b

φ n ( x) d x =

lim n→∞

b

a

lim φ n ( x) d x, a

n→∞

even though lim φ n ( x) exists and is continuous. n→∞

13. a. Verify that φ ( t) =

∞ t 2k k=1

k!

is a solution of the integral

equation (9). b. Verify that φ ( t) is also a solution of the initial value problem (6). ∞ tk c. Use the fact that = et to evaluate φ ( t) in terms of k! k=0 elementary functions. d. Solve initial value problem (6) as a separable equation. e. Solve initial value problem (6) as a first order linear equation. In Problems 14 through 17, we indicate how to prove that the sequence {φ n ( t) }, defined by equations (4) through (7), converges. ∞ t 2k a. Verify that φ ( t) = is a solution of the integral k! k=1 equation (9). b. Verify that φ ( t) is also a solution of the initial value problem (6). ∞ tk = et to evaluate φ ( t) in terms of c. Use the fact that k! k=0 elementary functions. d. Solve initial value problem (6) as a separable equation. e. Solve initial value problem (6) as a first order linear equation.

Boyce 9131 Ch02 2

September 29, 2016

17:16

91

2.9 First-Order Difference Equations

91

14. If ∂ f /∂ y is continuous in the rectangle D, show that there is a

a. Show that

positive constant K such that

|φ n ( t) | ≤ |φ 1 ( t) | + |φ 2 ( t) − φ 1 ( t) | + · · · + |φ n ( t) − φ n−1 ( t) |.

| f ( t, y1 ) − f ( t, y2 ) | ≤ K |y1 − y2 |,

(31)

where ( t, y1 ) and ( t, y2 ) are any two points in D having the same t coordinate. This inequality is known as a Lipschitz22 condition. Hint: Hold t fixed and use the mean value theorem on f as a function of y only. Choose K to be the maximum value of |∂ f /∂ y| in D.

15. If φ n−1 ( t) and φ n ( t) are members of the sequence {φ n ( t) }, use the result of Problem 14 to show that

f ( t, φ n ( t) ) − f t, φ n−1 ( t) ≤ K φ n ( t) − φ n−1 ( t) .

b. Use the results of Problem 16 to show that 2 |φ n ( t) | ≤

M K

Kh +

( K h) ( K h) n + ··· + 2! n!

.

c. Show that the sum in part b converges as n → ∞ and, hence,

the sum in part a also converges as n → ∞. Conclude therefore that the sequence {φ n ( t) } converges since it is the sequence of partial sums of a convergent infinite series.

18. In this problem we deal with the question of uniqueness of the solution of the integral equation (3)

16. a. Show that if |t| ≤ h, then |φ 1 ( t) | ≤ M|t|,

|φ n ( t) − φ n−1 ( t) | ≤

17. Note that

n!

|t|

b. Show that

h MK ≤ . n!

( f ( s, φ ( s) ) − f ( s, ψ ( s) ) ) ds.

( f ( s, φ ( s) ) − f ( s, ψ ( s) ) ) ds.

c. Use the result of Problem 14 to show that t

22 The German mathematician Rudolf Lipschitz (1832--1903), professor at the University of Bonn for many years, worked in several areas of mathematics. The inequality (i) can replace the hypothesis that ∂ f /∂ y is continuous in Theorem 2.8.1; this results in a slightly stronger theorem.

0

where K is an upper bound for |∂ f /∂ y| in D. This is the same as equation (30), and the rest of the proof may be constructed as indicated in the text.

First-Order Difference Equations

Although a continuous model leading to a differential equation is reasonable and attractive for many problems, there are some cases in which a discrete model may be more natural. For instance, the continuous model of compound interest used in Section 2.3 is only an approximation to the actual discrete process. Similarly, sometimes population growth may be described more accurately by a discrete model than by a continuous model. This is true, for example, of species whose generations do not overlap and that propagate at regular intervals, such as at particular times of the calendar year. Then the population yn+1 of the species in the year n + 1 is some function of n and the population yn in the preceding year; that is, n = 0, 1, 2, . . . .

|φ ( s) − ψ ( s) | ds,

|φ ( t) − ψ ( t) | ≤ K

.............................................................................................................................

yn+1 = f ( n, yn ) ,

t

0

φ n ( t) = φ 1 ( t) + φ 2 ( t) − φ 1 ( t) + · · · + φ n ( t) − φ n−1 ( t) .

2.9

t

|φ ( t) − ψ ( t) | ≤

n−1 n

that, for t ≥ 0,

0

c. Show, by mathematical induction, that MK

a. Suppose that φ and ψ are two solutions of equation (3). Show φ ( t) − ψ ( t) =

M K |t|2 . |φ 2 ( t) − φ 1 ( t) | ≤ 2 n

f ( s, φ ( s) ) ds. 0

where M is chosen so that | f ( t, y) | ≤ M for ( t, y) in D. b. Use the results of Problem 15 and part a of Problem 16 to show that

n−1

t

φ ( t) =

(1)

Equation (1) is called a first-order difference equation. It is first-order because the value of yn+1 depends on the value of yn but not on earlier values yn−1 , yn−2 , and so forth. As for differential equations, the difference equation (1) is linear if f is a linear function of yn ; otherwise, it is nonlinear. A solution of the difference equation (1) is a sequence of numbers y0 , y1 , y2 , . . . that satisfy the equation for each n. In addition to the difference equation itself, there may also be an initial condition y0 = α (2) that prescribes the value of the first term of the solution sequence. We now assume temporarily that the function f in equation (1) depends only on yn , but not on n. In this case yn+1 = f ( yn ) , n = 0, 1, 2, . . . . (3)

Boyce 9131 Ch02 2

92

September 29, 2016

17:16

92

CHAPTER 2 First-Order Differential Equations

If y0 is given, then successive terms of the solution can be found from equation (3). Thus y1 = f ( y0 ) , and y2 = f ( y1 ) = f ( f ( y0 ) ) . The quantity f ( f ( y0 ) ) is called the second iterate of the difference equation and is sometimes denoted by f 2 ( y0 ) . Similarly, the third iterate y3 is given by y3 = f ( y2 ) = f f f ( y0 ) = f 3 ( y0 ) , and so on. In general, the n th iterate yn is yn = f ( yn−1 ) = f n ( y0 ) . This procedure is referred to as iterating the difference equation. It is often of primary interest to determine the behavior of yn as n → ∞. In particular, does yn approach a limit, and if so, what is it? Solutions for which yn has the same value for all n are called equilibrium solutions. They are frequently of special importance, just as in the study of differential equations. If equilibrium solutions exist, you can find them by setting yn+1 equal to yn in equation (3) and solving the resulting equation yn = f ( yn )

(4)

for yn . Linear Equations. Suppose that the population of a certain species in a given region in year n + 1, denoted by yn+1 , is a positive multiple ρ n of the population yn in year n; that is, yn+1 = ρ n yn ,

n = 0, 1, 2, . . . .

(5)

Note that the reproduction rate ρ n may differ from year to year. The difference equation (5) is linear and can easily be solved by iteration. We obtain y1 = ρ 0 y0 , y2 = ρ 1 y1 = ρ 1 ρ 0 y0 , and, in general, yn = ρ n−1 · · · ρ 0 y0 ,

n = 1, 2, . . . .

(6)

Thus, if the initial population y0 is given, then the population of each succeeding generation is determined by equation (6). Although for a population problem ρ n is intrinsically positive, the solution (6) is also valid if ρ n is negative for some or all values of n. Note, however, that if ρ n is zero for some n, then yn+1 and all succeeding values of y are zero; in other words, the species has become extinct. If the reproduction rate ρ n has the same value ρ for each n, then the difference equation (5) becomes yn+1 = ρ yn

(7)

yn = ρ n y0 .

(8)

and its solution is Equation (7) also has an equilibrium solution, namely, yn = 0 for all n, corresponding to the initial value y0 = 0. The limiting behavior of yn is easy to determine from equation (8). In fact, ⎧ if |ρ| < 1; ⎪ ⎨0, if ρ = 1; lim yn = y0 , (9) ⎪ n→∞ ⎩ does not exist, otherwise. In other words, the equilibrium solution yn = 0 is asymptotically stable for |ρ| < 1 and unstable for |ρ| > 1.

Boyce 9131 Ch02 2

September 29, 2016

17:16

93

2.9 First-Order Difference Equations

Now we will modify the population model represented by equation (5) to include the effect of immigration or emigration. If bn is the net increase in population in year n due to immigration, then the population in year n + 1 is the sum of the part of the population resulting from natural reproduction and the part due to immigration. Thus yn+1 = ρ yn + bn ,

n = 0, 1, 2, . . . ,

(10)

where we are now assuming that the reproduction rate ρ is constant. We can solve equation (10) by iteration in the same manner as before. We have y1 = ρ y0 + b0 , y2 = ρ ( ρ y0 + b0 ) + b1 = ρ 2 y0 + ρb0 + b1 , y3 = ρ ( ρ 2 y0 + ρb0 + b1 ) + b2 = ρ 3 y0 + ρ 2 b0 + ρb1 + b2 , and so forth. In general, we obtain yn = ρ y0 + ρ n

n−1

b0 + · · · + ρbn−2 + bn−1 = ρ y0 + n

n−1

ρ n−1− j b j .

(11)

j=0

Note that the first term on the right-hand side of equation (11) represents the descendants of the original population, while the other terms represent the population in year n resulting from immigration in all preceding years. In the special case where bn = b = 0 for all n, the difference equation is yn+1 = ρ yn + b,

(12)

yn = ρ n y0 + ( 1 + ρ + ρ 2 + · · · + ρ n−1 ) b.

(13)

and from equation (11) its solution is

If ρ = 1, we can write this solution in the more compact form yn = ρ n y0 +

1 − ρn b, 1−ρ

(14)

where again the two terms on the right-hand side are the effects of the original population and of immigration, respectively. Rewriting equation (14) as b b yn = ρ n y0 − (15) + 1−ρ 1−ρ makes the long-time behavior of yn more evident. It follows from equation (15) that yn → b/( 1−ρ) if |ρ| < 1. If |ρ| > 1 or if ρ = −1 then yn has no limit unless y0 = b/( 1−ρ) . The quantity b/( 1 − ρ) , for ρ = 1, is an equilibrium solution of equation (12), as can readily be seen directly from that equation. Of course, equation (14) is not valid for ρ = 1. To deal with that case, we must return to equation (13) and let ρ = 1 there. It follows that yn = y0 + nb,

(16)

so in this case yn becomes unbounded as n → ∞. The same model also provides a framework for solving many problems of a financial character. For such problems, yn is the account balance in the nth time period, ρ n = 1 + rn , where rn is the interest rate for that period, and bn is the amount deposited or withdrawn. The following example is typical.

EXAMPLE 1 A recent college graduate takes out a $10,000 loan to purchase a car. If the interest rate is 12%, what monthly payment is required to pay off the loan in 4 years? Solution:

▼

The relevant difference equation is equation (12), where yn is the loan balance outstanding in the n th month, ρ = 1 + r , where r is the interest rate per month, and b is the effect of the monthly payment.

93

Boyce 9131 Ch02 2

94

September 29, 2016

17:16

94

CHAPTER 2 First-Order Differential Equations

▼ Note that ρ = 1.01, corresponding to a monthly interest rate of 1%. Since payments reduce the loan

balance, b must be negative; the actual payment is |b|. The solution of the difference equation (12) with this value for ρ and the initial condition y0 = 10,000 is given by equation (15); that is, yn = ( 1.01) n ( 10,000 + 100b) − 100b.

(17)

The value of b needed to pay off the loan in 4 years is found by setting y48 = 0 and solving for b. This gives b = −100

( 1.01) 48 = −263.34. ( 1.01) 48 − 1

(18)

The total amount paid on the loan is 48 times |b|, or $12,640.32. Of this amount, $10,000 is repayment of the principal and the remaining $2640.32 is interest.

Nonlinear Equations. Nonlinear difference equations are much more complicated and have much more varied solutions than linear equations. We will restrict our attention to a single equation, the logistic difference equation

yn yn+1 = ρ yn 1 − (19) , k which is analogous to the logistic differential equation

dy y = ry 1 − dt K

(20)

that was discussed in Section 2.5. Note that if the derivative dy/dt in equation (20) is replaced by the difference quotient ( yn+1 − yn ) / h, then equation (20) reduces to equation (19) with ρ = 1 + hr and k = ( 1 + hr ) K / hr . To simplify equation (19) a little more, we can scale the variable yn by introducing the new variable u n = yn / k. Then equation (19) becomes u n+1 = ρu n ( 1 − u n ) ,

(21)

where ρ is a positive parameter. We begin our investigation of equation (21) by seeking the equilibrium, or constant, solutions. These can be found by setting u n+1 equal to u n in equation (21), which corresponds to setting dy/dt equal to zero in equation (20). The resulting equation is u n = ρu n − ρu 2n ,

(22)

so it follows that the equilibrium solutions of equation (21) are u n = 0,

un =

ρ −1 . ρ

(23)

The next question is whether the equilibrium solutions are asymptotically stable or unstable. That is, for an initial condition near one of the equilibrium solutions, does the resulting solution sequence approach or depart from the equilibrium solution? One way to examine this question is by approximating equation (21) by a linear equation in the neighborhood of an equilibrium solution. For example, near the equilibrium solution 2 u n = 0, the quantity u n is small compared to u n itself, so we assume that we can neglect the quadratic term in equation (21) in comparison with the linear terms. This leaves us with the linear difference equation u n+1 = ρu n ,

(24)

which is presumably a good approximation to equation (21) for u n sufficiently near zero. However, equation (24) is the same as equation (7), and we have already concluded, in equation (9), that u n → 0 as n → ∞ if and only if |ρ| < 1, or (since ρ must be positive) for 0 < ρ < 1. Thus the equilibrium solution u n = 0 is asymptotically stable for the linear approximation (24) for this set of ρ values, so we conclude that it is also asymptotically stable for the full nonlinear equation (21). The previous conclusion is correct, although our argument is not complete. What is lacking is a theorem stating that the solutions of the nonlinear equation (21) resemble those of

Boyce 9131 Ch02 2

September 29, 2016

17:16

95

2.9 First-Order Difference Equations

the linear equation (24) near the equilibrium solution u n = 0. We will not take time to discuss this issue here; the same question is treated for differential equations in Section 9.3. Now consider the other equilibrium solution u n = ( ρ − 1) /ρ. To study solutions in the neighborhood of this point, we write un =

ρ −1 + vn , ρ

(25)

where we assume that v n is small. By substituting from equation (25) in equation (21) and simplifying the resulting equation, we eventually obtain v n+1 = ( 2 − ρ) v n − ρv n2 .

(26)

Since v n is small, we again neglect the quadratic term in comparison with the linear terms and thereby obtain the linear equation v n+1 = ( 2 − ρ) v n .

(27)

Referring to equation (9) once more, we find that v n → 0 as n → ∞ for |2 − ρ| < 1, or in other words for 1 < ρ < 3. Therefore, we conclude that for this range of values of ρ, the equilibrium solution u n = ( ρ − 1) /ρ is asymptotically stable. Figure 2.9.1 contains the graphs of solutions of equation (21) for ρ = 0.8, ρ = 1.5, and ρ = 2.8, respectively. Observe that the solution converges to zero for ρ = 0.8 and to the nonzero equilibrium solution for ρ = 1.5 and ρ = 2.8. The convergence is (eventually) monotone for ρ = 0.8 and for ρ = 1.5 and is oscillatory for ρ = 2.8. The graphs shown are for particular initial conditions, but the graphs for other initial conditions are similar.

un

un

un

0.8

0.8

0.8

0.6

0.6

0.4

0.4

0.4

0.2

0.2

0.2

2

4

6 ( a)

8

n

un =

2

0.5 1.5

=

4

un =

1.8 2.8

~ = 0.6429

0.6

1 3

6

8

n

2

( b)

FIGURE 2.9.1 Solutions of u n+1 = ρu n ( 1 − u n ) : (a) ρ = 0.8; (b) ρ = 1.5; (c) ρ = 2.8.

Another way of displaying the solution of a difference equation is shown in Figure 2.9.2. In each part of this figure, the graphs of the parabola y = ρ x( 1 − x) and of the straight line y = x are shown. The equilibrium solutions correspond to the points of intersection of these two curves. The piecewise linear graph consisting of successive vertical and horizontal line segments, sometimes called a stairstep or cobweb diagram, represents the solution sequence. The sequence starts at the point u 0 on the x-axis. The vertical line segment drawn upward to the parabola at u 0 corresponds to the calculation of ρu 0 ( 1 − u 0 ) = u 1 . This value is then transferred from the y-axis to the x-axis; this step is represented by the horizontal line segment from the parabola to the line y = x. Then the process is repeated over and over again. Clearly, the sequence converges to the origin in Figure 2.9.2a and to the nonzero equilibrium solution in the other two cases. To summarize our results so far: the difference equation (21) has two equilibrium solutions, u n = 0 and u n = ( ρ − 1) /ρ; the former is asymptotically stable for 0 ≤ ρ < 1, and the latter is asymptotically stable for 1 < ρ < 3. When ρ = 1, the two equilibrium solutions coincide at u = 0; this solution can be shown to be asymptotically stable. In Figure 2.9.3 the parameter ρ is plotted on the horizontal axis and u on the vertical axis. The equilibrium solutions u = 0 and u = ( ρ − 1) /ρ are shown. The intervals in which each one is asymptotically stable are indicated by the solid portions of the curves. There is an exchange of stability from one equilibrium solution to the other at ρ = 1.

4

6 ( c)

8

n

95

Boyce 9131 Ch02 2

96

September 29, 2016

17:16

96

CHAPTER 2 First-Order Differential Equations

y

y

1

1

0.8

0.8

y=x

0.6

0.6

0.4

0.4 y = ρ x (1 – x)

0.2

y=x

y = ρ x (1 – x) 1, 1 3 3

0.2

u0 = 0.85

u0 = 0.3 0.2

0.4

0.6

0.8

1

0.2

x

0.4

(a)

0.6

0.8

1 x

(b)

y 1 y=x

0.8 y = ρ x (1 – x)

(0.6429..., 0.6429...)

0.6 0.4 0.2

u0 = 0.3 0.2

0.4

0.6 (c)

0.8

1 x

FIGURE 2.9.2 Iterates of u n+1 = ρu n ( 1 − u n ) : (a) ρ = 0.8; (b) ρ = 1.5; (c) ρ = 2.8.

u 1 u = (ρ – 1)/ρ 0.5

Asymptotically stable u=0 1

–0.5

2

3

ρ

Unstable

FIGURE 2.9.3 Exchange of stability for u n+1 = ρu n ( 1 − u n ) .

For ρ > 3, neither of the equilibrium solutions is stable, and the solutions of equation (21) exhibit increasing complexity as ρ increases. For ρ somewhat greater than 3, the sequence u n rapidly approaches a steady oscillation of period 2; that is, u n oscillates back and forth between two distinct values. For ρ = 3.2, a solution is shown in Figure 2.9.4. For n greater than about

Boyce 9131 Ch02 2

September 29, 2016

17:16

97

2.9 First-Order Difference Equations

un 0.7995

0.8 0.6

0.5130

0.4 0.2 10

20

30

(a)

40

n

y 1 0.8

0.7995 y = ρ x (1 – x)

0.6 0.5130 0.4

0.5130

0.7995

y=x 0.2

0.2

0.4

0.6

0.8

1 x

(b)

FIGURE 2.9.4 A solution of u n+1 = ρu n ( 1 − u n ) for ρ = 3.2; period 2. (a) u n versus n; (b) the cobweb diagram shows the iterates are in a two-cycle.

20, the solution alternates between the values 0.5130 and 0.7995. The graph is drawn for the particular initial condition u 0 = 0.3, but it is similar for all other initial values between 0 and 1. Figure 2.9.4b also shows the same steady oscillation as a rectangular path that is traversed repeatedly in the clockwise direction. At about ρ = 3.449, each state in the oscillation of period 2 separates into two distinct states, and the solution becomes periodic with period 4; see Figure 2.9.5, which shows a solution of period 4 for ρ = 3.5. As ρ increases further, periodic solutions of period 8, 16, . . . appear. The transition from solutions with one period to solutions with a new period that occurs at a certain parameter value is called a bifurcation; the value of the parameter where the bifurcation occurs is called a bifurcation value of the parameter. The values of ρ at which the successive period doublings occur approach a limit that is approximately 3.57. For ρ > 3.57, the solutions possess some regularity but no discernible detailed pattern for most values of ρ. For example, a solution for ρ = 3.65 is shown in Figure 2.9.6. It oscillates between approximately 0.3 and 0.9, but its fine structure is unpredictable. The term chaotic is used to describe this situation. One of the features of chaotic solutions is extreme sensitivity to the initial conditions. This is illustrated in Figure 2.9.7, where two solutions of equation (21) for ρ = 3.65 are shown. One solution is the same as that in Figure 2.9.6 and has the initial value u 0 = 0.3, while the other solution has the initial value u 0 = 0.305. For about 15 iterations the two solutions remain close and are hard to distinguish from each other in the figure. After that, although they continue to wander about in approximately the same set of values, their graphs are quite dissimilar. It would certainly not be possible to use one of these solutions to estimate the value of the other for values of n larger than about 15.

97

Boyce 9131 Ch02 2

98

September 29, 2016

17:16

98

CHAPTER 2 First-Order Differential Equations

un 0.8 0.4

4

8

12

16

20 24 (a)

28

32

36

40

n

y 1 y=x y = ρ x (1 – x)

0.5

0.5009 0.3828

0.5

0.8269 0.8750

1

(b)

x

FIGURE 2.9.5 A solution of u n+1 = ρu n ( 1 − u n ) for ρ = 3.5; period 4. (a) u n versus n; (b) the cobweb diagram shows the iterates are in a four-cycle.

un 0.9 0.8 0.7 0.6 0.5 0.4 0.3 10

20

30

40

50

60 n

FIGURE 2.9.6 A solution of u n+1 = ρu n ( 1 − u n ) for ρ = 3.65; a chaotic solution.

It is only comparatively recently that chaotic solutions of difference and differential equations have become widely known. Equation (20) was one of the first instances of mathematical chaos to be found and studied in detail, by Robert May23 in 1974. On the basis ......................................................................................................................................................................... 23 Robert M. May (1936--) was born in Sydney, Australia, and received his education at the University of Sydney with a doctorate in theoretical physics in 1959. His interests soon turned to population dynamics and theoretical ecology; the work cited in the text is described in two papers listed in the References at the end of this chapter. He has held professorships at Sydney, at Princeton, at Imperial College (London), and (since 1988) at Oxford.

Boyce 9131 Ch02 2

September 29, 2016

17:16

99

Problems

un 0.9 0.8

+ + +

+

0.7 0.6 0.5

+ + + ++ + + + + + + + +

+ +

+ + 10

+

+

+

20

+

+

+

+

++

+

+

+

+

+

+ + +

+ +

+

+

0.4 0.3

+

+

+

+

+ + +

+ 30

+ +

+ 40

99

+

+ 50

+

++

60 n

FIGURE 2.9.7 Two solutions of u n+1 = ρu n ( 1 − u n ) for ρ = 3.65; u 0 = 0.3 and u 0 = 0.305.

of his analysis of this equation as a model of the population of certain insect species, May suggested that if the growth rate ρ is too large, then it will be impossible to make effective long-range predictions about these insect populations. The occurrence of chaotic solutions in seemingly simple problems has stimulated an enormous amount of research, but many questions remain unanswered. It is increasingly clear, however, that chaotic solutions are much more common than was suspected at first and that they may be a part of the investigation of a wide range of phenomena.

Problems In each of Problems 1 through 4, solve the given difference equation in terms of the initial value y0 . Describe the behavior of the solution as n → ∞.

1. yn+1 = −0.9yn n+3 yn 2. yn+1 = n+1

3. yn+1 = ( −1) yn 4. yn+1 = 0.5yn + 6 5. An investor deposits $1000 in an account paying interest at a n+1

rate of 8%, compounded monthly, and also makes additional deposits of $25 per month. Find the balance in the account after 3 years.

6. A certain college graduate borrows $8000 to buy a car. The lender charges interest at an annual rate of 10%. What monthly payment rate is required to pay off the loan in 3 years? Compare your result with that of Problem 7 in Section 2.3. 7. A homebuyer takes out a mortgage of $100,000 with an interest rate of 9%. What monthly payment is required to pay off the loan in 30 years? In 20 years? What is the total amount paid during the term of the loan in each of these cases? 8. If the interest rate on a 20-year mortgage is fixed at 10% and if a monthly payment of $1000 is the maximum that the buyer can afford, what is the maximum mortgage loan that can be made under these conditions?

9. A homebuyer wishes to finance the purchase with a $95,000 mortgage with a 20-year term. What is the maximum interest rate the buyer can afford if the monthly payment is not to exceed $900? The Logistic Difference Equation. Problems 10 through 15 deal with the difference equation (21), u n+1 = ρu n ( 1 − u n ) .

10. Carry out the details in the linear stability analysis of the equilibrium solution u n = ( ρ − 1) /ρ. That is, derive the difference equation (26) in the text for the perturbation v n .

11.

N a. For ρ = 3.2, plot or calculate the solution of the logistic equation (21) for several initial conditions, say, u 0 = 0.2, 0.4, 0.6, and 0.8. Observe that in each case the solution approaches a steady oscillation between the same two values. This illustrates that the long-term behavior of the solution is independent of the initial conditions. N b. Make similar calculations and verify that the nature of the solution for large n is independent of the initial condition for other values of ρ, such as 2.6, 2.8, and 3.4.

12. Assume that ρ > 1 in equation (21). G a. Draw a qualitatively correct stairstep diagram and thereby show that if u 0 < 0, then u n → −∞ as n → ∞. G b. In a similar way, determine what happens as n → ∞ if u 0 > 1.

Boyce 9131 Ch02 2

100

September 29, 2016

17:16

100

CHAPTER 2 First-Order Differential Equations

b. Let δ n = ( ρ n − ρ n−1)/( ρ n+1 − ρ n ) . It can be shown that δ n

13. The solutions of equation (21) change from convergent

approaches a limit δ as n → ∞, where δ ∼ = 4.6692 is known as the Feigenbaum24 number. Determine the percentage difference between the limiting value δ and δ 2 , as calculated in part a. c. Assume that δ 3 = δ and use this relation to estimate ρ 4 , the value of ρ at which solutions of period 16 appear. G d. By plotting or calculating solutions near the value of ρ 4 found in part c, try to detect the appearance of a period 16 solution. e. Observe that

sequences to periodic oscillations of period 2 as the parameter ρ passes through the value 3. To see more clearly how this happens, carry out the following calculations. N a. Plot or calculate the solution for ρ = 2.9, 2.95, and 2.99, respectively, using an initial value u 0 of your choice in the interval ( 0, 1) . In each case estimate how many iterations are required for the solution to get “very close” to the limiting value. Use any convenient interpretation of what “very close” means in the preceding sentence. N b. Plot or calculate the solution for ρ = 3.01, 3.05, and 3.1, respectively, using the same initial condition as in part a. In each case estimate how many iterations are needed to reach a steady-state oscillation. Also find or estimate the two values in the steady-state oscillation.

ρ n = ρ 1 + ( ρ 2 − ρ 1 ) + ( ρ 3 − ρ 2 ) + · · · + ( ρ n − ρ n−1 ) . Assuming that ρ 4 − ρ 3 = ( ρ 3 − ρ 2 ) δ −1 ,

N

14. By calculating or plotting the solution of equation (21) for different values of ρ, estimate the value of ρ at which the solution changes from an oscillation of period 2 to one of period 4. In the same way, estimate the value of ρ at which the solution changes from period 4 to period 8. N

15. Let ρ k be the value of ρ at which the solution of

equation (21) changes from period 2k−1 to period 2k . Thus, as noted in the text, ρ 1 = 3, ρ 2 ∼ = 3.449, and ρ 3 ∼ = 3.544. a. Using these values of ρ 1 , ρ 2 , and ρ 3 , or those you found in Problem 14, calculate ( ρ 2 − ρ 1 ) /( ρ 3 − ρ 2 ) .

ρ 5 − ρ 4 = ( ρ 3 − ρ 2 ) δ −2 ,

and so forth, express ρ n as a geometric sum. Then find the limit ρ n as n → ∞. This is an estimate of the value of ρ at which the onset of chaos occurs in the solution of the logistic equation (21). .............................................................................................................................. 24 This result for the logistic difference equation was discovered in August 1975 by Mitchell Feigenbaum (1944--), while he was working at the Los Alamos National Laboratory. Within a few weeks he had established that the same limiting value also appears in a large class of period-doubling difference equations. Feigenbaum, who has a doctorate in physics from M.I.T., is now at Rockefeller University.

Chapter Review Problems Miscellaneous Problems. One of the difficulties in solving firstorder differential equations is that there are several methods of solution, each of which can be used on a certain type of equation. It may take some time to become proficient in matching solution methods with equations. The first 24 of the following problems are presented to give you some practice in identifying the method or methods applicable to a given equation. The remaining problems involve certain types of equations that can be solved by specialized methods. In each of Problems 1 through 24, solve the given differential equation. If an initial condition is given, also find the solution that satisfies it. dy x 3 − 2y = 1. dx x 1 + cos x dy = 2. dx 2 − sin y dy 2x + y , y( 0) = 0 3. = dx 3 + 3y 2 − x dy = 3 − 6x + y − 2x y 4. dx dy 2x y + y 2 + 1 5. =− dx x 2 + 2x y dy 6. x + x y = 1 − y, y( 1) = 0 dx dy sin x 7. x + 2y = , y( 2) = 1 dx x dy 2x y + 1 8. =− 2 dx x + 2y dy 2 =0 9. ( x y + x y − y) + ( x 2 y − 2x 2 ) dx

dy =0 dx dy ( x + y) + ( x + 2y) = 0, y( 2) = 3 dx dy = y − ye x ( e x + 1) dx dy e−x cos y − e2y cos x = dx −e−x sin y + 2e2y sin x dy = e2x + 3y dx dy 2 + 2y = e−x −2x , y( 0) = 3 dx dy 3x 2 − 2y − y 3 = dx 2x + 3x y 2 y = e x+y 2y 2 + 6x y − 4 dy + 2 =0 dx 3x + 4x y + 3y 2 dy + ( t + 1) y = e2t t dt x y = y + xe y/ x dy x Hint: Let u = x 2 . = 2 dx x y + y3 x+y dy = dx x−y

10. ( x 2 + y) + ( x + e y ) 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22.

dy =0 dx y( 1) = 2

23. ( 3y 2 + 2x y) − ( 2x y + x 2 ) 24. x y + y − y 2 e2x = 0,

Boyce 9131 Ch02 2

September 29, 2016

17:16

101

References

25. Riccati Equations. The equation dy = q1 ( t) + q2 ( t) y + q3 ( t) y 2 dt is known as a Riccati25 equation. Suppose that some particular solution y1 of this equation is known. A more general solution containing one arbitrary constant can be obtained through the substitution y = y1 ( t) +

1 . v( t)

Show that v( t) satisfies the first-order linear equation dv = −( q2 + 2q3 y1 ) v − q3 . dt Note that v( t) will contain a single arbitrary constant.

26. Verify that the given function is a particular solution of the given Riccati equation. Then use the method of Problem 25 to solve the following Riccati equations: a. y = 1 + t 2 − 2t y + y 2 ; y1 ( t) = t 1 y 1 b. y = − 2 − + y 2 ; y1 ( t) = t t t 2 cos2 t − sin2 t + y 2 dy = ; y1 ( t) = sin t c. dt 2 cos t 27. The propagation of a single action in a large population (for example, drivers turning on headlights at sunset) often depends partly on external circumstances (gathering darkness) and partly on a tendency to imitate others who have already performed the action in question. In this case the proportion y( t) of people who have performed the action can be described26 by the equation dy/dt = ( 1 − y) ( x( t) + by) ,

(28)

where x( t) measures the external stimulus and b is the imitation coefficient. a. Observe that equation (28) is a Riccati equation and that y1 ( t) = 1 is one solution. Use the transformation suggested in Problem 25, and find the linear equation satisfied by v( t) . b. Find v( t) in the case that x( t) = at, where a is a constant. Leave your answer in the form of an integral. ............................................................................................................................. 25 Riccati

equations are named for Jacopo Francesco Riccati (1676--1754), a Venetian nobleman, who declined university appointments in Italy, Austria, and Russia to pursue his mathematical studies privately at home. Riccati studied these equations extensively; however, it was Euler (in 1760) who discovered the result stated in this problem. 26 See

Anatol Rapoport, “Contribution to the Mathematical Theory of Mass Behavior: I. The Propagation of Single Acts,” Bulletin of Mathematical Biophysics 14 (1952), pp. 159--169, and John Z. Hearon, “Note on the Theory of Mass Behavior,” Bulletin of Mathematical Biophysics 17 (1955), pp. 7--13.

101

Some Special Second-Order Differential Equations. Second-order differential equations involve the second derivative of the unknown function and have the general form y = f ( t, y, y ) . Usually, such equations cannot be solved by methods designed for first-order equations. However, there are two types of second-order equations that can be transformed into first-order equations by a suitable change of variable. The resulting equation can sometimes be solved by the methods presented in this chapter. Problems 28 through 37 deal with these types of equations. Equations with the Dependent Variable Missing. For a secondorder differential equation of the form y = f ( t, y ) , the substitution v = y , v = y leads to a first-order differential equation of the form v = f ( t, v) . If this equation can be solved for v, then y can be obtained by integrating dy/dt = v. Note that one arbitrary constant is obtained in solving the first-order equation for v, and a second is introduced in the integration for y. In each of Problems 28 through 31, use this substitution to solve the given equation.

28. 29. 30. 31.

t 2 y + 2t y − 1 = 0, t y + y = 1,

t>0

t>0

y + t ( y ) 2 = 0 2t 2 y + ( y ) 3 = 2t y ,

t>0

Equations with the Independent Variable Missing. Consider second-order differential equations of the form y = f ( y, y ) , in which the independent variable t does not appear explicitly. If we let v = y , then we obtain dv/dt = f ( y, v) . Since the righthand side of this equation depends on y and v, rather than on t and v, this equation contains too many variables. However, if we think of y as the independent variable, then by the chain rule, dv/dt = ( dv/dy) ( dy/dt) = v( dv/dy) . Hence the original differential equation can be written as v( dv/dy) = f ( y, v) . Provided that this first-order equation can be solved, we obtain v as a function of y. A relation between y and t results from solving dy/dt = v( y) , which is a separable equation. Again, there are two arbitrary constants in the final result. In each of Problems 32 through 35, use this method to solve the given differential equation.

32. 33. 34. 35.

yy + ( y ) 2 = 0 y + y = 0 yy − ( y ) 3 = 0 y + ( y ) 2 = 2e−y

Hint: In Problem 35 the transformed equation is a Bernoulli equation. See Problem 23 in Section 2.4. In each of Problems 36 through 37, solve the given initial value problem using the methods of Problems 28 through 35.

36. y y = 2, y( 0) = 1, y ( 0) = 2 37. ( 1 + t 2 ) y + 2t y + 3t −2 = 0, y( 1) = 2, y ( 1) = −1

References The two books mentioned in Section 2.5 are Bailey, N. T. J., The Mathematical Theory of Infectious Diseases and Its Applications (2nd ed.) (New York: Hafner Press, 1975). Clark, Colin W., Mathematical Bioeconomics (2nd ed.) (New York: Wiley-Interscience, 1990).

A good introduction to population dynamics, in general, is Frauenthal, J. C., Introduction to Population Modeling (Boston: Birkhauser, 1980). A fuller discussion of the proof of the fundamental existence and uniqueness theorem can be found in many more advanced books on differential equations. Two that are reasonably accessible to elementary readers are

Boyce 9131 Ch02 2

102

September 29, 2016

17:16

102

CHAPTER 2 First-Order Differential Equations

Coddington, E. A., An Introduction to Ordinary Differential Equations (Englewood Cliffs, NJ: Prentice-Hall, 1961; New York: Dover, 1989). Brauer, F., and Nohel, J., The Qualitative Theory of Ordinary Differential Equations (New York: Benjamin, 1969; New York: Dover, 1989). A valuable compendium of methods for solving differential equations is Zwillinger, D., Handbook of Differential Equations (3rd ed.) (San Diego: Academic Press, 1998). For further discussion and examples of nonlinear phenomena, including bifurcations and chaos, see Strogatz, Steven H., Nonlinear Dynamics and Chaos (Reading, MA: Addison-Wesley, 1994).

A general reference on difference equations is Mickens, R. E., Difference Equations, Theory and Applications (2nd ed.) (New York: Van Nostrand Reinhold, 1990). Two papers by Robert May cited in the text are R. M. May, “Biological Populations with Nonoverlapping Generations: Stable Points, Stable Cycles, and Chaos,” Science 186 (1974), pp. 645--647; ''Biological Populations Obeying Difference Equations: Stable Points, Stable Cycles, and Chaos,'' Journal of Theoretical Biology 51 (1975), pp. 511--524. An elementary treatment of chaotic solutions of difference equations is Devaney, R. L., Chaos, Fractals, and Dynamics (Reading, MA: Addison-Wesley, 1990).

Boyce 9131 Ch03 2

September 29, 2016

17:28

103

CHAPTER 3 Second-Order Linear Differential Equations Linear differential equations of second order are of crucial importance in the study of differential equations for two main reasons. The first is that linear equations have a rich theoretical structure that underlies a number of systematic methods of solution. Further, a substantial portion of this structure and of these methods is understandable at a fairly elementary mathematical level. In order to present the key ideas in the simplest possible context, we describe them in this chapter for second-order equations. The second reason to study second-order linear differential equations is that they are vital to any serious investigation of the classical areas of mathematical physics. One cannot go very far in the development of fluid mechanics, heat conduction, wave motion, or electromagnetic phenomena without finding it necessary to solve second-order linear differential equations. We illustrate this at the end of this chapter with a discussion of the oscillations of some basic mechanical and electrical systems.

Homogeneous Differential Equations with Constant Coefficients 3.1

Many second-order ordinary differential equations have the form d2 y dy = f t, y, , dt dt 2

(1)

where f is some given function. Usually, we will denote the independent variable by t since time is often the independent variable in physical problems, but sometimes we will use x instead. We will use y, or occasionally some other letter, to designate the dependent variable. Equation (1) is said to be linear if the function f has the form dy dy f t, y, = g( t) − p( t) − q( t) y, (2) dt dt that is, if f is linear in y and dy/dt. In equation (2) g, p, and q are specified functions of the independent variable t but do not depend on y. In this case we usually rewrite equation (1) as y + p( t) y + q( t) y = g( t) ,

(3)

where the primes denote differentiation with respect to t. Instead of equation (3), we sometimes see the equation P( t) y + Q( t) y + R( t) y = G( t) .

(4)

Of course, if P( t) = 0, we can divide equation (4) by P( t) and thereby obtain equation (3) with p( t) =

Q( t) , P( t)

q( t) =

R( t) , P( t)

g( t) =

G( t) . P( t)

(5) 103

Boyce 9131 Ch03 2

104

September 29, 2016

17:28

104

CHAPTER 3 Second-Order Linear Differential Equations

In discussing equation (3) and in trying to solve it, we will restrict ourselves to intervals in which p, q, and g are continuous functions.1 If equation (1) is not of the form (3) or (4), then it is called nonlinear. Analytical investigations of nonlinear equations are relatively difficult, so we will have little to say about them in this book. Numerical or geometical approaches are often more appropriate, and these are discussed in Chapters 8 and 9. An initial value problem consists of a differential equation such as equations (1), (3), or (4) together with a pair of initial conditions y( t0 ) = y0 ,

y ( t0 ) = y0 ,

y0

(6)

where y0 and are given numbers prescribing values for y and y at the initial point t0 . Observe that the initial conditions for a second-order differential equation identify not only a particular point ( t0 , y0 ) through which the graph of the solution must pass, but also the slope y0 of the graph at that point. It is reasonable to expect that two initial conditions are needed for a second-order differential equation because, roughly speaking, two integrations are required to find a solution and each integration introduces an arbitrary constant. Presumably, two initial conditions will suffice to determine values for these two constants. A second-order linear differential equation is said to be homogeneous if the term g( t) in equation (3), or the term G( t) in equation (4), is zero for all t. Otherwise, the equation is called nonhomogeneous. Alternatively, the nonhomogeneous term g( t) , or G( t) , is sometimes called the forcing function because in many applications it describes an externally applied force. We begin our discussion with homogeneous equations, which we will write in the form P( t) y + Q( t) y + R( t) y = 0.

(7)

Later, in Sections 3.5 and 3.6, we will show that once the homogeneous equation has been solved, it is always possible to solve the corresponding nonhomogeneous equation (4), or at least to express the solution in terms of an integral. Thus the problem of solving the homogeneous equation is the more fundamental one. In this chapter we will concentrate our attention on equations in which the functions P, Q, and R are constants. In this case, equation (7) becomes ay + by + cy = 0,

(8)

where a, b, and c are given constants. It turns out that equation (8) can always be solved easily in terms of the elementary functions of calculus. On the other hand, it is usually much more difficult to solve equation (7) if the coefficients are not constants, and a treatment of that case is deferred until Chapter 5. Before taking up equation (8), let us first gain some experience by looking at a simple example that in many ways is typical.

EXAMPLE 1 Solve the equation y − y = 0.

(9)

Also find the solution that satisfies the initial conditions y( 0) = 2,

y ( 0) = −1.

(10)

Solution: Observe that equation (9) is just equation (8) with a = 1, b = 0, and c = −1. In words, equation (9) says that we seek a function with the property that the second derivative of the function is the same as the function itself. Do any of the functions that you studied in calculus have this property? A little thought will probably produce at least one such function, namely, y1 ( t) = et , the exponential function. A little more thought may also produce a second function, y2 ( t) = e−t . Some further experimentation reveals that constant multiples of these two solutions are also solutions.

▼ ......................................................................................................................................................................... 1 There is a corresponding treatment of higher-order linear equations in Chapter 4. If you wish, you may read the appropriate parts of Chapter 4 in parallel with Chapter 3.

Boyce 9131 Ch03 2

September 29, 2016

17:28

105

3.1 Homogeneous Differential Equations with Constant Coefficients

▼ For example, the functions 2et and 5e−t also satisfy equation (9), as you can verify by calculating their second derivatives. In the same way, the functions c1 y1 ( t) = c1 et and c2 y2 ( t) = c2 e−t satisfy the differential equation (9) for all values of the constants c1 and c2 . Next, it is vital to notice that the sum of any two solutions of equation (9) is also a solution. In particular, since c1 y1 ( t) and c2 y2 ( t) are solutions of equation (9) for any values of c1 and c2 , so is the function y = c1 y1 ( t) + c2 y2 ( t) = c1 et + c2 e−t .

(11)

Again, this can be verified by calculating the second derivative y from equation (11). We have y = c1 et − c2 e−t and y = c1 et + c2 e−t ; thus y is the same as y, and equation (9) is satisfied. Let us summarize what we have done so far in this example. Once we notice that the functions y1 ( t) = et and y2 ( t) = e−t are solutions of equation (9), it follows that the general linear combination (11) of these functions is also a solution. Since the coefficients c1 and c2 in equation (11) are arbitrary, this expression represents an infinite two-parameter family of solutions of the differential equation (9). We now turn to the task of picking out a particular member of this infinite family of solutions that also satisfies the given pair of initial conditions (10). In other words, we seek the solution that passes through the point (0, 2) and at that point has the slope −1. First, to ensure the solution passes through the point ( 0, 2) , we set t = 0 and y = 2 in equation (11); this gives the equation c1 + c2 = 2.

(12)

Next, we differentiate equation (11) with the result that y = c1 et − c2 e−t .

(13)

Then, to enforce the condition that the slope at ( 0, 2) is −1, we set t = 0 and y = −1 in equation (13); this yields the equation c1 − c2 = −1.

(14)

By solving equations (12) and (14) simultaneously for c1 and c2 , we find that c1 =

1 , 2

c2 =

3 . 2

Finally, inserting these values in equation (11), we obtain 1 3 y = et + e−t , 2 2

(15)

(16)

the solution of the initial value problem consisting of the differential equation (9) and the initial conditions (10).

What conclusions can we draw from the preceding example that will help us to deal with the more general equation (8), ay + by + cy = 0, whose coefficients a, b, and c are arbitrary (real) constants? In the first place, in the example the solutions were exponential functions. Further, once we had identified two solutions, we were able to use a linear combination of them to satisfy the given initial conditions as well as the differential equation itself. It turns out that by exploiting these two ideas, we can solve equation (8) for any values of its coefficients and also satisfy any given set of initial conditions for y and y . We start by seeking exponential solutions of the form y = er t , where r is a parameter to be determined. Then it follows that y = r er t and y = r 2 er t . By substituting these expressions for y, y , and y in equation (8), we obtain ( ar 2 + br + c) er t = 0. Since er t = 0, this condition is satisfied only when the other factor is zero: ar 2 + br + c = 0.

(17)

Equation (17) is called the characteristic equation for the differential equation (8). Its significance lies in the fact that if r is a root of the polynomial equation (17), then y = er t is a solution of the differential equation (8). Since equation (17) is a quadratic equation with real coefficients, it has two roots, which may be real and different, complex conjugates, or real but repeated. We consider the first case here and the latter two cases in Sections 3.3 and 3.4, respectively.

105

Boyce 9131 Ch03 2

106

September 29, 2016

17:28

106

CHAPTER 3 Second-Order Linear Differential Equations

Assuming that the roots of the characteristic equation (17) are real and different, let them be denoted by r1 and r2 , where r1 = r2 . Then y1 ( t) = er1 t and y2 ( t) = er2 t are two solutions of equation (8). Just as in Example 1, it now follows that y = c1 y1 ( t) + c2 y2 ( t) = c1 er1 t + c2 er2 t

(18)

is also a solution of equation (8). To verify that this is so, we can differentiate the expression in equation (18); hence y = c1r1 er1 t + c2r2 er2 t

(19)

y = c1r1 er1 t + c2r2 er2 t .

(20)

and 2

2

Substituting these expressions for y, y , and y in equation (8) and rearranging terms, we obtain 2 2 ay + by + cy = c1 ar1 + br1 + c er1 t + c2 ar2 + br2 + c er2 t . (21) 2

The fact that r1 is a root of equation (17) means that ar1 + br1 + c = 0. Since r2 is also a root 2 ar2

of the characteristic equation (17), it follows that + br2 + c = 0 as well. This completes the verification that y as given by equation (18) is indeed a solution of equation (8). Now suppose that we want to find the particular member of the family of solutions (18) that satisfies the initial conditions (6) y( t0 ) = y0 ,

y ( t0 ) = y0 .

By substituting t = t0 and y = y0 in equation (18), we obtain c1 er1 t0 + c2 er2 t0 = y0 .

(22)

Similarly, setting t = t0 and y = y0 in equation (19) gives c1r1 er1 t0 + c2r2 er2 t0 = y0 .

(23)

On solving equations (22) and (23) simultaneously for c1 and c2 , we find that c1 =

y0 − y0r2 r1 − r2

e−r1 t0 ,

c2 =

y0r1 − y0 r1 − r2

e−r2 t0 .

(24)

Since the roots of the characteristic equation (17) are assumed to be different, r1 −r2 = 0 so that the expressions in equation (24) always make sense. Thus, no matter what initial conditions are assigned---that is, regardless of the values of t0 , y0 , and y0 in equations (6)---it is always possible to determine c1 and c2 so that the initial conditions are satisfied. Moreover, there is only one possible choice of c1 and c2 for each set of initial conditions. With the values of c1 and c2 given by equation (24), the expression (18) is the solution of the initial value problem ay + by + cy = 0,

y( t0 ) = y0 , y ( t0 ) = y0 .

(25)

It is possible to show, on the basis of the fundamental theorem cited in the next section, that all solutions of equation (8) are included in the expression (18), at least for the case in which the roots of equation (17) are real and different. Therefore, we call equation (18) the general solution of equation (8). The fact that any possible initial conditions can be satisfied by the proper choice of the constants in equation (18) makes more plausible the idea that this expression does include all solutions of equation (8). Let us now look at some further examples.

EXAMPLE 2 Find the general solution of y + 5y + 6y = 0.

▼

(26)

Boyce 9131 Ch03 2

September 29, 2016

17:28

107

3.1 Homogeneous Differential Equations with Constant Coefficients

▼ Solution: We assume that y = er t , and it then follows that r must be a root of the characteristic equation r 2 + 5r + 6 = (r + 2) (r + 3) = 0. Thus the possible values of r are r1 = −2 and r2 = −3; the general solution of equation (26) is y = c1 e−2t + c2 e−3t .

(27)

EXAMPLE 3 Find the solution of the initial value problem y + 5y + 6y = 0,

y( 0) = 2, y ( 0) = 3.

(28)

Solution: The general solution of the differential equation was found in Example 2 and is given by equation (27). To satisfy the first initial condition, we set t = 0 and y = 2 in equation (27); thus c1 and c2 must satisfy c1 + c2 = 2.

(29)

To use the second initial condition, we must first differentiate equation (27). This gives y = −2c1 e−2t − 3c2 e−3t . Then, setting t = 0 and y = 3, we obtain −2c1 − 3c2 = 3.

(30)

By solving equations (29) and (30), we find that c1 = 9 and c2 = −7. Using these values in the expression (27), we obtain the solution y = 9e−2t − 7e−3t

(31)

of the initial value problem (28). The graph of the solution is shown in Figure 3.1.1. y

2 y = 9e–2 t – 7e–3t 1

0.5

1

1.5

t

2

FIGURE 3.1.1 Solution of the initial value problem (28): y + 5y + 6y = 0, y( 0) = 2, y ( 0) = 3.

EXAMPLE 4 Find the solution of the initial value problem 4y − 8y + 3y = 0,

y( 0) = 2, y ( 0) =

Solution: If y = er t , then we obtain the characteristic equation 4r 2 − 8r + 3 = 0

▼

1 . 2

(32)

107

Boyce 9131 Ch03 2

108

September 29, 2016

17:28

108

CHAPTER 3 Second-Order Linear Differential Equations

▼

whose roots are r =

1 3 and r = . Therefore, the general solution of the differential equation is 2 2 y = c1 e3t/2 + c2 et/2 .

(33)

Applying the initial conditions, we obtain the following two equations for c1 and c2 : c1 + c2 = 2,

3 1 1 c1 + c2 = . 2 2 2

1 5 The solution of these equations is c1 = − , c2 = , so the solution of the initial value problem (32) 2 2 is 1 5 y = − e3t/2 + et/2 . 2 2

(34)

Figure 3.1.2 shows the graph of the solution. y

2 y=–

1 3t/2 e 2

+

5 t/2 e 2

1

0.5

1

1.5

2

t

–1 FIGURE 3.1.2 Solution of the initial value problem (32): 4y − 8y + 3y = 0, y( 0) = 2, y ( 0) = 1/2.

EXAMPLE 5 The solution (31) of the initial value problem (28) initially increases (because its initial slope is positive), but eventually approaches zero (because both terms involve negative exponential functions). Therefore, the solution must have a maximum point, and the graph in Figure 3.1.1 confirms this. Determine the location of this maximum point. Solution: The coordinates of the maximum point can be estimated from the graph, but to find them more precisely, we seek the point where the solution has a horizontal tangent line. By differentiating the solution (31), y = 9e−2t − 7e−3t , with respect to t, we obtain y = −18e−2t + 21e−3t .

(35)

Setting y equal to zero and multiplying by e3t , we find that the critical value tm satisfies et = 7/6; hence tm = ln( 7/6) ∼ = 0.15415.

(36)

The corresponding maximum value ym is given by ym = 9e−2tm − 7e−3tm =

108 ∼ = 2.20408. 49

(37)

In this example the initial slope is 3, but the solution of the given differential equation behaves in a similar way for any other positive initial slope. In Problem 19 you are asked to determine how the coordinates of the maximum point depend on the initial slope.

Boyce 9131 Ch03 2

September 29, 2016

17:28

109

3.1 Homogeneous Differential Equations with Constant Coefficients

109

Returning to the equation ay + by + cy = 0 with arbitrary coefficients, recall that when r1 = r2 , its general solution (18) is the sum of two exponential functions. Therefore, the solution has a relatively simple geometrical behavior: as t increases, the magnitude of the solution either tends to zero (when both exponents are negative) or else exhibits unbounded growth (when at least one exponent is positive). These two cases are illustrated by the solutions of Examples 3 and 4, which are shown in Figures 3.1.1 and 3.1.2, respectively. Note that whether a growing solution approaches +∞ or −∞ as t → ∞ is determined by the sign of the coefficient of the exponential for the larger root of the characteristic equation. (See Problem 21.) There is also a third case that occurs less often: the solution approaches a constant when one exponent is zero and the other is negative. In Sections 3.3 and 3.4, respectively, we return to the problem of solving the equation ay +by +cy = 0 when the roots of the characteristic equation either are complex conjugates or are real and equal. In the meantime, in Section 3.2, we provide a systematic discussion of the mathematical structure of the solutions of all second-order linear homogeneous equations.

Problems In each of Problems 1 through 6, find the general solution of the given differential equation.

1. y + 2y − 3y = 0 2. y + 3y + 2y = 0

16. Solve the initial value problem y − y − 2y = 0, y( 0) = α , y ( 0) = 2. Then find α so that the solution approaches zero as t → ∞.

4. y + 5y = 0

In each of Problems 17 and 18, determine the values of α , if any, for which all solutions tend to zero as t → ∞; also determine the values of α , if any, for which all (nonzero) solutions become unbounded as t → ∞.

5. 4y − 9y = 0

17. y − ( 2α − 1) y + α ( α − 1) y = 0

6. y − 2y − 2y = 0

18. y + ( 3 − α ) y − 2( α − 1) y = 0

3. 6y − y − y = 0

In each of Problems 7 through 12, find the solution of the given initial value problem. Sketch the graph of the solution and describe its behavior as t increases. G G G G G G

7. y + y − 2y = 0,

y( 0) = 1, y ( 0) = 1

8. y + 4y + 3y = 0, 9. 10. 11. 12.

y + 3y = 0,

y + 8y − 9y = 0, 4y − y = 0,

y( 0) = 2, y ( 0) = −1

y( 0) = −2, y ( 0) = 3

2y + y − 4y = 0,

y( 0) = 0, y ( 0) = 1 y( 1) = 1, y ( 1) = 0

y( −2) = 1, y ( −2) = −1

13. Find a differential equation whose general solution is y = c1 e2t + c2 e−3t . G

14. Find the solution of the initial value problem y − y = 0,

y( 0) =

5 3 , y ( 0) = − . 4 4

Plot the solution for 0 ≤ t ≤ 2 and determine its minimum value.

15. Find the solution of the initial value problem 2y − 3y + y = 0,

y( 0) = 2, y ( 0) =

1 . 2

Then determine the maximum value of the solution and also find the point where the solution is zero.

19. Consider the initial value problem (see Example 5) y + 5y + 6y = 0,

y( 0) = 2, y ( 0) = β ,

where β > 0. a. Solve the initial value problem. b. Determine the coordinates tm and ym of the maximum point of the solution as functions of β . c. Determine the smallest value of β for which ym ≥ 4. d. Determine the behavior of tm and ym as β → ∞.

20. Consider the equation ay + by + cy = d, where a, b, c, and d are constants. a. Find all equilibrium, or constant, solutions of this differential equation. b. Let ye denote an equilibrium solution, and let Y = y − ye . Thus Y is the deviation of a solution y from an equilibrium solution. Find the differential equation satisfied by Y .

21. Consider the equation ay + by + cy = 0, where a, b, and c are constants with a > 0. Find conditions on a, b, and c such that the roots of the characteristic equation are: a. real, different, and negative. b. real with opposite signs. c. real, different, and positive. In each case, determine the behavior of the solution as t → ∞.

Boyce 9131 Ch03 2

110

September 29, 2016

17:28

110

CHAPTER 3 Second-Order Linear Differential Equations

Solutions of Linear Homogeneous Equations; the Wronskian 3.2

In the preceding section we showed how to solve some differential equations of the form ay + by + cy = 0, where a, b, and c are constants. Now we build on those results to provide a clearer picture of the structure of the solutions of all second-order linear homogeneous equations. In turn, this understanding will assist us in finding the solutions of other problems that we will encounter later. To discuss general properties of linear differential equations, it is helpful to introduce a differential operator notation. Let p and q be continuous functions on an open interval I ---that is, for α < t < β . The cases for α = −∞, or β = ∞, or both, are included. Then, for any function φ that is twice differentiable on I , we define the differential operator L by the equation L[φ ] = φ + pφ + qφ .

(1)

It is important to understand that the result of applying the operator L to a function φ is another function, which we refer to as L[φ ]. The value of L[φ ] at a point t is L[φ ]( t) = φ ( t) + p( t) φ ( t) + q( t) φ ( t) . For example, if p( t) = t 2 , q( t) = 1 + t, and φ ( t) = sin 3t, then L[φ ]( t) = ( sin 3t) + t 2 ( sin 3t) + ( 1 + t) sin 3t = −9 sin 3t + 3t 2 cos 3t + ( 1 + t) sin 3t. The operator L is often written as L = D 2 + p D + q, where D is the derivative operator, that is, D[φ ] = φ . In this section we study the second-order linear homogeneous differential equation L[φ ]( t) = 0. Since it is customary to use the symbol y to denote φ ( t) , we will usually write this equation in the form L[y] = y + p( t) y + q( t) y = 0.

(2)

With equation (2) we associate a set of initial conditions y( t0 ) = y0 ,

y ( t0 ) = y0 ,

(3)

where t0 is any point in the interval I , and y0 and y0 are given real numbers. We would like to know whether the initial value problem (2), (3) always has a solution, and whether it may have more than one solution. We would also like to know whether anything can be said about the form and structure of solutions that might be helpful in finding solutions of particular problems. Answers to these questions are contained in the theorems in this section. The fundamental theoretical result for initial value problems for second-order linear equations is stated in Theorem 3.2.1, which is analogous to Theorem 2.4.1 for first-order linear equations. The result applies equally well to nonhomogeneous equations, so the theorem is stated in that form.

Theorem 3.2.1 | (Existence and Uniqueness Theorem) Consider the initial value problem y + p( t) y + q( t) y = g( t) ,

y( t0 ) = y0 , y ( t0 ) = y0 ,

(4)

where p, q, and g are continuous on an open interval I that contains the point t0 . This problem has exactly one solution y = φ ( t) , and the solution exists throughout the interval I .

Boyce 9131 Ch03 2

September 29, 2016

17:28

111

3.2 Solutions of Linear Homogeneous Equations; the Wronskian

We emphasize that the theorem says three things: 1. The initial value problem has a solution; in other words, a solution exists. 2. The initial value problem has only one solution; that is, the solution is unique. 3. The solution φ is defined throughout the interval I where the coefficients are continuous and is at least twice differentiable there. For some problems some of these assertions are easy to prove. For instance, we found in Example 1 of Section 3.1 that the initial value problem y − y = 0,

y( 0) = 2, y ( 0) = −1

(5)

has the solution 1 t 3 −t (6) e + e . 2 2 The fact that we found a solution certainly establishes that a solution exists for this initial value problem. Further, the solution (6) is twice differentiable, indeed differentiable any number of times, throughout the interval ( −∞, ∞) where the coefficients in the differential equation are continuous. On the other hand, it is not obvious, and is more difficult to show, that the initial value problem (5) has no solutions other than the one given by equation (6). Nevertheless, Theorem 3.2.1 states that this solution is indeed the only solution of the initial value problem (5). For most problems of the form (4), it is not possible to write down a useful expression for the solution. This is a major difference between first-order and second-order linear differential equations. Therefore, all parts of the theorem must be proved by general methods that do not involve having such an expression. The proof of Theorem 3.2.1 is fairly difficult, and we do not discuss it here.2 We will, however, accept Theorem 3.2.1 as true and make use of it whenever necessary. y=

EXAMPLE 1 Find the longest interval in which the solution of the initial value problem ( t 2 − 3t) y + t y − ( t + 3) y = 0,

y( 1) = 2, y ( 1) = 1

is certain to exist. Solution: If the given differential equation is written in the form of equation (4), then 1 t +3 p( t) = , q( t) = − , and g( t) = 0. t −3 t ( t − 3) The only points of discontinuity of the coefficients are t = 0 and t = 3. Therefore, the longest open interval, containing the initial point t = 1, in which all the coefficients are continuous is 0 < t < 3. Thus this is the longest interval in which Theorem 3.2.1 guarantees that the solution exists.

EXAMPLE 2 Find the unique solution of the initial value problem y + p( t) y + q( t) y = 0,

y( t0 ) = 0, y ( t0 ) = 0,

where p and q are continuous in an open interval I containing t0 . Solution: The function y = φ ( t) = 0 for all t in I certainly satisfies the differential equation and initial conditions. By the uniqueness part of Theorem 3.2.1, it is the only solution of the given problem.

......................................................................................................................................................................... 2A

proof of Theorem 3.2.1 can be found, for example, in Chapter 6, Section 8 of the book by Coddington listed in the references at the end of this chapter.

111

Boyce 9131 Ch03 2

112

September 29, 2016

17:28

112

CHAPTER 3 Second-Order Linear Differential Equations

Let us now assume that y1 and y2 are two solutions of equation (2); in other words, L[y1 ] = y1 + py1 + qy1 = 0, and similarly for y2 . Then, just as in the examples in Section 3.1, we can generate more solutions by forming linear combinations of y1 and y2 . We state this result as a theorem.

Theorem 3.2.2 | (Principle of Superposition) If y1 and y2 are two solutions of the differential equation (2), L[y] = y + p( t) y + q( t) y = 0, then the linear combination c1 y1 + c2 y2 is also a solution for any values of the constants c1 and c2 .

A special case of Theorem 3.2.2 occurs if either c1 or c2 is zero. Then we conclude that any constant multiple of a solution of equation (2) is also a solution. To prove Theorem 3.2.2, we need only substitute y = c1 y1 ( t) + c2 y2 ( t)

(7)

for y in equation (2). By calculating the indicated derivatives and rearranging terms, we obtain L[c1 y1 + c2 y2 ] = [c1 y1 + c2 y2 ] + p[c1 y1 + c2 y2 ] + q[c1 y1 + c2 y2 ] = c1 y1 + c2 y2 + c1 py1 + c2 py2 + c1 qy1 + c2 qy2 = c1 [y1 + py1 + qy1 ] + c2 [y2 + py2 + qy2 ] = c1 L[y1 ] + c2 L[y2 ]. Since L[y1 ] = 0 and L[y2 ] = 0, it follows that L[c1 y1 + c2 y2 ] = 0 also. Therefore, regardless of the values of c1 and c2 , the function y as given by equation (7) satisfies the differential equation (2), and the proof of Theorem 3.2.2 is complete. Theorem 3.2.2 states that, beginning with only two solutions of equation (2), we can construct an infinite family of solutions by means of equation (7). The next question is whether all solutions of equation (2) are included in equation (7) or whether there may be other solutions of a different form. We begin to address this question by examining whether the constants c1 and c2 in equation (7) can be chosen so as to satisfy the initial conditions (3). These initial conditions require c1 and c2 to satisfy the equations c1 y1 ( t0 ) + c2 y2 ( t0 ) = y0 ,

(8)

c1 y1 ( t0 ) + c2 y2 ( t0 ) = y0 . The determinant of coefficients of the system (8) is y1 ( t0 ) y2 ( t0 ) W = = y1 ( t0 ) y2 ( t0 ) − y1 ( t0 ) y2 ( t0 ) . y1 ( t0 ) y2 ( t0 )

(9)

If W = 0, then equations (8) have a unique solution ( c1 , c2 ) regardless of the values of y0 and y0 . This solution is given by c1 =

y0 y2 ( t0 ) − y0 y2 ( t0 ) y1 ( t0 ) y2 ( t0 ) − y1 ( t0 ) y2 ( t0 )

or, in terms of determinants,

y0 y0

c1 = y1 ( t0 ) y1 ( t0 )

y2 ( t0 ) y2 ( t0 )

,

, y2 ( t0 ) y2 ( t0 )

c2 =

−y0 y1 ( t0 ) + y0 y1 ( t0 ) y1 ( t0 ) y2 ( t0 ) − y1 ( t0 ) y2 ( t0 ) y1 ( t0 ) y1 ( t0 )

c2 = y1 ( t0 ) y1 ( t0 )

y0 y0

. y2 ( t0 ) y2 ( t0 )

,

(10)

(11)

With these values for c1 and c2 , the linear combination y = c1 y1 ( t) + c2 y2 ( t) satisfies the initial conditions (3) as well as the differential equation (2). Note that the denominator in the expressions for c1 and c2 is the nonzero determinant W .

Boyce 9131 Ch03 2

September 29, 2016

17:28

113

3.2 Solutions of Linear Homogeneous Equations; the Wronskian

On the other hand, if W = 0, then the denominators appearing in equations (10) and (11) are zero. In this case equations (8) have no solution unless y0 and y0 have values that also make the numerators in equations (10) and (11) equal to zero. Thus, if W = 0, there are many initial conditions that cannot be satisfied no matter how c1 and c2 are chosen. The determinant W is called the Wronskian3 determinant, or simply the Wronskian, of the solutions y1 and y2 . Sometimes we use the more extended notation W [y1 , y2 ]( t0 ) to stand for the expression on the right-hand side of equation (9), thereby emphasizing that the Wronskian depends on the functions y1 and y2 , and that it is evaluated at the point t0 . The preceding argument establishes the following result.

Theorem 3.2.3 Suppose that y1 and y2 are two solutions of equation (2) L[y] = y + p( t) y + q( t) y = 0, and that the initial conditions (3) y( t0 ) = y0 ,

y ( t0 ) = y0

are assigned. Then it is always possible to choose the constants c1 , c2 so that y = c1 y1 ( t) + c2 y2 ( t) satisfies the differential equation (2) and the initial conditions (3) if and only if the Wronskian W [y1 , y2 ] = y1 y2 − y1 y2 is not zero at t0 .

EXAMPLE 3 In Example 2 of Section 3.1 we found that y1 ( t) = e−2t and y2 ( t) = e−3t are solutions of the differential equation y + 5y + 6y = 0. Find the Wronskian of y1 and y2 . Solution: The Wronskian of these two functions is

W e

−2t

,e

−3t

e−2t = −2e−2t

e−3t = −e−5t . −3e−3t

Since W is nonzero for all values of t, the functions y1 and y2 can be used to construct solutions of the given differential equation, together with initial conditions prescribed at any value of t. One such initial value problem was solved in Example 3 of Section 3.1.

The next theorem justifies the term “general solution” that we introduced in Section 3.1 for the linear combination c1 y1 + c2 y2 . ......................................................................................................................................................................... 3 Wronskian determinants are named for Jósef Maria Hoëné-Wronski (1776--1853), who was born in Poland but spent

most of his life in France. Wronski was a gifted but troubled man, and his life was marked by frequent heated disputes with other individuals and institutions.

113

Boyce 9131 Ch03 2

114

September 29, 2016

17:28

114

CHAPTER 3 Second-Order Linear Differential Equations

Theorem 3.2.4 Suppose that y1 and y2 are two solutions of the second-order linear differential equation (2), L[y] = y + p( t) y + q( t) y = 0. Then the two-parameter family of solutions y = c1 y1 ( t) + c2 y2 ( t) with arbitrary coefficients c1 and c2 includes every solution of equation (2) if and only if there is a point t0 where the Wronskian of y1 and y2 is not zero.

Let the function φ be any solution of equation (2). To prove the theorem, we must determine whether φ is included in the linear combinations c1 y1 + c2 y2 . That is, we must determine whether there are values of the constants c1 and c2 that make the linear combination the same as φ . Let t0 be a point where the Wronskian of y1 and y2 is nonzero. Then evaluate φ and φ at this point and call these values y0 and y0 , respectively; that is, y0 = φ ( t0 ) ,

y0 = φ ( t0 ) .

Next, consider the initial value problem y + p( t) y + q( t) y = 0,

y( t0 ) = y0 , y ( t0 ) = y0 .

(12)

The function φ is certainly a solution of this initial value problem. Further, because we are assuming that W [y1 , y2 ]( t0 ) is nonzero, it is possible (by Theorem 3.2.3) to choose c1 and c2 such that y = c1 y1 ( t) + c2 y2 ( t) is also a solution of the initial value problem (9). In fact, the proper values of c1 and c2 are given by equations (10) or (11). The uniqueness part of Theorem 3.2.1 guarantees that these two solutions of the same initial value problem are actually the same function; thus, for the proper choice of c1 and c2 , φ ( t) = c1 y1 ( t) + c2 y2 ( t) ,

(13)

and therefore φ is included in the family of functions c1 y1 + c2 y2 . Finally, since φ is an arbitrary solution of equation (2), it follows that every solution of this equation is included in this family. Now suppose that there is no point t0 where the Wronskian is nonzero. Thus W [y1 , y2 ]( t0 ) = 0 for every point t0 . Then (by Theorem 3.2.3) there are values of y0 and y0 such that no values of c1 and c2 satisfy the system (8). Select a pair of such values for y0 and y0 and choose the solution φ ( t) of equation (2) that satisfies the initial condition (3). Observe that this initial value problem is guaranteed to have a solution by Theorem 3.2.l. However, this solution is not included in the family y = c1 y1 +c2 y2 . Thus, in cases where W [y1 , y2 ]( t0 ) = 0 for every t0 , the linear combinations of y1 and y2 do not include all solutions of equation (2). This completes the proof of Theorem 3.2.4. Theorem 3.2.4 states that the Wronskian of y1 and y2 is not everywhere zero if and only if the linear combination c1 y1 + c2 y2 contains all solutions of equation (2). It is therefore natural (and we have already done this in the preceding section) to call the expression y = c1 y1 ( t) + c2 y2 ( t) with arbitrary constant coefficients the general solution of equation (2). The solutions y1 and y2 are said to form a fundamental set of solutions of equation (2) if and only if their Wronskian is nonzero. We can restate the result of Theorem 3.2.4 in slightly different language: to find the general solution, and therefore all solutions, of an equation of the form (2), we need only find two solutions of the given equation whose Wronskian is nonzero. We did precisely this in several examples in Section 3.1, although there we did not calculate the Wronskians. You should now go back and do that, thereby verifying that all the solutions we called “general solutions” in Section 3.1 do satisfy the necessary Wronskian condition. Now that you have a little experience verifying the nonzero Wronskian condition for the examples from Section 3.1, the following example handles all second-order linear differential equations whose characteristic polynomial has two distinct real roots.

Boyce 9131 Ch03 2

September 29, 2016

17:28

115

3.2 Solutions of Linear Homogeneous Equations; the Wronskian

EXAMPLE 4 r t

r t

Suppose that y1 ( t) = e 1 and y2 ( t) = e 2 are two solutions of an equation of the form (2). Show that if r1 = r2 , then y1 and y2 form a fundamental set of solutions of equation (2). Solution: We calculate the Wronskian of y1 and y2 :

e 1 W = r1 t r1 e

e 2 r t = (r2 − r1 ) exp[(r1 + r2 ) t]. r2 e 2

r t

r t

Since the exponential function is never zero, and since we are assuming that r2 − r1 = 0, it follows that W is nonzero for every value of t. Consequently, y1 and y2 form a fundamental set of solutions of equation (2).

EXAMPLE 5 Show that y1 ( t) = t 1/2 and y2 ( t) = t −1 form a fundamental set of solutions of 2t 2 y + 3t y − y = 0,

t > 0.

(14)

Solution: We will show how to solve equation (14) later (see Problem 25 in Section 3.3). However, at this stage we can verify by direct substitution that y1 and y2 are solutions of the differential equation 1 1 (14). Since y1 ( t) = t −1/2 and y1 ( t) = − t −3/2 , we have 2 4

2t

2

1 − t −3/2 4

+ 3t

1 −1/2 t 2

−t

1/2

=

1 3 − + − 1 t 1/2 = 0. 2 2

Similarly, y2 ( t) = −t −2 and y2 ( t) = 2t −3 , so

2t 2 2t −3 + 3t −t −2 − t −1 = ( 4 − 3 − 1) t −1 = 0. Next we calculate the Wronskian W of y1 and y2 :

t 1/2 t −1 = − 3 t −3/2 . W = 1 −1/2 −2 2 t −t 2

(15)

Since W = 0 for t > 0, we conclude that y1 and y2 form a fundamental set of solutions there. Thus the general solution of differential equation (14) is y( t) = c1 t 1/2 + c2 t −1 for t > 0.

In several cases we have been able to find a fundamental set of solutions, and therefore the general solution, of a given differential equation. However, this is often a difficult task, and the question arises as to whether a differential equation of the form (2) always has a fundamental set of solutions. The following theorem provides an affirmative answer to this question.

Theorem 3.2.5 Consider the differential equation (2), L[y] = y + p( t) y + q( t) y = 0, whose coefficients p and q are continuous on some open interval I . Choose some point t0 in I . Let y1 be the solution of equation (2) that also satisfies the initial conditions y( t0 ) = 1,

y ( t0 ) = 0,

and let y2 be the solution of equation (2) that satisfies the initial conditions y( t0 ) = 0,

y ( t0 ) = 1.

Then y1 and y2 form a fundamental set of solutions of equation (2).

115

Boyce 9131 Ch03 2

116

September 29, 2016

17:28

116

CHAPTER 3 Second-Order Linear Differential Equations

First observe that the existence of the functions y1 and y2 is ensured by the existence part of Theorem 3.2.1. To show that they form a fundamental set of solutions, we need only calculate their Wronskian at t0 : y1 ( t0 ) y2 ( t0 ) 1 0 = 1. W ( y1 , y2 ) ( t0 ) = = y1 ( t0 ) y2 ( t0 ) 0 1 Since their Wronskian is not zero at the point t0 , the functions y1 and y2 do form a fundamental set of solutions, thus completing the proof of Theorem 3.2.5. Note that the potentially difficult part of this proof, demonstrating the existence of a pair of solutions, is taken care of by reference to Theorem 3.2.1. Note also that Theorem 3.2.5 does not address the question of how to find the solutions y1 and y2 by solving the specified initial value problems. Nevertheless, it may be reassuring to know that a fundamental set of solutions always exists.

EXAMPLE 6 Find the fundamental set of solutions y1 and y2 specified by Theorem 3.2.5 for the differential equation y − y = 0,

(16)

using the initial point t0 = 0. Solution: In Section 3.1 we noted that two solutions of equation (16) are y1 ( t) = et and y2 ( t) = e−t . The Wronskian of these solutions is W [y1 , y2 ]( t) = −2 = 0, so they form a fundamental set of solutions. However, they are not the fundamental solutions indicated by Theorem 3.2.5 because they do not satisfy the initial conditions mentioned in that theorem at the point t = 0. To find the fundamental solutions specified by the theorem, we need to find the solutions satisfying the proper initial conditions. Let us denote by y3 ( t) the solution of equation (16) that satisfies the initial conditions y( 0) = 1,

y ( 0) = 0.

(17)

The general solution of equation (16) is y = c1 et + c2 e−t ,

(18)

and the initial conditions (17) are satisfied if c1 = 1/2 and c2 = 1/2. Thus y3 ( t) =

1 t 1 −t e + e = cosh t. 2 2

Similarly, if y4 ( t) satisfies the initial conditions y( 0) = 0,

y ( 0) = 1,

(19)

then y4 ( t) =

1 t 1 −t e − e = sinh t. 2 2

Since the Wronskian of y3 and y4 is W [y3 , y4 ]( t) = cosh2 t − sinh2 t = 1, these functions also form a fundamental set of solutions, as stated by Theorem 3.2.5. Therefore, the general solution of equation (16) can be written as y = k1 cosh t + k2 sinh t,

(20)

as well as in the form (18). We have used k1 and k2 for the arbitrary constants in equation (20) because they are not the same as the constants c1 and c2 in equation (18). One purpose of this example is to make it clear that a given differential equation has more than one fundamental set of solutions; indeed, it has infinitely many (see Problem 16). As a rule, you should choose the set that is most convenient.

Boyce 9131 Ch03 2

September 29, 2016

17:28

117

3.2 Solutions of Linear Homogeneous Equations; the Wronskian

In the next section we will encounter equations that have complex-valued solutions. The following theorem is fundamental in dealing with such equations and their solutions.

Theorem 3.2.6 Consider again the second-order linear differential equation (2), L[y] = y + p( t) y + q( t) y = 0, where p and q are continuous real-valued functions. If y = u( t) +iv( t) is a complex-valued solution of differential equation (2), then its real part u and its imaginary part v are also solutions of this equation.

To prove this theorem, we substitute u( t) + iv( t) for y in L[y], obtaining L[y]( t) = u ( t) + iv ( t) + p( t) ( u ( t) + iv ( t) ) + q( t) ( u( t) + iv( t) ) .

(21)

Then, by separating equation (21) into its real and imaginary parts---and this is where we need to know that p( t) and q( t) are real-valued---we find that L[y]( t) = ( u ( t) + p( t) u ( t) + q( t) u( t) ) + i( v ( t) + p( t) v ( t) + q( t) v( t) ) = L[u]( t) + i L[v]( t) . Recall that a complex number is zero if and only if its real and imaginary parts are both zero. We know that L[y] = 0 because y is a solution of equation (2). Therefore, both L[u] = 0 and L[v] = 0; consequently, the two real-valued functions u and v are also solutions of equation (2), so the theorem is established. We will see examples of the use of Theorem 3.2.6 in Section 3.3. Incidentally, the complex conjugate y of a solution y is also a solution. While this can be proved by an argument similar to the one just used to prove Theorem 3.2.6, it is also a consequence of Theorem 3.2.2 since y = u( t) −iv( t) is a linear combination of two solutions. Now let us examine further the properties of the Wronskian of two solutions of a secondorder linear homogeneous differential equation. The following theorem, perhaps surprisingly, gives a simple explicit formula for the Wronskian of any two solutions of any such equation, even if the solutions themselves are not known.

Theorem 3.2.7 | (Abel’s Theorem)4 If y1 and y2 are solutions of the second-order linear differential equation L[y] = y + p( t) y + q( t) y = 0,

(22)

where p and q are continuous on an open interval I , then the Wronskian W [y1 , y2 ]( t) is given by

W [y1 , y2 ]( t) = c exp −

p( t) dt ,

(23)

where c is a certain constant that depends on y1 and y2 , but not on t. Further, W [y1 , y2 ]( t) either is zero for all t in I (if c = 0) or else is never zero in I (if c = 0) .

To prove Abel’s theorem, we start by noting that y1 and y2 satisfy y1 + p( t) y1 + q( t) y1 = 0, y2 + p( t) y2 + q( t) y2 = 0.

(24)

If we multiply the first equation by −y2 , multiply the second by y1 , and add the resulting equations, we obtain ( y1 y2 − y1 y2 ) + p( t) ( y1 y2 − y1 y2 ) = 0. (25) ......................................................................................................................................................................... 4 The

result in Theorem 3.2.7 was derived by the Norwegian mathematician Niels Henrik Abel (1802--1829) in 1827 and is known as Abel’s formula. Abel also showed that there is no general formula for solving a quintic, or fifth degree, polynomial equation in terms of explicit algebraic operations on the coefficients, thereby resolving a question that had remained unanswered since the sixteenth century. His greatest contributions, however, were in analysis, particularly in the study of elliptic functions. Unfortunately, his work was not widely noticed until after his death. The distinguished French mathematician Legendre called it a “monument more lasting than bronze.”

117

Boyce 9131 Ch03 2

118

September 29, 2016

17:28

118

CHAPTER 3 Second-Order Linear Differential Equations

Next, we let W ( t) = W [y1 , y2 ]( t) and observe that W = y1 y2 − y1 y2 .

(26)

Then we can write equation (25) in the form W + p( t) W = 0.

(27)

Equation (27) can be solved immediately since it is both a first-order linear differential equation (Section 2.1) and a separable differential equation (Section 2.2). Thus W ( t) = c exp − p( t) dt , (28) where c is a constant. The value of c depends on which pair of solutions of equation (22) is involved. However, since the exponential function is never zero, W ( t) is not zero unless c = 0, in which case W ( t) is zero for all t. This completes the proof of Theorem 3.2.7. Note that the Wronskians of any two fundamental sets of solutions of the same differential equation can differ only by a multiplicative constant, and that the Wronskian of any fundamental set of solutions can be determined, up to a multiplicative constant, without solving the differential equation. Further, since under the conditions of Theorem 3.2.7 the Wronskian W is either always zero or never zero, you can determine which case actually occurs by evaluating W at any single convenient value of t.

EXAMPLE 7 In Example 5 we verified that y1 ( t) = t 1/2 and y2 ( t) = t −1 are solutions of the equation 2t 2 y + 3t y − y = 0,

t > 0.

(29)

Verify that the Wronskian of y1 and y2 is given by Abel’s formula (23). Solution: 3 From the example just cited we know that W [y1 , y2 ]( t) = − t −3/2 . To use equation (23), we must 2 write the differential equation (29) in the standard form with the coefficient of y equal to 1. Thus we obtain 3 1 y + y − 2 y = 0, 2t 2t 3 so p( t) = . Hence 2t

W [y1 , y2 ]( t) = c exp −

3 dt 2t

3 = c exp − ln t 2

= c t −3/2 .

(30)

Equation (30) gives the Wronskian of any pair of solutions of equation (29). For the particular 3 solutions given in this example, we must choose c = − . 2

Summary. We can summarize the discussion in this section as follows: to find the general solution of the differential equation y + p( t) y + q( t) y = 0,

α < t < β,

we must first find two functions y1 and y2 that satisfy the differential equation in α < t < β . Then we must make sure that there is a point in the interval where the Wronskian W of y1 and y2 is nonzero. Under these circumstances y1 and y2 form a fundamental set of solutions, and the general solution is y = c1 y1 ( t) + c2 y2 ( t) , where c1 and c2 are arbitrary constants. If initial conditions are prescribed at a given point in α < t < β , then c1 and c2 can be chosen so as to satisfy these conditions.

Boyce 9131 Ch03 2

September 29, 2016

17:28

119

3.2 Solutions of Linear Homogeneous Equations; the Wronskian

119

Problems In each of Problems 1 through 5, find the Wronskian of the given pair of functions.

1. 2. 3. 4. 5.

e2t , e−3t/2

set of solutions. b. Let y3 ( t) = −2e2t , y4 ( t) = y1 ( t) + 2y2 ( t) , and y5 ( t) = 2y1 ( t) − 2y3 ( t) . Are y3 ( t) , y4 ( t) , and y5 ( t) also solutions of the given differential equation? c. Determine whether each of the following

pairs forms a fundamental set of solutions: y1 ( t) , y3 ( t) ; y2 ( t) , y3 ( t) ;

cos t, sin t e−2t , te−2t et sin t, et cos t cos2 θ , 1 + cos( 2θ )

In each of Problems 6 through 9, determine the longest interval in which the given initial value problem is certain to have a unique twicedifferentiable solution. Do not attempt to find the solution.

6. t y + 3y = t,

22. Consider the equation y − y − 2y = 0. a. Show that y1 ( t) = e−t and y2 ( t) = e2t form a fundamental

y( 1) = 1, y ( 1) = 2

7. t ( t − 4) y + 3t y + 4y = 2,

y( 3) = 0, y ( 3) = −1

8. y + ( cos t) y + 3( ln |t|) y = 0,

y( 2) = 3, y ( 2) = 1

9. ( x − 2) y + y + ( x − 2) ( tan x) y = 0,

y( 3) = 1, y ( 3) = 2

10. Verify that y1 ( t) = t and y2 ( t) = t 2

−1

y1 ( t) , y4 ( t) ;

y4 ( t) , y5 ( t) .

In each of Problems 23 through 25, find the Wronskian of two solutions of the given differential equation without solving the equation.

23. t 2 y − t ( t + 2) y + ( t + 2) y = 0 24. ( cos t) y + ( sin t) y − t y = 0 25. ( 1 − x 2 ) y − 2x y + α ( α + 1) y = 0,

Legendre’s equation

are two solutions of the differential equation t 2 y − 2y = 0 for t > 0. Then show that y = c1 t 2 + c2 t −1 is also a solution of this equation for any c1 and c2 .

26. Show that if p is differentiable and p( t) > 0, then the

11. Verify that y1 ( t) = 1 and y2 ( t) = t 1/2 are solutions of the

a fundamental set of solutions and if W [y1 , y2 ]( 1) = 2, find the value of W [y1 , y2 ]( 5) .

differential equation yy + ( y ) 2 = 0 for t > 0. Then show that y = c1 + c2 t 1/2 is not, in general, a solution of this equation. Explain why this result does not contradict Theorem 3.2.2.

12. Show that if y = φ ( t) is a solution of the differential equation

y + p( t) y + q( t) y = g( t) , where g( t) is not always zero, then y = cφ ( t) , where c is any constant other than 1, is not a solution. Explain why this result does not contradict the remark following Theorem 3.2.2. 13. Can y = sin( t 2 ) be a solution on an interval containing t = 0 of an equation y + p( t) y + q( t) y = 0 with continuous coefficients? Explain your answer.

14. If the Wronskian W of f and g is 3e4t , and if f ( t) = e2t , find g( t) .

15. If the Wronskian of f and g is t cos t − sin t, and if u = f + 3g, v = f − g, find the Wronskian of u and v.

16. Assume that y1 and y2 are a fundamental set of solutions of y + p( t) y + q( t) y = 0 and let y3 = a1 y1 + a2 y2 and y4 = b1 y1 + b2 y2 , where a1 , a2 , b1 , and b2 are any constants. Show that W [y3 , y4 ] = ( a1 b2 − a2 b1 ) W [y1 , y2 ]. Are y3 and y4 also a fundamental set of solutions? Why or why not? In each of Problems 17 and 18, find the fundamental set of solutions specified by Theorem 3.2.5 for the given differential equation and initial point.

17. y + y − 2y = 0, t0 = 0 18. y + 4y + 3y = 0, t0 = 1 In each of Problems 19 through 21, verify that the functions y1 and y2 are solutions of the given differential equation. Do they constitute a fundamental set of solutions?

19. y + 4y = 0; y1 ( t) = cos( 2t) , y2 ( t) = sin( 2t) 20. y − 2y + y = 0; y1 ( t) = et , y2 ( t) = tet 21. x 2 y − x( x + 2) y + ( x + 2) y = 0, x > 0; y1 ( x) = x, y2 ( x) = xe x

Wronskian W ( t) of two solutions of [ p( t) y ] + q( t) y = 0 is W ( t) = c/ p( t) , where c is a constant.

27. If the differential equation t y + 2y + tet y = 0 has y1 and y2 as 28. If the Wronskian of any two solutions of y + p( t) y +q( t) y = 0 is constant, what does this imply about the coefficients p and q?

In Problems 29 and 30, assume that p and q are continuous and that the functions y1 and y2 are solutions of the differential equation y + p( t) y + q( t) y = 0 on an open interval I .

29. Prove that if y1 and y2 are zero at the same point in I , then they cannot be a fundamental set of solutions on that interval.

30. Prove that if y1 and y2 have a common point of inflection t0 in I , then they cannot be a fundamental set of solutions on I unless both p and q are zero at t0 .

31. Exact Equations. The equation P( x) y + Q( x) y + R( x) y = 0 is said to be exact if it can be written in the form

( P( x) y ) + ( f ( x) y) = 0, where f ( x) is to be determined in terms of P( x) , Q( x) , and R( x) . The latter equation can be integrated once immediately, resulting in a first-order linear equation for y that can be solved as in Section 2.1. By equating the coefficients of the preceding equations and then eliminating f ( x) , show that a necessary condition for exactness is P ( x) − Q ( x) + R( x) = 0. It can be shown that this is also a sufficient condition. In each of Problems 32 through 34, use the result of Problem 31 to determine whether the given equation is exact. If it is, then solve the equation.

32. y + x y + y = 0 33. x y − ( cos x) y + ( sin x) y = 0, x > 0 34. x 2 y + x y − y = 0, x > 0

Boyce 9131 Ch03 2

120

September 29, 2016

17:28

120

CHAPTER 3 Second-Order Linear Differential Equations

35. The Adjoint Equation. If a second-order linear homogeneous equation is not exact, it can be made exact by multiplying by an appropriate integrating factor μ ( x) . Thus we require that μ ( x) be such that μ ( x) P( x) y + μ ( x) Q( x) y + μ ( x) R( x) y = 0 can be written in the form ( μ ( x) P( x) y ) + ( f ( x) y) = 0. By equating coefficients in these two equations and eliminating f ( x) , show that the function μ must satisfy Pμ + ( 2P − Q) μ + ( P − Q + R) μ = 0. This equation is known as the adjoint of the original equation and is important in the advanced theory of differential equations. In general,

the problem of solving the adjoint differential equation is as difficult as that of solving the original equation, so only occasionally is it possible to find an integrating factor for a second-order equation. In each of Problems 36 and 37, use the result of Problem 35 to find the adjoint of the given differential equation.

36. x 2 y + x y + ( x 2 − ν 2 ) y = 0, Bessel’s equation 37. y − x y = 0, Airy’s equation 38. A second-order linear equation P( x) y + Q( x) y + R( x) y = 0 is said to be self-adjoint if its adjoint is the same as the original equation. Show that a necessary condition for this equation to be self-adjoint is that P ( x) = Q( x) . Determine whether each of the equations in Problems 36 and 37 is self-adjoint.

Complex Roots of the Characteristic Equation 3.3

We continue our discussion of the second-order linear differential equation ay + by + cy = 0,

(1)

where a, b, and c are given real numbers. In Section 3.1 we found that if we seek solutions of the form y = er t , then r must be a root of the characteristic equation ar 2 + br + c = 0.

(2)

We showed in Section 3.1 that if the roots r1 and r2 are real and different, which occurs whenever the discriminant b2 − 4ac is positive, then the general solution of equation (1) is y = c 1 e r 1 t + c2 e r 2 t .

(3)

Suppose now that b2 − 4ac is negative. Then the roots of equation (2) are conjugate complex numbers; we denote them by r1 = λ + iμ ,

r2 = λ − iμ ,

(4)

where λ and μ are real. The corresponding expressions for y are y1 ( t) = exp( ( λ + iμ ) t) ,

y2 ( t) = exp( ( λ − iμ ) t) .

(5)

Our first task is to explore what is meant by these expressions, which involve evaluating the exponential function for a complex exponent. For example, if λ = −1, μ = 2, and t = 3, then from equation (5), y1 ( 3) = e−3+6i .

(6)

What does it mean to raise the number e to a complex power? The answer is provided by an important relation known as Euler’s formula. Euler’s Formula. To assign a meaning to the expressions in equations (5), we need to give a definition of the complex exponential function. Of course, we want the definition to reduce to the familiar real exponential function when the exponent is real. There are several ways to discover how this extension of the exponential function should be defined. Here we use a method based on infinite series; an alternative is outlined in Problem 20. Recall from calculus that the Taylor series for et about t = 0 is ∞

et = 1 + t +

tn t2 tn + ··· + + ··· = , 2 n! n! n=0

−∞ < t < ∞.

(7)

Boyce 9131 Ch03 2

September 29, 2016

17:28

121

3.3 Complex Roots of the Characteristic Equation

If we now assume that we can substitute it for t in equation (7), then we have ∞ ( it) n eit = . n!

(8)

n=0

To simplify this series, we write ( it) n = i n t n and make use of the facts that i 2 = −1, i 3 = −i, i 4 = 1, and so forth. When n is even, there is an integer k with n = 2k; in this case i n = i 2k = ( −1) k . And when n is odd, n = 2k + 1, so i n = i 2k+1 = i( −1) k . This suggests separating the terms in the right-hand side of (8) into its real and imaginary parts. The result is5 ∞ ∞ ( −1) k t 2k ( −1) k t 2k+1 it e = (9) +i . ( 2k) ! ( 2k + 1) ! k=0

k=0

The first series in equation (9) is precisely the Taylor series for cos t about t = 0, and the second is the Taylor series for sin t about t = 0. Thus we have eit = cos t + i sin t.

(10)

Equation (10) is known as Euler’s formula and is an extremely important mathematical relationship. Although our derivation of equation (10) is based on the unverified assumption that the series (7) can be used for complex as well as real values of the independent variable, our intention is to use this derivation only to make equation (10) seem plausible. We now put matters on a firm foundation by adopting equation (10) as the definition of eit . In other words, whenever we write eit , we mean the expression on the right-hand side of equation (10). There are some variations of Euler’s formula that are also worth noting. If we replace t by −t in equation (10) and recall that cos( −t) = cos t and sin( −t) = − sin t, then we have e−it = cos t − i sin t.

(11)

Further, if t is replaced by μ t in equation (10), then we obtain a generalized version of Euler’s formula, namely, eiμ t = cos( μ t) + i sin( μ t) . (12) Next, we want to extend the definition of the exponential function to arbitrary complex exponents of the form ( λ+iμ ) t. Since we want the usual properties of the exponential function to hold for complex exponents, we certainly want exp( ( λ + iμ ) t) to satisfy e( λ+iμ ) t = eλt eiμ t .

(13)

Then, substituting for eiμ t from equation (12), we obtain e( λ+iμ ) t = eλt ( cos( μ t) + i sin( μ t) ) = eλt cos( μ t) + ieλt sin( μ t) .

(14)

We now take equation (14) as the definition of exp[( λ + iμ ) t]. The value of the exponential function with a complex exponent is a complex number whose real and imaginary parts are given by the terms on the right-hand side of equation (14). Observe that the real and imaginary parts of exp( ( λ + iμ ) t) are expressed entirely in terms of elementary real-valued functions. For example, the quantity in equation (6) has the value e−3+6i = e−3 cos 6 + ie−3 sin 6 ∼ = 0.0478041 − 0.0139113i. With the definitions (10) and (14), it is straightforward to show that the usual laws of exponents are valid for the complex exponential function. You can also use equation (14) to verify that the differentiation formula d rt ( e ) = r er t dt

(15)

holds for complex values of r . ......................................................................................................................................................................... 5 Recall

from calculus that the reordering of terms in the right-hand side of equation (9) is allowed because the series converges absolutely for all −∞ < t < ∞.

121

Boyce 9131 Ch03 2

122

September 29, 2016

17:28

122

CHAPTER 3 Second-Order Linear Differential Equations

EXAMPLE 1 Find the general solution of the differential equation y + y + 9.25y = 0.

(16)

Also find the solution that satisfies the initial conditions y( 0) = 2,

y ( 0) = 8,

(17)

and draw its graph for 0 < t < 10. Solution: The characteristic equation for equation (16) is r 2 + r + 9.25 = 0 so its roots are 1 r1 = − + 3i, 2

1 r2 = − − 3i. 2

Therefore, two solutions of equation (16) are

y1 ( t) = exp and

y2 ( t) = exp

1 − + 3i t 2

= e−t/2 ( cos( 3t) + i sin( 3t) )

(18)

= e−t/2 ( cos( 3t) − i sin( 3t) ) .

(19)

1 − − 3i t 2

You can verify that the Wronskian W [y1 , y2 ]( t) = −6ie−t , which is not zero, so the general solution of equation (15) can be expressed as a linear combination of y1 ( t) and y2 ( t) with arbitrary coefficients. However, the initial value problem (16), (17) has only real coefficients, and it is often desirable to express the solution of such a problem in terms of real-valued functions. To do this we can make use of Theorem 3.2.6, which states that the real and imaginary parts of a complex-valued solution of equation (16) are also solutions of the same differential equation. Thus, starting from y1 ( t) , we obtain u( t) = e−t/2 cos( 3t) ,

v( t) = e−t/2 sin( 3t)

(20)

as real-valued solutions6 of equation (16). On calculating the Wronskian of u( t) and v( t) , we find that W [u, v]( t) = 3e−t , which is not zero; thus u( t) and v( t) form a fundamental set of solutions, and the general solution of equation (16) can be written as

y = c1 u( t) + c2 v( t) = e−t/2 c1 cos( 3t) + c2 sin( 3t) ,

(21)

where c1 and c2 are arbitrary constants. To satisfy the initial conditions (17), we first substitute t = 0 and y = 2 in the solution (20) with the result that c1 = 2. Then, by differentiating equation (21), setting t = 0, and setting y = 8, 1 we obtain − c1 + 3c2 = 8 so that c2 = 3. Thus the solution of the initial value problem (16), (17) is 2 y = e−t/2 ( 2 cos( 3t) + 3 sin( 3t) ) .

(22)

The graph of this solution is shown in Figure 3.3.1. From the graph we see that the solution of this problem oscillates, with period 2π/3 and a decaying amplitude. The sine and cosine factors control the oscillatory nature of the solution, and the negative exponential factor in each term causes the magnitude of the oscillations to decrease toward zero as time increases. ..................................................................................................................................................................................... 6 If

you are not completely sure that u( t) and v( t) are solutions of the given differential equation, you should substitute these functions into equation (16) and confirm that they satisfy it. (See Problem 23.)

▼

Boyce 9131 Ch03 2

September 29, 2016

17:28

123

3.3 Complex Roots of the Characteristic Equation

▼

y 3

2

1

2

4

6

8

10 t

–1

FIGURE 3.3.1 Solution of the initial value problem (16), (17): y + y + 9.25y = 0, y( 0) = 2, y ( 0) = 8.

Complex Roots; The General Case. The functions y1 ( t) and y2 ( t) , given by equations (5) and with the meaning expressed by equation (14), are solutions of equation (1) when the roots of the characteristic equation (2) are complex numbers λ ± iμ . However, the solutions y1 and y2 are complex-valued functions, whereas in general we would prefer to have real-valued solutions because the differential equation itself has real coefficients. Just as in Example 1, we can use Theorem 3.2.6 to find a fundamental set of real-valued solutions by choosing the real and imaginary parts of either y1 ( t) or y2 ( t) . In this way we obtain the solutions u( t) = eλt cos( μ t) ,

v( t) = eλt sin( μ t) .

(23)

By direct computation (see Problem 19), you can show that the Wronskian of u and v is W [u, v]( t) = μ e2λt .

(24)

Thus, as long as μ = 0, the Wronskian W is not zero, so u and v form a fundamental set of solutions. (Of course, if μ = 0, then the roots are real and equal and the discussions in this section, and in Section 3.1, are not applicable.) Consequently, if the roots of the characteristic equation are complex numbers λ ± iμ , with μ = 0, then the general solution of equation (1) is y = c1 eλt cos( μ t) + c2 eλt sin( μ t) ,

(25)

where c1 and c2 are arbitrary constants. Note that the solution (25) can be written down as soon as the values of λ and μ are known. Let us now consider some further examples.

EXAMPLE 2 Find the solution of the initial value problem 16y − 8y + 145y = 0,

▼

y( 0) = −2, y ( 0) = 1.

(26)

123

Boyce 9131 Ch03 2

124

September 29, 2016

17:28

124

CHAPTER 3 Second-Order Linear Differential Equations

▼ Solution: The characteristic equation is 16r 2 − 8r + 145 = 0 and its roots are r =

1 ± 3i. Thus the general 4

solution of the differential equation is y( t) = c1 et/4 cos( 3t) + c2 et/4 sin( 3t) .

(27)

To apply the first initial condition, we set t = 0 in equation (27); this gives y( 0) = c1 = −2. For the second initial condition, we must differentiate equation (27) before substituting t = 0. In this way we find that y ( 0) = from which we determine that c2 =

1 c1 + 3c2 = 1, 4

1 . Using these values of c1 and c2 in the general solution (27), 2

we obtain 1 y = −2et/4 cos( 3t) + et/4 sin( 3t) 2

(28)

as the solution of the initial value problem (26). The graph of this solution is shown in Figure 3.3.2. In this case we observe that the solution is a growing oscillation. Again the trigonometric factors in equation (28) determine the oscillatory part of the solution (again with period 2π/3), while the exponential factor (with a positive exponent this time) causes the magnitude of the oscillation to increase with time. y 10

y = –2e t/4 cos (3t) +

1 t/4 e 2

sin (3t)

5

2

4

6

8 t

–5

–10 FIGURE 3.3.2 Solution of the initial value problem (26): 16y − 8y + 145y = 0, y( 0) = −2, y ( 0) = 1.

EXAMPLE 3 Find the general solution of y + 9y = 0.

(29)

Solution: The characteristic equation is r 2 + 9 = 0 with the roots r = ±3i; thus λ = 0 and μ = 3. The general solution is y = c1 cos( 3t) + c2 sin( 3t) .

▼

(30)

Boyce 9131 Ch03 2

September 29, 2016

17:28

125

3.3 Complex Roots of the Characteristic Equation

▼

125

y 3

2

1

t

0 1

2

3

4

5

6

7

8

–1

–2

–3 FIGURE 3.3.3 Solutions of equation (29): y + 9y = 0, with two sets of

initial conditions: y( 0) = 1, y ( 0) = 2 (dashed, green) and y( 0) = 2, y ( 0) = 8 (solid, blue). Both solutions have the same period, but different amplitudes and phase shifts.

Note that if the real part of the roots is zero, as in this example, then there is no exponential factor in the solution. Figure 3.3.3 shows the graph of two solutions of equation (28) with different initial conditions. In each case the solution is a pure oscillation with period 2π/3 but whose amplitude and phase shift are determined by the initial conditions. Since there is no exponential factor in the solution (30), the amplitude of each oscillation remains constant in time.

Problems In each of Problems 1 through 4, use Euler’s formula to write the given expression in the form a + ib.

1. 2. 3. 4.

exp( 2 − 3i)

21−i

G

N

y − 2y + 6y = 0 y + 2y + 2y = 0 y + 2y + 1.25y = 0 9y + 9y − 4y = 0 y + 4y + 6.25y = 0

12. y + 4y = 0, y( 0) = 0, y ( 0) = 1 13. y − 2y + 5y = 0, y( π/2) = 0, y ( π/2) = 2

N

u( 0) = 2, u ( 0) = 0.

a. Find the solution u( t) of this problem. b. For t > 0, find the first time at which |u( t) | = 10. 17. Consider the initial value problem 5u + 2u + 7u = 0,

y − 2y + 2y = 0

y + 6y + 13y = 0

14. y + y = 0, y( π/3) = 2, y ( π/3) = −4 15. y + 2y + 2y = 0, y( π/4) = 2, y ( π/4) = −2 16. Consider the initial value problem 3u − u + 2u = 0,

e2−( π/2) i

In each of Problems 12 through 15, find the solution of the given initial value problem. Sketch the graph of the solution and describe its behavior for increasing t. G

G N

eiπ

In each of Problems 5 through 11, find the general solution of the given differential equation.

5. 6. 7. 8. 9. 10. 11.

G

u( 0) = 2, u ( 0) = 1.

a. Find the solution u( t) of this problem. b. Find the smallest T such that |u( t) | ≤ 0.1 for all t > T . 18. Consider the initial value problem y + 2y + 6y = 0,

y( 0) = 2, y ( 0) = α ≥ 0.

a. Find the solution y( t) of this problem. b. Find α such that y = 0 when t = 1. c. Find, as a function of α , the smallest positive value of t for which y = 0.

d. Determine the limit of the expression found in part c as α → ∞.

Boyce 9131 Ch03 2

126

September 29, 2016

17:28

126

CHAPTER 3 Second-Order Linear Differential Equations

19. Show that W eλt cos( μ t) , eλt sin( μ t) = μ e2λt . 20. In this problem we outline a different derivation of Euler’s

b. Use the results of part a to transform equation (33) into d2 y dy + ( α − 1) + β y = 0. 2 d x dx

formula.

a. Show that y1 ( t) = cos t and y2 ( t) = sin t are a fundamental

Observe that differential equation (34) has constant coefficients. If y1 ( x) and y2 ( x) form a fundamental set of solutions of equation (34), then y1 ( ln t) and y2 ( ln t) form a fundamental set of solutions of equation (33).

set of solutions of y + y = 0; that is, show that they are solutions and that their Wronskian is not zero.

b. Show (formally) that y = eit is also a solution of y + y = 0. Therefore, eit = c1 cos t + c2 sin t

(31)

for some constants c1 and c2 . Why is this so? c. Set t = 0 in equation (31) to show that c1 = 1. d. Assuming that equation (15) is true, differentiate equation (31) and then set t = 0 to conclude that c2 = i. Use the values of c1 and c2 in equation (31) to arrive at Euler’s formula.

21. Using Euler’s formula, show that eit + e−it = cos t, 2

eit − e−it = sin t. 2i

22. If er t is given by equation (14), show that e(r1 +r2 ) t = er1 t er2 t for any complex numbers r1 and r2 .

23. Consider the differential equation ay + by + cy = 0, where b2 − 4ac < 0 and the characteristic equation has complex roots λ ± iμ . Substitute the functions u( t) = eλt cos( μ t) and v( t) = eλt sin( μ t) for y in the differential equation and thereby confirm that they are solutions.

24. If the functions y1 and y2 are a fundamental set of solutions of

y + p( t) y + q( t) y = 0, show that between consecutive zeros of y1 there is one and only one zero of y2 . Note that this result is illustrated by the solutions y1 ( t) = cos t and y2 ( t) = sin t of the equation y + y = 0. Hint: Suppose that t1 and t2 are two zeros of y1 between which there are no zeros of y2 . Apply Rolle’s theorem to y1 / y2 to reach a contradiction. Change of Variables. Sometimes a differential equation with variable coefficients, y + p( t) y + q( t) y = 0,

dy d2 y + αt + β y = 0, dt dt 2

t > 0,

26. 27. 28. 29. 30. 31. 32.

t 2 y + t y + y = 0 t 2 y + 4t y + 2y = 0 t 2 y − 4t y − 6y = 0 t 2 y − 4t y + 6y = 0 t 2 y + 3t y − 3y = 0 t 2 y + 7t y + 10y = 0

In this problem we determine conditions on p and q that enable equation (32) to be transformed into an equation with constant coefficients by a change of the independent variable. Let x = u( t) be the new independent variable, with the relation between x and t to be specified later. a. Show that d2 y = dt 2

dy d x dy = , dt dt d x

(33)

where α and β are real constants, is called an Euler equation. a. Let x = ln t and calculate dy/dt and d 2 y/dt 2 in terms of dy/d x and d 2 y/d x 2 .

dx dt

2

d2 y d 2 x dy . + 2 2 dx dt d x

b. Show that the differential equation (32) becomes 2 2 2 dx dt

d y + dx2

dx d x + p( t) dt dt 2

dy + q( t) y = 0. dx

(35)

c. In order for equation (35) to have constant coefficients, the coefficients of d 2 y/d x 2 , dy/d x, and y must all be proportional. If q( t) > 0, then we can choose the constant of proportionality to be 1; hence, after integrating with respect to t,

x = u( t) =

( q( t) ) 1/2 dt .

(36)

d. With x chosen as in part c, show that the coefficient of dy/d x in equation (35) is also a constant, provided that the expression q ( t) + 2 p( t) q( t) 2( q( t) ) 3/2

(37)

is a constant. Thus equation (32) can be transformed into an equation with constant coefficients by a change of the independent variable, provided that the function ( q +2 pq) /q 3/2 is a constant. e. How must the analysis and results in d be modified if q( t) < 0? In each of Problems 33 through 36, try to transform the given equation into one with constant coefficients by the method of Problem 32. If this is possible, find the general solution of the given equation. 2

25. Euler Equations. An equation of the form t2

In each of Problems 26 through 31, use the method of Problem 25 to solve the given equation for t > 0.

(32)

can be put in a more suitable form for finding a solution by making a change of the independent variable. We explore these ideas in Problems 25 through 36. In particular, in Problem 25 we show that a class of equations known as Euler equations can be transformed into equations with constant coefficients by a simple change of the independent variable. Problems 26 through 31 are examples of this type of equation. Problem 32 determines conditions under which the more general equation (32) can be transformed into a differential equation with constant coefficients. Problems 33 through 36 give specific applications of this procedure.

(34)

33. y + t y + e−t y = 0, −∞ < t < ∞ 34. y + 3t y + t 2 y = 0, −∞ < t < ∞ 35. t y + ( t 2 − 1) y + t 3 y = 0, 0 < t < ∞ 2

36. y + t y − e−t y = 0

Boyce 9131 Ch03 2

September 29, 2016

17:28

127

3.4 Repeated Roots; Reduction of Order

3.4

Repeated Roots; Reduction of Order

In Sections 3.1 and 3.3 we showed how to solve the equation ay + by + cy = 0

(1)

when the roots of the characteristic equation ar 2 + br + c = 0

(2)

either are real and different or are complex conjugates. Now we consider the third possibility, namely, that the two roots r1 and r2 are equal. This case is transitional between the other two and occurs when the discriminant b2 − 4ac is zero. Then it follows from the quadratic formula that b r1 = r2 = − . (3) 2a The difficulty is immediately apparent; both roots yield the same solution y1 ( t) = e−bt/( 2a)

(4)

of the differential equation (1), and it is not obvious how to find a second solution.

EXAMPLE 1 Solve the differential equation y + 4y + 4y = 0.

(5)

Solution: The characteristic equation is r 2 + 4r + 4 = (r + 2) 2 = 0, so r1 = r2 = −2. Therefore, one solution of equation (5) is y1 ( t) = e−2t . To find the general solution of equation (5), we need a second solution that is not a constant multiple of y1 . This second solution can be found in several ways (see Problems 15 through 17); here we use a method originated by d’Alembert7 in the eighteenth century. Recall that since y1 ( t) is a solution of equation (1), so is cy1 ( t) for any constant c. The basic idea is to generalize this observation by replacing c by a function v( t) and then trying to determine v( t) so that the product v( t) y1 ( t) is also a solution of equation (1). To carry out this program, we substitute y = v( t) y1 ( t) in equation (5) and use the resulting equation to find v( t) . Starting with y = v( t) y1 ( t) = v( t) e−2t ,

(6)

y = v ( t) e−2t − 2v( t) e−2t

(7)

y = v ( t) e−2t − 4v ( t) e−2t + 4v( t) e−2t .

(8)

we differentiate once to find

and a second differentiation yields

By substituting the expressions in equations (6), (7), and (8) in equation (5) and collecting terms, we obtain ( v ( t) − 4v ( t) + 4v( t) + 4v ( t) − 8v( t) + 4v( t) ) e−2t = 0, ..................................................................................................................................................................................... 7 Jean d’Alembert (1717--1783), a French mathematician, was a contemporary of Euler and Daniel Bernoulli and is known primarily for his work in mechanics and differential equations. D’Alembert’s principle in mechanics and d’Alembert’s paradox in hydrodynamics are named for him, and the wave equation first appeared in his paper on vibrating strings in 1747. In his later years he devoted himself primarily to philosophy and to his duties as science editor of Diderot’s Encyclopédie.

▼

127

Boyce 9131 Ch03 2

128

September 29, 2016

17:28

128

CHAPTER 3 Second-Order Linear Differential Equations

▼ which simplifies to v ( t) = 0.

(9)

Therefore, v ( t) = c1 and v( t) = c1 t + c2 ,

(10)

where c1 and c2 are arbitrary constants. Finally, substituting for v( t) in equation (6), we obtain y = c1 te−2t + c2 e−2t .

(11)

The second term on the right-hand side of equation (11) corresponds to the original solution y1 ( t) = exp( −2t) , but the first term arises from a second solution, namely, y2 ( t) = t exp( −2t) . We can verify that these two solutions form a fundamental set by calculating their Wronskian:

e−2t W [y1 , y2 ]( t) = −2e−2t

te−2t −4t − 2te−4t + 2te−4t −2t = e ( 1 − 2t) e

= e−4t = 0. Therefore, y1 ( t) = e−2t ,

y2 ( t) = te−2t

(12)

form a fundamental set of solutions of equation (5), and the general solution of that equation is given by equation (11). Note that both y1 ( t) and y2 ( t) tend to zero as t → ∞; consequently, all solutions of equation (5) behave in this way. The graphs of typical solutions are shown in Figure 3.4.1. y 2.0 1.8 1.6 1.4 1.2 1.0 0.8 0.6 0.4 0.2 t 0

0.5

1

1.5

2

2.5

FIGURE 3.4.1 Three solutions of equation (5): y + 4y + 4y = 0, with different sets of initial conditions: y( 0) = 2, y ( 0) = 1 (blue, dashed); y( 0) = 1, y ( 0) = 1 (green, solid); y( 0) = 1/2, y ( 0) = 1 (red).

The procedure used in Example 1 can be extended to a general equation whose characteristic equation has repeated roots. That is, we assume that the coefficients in equation (1) satisfy b2 − 4ac = 0, in which case y1 ( t) = e−bt/( 2a) is a solution. To find a second solution, we assume that y = v( t) y1 ( t) = v( t) e−bt/( 2a)

(13)

and substitute for y in equation (1) to determine v( t) . We have y = v ( t) e−bt/( 2a) −

b v( t) e−bt/( 2a) 2a

(14)

Boyce 9131 Ch03 2

September 29, 2016

17:28

129

3.4 Repeated Roots; Reduction of Order

and b b2 v ( t) e−bt/( 2a) + 2 v( t) e−bt/( 2a) . a 4a Then, by substituting in equation (1), we obtain b b2 b a v ( t) − v ( t) + 2 v( t) + b v ( t) − v( t) + cv( t) e−bt/( 2a) = 0. a 2a 4a y = v ( t) e−bt/( 2a) −

(15)

(16)

Canceling the factor e−b/( 2a) , which is nonzero, and rearranging the remaining terms, we find that b2 b2 av ( t) + ( −b + b) v ( t) + − + c v( t) = 0. (17) 4a 2a The term involving v ( t) is obviously zero. Further, the coefficient of v( t) is c − b2 /( 4a) , which is also zero because b2 − 4ac = 0 in the problem that we are considering. Thus, just as in Example 1, equation (17) reduces to v ( t) = 0, so v( t) = c1 + c2 t. Hence, from equation (13), we have y = c1 e−bt/( 2a) + c2 te−bt/( 2a) .

(18)

Thus y is a linear combination of the two solutions y1 ( t) = e−bt/( 2a) , The Wronskian of these two solutions is e−bt/( 2a) W ( y1 , y2 ) ( t) = b −bt/( 2a) − e 2a

y2 ( t) = te−bt/( 2a) . te−bt/( 2a) −bt/a . bt −bt/( 2a) = e 1− e 2a

(19)

(20)

Since W [y1 , y2 ]( t) is never zero, the solutions y1 and y2 given by equation (19) are a fundamental set of solutions. Further, equation (18) is the general solution of equation (1) when the roots of the characteristic equation are equal. In other words, in this case there is one exponential solution corresponding to the repeated root and a second solution that is obtained by multiplying the exponential solution by t.

EXAMPLE 2 Find the solution of the initial value problem y − y +

y = 0, 4

y( 0) = 2, y ( 0) =

1 . 3

(21)

Solution: The characteristic equation is r2 − r +

1 = 0, 4

so the roots are r1 = r2 = 1/2. Thus the general solution of the differential equation is y = c1 et/2 + c2 tet/2 .

▼

(22)

129

Boyce 9131 Ch03 2

130

September 29, 2016

17:28

130

CHAPTER 3 Second-Order Linear Differential Equations

▼ The first initial condition requires that y( 0) = c1 = 2. To satisfy the second initial condition, we first differentiate equation (22) and then set t = 0. This gives y ( 0) =

1 1 c 1 + c2 = , 2 3

so c2 = −2/3. Thus the solution of the initial value problem is 2 y = 2et/2 − tet/2 . 3

(23)

The graph of this solution is shown by the blue curve in Figure 3.4.2. y 4 y'(0) = 2: y = 2et/2 + te t/2 3 y'(0) =

1 3

: y = 2et/2 –

2 t/2 te 3

2

1

1

2

3

t

–1 FIGURE 3.4.2 Solutions of y − y + y/4 = 0, y( 0) = 2,

with y ( 0) = 1/3 (blue) and with y ( 0) = 2 (red).

Let us now modify the initial value problem (16) by changing the initial slope; to be specific, let the second initial condition be y ( 0) = 2. The solution of this modified problem is y = 2et/2 + tet/2 , and its graph is shown by the red curve in Figure 3.4.2. The graphs shown in this figure suggest that there is a critical initial slope, with a value between 1/3 and 2, that separates solutions that increase as t → ∞ from those that ultimately decrease as t → ∞. In Problem 12 you are asked to determine this critical initial slope.

The asymptotic behavior of solutions is similar in this case to that when the roots are real and different. If the exponents are either positive or negative, then the magnitude of the solution grows or decays accordingly; the linear factor t has little influence. A decaying solution is shown in Figure 3.4.1 and growing solutions in Figure 3.4.2. However, if the repeated root is zero, then the differential equation is y = 0 and the general solution is a linear function of t. Summary. We can now summarize the results that we have obtained for second-order linear homogeneous equations with constant coefficients ay + by + cy = 0.

(24)

Let r1 and r2 be the roots of the corresponding characteristic equation ar 2 + br + c = 0.

(25)

Boyce 9131 Ch03 2

September 29, 2016

17:28

131

3.4 Repeated Roots; Reduction of Order

If r1 and r2 are real but not equal, then the general solution of differential equation (24) is y = c1 er1 t + c2 er2 t .

(26)

If r1 and r2 are complex conjugates λ ± iμ , then the general solution is y = c1 eλt cos( μ t) + c2 eλt sin( μ t) .

(27)

If r1 = r2 , then the general solution is y = c1 er1 t + c2 ter1 t .

(28)

Reduction of Order. It is worth noting that the procedure used in this section for equations with constant coefficients is more generally applicable. Suppose that we know one solution y1 ( t) , not everywhere zero, of y + p( t) y + q( t) y = 0.

(29)

y = v( t) y1 ( t) ;

(30)

To find a second solution, let then y = v ( t) y1 ( t) + v( t) y1 ( t) and y = v ( t) y1 ( t) + 2v ( t) y1 ( t) + v( t) y1 ( t) . Substituting for y, y , and y in equation (29) and collecting terms, we find that y1 v + ( 2y1 + py1 ) v + ( y1 + py1 + qy1 ) v = 0.

(31)

Since y1 is a solution of equation (29), the coefficient of v in equation (31) is zero so that equation (31) becomes y1 v + ( 2y1 + py1 ) v = 0.

(32)

Despite its appearance, equation (32) is actually a first-order differential equation for the function v and can be solved either as a first-order linear equation or as a separable equation. Once v has been found, then v is obtained by an integration. Finally, y is determined from equation (30). This procedure is called the method of reduction of order, because the crucial step is the solution of a first-order differential equation for v rather than the original secondorder differential equation for y. Although it is possible to write down a formula for v( t) , we will instead illustrate how this method works by an example.

EXAMPLE 3 Given that y1 ( t) = t −1 is a solution of 2t 2 y + 3t y − y = 0,

t > 0,

(33)

find a fundamental set of solutions. Solution: We set y = v( t) t −1 ; then y = v t −1 − vt −2 ,

y = v t −1 − 2v t −2 + 2vt −3 .

Substituting for y, y , and y in equation (33) and collecting terms, we obtain

2t 2 v t −1 − 2v t −2 + 2vt −3 + 3t v t −1 − vt −2 − vt −1

= 2tv + ( −4 + 3) v + 4t −1 − 3t −1 − t −1 v = 2tv − v = 0.

▼

(34)

131

Boyce 9131 Ch03 2

132

September 29, 2016

17:28

132

CHAPTER 3 Second-Order Linear Differential Equations

▼ Note that the coefficient of v is zero, as it should be; this provides a useful check on our algebraic calculations. If we let w = v , then the second-order linear differential equation (34) reduces to the separable first-order differential equation 2tw − w = 0. Separating the variables and solving for w( t) , we find that w( t) = v ( t) = ct 1/2 ; then, one final integration yields v( t) =

2 3/2 ct + k. 3

It follows that y = v( t) t −1 =

2 1/2 + kt −1 , ct 3

(35)

where c and k are arbitrary constants. The second term on the right-hand side of equation (35) is a multiple of y1 ( t) and can be dropped, but the first term provides a new solution y2 ( t) = t 1/2 . You can verify that the Wronskian of y1 and y2 is W [y1 , y2 ]( t) =

3 −3/2 t = 0 for t > 0. 2

(36)

Consequently, y1 and y2 form a fundamental set of solutions of equation (33) for t > 0.

Problems In each of Problems 1 through 8, find the general solution of the given differential equation.

1. 2. 3. 4. 5. 6. 7. 8.

y − 2y + y = 0 9y + 6y + y = 0

4y − 4y − 3y = 0 y − 2y + 10y = 0 4y + 17y + 4y = 0 2y + 2y + y = 0

9y − 12y + 4y = 0,

y( 0) = 2, y ( 0) = −1

y − 6y + 9y = 0,

y( 0) = 0, y ( 0) = 2

y + 4y + 4y = 0,

y( −1) = 2, y ( −1) = 1

Consider the following modification of the initial value problem in Example 2: y − y +

y( 0) = 1, y ( 0) = 2.

14. Consider the equation ay + by + cy = 0. If the roots of the

16y + 24y + 9y = 0

4y + 4y + y = 0,

find the solution as a function of b. d. Find the coordinates ( t M , y M ) of the maximum point in terms of b. Describe the dependence of t M and y M on b as b increases.

y − 6y + 9y = 0

13. Consider the initial value problem

a. Solve the initial value problem and plot the solution. b. Determine the coordinates ( t M , y M ) of the maximum point. c. Change the second initial condition to y ( 0) = b > 0 and

In each of Problems 9 through 11, solve the given initial value problem. Sketch the graph of the solution and describe its behavior for increasing t.

9. 10. 11. 12.

N

y = 0, 4

y( 0) = 2, y ( 0) = b.

Find the solution as a function of b, and then determine the critical value of b that separates solutions that remain positive for all t > 0 from those that eventually become negative.

corresponding characteristic equation are real, show that a solution to the differential equation either is everywhere zero or else can take on the value zero at most once. Problems 15 through 17 indicate other ways of finding the second solution when the characteristic equation has repeated roots. 15. a. Consider the equation y + 2ay + a 2 y = 0. Show that the roots of the characteristic equation are r1 = r2 = −a so that one solution of the equation is e−at .

b. Use Abel’s formula [equation (23) of Section 3.2] to show that the Wronskian of any two solutions of the given equation is W ( t) = y1 ( t) y2 ( t) − y1 ( t) y2 ( t) = c1 e−2at , where c1 is a constant.

c. Let y1 ( t) = e−at and use the result of part b to obtain a differential equation satisfied by a second solution y2 ( t) . By solving this equation, show that y2 ( t) = te−at .

Boyce 9131 Ch03 2

September 29, 2016

17:28

133

3.5 Nonhomogeneous Equations; Method of Undetermined Coefficients

16. Suppose that r1 and r2 are roots of ar 2 + br + c = 0 and that

r1 = r2 ; then exp(r1 t) and exp(r2 t) are solutions of the differential equation ay + by + cy = 0. Show that φ ( t; r1 , r2 ) =

e

r2 t

r t

−e 1 r2 − r1

133

satisfies ( y2 / y1 ) = W [y1 , y2 ]/ y1 , where W [y1 , y2 ] is the Wronskian of y1 and y2 . Then use Abel’s formula (equation (23) of Section 3.2) to determine y2 . 2

In each of Problems 25 through 27, use the method of Problem 24 to find a second independent solution of the given equation.

is also a solution of the equation for r2 = r1 . Then think of r1 as fixed, and use l’Hôpital’s rule to evaluate the limit of φ ( t; r1 , r2 ) as r2 → r1 , thereby obtaining the second solution in the case of equal roots.

25. t 2 y + 3t y + y = 0, t > 0; y1 ( t) = t −1 26. t y − y + 4t 3 y = 0, t > 0; y1 ( t) = sin( t 2 ) 27. x 2 y + x y +( x 2 −0.25) y = 0, x > 0; y1 ( x) = x −1/2 sin x

17. a. If ar 2 + br + c = 0 has equal roots r1 , show that rt r t rt rt 2 rt

Behavior of Solutions as t → ∞. Problems 28 through 30 are concerned with the behavior of solutions as t → ∞.

= a( e ) + b( e ) + ce = a(r − r1 ) e .

L e

(37)

Since the right-hand side of equation (37) is zero when r = r1 , it follows that exp(r1 t) is a solution of L[y] = ay + by + cy = 0. b. Differentiate equation (37) with respect to r , and interchange differentiation with respect to r and with respect to t, thus showing that

28. If a, b, and c are positive constants, show that all solutions of ay + by + cy = 0 approach zero as t → ∞.

29. a. If a > 0 and c > 0, but b = 0, show that the result of Problem 28 is no longer true, but that all solutions are bounded as t → ∞. b. If a > 0 and b > 0, but c = 0, show that the result of Problem 28 is no longer true, but that all solutions approach a constant that depends on the initial conditions as t → ∞. Determine this constant for the initial conditions y( 0) = y0 , y ( 0) = y0 .

∂ rt ∂ rt = L ter t L e =L e ∂r ∂r = ater t (r − r1 ) 2 + 2aer t (r − r1 ) .

(38)

Since the right-hand side of equation (36) is zero when r = r1 , conclude that t exp(r1 t) is also a solution of L[y] = 0. In each of Problems 18 through 22, use the method of reduction of order to find a second solution of the given differential equation.

18. 19. 20. 21. 22. 23.

t 2 y − 4t y + 6y = 0, t > 0; 2

2

t y + 2t y − 2y = 0, t > 0; t y + 3t y + y = 0, t > 0;

x y − y + 4x y = 0, x > 0; 3

y1 ( t) = t 2 y1 ( t) = t y1 ( t) = t −1 y1 ( x) = sin( x 2 )

x 2 y + x y +( x 2 −0.25) y = 0, x > 0;

y1 ( x) = x −1/2 sin x

The differential equation y + δ ( x y + y) = 0

arises in the study of the turbulent flow of a uniform stream past a circular cylinder. Verify that y1 ( x) = exp( −δ x 2 /2) is one solution, and then find the general solution in the form of an integral.

24. The method of Problem 15 can be extended to second-order equations with variable coefficients. If y1 is a known nonvanishing solution of y + p( t) y + q( t) y = 0, show that a second solution y2

30. Show that y = sin t is a solution of y + ( k sin2 t) y + ( 1 − k cos t sin t) y = 0 for any value of the constant k. If 0 < k < 2, show that 1 − k cos t sin t > 0 and k sin2 t ≥ 0. Thus observe that even though the coefficients of this variable-coefficient differential equation are nonnegative (and the coefficient of y is zero only at the points t = 0, π, 2π, . . . ) , it has a solution that does not approach zero as t → ∞. Compare this situation with the result of Problem 28. Thus we observe a not unusual situation in the study of differential equations: equations that are apparently very similar can have quite different properties. Euler Equations. In each of Problems 31 through 34, use the substitution introduced in Problem 25 in Section 3.3 to solve the given differential equation.

31. 32. 33. 34.

t 2 y − 3t y + 4y = 0, 2

t>0

t y + 2t y + 0.25y = 0, t 2 y + 3t y + y = 0, 4t 2 y − 8t y + 9y = 0,

Nonhomogeneous Equations; Method of Undetermined Coefficients 3.5

We now turn our attention to the nonhomogeneous second-order linear differential equation L[y] = y + p( t) y + q( t) y = g( t) ,

(1)

where p, q, and g are given (continuous) functions on the open interval I . The equation L[y] = y + p( t) y + q( t) y = 0,

(2)

t>0

t>0 t>0

Boyce 9131 Ch03 2

134

September 29, 2016

17:28

134

CHAPTER 3 Second-Order Linear Differential Equations

in which g( t) = 0 and p and q are the same as in equation (1), is called the homogeneous differential equation corresponding to equation (1). The following two results describe the structure of solutions of the nonhomogeneous equation (1) and provide a foundation for constructing its general solution.

Theorem 3.5.1 If Y1 and Y2 are two solutions of the nonhomogeneous linear differential equation (1), then their difference Y1 − Y2 is a solution of the corresponding homogeneous differential equation (2). If, in addition, y1 and y2 form a fundamental set of solutions of equation (2), then Y1 ( t) − Y2 ( t) = c1 y1 ( t) + c2 y2 ( t) ,

(3)

where c1 and c2 are certain constants.

To prove this result, note that Y1 and Y2 satisfy the equations L[Y1 ]( t) = g( t) ,

L[Y2 ]( t) = g( t) .

(4)

Subtracting the second of these equations from the first, we have L[Y1 ]( t) − L[Y2 ]( t) = g( t) − g( t) = 0.

(5)

However, L[Y1 ] − L[Y2 ] = L[Y1 − Y2 ], so equation (5) becomes L[Y1 − Y2 ]( t) = 0.

(6)

Equation (6) states that Y1 − Y2 is a solution of equation (2). Finally, since by Theorem 3.2.4 all solutions of equation (2) can be expressed as linear combinations of a fundamental set of solutions, it follows that the solution Y1 − Y2 can be so written. Hence equation (3) holds and the proof is complete.

Theorem 3.5.2 The general solution of the nonhomogeneous equation (1) can be written in the form y = φ ( t) = c1 y1 ( t) + c2 y2 ( t) + Y ( t) ,

(7)

where y1 and y2 form a fundamental set of solutions of the corresponding homogeneous equation (2), c1 and c2 are arbitrary constants, and Y is any solution of the nonhomogeneous equation (1).

The proof of Theorem 3.5.2 follows quickly from Theorem 3.5.1. Note that equation (3) holds if we identify Y1 with an arbitrary solution φ of equation (1) and Y2 with the specific solution Y . From equation (3) we thereby obtain φ ( t) − Y ( t) = c1 y1 ( t) + c2 y2 ( t) ,

(8)

which is equivalent to equation (7). Since φ is an arbitrary solution of equation (1), the expression on the right-hand side of equation (7) includes all solutions of equation (1); thus it is natural to call it the general solution of equation (1). In somewhat different words, Theorem 3.5.2 states that to solve the nonhomogeneous equation (1), we must do three things: 1. Find the general solution c1 y1 ( t) + c2 y2 ( t) of the corresponding homogeneous equation. This solution is frequently called the complementary solution and may be denoted by yc ( t) . 2. Find any solution Y ( t) of the nonhomogeneous equation. Often this solution is referred to as a particular solution. 3. Form the sum of the functions found in steps 1 and 2. We have already discussed how to find yc ( t) , at least when the homogeneous equation (2) has constant coefficients. Therefore, in the remainder of this section and Section 3.6, we

Boyce 9131 Ch03 2

September 29, 2016

17:28

135

3.5 Nonhomogeneous Equations; Method of Undetermined Coefficients

will focus on finding a particular solution Y ( t) of the nonhomogeneous linear differential equation (1). There are two methods that we wish to discuss. They are known as the method of undetermined coefficients (discussed here) and the method of variation of parameters (see Section 3.6). Each has some advantages and some possible shortcomings. Method of Undetermined Coefficients. The method of undetermined coefficients requires us to make an initial assumption about the form of the particular solution Y ( t) , but with the coefficients left unspecified. We then substitute the assumed expression into the nonhomogeneous differential equation (1) and attempt to determine the coefficients so as to satisfy that equation. If we are successful, then we have found a solution of the differential equation (1) and can use it for the particular solution Y ( t) . If we cannot determine the coefficients, then this means that there is no solution of the form that we assumed. In this case we may modify the initial assumption and try again. The main advantage of the method of undetermined coefficients is that it is straightforward to execute once the assumption is made about the form of Y ( t) . Its major limitation is that it is useful primarily for equations for which we can easily write down the correct form of the particular solution in advance. For this reason, this method is usually used only for problems in which the homogeneous equation has constant coefficients and the nonhomogeneous term is restricted to a relatively small class of functions. In particular, we consider only nonhomogeneous terms that consist of polynomials, exponential functions, sines, and cosines. Despite this limitation, the method of undetermined coefficients is useful for solving many problems that have important applications. However, the algebraic details may become tedious, and a computer algebra system can be very helpful in practical applications. We will illustrate the method of undetermined coefficients by several simple examples and then summarize some rules for using it.

EXAMPLE 1 Find a particular solution of y − 3y − 4y = 3e2t .

(9)

Solution: We seek a function Y such that the combination Y ( t) − 3Y ( t) − 4Y ( t) is equal to 3e2t . Since the exponential function reproduces itself through differentiation, the most plausible way to achieve the desired result is to assume that Y ( t) is some multiple of e2t , Y ( t) = Ae2t , where the coefficient A is yet to be determined. To find A, we calculate the first two derivatives of Y : Y ( t) = 2Ae2t ,

Y ( t) = 4Ae2t ,

and substitute for y, y , and y in the nonhomogeneous differential equation (9). We obtain Y − 3Y − 4Y = ( 4A − 6A − 4A) e2t = 3e2t . 1 Hence −6Ae2t must equal 3e2t , so −6A = 3 and we conclude that A = − . Thus a particular 2 solution is 1 Y ( t) = − e2t . 2

(10)

y − 3y − 4y = 2 sin t.

(11)

EXAMPLE 2 Find a particular solution of

▼

135

Boyce 9131 Ch03 2

136

September 29, 2016

17:28

136

CHAPTER 3 Second-Order Linear Differential Equations

▼ Solution: By analogy with Example 1, let us assume that Y ( t) = A sin t, where A is a constant to be determined. On substituting this guess in equation (11) we obtain Y − 3Y − 4Y = −A sin t − 3A cos t − 4A sin t = 2 sin t, or, moving all terms to the left-hand side and collecting the terms involving sin t and cos t, we arrive at, ( 2 + 5A) sin t + 3A cos t = 0.

(12)

We want equation (12) to hold for all t. Thus it must hold for two specific points, such as t = 0 π . At these points equation (12) reduces to 3A = 0 and 2 + 5A = 0, respectively. and t = 2 These contradictory requirements mean that there is no choice of the constant A that makes π equation (12) true for t = 0 and t = , much less for all t. Thus we conclude that our assumption 2 concerning Y ( t) is inadequate. The appearance of the cosine term in equation (12) suggests that we modify our original assumption to include a cosine term in Y ( t) ; that is, Y ( t) = A sin t + B cos t, where A and B are the undetermined coefficients. Then Y ( t) = A cos t − B sin t,

Y ( t) = −A sin t − B cos t.

By substituting these expressions for y, y , and y in equation (11) and collecting terms, we obtain Y − 3Y − 4Y = ( −A + 3B − 4A) sin t + ( −B − 3A − 4B) cos t = 2 sin t.

(13)

Now, working exactly as with the first guess, move all terms to the left-hand side and evaluate t = 0 π to find that A and B must satisfy the equations and t = 2 −5A + 3B − 2 = 0,

−3A − 5B = 0.

Solving these algebraic equations for A and B, we obtain A = −

3 5 and B = ; hence a particular 17 17

solution of equation (11) is Y ( t) = −

3 5 sin t + cos t. 17 17

The method illustrated in the preceding examples can also be used when the right-hand side of the equation is a polynomial. Thus, to find a particular solution of y − 3y − 4y = 4t 2 − 1,

(14)

we initially assume that Y ( t) is a polynomial of the same degree as the nonhomogeneous term; that is, Y ( t) = At 2 + Bt + C. To summarize our conclusions up to this point: if the nonhomogeneous term g( t) in differential equation (1) is an exponential function eα t , then assume that Y ( t) is proportional to the same exponential function; if g( t) is sin( β t) or cos( β t) , then assume that Y ( t) is a linear combination of sin( β t) and cos( β t) ; if g( t) is a polynomial of degree n, then assume that Y ( t) is a polynomial of degree n. The same principle extends to the case where g( t) is a product of any two, or all three, of these types of functions, as the next example illustrates.

EXAMPLE 3 Find a particular solution of y − 3y − 4y = −8et cos( 2t) .

▼

(15)

Boyce 9131 Ch03 2

September 29, 2016

17:28

137

3.5 Nonhomogeneous Equations; Method of Undetermined Coefficients

▼ Solution: In this case we assume that Y ( t) is the product of et and a linear combination of cos( 2t) and sin( 2t) ; that is, Y ( t) = Aet cos( 2t) + Bet sin( 2t) . The algebra is more tedious in this example, but it follows that Y ( t) = ( A + 2B) et cos( 2t) + ( −2A + B) et sin( 2t) and Y ( t) = ( −3A + 4B) et cos( 2t) + ( −4A − 3B) et sin( 2t) . By substituting these expressions in equation (15), we find that A and B must satisfy 10A + 2B = 8, Hence A =

2A − 10B = 0.

2 10 and B = ; therefore, a particular solution of equation (15) is 13 13 Y ( t) =

10 t 2 e cos( 2t) + et sin( 2t) . 13 13

Now suppose that g( t) is the sum of two terms, g( t) = g1 ( t) + g2 ( t) , and suppose that Y1 and Y2 are solutions of the equations ay + by + cy = g1 ( t)

(16)

ay + by + cy = g2 ( t) ,

(17)

and respectively. Then Y1 + Y2 is a solution of the equation ay + by + cy = g( t) .

(18)

To prove this statement, substitute Y1 ( t) + Y2 ( t) for y in equation (18) and make use of equations (16) and (17). A similar conclusion holds if g( t) is the sum of any finite number of terms. The practical significance of this result is that for an equation whose nonhomogeneous function g( t) can be expressed as a sum, you can consider instead several simpler equations and then add together the results. The following example is an illustration of this procedure.

EXAMPLE 4 Find a particular solution of y − 3y − 4y = 3e2t + 2 sin t − 8et cos( 2t) .

(19)

Solution: By splitting up the right-hand side of equation (19), we obtain the three equations y − 3y − 4y = 3e2t , y − 3y − 4y = 2 sin t, and y − 3y − 4y = −8et cos( 2t) . Solutions of these three equations have been found in Examples 1, 2, and 3, respectively. Therefore, a particular solution of equation (19) is their sum, namely, 1 3 2 5 10 Y ( t) = − e2t + cos t − sin t + et cos( 2t) + et sin( 2t) . 2 17 17 13 13

The procedure illustrated in these examples enables us to solve a fairly large class of problems in a reasonably efficient manner. However, there is one difficulty that sometimes occurs. The next example illustrates how it arises.

137

Boyce 9131 Ch03 2

138

September 29, 2016

17:28

138

CHAPTER 3 Second-Order Linear Differential Equations

EXAMPLE 5 Find a particular solution of y − 3y − 4y = 2e−t .

(20)

Solution: Proceeding as in Example 1, we assume that Y ( t) = Ae−t . By substituting in equation (20), we obtain Y − 3Y − 4Y = ( A + 3A − 4A) e−t = 2e−t .

(21)

Since the left-hand side of equation (21) is zero, there is no choice of A for which 0 = 2e−t . Therefore, there is no particular solution of equation (20) of the assumed form. The reason for this possibly unexpected result becomes clear if we solve the homogeneous equation y − 3y − 4y = 0

(22)

that corresponds to equation (20). The two functions in a fundamental set of solutions of equation (22) are y1 ( t) = e−t and y2 ( t) = e4t . Thus our assumed particular solution of equation (20) is actually a solution of the homogeneous equation (22); consequently, it cannot possibly be a solution of the nonhomogeneous equation (20). To find a solution of equation (20), we must therefore consider functions of a somewhat different form. At this stage, we have several possible alternatives. One is simply to try to guess the proper form of the particular solution of equation (20). Another is to solve this equation in some different way and then to use the result to guide our assumptions if this situation arises again in the future; see Problems 22 and 27 for other solution methods. Still another possibility is to seek a simpler equation where this difficulty occurs and to use its solution to suggest how we might proceed with equation (20). Adopting the latter approach, we look for a first-order equation analogous to equation (20). One possibility is the linear equation y + y = 2e−t .

(23) −t

−t

If we try to find a particular solution of equation (23) of the form Ae , we will fail because e is a solution of the homogeneous equation y + y = 0. However, from Section 2.1 we already know how to solve equation (23). An integrating factor is μ ( t) = et , and by multiplying by μ ( t) and then integrating both sides, we obtain the solution y = 2te−t + ce−t .

(24)

The second term on the right-hand side of equation (24) is the general solution of the homogeneous equation y + y = 0, but the first term is a solution of the full nonhomogeneous equation (23). Observe that it involves the exponential factor e−t multiplied by the factor t. This is the clue that we were looking for. We now return to equation (20) and assume a particular solution of the form Y ( t) = Ate−t . Then Y ( t) = Ae−t − Ate−t ,

Y ( t) = −2Ae−t + Ate−t .

(25)

Substituting these expressions for y, y , and y in equation (20), we obtain Y − 3Y − 4Y = ( −2A − 3A) e−t + ( A + 3A − 4A) te−t = 2e−t . 2 The coefficient of te−t is zero, and from the terms involving et we have −5A = 2, so A = − . 5 Thus a particular solution of equation (20) is 2 Y ( t) = − te−t . 5

(26)

The outcome of Example 5 suggests a modification of the principle stated previously: if the assumed form of the particular solution duplicates a solution of the corresponding homogeneous equation, then modify the assumed particular solution by multiplying it by t. Occasionally, this modification will be insufficient to remove all duplication with the solutions of the homogeneous equation, in which case it is necessary to multiply by t a second time. For a second-order equation, it will never be necessary to carry the process further than this. Summary. We now summarize the steps involved in finding the solution of an initial value problem consisting of a nonhomogeneous linear differential equation of the form ay + by + cy = g( t) ,

(27)

Boyce 9131 Ch03 2

September 29, 2016

17:28

139

3.5 Nonhomogeneous Equations; Method of Undetermined Coefficients

where the coefficients a, b, and c are constants, together with a given set of initial conditions. 1. Find the general solution of the corresponding homogeneous equation. 2. Make sure that the function g( t) in equation (27) belongs to the class of functions discussed in this section; that is, be sure it involves nothing more than exponential functions, sines, cosines, polynomials, or sums or products of such functions. If this is not the case, use the method of variation of parameters (discussed in Section 3.6). 3. If g( t) = g1 ( t) + · · · + gn ( t) ---that is, if g( t) is a sum of n terms---then form n subproblems, each of which contains only one of the terms g1 ( t) , . . . , gn ( t) . The i th subproblem consists of the equation ay + by + cy = gi ( t) , where i runs from 1 to n. 4. For the i th subproblem, assume a particular solution Yi ( t) consisting of the appropriate exponential function, sine, cosine, polynomial, or combination thereof. If there is any duplication in the assumed form of Yi ( t) with the solutions of the homogeneous equation (found in step 1), then multiply Yi ( t) by t, or (if necessary) by t 2 , so as to remove the duplication. See Table 3.5.1. T A B L E 3.5.1

The Particular Solution of ay + by + cy = gi (t)

gi (t)

Yi (t)

Pn ( t) = a0 t n + a1 t n−1 + · · · + an

t s ( A0 t n + A1 t n−1 + · · · + An )

Pn ( t) eα t

t s ( A0 t n + A1 t n−1 + · · · + An ) eα t

Pn ( t) eα t

sin β t cos β t

t s ( A0 t n + A1 t n−1 + · · · + An ) eα t cos( β t) + ( B0 t n + B1 t n−1 + · · · + Bn ) eα t sin( β t)

Notes: Here, s is the smallest nonnegative integer ( s = 0, 1, or 2) that will ensure that no term in Yi ( t) is a solution of the corresponding homogeneous equation. Equivalently, for the three cases, s is the number of times 0 is a root of the characteristic equation, α is a root of the characteristic equation, and α + iβ is a root of the characteristic equation, respectively.

5. Find a particular solution Yi ( t) for each of the subproblems. Then Y1 ( t) + · · · + Yn ( t) is a particular solution of the full nonhomogeneous equation (27). 6. Form the sum of the general solution of the homogeneous equation (step 1) and the particular solution of the nonhomogeneous equation (step 5). This is the general solution of the nonhomogeneous equation. 7. When initial conditions are provided, use them to determine the values of the arbitrary constants remaining in the general solution. For some problems this entire procedure is easy to carry out by hand, but often the algebraic calculations are lengthy. Once you understand clearly how the method works, a computer algebra system can be of great assistance in executing the details. The method of undetermined coefficients is self-correcting in the sense that if you assume too little for Y ( t) , then a contradiction is soon reached that usually points the way to the modification that is needed in the assumed form. On the other hand, if you assume too many terms, then some unnecessary work is done and some coefficients turn out to be zero, but at least the correct answer is obtained. Proof of the Method of Undetermined Coefficients. In the preceding discussion we have described the method of undetermined coefficients on the basis of several examples. To prove that the procedure always works as stated, we now give a general argument, in which we consider three cases corresponding to different forms for the nonhomogeneous term g( t) . Case 1: g(t) = Pn (t) = a0 tn + a1 tn−1 + · · · + an . In this case equation (27) becomes ay + by + cy = a0 t n + a1 t n−1 + · · · + an .

(28)

139

Boyce 9131 Ch03 2

140

September 29, 2016

17:28

140

CHAPTER 3 Second-Order Linear Differential Equations

To obtain a particular solution, we assume that Y ( t) = A0 t n + A1 t n−1 + · · · + An−2 t 2 + An−1 t + An .

(29)

Substituting in equation (28), we obtain a n( n − 1) A0 t n−2 + · · · + 2An−2 + b( n A0 t n−1 + · · · + An−1 ) + c( A0 t n + A1 t n−1 + · · · + An ) = a0 t n + · · · + an .

(30)

Equating the coefficients of like powers of t, beginning with t n , leads to the following sequence of equations: c A 0 = a0 , c A1 + nb A0 = a1 , . . . c An + b An−1 + 2a An−2 = an . Provided that c = 0, the solution of the first equation is A0 = a0 /c, and the remaining equations determine A1 , . . . , An successively. If c = 0 but b = 0, then the polynomial on the left-hand side of equation (30) is of degree n − 1, and we cannot satisfy equation (30). To be sure that aY ( t) + bY ( t) is a polynomial of degree n, we must choose Y ( t) to be a polynomial of degree n + 1. Hence we assume that Y ( t) = t ( A0 t n + · · · + An ) . Substituting this guess into equation (28), with c = 0, and simplifying yields aY + bY = b A0 ( n + 1) t n + ( a A0 ( n + 1) n + b A1 n) t n−1 + · · · = a0 t n + a1 t n−1 + · · · + an . There is no constant term in this expression for Y ( t) , but there is no need to include such a term since a constant is a solution of the homogeneous equation when c = 0. Since b = 0, we find A0 = a0 /( b( n +1) ) , and the other coefficients A1 , . . . , An can be determined similarly. If both c and b are zero, then the characteristic equation is ar 2 = 0 and r = 0 is a repeated root. Thus y1 = e0t = 1 and y2 = te0t = t form a fundamental set of solutions of the corresponding homogeneous equation. This leads us to assume that Y ( t) = t 2 ( A0 t n + · · · + An ) . The term aY ( t) gives rise to a term of degree n, and we can proceed as before. Again the constant and linear terms in Y ( t) are omitted since, in this case, they are both solutions of the homogeneous equation. Case 2: g(t) = eαt Pn (t). The problem of determining a particular solution of ay + by + cy = eα t Pn ( t)

(31)

can be reduced to the preceding case by a substitution. Let Y ( t) = eα t u( t) ; then Y ( t) = eα t ( u ( t) + α u( t) ) and Y ( t) = eα t ( u ( t) + 2α u ( t) + α 2 u( t) ) . Substituting for y, y , and y in equation (31), canceling the factor eα t , and collecting terms, we obtain au ( t) + ( 2aα + b) u ( t) + ( aα 2 + bα + c) u( t) = Pn ( t) .

(32)

The determination of a particular solution of equation (32) is precisely the same problem, except for the names of the constants, as solving equation (28). Therefore, if aα 2 + bα + c is not zero, we assume that u( t) = A0 t n + · · · + An ; hence a particular solution of equation (31) is of the form Y ( t) = eα t ( A0 t n + A1 t n−1 + · · · + An ) .

(33)

Boyce 9131 Ch03 2

September 29, 2016

17:28

141

3.5 Nonhomogeneous Equations; Method of Undetermined Coefficients

141

On the other hand, if aα 2 + bα + c is zero but 2aα + b is not, we must take u( t) to be of the form t ( A0 t n + · · · + An ) . The corresponding form for Y ( t) is t times the expression on the right-hand side of equation (33). Note that if aα 2 + bα + c is zero, then eα t is a solution of the homogeneous equation. If both aα 2 + bα + c and 2aα + b are zero (and this implies that both eα t and teα t are solutions of the homogeneous equation), then the correct form for u( t) is t 2 ( A0 t n + · · · + An ) . Hence Y ( t) is t 2 times the expression on the right-hand side of equation (33). Case 3: g(t) = eαt Pn (t) cos ( βt) or eαt Pn (t) sin ( βt) . These two cases are similar, so we consider only the latter. We can reduce this problem to the preceding one by noting that, as a consequence of Euler’s formula, sin( β t) = ( eiβ t − e−iβ t ) /( 2i) . Hence g( t) is of the form g( t) = Pn ( t)

e( α +iβ ) t − e( α −iβ ) t , 2i

and we should choose Y ( t) = e( α +iβ ) t ( A0 t n + · · · + An ) + e( α −iβ ) t ( B0 t n + · · · + Bn ) , or, equivalently, Y ( t) = eα t ( A0 t n + · · · + An ) cos( β t) + eα t ( B0 t n + · · · + Bn ) sin( β t) . Usually, the latter form is preferred because it does not involve the use of complex-valued coefficients. If α ± iβ satisfy the characteristic equation corresponding to the homogeneous equation, we must, of course, multiply each of the polynomials by t to increase their degrees by 1. If the nonhomogeneous function involves both cos( β t) and sin( β t) , it is usually convenient to treat these terms together, since each one individually may give rise to the same form for a particular solution. For example, if g( t) = t sin t + 2 cos t, the form for Y ( t) would be Y ( t) = ( A0 t + A1 ) sin t + ( B0 t + B1 ) cos t, provided that sin t and cos t are not solutions of the homogeneous equation.

Problems In each of Problems 1 through 10, find the general solution of the given differential equation.

1. 2. 3. 4. 5. 6. 7.

y − 2y − 3y = 3e2t y − y − 2y = −2t + 4t 2 y + y − 6y = 12e3t + 12e−2t y − 2y − 3y = −3te−t y + 2y = 3 + 4 sin( 2t) y + 2y + y = 2e−t y + y = 3 sin( 2t) + t cos( 2t)

8. u + ω 02 u = cos( ω t) ,

9. u +

2 ω0u

2

ω 2 = ω 0

= cos( ω 0 t)

10. y + y + 4y = 2 sinh t

Hint: sinh t = ( et − e−t ) /2

In each of Problems 11 through 15, find the solution of the given initial value problem.

11. 12. 13. 14.

y + y − 2y = 2t, y + 4y = t 2 + 3et ,

y( 0) = 0, y ( 0) = 1 y( 0) = 0, y ( 0) = 2

y − 2y + y = te + 4, t

y + 4y = 3 sin( 2t) ,

y( 0) = 1, y ( 0) = 1

y( 0) = 2, y ( 0) = −1

15. y + 2y + 5y = 4e−t cos( 2t) ,

y( 0) = 1, y ( 0) = 0

In each of Problems 16 through 21: a. Determine a suitable form for Y ( t) if the method of undetermined coefficients is to be used. N b. Use a computer algebra system to find a particular solution of the given equation.

16. 17. 18. 19. 20. 21. 22.

y + 3y = 2t 4 + t 2 e−3t + sin( 3t) y − 5y + 6y = et cos( 2t) + e2t ( 3t + 4) sin t y + 2y + 2y = 3e−t + 2e−t cos t + 4e−t t 2 sin t y + 4y = t 2 sin( 2t) + ( 6t + 7) cos( 2t) y + 3y + 2y = et ( t 2 + 1) sin( 2t) + 3e−t cos t + 4et y + 2y + 5y = 3te−t cos( 2t) − 2te−2t cos t Consider the equation y − 3y − 4y = 2e−t

(34)

from Example 5. Recall that y1 ( t) = e−t and y2 ( t) = e4t are solutions of the corresponding homogeneous equation. Adapting the method of reduction of order (Section 3.4), seek a solution of the nonhomogeneous equation of the form Y ( t) = v( t) y1 ( t) = v( t) e−t , where v( t) is to be determined.

Boyce 9131 Ch03 2

142

September 29, 2016

17:28

142

CHAPTER 3 Second-Order Linear Differential Equations

a. Substitute Y ( t) , Y ( t) , and Y ( t) into equation (34) and

show that v( t) must satisfy v − 5v = 2. b. Let w( t) = v ( t) and show that w( t) satisfies w − 5w = 2. Solve this equation for w( t) . c. Integrate w( t) to find v( t) and then show that 2 1 Y ( t) = − te−t + c1 e4t + c2 e−t . 5 5 The first term on the right-hand side is the desired particular solution of the nonhomogeneous equation. Note that it is a product of t and e−t .

23. Determine the general solution of y + λ 2 y =

N

26. If g( t) = d, a constant, show that every solution of equation (35)

approaches d/c as t → ∞. What happens if c = 0? What if b = 0 also?

27. In this problem we indicate an alternative procedure8 for solving the differential equation y + by + cy = ( D 2 + bD + c) y = g( t) ,

where b and c are constants, and D denotes differentiation with respect to t. Let r1 and r2 be the zeros of the characteristic polynomial of the corresponding homogeneous equation. These roots may be real and different, real and equal, or conjugate complex numbers. a. Verify that equation (36) can be written in the factored form ( D − r1 ) ( D − r2 ) y = g( t) ,

am sin( mπ t) ,

where r1 + r2 = −b and r1 r2 = c.

m=1

b. Let u = ( D − r2 ) y. Then show that the solution of

where λ > 0 and λ = mπ for m = 1, . . . , N . N

equation (36) can be found by solving the following two firstorder equations:

24. In many physical problems the nonhomogeneous term may

be specified by different formulas in different time periods. As an example, determine the solution y = φ ( t) of

y + y =

t, π eπ −t ,

(36)

0 ≤ t ≤ π, t > π,

( D − r1 ) u = g( t) ,

( D − r2 ) y = u( t) .

In each of Problems 28 through 30, use the method of Problem 27 to solve the given differential equation.

satisfying the initial conditions y( 0) = 0 and y ( 0) = 1. Assume that y and y are also continuous at t = π . Plot the nonhomogeneous term and the solution as functions of time. Hint: First solve the initial value problem for t ≤ π ; then solve for t > π , determining the constants in the latter solution from the continuity conditions at t = π . Behavior of Solutions as t → ∞. In Problems 25 and 26, we continue the discussion started with Problems 28 through 30 of Section 3.4. Consider the differential equation ay + by + cy = g( t) ,

(35)

where a, b, and c are positive.

28. y − 3y − 4y = 3e2t (see Example 1) 29. y + 2y + y = 2e−t (see Problem 6) 30. y + 2y = 3 + 4 sin( 2t) (see Problem 5) .............................................................................................................................. 8 R.

S. Luthar, “Another Approach to a Standard Differential Equation,” Two Year College Mathematics Journal 10 (1979), pp. 200--201. Also see D. C. Sandell and F. M. Stein, “Factorization of Operators of Second-Order Linear Homogeneous Ordinary Differential Equations,” Two Year College Mathematics Journal 8 (1977), pp. 132--141, for a more general discussion of factoring operators.

25. If Y1 ( t) and Y2 ( t) are solutions of equation (35), show that Y1 ( t) − Y2 ( t) → 0 as t → ∞. Is this result true if b = 0?

3.6

Variation of Parameters

In this section we describe a second method of finding a particular solution of a nonhomogeneous equation. This method, variation of parameters, is due to Lagrange and complements the method of undetermined coefficients rather well. The main advantage of variation of parameters is that it is a general method; in principle at least, it can be applied to any equation, and it requires no detailed assumptions about the form of the solution. In fact, later in this section we use this method to derive a formula for a particular solution of an arbitrary second-order linear nonhomogeneous differential equation. On the other hand, the method of variation of parameters eventually requires us to evaluate certain integrals involving the nonhomogeneous term in the differential equation, and this may present difficulties. Before looking at this method in the general case, we illustrate its use in an example.

EXAMPLE 1 Find the general solution of y + 4y = 8 tan t

▼

− π/2 < t < π/2.

(1)

Boyce 9131 Ch03 2

September 29, 2016

17:28

143

3.6 Variation of Parameters

▼ Solution: Observe that this problem is not a good candidate for the method of undetermined coefficients, as described in Section 3.5, because the nonhomogeneous term g( t) = 8 tan t involves a quotient (rather than a sum or a product) of sin t and cos t. Therefore, the method of undetermined coefficients cannot be applied; we need a different approach. Observe also that the homogeneous equation corresponding to equation (1) is y + 4y = 0,

(2)

and that the general solution of equation (2) is yc ( t) = c1 cos( 2t) + c2 sin( 2t) .

(3)

The basic idea in the method of variation of parameters is similar to the method of reduction of order introduced at the end of Section 3.4. In the general solution (3), replace the constants c1 and c2 by functions u 1 ( t) and u 2 ( t) , respectively, and then determine these functions so that the resulting expression y = u 1 ( t) cos( 2t) + u 2 ( t) sin( 2t)

(4)

is a solution of the nonhomogeneous equation (1). To determine u 1 and u 2 , we need to substitute for y from equation (4) in differential equation (1). However, even without carrying out this substitution, we can anticipate that the result will be a single equation involving some combination of u 1 , u 2 , and their first two derivatives. Since there is only one equation and two unknown functions, we can expect that there are many possible choices of u 1 and u 2 that will meet our needs. Alternatively, we may be able to impose a second condition of our own choosing, thereby obtaining two equations for the two unknown functions u 1 and u 2 . We will soon show (following Lagrange) that it is possible to choose this second condition in a way that makes the computation markedly more efficient.9 Returning now to equation (4), we differentiate it and rearrange the terms, thereby obtaining y = −2u 1 ( t) sin( 2t) + 2u 2 ( t) cos( 2t) + u 1 ( t) cos( 2t) + u 2 ( t) sin( 2t) .

(5)

Keeping in mind the possibility of choosing a second condition on u 1 and u 2 , let us require the sum of the last two terms on the right-hand side of equation (5) to be zero; that is, we require that u 1 ( t) cos( 2t) + u 2 ( t) sin( 2t) = 0.

(6)

It then follows from equation (5) that y = −2u 1 ( t) sin( 2t) + 2u 2 ( t) cos( 2t) .

(7)

Although the ultimate effect of the condition (6) is not yet clear, the removal of the terms involving u 1 and u 2 has simplified the expression for y . Further, by differentiating equation (7), we obtain y = −4u 1 ( t) cos( 2t) − 4u 2 ( t) sin( 2t) − 2u 1 ( t) sin( 2t) + 2u 2 ( t) cos( 2t) .

(8)

Then, substituting for y and y in equation (1) from equations (4) and (8), respectively, we find that y + 4y = − 4u 1 ( t) cos( 2t) − 4u 2 ( t) sin( 2t) − 2u 1 ( t) sin( 2t) + 2u 2 ( t) cos( 2t) + 4u 1 ( t) cos( 2t) + 4u 2 ( t) sin( 2t) = 8 tan t. Hence u 1 and u 2 must satisfy −2u 1 ( t) sin( 2t) + 2u 2 ( t) cos( 2t) = 8 tan t.

(9)

Summarizing our results to this point, we want to choose u 1 and u 2 so as to satisfy equations (6) and (9). These equations can be viewed as a pair of linear algebraic equations for the two unknown quantities u 1 ( t) and u 2 ( t) . Equations (6) and (9) can be solved in various ways. For example, solving equation (6) for u 2 ( t) , we have cos( 2t) . sin( 2t) Then, substituting for u 2 ( t) in equation (9) and simplifying, we obtain u 2 ( t) = −u 1 ( t)

u 1 ( t) = −

8 tan t sin( 2t) = −8 sin2 t. 2

(10)

(11)

..................................................................................................................................................................................... 9 An alternate, and more mathematically appealing, derivation of the second condition can be found in Problems 17 to 19 in Section 7.9.

▼

143

Boyce 9131 Ch03 2

144

September 29, 2016

17:28

144

CHAPTER 3 Second-Order Linear Differential Equations

▼ Further, putting this expression for u 1 ( t) back in equation (10) and using the double-angle formulas, we find that

u 2 ( t) =

8 sin2 t cos( 2t) 1 sin t ( 2 cos2 t − 1) =4 = 4 sin t 2 cos t − sin( 2t) cos t cos t

.

(12)

Having obtained u 1 ( t) and u 2 ( t) , we next integrate so as to find u 1 ( t) and u 2 ( t) . The result is u 1 ( t) = 4 sin t cos t − 4t + c1

(13)

u 2 ( t) = 4 ln( cos t) − 4 cos2 t + c2 .

(14)

and

On substituting these expressions in equation (4), we have y = ( 4 sin t cos t) cos( 2t) + ( 4 ln( cos t) − 4 cos2 t) sin( 2t) + c1 cos( 2t) + c2 sin( 2t) . Finally, by using the double-angle formulas once more, we obtain y = −2 sin( 2t) − 4t cos( 2t) + 4 ln( cos t) sin( 2t) + c1 cos( 2t) + c2 sin( 2t) .

(15)

The terms in equation (15) involving the arbitrary constants c1 and c2 are the general solution of the corresponding homogeneous equation, while the other three terms are a particular solution of the nonhomogeneous equation (1). Thus equation (15) is the general solution of equation (1).

The particular solution identified at the end of Example 1 corresponds to choosing both c1 , and c2 to be zero in equation (15). Any other choice of c1 and c2 is also a particular solution of the same nonhomogeneous differential equation. Notice, in particular, that choosing c1 = 0 and c2 = 2 in equation (15) yields a particular solution with only two terms: −4t cos( 2t) + 4 ln( cos t) sin( 2t) . We conclude this first look at the method of variation of parameters with the observation that the particular solution involves terms that might be difficult to anticipate. This explains why the method of undetermined coefficients is not a good candidate for this problem, and why the method of variation of parameters is needed. In the preceding example the method of variation of parameters worked well in determining a particular solution, and hence the general solution, of equation (1). The next question is whether this method can be applied effectively to an arbitrary equation. Therefore, we consider y + p( t) y + q( t) y = g( t) ,

(16)

where p, q, and g are given continuous functions. As a starting point, we assume that we know the general solution yc ( t) = c1 y1 ( t) + c2 y2 ( t)

(17)

of the corresponding homogeneous equation y + p( t) y + q( t) y = 0.

(18)

This is a major assumption. So far we have shown how to solve equation (18) only if it has constant coefficients. If equation (18) has coefficients that depend on t, then usually the methods described in Chapter 5 must be used to obtain yc ( t) . The crucial idea, as illustrated in Example 1, is to replace the constants c1 and c2 in equation (17) by functions u 1 ( t) and u 2 ( t) , respectively; thus we have y = u 1 ( t) y1 ( t) + u 2 ( t) y2 ( t) .

(19)

Then we try to determine u 1 ( t) and u 2 ( t) so that the expression in equation (19) is a solution of the nonhomogeneous equation (16) rather than the homogeneous equation (18). Thus we differentiate equation (19), obtaining y = u 1 ( t) y1 ( t) + u 1 ( t) y1 ( t) + u 2 ( t) y2 ( t) + u 2 ( t) y2 ( t) .

(20)

Boyce 9131 Ch03 2

September 29, 2016

17:28

145

3.6 Variation of Parameters

As in Example 1, we now set the terms involving u 1 ( t) and u 2 ( t) in equation (20) equal to zero; that is, we require that u 1 ( t) y1 ( t) + u 2 ( t) y2 ( t) = 0.

(21)

Then, from equation (20), we have y = u 1 ( t) y1 ( t) + u 2 ( t) y2 ( t) .

(22)

Further, by differentiating again, we obtain y = u 1 ( t) y1 ( t) + u 1 ( t) y1 ( t) + u 2 ( t) y2 ( t) + u 2 ( t) y2 ( t) .

(23)

Now we substitute for y, y , and y in equation (16) from equations (19), (22), and (23), respectively. After rearranging the terms in the resulting equation, we find that u 1 ( t) y1 ( t) + p( t) y1 ( t) + q( t) y1 ( t) + u 2 ( t) y2 ( t) + p( t) y2 ( t) + q( t) y2 ( t) + u 1 ( t) y1 ( t) + u 2 ( t) y2 ( t) = g( t) .

(24)

Each of the expressions in parentheses in the first two lines of equation (24) is zero because both y1 and y2 are solutions of the homogeneous equation (18). Therefore, equation (24) reduces to u 1 ( t) y1 ( t) + u 2 ( t) y2 ( t) = g( t) .

(25)

Equations (21) and (25) form a system of two linear algebraic equations for the derivatives u 1 ( t) and u 2 ( t) of the unknown functions. They correspond exactly to equations (6) and (9) in Example 1. Solving the system of equations (21), (25), we obtain y2 ( t) g( t) y1 ( t) g( t) u 1 ( t) = − , u 2 ( t) = , (26) W [y1 , y2 ]( t) W [y1 , y2 ]( t) where W [y1 , y2 ] is the Wronskian of y1 and y2 . Note that division by W [y1 , y2 ] is permissible since y1 and y2 are a fundamental set of solutions, and therefore their Wronskian is nonzero. By integrating equations (26), we find the desired functions u 1 ( t) and u 2 ( t) , namely, y2 ( t) g( t) y1 ( t) g( t) (27) u 1 ( t) = − dt + c1 , u 2 ( t) = dt + c2 . W [y1 , y2 ]( t) W [y1 , y2 ]( t) If the integrals in equations (27) can be evaluated in terms of elementary functions, then we substitute the results in equation (19), thereby obtaining the general solution of equation (16). More generally, the solution can always be expressed in terms of integrals, as stated in the following theorem.

Theorem 3.6.1 Consider the nonhomogeneous second-order linear differential equation y + p( t) y + q( t) y = g( t) .

(28)

If the functions p, q, and g are continuous on an open interval I , and if the functions y1 and y2 form a fundamental set of solutions of the corresponding homogeneous equation y + p( t) y + q( t) y = 0, then a particular solution of equation (28) is

t

Y ( t) = −y1 ( t) t0

y2 ( s) g( s) ds + y2 ( t) W [y1 , y2 ]( s)

(29)

t

t0

y1 ( s) g( s) ds , W [y1 , y2 ]( s)

(30)

where t0 is any conveniently chosen point in I . The general solution is y = c1 y1 ( t) + c2 y2 ( t) + Y ( t) , as prescribed by Theorem 3.5.2.

(31)

145

Boyce 9131 Ch03 2

146

September 29, 2016

17:28

146

CHAPTER 3 Second-Order Linear Differential Equations

By examining the expression (30) and reviewing the process by which we derived it, we can see that there may be two major difficulties in carrying out the method of variation of parameters. As we have mentioned earlier, one is the determination of functions y1 ( t) and y2 ( t) that form a fundamental set of solutions of the homogeneous equation (29) when the coefficients in that equation are not constants. The other possible difficulty lies in the evaluation of the integrals appearing in equation (30). This depends entirely on the nature of the functions y1 , y2 , and g. In using equation (30), be sure that the differential equation is exactly in the form (28); otherwise, the nonhomogeneous term g( t) will not be correctly identified. A major advantage of the method of variation of parameters is that equation (30) provides an expression for the particular solution Y ( t) in terms of an arbitrary forcing function g( t) . This expression is a good starting point if you wish to investigate the effect of variations in the forcing function, or if you wish to analyze the response of a system to a number of different forcing functions. (See Problems 18 to 22.)

Problems In each of Problems 1 through 3, use the method of variation of parameters to find a particular solution of the given differential equation. Then check your answer by using the method of undetermined coefficients.

1. y − 5y + 6y = 2et 2. y − y − 2y = 2e−t 3. 4y − 4y + y = 16et/2

4. 5. 6. 7. 8. 9.

y + y = tan t,

0 < t < π/2

y + 4y + 4y = t

0 < t < π/6

−2 −2t

e

4y + y = 2 sec ( t/2) ,

t>0

,

−π < t < π

y − 2y + y = et /( 1 + t 2 )

L[v] = g( t) ,

y( t0 ) = y0 , y ( t0 ) = y0 (32)

v( t0 ) = 0,

u ( t0 ) = y0 ,

v ( t0 ) = 0,

(33) (34)

respectively. In other words, the nonhomogeneities in the differential equation and in the initial conditions can be dealt with separately. Observe that u is easy to find if a fundamental set of solutions of L[u] = 0 is known. And, as shown in Problem 16, the function v is given by equation (30).

18. a. Use the result of Problem 16 to show that the solution of the initial value problem

y − 5y + 6y = g( t)

y + y = g( t) ,

In each of Problems 10 through 15, verify that the given functions y1 and y2 satisfy the corresponding homogeneous equation; then find a particular solution of the given nonhomogeneous equation. In Problems 14 and 15, g is an arbitrary continuous function.

10. t 2 y − 2y = 3t 2 − 1, t > 0; y1 ( t) = t 2 , y2 ( t) = t −1 11. t 2 y − t ( t + 2) y + ( t + 2) y = 2t 3 , t > 0; y1 ( t) = t, y2 ( t) = tet

12. t y −( 1+t) y +y = t 2 e2t , t > 0; y1 ( t) = 1+t, y2 ( t) = et 13. x 2 y − 3x y + 4y = x 2 ln x, x > 0; y1 ( x) = x 2 , y2 ( x) = x 2 ln x 2

u( t0 ) = y0 ,

L[u] = 0,

y + 9y = 9 sec2 ( 3t) ,

L[y] = y + p( t) y + q( t) y = g( t) ,

can be written as y = u( t) + v( t) , where u and v are solutions of the two initial value problems

In each of Problems 4 through 9, find the general solution of the given differential equation. In Problems 9, g is an arbitrary continuous function.

17. Show that the solution of the initial value problem

y( t0 ) = 0, y ( t0 ) = 0

(35)

is

t

y=

sin( t − s) g( s) ds .

(36)

t0

b. Use the result of Problem 17 to find the solution of the initial value problem y + y = g( t) ,

y( 0) = y0 , y ( 0) = y0 .

19. Use the result of Problem 16 to find the solution of the initial value problem

14. x y + x y + ( x −

1 )y 4

= 3x sin x, x > 0; y1 ( x) = x −1/2 sin x, y2 ( x) = x −1/2 cos x 2

3/2

L[y] = g( t) ,

y( t0 ) = 0, y ( t0 ) = 0,

15. x 2 y + x y + ( x 2 − 0.25) y = g( x) , x > 0; y1 ( x) = x −1/2 sin x, y2 ( x) = x −1/2 cos x

where L[y] = ( D − a) ( D − b) y for real numbers a and b with a = b. Note that L[y] = y − ( a + b) y + aby.

16. By choosing the lower limit of integration in equation (30) in the

20. Use the result of Problem 16 to find the solution of the initial

text as the initial point t0 , show that Y ( t) becomes

t

Y ( t) = t0

y1 ( s) y2 ( t) − y1 ( t) y2 ( s) g( s) ds . y1 ( s) y2 ( s) − y1 ( s) y2 ( s)

Show that Y ( t) is a solution of the initial value problem L[y] = g( t) ,

y( t0 ) = 0, y ( t0 ) = 0.

value problem L[y] = g( t) ,

y( t0 ) = 0, y ( t0 ) = 0,

where L[y] = ( D − ( λ + iμ ) ) ( D − ( λ − iμ ) ) y; that is, L[y] = y −2λ y +( λ 2 +μ 2 ) y. Note that the roots of the characteristic equation are λ ± iμ .

Boyce 9131 Ch03 2

September 29, 2016

17:28

147

3.7 Mechanical and Electrical Vibrations

21. Use the result of Problem 16 to find the solution of the initial value problem L[y] = g( t) ,

initial point t0 to the current value t. The integral in equation (37) is called the convolution of K and g, and K is referred to as the kernel.

23. The method of reduction of order (Section 3.4) can also be used

y( t0 ) = 0, y ( t0 ) = 0,

for the nonhomogeneous equation

where L[y] = ( D − a) 2 y, that is, L[y] = y − 2ay + a 2 y, and a is any real number.

22. By combining the results of Problems 19 through 21, show that the solution of the initial value problem L[y] = ( D 2 + bD + c) y = g( t) ,

y( t0 ) = 0, y ( t0 ) = 0,

K ( t − s) g( s) ds,

(37)

t0

where the function K depends only on the solutions y1 and y2 of the corresponding homogeneous equation and is independent of the nonhomogeneous term. Once K is determined, all nonhomogeneous problems involving the same differential operator L are reduced to the evaluation of an integral. Note also that although K depends on both t and s, only the combination t − s appears, so K is actually a function of a single variable. When we think of g( t) as the input to the problem and of φ ( t) as the output, it follows from equation (37) that the output depends on the input over the entire interval from the

3.7

(39)

Equation (39) is a first-order linear differential equation for v . By solving equation (39) for v , integrating the result to find v, and then multiplying by y1 ( t) , you can find the general solution of equation (38). This method simultaneously finds both the second homogeneous solution and a particular solution. In each of Problems 24 through 26, use the method outlined in Problem 23 to solve the given differential equation.

24. t 2 y − 2t y + 2y = 4t 2 , t > 0; 2

25. t y + 7t y + 5y = t, t > 0;

y1 ( t) = t y1 ( t) = t −1

26. t y − ( 1 + t) y + y = t 2 e2t , t > 0; Problem 12)

Mechanical and Electrical Vibrations

One of the reasons why second-order linear differential equations with constant coefficients are worth studying is that they serve as mathematical models of many important physical processes. Two important areas of application are the fields of mechanical and electrical oscillations. For example, the motion of a mass on a vibrating spring, the torsional oscillations of a shaft with a flywheel, the flow of electric current in a simple series circuit, and many other physical problems are all described by the solution of an initial value problem of the form ay + by + cy = g( t) ,

(38)

provided one solution y1 of the corresponding homogeneous equation is known. Let y = v( t) y1 ( t) and show that y satisfies equation (38) if v is a solution of

t

y = φ ( t) =

y + p( t) y + q( t) y = g( t) ,

y1 ( t) v + 2y1 ( t) + p( t) y1 ( t) v = g( t) .

where b and c are constants, can be written in the form

147

y( 0) = y0 , y ( 0) = y0 .

(1)

This illustrates a fundamental relationship between mathematics and physics: many physical problems may have mathematically equivalent models. Thus, once we know how to solve the initial value problem (1), it is only necessary to make appropriate interpretations of the constants a, b, and c, and of the functions y and g, to obtain solutions of different physical problems. We will study the motion of a mass on a spring in detail because understanding the behavior of this simple system is the first step in the investigation of more complex vibrating systems. Further, the principles involved are common to many problems. Consider a mass m hanging at rest on the end of a vertical spring of original length l, as shown in Figure 3.7.1. The mass causes an elongation L of the spring in the downward (positive) direction. In this static situation there are two forces acting at the point where the mass is attached to the spring; see Figure 3.7.2. The gravitational force, or weight of the mass, acts downward and has magnitude w = mg, where g is the acceleration due to gravity. There is also a force Fs , due to the spring, that acts upward. If we assume that the elongation L of the spring is small, the spring force is very nearly proportional to L; this is known as Hooke’s10 law. Thus we write Fs = −k L, where the constant of proportionality k is called the ......................................................................................................................................................................... 10 Robert Hooke (1635--1703) was an English scientist with wide-ranging interests. His most important book, Micrographia, was published in 1665 and described a variety of microscopical observations. Hooke first published his law of elastic behavior in 1676 as ceiiinosssttuv; in 1678 he gave the interpretation ut tensio sic vis, which means, roughly, “as the force so is the displacement.”

y1 ( t) = 1 + t

(see

Boyce 9131 Ch03 2

148

September 29, 2016

17:28

148

CHAPTER 3 Second-Order Linear Differential Equations

spring constant, and the minus sign is due to the fact that the spring force acts in the upward (negative) direction. Since the mass is in equilibrium, the two forces balance each other, which means that w + Fs = mg − k L = 0.

(2)

For a given weight w = mg, you can measure L and then use equation (2) to determine k. Note that k has the units of force per unit length.

l l+L+u L m m

u

FIGURE 3.7.1 A spring-mass system.

Fs = –kL

w = mg FIGURE 3.7.2 Force diagram for a spring-mass system.

In the corresponding dynamic problem, we are interested in studying the motion of the mass when it is acted on by an external force or is initially displaced. Let u( t) , measured positive in the downward direction, denote the displacement of the mass from its equilibrium position at time t; see Figure 3.7.1. Then u( t) is related to the forces acting on the mass through Newton’s law of motion mu ( t) = f ( t) ,

(3)

where u is the acceleration of the mass and f is the net force acting on the mass. Observe that both u and f are functions of time. In this dynamic problem there are now four separate forces that must be considered. 1. The weight w = mg of the mass always acts downward. 2. The spring force Fs is assumed to be proportional to the total elongation L + u of the spring and always acts to restore the spring to its natural position. If L + u > 0, then the spring is extended, and the spring force is directed upward. In this case Fs = −k( L + u) .

(4)

On the other hand, if L + u < 0, then the spring is compressed a distance |L + u|, and the spring force, which is now directed downward, is given by Fs = k|L +u|. However, when L +u < 0, it follows that |L +u| = −( L +u) , so Fs is again given by equation (4). Thus, regardless of the position of the mass, the force exerted by the spring is always expressed by equation (4). 3. The damping or resistive force Fd always acts in the direction opposite to the direction of motion of the mass. This force may arise from several sources: resistance from the air or other medium in which the mass moves, internal energy dissipation due to the extension or compression of the spring, friction between the mass and the guides (if any) that constrain its motion to one dimension, or a mechanical device (dashpot) that imparts a resistive force to the mass. In any case, we assume that the resistive force is proportional to the speed |du/dt| of the mass; this is usually referred to as viscous damping. If du/dt > 0, then u is increasing, so the mass is moving downward. Then Fd is directed

Boyce 9131 Ch03 2

September 29, 2016

17:28

149

3.7 Mechanical and Electrical Vibrations

upward and is given by Fd ( t) = −γ u ( t) ,

(5)

where γ is a positive constant of proportionality known as the damping constant. On the other hand, if du/dt < 0, then u is decreasing, the mass is moving upward, and Fd is directed downward. In this case, Fd = γ |u ( t) |; since |u ( t) | = −u ( t) , it follows that Fd ( t) is again given by equation (5). Thus, regardless of the direction of motion of the mass, the damping force is always expressed by equation (5). The damping force may be rather complicated, and the assumption that it is modeled adequately by equation (5) may be open to question. Some dashpots do behave as equation (5) states, and if the other sources of dissipation are small, it may be possible to neglect them altogether or to adjust the damping constant γ to approximate them. An important benefit of the assumption (5) is that it leads to a linear (rather than a nonlinear) differential equation. In turn, this means that a thorough analysis of the system is straightforward, as we will show in this section and in Section 3.8. 4. An applied external force F( t) is directed downward or upward as F( t) is positive or negative. This could be a force due to the motion of the mount to which the spring is attached, or it could be a force applied directly to the mass. Often the external force is periodic. Taking account of these forces, we can now rewrite Newton’s law (3) as mu ( t) = w + Fs ( t) + Fd ( t) + F( t) = mg − k( L + u( t) ) − γ u ( t) + F( t) .

(6)

Since mg − k L = 0 by equation (2), it follows that the equation of motion of the mass is mu ( t) + γ u ( t) + ku( t) = F( t) ,

(7)

where the constants m, γ , and k are positive. Note that equation (7) has the same form as equation (1), that is, it is a nonhomogeneous second-order linear differential equation with constant coefficients. It is important to understand that equation (7) is only an approximate equation for the displacement u( t) . In particular, both equations (4) and (5) should be viewed as approximations for the spring force and the damping force, respectively. In our derivation we have also neglected the mass of the spring in comparison with the mass of the attached body. The complete formulation of the vibration problem requires that we specify two initial conditions, namely, the initial position u 0 and the initial velocity v 0 of the mass: u( 0) = u 0 ,

u ( 0) = v 0 .

(8)

It follows from Theorem 3.2.1 that these conditions give a mathematical problem that has a unique solution for any values of the constants u 0 and v 0 . This is consistent with our physical intuition that if the mass is set in motion with a given initial displacement and velocity, then its position will be determined uniquely at all future times. The position of the mass is given (approximately) by the solution of the second-order linear differential equation (7) subject to the prescribed initial conditions (8).

EXAMPLE 1 A mass weighing 4 lb stretches a spring 2 in. Suppose that the mass is given an additional 6-in displacement in the positive direction and then released. The mass is in a medium that exerts a viscous resistance of 6 lb when the mass has a velocity of 3 ft/s. Under the assumptions discussed in this section, formulate the initial value problem that governs the motion of the mass. Solution: The required initial value problem consists of the differential equation (7) and initial conditions (8), so our task is to determine the various constants that appear in these equations. The first step is to

▼

149

Boyce 9131 Ch03 2

150

September 29, 2016

17:28

150

CHAPTER 3 Second-Order Linear Differential Equations

▼ choose the units of measurement. Based on the statement of the problem, it is natural to use the English rather than the metric system of units. The only time unit mentioned is the second, so we will measure t in seconds. On the other hand, both the foot and the inch appear in the statement as units of length. It is immaterial which one we use, but having made a choice, we must be consistent. To be definite, let us measure the displacement u in feet. Since nothing is said in the statement of the problem about an external force, we assume that F( t) = 0. To determine m, note that m=

w 1 lb · s2 4 lb = = . g 8 ft 32 ft/s2

The damping coefficient γ is determined from the statement that γ u is equal to 6 lb when u is 3 ft/s. Therefore, γ =

6 lb lb · s =2 . 3 ft/s ft

The spring constant k is found from the statement that the mass stretches the spring by 2 in or

1 ft. 6

Thus k=

4 lb lb = 24 . 1/6 ft ft

Consequently, differential equation (7) becomes 1 u + 2u + 24u = 0, 8 or u + 16u + 192u = 0.

(9)

The initial conditions are 1 (10) , u ( 0) = 0. 2 The second initial condition is implied by the word “released” in the statement of the problem, which we interpret to mean that the mass is set in motion with no initial velocity. u( 0) =

Undamped Free Vibrations. If there is no external force, then F( t) = 0 in equation (7). Let us also suppose that there is no damping so that γ = 0; this is an idealized configuration of the system, seldom (if ever) completely attainable in practice. However, if the actual damping is very small, then the assumption of no damping may yield satisfactory results over short to moderate time intervals. In this case the equation of motion (7) reduces to mu + ku = 0.

(11)

The characteristic equation for equation (11) is mr 2 + k = 0

and its roots are r = ±i

k/ m. Thus the general solution of equation (11) is u = A cos( ω 0 t) + B sin( ω 0 t) ,

(12)

where k (13) . m The arbitrary constants A and B can be determined if initial conditions of the form (8) are given. In discussing the solution of equation (11), it is convenient to rewrite equation (12) in the form 2

ω0 =

or

u = R cos( ω 0 t − δ ) ,

(14)

u = R cos δ cos ω 0 t + R sin δ sin ω 0 t .

(15)

Boyce 9131 Ch03 2

September 29, 2016

17:28

151

3.7 Mechanical and Electrical Vibrations

By comparing equation (15) with equation (12), we find that A, B, R, and δ are related by the equations A = R cos δ , Thus

B = R sin δ .

(16)

B (17) . A In calculating δ , we must take care to choose the correct quadrant; this can be done by checking the signs of cos δ and sin δ in equations (16). The graph of equation (14), or the equivalent equation (12), for a typical set of initial conditions is shown in Figure 3.7.3. The graph is a displaced cosine wave that describes a periodic, or simple harmonic, motion of the mass. The period of the motion is m 1/2 2π T = = 2π . (18) ω0 k The circular frequency ω 0 = k/ m, measured in radians per unit time, is called the natural frequency of the vibration. The maximum displacement R of the mass from equilibrium is the amplitude of the motion. The dimensionless parameter δ is called the phase, or phase angle, and measures the displacement of the wave from its normal position corresponding to δ = 0. R=

A2 + B 2 ,

tan δ =

u R R cos δ

δ

δ +π

δ + 2π

ω 0t

–R FIGURE 3.7.3 Simple harmonic motion; u = R cos( ω 0 t −δ ) . Note that the horizontal axis is labeled as ω 0 t.

Note that the motion described by equation (14) has a constant amplitude that does not diminish with time. This reflects the fact that, in the absence of damping, there is no way for the system to dissipate the energy imparted to it by the initial displacement and velocity. Further, for a given mass m and spring constant k, the system always vibrates at the same frequency ω 0 , regardless of the initial conditions. However, the initial conditions do help to determine the amplitude of the motion. Finally, observe from equation (18) that the period T increases as the mass m increases, so larger masses vibrate more slowly. On the other hand, the period T decreases as the spring constant k increases, which means that stiffer springs cause the system to vibrate more rapidly.

EXAMPLE 2 Suppose that a mass weighing 10 lb stretches a spring 2 in. If the mass is displaced an additional 2 in and is then set in motion with an initial upward velocity of 1 ft/s, determine the position of the mass at any later time. Also determine the period, amplitude, and phase of the motion. Solution: The spring constant is k = 10 lb/2 in = 60 lb/ft, and the mass is m = w/g = 10/32 lb·s2 /ft. Hence the equation of motion reduces to u + 192u = 0,

▼

(19)

151

Boyce 9131 Ch03 2

152

September 29, 2016

17:28

152

CHAPTER 3 Second-Order Linear Differential Equations

▼ and the general solution is

u = A cos( 8

3t) + B sin( 8

3t) .

The solution satisfying the initial conditions u( 0) = 1/6 ft and u ( 0) = −1 ft/s is u=

1 1 cos( 8 3t) − sin( 8 3t) . 6 8 3

(20)

The natural frequency is ω 0 = 8 3 ∼ = 13.856 rad/s, so the period is T = 2π/ω 0 ∼ = 0.453 s. The amplitude R and phase δ are found from equations (17). We have R2 =

1 1 19 + = , so R ∼ = 0.182ft. 36 192 576

The second of equations (17) yields tan δ = − 3/4. There are two solutions of this equation, one in the second quadrant and one in the fourth. In the present problem, cos δ > 0 and sin δ < 0, so δ is in the fourth quadrant. In fact,

3

δ = − arctan

4

∼ = − 0.40864 rad.

The graph of the solution (20) is shown in Figure 3.7.4. u u = 0.182 cos(8√3 t + 0.409) R

δ+π

δ

δ + 2π

δ + 4π

ω 0t

–R

FIGURE 3.7.4 An undamped free vibration: u + 192u = 0, u( 0) = 1/6, u ( 0) = −1. Note that the scale for the horizontal axis is ω 0 t.

Damped Free Vibrations. When the effects of damping are included, the differential equation governing the motion of the mass is mu + γ u + ku = 0.

(21)

We are especially interested in examining the effect of variations in the damping coefficient γ for given values of the mass m and spring constant k. The corresponding characteristic equation is mr 2 + γ r + k = 0, and its roots are r1 , r2 =

−γ ±

⎛ γ2

− 4km

2m

=

γ ⎝ −1 ± 2m

⎞

1−

4km ⎠ . γ2

(22)

Depending on the sign of γ 2 − 4km, the solution u has one of the following forms: γ 2 − 4km > 0, u = Aer1 t + Ber2 t ;

(23)

γ 2 − 4km = 0, u = ( A + Bt) e−γ t/( 2m) ;

(24)

1/2 1 4km − γ 2 γ 2 − 4km < 0, u = e−γ t/( 2m) ( A cos( μ t) + B sin( μ t) ) , μ = > 0. 2m (25)

Boyce 9131 Ch03 2

September 29, 2016

17:28

153

3.7 Mechanical and Electrical Vibrations

Since m, γ , and k are positive, γ 2 − 4km is always less than γ 2 . Hence, if γ 2 − 4km ≥ 0, then the values of r1 and r2 given by equation (22) are negative. If γ 2 − 4km < 0, then the values of r1 and r2 are complex, but with negative real part. Thus, in all cases, the solution u tends to zero as t → ∞; this occurs regardless of the values of the arbitrary constants A and B ---that is, regardless of the initial conditions. This confirms our intuitive expectation, namely, that damping gradually dissipates the energy initially imparted to the system, and consequently the motion dies out with increasing time. The most interesting case is the third one, which occurs when the damping is small. If we let A = R cos δ and B = R sin δ in equation (25), then we obtain u = Re−γ t/( 2m) cos( μ t − δ ) .

(26)

The displacement u lies between the curves u = ±Re−γ t/( 2m) ; hence it resembles a cosine wave whose amplitude decreases as t increases. A typical example is sketched in Figure 3.7.5. The motion is called a damped oscillation or a damped vibration. The amplitude factor R depends on m, γ , k, and the initial conditions.

u Re– γ t/(2m) R cos δ

δ

δ +π −Re–

δ + 2π

δ + 3π

μt

t/(2m)

FIGURE 3.7.5 Damped vibration; u = Re−γ t/2m cos( μ t −δ ) .

Note that the scale for the horizontal axis is μ t.

Although the motion is not periodic, the parameter μ determines the frequency with which the mass oscillates back and forth; consequently, μ is called the quasi-frequency. By comparing μ with the frequency ω 0 of undamped motion, we find that μ ( 4km − γ 2 ) 1/2 /( 2m) = = ω0 k/ m

γ2 1− 4km

1/2

γ2 ∼ . =1− 8km

(27)

The last approximation is valid when γ 2 /4km is small; we refer to this situation as “small damping.” Thus the effect of small damping is to reduce slightly the frequency of the oscillation. By analogy with equation (18), the quantity Td = 2π/μ is called the quasi-period of the motion. It is the time between successive maxima or successive minima of the position of the mass, or between successive passages of the mass through its equilibrium position while going in the same direction. The relation between Td and T is given by ω0 Td = = T μ

γ2 1− 4km

−1/2

γ2 ∼ , =1+ 8km

(28)

where again the last approximation is valid when γ 2 /4km is small. Thus small damping increases the quasi-period. Equations (27) and (28) reinforce the significance of the dimensionless ratio γ 2 /( 4km) . It is not the magnitude of γ alone that determines whether damping is large or small, but the magnitude of γ 2 compared to 4km. When γ 2 /( 4km) is small, then damping has a small effect on the quasi-frequency and quasi-period of the motion. On the other hand, if we want to study the detailed motion of the mass for all time, then we can never neglect the damping force, no matter how small.

153

Boyce 9131 Ch03 2

154

September 29, 2016

17:28

154

CHAPTER 3 Second-Order Linear Differential Equations

As γ 2 /( 4km) increases, the quasi-frequency μ√ decreases and the quasi-period Td increases. In fact, μ → 0 and Td → ∞ as γ → 2 km. As indicated by equations√(23), (24), and (25), the nature √ of the solution changes as γ passes through the value 2 km. The motion with γ = 2 km is said to be critically damped. For larger values of γ , √ γ > 2 km, the motion is said to be overdamped. In these cases, given by equations (24) and (23), respectively, the mass may pass through its equilibrium position at most once (see Figure 3.7.6) and then creep back to it. The mass does not oscillate about the equilibrium, as it does for small γ . Two typical examples of critically damped motion are shown in Figure 3.7.6, and the situation is discussed further in Problems 15 and 16. u 2 u(0) = 12 , u'(0) = u= 1

2

4

(

1 2

7 4

)

+ 2t e–t /2

t

8 10 u(0) = 12 , u'

6

u=

–1

(

1 2

)

3 2

–

7 4

t e– t/2

FIGURE 3.7.6 Critically damped motions: u + u + 0.25u = 0;

u = ( A + Bt) e−t/2 . The solid blue curve is the solution satisfying u( 0) = 1/2, u ( 0) = 7/4; the dashed green curve satisfies u( 0) = 1/2, u ( 0) = −7/4.

EXAMPLE 3 The motion of a certain spring-mass system is governed by the differential equation 1 u + u + u = 0, 8

(29)

where u is measured in feet and t in seconds. If u( 0) = 2 and u ( 0) = 0, determine the position of the mass at any time. Find the quasi-frequency and the quasi-period, as well as the time at which the mass first passes through its equilibrium position. Also find the time τ such that |u( t) | < 0.1 for all t > τ. Solution: The solution of equation (29) is u( t) = e−t/16

255 t 16

A cos

255 t 16

+ B sin

.

To satisfy the initial conditions, we must choose A = 2 and B = 2/ initial value problem is

u = e−t/16 2 cos

32

e−t/16 cos = 255

▼

255 t 16

2

+

255

255; hence the solution of the

sin

255 t 16

255 t −δ , 16

(30)

Boyce 9131 Ch03 2

September 29, 2016

17:28

155

3.7 Mechanical and Electrical Vibrations

▼ u u" + u u" + 0.125 u' + u = 0

2

1

10

20

30

40

50

t

–1

–2

FIGURE 3.7.7 Vibration with small damping (solid blue curve) and with no damping (dashed green curve). Both motions have the same initial conditions: u( 0) = 2, u ( 0) = 0.

where δ is in the first quadrant with tan δ = 1/ 255, so δ ∼ = 0.06254. The displacement of the mass as a function of time is shown in Figure 3.7.7. For purposes of comparison, we also show the motion if the damping term is neglected. The quasi-frequency is μ = 255/16 ∼ = 0.998, and the quasi-period is Td = 2π/μ ∼ = 6.295 s. These values differ only slightly from the corresponding values (1 and 2π , respectively) for the undamped oscillation. This is evident also from the graphs in Figure 3.7.7, which rise and fall almost together. The damping coefficient is small in this example: only one-sixteenth of the critical value, in fact. Nevertheless, the amplitude of the oscillation is reduced rather rapidly. u u = 0.1

0.1

u=

32 √255

e–t/16 cos

(

√255 16

)

t – 0.06254

0.05

τ 40

45

50

55

60

t

–0.05 u = – 0.1

–0.1

–0.15 FIGURE 3.7.8 Solution of Example 3 for 40 ≤ t ≤ 60; determination of the time τ after which |u( t) | < 0.1.

Figure 3.7.8 shows the graph of the solution for 40 ≤ t ≤ 60, together with the graphs of u = ±0.1. From the graph it appears that τ is about 47.5, and by a more precise calculation we find that τ ∼ = 47.5149 s. To find the time at which the mass first passes through its equilibrium position, we refer to equation (30) and set 255t/16 − δ equal to π/2, the smallest positive zero of the cosine function. Then, by solving for t, we obtain 16 t= 255

π +δ 2

∼ = 1.637 s.

155

Boyce 9131 Ch03 2

156

September 29, 2016

17:28

156

CHAPTER 3 Second-Order Linear Differential Equations

Resistance R

Capacitance C

Inductance L

I

Impressed voltage E(t) FIGURE 3.7.9 A simple electric circuit.

Electric Circuits. A second example of the occurrence of second-order linear differential equations with constant coefficients is their use as a model of the flow of electric current in the simple series circuit shown in Figure 3.7.9. The current I , measured in amperes (A), is a function of time t. The resistance R in ohms ( Ω ) , the capacitance C in farads (F), and the inductance L in henrys (H) are all positive and are assumed to be known constants. The impressed voltage E in volts (V) is a given function of time. Another physical quantity that enters the discussion is the total charge Q in coulombs (C) on the capacitor at time t. The relation between charge Q and current I is dQ I = (31) . dt The flow of current in the circuit is governed by Kirchhoff’s11 second law: In a closed circuit the impressed voltage is equal to the sum of the voltage drops in the rest of the circuit. According to the elementary laws of electricity, we know that The voltage drop across the resistor is R I . Q . C dI The voltage drop across the inductor is L . dt Hence, by Kirchhoff’s law, dI 1 L (32) + R I + Q = E( t) . dt C The units for voltage, resistance, current, charge, capacitance, inductance, and time are all related: The voltage drop across the capacitor is

1 volt = 1 ohm · 1 ampere = 1 coulomb/1 farad = 1 henry · 1 ampere/1 second. Substituting for I from equation (31), we obtain the differential equation 1 L Q + R Q + Q = E( t) C for the charge Q. The initial conditions are Q( t0 ) = Q 0 ,

Q ( t0 ) = I ( t 0 ) = I 0 .

(33)

(34)

Thus to know the charge at any time it is sufficient to know the charge on the capacitor and the current in the circuit at some initial time t0 . Alternatively, we can obtain a differential equation for the current I by differentiating equation (33) with respect to t, and then substituting for d Q/dt from equation (31). The result is 1 L I + R I + I = E ( t) , (35) C with the initial conditions I ( t0 ) = I 0 ,

I ( t0 ) = I0 .

(36)

......................................................................................................................................................................... 11 Gustav Kirchhoff (1824--1887) was a German physicist and professor at Breslau, Heidelberg, and Berlin. He formulated the basic laws of electric circuits about 1845 while still a student at Albertus University in his native Königsberg. In 1857 he discovered that an electric current in a resistanceless wire travels at the speed of light. He is also famous for fundamental work in electromagnetic absorption and emission and was one of the founders of spectroscopy.

Boyce 9131 Ch03 2

September 29, 2016

17:28

157

3.7 Mechanical and Electrical Vibrations

157

From equation (32) it follows that I0

=

E( t0 ) − R I0 −

Q0 C .

(37)

L Hence I0 is also determined by the initial charge and current, which are physically measurable quantities. The most important conclusion from this discussion is that the flow of current in the circuit is described by an initial value problem of precisely the same form as the one that describes the motion of a spring-mass system. This is a good example of the unifying role of mathematics: once you know how to solve second-order linear equations with constant coefficients, you can interpret the results in terms of mechanical vibrations, electric circuits, or any other physical situation that leads to the same problem.

Problems In each of Problems 1 and 2, determine ω 0 , R, and δ so as to write the given expression in the form u = R cos( ω 0 t − δ ) .

1. u = 3 cos( 2t) + 4 sin( 2t) 2. u = −2 cos( π t) − 3 sin( π t) 3. A mass of 100 g stretches a spring 5 cm. If the mass is set in motion from its equilibrium position with a downward velocity of 10 cm/s, and if there is no damping, determine the position u of the mass at any time t. When does the mass first return to its equilibrium position?

4. A mass weighing 3 lb stretches a spring 3 in. If the mass is

can be expressed as the sum u = v + w, where v satisfies the initial conditions v( t0 ) = u 0 , v ( t0 ) = 0, w satisfies the initial conditions w( t0 ) = 0, w ( t0 ) = u 0 , and both v and w satisfy the same differential equation as u. This is another instance of superposing solutions of simpler problems to obtain the solution of a more general problem.

11. a. Show that A cos ω 0 t + B sin ω 0 t can be written in the form r sin( ω 0 t − θ ) . Determine r and θ in terms of A and B. b. If R cos( ω 0 t − δ ) = r sin( ω 0 t − θ ) , determine the relationship among R, r , δ , and θ .

12. If a series circuit has a capacitor of C = 0.8 × 10−6 F and an

inductor of L = 0.2 H, find the resistance R so that the circuit is critically damped.

pushed upward, contracting the spring a distance of 1 in and then set in motion with a downward velocity of 2 ft/s, and if there is no damping, find the position u of the mass at any time t. Determine the frequency, period, amplitude, and phase of the motion.

13. Assume that the system described by the differential equation

G 5. A mass of 20 g stretches a spring 5 cm. Suppose that the mass is also attached to a viscous damper with a damping constant of 400 dyn·s/cm. If the mass is pulled down an additional 2 cm and then released, find its position u at any time t. Plot u versus t. Determine the quasi-frequency and the quasi-period. Determine the ratio of the quasiperiod to the period of the corresponding undamped motion. Also find the time τ such that |u( t) | < 0.05 cm for all t > τ .

14. Assume that the system described by the differential equation

6. A spring is stretched 10 cm by a force of 3 N. A mass of 2 kg is hung from the spring and is also attached to a viscous damper that exerts a force of 3 N when the velocity of the mass is 5 m/s. If the mass is pulled down 5 cm below its equilibrium position and given an initial downward velocity of 10 cm/s, determine its position u at any time t. Find the quasi-frequency μ and the ratio of μ to the natural frequency of the corresponding undamped motion. 7. A series circuit has a capacitor of 10−5 F, a resistor of 3×102 Ω , and an inductor of 0.2 H. The initial charge on the capacitor is 10−6 C and there is no initial current. Find the charge Q on the capacitor at any time t. 8. A vibrating system satisfies the equation u + γ u + u = 0. Find the value of the damping coefficient γ for which the quasi-period of the damped motion is 50% greater than the period of the corresponding undamped motion. 9. Show that the period of motion of an undamped vibration of a mass hanging from a vertical spring is 2π L/g, where L is the elongation of the spring due to the mass, and g is the acceleration due to gravity. 10. Show that the solution of the initial value problem mu + γ u + ku = 0,

u( t0 ) = u 0 , u ( t0 ) = u 0

mu + γ u + ku = 0 is either critically damped or overdamped. Show that the mass can pass through the equilibrium position at most once, regardless of the initial conditions. Hint: Determine all possible values of t for which u = 0.

mu +γ u +ku = 0 is critically damped and that the initial conditions are u( 0) = u 0 , u ( 0) = v 0 . If v 0 = 0, show that u → 0 as t → ∞ but that u is never zero. If u 0 is positive, determine a condition on v 0 that will ensure that the mass passes through its equilibrium position after it is released.

15. Logarithmic Decrement. a. For the damped oscillation described by equation (26), show that the time between successive maxima is Td = 2π/μ . b. Show that the ratio of the displacements at two successive maxima is given by exp( γ Td /( 2m) ) . Observe that this ratio does not depend on which pair of maxima is chosen. The natural logarithm of this ratio is called the logarithmic decrement and is denoted by Δ . c. Show that Δ = π γ /( mμ ) . Since m, μ , and Δ are quantities that can be measured easily for a mechanical system, this result provides a convenient and practical method for determining the damping constant of the system, which is more difficult to measure directly. In particular, for the motion of a vibrating mass in a viscous fluid, the damping constant depends on the viscosity of the fluid; for simple geometric shapes the form of this dependence is known, and the preceding relation allows the experimental determination of the viscosity. This is one of the most accurate ways of determining the viscosity of a gas at high pressure.

Boyce 9131 Ch03 2

158

September 29, 2016

17:28

158

CHAPTER 3 Second-Order Linear Differential Equations

16. Referring to Problem 15, find the logarithmic decrement of the system in Problem 5.

17. The position of a certain spring-mass system satisfies the initial value problem 3 u + ku = 0, 2

u( 0) = 2, u ( 0) = v.

If the period and amplitude of the resulting motion are observed to be π and 3, respectively, determine the values of k and v.

18. Consider the initial value problem mu + γ u + ku = 0,

u( 0) = u 0 , u ( 0) = v 0 .

Assume that γ 2 < 4km. a. Solve the initial value problem. b. Write the solution in the form u( t) = Re−γ t/( 2m) cos( μ t −δ ) . Determine R in terms of m, γ , k, u 0 , and v 0 . c. Investigate the dependence of R on the damping coefficient γ for fixed values of the other parameters.

b. Solve the given initial value problem. c. Using the solution in part b, determine the total energy in the system at any time t. Your result should confirm the principle of conservation of energy for this system.

23. Suppose that a mass m slides without friction on a horizontal surface. The mass is attached to a spring with spring constant k, as shown in Figure 3.7.10, and is also subject to viscous air resistance with coefficient γ . Show that the displacement u( t) of the mass from its equilibrium position satisfies equation (21). How does the derivation of the equation of motion in this case differ from the derivation given in the text? u(t) k m

FIGURE 3.7.10 A spring-mass system.

19. A cubic block of side l and mass density ρ per unit volume is floating in a fluid of mass density ρ 0 per unit volume, where ρ 0 > ρ. If the block is slightly depressed and then released, it oscillates in the vertical direction. Assuming that the viscous damping of the fluid and air can be neglected, derive the differential equation of motion and determine the period of the motion. Hint: Use Archimedes’12 principle: an object that is completely or partially submerged in a fluid is acted on by an upward (buoyant) force equal to the weight of the displaced fluid.

20. The position of a certain undamped spring-mass system satisfies the initial value problem u + 2u = 0,

u( 0) = 0, u ( 0) = 2.

a. Find the solution of this initial value problem. G b. Plot u versus t and u versus t on the same axes. G c. Plot u versus u; that is, plot u( t) and u ( t) parametrically with t as the parameter. This plot is known as a phase plot, and the uu -plane is called the phase plane. Observe that a closed curve in the phase plane corresponds to a periodic solution u( t) . What is the direction of motion on the phase plot as t increases?

21. The position of a certain spring-mass system satisfies the initial value problem 1 u + u + 2u = 0, u( 0) = 0, u ( 0) = 2. 4 a. Find the solution of this initial value problem. G b. Plot u versus t and u versus t on the same axes. G c. Plot u versus u in the phase plane (see Problem 20). Identify several corresponding points on the curves in parts b and c. What is the direction of motion on the phase plot as t increases?

22. In the absence of damping, the motion of a spring-mass system satisfies the initial value problem mu + ku = 0,

u( 0) = a, u ( 0) = b.

a. Show that the kinetic energy initially imparted to the mass is mb2 /2 and that the potential energy initially stored in the spring is ka 2 /2, so initially the total energy in the system is ( ka 2 + mb2 ) /2.

24. In the spring-mass system of Problem 23, suppose that the spring force is not given by Hooke’s law but instead satisfies the relation Fs = −( ku + u 3 ) , where k > 0 and is small but may be of either sign. The spring is called a hardening spring if > 0 and a softening spring if < 0. Why are these terms appropriate? a. Show that the displacement u( t) of the mass from its equilibrium position satisfies the differential equation mu + γ u + ku + u 3 = 0. Suppose that the initial conditions are u( 0) = 0,

u ( 0) = 1.

In the remainder of this problem, assume that m = 1, k = 1, and γ = 0. b. Find u( t) when = 0 and also determine the amplitude and period of the motion. G c. Let = 0.1. Plot a numerical approximation to the solution. Does the motion appear to be periodic? Estimate the amplitude and period. G d. Repeat part c for = 0.2 and = 0.3. G e. Plot your estimated values of the amplitude A and the period T versus . Describe the way in which A and T , respectively, depend on . G f. Repeat parts c, d, and e for negative values of . .............................................................................................................................. 12 Archimedes (287--212 BCE) was the foremost of the ancient Greek mathematicians. He lived in Syracuse on the island of Sicily. His most notable discoveries were in geometry, but he also made important contributions to hydrostatics and other branches of mechanics. His method of exhaustion is a precursor of the integral calculus developed by Newton and Leibniz almost two millennia later. He died at the hands of a Roman soldier during the Second Punic War.

Boyce 9131 Ch03 2

September 29, 2016

17:28

159

3.8 Forced Periodic Vibrations

3.8

Forced Periodic Vibrations

We will now investigate the situation in which a periodic external force is applied to a springmass system. The behavior of this simple system models that of many oscillatory systems with an external force due, for example, to a motor attached to the system. We will first consider the case in which damping is present and will look later at the idealized special case in which there is assumed to be no damping. Forced Vibrations with Damping. The algebraic calculations can be fairly complicated in this kind of problem, so we will begin with a relatively simple example.

EXAMPLE 1 Suppose that the motion of a certain spring-mass system satisfies the differential equation 5 u + u + u = 3 cos t 4

(1)

u ( 0) = 3.

(2)

and the initial conditions u( 0) = 2,

Find the solution of this initial value problem and describe the behavior of the solution for large t. Solution: The homogeneous equation corresponding to equation (1) has the characteristic equation 1 5 = 0 with roots r = − ± i. Thus a general solution u c ( t) of this homogeneous r2 + r + 4 2 equation is u c ( t) = c1 e−t/2 cos t + c2 e−t/2 sin t.

(3)

A particular solution of equation (1) has the form U ( t) = A cos t + B sin t, where A and B are found by substituting U ( t) for u in equation (1). We have U ( t) = −A sin t + B cos t and U ( t) = −A cos t − B sin t. Thus, from equation (1) we obtain

1 A+B 4

cos t +

−A +

1 B 4

sin t = 3 cos t.

Consequently, A and B must satisfy the equations 1 1 A + B = 3, −A + B = 0, 4 4 48 12 and B = . Therefore, the particular solution is with the result that A = 17 17 12 48 U ( t) = cos t + sin t, (4) 17 17 and the general solution of equation (1) is 48 12 cos t + sin t. (5) u = u c ( t) + U ( t) = c1 e−t/2 cos t + c2 e−t/2 sin t + 17 17 The remaining constants c1 and c2 are determined by the initial conditions (2). From equation (5), and its first derivative, we have u( 0) = c1 +

12 = 2, 17

1 48 u ( 0) = − c1 + c2 + = 3, 2 17

22 14 and c2 = . Thus we finally arrive at the solution of the given initial value problem 17 17 (1), (2), namely, 14 12 48 22 −t/2 cos t + e−t/2 sin t + e cos t + sin t. (6) u= 17 17 17 17 The graph of the solution (6) is shown by the green curve in Figure 3.8.1. so c1 =

▼

159

Boyce 9131 Ch03 2

160

September 29, 2016

17:28

160

CHAPTER 3 Second-Order Linear Differential Equations

▼

It is important to note that the solution consists of two distinct parts. The first two terms on the right-hand side of equation (6) contain the exponential factor e−t/2 ; as a result, they rapidly approach zero. It is customary to call these terms the transient solution. The remaining terms in equation (6) involve only sines and cosines, so they represent an oscillation that continues indefinitely. We refer to them as the steady-state solution. The dotted red and dashed blue curves in Figure 3.8.1 show the transient and the steady-state parts of the solution, respectively. The transient part comes from the solution of the homogeneous equation corresponding to equation (1) and is needed to satisfy the initial conditions. The steady-state solution is the particular solution of the full nonhomogeneous equation. After a fairly short time, the transient solution is vanishingly small and the full solution is essentially indistinguishable from the steady state.

u

Full solution

3

2

Steady-state solution Transient solution

1

4

8

12

16 t

–1

–2

–3 FIGURE 3.8.1 Solution of the initial value problem (1), (2): u + u + 5u/4 = 3 cos t, u( 0) = 2, u ( 0) = 3. The full solution (solid green) is the sum of the transient solution (dotted red) and steady-state solution (dashed blue).

The equation of motion of a general spring-mass system subject to an external force F( t) is equation (7) in Section 3.7: mu ( t) + γ u ( t) + ku( t) = F( t) ,

(7)

where m, γ , and k are the mass, damping coefficient, and spring constant of the spring-mass system. Suppose now that the external force is given by F0 cos( ω t) , where F0 and ω are positive constants representing the amplitude and frequency, respectively, of the force. Then equation (7) becomes mu + γ u + ku = F0 cos( ω t) .

(8)

Solutions of equation (8) behave very much like the solution in the preceding example. The general solution of equation (8) must have the form u = c1 u 1 ( t) + c2 u 2 ( t) + A cos( ω t) + B sin( ω t) = u c ( t) + U ( t) .

(9)

The first two terms on the right-hand side of equation (9) are the general solution u c ( t) of the homogeneous equation corresponding to equation (8), and the latter two terms are a particular solution U ( t) of the full nonhomogeneous equation. The coefficients A and B can be found, as usual, by substituting these terms into the differential equation (8), while the arbitrary constants c1 and c2 are available to satisfy initial conditions, if any are prescribed. The solutions u 1 ( t)

Boyce 9131 Ch03 2

September 29, 2016

17:28

161

3.8 Forced Periodic Vibrations

and u 2 ( t) of the homogeneous equation depend on the roots r1 and r2 of the characteristic equation mr 2 + γ r + k = 0. Since m, γ , and k are all positive, it follows that r1 and r2 either are real and negative or are complex conjugates with a negative real part. In either case, both u 1 ( t) and u 2 ( t) approach zero as t → ∞. Since u c ( t) dies out as t increases, it is called the transient solution. In many applications, it is of little importance and (depending on the value of γ ) may well be undetectable after only a few seconds. The remaining terms in equation (9)---namely, U ( t) = A cos( ω t) + B sin( ω t) ---do not die out as t increases but persist indefinitely, or as long as the external force is applied. They represent a steady oscillation with the same frequency as the external force and are called the steady-state solution or the forced response of the system. The transient solution enables us to satisfy whatever initial conditions may be imposed. With increasing time, the energy put into the system by the initial displacement and velocity is dissipated through the damping force, and the motion then becomes the response of the system to the external force. Without damping, the effect of the initial conditions would persist for all time. It is convenient to express U ( t) as a single trigonometric term rather than as a sum of two terms. Recall that we did this for other similar expressions in Section 3.7. Thus we write U ( t) = R cos( ω t − δ ) .

(10)

The amplitude R and phase δ depend directly on A and B and indirectly on the parameters in the differential equation (8). It is possible to show, by straightforward but somewhat lengthy algebraic computations, that R=

2

F0 , Δ

cos δ =

m( ω 0 − ω 2 ) Δ

,

and sin δ =

γω , Δ

(11)

where Δ =

2

2

m 2 ( ω 0 − ω 2 ) 2 + γ 2 ω 2 and ω 0 =

k . m

(12)

Recall that ω 0 is the natural frequency of the unforced system in the absence of damping. We now investigate how the amplitude R of the steady-state oscillation depends on the frequency ω of the external force. Substituting from equation (12) into the expression for R in equation (11) and executing some algebraic manipulations, we find that ⎞ ⎛ 2 2 2 −1/2 Rk ω ω γ2 ⎠ =⎝ 1− +Γ where Γ = . F0 ω0 ω0 mk

(13)

Observe that the quantity Rk/ F0 is the ratio of the amplitude R of the forced response to F0 / k, the static displacement of the spring produced by a force F0 . For low frequency excitation---that is, as ω → 0---it follows from equation (13) that Rk/ F0 → 1 or R → F0 / k. At the other extreme, for very high frequency excitation, equation (13) implies that R → 0 as ω → ∞. At an intermediate value of ω the amplitude may have a maximum. To find this maximum point, we can differentiate R with respect to ω and set the result equal to zero. In this way we find that the maximum amplitude occurs when ω = ω max , where 2 ω max

=

2 ω0

γ2 γ2 2 − = ω0 1 − . 2mk 2m 2

(14)

Note that ω max < ω 0 and that ω max is close to ω 0 when γ is small. The maximum value of R is F0 F0 γ2 ∼ Rmax = (15) 1+ , = γ ω0 8mk γ ω 0 1 − ( γ 2 /4mk)

161

Boyce 9131 Ch03 2

162

September 29, 2016

17:28

162

CHAPTER 3 Second-Order Linear Differential Equations

where the last expression is an approximation that is valid when γ is small (see Problem 5). If γ2 > 2, then ω max as given by equation (14) is imaginary; in this case the maximum value mk of R occurs for ω = 0, and R is a monotone decreasing function of ω . Recall that critical γ2 damping occurs when = 4. mk F0 For small γ it follows from equation (15) that Rmax ∼ . Thus, for lightly damped = γ ω0 systems, the amplitude R of the forced response when ω is near ω 0 is quite large even for relatively small external forces, and the smaller the value of γ , the more pronounced is this effect. This phenomenon is known as resonance, and it is often an important design consideration. Resonance can be either good or bad, depending on the circumstances. It must be taken very seriously in the design of structures, such as buildings and bridges, where it can produce instabilities that might lead to the catastrophic failure of the structure. On the other hand, resonance can be put to good use in the design of instruments, such as seismographs, that are intended to detect weak periodic incoming signals. Rk/F0 10

Γ→0

8 Γ = 0.015625 6

4

Γ = 0.1 Γ=2

Γ = 0.5

2

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2 ω /ω 0

FIGURE 3.8.2 Forced vibration with damping: amplitude of steady-state response versus frequency of driving force for several values of the dimensionless damping parameter Γ = γ 2 / mk.

Figure 3.8.2 contains some representative graphs of

Rk versus F0

ω ω0

for several values of

γ2 . We refer to Γ as a damping parameter, as the following examples will explain. mk The graph corresponding to Γ = 0.015625 is included because this is the value of Γ that occurs in Example 2 below. Note particularly the sharp peak in the curve corresponding to ω Γ = 0.015625 near = 1. The limiting case as Γ → 0 is also shown. It follows from ω0 F0 as γ → 0 and hence equation (13), or from equations (11) and (12), that R → 2 m ω 0 − ω 2 Rk is asymptotic to the vertical line ω = ω 0 , as shown in the figure. As the damping in the F0 system increases, the peak response gradually diminishes. Figure 3.8.2 also illustrates the usefulness of dimensionless variables. You can easily Rk ω verify that each of the quantities , , and Γ is dimensionless (see Problem 9d). The F0 ω 0 importance of this observation is that the number of significant parameters in the problem has been reduced to three rather than the five that appear in equation (8). Thus only one family of curves, of which a few are shown in Figure 3.8.2, is needed to describe the response-versusfrequency behavior of all systems governed by equation (8). The phase angle δ also depends in an interesting way on ω . For ω near zero, it follows from equations (11) and (12) that cos δ ∼ = 1 and sin δ ∼ = 0. Thus δ ∼ = 0, and the response is nearly in phase with the excitation, meaning that they rise and fall together and, in particular, assume their respective maxima nearly together and their respective minima nearly together. Γ =

Boyce 9131 Ch03 2

September 29, 2016

17:28

163

3.8 Forced Periodic Vibrations

For ω = ω 0 we find that cos δ = 0 and sin δ = 1, so δ = π/2. In this case the response lags behind the excitation by π/2; that is, the peaks of the response occur π/2 later than the peaks of the excitation, and similarly for the valleys. Finally, for ω very large, we have cos δ ∼ = −1 and sin δ ∼ = 0. Thus δ ∼ = π so that the response is nearly out of phase with the excitation; this means that the response is minimum when the excitation is maximum, and vice versa. Figure 3.8.3 shows the graphs of δ versus ω /ω 0 for several values of Γ. For small damping, the phase transition from near δ = 0 to near δ = π occurs rather abruptly, whereas for larger values of the damping parameter, the transition takes place more gradually. δ

π Γ = 0.015625 Γ = 0.1 Γ = 0.5

π 2

Γ=2

ω /ω 0 FIGURE 3.8.3 Forced vibration with damping: phase of steady-state response versus frequency of driving force for several values of the dimensionless damping parameter Γ = γ 2 / mk.

EXAMPLE 2 Consider the initial value problem 1 u + u + u = 3 cos( ω t) , 8

u( 0) = 2, u ( 0) = 0.

(16)

Show plots of the solution for different values of the forcing frequency ω , and compare them with corresponding plots of the forcing function. Solution: For this system we have ω 0 = 1 and Γ = 1/64 = 0.015625. Its unforced motion was discussed in Example 3 of Section 3.7, and Figure 3.7.7 shows the graph of the solution of the unforced problem. Figures 3.8.4, 3.8.5, and 3.8.6 show the solution of the forced problem (16) for ω = 0.3, ω = 1, and ω = 2, respectively. The graph of the corresponding forcing function is also shown in each figure. In this example the static displacement, F0 / k, is equal to 3. Figure 3.8.4 shows the low frequency case, ω /ω 0 = 0.3. After the initial transient response is substantially damped out, the remaining steady-state response is essentially in phase with the excitation, and the amplitude of the response is somewhat larger than the static displacement. To be specific, R ∼ = 0.041185. = 3.2939 and δ ∼ The resonant case, ω /ω 0 = 1, is shown in Figure 3.8.5. Here, the amplitude of the steady-state response is eight times the static displacement, and the figure also shows the predicted phase lag of π/2 relative to the external force. The case of comparatively high frequency excitation is shown in Figure 3.8.6. Observe that the amplitude of the steady forced response is approximately one-third the static displacement and that the phase difference between the excitation and the response is approximately π . More precisely, we find that R ∼ = 0.99655 and that δ ∼ = 3.0585.

▼

163

Boyce 9131 Ch03 2

164

September 29, 2016

17:28

164

CHAPTER 3 Second-Order Linear Differential Equations

▼

u 3 2 1 10

20

30

40

50

60

70

80 t

–1 –2 –3 Solution

Forcing function

FIGURE 3.8.4 A forced vibration with damping; the solution (solid blue)

of equation (16) with ω = 0.3: u + 18 u + u = 3 cos( 0.3t) , u( 0) = 2, u ( 0) = 0. The dashed red curve is the external force: F( t) = 3 cos( 0.3t) . u 20

10

10

20

30

40

50

60 t

–10

–20 Forcing function

Solution

FIGURE 3.8.5 A forced vibration with damping; the solution (solid blue)

of equation (16) with ω = 1: u + 18 u + u = 3 cos t, u( 0) = 2, u ( 0) = 0. The dashed red curve is the external force: F( t) = 3 cos t. u 3 2 1

10

20

30

40

50 t

–1 –2 –3 Forcing function

Solution

FIGURE 3.8.6 A forced vibration with damping; the solution (solid blue)

of equation (16) with ω = 2: u + 18 u + u = 3 cos( 2t) , u( 0) = 2, u ( 0) = 0. The dashed red curve is the external force: F( t) = 3 cos( 2t) .

Boyce 9131 Ch03 2

September 29, 2016

17:28

165

3.8 Forced Periodic Vibrations

Forced Vibrations Without Damping. We now assume that γ = 0 in equation (8), thereby obtaining the equation of motion of an undamped forced oscillator, mu + ku = F0 cos( ω t) .

(17)

The form of the general solution of equation (17) is different, depending on whether the forcing frequency ω is different from or equal to the natural frequency ω 0 = k/ m of the unforced system. First consider the case ω = ω 0 ; then the general solution of equation (17) is

F0 u = c1 cos ω 0 t + c2 sin ω 0 t + (18) cos( ω t) . 2 m( ω 0 − ω 2 ) The constants c1 and c2 are determined by the initial conditions. The resulting motion is, in general, the sum of two periodic motions of different frequencies (ω 0 and ω ) and different amplitudes as well. It is particularly interesting to suppose that the mass is initially at rest so that the initial conditions are u( 0) = 0 and u ( 0) = 0. Then the energy driving the system comes entirely from the external force, with no contribution from the initial conditions. In this case it turns out that the constants c1 and c2 in equation (18) are given by F0 c1 = − (19) , c2 = 0, 2 m( ω 0 − ω 2 ) and the solution of equation (17) is u=

F0 2 m( ω 0

− ω 2)

cos( ω t) − cos( ω 0 t) .

(20)

This is the sum of two periodic functions of different periods but the same amplitude. Making 1 1 use of the trigonometric identities for cos( A± B) with A = ( ω 0 +ω ) t and B = ( ω 0 −ω ) t, 2 2 we can write equation (20) in the form 1

2F0 2 1 u= (21) ω 0 − ω 2 sin ω 0 − ω ) t sin ( ω 0 + ω ) t . m 2 2 If |ω 0 − ω | is small, then ω 0 + ω is much greater than |ω 0 − ω |. Consequently, 1 1 sin ( ω 0 + ω ) t is a rapidly oscillating function compared to sin ( ω 0 − ω ) t . Thus the 2 2 1 motion is a rapid oscillation with frequency ( ω 0 + ω ) but with a slowly varying sinusoidal 2 amplitude

2F0 1 . sin − ω t ω 0 2 2 m ω 0 − ω 2 This type of motion, possessing a periodic variation of amplitude, exhibits what is called a beat. For example, such a phenomenon occurs in acoustics when two tuning forks of nearly equal frequency are excited simultaneously. In this case the periodic variation of amplitude is quite apparent to the unaided ear. In electronics, the variation of the amplitude with time is called amplitude modulation.

EXAMPLE 3 Solve the initial value problem u + u =

1 cos( 0.8t) , 2

u( 0) = 0, u ( 0) = 0,

(22)

and plot the solution. Solution: In this case ω 0 = 1, ω = 0.8, and F0 =

1 , so from equation (21) the solution of the given problem is 2

u = 2.778 sin( 0.1t) sin( 0.9t) .

▼

(23)

165

Boyce 9131 Ch03 2

166

September 29, 2016

17:28

166

CHAPTER 3 Second-Order Linear Differential Equations

▼ A graph of this solution is shown in Figure 3.8.7. The amplitude variation has a slow frequency of

0.1 and a corresponding slow period of 2π/0.1 = 20π . Note that a half-period of 10π corresponds to a single cycle of increasing and then decreasing amplitude. The displacement of the spring-mass system oscillates with a relatively fast frequency of 0.9, which is only slightly less than the natural frequency ω 0 . Now imagine that the forcing frequency ω is increased, say, to ω = 0.9. Then the slow frequency is halved to 0.05, and the corresponding slow half-period is doubled to 20π . The multiplier 2.7778 also increases substantially, to 5.263. However, the fast frequency is only marginally increased, to 0.95. Can you visualize what happens as ω takes on values closer and closer to the natural frequency ω 0 = 1? u u = 2.778 sin (0.1t) u = 2.778 sin (0.1t) sin (0.9t)

3 2 1

10

20

30

40

50

60

t

–1 –2 u = –2.77778 sin (0.1t) –3

1 cos( 0.8t) , 2 u( 0) = 0, u ( 0) = 0 is u = 2.778 sin ( 0.1t) sin( 0.9t) . The dashed red curve is the external 1 force F( t) = cos( 0.8t) . 2 FIGURE 3.8.7 A beat; the solution (solid blue) of equation (22): u + u =

Now let us return to equation (17) and consider the case of resonance, where ω = ω 0 ; that is, the frequency of the forcing function is the same as the natural frequency of the system. Then the nonhomogeneous term F0 cos( ω t) is a solution of the homogeneous equation. In this case the solution of equation (17) is u = c1 cos ω 0 t + c2 sin ω 0 t +

F0 t sin( ω 0 t) . 2mω 0

(24)

Consider the following example.

EXAMPLE 4 Solve the initial value problem u + u =

1 cos t, 2

u( 0) = 0,

u ( 0) = 0,

and plot the graph of the solution. Solution: The general solution of the differential equation is u = c1 cos t + c2 sin t +

▼

t sin t, 4

(25)

Boyce 9131 Ch03 2

September 29, 2016

17:28

167

3.8 Forced Periodic Vibrations

167

▼ and the initial conditions require that c1 = c2 = 0. Thus the solution of the given initial value problem is u=

t sin t. 4

(26)

The graph of the solution is shown in Figure 3.8.8. u 10 u = t sin t 4 u= t 4

5

10

20

30

40

t

u = –t 4

–5

–10 FIGURE 3.8.8 Resonance; the solution (solid blue) of equation (25):

u + u =

1 t cos t, u( 0) = 0, u ( 0) = 0 is u = sin t. 2 4

Because of the term t sin( ω 0 t) , the solution (24) predicts that the motion will become unbounded as t → ∞ regardless of the values of c1 and c2 , and Figure 3.8.8 bears this out. Of course, in reality, unbounded oscillations do not occur, because the spring cannot stretch infinitely far. Moreover, as soon as u becomes large, the mathematical model on which equation (17) is based is no longer valid, since the assumption that the spring force depends linearly on the displacement requires that u be small. As we have seen, if damping is included in the model, the predicted motion remains bounded; however, the response to the input function F0 cos( ω t) may be quite large if the damping is small and ω is close to ω 0 .

Problems In each of Problems 1 through 3, write the given expression as a product of two trigonometric functions of different frequencies.

1. 2. 3. 4.

sin( 7t) − sin( 6t) cos( π t) + cos( 2π t) sin( 3t) + sin( 4t)

A mass of 5 kg stretches a spring 10 cm. The mass is acted on by an external force of 10 sin( t/2) N (newtons) and moves in a medium that imparts a viscous force of 2 N when the speed of the mass is 4 cm/s. If the mass is set in motion from its equilibrium position with an initial velocity of 3 cm/s, formulate the initial value problem describing the motion of the mass.

5. a.

Find the solution of the initial value problem in Problem 4. b. Identify the transient and steady-state parts of the solution. G c. Plot the graph of the steady-state solution.

N d. If the given external force is replaced by a force of 2 cos( ω t) of frequency ω , find the value of ω for which the amplitude of the forced response is maximum. N 6. A mass that weighs 8 lb stretches a spring 6 in. The system is acted on by an external force of 8 sin( 8t) lb. If the mass is pulled down 3 in and then released, determine the position of the mass at any time. Determine the first four times at which the velocity of the mass is zero.

7. A spring is stretched 6 in by a mass that weighs 8 lb. The mass is attached to a dashpot mechanism that has a damping constant of 1 lb·s/ft and is acted on by an external force of 4 cos( 2t) lb. 4 a. Determine the steady-state response of this system. b. If the given mass is replaced by a mass m, determine the value of m for which the amplitude of the steady-state response is maximum.

Boyce 9131 Ch03 2

168

September 29, 2016

17:28

168

CHAPTER 3 Second-Order Linear Differential Equations

8. A spring-mass system has a spring constant of 3 N/m. A mass of 2 kg is attached to the spring, and the motion takes place in a viscous fluid that offers a resistance numerically equal to the magnitude of the instantaneous velocity. If the system is driven by an external force of ( 3 cos( 3t) − 2 sin( 3t) ) N, determine the steady-state response. Express your answer in the form R cos( ω t − δ ) .

to 1? Note that the natural frequency of the unforced system is ω 0 = 1.

14. Consider the vibrating system described by the initial value problem

u( 0) = 1, u ( 0) = 1.

a. Find the solution for ω = 1. G b. Plot the solution u( t) versus t for ω = 0.7, ω = 0.8, and

9. In this problem we ask you to supply some of the details in the analysis of a forced damped oscillator. a. Derive equations (10), (11), and (12) for the steady-state solution of equation (8). b. Derive the expression in equation (13) for Rk/ F0 . 2 c. Show that ω max and Rmax are given by equations (14) and (15), respectively. d. Verify that Rk/ F0 , ω /ω 0 , and Γ = γ 2 /( mk) are all dimensionless quantities.

u + u = 3 cos( ω t) ,

ω = 0.9. Compare the results with those of Problem 13; that is, describe the effect of the nonzero initial conditions.

G

15. For the initial value problem in Problem 13, plot u versus

u for ω = 0.7, ω = 0.8, and ω = 0.9. (Recall that such a plot is called a phase plot.) Use a t interval that is long enough so that the phase plot appears as a closed curve. Mark your curve with arrows to show the direction in which it is traversed as t increases.

10. Find the velocity of the steady-state response given by

Problems 16 through 18 deal with the initial value problem

equation (10). Then show that the velocity is maximum when ω = ω 0 .

1 u + u + 4u = F( t) , u( 0) = 2, u ( 0) = 0. 8 In each of these problems: G a. Plot the given forcing function F( t) versus t, and also plot the solution u( t) versus t on the same set of axes. Use a t interval that is long enough so the initial transients are substantially eliminated. Observe the relation between the amplitude and phase of the forcing term and the amplitude and phase of the response. Note that ω 0 = k/ m = 2. G b. Draw the phase plot of the solution; that is, plot u versus u.

11. Find the solution of the initial value problem u + u = F( t) , where F( t) =

⎧ ⎨ F0 t,

u( 0) = 0, u ( 0) = 0,

F0 ( 2π − t) , ⎩ 0,

0 ≤ t ≤ π, π < t ≤ 2π, 2π < t.

Hint: Treat each time interval separately, and match the solutions in the different intervals by requiring u and u to be continuous functions of t. −6

12. A series circuit has a capacitor of 0.25 × 10 F, a resistor of 5×103 Ω , and an inductor of 1 H. The initial charge on the capacitor is zero. If a 12 V battery is connected to the circuit and the circuit is closed at t = 0, determine the charge on the capacitor at t = 0.001 s, at t = 0.01 s, and at any time t. Also determine the limiting charge as t → ∞. N

N

13. Consider the forced but undamped system described by the initial value problem u + u = 3 cos( ω t) ,

u( 0) = 0, u ( 0) = 0.

a. Find the solution u( t) for ω = 1. G b. Plot the solution u( t) versus t for ω = 0.7, ω = 0.8, and

ω = 0.9. Describe how the response u( t) changes as ω varies in this interval. What happens as ω takes on values closer and closer

16. F( t) = 3 cos( t/4) 17. F( t) = 3 cos( 2t) 18. F( t) = 3 cos( 6t) G 19. A spring-mass system with a hardening spring (Problem 24 of Section 3.7) is acted on by a periodic external force. In the absence of damping, suppose that the displacement of the mass satisfies the initial value problem 1 u + u + u 3 = cos ω t, 5

u( 0) = 0, u ( 0) = 0.

a. Let ω = 1 and plot a computer-generated solution of the given problem. Does the system exhibit a beat?

b. Plot the solution for several values of ω between 1/2 and 2. Describe how the solution changes as ω increases.

References Coddington, E. A., An Introduction to Ordinary Differential Equations (Englewood Cliffs, NJ: Prentice-Hall, 1961; New York: Dover, 1989). There are many books on mechanical vibrations and electric circuits. One that deals with both is Close, C. M., and Frederick, D. K., Modeling and Analysis of Dynamic Systems (3rd ed.) (New York: Wiley, 2001).

A classic book on mechanical vibrations is Den Hartog, J. P., Mechanical Vibrations (4th ed.) (New York: McGraw-Hill, 1956; New York: Dover, 1985). An intermediate-level book is Thomson, W. T., Theory of Vibrations with Applications (5th ed.) (Englewood Cliffs, NJ: Prentice-Hall, 1997). An elementary book on electric circuits is Bobrow, L. S., Elementary Linear Circuit Analysis (New York: Oxford University Press, 1996).

Boyce 9131 Ch04 2

September 29, 2016

17:26

169

CHAPTER 4 Higher-Order Linear Differential Equations The theoretical structure and methods of solution developed in the preceding chapter for second-order linear equations extend directly to linear equations of third and higher order. In this chapter we briefly review this generalization, taking particular note of those instances where new phenomena may appear, because of the greater variety of situations that can occur for equations of higher order.

General Theory of nth Order Linear Differential Equations 4.1

An n th order linear differential equation is an equation of the form P0 ( t)

dn y d n−1 y dy + P ( t) + · · · + Pn−1 ( t) + Pn ( t) y = G( t) . 1 n n−1 dt dt dt

(1)

We assume that the functions P0 , . . . , Pn , and G are continuous real-valued functions on some interval I : α < t < β , and that P0 is nowhere zero in this interval. Then, dividing equation (1) by P0 ( t) , we obtain L[y] =

dn y d n−1 y dy + pn ( t) y = g( t) . + p1 ( t) n−1 + · · · + pn−1 ( t) n dt dt dt

(2)

The linear differential operator L of order n defined by equation (2) is similar to the secondorder operator introduced in Chapter 3. The mathematical theory associated with equation (2) is completely analogous to that for the second-order linear equation; for this reason we simply state the results for the n th order problem. The proofs of most of the results are also similar to those for the second-order equation and are usually left as exercises. Since equation (2) involves the n th derivative of y with respect to t, it will, so to speak, require n integrations to solve equation (2). Each of these integrations introduces an arbitrary constant. Hence we expect that to obtain a unique solution it is necessary to specify n initial conditions y( t0 ) = y0 , y ( t0 ) = y0 , . . . , y ( n−1) ( t0 ) = y0

( n−1)

,

(3)

( n−1)

where t0 may be any point in the interval I and y0 , y0 , . . . , y0 are any prescribed real constants. The following theorem, which is similar to Theorem 3.2.1, guarantees that the initial value problem (2), (3) has a solution and that it is unique.

Theorem 4.1.1 If the functions p1 , p2 , . . . , pn , and g are continuous on the open interval I , then there exists exactly one solution y = φ ( t) of the differential equation (2) that also satisfies the initial conditions (3), where t0 is any point in I . This solution exists throughout the interval I . 169

Boyce 9131 Ch04 2

170

September 29, 2016

17:26

170

CHAPTER 4 Higher-Order Linear Differential Equations

We will not give a proof of this theorem here. However, if the coefficients p1 , . . . , pn are constants, then we can construct the solution of the initial value problem (2), (3) much as in Chapter 3; see Sections 4.2 through 4.4. Even though we may find a solution in this case, we do not know that it is unique without the use of Theorem 4.1.1. A proof of the theorem can be found in Ince (Section 3.32) or Coddington (Chapter 6). The Homogeneous Equation. As in the corresponding second-order problem, we first discuss the homogeneous equation L[y] = y ( n) + p1 ( t) y ( n−1) + · · · + pn−1 ( t) y + pn ( t) y = 0.

(4)

If the functions y1 , y2 , . . . , yn are solutions of the differential equation (4), then it follows by direct computation that the linear combination y = c1 y1 ( t) + c2 y2 ( t) + · · · + cn yn ( t) ,

(5)

where c1 , . . . , cn are arbitrary constants, is also a solution of equation (4). It is then natural to ask whether every solution of equation (4) can be expressed as a linear combination of y1 , . . . , yn . This will be true if, regardless of the initial conditions (3) that are prescribed, it is possible to choose the constants c1 , . . . , cn so that the linear combination (5) satisfies the initial conditions. That is, for any choice of the point t0 in I , and for any ( n−1) choice of y0 , y0 , . . . , y0 , we must be able to determine c1 , . . . , cn so that the equations c1 y1 ( t0 ) + · · · + cn yn ( t0 ) = y0 c1 y1 ( t0 ) + · · · + cn yn ( t0 ) = y0 . . . ( n−1)

c1 y1

(6)

( n−1)

( t0 ) + · · · + cn yn( n−1) ( t0 ) = y0

are satisfied. The system (6) of n linear algebraic equations can be solved uniquely for the n constants c1 , . . . , cn , provided that the determinant of the coefficient matrix is not zero. On the other hand, if the determinant of the coefficient matrix is zero, then it is always possible to ( n−1) choose values of y0 , y0 , . . . , y0 so that equations (6) do not have a solution. Therefore a necessary and sufficient condition for the existence of a solution of equations (6) for arbitrary ( n−1) values of y0 , y0 , . . . , y0 is that the Wronskian y1 y 1 W [y1 , . . . , yn ] = .. . ( n−1) y 1

y2

···

y2 . . . ( n−1) y2

··· ···

yn . . . ( n−1) yn yn

(7)

is not zero at t = t0 . Since t0 can be any point in the interval I , it is necessary and sufficient that W [y1 , y2 , . . . , yn ] be nonzero at every point in the interval. Just as for the second-order linear equation, it can be shown that if y1 , y2 , . . . , yn are solutions of equation (4), then W [y1 , y2 , . . . , yn ] either is zero for every t in the interval I or else is never zero there; see Problem 15. Hence we have the following theorem.

Theorem 4.1.2 If the functions p1 , p2 , . . . , pn are continuous on the open interval I , if the functions y1 , y2 , . . . , yn are solutions of equation (4), and if W [y1 , y2 , . . . , yn ]( t) = 0 for at least one point in I , then every solution of equation (4) can be expressed as a linear combination of the solutions y1 , y2 , . . . , yn .

Solutions y1 , . . . , yn of equation (4) whose Wronskian is nonzero are said to form a fundamental set of solutions. The existence of a fundamental set of solutions can be demonstrated in the same way as for the second-order linear equation (see Theorem 3.2.5).

Boyce 9131 Ch04 2

September 29, 2016

17:26

171

4.1 General Theory of nth Order Linear Differential Equations

Since all solutions of the homogeneous n th order linear differential equation (4) are of the form (5), we use the term general solution to refer to an arbitrary linear combination of any fundamental set of solutions of equation (4). Linear Dependence and Independence. We now explore the relationship between fundamental sets of solutions and the concept of linear independence, a central idea in the study of linear algebra. The functions f 1 , f 2 , . . . , f n are said to be linearly dependent on an interval I if there exists a set of constants k1 , k2 , . . . , kn , not all zero, such that k1 f 1 ( t) + k2 f 2 ( t) + · · · + kn f n ( t) = 0

(8)

for all t in I . The functions f 1 , . . . , f n are said to be linearly independent on I if they are not linearly dependent there.

EXAMPLE 1 Determine whether the functions f 1 ( t) = 1, f 2 ( t) = t, and f 3 ( t) = t 2 are linearly independent or dependent on the interval I : −∞ < t < ∞. Solution: Form the linear combination k1 f 1 ( t) + k2 f 2 ( t) + k3 f 3 ( t) = k1 + k2 t + k3 t 2 , and set it equal to zero to obtain k1 + k2 t + k3 t 2 = 0.

(9)

If equation (9) is to hold for all t in I , then it must certainly be true at any three distinct points in I . Any three points will serve our purpose, but it is convenient to choose t = 0, t = 1, and t = −1. Evaluating equation (9) at each of these points, we obtain the system of equations k1

= 0,

k1 + k2 + k3 = 0,

(10)

k1 − k2 + k3 = 0. From the first of equations (10) we note that k1 = 0; then from the other two equations it follows that k2 = k3 = 0 as well. Therefore, there is no set of constants k1 , k2 , k3 , not all zero, for which equation (9) holds even at the three chosen points, much less throughout I . Thus the given functions are not linearly dependent on I , so they must be linearly independent. Indeed, they are linearly independent on any interval. This can be established just as in this example, possibly using a different set of three points.

EXAMPLE 2 Determine whether the functions f 1 ( t) = 1, f 2 ( t) = 2 + t, f 3 ( t) = 3 − t 2 , and f 4 ( t) = 4t + t 2 are linearly independent or dependent on any interval I . Solution: Form the linear combination k1 f 1 ( t) + k2 f 2 ( t) + k3 f 3 ( t) + k4 f 4 ( t) = k1 + k2 ( 2 + t) + k3 ( 3 − t 2 ) + k4 ( 4t + t 2 ) = ( k1 + 2k2 + 3k3 ) + ( k2 + 4k4 ) t + ( −k3 + k4 ) t 2 . (11)

▼

171

Boyce 9131 Ch04 2

172

September 29, 2016

17:26

172

CHAPTER 4 Higher-Order Linear Differential Equations

▼ For this expression to be zero throughout an interval, it is certainly sufficient to require that k1 + 2k2 + 3k3 = 0, k2 + 4k4 = 0, −k3 + k4 = 0. These three equations, with four unknowns, have many solutions. For instance, if k4 = 1, then k3 = 1, k2 = −4, and k1 = 5. If we use these values for the coefficients in equation (11), then these functions satisfy the linear relation 5 f 1 ( t) − 4 f 2 ( t) + f 3 ( t) + f 4 ( t) = 0 for each value of t. Thus the given functions are linearly dependent on every interval.

The concept of linear independence provides an alternative characterization of fundamental sets of solutions of the homogeneous equation (4). Suppose that the functions y1 , . . . , yn are solutions of equation (4) on an interval I , and consider the equation k1 y1 ( t) + · · · + kn yn ( t) = 0.

(12)

By differentiating equation (12) repeatedly, we obtain the additional n − 1 equations k1 y1 ( t) + · · · + kn yn ( t) = 0, . . . ( n−1) k1 y1 ( t) + · · · + kn yn( n−1) ( t) = 0.

(13)

The system consisting of equations (12) and (13) is a system of n linear algebraic equations for the n unknowns k1 , . . . , kn . The determinant of coefficients for this system is the Wronskian W [y1 , . . . , yn ]( t) of y1 , . . . , yn . This leads to the following theorem.

Theorem 4.1.3 If y1 ( t) , . . . , yn ( t) form a fundamental set of solutions of the homogeneous n th order linear differential equation (4) L[y] = y ( n) + p1 ( t) y ( n−1) + · · · + pn−1 ( t) y + pn ( t) y = 0 on an interval I , then y1 ( t) , . . . , yn ( t) are linearly independent on I . Conversely, if y1 ( t) , . . . , yn ( t) are linearly independent solutions of equation (4) on I , then they form a fundamental set of solutions on I .

To prove this theorem, first suppose that y1 ( t) , . . . , yn ( t) form a fundamental set of solutions of the homogeneous differential equation (4) on I . Then the Wronskian W [y1 , . . . , yn ]( t) = 0 for every t in I . Hence the system (12), (13) has only the solution k1 = · · · = kn = 0 for every t in I . Thus y1 ( t) , . . . , yn ( t) cannot be linearly dependent on I and must therefore be linearly independent there. To demonstrate the converse, let y1 ( t) , . . . , yn ( t) be linearly independent on I . To show that they form a fundamental set of solutions, we need to show that their Wronskian is never zero in I . Suppose that this is not true; then there is at least one point t0 where the Wronskian is zero. At this point the system (12), (13) has a nonzero solution; let us denote it by k1∗ , . . . , kn∗ . Now form the linear combination φ ( t) = k1∗ y1 ( t) + · · · + kn∗ yn ( t) .

(14)

Then y = φ ( t) satisfies the initial value problem L[y] = 0,

y( t0 ) = 0, y ( t0 ) = 0, . . . , y ( n−1) ( t0 ) = 0.

(15)

The function φ satisfies the differential equation because it is a linear combination of solutions; it satisfies the initial conditions because these are just the equations in the system (12), (13) evaluated at t0 . However, the function y( t) = 0 for all t in I is also a solution of this initial value problem, and by Theorem 4.1.1, the solution to the initial value problem (15) is unique. Thus φ ( t) = 0 for all t in I . Consequently, y1 ( t) , . . . , yn ( t) are linearly dependent on I , which is a contradiction. Hence the assumption that there is a point where

Boyce 9131 Ch04 2

September 29, 2016

17:26

173

4.1 General Theory of nth Order Linear Differential Equations

173

the Wronskian is zero is untenable. Therefore, the Wronskian is never zero on I , as was to be proved. Note that for a set of functions f 1 , . . . , f n that are not solutions of the homogeneous linear differential equation (4), the converse part of Theorem 4.1.3 is not necessarily true. They may be linearly independent on I even though the Wronskian is zero at some points, or even every point, but with different sets of constants k1 , . . . , kn at different points. See Problem 18 for an example. The Nonhomogeneous Equation. Now consider the nonhomogeneous equation (2) L[y] = y ( n) + p1 ( t) y ( n−1) + · · · + pn ( t) y = g( t) . If Y1 and Y2 are any two solutions of equation (2), then it follows immediately from the linearity of the operator L that L[Y1 − Y2 ]( t) = L[Y1 ]( t) − L[Y2 ]( t) = g( t) − g( t) = 0. Hence the difference of any two solutions of the nonhomogeneous equation (2) is a solution of the homogeneous differential equation (4). Since any solution of the homogeneous equation can be expressed as a linear combination of a fundamental set of solutions y1 , . . . , yn , it follows that any solution of the nonhomogeneous differential equation (2) can be written as y = c1 y1 ( t) + c2 y2 ( t) + · · · + cn yn ( t) + Y ( t) ,

(16)

where Y is some particular solution of the nonhomogeneous differential equation (2). The linear combination (16) is called the general solution of the nonhomogeneous equation (2). Thus the primary problem is to determine a fundamental set of solutions y1 , . . . , yn of the homogeneous n th order linear differential equation (4). If the coefficients are constants, this is a fairly simple problem; it is discussed in the next section. If the coefficients are not constants, it is usually necessary to use numerical methods such as those in Chapter 8 or series methods similar to those in Chapter 5. These tend to become more cumbersome as the order of the equation increases. To find a particular solution Y ( t) in equation (16), the methods of undetermined coefficients and variation of parameters are again available. They are discussed and illustrated in Sections 4.3 and 4.4, respectively. The method of reduction of order (Section 3.4) also applies to n th order linear differential equations. If y1 is one solution of equation (4), then the substitution y = v( t) y1 ( t) leads to a linear differential equation of order n − 1 for v (see Problem 19 for the case when n = 3). However, if n ≥ 3, the reduced equation is itself at least of second order, and only rarely will it be significantly simpler than the original equation. Thus, in practice, reduction of order is seldom useful for equations of higher than second order.

Problems In each of Problems 1 through 4, determine intervals in which solutions are sure to exist.

1. 2. 3. 4.

y ( 4) + 4y + 3y = t t ( t − 1) y ( 4) + et y + 4t 2 y = 0 ( x − 1) y ( 4) + ( x + 1) y + ( tan x) y = 0 ( x 2 − 4) y ( 6) + x 2 y + 9y = 0

In each of Problems 5 through 7, determine whether the given functions are linearly dependent or linearly independent. If they are linearly dependent, find a linear relation among them.

5. 6.

f 1 ( t) = 2t − 3, f 2 ( t) = t 2 + 1, f 3 ( t) = 2t 2 − t f 1 ( t) = 2t − 3, f 2 ( t) = 2t 2 + 1, f 3 ( t) = 3t 2 + t

f 1 ( t) = 2t − 3, f 2 ( t) = t 2 + 1, f 3 ( t) = 2t 2 − t, f 4 ( t) = t 2 + t + 1

7.

In each of Problems 8 through 11, verify that the given functions are solutions of the differential equation, and determine their Wronskian.

8. y ( 4) + y = 0; 1, t, cos t, sin t 9. y + 2y − y − 2y = 0; et , e−t , e−2t 10. x y − y = 0;

1, x, x 3

11. x 3 y + x 2 y − 2x y + 2y = 0; x, x 2 , 1/ x 12. a. Show that W [5, sin2 t, cos( 2t) ] = 0 for all t by directly evaluating the Wrosnkian.

b. Establish the same result without direct evaluation of the Wronskian.

Boyce 9131 Ch04 2

174

September 29, 2016

17:26

174

CHAPTER 4 Higher-Order Linear Differential Equations

13. Verify that the differential operator defined by L[y] = y

( n)

+ p1 ( t) y

c. Show that

+ · · · + pn ( t) y

( n−1)

W [y1 , y2 , y3 ]( t) = c exp −

is a linear differential operator. That is, show that

p1 ( t) dt .

It follows that W is either always zero or nowhere zero on I .

L[c1 y1 + c2 y2 ] = c1 L[y1 ] + c2 L[y2 ],

d. Generalize this argument to the n th order equation

where y1 and y2 are n-times-differentiable functions and c1 and c2 are arbitrary constants. Hence, show that if y1 , y2 , . . . , yn are solutions of L[y] = 0, then the linear combination c1 y1 + · · · + cn yn is also a solution of L[y] = 0.

y ( n) + p1 ( t) y ( n−1) + · · · + pn ( t) y = 0 with solutions y1 , . . . , yn . That is, establish Abel’s formula

W [y1 , . . . , yn ]( t) = c exp −

p1 ( t) dt

(17)

14. Let the linear differential operator L be defined by for this case.

L[y] = a0 y ( n) + a1 y ( n−1) + · · · + an y, where a0 , a1 , . . . , an are real constants. a. Find L[t n ]. b. Find L[er t ]. c. Determine four solutions of the equation y ( 4) −5y +4y = 0. Do you think the four solutions form a fundamental set of solutions? Why?

15. In this problem we show how to generalize Theorem 3.2.7 (Abel’s theorem) to higher-order equations. We first outline the procedure for the third-order equation y + p1 ( t) y + p2 ( t) y + p3 ( t) y = 0.

1

y2 y2 y2

y3 . y y3

on 0 < t < 1.

b. Show that f ( t) and g( t) are linearly dependent on −1 < t < 0. −1 < t < 1.

d. Show that W [ f, g]( t) is zero for all t in −1 < t < 1. e. Explain why the results in c and d do not contradict Theorem 4.1.3.

19. Show that if y1 is a solution of y + p1 ( t) y + p2 ( t) y + p3 ( t) y = 0,

3

Hint: The derivative of a 3-by-3 determinant is the sum of three 3-by-3 determinants obtained by differentiating the first, second, and third rows, respectively. b. Substitute for y1 , y2 , and y3 from the differential equation; multiply the first row by p3 , multiply the second row by p2 , and add these to the last row to obtain W = − p1 ( t) W.

16. y + 2y − y − 3y = 0 17. t y + 2y − y + t y = 0 18. Let f ( t) = t 2 |t| and g( t) = t 3 . a. Show that the functions f ( t) and g( t) are linearly dependent

c. Show that f ( t) and g( t) are linearly independent on

Let y1 , y2 , and y3 be solutions of this equation on an interval I . a. If W = W [y1 , y2 , y3 ], show that

y1 W = y1 y

In each of Problems 16 and 17, use Abel’s formula (17) to find the Wronskian of a fundamental set of solutions of the given differential equation.

then the substitution y = y1 ( t) v( t) leads to the following secondorder equation for v : y1 v + ( 3y1 + p1 y1 ) v + ( 3y1 + 2 p1 y1 + p2 y1 ) v = 0. In each of Problems 20 and 21, use the method of reduction of order (Problem 19) to solve the given differential equation.

20. ( 2 − t) y + ( 2t − 3) y − t y + y = 0, t < 2; y1 ( t) = et 21. t 2 ( t + 3) y − 3t ( t + 2) y + 6( 1 + t) y − 6y = 0, t > 0; y1 ( t) = t 2 , y2 ( t) = t 3

Homogeneous Differential Equations with Constant Coefficients 4.2

Consider the n th order linear homogeneous differential equation L[y] = a0 y ( n) + a1 y ( n−1) + · · · + an−1 y + an y = 0,

(1)

where a0 , a1 , . . . , an are real constants and a0 = 0. From our knowledge of second-order linear equations with constant coefficients, it is natural to anticipate that y = er t is a solution of equation (1) for suitable values of r . Indeed, L[er t ] = er t ( a0r n + a1r n−1 + · · · + an−1r + an ) = er t Z (r )

(2)

Boyce 9131 Ch04 2

September 29, 2016

17:26

175

4.2 Homogeneous Differential Equations with Constant Coefficients

for all r , where Z (r ) = a0r n + a1r n−1 + · · · + an−1r + an .

(3)

For those values of r for which Z (r ) = 0, it follows that L[er t ] = 0 and y = er t is a solution of equation (1). The polynomial Z (r ) is called the characteristic polynomial, and the equation Z (r ) = 0 is the characteristic equation of the differential equation (1). Since a0 = 0, we know that Z (r ) is a polynomial of degree n and therefore has n zeros,1 say, r1 , r2 , . . . , rn , some of which may be equal and some of which may be complex-valued. Hence we can write the characteristic polynomial in the form Z (r ) = a0 (r − r1 ) (r − r2 ) · · · (r − rn ) .

(4)

Real and Unequal Roots. If the roots of the characteristic equation are real and no two are equal, then we have n distinct solutions er1 t , er2 t , . . . , ern t of equation (1). If these functions are linearly independent, then the general solution of the homogeneous n th order linear differential equation (1) is y = c1 er1 t + c2 er2 t + · · · + cn ern t .

(5)

One way to establish the linear independence of er1 t , er2 t , . . . , ern t is to evaluate their Wronskian determinant; another way is outlined in Problem 30.

EXAMPLE 1 Find the general solution of y ( 4) + y − 7y − y + 6y = 0.

(6)

Also find the solution that satisfies the initial conditions y( 0) = 1,

y ( 0) = 0,

y ( 0) = −2,

y ( 0) = −1.

(7)

Plot its graph and determine the behavior of the solution as t → ∞. Solution: Assuming that y = er t , we must determine r by solving the polynomial equation r 4 + r 3 − 7r 2 − r + 6 = 0.

(8)

The roots of this equation are r1 = 1, r2 = −1, r3 = 2, and r4 = −3. Therefore, the general solution of differential equation (6) is y = c1 et + c2 e−t + c3 e2t + c4 e−3t .

(9)

The initial conditions (7) require that c1 , . . . , c4 satisfy the four equations c 1 + c2 + c 3 +

c4 =

1,

c1 − c2 + 2c3 − 3c4 =

0,

c1 + c2 + 4c3 + 9c4 = −2,

(10)

c1 − c2 + 8c3 − 27c4 = −1.

▼ ......................................................................................................................................................................... 1 An

important question in mathematics for more than 200 years was whether every polynomial equation has at least one root. The affirmative answer to this question, the fundamental theorem of algebra, was given by Carl Friedrich Gauss (1777--1855) in his doctoral dissertation in 1799, although his proof does not meet modern standards of rigor. Several other proofs have been discovered since, including three by Gauss himself. Today, students often meet the fundamental theorem of algebra in a first course on complex variables, where it can be established as a consequence of some of the basic properties of complex analytic functions.

175

Boyce 9131 Ch04 2

176

September 29, 2016

17:26

176

CHAPTER 4 Higher-Order Linear Differential Equations

▼ By solving this system of four linear algebraic equations, we find that c1 =

11 , 8

c2 =

5 , 12

2 c3 = − , 3

1 c4 = − . 8

Thus the solution of the initial value problem is y=

11 t 5 2 1 e + e−t − e2t − e−3t . 8 12 3 8

(11)

The graph of the solution is shown in Figure 4.2.1. Observe that the dominant term, as t → ∞, in 2 the solution is e2t . As a result, we conclude that the solution approaches −∞ as t → ∞. 3 y

1

0.5

1

t

–1

FIGURE 4.2.1 Solution of the initial value problem (6), (7):

y ( 4) + y − 7y − y + 6y = 0, y( 0) = 1, y ( 0) = 0, y ( 0) = −2, y ( 0) = −1.

As Example 1 illustrates, the procedure for solving an n th order linear differential equation with constant coefficients depends on finding the roots of a corresponding n th degree polynomial equation. If initial conditions are prescribed, then a system of n linear algebraic equations must be solved to determine the proper values of the constants c1 , . . . , cn . Each of these tasks becomes much more complicated as n increases, and we have omitted the detailed calculations in Example 1. Computer assistance can be very helpful in such problems. For third and fourth degree polynomials there are formulas,2 analogous to the formula for quadratic equations but more complicated, that give exact expressions for the roots. Root-finding algorithms are readily available on calculators and computers. Sometimes they are included in the differential equation solver, so that the process of factoring the characteristic polynomial is hidden and the solution of the differential equation is produced automatically. If you are faced with the need to factor the characteristic polynomial by hand, here is one result that is sometimes helpful. Suppose that the polynomial a0r n + a1r n−1 + · · · + an−1r + an = 0

(12)

has integer coefficients. If r = p/q is a rational root, where p and q have no common factors, then p must be a factor of an , and q must be a factor of a0 . For example, in equation (8) the factors of a0 are ±1 and the factors of an are ±1, ±2, ±3, and ±6. Thus the only possible rational roots of this equation are ±1, ±2, ±3, and ±6. By testing these possible roots, we find that 1, −1, 2, and −3 are actual roots. In this case there are no other roots, since the polynomial ......................................................................................................................................................................... 2 The method for solving the cubic equation was apparently discovered by Scipione dal Ferro (1465--1526) about 1500, although it was first published in 1545 by Girolamo Cardano (1501--1576) in his Ars Magna. This book also contains a method for solving quartic equations that Cardano attributes to his pupil Ludovico Ferrari (1522--1565). The question of whether analogous formulas exist for the roots of higher degree equations remained open for more than two centuries, until 1826, when Niels Abel showed that no general solution formulas can exist for polynomial equations of degree five or higher. A more general theory was developed by Evariste Galois (1811--1832) in 1831, but unfortunately it did not become widely known for several decades.

Boyce 9131 Ch04 2

September 29, 2016

17:26

177

4.2 Homogeneous Differential Equations with Constant Coefficients

is of fourth degree. If some of the roots are irrational or complex, as is usually the case, then this process will not find them, but at least the degree of the polynomial can be reduced by dividing the polynomial by the factors corresponding to the rational roots. If the roots of the characteristic equation are real and different, we have seen that the general solution (5) is simply a sum of exponential functions. For large values of t the solution is dominated by the term corresponding to the algebraically largest root. If this root is positive, then solutions become exponentially unbounded, approaching +∞ or −∞ depending on the sign of the coefficient of the dominant term in the solution. If the largest root is negative, then solutions tend exponentially to zero. Finally, if this root is zero, then solutions approach a nonzero constant as t becomes large. Of course, for certain initial conditions, the coefficient of the otherwise dominant term may be zero; then the nature of the solution for large t is determined by the next largest root. Complex Roots. Since the coefficients a0 , a1 , a2 , . . . , an are real numbers, if the characteristic equation has complex roots, they must occur in conjugate pairs, λ ± iμ . That is, when r = λ + iμ is a root of the characteristic equation, so is r- = λ − iμ . Provided that none of the roots is repeated, equation (5) still gives the form of the general solution of equation (1). However, just as for the second-order differential equation (Section 3.3), we can replace the complex-valued solutions e( λ+iμ ) t and e( λ−iμ ) t by the real-valued solutions eλt cos( μ t) ,

eλt sin( μ t)

(13)

obtained as the real and imaginary parts of e( λ+iμ ) t . Thus, even though some of the roots of the characteristic equation are complex, it is still possible to express the general solution of equation (1) as a linear combination of real-valued solutions.

EXAMPLE 2 Find the general solution of y ( 4) − y = 0.

(14)

Also find the solution that satisfies the initial conditions y( 0) =

7 , 2

y ( 0) = −4,

y ( 0) =

5 , 2

y ( 0) = −2

(15)

and draw its graph. Solution: Substituting er t for y, we find that the characteristic equation is r 4 − 1 = (r 2 − 1) (r 2 + 1) = 0. Therefore, the roots are r = 1, −1, i, and −i, and the general solution of equation (14) is y = c1 et + c2 e−t + c3 cos t + c4 sin t. If we impose the initial conditions (15), we obtain (see Problem 26a) c1 = 0,

c2 = 3,

c3 =

1 , 2

c4 = −1;

thus the solution of the given initial value problem is y = 3e−t +

1 cos t − sin t. 2

(16)

The graph of this solution is shown in Figure 4.2.2. Observe that the initial conditions (15) cause the coefficient c1 of the exponentially growing term in the general solution to be zero. Therefore, this term is absent in the solution (16), which describes an exponential decay to a steady oscillation, as Figure 4.2.2 shows. However, if the initial conditions are changed slightly, then c1 is likely to be nonzero, and the nature of the solution changes enormously. For example, if the first three initial conditions remain the same, but the value of y ( 0)

▼

177

Boyce 9131 Ch04 2

178

September 29, 2016

17:26

178

CHAPTER 4 Higher-Order Linear Differential Equations

y

▼

2 2

4

6

8

10 12

14 t

–2

FIGURE 4.2.2 Solution of the initial value

problem (14), (15): y ( 4) − y = 0, y( 0) = y ( 0) = −4, y ( 0) = is changed from −2 to −

7 , 2

5 , y ( 0) = −2. 2

15 , then the solution of the initial value problem becomes (see Problem 8

26b) y=

17 1 t 95 −t 1 e + e + cos t − sin t. 32 32 2 16

(17)

The coefficients in the solution (17) differ only slightly from those in the solution (16), but the 1 exponentially growing term, even with the relatively small coefficient of , completely dominates 32 the solution by the time t is larger than about 4 or 5. This is clearly seen in Figure 4.2.3, which shows the graphs of the two solutions (16) and (17). y 8 6 4 2 2

4

6

t

FIGURE 4.2.3 Two

solutions of the homogeneous differential equation (14). The blue curve satsifies the initial conditions (15); the red curve satisfies the modified problem in which the last initial condition is changed to y ( 0) = −15/8.

Repeated Roots. If the roots of the characteristic equation are not distinct---that is, if some of the roots are repeated---then the solution (5) is clearly not the general solution of equation (1). Recall that if r1 is a repeated root for the second-order linear equation a0 y + a1 y + a2 y = 0, then two linearly independent solutions are er1 t and ter1 t . For an equation of order n, if a root of Z (r ) = 0, say r = r1 , has multiplicity s (where s ≤ n), then er1 t , ter1 t , t 2 er1 t , . . . , t s−1 er1 t

(18)

are corresponding solutions of equation (1). See Problem 31 for a proof of this statement, which is valid whether the repeated root is real or complex. Note that a complex root can be repeated only if the differential equation (1) is of order four or higher. If a complex root λ + iμ is repeated s times, the complex conjugate λ − iμ is also repeated s times. Corresponding to these 2s complex-valued solutions, we can find 2s real-valued solutions by noting that the real and imaginary parts of e( λ+iμ ) t , te( λ+iμ ) t , . . . , t s−1 e( λ+iμ ) t are also linearly independent solutions: eλt cos( μ t) , eλt sin( μ t) , teλt cos( μ t) , teλt sin( μ t) , . . . , t s−1 eλt cos( μ t) , t s−1 eλt sin( μ t) . Hence the general solution of equation (1) can always be expressed as a linear combination of n real-valued solutions. Consider the following example.

Boyce 9131 Ch04 2

September 29, 2016

17:26

179

4.2 Homogeneous Differential Equations with Constant Coefficients

EXAMPLE 3 Find the general solution of y ( 4) + 2y + y = 0.

(19)

Solution: The characteristic equation is r 4 + 2r 2 + 1 = (r 2 + 1) (r 2 + 1) = 0. Since r 2 + 1 = (r − i) (r + i) , it follows that the roots of the characteristic equation are r1 = i and r2 = −i. Each of these roots has multiplicity 2. Thus the general solution of equation (19) is y = c1 cos t + c2 sin t + c3 t cos t + c4 t sin t.

In determining the roots of the characteristic equation, it may be necessary to compute the cube roots, the fourth roots, or even higher roots of a (possibly complex) number. This can usually be done most conveniently by using Euler’s formula eit = cos t + i sin t and the algebraic laws given in Section 3.3. This is illustrated in the following example.

EXAMPLE 4 Find the general solution of y ( 4) + y = 0.

(20)

Solution: The characteristic equation is r 4 + 1 = 0. To solve the equation, we must compute the fourth roots of −1. Now −1, thought of as a complex number, is −1 + 0i. It has magnitude 1 and polar angle π : −1 = cos π + i sin π = eiπ . Moreover, because sin( x) and cos( x) both have period 2π , the angle is determined only up to a multiple of 2π : −1 = cos( π + 2mπ ) + i sin( π + 2mπ ) = ei( π +2mπ ) , where m is zero or any positive or negative integer. Now, by the properties of exponents,

( −1) 1/4 = ei( π +2mπ )

1/4

= ei( π/4+mπ/2) = cos

mπ π + 4 2

+ i sin

mπ π + 4 2

.

The four fourth roots of −1 are obtained by setting m = 0, 1, 2, and 3; they are −1 + i

1+i

,

2

2

,

−1 − i

2

1−i

.

,

2

It is easy to verify that, for any other value of

m, we obtain one of these four roots. For example, corresponding to m = 4, we obtain ( 1 + i) / 2. The general solution of the homogeneous fourth-order linear differential equation (20) is y=e

t/

√

2

c1 cos

t

2

+ c2 sin

t

2

+e

−t/

√

2

c3 cos

t

2

+ c4 sin

t

. 2 (21)

179

Boyce 9131 Ch04 2

180

September 29, 2016

17:26

180

CHAPTER 4 Higher-Order Linear Differential Equations

In conclusion, we note that the problem of finding all the roots of a polynomial equation may not be entirely straightforward, even with computer assistance. In particular, it may be difficult to determine whether two roots are equal or merely very close together. Recall that the form of the general solution is different in these two cases. If the constants a0 , a1 , . . . , an in equation (1) are complex numbers, the solution of equation (1) is still of the form (4). In this case, however, the roots of the characteristic equation are, in general, complex numbers, and it is no longer true that the complex conjugate of a root is also a root. The corresponding solutions are complex-valued.

Problems In each of Problems 1 through 4, express the given complex number in polar form R( cos θ + i sin θ ) = Reiθ .

1. 2. 3. 4.

1+i −1 +

3i

−3

3−i

In each of Problems 5 through 7, follow the procedure in Example 4 to determine the indicated roots of the given complex number. 1/3

5. 1 6. ( 1 − i) 1/2 7. ( 2( cos( π/3) + i sin( π/3) ) ) 1/2

8. 9. 10. 11. 12. 13. 14. 15. 16. 17. N N

y −y −y +y=0 y ( 4) − 4y + 4y = 0 y ( 6) + y = 0 y ( 6) − 3y ( 4) + 3y − y = 0 y ( 6) − y = 0 y

+ 8y

+ 16y = 0

y ( 4) + 2y + y = 0 y + 5y + 6y + 2y = 0

18. y ( 4) − 7y + 6y + 30y − 36y = 0 19. 12y ( 4) + 31y + 75y + 37y + 5y = 0

In each of Problems 20 through 25, find the solution of the given initial value problem, and plot its graph. How does the solution behave as t → ∞? G G

20. y + y = 0; y( 0) = 0, y ( 0) = 1, y ( 0) = 2 21. y ( 4) + y = 0; y( 0) = 0, y ( 0) = 0,

y ( 0) = −1, y ( 0) = 0 G

y( 1) = −1, y ( 1) = 2,

y ( 1) = 0, y ( 1) = 0

27. Show that the general solution of y ( 4) − y = 0 can be written as Determine the solution satisfying the initial conditions y( 0) = 0, y ( 0) = 0, y ( 0) = 1, y ( 0) = 1. Why is it convenient to use the solutions cosh t and sinh t rather than et and e−t ?

28. Consider the equation y ( 4) − y = 0. a. Use Abel’s formula (Problem 15d of Section 4.1) to find

consisting of two unit masses suspended from springs with spring constants 3 and 2, respectively. Assume that there is no damping in the system. a. Show that the displacements u 1 and u 2 of the masses from their respective equilibrium positions satisfy the equations u 1 + 5u 1 = 2u 2 ,

u 2 + 2u 2 = 2u 1 .

(22)

b. Solve the first of equations (22) for u 2 and substitute into the second equation, thereby obtaining the following fourth-order equation for u 1 : u 1 + 7u 1 + 6u 1 = 0. ( 4)

(23)

Find the general solution of equation (23).

22. y ( 4) − 4y + 4y = 0;

c. Suppose that the initial conditions are u 1 ( 0) = 1,

− y − 9y + 4y + 4y = 0; y ( 0) = 0, y ( 0) = −2, y ( 0) = 0 ( 4)

7 5 , y ( 0) = −4, y ( 0) = , y ( 0) = −2. 2 2 7 ( 4) N b. Find the solution to y − y = 0, y( 0) = , y ( 0) = −4, 2 5 15 y ( 0) = , y ( 0) = − . 2 8 Note: These are the initial value problems considered in Example 2. y ( 4) − y = 0, y( 0) =

29. Consider the spring-mass system, shown in Figure 4.2.4,

y ( 5) − 3y ( 4) + 3y − 3y + 2y = 0 ( 4)

a. Verify that y( t) = 3e−t + 12 cos t − sin t is the solution to

the Wronskian of a fundamental set of solutions of the given equation. b. Determine the Wronskian of the solutions et , e−t , cos t, and sin t. c. Determine the Wronskian of the solutions cosh t, sinh t, cos t, and sin t.

y − 3y + 3y − y = 0

( 8)

C

y = c1 cos t + c2 sin t + c3 cosh t + c4 sinh t.

In each of Problems 8 through 19, find the general solution of the given differential equation.

26.

y( 0) = −2,

G

23. 2y

G

24. 4y +y +5y = 0;

G

25. 6y +5y + y = 0; y( 0) = −2, y ( 0) = 2, y ( 0) = 0

y( 0) = 2, y ( 0) = 1, y ( 0) = −1

u 1 ( 0) = 0,

u 2 ( 0) = 2,

u 2 ( 0) = 0.

(24)

Use the first of equations (22) and the initial conditions (24) to ( 0) . Then show that the solution obtain values for u 1 ( 0) and u 1 of equation (23) that satisfies the four initial conditions on u 1 is u 1 ( t) = cos t. Show that the corresponding solution u 2 is u 2 ( t) = 2 cos t.

Boyce 9131 Ch04 2

September 29, 2016

17:26

181

4.3 The Method of Undetermined Coefficients

b. Multiply the result of part a by e−(r2 −r1 ) t and differentiate

d. Now suppose that the initial conditions are u 1 ( 0) = 0,

u 1 ( 0) = −2,

u 2 ( 0) = 1,

181

u 2 ( 0) = 0. (25)

with respect to t to obtain c3 (r3 − r2 ) (r3 − r1 ) e

Proceed as in part c to show

that the corresponding

solutions are 6 t and u 2 ( t) = cos 6t . u 1 ( t) = −2 cos

e. Observe that the solutions obtained in parts c and d describe two distinct modes of vibration. In the first, the frequency of the motion is 1, and the two masses move in phase, both moving up or down together; the second mass moves

twice as far as the first. The second motion has frequency 6, and the masses move out of phase with each other, one moving down while the other is moving up, and vice versa. In this mode the first mass moves twice as far as the second. For other initial conditions, not proportional to either of equations (24) or (25), the motion of the masses is a combination of these two modes.

(r3 −r2 ) t

+ · · · + cn (rn − r2 ) (rn − r1 ) e

(rn −r2 ) t

= 0.

c. Continue the procedure from parts a and b, eventually obtaining cn (rn − rn−1 ) · · · (rn − r1 ) e

(rn −rn−1 ) t

= 0.

Hence cn = 0, and therefore, c1 e

r1 t

+ · · · + cn−1 e

rn−1 t

= 0.

d. Repeat the preceding argument to show that cn−1 = 0. In a similar way it follows that cn−2 = · · · = c1 = 0. Thus the r t functions e 1 , . . . , ern t are linearly independent.

31. In this problem we indicate one way to show that if r = r1 is a root of multiplicity s of the characteristic polynomial Z (r ) , then r t r t r t e 1 , te 1 , . . . , t s−1 e 1 are solutions of equation (1). This problem th extends to n order equations the method for second-order equations given in Problem 17 of Section 3.4. We start from equation (2) in the text

k1 = 3

u1

m1 = 1

L[er t ] = er t Z (r )

and differentiate repeatedly with respect to r , setting r = r1 after each differentiation. a. Recall that if r1 is a root of multiplicity s, then Z (r ) = (r − r1 ) s q(r ) , where q(r ) is a polynomial of degree n − s and q(r1 ) = 0. Show that Z (r1 ) , Z (r1 ) , . . . , Z ( s−1) (r1 ) are all zero, but Z ( s) (r1 ) = 0. b. By differentiating equation (27) repeatedly with respect to r , show that

k2 = 2

u2

m2 = 1

FIGURE 4.2.4 A two-spring, two-mass system.

30. In this problem we outline one way to show that if r1 , . . . , rn r t

r1 t

+ · · · + cn ern t = 0,

−∞ < t < ∞

(26)

and show that all the constants are zero. a. Multiply equation (26) by e−r1 t and differentiate with respect to t, thereby obtaining c2 (r2 − r1 ) e

4.3

(r2 −r1 ) t

+ · · · + cn (rn − r1 ) e

(rn −r1 ) t

= 0.

∂ ∂ rt L[er t ] = L e = L[ter t ], ∂r ∂r . . .

are all real and different, then e 1 , . . . , ern t are linearly independent on −∞ < t < ∞. To do this, we consider the linear relation c1 e

(27)

∂ s−1 L[er t ] = L[t s−1 er t ]. ∂ r s−1

c. Show that er1 t , ter1 t , . . . , t s−1 er1 t equation (27).

The Method of Undetermined Coefficients

A particular solution Y of the nonhomogeneous n th order linear differential equation with constant coefficients L[y] = a0 y ( n) + a1 y ( n−1) + · · · + an−1 y + an y = g( t)

(1)

can be obtained by the method of undetermined coefficients, provided the nonhomogeneous term g( t) is of an appropriate form. Although the method of undetermined coefficients is not as general as the method of variation of parameters described in the next section, it is usually much easier to use when it is applicable.

are

solutions

of

Boyce 9131 Ch04 2

182

September 29, 2016

17:26

182

CHAPTER 4 Higher-Order Linear Differential Equations

Just as for the second-order linear differential equation, when the constant coefficient linear differential operator L is applied to a polynomial A0 t m + A1 t m−1 + · · · + Am , an exponential function eα t , a linear combination of sine and cosine functions a1 cos( β t) + a2 sin( β t) , the result is a polynomial, an exponential function, or a linear combination of sine and cosine functions, respectively. Hence, if g( t) is a sum of polynomials, exponentials, sines, and cosines, or products of such functions, we can expect that it is possible to find Y ( t) by choosing a suitable combination of polynomials, exponentials, and so forth, multiplied by a number of undetermined constants. The constants are then determined by substituting the assumed expression into the nonhomogeneous linear differential equation (1). The main difference in using this method for higher-order equations stems from the fact that roots of the characteristic polynomial equation may have multiplicity greater than 2. Consequently, terms proposed for the nonhomogeneous part of the solution may need to be multiplied by higher powers of t to make them different from terms in the solution of the corresponding homogeneous equation. The following examples illustrate this. In these examples we have omitted numerous straightforward algebraic steps, because our main goal is to show how to arrive at the correct form for the assumed solution.

EXAMPLE 1 Find the general solution of y − 3y + 3y − y = 4et .

(2)

Solution: The characteristic polynomial for the homogeneous equation corresponding to equation (2) is r 3 − 3r 2 + 3r − 1 = (r − 1) 3 , so the general solution of the homogeneous equation is yc ( t) = c1 et + c2 tet + c3 t 2 et .

(3)

To find a particular solution Y ( t) of equation (2), we start by assuming that Y ( t) = Aet . However, since et , tet , and t 2 et are all solutions of the homogeneous equation, we must multiply this initial choice by t 3 . Thus our final assumption is that Y ( t) = At 3 et , where A is an undetermined coefficient. To find the correct value for A, differentiate Y ( t) three times, substitute for y and its derivatives in equation (2), and collect terms in the resulting equation. In this way we obtain 6Aet = 4et . Thus A =

2 and the particular solution is 3 Y ( t) =

2 3 t t e. 3

(4)

The general solution of the nonhomogeneous differential equation (2) is the sum of yc ( t) from equation (3) and Y ( t) from equation (4): 2 y = c1 et + c2 tet + c3 t 2 et + t 3 et . 3

EXAMPLE 2 Find a particular solution of the equation y ( 4) + 2y + y = 3 sin t − 5 cos t.

▼

(5)

Boyce 9131 Ch04 2

September 29, 2016

17:26

183

4.3 The Method of Undetermined Coefficients

▼ Solution: The general solution of the homogeneous equation was found in Example 3 of Section 4.2; it is yc ( t) = c1 cos t + c2 sin t + c3 t cos t + c4 t sin t,

(6)

corresponding to the roots r = i, i, −i, and −i of the characteristic equation. Our initial assumption for a particular solution is Y ( t) = A sin t + B cos t, but we must multiply this choice by t 2 to make it different from all solutions of the homogeneous equation. Thus our final assumption is Y ( t) = At 2 sin t + Bt 2 cos t. Next, we differentiate Y ( t) four times, substitute into the differential equation (5), and collect terms, obtaining finally −8A sin t − 8B cos t = 3 sin t − 5 cos t. 3 5 Thus A = − , B = , and the particular solution of equation (4) is 8 8 3 5 Y ( t) = − t 2 sin t + t 2 cos t. 8 8

(7)

If g( t) is a sum of several terms, it may be easier in practice to compute separately the particular solution corresponding to each term in g( t) . In the same way as for secondorder differential equations, the particular solution of the complete problem is the sum of the particular solutions of the individual component problems. This is illustrated in the following example.

EXAMPLE 3 Find a particular solution of y − 4y = t + 3 cos t + e−2t .

(8)

Solution: First we solve the homogeneous equation. The characteristic equation is r 3 − 4r = 0, and the roots are r = 0, ±2; hence yc ( t) = c1 + c2 e2t + c3 e−2t . We can write a particular solution of equation (8) as the sum of particular solutions of the differential equations y − 4y = t,

y − 4y = 3 cos t,

y − 4y = e−2t .

Our initial choice for a particular solution Y1 ( t) of the first equation is A0 t + A1 , but a constant is a solution of the homogeneous equation, so we multiply by t. Thus Y1 ( t) = t ( A0 t + A1 ) . For the second equation we choose Y2 ( t) = B cos t + C sin t, and there is no need to modify this initial choice since sin t and cos t are not solutions of the homogeneous equation. Finally, for the third equation, since e−2t is a solution of the homogeneous equation, we assume that Y3 ( t) = Ete−2t . The constants are determined by substituting into the individual differential equations; they are 1 3 1 A0 = − , A1 = 0, B = 0, C = − , and E = . Hence a particular solution of equation (8) is 8 5 8 1 3 1 Y ( t) = − t 2 − sin t + te−2t . 8 5 8

(9)

183

Boyce 9131 Ch04 2

184

September 29, 2016

17:26

184

CHAPTER 4 Higher-Order Linear Differential Equations

You should keep in mind that the amount of algebra required to calculate the coefficients may be quite substantial for higher-order equations, especially if the nonhomogeneous term is even moderately complicated. A computer algebra system can be extremely helpful in executing these algebraic calculations. The method of undetermined coefficients can be used whenever it is possible to guess the correct form for Y ( t) . However, this is usually impossible for differential equations not having constant coefficients, or for nonhomogeneous terms other than the type described previously. For more complicated problems we can use the method of variation of parameters, which is discussed in the next section.

Problems In each of Problems 1 through 6, determine the general solution of the given differential equation.

1. 2. 3. 4. 5. 6.

y − y − y + y = 2e−t + 3 y ( 4) − y = 3t + cos t y + y + y + y = e−t + 4t y ( 4) − 4y = t 2 + et y ( 4) + 2y + y = 3 + cos 2t y ( 6) + y = t

In each of Problems 7 through 9, find the solution of the given initialvalue problem. Then plot a graph of the solution. G G

7. y + 4y = t; y( 0) = y ( 0) = 0, y ( 0) = 1 8. y ( 4) + 2y + y = 3t + 4; y( 0) = y ( 0) = 0,

In each of Problems 10 through 13, determine a suitable form for Y ( t) if the method of undetermined coefficients is to be used. Do not evaluate the constants. y − 2y + y = t 3 + 2et y − y = te−t + 2 cos t y ( 4) − y − y + y = t 2 + 4 + t sin t y ( 4) + 2y + 2y = 3et + 2te−t + e−t sin t

Consider the nonhomogeneous n th order linear differential equation a0 y

( n)

+ a1 y

( n−1)

+ · · · + an y = g( t) ,

(10)

where a0 , . . . , an are constants. Verify that if g( t) is of the form eα t ( b0 t m + · · · + bm ) , then the substitution y = eα t u( t) reduces equation (10) to the form k0 u ( n) + k1 u ( n−1) + · · · + kn u = b0 t m + · · · + bm ,

( D − a) ( D − b) f = ( D − b) ( D − a) f

Y ( t) of

y( 0) = 3, y ( 0) = 0, y ( 0) = −1, y ( 0) = 2

10. 11. 12. 13. 14.

obey the commutative law. That is, show that

16. Consider the problem of finding the form of a particular solution

9. y ( 4) + 2y + y + 8y − 12y = 12 sin t − e−t ;

15. Show that linear differential operators with constant coefficients

for any twice-differentiable function f and any constants a and b. The result extends at once to any finite number of factors.

y ( 0) = y ( 0) = 1 G

observation that exponential, polynomial, or sinusoidal terms (or sums and products of such terms) can be viewed as solutions of certain linear homogeneous differential equations with constant coefficients. It is d convenient to use the symbol D for . Then, for example, e−t is a dt solution of ( D + 1) y = 0; the differential operator D + 1 is said to annihilate, or to be an annihilator of, e−t . In the same way, D 2 + 4 is an annihilator of sin 2t or cos 2t, ( D − 3) 2 = D 2 − 6D + 9 is an annihilator of e3t or te3t , and so forth.

(11)

where k0 , . . . , kn are constants. Determine k0 and kn in terms of the a's and α . Thus the problem of determining a particular solution of the original equation is reduced to the simpler problem of determining a particular solution of an equation with constant coefficients and a polynomial for the nonhomogeneous term. Method of Annihilators. In Problems 15 through 17, we consider another way of arriving at the proper form of Y ( t) for use in the method of undetermined coefficients. The procedure is based on the

( D − 2) 3 ( D + 1) Y = 3e2t − te−t ,

(12)

where the left-hand side of the equation is written in a form corresponding to the factorization of the characteristic polynomial. a. Show that D − 2 and ( D + 1) 2 , respectively, are annihilators of the terms on the right-hand side of equation (12), and that the combined operator ( D − 2) ( D + 1) 2 annihilates both terms on the right-hand side of equation (12) simultaneously. b. Apply the operator ( D − 2) ( D + 1) 2 to equation (12) and use the result of Problem 15 to obtain ( D − 2) 4 ( D + 1) 3 Y = 0.

(13)

Thus Y is a solution of the homogeneous equation (13). By solving equation (13), show that Y ( t) = c1 e2t + c2 te2t + c3 t 2 e2t + c4 t 3 e2t + c5 e−t + c6 te−t + c7 t 2 e−t ,

(14)

where c1 , . . . , c7 are constants, as yet undetermined. c. Observe that e2t , te2t , t 2 e2t , and e−t are solutions of the homogeneous equation corresponding to equation (12); hence these terms are not useful in solving the nonhomogeneous equation. Therefore, choose c1 , c2 , c3 , and c5 to be zero in equation (14), so that Y ( t) = c4 t 3 e2t + c6 te−t + c7 t 2 e−t .

(15)

This is the form of the particular solution Y of equation (12). The values of the coefficients c4 , c6 , and c7 can be found by substituting from equation (15) in the differential equation (12).

Boyce 9131 Ch04 2

September 29, 2016

17:26

185

4.4 The Method of Variation of Parameters

Summary of the Method of Annihilators. Suppose that

which is a homogeneous equation of higher-order.

L( D) y = g( t) , (16) where L( D) is a linear differential operator with constant coefficients, and g( t) is a sum or product of exponential, polynomial, or sinusoidal terms. To find the form of a particular solution of equation (16), you can proceed as follows: a. Find a differential operator H ( D) with constant coefficients that annihilates g( t) ---that is, an operator such that H ( D) g( t) = 0. b. Apply H ( D) to equation (16), obtaining H ( D) L( D) y = 0,

4.4

c. Solve equation (17). d. Eliminate from the solution found in step c the terms that also appear in the solution of L( D) y = 0. The remaining terms constitute the correct form of a particular solution of equation (16).

17. Use the method of annihilators to find the form of a particular solution Y ( t) for each of the equations in Problems 10 through 13. Do not evaluate the coefficients.

(17)

The Method of Variation of Parameters

The method of variation of parameters for determining a particular solution of the nonhomogeneous n th order linear differential equation L[y] = y ( n) + p1 ( t) y ( n−1) + · · · + pn−1 ( t) y + pn ( t) y = g( t)

(1)

is a direct extension of the method for the second-order differential equation (see Section 3.6). As before, to use the method of variation of parameters, it is first necessary to solve the corresponding homogeneous differential equation. In general, this may be difficult unless the coefficients are constants. However, the method of variation of parameters is still more general than the method of undetermined coefficients in that it leads to an expression for the particular solution for any continuous function g, whereas the method of undetermined coefficients is restricted in practice to a limited class of functions g. Suppose then that we know a fundamental set of solutions y1 , y2 , . . . , yn of the homogeneous equation. Then the general solution of the homogeneous equation is yc ( t) = c1 y1 ( t) + c2 y2 ( t) + · · · + cn yn ( t) .

(2)

The method of variation of parameters for determining a particular solution of equation (1) rests on the possibility of determining n functions u 1 , u 2 , . . . , u n such that Y ( t) is of the form Y ( t) = u 1 ( t) y1 ( t) + u 2 ( t) y2 ( t) + · · · + u n ( t) yn ( t) .

(3)

Since we have n functions to determine, we will have to specify n conditions. One of these is clearly that Y satisfy equation (1). The other n − 1 conditions are chosen so as to make the calculations as simple as possible. Since we can hardly expect a simplification in determining Y if we must solve high order differential equations for u 1 , . . . , u n , it is natural to impose conditions to suppress the terms that lead to higher derivatives of u 1 , . . . , u n . From equation (3) we obtain Y = ( u 1 y1 + u 2 y2 + · · · + u n yn ) + ( u 1 y1 + u 2 y2 + · · · + u n yn ) ,

(4)

where we have omitted the independent variable t on which each function in equation (4) depends. Thus the first condition that we impose is that u 1 y1 + u 2 y2 + · · · + u n yn = 0.

(5)

It follows that the expression (4) for Y reduces to Y = u 1 y1 + u 2 y2 + · · · + u n yn .

(6)

We continue this process by calculating the successive derivatives Y , . . . , Y ( n−1) . After each differentiation we set equal to zero the sum of terms involving derivatives of u 1 , . . . , u n . In this way we obtain n − 2 further conditions similar to equation (5); that is, u 1 y1

( m)

+ u 2 y2

( m)

+ · · · + u n yn( m) = 0,

m = 1, 2, . . . , n − 2.

(7)

As a result of these conditions, it follows that the expressions for Y , . . . , Y ( n−1) reduce to ( m)

Y ( m) = u 1 y1

( m)

+ u 2 y2

+ · · · + u n yn( m) ,

185

m = 2, 3, . . . , n − 1.

(8)

Boyce 9131 Ch04 2

186

September 29, 2016

17:26

186

CHAPTER 4 Higher-Order Linear Differential Equations

Finally, we need to impose the condition that Y must be a solution of equation (1). By differentiating Y ( n−1) from equation (8), we obtain Y ( n) = ( u 1 y1 + · · · + u n yn( n) ) + ( u 1 y1 ( n)

( n−1)

+ · · · + u n yn( n−1) ) .

(9)

To satisfy the differential equation we substitute for Y and its derivatives in equation (1) from equations (3), (6), (8), and (9). Then we group the terms involving each of the functions y1 , . . . , yn and their derivatives. It then follows that most of the terms in the equation drop out because each of y1 , . . . , yn is a solution of equation (1) and therefore L[yi ] = 0, i = 1, 2, . . . , n. The remaining terms yield the relation u 1 y1

( n−1)

+ u 2 y2

( n−1)

+ · · · + u n yn( n−1) = g.

(10)

Equation (10), equation (5), and the n − 2 equations (7) provide n simultaneous linear nonhomogeneous algebraic equations for u 1 , u 2 , . . . , u n : y1 u 1 + y2 u 2 + · · · + yn u n = 0, y1 u 1 + y2 u 2 + · · · + yn u n = 0, y1 u 1 + y2 u 2 + · · · + yn u n = 0, . . . ( n−1) ( n−1) u 1 + · · · + yn u n = g. y1

(11)

The system (11) is a linear algebraic system for the unknown quantities u 1 , . . . , u n . By solving this system and then integrating the resulting expressions, you can obtain the coefficients u 1 , . . . , u n . A sufficient condition for the existence of a solution of the system of equations (11) is that the determinant of coefficients is nonzero for each value of t. However, the determinant of coefficients is precisely W [y1 , y2 , . . . , yn ], and it is nowhere zero since y1 , . . . , yn is a fundamental set of solutions of the homogeneous equation. Hence it is possible to determine u 1 , . . . , u n . Using Cramer’s3 rule, we can write the solution of the system of equations (11) in the form u m ( t) =

g( t) Wm ( t) , W ( t)

m = 1, 2, . . . , n.

(12)

Here W ( t) = W [y1 , y2 , . . . , yn ]( t) , and Wm is the determinant obtained from W by replacing the m th column by the column ( 0, 0, . . . , 0, 1) T . With this notation a particular solution of equation (1) is given by t n g( s) Wm ( s) Y ( t) = ym ( t) (13) ds, W ( s) t0 m=1

where t0 is arbitrary.

EXAMPLE 1 Given that y1 ( t) = et , y2 ( t) = tet , and y3 ( t) = e−t are solutions of the homogeneous equation corresponding to y − y − y + y = g( t) ,

(14)

determine a particular solution of equation (14) in terms of an integral.

▼ ......................................................................................................................................................................... 3 Cramer’s rule is credited to the Swiss mathematician Gabriel Cramer (1704--1752), professor at the Académie de Calvin in Geneva, who published it in a general form (but without proof) in 1750. For small systems the result had been known earlier.

Boyce 9131 Ch04 2

September 29, 2016

17:26

187

4.4 The Method of Variation of Parameters

▼ Solution: We use equation (13). First, we have

t e t t −t W ( t) = W [e , te , e ]( t) = et e t

−e−t . e−t e−t

tet ( t + 1) et ( t + 2) et

Factoring et from each of the first two columns and e−t from the third column, we obtain

1 W ( t) = et 1 1

1

t

1

−1.

t +1 t +2

Then, by subtracting the first row from the second and third rows, we have

1 W ( t) = et 0 0

1

t

0

1 −2. 2

Finally, evaluating the latter determinant by minors associated with the first column, we find that W ( t) = 4et . Next,

0 W1 ( t) = 0 1

e−t

tet

e−t

−e−t .

( t + 1) et ( t + 2) et

Using minors associated with the first column, we obtain

= −2t − 1. −e−t

tet W1 ( t) = ( t + 1) et In a similar way,

and

e t W2 ( t) = et t e e t W3 ( t) = et t e

e−t

e−t

t e 0 −e = − t e 1 e−t 0

tet ( t + 1) et ( t + 2) et

−t

0

Substituting these results in equation (13), we have

t

Y ( t) = et

=

1 4

t0

g( s) ( −1 − 2s) ds + tet 4es

t

t e

1

0 =

t

t0

=2 −e−t e−t

= e2t . t ( t + 1) e tet

et

g( s) ( 2) ds + e−t 4es

t

t0

g( s) e2s ds 4es

et−s ( −1 + 2( t − s) ) + e−( t−s) g( s) ds.

(15)

t0

Depending on the specific function g( t) , it may or may not be possible to evaluate the integrals in equation (15) in terms of elementary functions.

Although the procedure is straightforward, the algebraic computations involved in determining Y ( t) from equation (13) become more and more complicated as n increases. In some cases the calculations may be simplified to some extent by using Abel’s identity (Problem 15d of Section 4.1), W ( t) = W [y1 , . . . , yn ]( t) = c exp − p1 ( t) dt . The constant c can be determined by evaluating W at some convenient point.

187

Boyce 9131 Ch04 2

188

September 29, 2016

17:26

188

CHAPTER 4 Higher-Order Linear Differential Equations

Problems In each of Problems 1 through 4, use the method of variation of parameters to determine the general solution of the given differential equation. π π 1. y + y = tan t, − < t < 2 2 2. y − y = t

3. y − 2y − y + 2y = e4t 4. y − y + y − y = e−t sin t

In each of Problems 7 and 8, find the solution of the given initial-value problem. Then plot a graph of the solution.

7. y − y + y − y = sec t;

y ( 0) = 1 G

y

8. y − y = tan t; π 4

y( 0) = 2, y ( 0) = −1,

y

π 4

equation corresponding to x 3 y + x 2 y − 2x y + 2y = 2x 4 ,

x > 0,

determine a particular solution.

10. Find a formula involving integrals for a particular solution of the differential equation

In each of Problems 5 and 6, find the general solution of the given differential equation. Leave your answer in terms of one or more integrals. π π 5. y − y + y − y = sec t, − < t < 2 2 6. y − y = csc t, 0 < t < π

G

9. Given that x, x 2 , and 1/ x are solutions of the homogeneous

π 4

11. Find a formula involving integrals for a particular solution of the differential equation y ( 4) − y = g( t) . Hint: The functions sin t, cos t, sinh t, and cosh t form a fundamental set of solutions of the homogeneous equation.

12. Find a formula involving integrals for a particular solution of the differential equation y − 3y + 3y − y = g( t) .

= 2, y

y − y + y − y = g( t) .

= 1,

If g( t) = t −2 et , determine Y ( t) .

= −1

References Coddington, E. A., An Introduction to Ordinary Differential Equations (Englewood Cliffs, NJ: Prentice-Hall, 1961; New York: Dover, 1989). Coddington, E. A. and Carlson, R., Linear Ordinary Differential Equations (Philadelphia, PA: Society for Industrial and Applied Mathematics, 1997).

Ince, E. L., Ordinary Differential Equations (London: Longmans, Green, 1927; New York: Dover, 1956).

Boyce 9131 Ch05 2

September 29, 2016

17:30

189

CHAPTER 5 Series Solutions of Second-Order Linear Equations Finding the general solution of a linear differential equation depends on determining a fundamental set of solutions of the homogeneous equation. So far, we have given a systematic procedure for constructing fundamental solutions only when the equation has constant coefficients. To deal with the much larger class of equations that have variable coefficients, it is necessary to extend our search for solutions beyond the familiar elementary functions of calculus. The principal tool that we need is the representation of a given function by a power series. The basic idea is similar to that in the method of undetermined coefficients: we assume that the solutions of a given differential equation have power series expansions, and then we attempt to determine the coefficients so as to satisfy the differential equation.

5.1

Review of Power Series

In this chapter we discuss the use of power series to construct fundamental sets of solutions of second-order linear differential equations whose coefficients are functions of the independent variable. We begin by summarizing very briefly the pertinent results about power series that we need. Readers who are familiar with power series may go on to Section 5.2. Those who need more details than are presented here should consult a book on calculus. ∞ 1. A power series an ( x − x0 ) n is said to converge at a point x if n=0

lim m→∞

m

an ( x − x 0 ) n

n=0

exists for that x. The series certainly converges for x = x0 ; it may converge for all x, or it may converge for some values of x and not for others. ∞ 2. The power series an ( x − x0 ) n is said to converge absolutely at a point x if the n=0

associated power series

∞ n=0

|an ( x − x0 ) n | =

∞

|an ||x − x0 |n

n=0

converges. It can be shown that if the power series converges absolutely, then the power series also converges; however, the converse is not necessarily true. 3. One of the most useful tests for the absolute convergence of a power series is the ratio test: If an = 0, and if, for a fixed value of x, a ( x − x ) n+1 an+1 0 n+1 = |x − x0 |L , lim = |x − x | lim 0 n n→∞ an ( x − x 0 ) n→∞ an 189

Boyce 9131 Ch05 2

190

September 29, 2016

17:30

190

CHAPTER 5 Series Solutions of Second-Order Linear Equations

then the power series converges absolutely at that value of x if |x − x0 |L < 1 and diverges if |x − x0 |L > 1. If |x − x0 |L = 1, the ratio test is inconclusive.

EXAMPLE 1 For which values of x does the power series ∞

( −1) n+1 n( x − 2) n = ( x − 2) − 2( x − 2) 2 + 3( x − 2) 3 − · · ·

n=1

converge? Solution: We first test for absolute convergence using the ratio test. We have

( −1) n+2 ( n + 1) ( x − 2) n+1 = |x − 2| lim n + 1 = |x − 2|. n+1 n n ( −1) n( x − 2) n→∞ n→∞ lim

According to statement 3, the series converges absolutely for |x − 2| < 1, that is, for 1 < x < 3, and diverges for |x − 2| > 1. The values of x corresponding to |x − 2| = 1 are x = 1 and x = 3. The series diverges for each of these values of x since the nth term of the series does not approach zero as n → ∞. This power series converges (absolutely) for 1 < x < 3 and diverges for x ≤ 1 and for x ≥ 3.

4. If the power series

∞

an ( x − x0 ) n converges at x = x1 , it converges absolutely for

n=0

|x − x0 | < |x1 − x0 |; and if it diverges at x = x1 , it diverges for |x − x0 | > |x1 − x0 |. 5. For a typical power series, such as the one in Example 1, there is a positive number ∞ ρ, called the radius of convergence, such that an ( x − x0 ) n converges absolutely n=0

for |x − x0 | < ρ and diverges for |x − x0 | > ρ. The interval |x − x0 | < ρ is called the interval of convergence; it is indicated by the hatched lines in Figure 5.1.1. The series may either converge or diverge when |x − x0 | = ρ. Many important power series converge for all values of x. In this case it is customary to say that ρ is infinite and the interval of convergence is the entire real line. It is also possible for a power series to converge only at x0 . For such a series we say that ρ = 0 and the series has no interval of convergence. When these exceptional cases are taken into account, every power series has a nonnegative radius of convergence ρ, and if ρ > 0, then there is a (finite or infinite) interval of convergence centered at x0 . Series converges absolutely

Series diverges x0 – ρ

Series diverges x0 + ρ

x0

x

Series may converge or diverge FIGURE 5.1.1 The interval of convergence of a power series.

EXAMPLE 2 Determine the radius of convergence of the power series ∞ ( x + 1) n n=1

▼

n2n

.

Boyce 9131 Ch05 2

September 29, 2016

17:30

191

5.1 Review of Power Series

▼ Solution: We apply the ratio test:

( x + 1) n+1 n2n |x + 1| |x + 1| n = lim = . lim n+1 ( x + 1) n 2 n + 1 2 ( n + 1) 2 n→∞ n→∞

Thus the series converges absolutely for |x + 1| < 2, that is, for −3 < x < 1, and diverges for |x + 1| > 2. The radius of convergence of the power series is ρ = 2. Finally, we check the endpoints of the interval of convergence. At x = 1 the series becomes the harmonic series ∞ 1 n=1

n

,

which diverges. At x = −3 we have ∞ ( −3 + 1) n n=1

=

n2n

∞ ( −1) n

n

n=1

.

Recognizing this as the alternating harmonic series, we recall that it converges but does not converge absolutely. The power series is said to converge conditionally at x = −3. To summarize, the given power series converges for −3 ≤ x < 1 and diverges otherwise. It converges absolutely for −3 < x < 1 and has a radius of convergence of 2.

Suppose that

∞

an ( x − x0 ) n and

n=0

∞

bn ( x − x0 ) n converge to f ( x) and g( x) , respec-

n=0

tively, for |x − x0 | < ρ, ρ > 0.

6. The two series can be added or subtracted termwise, and f ( x) ± g( x) =

∞

( an ± bn ) ( x − x0 ) n ;

n=0

the resulting series converges at least for |x − x0 | < ρ. 7. The two series can be formally multiplied, and ⎛ ⎞⎛ ⎞ ∞ ∞ ∞ f ( x) g( x) = ⎝ a n ( x − x 0 ) n ⎠⎝ bn ( x − x0 ) n ⎠ = cn ( x − x 0 ) n , n=0

n=0

n=0

where cn = a0 bn + a1 bn−1 + · · · + an b0 . The resulting series converges at least for |x − x0 | < ρ. Further, if b0 = 0, then g( x0 ) = 0, and the series for f ( x) can be formally divided by the series for g( x) , and ∞

f ( x) = dn ( x − x 0 ) n . g( x) n=0

In most cases the coefficients dn can be most easily obtained by equating coefficients in the equivalent relation ⎤⎡ ⎤ ⎡ ∞ ∞ ∞ an ( x − x 0 ) n = ⎣ dn ( x − x0 ) n ⎦⎣ bn ( x − x0 ) n ⎦ n=0

n=0

n=0

⎛

=

⎞

∞ n ⎝ dk bn−k ⎠( x − x0 ) n . n=0

k=0

In the case of division, the radius of convergence of the resulting power series may be less than ρ.

191

Boyce 9131 Ch05 2

192

September 29, 2016

17:30

192

CHAPTER 5 Series Solutions of Second-Order Linear Equations

8. The function f is continuous and has derivatives of all orders for |x−x0 | < ρ. Moreover, f , f , . . . can be computed by differentiating the series termwise; that is, f ( x) = a1 + 2a2 ( x − x0 ) + · · · + nan ( x − x0 ) n−1 + · · · =

∞

nan ( x − x0 ) n−1 ,

n=1

f ( x) = 2a2 + 6a3 ( x − x0 ) + · · · + n( n − 1) an ( x − x0 ) n−2 + · · · =

∞

n( n − 1) an ( x − x0 ) n−2 ,

n=2

and so forth, and each of the series converges absolutely for |x − x0 | < ρ. 9. The value of an is given by an =

f ( n) ( x0 ) . n!

The series is called the Taylor1 series for the function f about x = x0 . ∞ ∞ an ( x − x 0 ) n = bn ( x − x0 ) n for each x in some open interval with center x0 , 10. If n=0

n=0

then an = bn for n = 0, 1, 2, 3, . . . . In particular, if

∞

an ( x − x0 ) n = 0 for each such

n=0

x, then a0 = a1 = · · · = an = · · · = 0.

A function f that has a Taylor series expansion about x = x0 f ( x) =

∞ f ( n) ( x0 ) ( x − x0 ) n , n! n=0

with a radius of convergence ρ > 0, is said to be analytic at x = x0 . All of the familiar functions of calculus are analytic except perhaps at certain easily recognized points. For example, sin x and e x are analytic everywhere, 1/ x is analytic except at x = 0, and tan x is analytic except at odd multiples of π/2. According to statements 6 and 7, if f and g are analytic at x0 , then f ± g, f · g, and f /g (provided that g( x0 ) = 0) are also analytic at x = x0 . In many respects the natural context for the use of power series is the complex plane. The methods and results of this chapter nearly always can be directly extended to differential equations in which the independent and dependent variables are complex-valued. Shift of Index of Summation. The index of summation in an infinite series is a dummy parameter just as the integration variable in a definite integral is a dummy variable. Thus it is immaterial which letter is used for the index of summation. For example, ∞ 2n x n n=0

n!

=

∞ 2jx j j=0

j!

.

Just as we make changes of the variable of integration in a definite integral, we find it convenient to make changes of summation indices in calculating series solutions of differential equations. We illustrate by several examples how to shift the summation index. ......................................................................................................................................................................... 1 Brook

Taylor (1685--1731), English mathematician, received his education at Cambridge University. His book Methodus incrementorum directa et inversa, published in 1715, includes a general statement of the expansion theorem that is named for him. This is a basic result in all branches of analysis, but its fundamental importance was not recognized until 1772 (by Lagrange). Taylor was also the first to use integration by parts, was one of the founders of the calculus of finite differences, and was the first to recognize the existence of singular solutions of differential equations.

Boyce 9131 Ch05 2

September 29, 2016

17:30

193

5.1 Review of Power Series

EXAMPLE 3 Write

∞

an x n as a series whose first term corresponds to n = 0 rather than n = 2.

n=2

Solution: Let m = n − 2; then n = m + 2, and n = 2 corresponds to m = 0. Hence ∞

an x n =

n=2

∞

am+2 x m+2 .

(1)

m=0

By writing out the first few terms of each of these series, you can verify that they contain precisely the same terms. Finally, in the series on the right-hand side of equation (1), we can replace the dummy index m by n, obtaining ∞

an x n =

n=2

∞

an+2 x n+2 .

(2)

n=0

In effect, we have shifted the index upward by 2 and have compensated by starting to count at a level 2 lower than originally.

EXAMPLE 4 Write the series ∞

( n + 2) ( n + 1) an ( x − x0 ) n−2

(3)

n=2

as a series whose generic term involves ( x − x0 ) n rather than ( x − x0 ) n−2 . Solution: Again, we shift the index by 2 so that n is replaced by n + 2 and start counting 2 lower. We obtain ∞

( n + 4) ( n + 3) an+2 ( x − x0 ) n .

(4)

n=0

You can readily verify that the terms in the series (3) and (4) are exactly the same.

EXAMPLE 5 Write the expression x2

∞

(r + n) an x r +n−1

(5)

n=0

as a series whose generic term involves x r +n . Solution: First, take the x 2 inside the summation, obtaining ∞ n=0

▼

(r + n) an x r +n+1 .

(6)

193

Boyce 9131 Ch05 2

194

September 29, 2016

17:30

194

CHAPTER 5 Series Solutions of Second-Order Linear Equations

▼ Next, shift the index down by 1 and start counting 1 higher. Thus ∞

(r + n) an x r +n+1 =

n=0

∞

(r + n − 1) an−1 x r +n .

(7)

n=1

Again, you can easily verify that the two series in equation (7) are identical and that both are exactly the same as the expression (5).

EXAMPLE 6 Assume that ∞

nan x

n−1

n=1

=

∞

an x n

(8)

n=0

for all x, and determine what this implies about the coefficients an . Solution: We want to use statement 10 to equate corresponding coefficients in the two series. In order to do this, we must first rewrite equation (8) so that the series display the same power of x in their generic terms. For instance, in the series on the left-hand side of equation (8), we can replace n by n + 1 and start counting 1 lower. Thus equation (8) becomes ∞

( n + 1) an+1 x n =

n=0

∞

an x n .

(9)

n=0

According to statement 10, we conclude that ( n + 1) an+1 = an ,

n = 0, 1, 2, 3, . . .

or an+1 =

an , n+1

n = 0, 1, 2, 3, . . . .

(10)

Hence, choosing successive values of n in equation (10), we have a1 = a0 ,

a2 =

a0 a1 = , 2 2

a3 =

a0 a2 = , 3 3!

and so forth. In general, an =

a0 , n!

n = 1, 2, 3, . . . .

(11)

Thus the relation (8) determines all the following coefficients in terms of a0 . Finally, using the coefficients given by equation (11), we obtain ∞ n=0

an x n =

∞ a0 n=0

n!

x n = a0

∞ xn n=0

n!

= a0 e x ,

where we have followed the usual convention that 0! = 1, and recalled that e x =

∞ xn n=0

values of x. (See Problem 8.)

n!

for all

Boyce 9131 Ch05 2

September 29, 2016

17:30

195

5.2 Series Solutions Near an Ordinary Point, Part I

195

Problems In each of Problems 1 through 6, determine the radius of convergence of the given power series.

1.

∞

15. Let y =

a. Compute y and y and write out the first four terms of each series, as well as the coefficient of x n in the general term. b. Show that if y = y, then the coefficients a0 and a1 are arbitrary, and determine a2 and a3 in terms of a0 and a1 . an , n = 0, 1, 2, 3, . . . . c. Show that an+2 = ( n + 2) ( n + 1)

( x − 3) n

∞ n n=0

3.

2n

4. 5.

xn

∞ x 2n

n!

n=0 ∞

In each of Problems 16 and 17, verify the given equation.

2n x n

16.

∞

n=0

n=0

∞ (x − x )n 0

∞

n

n=1

6.

3n

n=1

In each of Problems 7 through 13, determine the Taylor series about the point x0 for the given function. Also determine the radius of convergence of the series. x0 = 0

7. sin x, 9. x, 10. x 2 ,

x0 = 0 x0 = 1

12.

1 , 1−x

x0 = 0

13.

1 , 1−x

x0 = 2

14. Let y =

an−1 ( x − 1) n

n=1

ak+1 x k +

∞

ak x k+1 = a1 +

k=0

∞

( ak+1 + ak−1 ) x k

k=1

In each of Problems 18 through 22, rewrite the given expression as a single power series whose generic term involves x n .

18.

∞

n( n − 1) an x n−2

n=2

19. x

∞

nan x n−1 +

n=1

20. 21.

∞

∞

ak x k

k=0

m( m − 1) am x m−2 + x

∞

22. x

nan x n−1 + x

∞

∞

kak x k−1

k=1

n=1

x0 = 1

∞

∞

m=2

x0 = −1

11. ln x,

17.

an ( x − 1) n+1 =

k=0

∞ ( −1) n n 2 ( x + 2) n

8. e x ,

an x n .

n=0

n=0

2.

∞

∞

an x n

n=0

n( n − 1) an x n−2 +

n=2

∞

an x n

n=0

23. Determine the an so that the equation ∞

nx n .

nan x n−1 + 2

n=1

∞

an x n = 0

n=0

n=0

a. Compute y and write out the first four terms of the series. b. Compute y and write out the first four terms of the series.

is satisfied. Try to identify the function represented by the series ∞

an x n .

n=0

Series Solutions Near an Ordinary Point, Part I 5.2

In Chapter 3 we described methods of solving second-order linear differential equations with constant coefficients. We now consider methods of solving second-order linear equations when the coefficients are functions of the independent variable. In this chapter we will denote

Boyce 9131 Ch05 2

196

September 29, 2016

17:30

196

CHAPTER 5 Series Solutions of Second-Order Linear Equations

the independent variable by x. It is sufficient to consider the homogeneous equation P( x)

d2 y dy + Q( x) + R( x) y = 0, 2 dx dx

(1)

since the procedure for the corresponding nonhomogeneous equation is similar. Many problems in mathematical physics lead to equations of the form (1) having polynomial coefficients; examples include the Bessel equation x 2 y + x y + ( x 2 − ν 2 ) y = 0, where ν is a constant, and the Legendre equation ( 1 − x 2 ) y − 2x y + α ( α + 1) y = 0, where α is a constant. For this reason, as well as to simplify the algebraic computations, we primarily consider the case in which the functions P, Q, and R are polynomials. However, as we will see, the method of solution is also applicable when P, Q, and R are general analytic functions. For the present, then, suppose that P, Q, and R are polynomials and that there is no factor ( x −c) that is common to all three of them. If there is such a common factor ( x −c) , then divide it out before proceeding. Suppose also that we wish to solve equation (1) in the neighborhood of a point x0 . The solution of equation (1) in an interval containing x0 is closely associated with the behavior of P in that interval. A point x0 such that P( x0 ) = 0 is called an ordinary point. Since P is continuous, it follows that there is an open interval containing x0 in which P( x) is never zero. In that interval, which we will call I , we can divide equation (1) by P( x) to obtain y + p( x) y + q( x) y = 0,

(2)

where p( x) = Q( x) / P( x) and q( x) = R( x) / P( x) are continuous functions on I . Hence, according to the existence and uniqueness theorem, Theorem 3.2.1, there exists a unique solution of equation (1) in the interval I that also satisfies the initial conditions y( x0 ) = y0 and y ( x0 ) = y0 for arbitrary values of y0 and y0 . In this and the following section, we discuss the solution of equation (1) in the neighborhood of an ordinary point. On the other hand, if P( x0 ) = 0, then x0 is called a singular point of equation (1). In this case, because ( x − x0 ) is not a factor of P, Q, and R, at least one of Q( x0 ) and R( x0 ) is not zero. Consequently, at least one of the coefficients p and q in equation (2) becomes unbounded as x → x0 , and therefore Theorem 3.2.1 does not apply in this case. Sections 5.4 through 5.7 deal with finding solutions of equation (1) in the neighborhood of a singular point. We now take up the problem of solving equation (1) in the neighborhood of an ordinary point x0 . We look for solutions of the form y = a0 + a1 ( x − x 0 ) + · · · + an ( x − x 0 ) n + · · · =

∞

an ( x − x 0 ) n

(3)

n=0

and assume that the series converges in the interval |x − x0 | < ρ for some ρ > 0. While at first sight it may appear unattractive to seek a solution in the form of a power series, this is actually a convenient and useful form for a solution. Within their intervals of convergence, power series behave very much like polynomials and are easy to manipulate both analytically and numerically. Indeed, even if we can obtain a solution in terms of elementary functions, such as exponential or trigonometric functions, we are likely to need a power series or some equivalent expression if we want to evaluate the solution numerically or to plot its graph. The most practical way to determine the coefficients an is to substitute the series (3) and its derivatives for y, y , and y in equation (1). The following examples illustrate this process. The operations, such as differentiation, that are involved in the procedure are justified so long as we stay within the interval of convergence. The differential equations in these examples are also of considerable importance in their own right.

Boyce 9131 Ch05 2

September 29, 2016

17:30

197

5.2 Series Solutions Near an Ordinary Point, Part I

EXAMPLE 1 Find a series solution of the equation y + y = 0,

−∞ < x < ∞.

(4)

Solution: As we know, sin x and cos x form a fundamental set of solutions of this equation, so series methods are not needed to solve it. However, this example illustrates the use of power series in a relatively simple case. For equation (4), P( x) = 1, Q( x) = 0, and R( x) = 1; hence every point is an ordinary point. We look for a solution in the form of a power series about x0 = 0 ∞

y = a0 + a1 x + a2 x 2 + a3 x 3 + · · · + an x n + · · · =

an x n

(5)

n=0

and assume that the series converges in some interval |x| < ρ. Differentiating equation (5) term by term, we obtain y = a1 + 2a2 x + 3a3 x 2 + · · · + nan x n−1 + · · · =

∞

nan x n−1

(6)

n=1

and y = 2a2 + 3 · 2a3 x + · · · + n( n − 1) an x n−2 + · · · =

∞

n( n − 1) an x n−2 .

(7)

n=2

Substituting the series (5) and (7) for y and y in equation (4) gives ∞

n( n − 1) an x n−2 +

n=2

∞

an x n = 0.

n=0

To combine the two series, we need to rewrite at least one of them so that both series display the same generic term. (See Problem 22 in Section 5.1.) Thus, in the first sum, we shift the index of summation by replacing n by n + 2 and starting the sum at 0 rather than 2. We obtain ∞

( n + 2) ( n + 1) an+2 x n +

n=0

∞

an x n = 0

n=0

or ∞

( n + 2) ( n + 1) an+2 + an x n = 0.

n=0

For this equation to be satisfied for all x, the coefficient of each power of x must be zero; hence we conclude that ( n + 2) ( n + 1) an+2 + an = 0,

n = 0, 1, 2, 3, . . . .

(8)

Equation (8) is referred to as a recurrence relation. The successive coefficients can be evaluated one by one by writing the recurrence relation first for n = 0, then for n = 1, and so forth. In this example equation (8) relates each coefficient to the second one before it. Thus the evennumbered coefficients ( a0 , a2 , a4 , . . . ) and the odd-numbered ones ( a1 , a3 , a5 , . . . ) are determined separately. For the even-numbered coefficients we have a0 a0 a0 a0 a2 a4 a2 = − = − , a4 = − = + , a6 = − = − , ... . 2·1 2! 4·3 4! 6·5 6! These results suggest that in general, if n = 2k, then an = a2k =

( −1) k a0 , ( 2k) !

k = 1, 2, 3, . . . .

(9)

We can prove equation (9) by mathematical induction. First, observe that it is true for k = 1. Next, assume that it is true for an arbitrary value of k and consider the case k + 1. We have a2k+2 = −

▼

a2k ( −1) k ( −1) k+1 =− a0 = a0 . ( 2k + 2) ( 2k + 1) ( 2k + 2) ( 2k + 1) ( 2k) ! ( 2k + 2) !

197

Boyce 9131 Ch05 2

198

September 29, 2016

17:30

198

CHAPTER 5 Series Solutions of Second-Order Linear Equations

▼ Hence equation (9) is also true for k + 1, and consequently it is true for all positive integers k. Similarly, for the odd-numbered coefficients a1 a1 a1 a3 a3 = − = − , a5 = − =+ , 2·3 3! 5·4 5!

a7 = −

a1 a5 = − , ... , 7·6 7!

and in general, if n = 2k + 1, then2 an = a2k+1 =

( −1) k a1 , ( 2k + 1) !

k = 1, 2, 3, . . . .

(10)

Substituting these coefficients into equation (5), we have y = a0 + a1 x − +··· +

( −1) n a0 2n ( −1) n a1 2n+1 x + x + ··· ( 2n) ! ( 2n + 1) !

= a0 1 −

+ a1 = a0

a0 2 a1 3 a0 4 a1 5 x − x + x + x 2! 3! 4! 5!

x2 x4 ( −1) n 2n + + ··· + x + ··· 2! 4! ( 2n) !

∞ ( −1) n n=0

x3 x5 ( −1) n 2n+1 x− + ··· + + ··· + x 3! 5! ( 2n + 1) ! ( 2n) !

x 2n + a1

∞ ( −1) n n=0

( 2n + 1) !

x 2n+1 .

(11)

We identify two series solutions of equation (4): y1 ( x) =

∞ ( −1) n n=0

( 2n) !

x 2n and y2 ( x) =

∞ ( −1) n n=0

( 2n + 1) !

x 2n+1 .

Using the ratio test, we can show that the series for y1 ( x) and y2 ( x) converge for all x, and this justifies retroactively all of the steps used in obtaining these solutions. Indeed, the series for y1 ( x) is exactly the Taylor series for cos x about x = 0 and the series for y2 ( x) is the corresponding Taylor series for sin x. Thus, as we anticipated in equation (11) we have obtained the general solution of equation (4) in the form y = a0 cos x + a1 sin x. Notice that no conditions are imposed on a0 and a1 ; hence they are arbitrary. From equations (5) and (6) we see that y and y evaluated at x = 0 are a0 and a1 , respectively. Since the initial conditions y( 0) and y ( 0) can be chosen arbitrarily, it follows that a0 and a1 should be arbitrary until specific initial conditions are stated. Figures 5.2.1 and 5.2.2 show how the partial sums of the series solutions y1 ( x) and y2 ( x) approximate cos x and sin x, respectively. As the number of terms increases, the interval over which y 2

n = 4 n = 8 n = 12

n = 16

n = 20

1

2

4

–1

6

8

10

x

y = cos x

–2 n=2

n=6

n = 10 n = 14 n = 18

FIGURE 5.2.1 Polynomial approximations to y = cos x. The value of n is the degree of the approximating polynomial.

..................................................................................................................................................................................... 2 The

result given in equation (10) and other similar formulas in this chapter can be proved by an induction argument resembling the one just given for equation (9). We assume that the results are plausible and omit the inductive argument hereafter. (See Problem 16.)

▼

Boyce 9131 Ch05 2

September 29, 2016

17:30

199

5.2 Series Solutions Near an Ordinary Point, Part I

▼ the approximation is satisfactory becomes longer, and for each x in this interval the accuracy of the approximation improves. However, you should always remember that a truncated power series provides only a local approximation of the solution in a neighborhood of the initial point x = 0; it cannot adequately represent the solution for large |x|.

y

n=5

n = 9 n = 13 n = 17

n = 21

2

1

2

4

6

8

–1

10

x

y = sin x

–2 n=3

n=7

n = 11 n = 15 n = 19

FIGURE 5.2.2 Polynomial approximations to y = sin x. The value of n is the degree of the approximating polynomial.

In Example 1 we knew from the start that sin x and cos x form a fundamental set of solutions of equation (4). However, if we had not known this and had simply solved equation (4) using series methods, we would still have obtained the solution (11). In recognition of the fact that the differential equation (4) often occurs in applications, we might decide to give the two solutions of equation (11) special names, perhaps C( x) =

∞ ( −1) n n=0

( 2n) !

x 2n ,

S( x) =

∞ ( −1) n 2n+1 . x ( 2n + 1) !

(12)

n=0

Then we might ask what properties these functions have. For instance, can we be sure that C( x) and S( x) form a fundamental set of solutions? It follows at once from the series expansions that C( 0) = 1 and S( 0) = 0. By differentiating the series for C( x) and S( x) term by term, we find that S ( x) = C( x) ,

C ( x) = −S( x) .

(13)

Thus at x = 0, we have S ( 0) = 1 and C ( 0) = 0. Consequently, the Wronskian of C and S at x = 0 is 1 0 W [C, S]( 0) = (14) = 1, 0 1 so these functions do indeed form a fundamental set of solutions. By substituting −x for x in each of equations (12), we obtain C( −x) = C( x) and S( −x) = −S( x) . Moreover, by calculating with the infinite series,3 we can show that the functions C( x) and S( x) have all the usual analytical and algebraic properties of the cosine and sine functions, respectively. Although you probably first saw the sine and cosine functions defined in a more elementary manner in terms of right triangles, it is interesting that these functions can be defined as solutions of a certain simple second-order linear differential equation. To be precise, the function sin x can be defined as the unique solution of the initial-value problem y + y = 0, y( 0) = 0, y ( 0) = 1; similarly, cos x can be defined as the unique solution of the initial-value ......................................................................................................................................................................... 3 Such

an analysis is given in Section 24 of Knopp (see the References at the end of this chapter).

199

Boyce 9131 Ch05 2

200

September 29, 2016

17:30

200

CHAPTER 5 Series Solutions of Second-Order Linear Equations

problem y + y = 0, y( 0) = 1, y ( 0) = 0. Many other functions that are important in mathematical physics are also defined as solutions of certain initial-value problems. For most of these functions there is no simpler or more elementary way to approach them.

EXAMPLE 2 Find a series solution in powers of x of Airy’s4 equation y − x y = 0,

−∞ < x < ∞.

(15)

Solution: For this equation P( x) = 1, Q( x) = 0, and R( x) = −x; hence every point is an ordinary point. We assume that y=

∞

an x n

(16)

n=0

and that the series converges in some interval |x| < ρ. The series for y is given by equation (7); as explained in the preceding example, we can rewrite it as

y =

∞

( n + 2) ( n + 1) an+2 x n .

(17)

n=0

Substituting the series (16) and (17) for y and y into the left-hand side of equation (15), we obtain ∞

( n + 2) ( n + 1) an+2 x n − x

n=0

∞

an x n =

∞

n=0

( n + 2) ( n + 1) ax+2 x n −

n=0

∞

an x n+1 .

(18)

n=0

Next, we shift the index of summation in the second series on the right-hand side of equation (18) by replacing n by n − 1 and starting the summation at 1 rather than zero. Thus we write equation (15) as 2 · 1a2 +

∞

( n + 2) ( n + 1) an+2 x n −

n=1

∞

an−1 x n = 0.

n=1

Again, for this equation to be satisfied for all x in some interval, the coefficients of like powers of x must be zero; hence a2 = 0, and we obtain the recurrence relation ( n + 2) ( n + 1) an+2 − an−1 = 0

for n = 1, 2, 3, . . . .

(19)

Since an+2 is given in terms of an−1 , the a’s are determined in steps of three. Thus a0 determines a3 , which in turn determines a6 , . . . ; a1 determines a4 , which in turn determines a7 , . . . ; and a2 determines a5 , which in turn determines a8 , . . . . Since a2 = 0, we immediately conclude that a5 = a8 = a11 = · · · = 0. For the sequence a0 , a3 , a6 , a9 , . . . we set n = 1, 4, 7, 10, . . . in the recurrence relation: a3 =

a0 , 2·3

a6 =

a0 a3 = , 5·6 2·3·5·6

a9 =

a0 a6 = , ... . 8·9 2·3·5·6·8·9

These results suggest the general formula a3n =

a0 , 2 · 3 · 5 · 6 · · · ( 3n − 1) ( 3n)

n ≥ 4.

..................................................................................................................................................................................... 4 Sir George Biddell Airy (1801--1892), an English astronomer and mathematician, was director of the Greenwich

Observatory from 1835 to 1881. He studied the equation named for him in an 1838 paper on optics. One reason why Airy’s equation is of interest is that for x negative the solutions are similar to trigonometric functions, and for x positive they are similar to hyperbolic functions. Can you explain why it is reasonable to expect such behavior?

▼

Boyce 9131 Ch05 2

September 29, 2016

17:30

201

5.2 Series Solutions Near an Ordinary Point, Part I

▼

For the sequence a1 , a4 , a7 , a10 , . . . , we set n = 2, 5, 8, 11, . . . in the recurrence relation: a4 =

a1 , 3·4

a7 =

a4 a1 = , 6·7 3·4·6·7

a10 =

a7 a1 = , ... . 9 · 10 3 · 4 · 6 · 7 · 9 · 10

In general, we have a3n+1 =

a1 , 3 · 4 · 6 · 7 · · · ( 3n) ( 3n + 1)

n ≥ 4.

Thus the general solution of Airy’s equation is

y( x) = a0 1 +

x3 x6 x 3n + + ··· + + ··· 2·3 2·3·5·6 2 · 3 · · · ( 3n − 1) ( 3n)

x7 x 3n+1 x4 + + ··· + + ··· + a1 x + 3·4 3·4·6·7 3 · 4 · · · ( 3n) ( 3n + 1) = a0 y1 ( x) + a1 y2 ( x)

(20)

where y1 ( x) and y2 ( x) are the first and second bracketed expressions in equation (20). Having obtained these two series solutions, we can now investigate their convergence. Because of the rapid growth of the denominators of the terms in the series for y1 ( x) and for y2 ( x) , we might expect these series to have a large radius of convergence. Indeed, it is easy to use the ratio test to show that both of these series converge for all x; see Problem 17. Assume for the moment that the series for y1 and y2 do converge for all x. Then, by choosing first a0 = 1, a1 = 0 and then a0 = 0, a1 = 1, it follows that y1 and y2 are individually solutions of equation (15). Notice that y1 satisfies the initial conditions y1 ( 0) = 1, y1 ( 0) = 0 and that y2 satisfies the initial conditions y2 ( 0) = 0, y2 ( 0) = 1. Thus W [y1 , y2 ]( 0) = 1 = 0, and consequently y1 and y2 are a fundamental set of solutions. Hence the general solution of Airy’s equation is y = a0 y1 ( x) + a1 y2 ( x)

− ∞ < x < ∞.

In Figures 5.2.3 and 5.2.4, respectively, we show the graphs of the solutions y1 and y2 of Airy’s equation as well as graphs of several partial sums of the two series in equation (20). Again, the partial sums provide local approximations to the solutions in a neighborhood of the origin. Although the quality of the approximation improves as the number of terms increases, no polynomial can adequately represent y1 and y2 for large |x|. A practical way to estimate the interval in which a given partial sum is reasonably accurate is to compare the graphs of that partial sum and the next one, obtained by including one more term. As soon as the graphs begin to separate noticeably, we can be confident that the original partial sum is no longer accurate. For example, in Figure 5.2.3 the graphs for n = 24 and n = 27 begin to separate at about x = −9/2. Thus, beyond this point, the partial sum of degree 24 is worthless as an approximation to the solution. y 2 n = 48

36 42

24 30

12

n≥6

18

n=3

6

y = y1(x)

–10

–8

–6

39 n = 45

–4

27 33

–2

2

x

15 21

9

3

–2

FIGURE 5.2.3 Polynomial approximations to the solution y = y1 ( x) of Airy’s equation. The value of n is the degree of the approximating polynomial.

▼

201

Boyce 9131 Ch05 2

202

September 29, 2016

17:30

202

CHAPTER 5 Series Solutions of Second-Order Linear Equations

▼ y n≥4 n = 46

34 40

16

28

2

10

22

4

y = y2(x)

–10

–8

n = 49

–6

25

37 43

31

–2

–4

2

x

13 19

–2

7

FIGURE 5.2.4 Polynomial approximations to the solution y = y2 ( x) of Airy’s equation. The value of n is the degree of the approximating polynomial.

Observe that both y1 and y2 are monotone for x > 0 and oscillatory for x < 0. You can also see from the figures that the oscillations are not uniform but, rather, decay in amplitude and increase in frequency as the distance from the origin increases. In contrast to Example 1, the solutions y1 and y2 of Airy’s equation are not elementary functions that you have already encountered in calculus. However, because of their importance in some physical applications, these functions have been extensively studied, and their properties are well known to applied mathematicians and scientists.

EXAMPLE 3 Find a solution of Airy’s equation in powers of x − 1. Solution: The point x = 1 is an ordinary point of equation (15), and thus we look for a solution of the form y=

∞

an ( x − 1) n ,

n=0

where we assume that the series converges in some interval |x − 1| < ρ. Then y =

∞

nan ( x − 1) n−1 =

n=1

∞

( n + 1) an+1 ( x − 1) n ,

n=0

and y =

∞

n( n − 1) an ( x − 1) n−2 =

n=2

∞

( n + 2) ( n + 1) an+2 ( x − 1) n .

n=0

Substituting for y and y in equation (15), we obtain ∞ n=0

( n + 2) ( n + 1) an+2 ( x − 1) n = x

∞

an ( x − 1) n .

(21)

n=0

Now to equate the coefficients of like powers of ( x − 1) , we must express x, the coefficient of y in equation (15), in powers of x − 1; that is, we write x = 1 + ( x − 1) . Note that this is precisely the

▼

Boyce 9131 Ch05 2

September 29, 2016

17:30

203

5.2 Series Solutions Near an Ordinary Point, Part I

▼ Taylor series for x about x = 1. (See Problem 9 in Section 5.1.) Then equation (21) takes the form ∞ ∞ ( n + 2) ( n + 1) an+2 ( x − 1) n = ( 1 + ( x − 1) )

n=0

an ( x − 1) n

n=0

=

∞

an ( x − 1) n +

n=0

∞

an ( x − 1) n+1 .

n=0

Shifting the index of summation in the second series on the right gives ∞

( n + 2) ( n + 1) an+2 ( x − 1) n =

n=0

∞

an ( x − 1) n +

n=0

∞

an−1 ( x − 1) n .

n=1

Equating coefficients of like powers of x − 1, we obtain 2a2 = a0 , ( 3 · 2) a3 = a1 + a0 , ( 4 · 3) a4 = a2 + a1 , ( 5 · 4) a5 = a3 + a2 , . . . The general recurrence relation is ( n + 2) ( n + 1) an+2 = an + an−1

for n ≥ 1.

(22)

Solving for the first few coefficients an in terms of a0 and a1 , we find that a2 =

a0 , 2

a3 =

a1 a0 + , 6 6

a4 =

a2 a1 a0 a1 + = + , 12 12 24 12

a5 =

a3 a2 a0 a1 + = + . 20 20 30 120

Hence

y = a0 1 +

( x − 1) 3 ( x − 1) 4 ( x − 1) 5 ( x − 1) 2 + + + + ··· 2 6 24 30

+ a1

( x − 1) 4 ( x − 1) 5 ( x − 1) 3 + + + ··· . ( x − 1) + 6 12 120

(23)

In general, when the recurrence relation has more than two terms, as in equation (22), the determination of a formula for an in terms a0 and a1 will be fairly complicated, if not impossible. In this example such a formula is not readily apparent. Lacking such a formula, we cannot test the two series in equation (23) for convergence by direct methods such as the ratio test. However, we shall see in Section 5.3 that even without knowing the formula for an , it is possible to establish that the two series in equation (23) converge for all x. Further, they define functions y3 and y4 that are a fundamental set of solutions of the Airy equation (15). Thus y = a0 y3 ( x) + a1 y4 ( x) is the general solution of Airy’s equation for −∞ < x < ∞.

While Airy’s equation is not particularly complicated, Example 3 shows some of the complications encountered when looking for a power series solution expressed in powers of x − x0 with x0 = 0. There is an alternative. We can make the change of variable x − x0 = t, obtaining a new differential equation for y as a function of t, and then look for solutions of ∞ this new equation of the form an t n . When we have finished the calculations, we replace t n=0

by x − x0 (see Problem 15). In Examples 2 and 3 we have found two sets of solutions of Airy’s equation. The functions y1 and y2 defined by the series in equation (20) are a fundamental set of solutions of equation (15) for all x, and this is also true for the functions y3 and y4 defined by the series in equation (23). According to the general theory of second-order linear equations, each of the first two functions can be expressed as a linear combination of the latter two functions, and vice versa---a result that is certainly not obvious from an examination of the series alone.

203

Boyce 9131 Ch05 2

204

September 29, 2016

17:30

204

CHAPTER 5 Series Solutions of Second-Order Linear Equations

Finally, we emphasize that it is not particularly important if, as in Example 3, we are unable to determine the general coefficient an in terms of a0 and a1 . What is essential is that we can determine as many coefficients as we want. Thus we can find as many terms in the two series solutions as we want, even if we cannot determine the general term. While the task of calculating several coefficients in a power series solution is not difficult, it can be tedious. A symbolic manipulation package can be very helpful here; some are able to find a specified number of terms in a power series solution in response to a single command. With a suitable graphics package we can also produce plots such as those shown in the figures in this section.

Problems In each of Problems 1 through 11: a. Seek power series solutions of the given differential equation about the given point x0 ; find the recurrence relation that the coefficients must satisfy. b. Find the first four nonzero terms in each of two solutions y1 and y2 (unless the series terminates sooner). c. By evaluating the Wronskian W [y1 , y2 ]( x0 ) , show that y1 and y2 form a fundamental set of solutions. d. If possible, find the general term in each solution.

1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11.

y − y = 0, x0 = 0 y + 3y = 0, x0 = 0 y − x y − y = 0, x0 = 0 y − x y − y = 0, x0 = 1 y + k 2 x 2 y = 0, x0 = 0, k a constant ( 1 − x) y + y = 0, x0 = 0 y + x y + 2y = 0, x0 = 0 x y + y + x y = 0, x0 = 1 ( 3 − x 2 ) y − 3x y − y = 0,

2y + x y + 3y = 0,

x0 = 0 x0 = 2

In each of Problems 12 through 14: a. Find the first five nonzero terms in the solution of the given initial-value problem. G b. Plot the four-term and the five-term approximations to the solution on the same axes. c. From the plot in part b, estimate the interval in which the four-term approximation is reasonably accurate.

12. 13. 14. 15.

y − x y − y = 0,

y + x y + 2y = 0,

y( 0) = 2, y ( 0) = 1;

see Problem 3

y( 0) = 4, y ( 0) = −1; see Problem 7

( 1 − x) y + x y − y = 0,

y( 0) = −3, y ( 0) = 2

a. By making the change of variable x − 1 = t and assuming that y has a Taylor series in powers of t, find two series solutions of y + ( x − 1) 2 y + ( x 2 − 1) y = 0 in powers of x − 1. b. Show that you obtain the same result by assuming that y has a Taylor series in powers of x − 1 and also expressing the coefficient x 2 − 1 in powers of x − 1.

16. Prove equation (10).

of Airy’s equation about x = 0 converge for all x; see equation (20) of the text.

18. The Hermite Equation. The equation y − 2x y + λ y = 0,

−∞ < x < ∞,

where λ is a constant, is known as the Hermite5 equation. It is an important equation in mathematical physics. a. Find the first four nonzero terms in each of two solutions about x = 0 and show that they form a fundamental set of solutions. b. Observe that if λ is a nonnegative even integer, then one or the other of the series solutions terminates and becomes a polynomial. Find the polynomial solutions for λ = 0, 2, 4, 6, 8, and 10. Note that each polynomial is determined only up to a multiplicative constant. c. The Hermite polynomial Hn ( x) is defined as the polynomial solution of the Hermite equation with λ = 2n for which the coefficient of x n is 2n . Find H0 ( x) , H1 ( x) , . . . , H5 ( x) .

19. Consider the initial-value problem y = 1 − y 2 , y( 0) = 0. a. Show that y = sin x is the solution of this initial-value

x0 = 0

2y + ( x + 1) y + 3y = 0,

17. Show directly, using the ratio test, that the two series solutions

problem.

b. Look for a solution of the initial-value problem in the form of a power series about x = 0. Find the coefficients up to the term in x 3 in this series. In each of Problems 20 through 23, plot several partial sums in a series solution of the given initial-value problem about x = 0, thereby obtaining graphs analogous to those in Figures 5.2.1 through 5.2.4 (except that we do not know an explicit formula for the actual solution). G G G G

20. 21. 22. 23.

y + x y + 2y = 0, y( 0) = 0, y ( 0) = 1; see Problem 7 ( 4 − x 2 ) y + 2y = 0, y + x 2 y = 0,

y( 0) = 0, y ( 0) = 1

y( 0) = 1, y ( 0) = 0;

( 1 − x) y + x y − 2y = 0,

see Problem 5

y( 0) = 0, y ( 0) = 1

................................................................................................................................. 5 Charles Hermite (1822--1901) was an influential French analyst and algebraist. An inspiring teacher, he was professor at the École Polytechnique and the Sorbonne. He introduced the Hermite functions in 1864 and showed in 1873 that e is a transcendental number (that is, e is not a root of any polynomial equation with rational coefficients). His name is also associated with Hermitian matrices (see Section 7.3), some of whose properties he discovered.

Boyce 9131 Ch05 2

September 29, 2016

17:30

205

5.3 Series Solutions Near an Ordinary Point, Part II

Series Solutions Near an Ordinary Point, Part II 5.3

In the preceding section we considered the problem of finding solutions of P( x) y + Q( x) y + R( x) y = 0,

(1)

where P, Q, and R are polynomials, in the neighborhood of an ordinary point x0 . Assuming that equation (1) does have a solution y = φ ( x) and that φ has a Taylor series φ ( x) =

∞

an ( x − x 0 ) n

(2)

n=0

that converges for |x − x0 | < ρ, where ρ > 0, we found that the an can be determined by directly substituting the series (2) for y in equation (1). Let us now consider how we might justify the statement that if x0 is an ordinary point of equation (1), then there exist solutions of the form (2). We also consider the question of the radius of convergence of such a series. In doing this, we are led to a generalization of the definition of an ordinary point. Suppose, then, that there is a solution of equation (1) of the form (2). By differentiating equation (2) m times and setting x equal to x0 , we obtain m!am = φ ( m) ( x0 ) .

(3)

Hence, to compute an in the series (2), we must show that we can determine φ ( n) ( x0 ) for n = 0, 1, 2, . . . from the differential equation (1). Suppose that y = φ ( x) is a solution of equation (1) satisfying the initial conditions y( x0 ) = y0 , y ( x0 ) = y0 . Then a0 = y0 and a1 = y0 . If we are solely interested in finding a solution of equation (1) without specifying any initial conditions, then a0 and a1 remain arbitrary. To determine φ ( n) ( x0 ) and the corresponding an for n = 2, 3, . . . , we turn to equation (1) with the goal of finding a formula for φ ( x) , φ ( x) , . . . . Since φ is a solution of equation (1), we have P( x) φ ( x) + Q( x) φ ( x) + R( x) φ ( x) = 0. For the interval about x0 for which P is nonzero, we can write this equation in the form φ ( x) = − p( x) φ ( x) − q( x) φ ( x) ,

(4)

where p( x) = Q( x) / P( x) and q( x) = R( x) / P( x) . Observe that, at x = x0 , the right-hand side of equation (4) is known, thus allowing us to compute φ ( x0 ) : Setting x equal to x0 in equation (4) gives φ ( x0 ) = − p( x0 ) φ ( x0 ) − q( x0 ) φ ( x0 ) = − p( x0 ) a1 − q( x0 ) a0 . Hence, using equation (3) with m = 2, we find that a2 is given by 2!a2 = φ ( x0 ) = − p( x0 ) a1 − q( x0 ) a0 .

(5)

To determine a3 , we differentiate equation (4) and then set x equal to x0 , obtaining 3!a3 = φ ( x0 ) = −( p( x) φ ( x) + q( x) φ ( x) ) x=x 0

= −2! p( x0 ) a2 − ( p ( x0 ) + q( x0 ) ) a1 − q ( x0 ) a0 . Substituting for a2 from equation (5) gives a3 in terms of a1 and a0 .

(6)

205

Boyce 9131 Ch05 2

206

September 29, 2016

17:30

206

CHAPTER 5 Series Solutions of Second-Order Linear Equations

Since P, Q, and R are polynomials and P( x0 ) = 0, all the derivatives of p and q exist at x0 . Hence we can continue to differentiate equation (4) indefinitely, determining after each differentiation the successive coefficients a4 , a5 , . . . by setting x equal to x0 .

EXAMPLE 1 Let y = φ ( x) be a solution of the initial value problem ( 1 + x 2 ) y + 2x y + 4x 2 y = 0, y( 0) = 0, y ( 0) = 1. Determine φ ( 0) , φ ( 0) , and φ ( 4) ( 0) . Solution: To find φ ( 0) , simply evaluate the differential equation when x = 0: ( 1 + 02 ) φ ( 0) + 2 · 0 · φ ( 0) + 4 · 02 · φ ( 0) = 0, so φ ( 0) = 0. To find φ ( 0) , differentiate the differential equation with respect to x: ( 1 + x 2 ) φ ( x) + 2xφ ( x) + 2xφ ( x) + 2φ ( x) + 4x 2 φ ( x) + 8xφ ( x) = 0.

(7)

Then evaluate the resulting equation (7) at x = 0: φ ( 0) + 2φ ( 0) = 0. Thus φ ( 0) = −2φ ( 0) = −2 (because φ ( 0) = 1). Finally, to find φ ( 4) ( 0) , first differentiate equation (7) with respect to x:

1 + x 2 φ ( 4) ( x) + 2xφ ( x) + 4xφ ( x) + 4φ ( x) + 2 + 4x 2 φ ( x) + 8xφ ( x) + 8xφ ( x) + 8φ ( x) = 0.

Evaluating this equation at x = 0 we find φ ( 4) ( 0) + 6φ ( 0) + 8φ ( 0) = 0. Finally, using φ ( 0) = 0 and φ ( 0) = 0, we conclude that φ ( 4) ( 0) = 0.

Notice that the important property that we used in determining the an was that we could compute infinitely many derivatives of the functions p and q. It might seem reasonable to relax our assumption that the functions p and q are ratios of polynomials and simply require that they be infinitely differentiable in the neighborhood of x0 . Unfortunately, this condition is too weak to ensure that we can prove the convergence of the resulting series expansion for y = φ ( x) . What is needed is to assume that the functions p and q are analytic at x0 ; that is, they have Taylor series expansions that converge to them in some interval about the point x0 : ∞ p( x) = p0 + p1 ( x − x0 ) + · · · + pn ( x − x0 ) n + · · · = pn ( x − x0 ) n , (8) n=0

q( x) = q0 + q1 ( x − x0 ) + · · · + qn ( x − x0 ) n + · · · =

∞

qn ( x − x 0 ) n .

(9)

n=0

With this idea in mind, we can generalize the definitions of an ordinary point and a singular point of equation (1) as follows: if the functions p( x) = Q( x) /P( x) and q( x) = R( x) /P( x) are analytic at x0 , then the point x0 is said to be an ordinary point of the differential equation (1); otherwise, it is a singular point. Now let us turn to the question of the interval of convergence of the series solution. One possibility is actually to compute the series solution for each problem and then to apply one of the tests for convergence of an infinite series to determine its radius of convergence. Unfortunately, these tests require us to obtain an expression for the general coefficient an as a function of n, and this task is often quite difficult, if not impossible; recall Example 3 in Section 5.2. However, the question can be answered at once for a wide class of problems by the following theorem.

Boyce 9131 Ch05 2

September 29, 2016

17:30

207

5.3 Series Solutions Near an Ordinary Point, Part II

Theorem 5.3.1 If x0 is an ordinary point of the differential equation (1) P( x) y + Q( x) y + R( x) y = 0, that is, if p( x) = Q( x) / P( x) and q( x) = R( x) / P( x) are analytic at x0 , then the general solution of equation (1) is y=

∞

an ( x − x0 ) n = a0 y1 ( x) + a1 y2 ( x) ,

n=0

where a0 and a1 are arbitrary, and y1 and y2 are two power series solutions that are analytic at x0 . The solutions y1 and y2 form a fundamental set of solutions. Further, the radius of convergence for each of the series solutions y1 and y2 is at least as large as the minimum of the radii of convergence of the series for p and q.

To see that y1 and y2 are a fundamental set of solutions, note that they have the form y1 ( x) = 1+b2 ( x − x0 ) 2 +· · · and y2 ( x) = ( x − x0 ) +c2 ( x − x0 ) 2 + · · · , where b2 +c2 = a2 . Hence y1 satisfies the initial conditions y1 ( x0 ) = 1, y1 ( x0 ) = 0, and y2 satisfies the initial conditions y2 ( x0 ) = 0, y2 ( x0 ) = 1. Thus W [y1 , y2 ]( x0 ) = 1. Also note that although calculating the coefficients by successively differentiating the differential equation is excellent in theory, it is usually not a practical computational procedure. Rather, you should substitute the series (2) for y in the differential equation (1) and determine the coefficients so that the differential equation is satisfied, as in the examples in the preceding section. We will not prove this theorem, which in a slightly more general form was established by Fuchs.6 What is important for our purposes is that there is a series solution of the form (2) and that the radius of convergence of the series solution cannot be less than the smaller of the radii of convergence of the series for p and q; hence we need only determine these. This can be done in either of two ways. Again, one possibility is simply to compute the power series for p and q and then to determine the radii of convergence by using one of the convergence tests for infinite series. However, there is an easier way when P( x) , Q( x) , and R( x) are polynomials. It is shown in the theory of functions of a complex variable that the ratio of two polynomials, say, Q( x) / P( x) , has a convergent power series expansion about a point x = x0 if P( x0 ) = 0. Further, if we assume that any factors common to Q( x) and P( x) have been canceled, then the radius of convergence of the power series for Q( x) / P( x) about the point x0 is precisely the distance from x0 to the nearest zero of P( x) . In determining this distance, we must remember that P( x) = 0 may have complex roots, and these must also be considered.

EXAMPLE 2 What is the radius of convergence of the Taylor series for ( 1 + x 2 ) −1 about x = 0? Solution: One way to proceed is to find the Taylor series in question, namely, 1 = 1 − x 2 + x 4 − x 6 + · · · + ( −1) n x 2n + · · · . 1 + x2 Then it can be verified by the ratio test that ρ = 1. Another approach is to note that the zeros of 1 + x 2 are x = ±i. Since the distance in the complex plane from 0 to i or to −i is 1, the radius of convergence of the power series about x = 0 is 1.

......................................................................................................................................................................... 6 Lazarus Immanuel Fuchs (1833--1902), a German mathematician, was a student and later a professor at the University

of Berlin. He proved the result of Theorem 5.3.1 in 1866. His most important research was on singular points of linear differential equations. He recognized the significance of regular singular points (Section 5.4), and equations whose only singularities, including the point at infinity, are regular singular points are known as Fuchsian equations.

207

Boyce 9131 Ch05 2

208

September 29, 2016

17:30

208

CHAPTER 5 Series Solutions of Second-Order Linear Equations

EXAMPLE 3 What is the radius of convergence of the Taylor series for ( x 2 − 2x + 2) −1 about x = 0? about x = 1? Solution: First notice that x 2 − 2x + 2 = 0 has solutions x = 1 ± i. The distance in the complex plane from x = 0 to either x = 1 + i or x = 1 − i is

2; hence the radius of convergence of the Taylor series expansion

∞

an x n about

n=0

x = 0 is 2. The distance in the complex plane from x = 1 to either x = 1 + i or x = 1 − i is 1; hence the radius of convergence of the Taylor series expansion

∞

bn ( x − 1) n about x = 1 is 1.

n=0

According to Theorem 5.3.1, the series solutions of the Airy equation in Examples 2 and 3 of the preceding section converge for all values of x and x − 1, respectively, since in each problem P( x) = 1 and hence is never zero. A series solution may converge for a wider range of x than indicated by Theorem 5.3.1, so the theorem actually gives only a lower bound on the radius of convergence of the series solution. This is illustrated by the Legendre polynomial solution of the Legendre equation given in the next example.

EXAMPLE 4 Determine a lower bound for the radius of convergence of series solutions about x = 0 for the Legendre equation ( 1 − x 2 ) y − 2x y + α ( α + 1) y = 0, where α is a constant. Solution: Note that P( x) = 1 − x 2 , Q( x) = −2x, and R( x) = α ( α + 1) are polynomials, and that the zeros of P, namely, x = ±1, are a distance 1 from x = 0. Hence a series solution of the form

∞

an x n

n=0

converges at least for |x| < 1, and possibly for larger values of x. Indeed, it can be shown that if α is a positive integer, one of the series solutions terminates after a finite number of terms, that is, one solution is a polynomial, and hence converges not just for |x| < 1 but for all x. For example, if α = 1, the polynomial solution is y = x. See Problems 17 through 23 at the end of this section for a further discussion of the Legendre equation.

EXAMPLE 5 Determine a lower bound for the radius of convergence of series solutions of the differential equation ( 1 + x 2 ) y + 2x y + 4x 2 y = 0

(10)

about the point x = 0; about the point x = − 12 . Solution: Again P, Q, and R are polynomials, andP has zeros at x = ±i. The distance in the complex plane ∞ 5 1 1 . Hence in the first case the series an x n from 0 to ±i is 1, and from − to ±i is 1 + = 2 4 2 n=0 n ∞ 1 converges at least for |x| < 1, and in the second case the series bn x + converges at least 2 n=0

▼

for x +

1 < 2

2

5

.

Boyce 9131 Ch05 2

September 29, 2016

17:30

209

5.3 Series Solutions Near an Ordinary Point, Part II

▼

209

An interesting observation that we can make about equation (10) follows from Theorems 3.2.1 and 5.3.1. Suppose that initial conditions y( 0) = y0 and y ( 0) = y0 are given. Since 1 + x 2 = 0 for all x, we know from Theorem 3.2.1 that there exists a unique solution of the initial-value problem on −∞ < x < ∞. On the other hand, Theorem 5.3.1 only guarantees a series solution of the form

∞

n=0

an x n (with a0 = y0 , a1 = y0 ) for −1 < x < 1. The unique solution on the interval

−∞ < x < ∞ may not have a power series about x = 0 that converges for all x.

EXAMPLE 6 Can we determine a series solution about x = 0 for the differential equation y + ( sin x) y + ( 1 + x 2 ) y = 0, and if so, what is the radius of convergence? Solution: For this differential equation, p( x) = sin x and q( x) = 1 + x 2 . Recall from calculus that sin x has a Taylor series expansion about x = 0 that converges for all x. Further, q also has a Taylor series expansion about x = 0, namely, q( x) = 1 + x 2 , that converges for all x. Thus there is a series solution of the form y =

∞

an x n with a0 and a1 arbitrary, and the series converges for all x.

n=0

Problems In each of Problems 1 through 3, determine φ ( x0 ) , φ ( x0 ) , and φ ( 4) ( x0 ) for the given point x0 if y = φ ( x) is a solution of the given initial-value problem.

b. Show that if α is a nonnegative integer n, then there is a polynomial solution of degree n. These polynomials, when properly normalized, are called the Chebyshev polynomials. They are very useful in problems that require a polynomial approximation to a function defined on −1 ≤ x ≤ 1. c. Find a polynomial solution for each of the cases α = n = 0, 1, 2, 3.

1. y + x y + y = 0; y( 0) = 1, y ( 0) = 0 2. x 2 y + ( 1 + x) y + 3( ln x) y = 0; y( 1) = 2, y ( 1) = 0 3. y + x 2 y + ( sin x) y = 0; y( 0) = a0 , y ( 0) = a1 In each of Problems 4 through 6, determine a lower bound for the radius of convergence of series solutions about each given point x0 for the given differential equation.

4. 5. 6. 7.

y + 4y + 6x y = 0;

x0 = 0, x0 = 4

( x 2 − 2x − 3) y + x y + 4y = 0;

( 1 + x ) y + 4x y + y = 0; 3

x0 = 4, x0 = −4, x0 = 0

x0 = 0, x0 = 2

Determine a lower bound for the radius of convergence of series solutions about the given x0 for each of the differential equations in Problems 1 through 11 of Section 5.2.

8. The Chebyshev Equation. The Chebyshev7 differential equation is ( 1 − x 2 ) y − x y + α 2 y = 0, where α is a constant.

a. Determine two solutions in powers of x for |x| < 1, and show that they form a fundamental set of solutions. ............................................................................................................................. 7 Pafnuty

L. Chebyshev (1821--1894), the most influential nineteenth-century Russian mathematician, was for 35 years professor at the University of St. Petersburg, which produced a long line of distinguished mathematicians. His study of Chebyshev polynomials began in about 1854 as part of an investigation of the approximation of functions by polynomials. Chebyshev is also known for his work in number theory and probability.

For each of the differential equations in Problems 9 through 11, find the first four nonzero terms in each of two power series solutions about the origin. Show that they form a fundamental set of solutions. What do you expect the radius of convergence to be for each solution?

9. 10. 11. 12.

y + ( sin x) y = 0 e x y + x y = 0 ( cos x) y + x y − 2y = 0

Let y = x and y = x 2 be solutions of a differential equation P( x) y + Q( x) y + R( x) y = 0. Can you say whether the point x = 0 is an ordinary point or a singular point? Prove your answer.

First-Order Equations. The series methods discussed in this section are directly applicable to the first-order linear differential equation P( x) y + Q( x) y = 0 at a point x0 , if the function p = Q/ P has a Taylor series expansion about that point. Such a point is called an ordinary point, and further, the radius of convergence of the series y=

∞

an ( x − x0 ) n is at least as large as the radius of convergence

n=0

of the series for Q/ P. In each of Problems 13 through 16, solve the given differential equation by a series in powers of x and verify that a0 is arbitrary in each case. Problem 17 involves a nonhomogeneous differential equation to which series methods can be easily extended. Where possible, compare the series solution with the solution obtained by using the methods of Chapter 2.

Boyce 9131 Ch05 2

210

September 29, 2016

17:30

210

CHAPTER 5 Series Solutions of Second-Order Linear Equations

13. y − y = 0

14. y − x y = 0 15. ( 1 − x) y = y 16. y − y = x 2 The Legendre Equation. Problems 17 through 23 deal with the Legendre8 equation ( 1 − x 2 ) y − 2x y + α ( α + 1) y = 0. As indicated in Example 4, the point x = 0 is an ordinary point of this equation, and the distance from the origin to the nearest zero of P( x) = 1 − x 2 is 1. Hence the radius of convergence of series solutions about x = 0 is at least 1. Also notice that we need to consider only α > −1 because if α ≤ −1, then the substitution α = −( 1 + γ ) , where γ ≥ 0, leads to the Legendre equation ( 1 − x 2 ) y − 2x y + γ ( γ + 1) y = 0.

17. Show that two solutions of the Legendre equation for |x| < 1 are α ( α + 1) 2 α ( α − 2) ( α + 1) ( α + 3) 4 y1 ( x) = 1 − x + x 2! 4! +

∞

( −1) m

m=3

α · · · ( α − 2m + 2) ( α + 1) · · · ( α + 2m − 1) 2m x , ( 2m) !

( α − 1) ( α + 2) 3 x y2 ( x) = x − 3! ( α − 1) ( α − 3) ( α + 2) ( α + 4) 5 x + 5! +

∞

( −1) m

m=3

( α − 1) · · · ( α − 2m + 1) ( α + 2) · · · ( α + 2m) 2m+1 × . x ( 2m + 1) !

18. Show that if α is zero or a positive even integer 2n, the series solution y1 reduces to a polynomial of degree 2n containing only even powers of x. Find the polynomials corresponding to α = 0, 2, and 4. Show that if α is a positive odd integer 2n + 1, the series solution y2 reduces to a polynomial of degree 2n + 1 containing only odd powers of x. Find the polynomials corresponding to α = 1, 3, and 5.

20. The Legendre polynomials play an important role in mathematical physics. For example, in solving Laplace’s equation (the potential equation) in spherical coordinates, we encounter the equation d 2 F( ϕ) d F( ϕ) + cot ϕ + n( n + 1) F( ϕ) = 0, dϕ dϕ 2

where n is a positive integer. Show that the change of variable x = cos ϕ leads to the Legendre equation with α = n for y = f ( x) = F( arccos x) .

21. Show that for n = 0, 1, 2, 3, the corresponding Legendre polynomial is given by Pn ( x) =

1 2n n!

dn ( x 2 − 1) n . dxn

This formula, known as Rodrigues’s formula,9 is true for all positive integers n.

22. Show that the Legendre equation can also be written as

2 (1 − x ) y

= −α ( α + 1) y.

Then it follows that

( 1 − x 2 ) Pn ( x)

= −n( n + 1) Pn ( x)

and

( 1 − x 2 ) Pm ( x)

= −m( m + 1) Pm ( x) .

By multiplying the first equation by Pm ( x) and the second equation by Pn ( x) , integrating by parts, and then subtracting one equation from the other, show that

1

−1

Pn ( x) Pm ( x) d x = 0 if n = m.

This property of the Legendre polynomials is known as the orthogonality property. If m = n, it can be shown that the value of the preceding integral is 2/( 2n + 1) .

23. Given a polynomial f of degree n, it is possible to express f as a linear combination of P0 , P1 , P2 , . . . , Pn :

19. The Legendre polynomial Pn ( x) is defined as the polynomial

solution of the Legendre equation with α = n that also satisfies the condition Pn ( 1) = 1. a. Using the results of Problem 18, find the Legendre polynomials P0 ( x) , . . . , P5 ( x) . G b. Plot the graphs of P0 ( x) , . . . , P5 ( x) for −1 ≤ x ≤ 1. N c. Find the zeros of P0 ( x) , . . . , P5 ( x) .

0 < ϕ < π,

f ( x) =

n

ak Pk ( x) .

k=0

Using the result of Problem 22, show that ak =

2k + 1 2

1

f ( x) Pk ( x) d x. −1

.............................................................................................................................. 8 Adrien-Marie

Legendre (1752--1833) held various positions in the French Académie des Sciences from 1783 onward. His primary work was in the fields of elliptic functions and number theory. The Legendre functions, solutions of Legendre’s equation, first appeared in 1784 in his study of the attraction of spheroids.

.............................................................................................................................. 9 Benjamin Olinde Rodrigues (1795--1851) published this result as part of his doctoral thesis from the University of Paris in 1815. He then became a banker and social reformer but retained an interest in mathematics. Unfortunately, his later papers were not appreciated until the late twentieth century.

Boyce 9131 Ch05 2

September 29, 2016

17:30

211

5.4 Euler Equations; Regular Singular Points

Euler Equations; Regular Singular Points

5.4

In this section we will begin to consider how to solve equations of the form P( x) y + Q( x) y + R( x) y = 0

(1)

in the neighborhood of a singular point x0 . Recall that if the functions P, Q, and R are polynomials having no factors common to all three of them, then the singular points of equation (1) are the points for which P( x) = 0. Euler Equations. A relatively simple differential equation that has a singular point is the Euler equation10 L[y] = x 2 y + α x y + β y = 0,

(2)

where α and β are real constants. Then P( x) = x , Q( x) = α x, and R( x) = β . If β = 0, then P( x) , Q( x) , and R( x) have no common factors, so the only singular point of equation (2) is x = 0; all other points are ordinary points. For convenience we first consider the interval x > 0; later we extend our results to the interval x < 0. Observe that ( x r ) = r x r −1 and ( x r ) = r (r − 1) x r −2 . Hence, if we assume that equation (2) has a solution of the form y = xr , (3) then we obtain L[x r ] = x 2 ( x r ) + α x( x r ) + β x r 2

= x 2r (r − 1) x r −2 + α x(r x r −1 ) + β x r = x r (r (r − 1) + αr + β ) . If r is a root of the quadratic equation F(r ) = r (r − 1) + αr + β = 0,

(4)

(5)

then L[x ] is zero, and y = x is a solution of equation (2). The roots of equation (5) are −( α − 1) ± ( α − 1) 2 − 4β r1 , r2 = (6) , 2 and the quadratic polynomial F(r ) defined in equation (5) can also be written as F(r ) = (r − r1 ) (r − r2 ) . Mirroring the treatment of second-order linear differential equations with constant coefficients, we consider separately the cases in which the roots are real and different, real but equal, and complex conjugates. Indeed, the entire discussion of Euler equations is similar to the treatment of second-order linear equations with constant coefficients in Chapter 3, with er x replaced by x r . r

r

Real, Distinct Roots. If F(r ) = 0 has real roots r1 and r2 , with r1 = r2 , then y1 ( x) = x r1 r2 and y2 ( x) = x are solutions of equation (2). Since W [x r1 , x r2 ] = (r2 − r1 ) x r1 +r2 −1 is nonzero for r1 = r2 and x > 0, it follows that the general solution of equation (2) is y = c 1 x r 1 + c2 x r 2 ,

x > 0.

(7)

Note that if r is not a rational number, then x r is defined by x r = er ln x .

EXAMPLE 1 Solve 2x 2 y + 3x y − y = 0,

x > 0.

(8)

▼ ......................................................................................................................................................................... 10 This equation is sometimes called the Cauchy--Euler equation or the equidimensional equation. Euler studied it in about 1740, but its solution was known to Johann Bernoulli before 1700.

211

Boyce 9131 Ch05 2

212

September 29, 2016

17:30

212

CHAPTER 5 Series Solutions of Second-Order Linear Equations

▼ Solution: Substituting y = x r in equation (8) gives x r ( 2r (r − 1) + 3r − 1) = x r ( 2r 2 + r − 1) = x r ( 2r − 1) (r + 1) = 0. Hence r1 =

1 and r2 = −1, so the general solution of equation (8) is 2 y = c1 x 1/2 + c2 x −1 ,

x > 0.

(9)

Equal Roots. If the roots r1 and r2 are equal, then we obtain only one solution y1 ( x) = x r1 of the assumed form. A second solution can be obtained by the method of reduction of order, but for the purpose of our future discussion we consider an alternative method. Since r1 = r2 , it follows that F(r ) = (r −r1 ) 2 . Thus in this case, not only does F(r1 ) = 0 but also F (r1 ) = 0. This suggests differentiating equation (4) with respect to r and then setting r equal to r1 . By differentiating equation (4) with respect to r , we obtain ∂ ∂ r ∂ r L[x r ] = x (r − r1 ) 2 [x F(r ) ] = ∂r ∂r ∂r = (r − r1 ) 2 x r ln x + 2(r − r1 ) x r .

(10)

However, by interchanging differentiation with respect to x and with respect to r , we also obtain ∂ ∂ r L[x r ] = L x = L[x r ln x]. ∂r ∂r The right-hand side of equation (10) is zero for r = r1 ; consequently, L[x r1 ln x] = 0 also. Therefore, a second solution of equation (2) is y2 ( x) = x r1 ln x,

x > 0.

(11)

By evaluating the Wronskian of y1 and y2 , we find that W [x r1 , x r1 ln x] = x 2r1 −1 . Hence x r1 and x r1 ln x are a fundamental set of solutions for x > 0, and the general solution of equation (2) is y = ( c1 + c2 ln x) x r1 ,

x > 0.

(12)

EXAMPLE 2 Solve x 2 y + 5x y + 4y = 0,

x > 0.

(13)

Solution: Substituting y = x r in equation (13) gives

x r (r (r − 1) + 5r + 4) = x r r 2 + 4r + 4 = 0. Hence r1 = r2 = −2, and y = x −2 ( c1 + c2 ln x) ,

x>0

(14)

is the general solution of equation (13).

Complex Roots. Finally, suppose that the roots r1 and r2 of equation (5) are complex conjugates, say, r1 = λ + iμ and r2 = λ − iμ , with μ = 0. We must now explain what

Boyce 9131 Ch05 2

September 29, 2016

17:30

213

5.4 Euler Equations; Regular Singular Points

is meant by x r when r is complex. Remembering that x r = er ln x

(15)

when x > 0 and r is real, we can use this equation to define x r when r is complex. Then, using Euler’s formula for eiμ ln x , we obtain x λ+iμ = e( λ+iμ ) ln x = eλ ln x eiμ ln x = x λ eiμ ln x = x λ ( cos( μ ln x) + i sin( μ ln x) ) ,

x > 0.

(16)

With this definition of x r for complex values of r , it can be verified that the usual laws of algebra and differential calculus hold, and hence x r1 and x r2 are indeed solutions of equation (2). The general solution of equation (2) is y = c1 x λ+iμ + c2 x λ−iμ .

(17)

λ+iμ

λ−iμ

The disadvantage of this expression is that the functions x and x are complex-valued. Recall that we had a similar situation for the second-order differential equation with constant coefficients when the roots of the characteristic equation were complex. Just as we did then, we can use Theorem 3.2.6 to obtain real-valued solutions of equation (2) by taking the real and imaginary parts of x λ+iμ , namely, x λ cos( μ ln x) and x λ sin( μ ln x) .

(18)

A straightforward calculation shows (see Problem 29) that W x λ cos( μ ln x) , x λ sin( μ ln x) = μ x 2λ−1 . Hence these solutions form a fundamental set of solutions for x > 0, and the general solution of the Euler equation (2) is y = c1 x λ cos( μ ln x) + c2 x λ sin( μ ln x) ,

x > 0.

(19)

EXAMPLE 3 Solve x 2 y + x y + y = 0.

(20)

Solution: Substituting y = x r in equation (20) gives x r (r (r − 1) + r + 1) = x r (r 2 + 1) = 0. Hence r = ±i, and the general solution is y = c1 cos( ln x) + c2 sin( ln x) ,

x > 0.

(21)

The factor x λ does not appear explicitly in equation (21) because in this example λ = 0 and x λ = 1.

Now let us consider the qualitative behavior of the solutions of equation (2) near the singular point x = 0. This depends entirely on the values of the exponents r1 and r2 . First, if r is real and positive, then x r → 0 as x tends to zero through positive values. On the other hand, if r is real and negative, then x r becomes unbounded. Finally, if r = 0, then x r = 1. Figure 5.4.1 shows these possibilities for various values of r . If r is complex, then a typical solution is x λ cos( μ ln x) . This function becomes unbounded or approaches zero if λ is negative or positive, respectively, and also oscillates more and more rapidly as x → 0. This behavior is shown in Figures 5.4.2 and 5.4.3 for selected values of λ and μ . If λ = 0, the oscillation is of constant amplitude. Finally, if there are repeated roots, then one solution is of the form x r ln x, which tends to zero if r > 0 and becomes unbounded if r ≤ 0. An example of each case is shown in Figure 5.4.4.

213

Boyce 9131 Ch05 2

214

September 29, 2016

17:30

214

CHAPTER 5 Series Solutions of Second-Order Linear Equations

y y = x–1/2

y = x–3/2

2

y = x0 1 y = x1/2 y = x3/2 x FIGURE 5.4.1 Solutions of an Euler equation; real roots (μ = 0).

y 10

y 1

y = x–1/2 cos(5 ln x)

0.125

0.25

0.5 x

0.375

0.5 –1

–10

FIGURE 5.4.2 Solution of an Euler equation; complex roots with negative real part.

1

1.5

2 x

y = x1/2 cos(5 ln x)

FIGURE 5.4.3 Solution of an Euler equation; complex roots with positive real part.

y 1

0.5 y = x ln x –1

1

1.5

2 x

y = x –1 ln x

FIGURE 5.4.4 The two typical second solutions of an Euler equation with equal roots: r > 0 (red), r < 0 (blue).

The extension of the solutions of equation (2) into the interval x < 0 can be carried out in a relatively straightforward manner. The difficulty lies in understanding what is meant by x r when x is negative and r is not an integer; similarly, ln x has not been defined for x < 0. The solutions of the Euler equation that we have given for x > 0 can be shown to be valid for x < 0, but in general they are complex-valued. Thus in Example 1 the solution x 1/2 is imaginary for x < 0. It is always possible to obtain real-valued solutions of the Euler equation (2) in the interval x < 0 by making the following change of variable. Let x = −ξ , where ξ > 0, and let y = u( ξ ) . Then we have dy du dξ du d2 y du dξ d 2u d = =− , − = = . (22) dx dξ d x dξ dξ dξ d x dx2 dξ 2

Boyce 9131 Ch05 2

September 29, 2016

17:30

215

5.4 Euler Equations; Regular Singular Points

Thus, for x < 0, equation (2) takes the form ξ2

d 2u du + αξ + β u = 0, dξ dξ 2

But except for names of the variables, this is equations (7), (12), and (19), we have ⎧ r c 1 ξ 1 + c2 ξ r 2 ⎪ ⎪ ⎪ ⎪ ⎪ ⎨( c + c ln ξ ) ξ r1 1 2 u( ξ ) = ⎪ ⎪ c1 ξ λ cos( μ ln ξ ) + c2 ξ λ sin( μ ln ξ ) ⎪ ⎪ ⎪ ⎩

ξ > 0.

(23)

exactly the same as equation (2); from if r1 and r2 are real-valued and different if r1 and r2 are real-valued with r1 = r2 if r1,2 = λ ± iμ are complex-valued ( μ = 0) , (24)

depending on the nature of the zeros of F(r ) = r (r − 1) + αr + β = 0. To obtain u in terms of x, we replace ξ by −x in equations (24). We can combine the results for x > 0 and x < 0 by recalling that |x| = x when x > 0 and that |x| = −x when x < 0. Thus we need only replace x by |x| in equations (7), (12), and (19) to obtain real-valued solutions valid in any interval not containing the origin. Hence the general solution of the Euler equation (2) x 2 y + α x y + β y = 0 in any interval not containing the origin is determined by the roots r1 and r2 of the equation F(r ) = r (r − 1) + αr + β = 0 as follows. If the roots r1 and r2 are real and different, r1,2 = λ ± iμ , then y = c1 |x|r1 + c2 |x|r2 .

(25)

y = ( c1 + c2 ln |x|) |x|r1 .

(26)

If the roots are real and equal, then

If the roots are complex conjugates, r1,2 = λ ± iμ , then

y = |x|λ c1 cos( μ ln |x|) + c2 sin( μ ln |x|) .

(27)

The solutions of an Euler equation of the form ( x − x0 ) 2 y + α ( x − x0 ) y + β y = 0

(28)

are similar. If we look for solutions of the form y = ( x − x0 ) r , then the general solution is given by equation (25), equation (26), or equation (27) with x replaced by x −x0 . Alternatively, we can reduce equation (28) to the form of equation (2) by making the change of independent variable t = x − x0 . Regular Singular Points. We now return to a consideration of the general equation (1) P( x) y + Q( x) y + R( x) y = 0, where x0 is a singular point. This means that P( x0 ) = 0 and that at least one of Q and R is not zero at x0 . Unfortunately, if we attempt to use the methods of the preceding two sections to solve equation (1) in the neighborhood of a singular point x0 , we find that these methods fail. This is because the solution of equation (1) is often not analytic at x0 and consequently cannot be represented by a Taylor series in powers of x − x0 . Examples 1, 2, and 3 illustrate this fact; in each of these examples, the solution fails to have a power series expansion about the singular

215

Boyce 9131 Ch05 2

216

September 29, 2016

17:30

216

CHAPTER 5 Series Solutions of Second-Order Linear Equations

point x = 0. Therefore, to have any chance of solving equation (1) in the neighborhood of a singular point we must use a more general type of series expansion. Since the singular points of a differential equation are usually few in number, we might ask whether we can simply ignore them, especially since we already know how to construct solutions about ordinary points. However, this is not feasible. The singular points determine the principal features of the solution to a much larger extent than you might at first suspect. In the neighborhood of a singular point the solution often becomes large in magnitude or experiences rapid changes in magnitude. For example, the solutions found in Examples 1, 2, and 3 are illustrations of this fact. Thus the behavior of a physical system modeled by a differential equation frequently is most interesting in the neighborhood of a singular point. Often geometric singularities in a physical problem, such as corners or sharp edges, lead to singular points in the corresponding differential equation. Thus, although at first we might want to avoid the few points where a differential equation is singular, it is precisely at these points that it is necessary to study the solution most carefully. As an alternative to analytical methods, we can consider the use of numerical methods, which are discussed in Chapter 8. However, these methods are ill suited for the study of solutions near a singular point. Thus, even if we adopt a numerical approach, it is advantageous to combine it with the analytical methods of this chapter in order to examine the behavior of solutions near singular points. Without any additional information about the behavior of Q/ P and R/ P in the neighborhood of the singular point, it is impossible to describe the behavior of the solutions of equation (1) near x = x0 . It may be that there are two distinct solutions of equation (1) that remain bounded as x → x0 (as in Example 3); or there may be only one, with the other becoming unbounded as x → x0 (as in Example 1); or they may both become unbounded as x → x0 (as in Example 2). If equation (1) has solutions that become unbounded as x → x0 , it is often important to determine how these solutions behave as x → x0 . For example, does y → ∞ in the same way as ( x − x0 ) −1 or |x − x0 |−1/2 , or in some other manner? Our goal is to extend the method already developed for solving equation (1) near an ordinary point so that it also applies to the neighborhood of a singular point x0 . To do this in a reasonably simple manner, it is necessary to restrict ourselves to cases in which the singularities in the functions Q/ P and R/ P at x = x0 are not too severe---that is, to what we might call “weak singularities.” At this stage it is not clear exactly what is an acceptable singularity. However, as we develop the method of solution, you will see that the appropriate conditions (see also Section 5.6, Problem 16) to distinguish “weak singularities” are lim ( x − x0 )

Q( x) is finite P( x)

(29)

lim ( x − x0 ) 2

R( x) is finite. P( x)

(30)

x→x0

and

x→x0

This means that the singularity in Q/ P can be no worse than ( x − x0 ) −1 and the singularity in R/ P can be no worse than ( x − x0 ) −2 . Such a point is called a regular singular point of equation (1). For equations with more general coefficients than polynomials, x0 is a regular singular point of equation (1) if it is a singular point and if both11 ( x − x0 )

R( x) Q( x) and ( x − x0 ) 2 P( x) P( x)

(31)

have convergent Taylor series about x0 ---that is, if the functions in equation (31) are analytic at x = x0 . Equations (29) and (30) imply that this will be the case when P, Q, and R are polynomials. Any singular point of equation (1) that is not a regular singular point is called an irregular singular point of equation (1). ......................................................................................................................................................................... 11 The

functions given in equation (31) may not be defined at x0 , in which case their values at x0 are to be assigned as their limits as x → x0 .

Boyce 9131 Ch05 2

September 29, 2016

17:30

217

5.4 Euler Equations; Regular Singular Points

Observe that the conditions in equations (29) and (30) are satisfied by the Euler equation (28). Thus the singularity in an Euler equation is a regular singular point. Indeed, we will see that all equations of the form (1) behave very much like Euler equations near a regular singular point. That is, solutions near a regular singular point may include powers of x with negative or nonintegral exponents, logarithms, or sines or cosines of logarithmic arguments. In the following sections we discuss how to solve equation (1) in the neighborhood of a regular singular point. A discussion of the solutions of differential equations in the neighborhood of irregular singular points is more complicated and may be found in more advanced books.

EXAMPLE 4 Determine the singular points of the Legendre equation ( 1 − x 2 ) y − 2x y + α ( α + 1) y = 0

(32)

and determine whether they are regular or irregular. Solution: In this case P( x) = 1 − x 2 , so the singular points are x = 1 and x = −1. Observe that when we divide equation (32) by 1−x 2 , the coefficients of y and y are −2x/( 1−x 2 ) and α ( α +1) /( 1−x 2 ) , respectively. We consider the point x = 1 first. Thus, from equations (29) and (30), we calculate lim ( x − 1) x→1

−2x ( x − 1) ( −2x) 2x = lim =1 = lim ( 1 − x) ( 1 + x) 1 1 − x2 x→1 x→1 + x

and lim ( x − 1) 2 x→1

α ( α + 1) ( x − 1) 2 α ( α + 1) = lim 1 − x2 x→1 ( 1 − x) ( 1 + x) ( x − 1) ( −α ) ( α + 1) = 0. 1+x x→1

= lim

Since these limits are finite, the point x = 1 is a regular singular point. It can be shown in a similar manner that x = −1 is also a regular singular point.

EXAMPLE 5 Determine the singular points of the differential equation 2x( x − 2) 2 y + 3x y + ( x − 2) y = 0 and classify them as regular or irregular. Solution: Dividing the differential equation by 2x( x − 2) 2 , we have y +

3 1 y = 0, y + 2 2x( x − 2) 2( x − 2)

Q( x) 3 1 R( x) = = . The singular points are x = 0 and q( x) = P( x) P( x) 2x( x − 2) 2( x − 2) 2 and x = 2. Consider x = 0. We have so p( x) =

lim x p( x) = lim x x→0

▼

x→0

3 = 0, 2( x − 2) 2

217

Boyce 9131 Ch05 2

218

September 29, 2016

17:30

218

CHAPTER 5 Series Solutions of Second-Order Linear Equations

▼ and lim x 2 q( x) = lim x 2 x→0

x→0

1 = 0. 2x( x − 2)

Since these limits are finite, x = 0 is a regular singular point. For x = 2 we have lim ( x − 2) p( x) = lim ( x − 2) x→2

x→2

3 3 = lim , 2( x − 2) 2( x − 2) 2 x→2

so the limit does not exist; hence x = 2 is an irregular singular point.

EXAMPLE 6 Determine the singular points of

x−

π 2

2 y + ( cos x) y + ( sin x) y = 0

and classify them as regular or irregular. Solution: The only singular point is x =

π . To study it, we consider the functions 2

π x− 2

p( x) =

π x− 2

Q( x) cos x = P( x) x − π/2

and

π 2

2

2

R( x) π = sin x. 2 P( x) π Starting from the Taylor series for cos x about x = , we find that 2 x−

q( x) =

x−

( x − π/2) 2 ( x − π/2) 4 cos x = −1 + − + ···, x − π/2 3! 5! which converges for all x. Similarly, sin x is analytic at x =

π π . Therefore, we conclude that is a 2 2

regular singular point for this equation.

Problems In each of Problems 1 through 8, determine the general solution of the given differential equation that is valid in any interval not including the singular point.

1. 2. 3. 4. 5. 6. 7. 8.

x 2 y + 4x y + 2y = 0 ( x + 1) 2 y + 3( x + 1) y + 0.75y = 0 x 2 y − 3x y + 4y = 0 x 2 y − x y + y = 0 x 2 y + 6x y − y = 0 2x 2 y − 4x y + 6y = 0 x 2 y − 5x y + 9y = 0 ( x − 2) 2 y + 5( x − 2) y + 8y = 0

In each of Problems 9 through 11, find the solution of the given initialvalue problem. Plot the graph of the solution and describe how the solution behaves as x → 0. G G G

9. 2x 2 y + x y − 3y = 0, y( 1) = 1, y ( 1) = 4 10. 4x 2 y + 8x y + 17y = 0, y( 1) = 2, y ( 1) = −3 11. x 2 y − 3x y + 4y = 0, y( −1) = 2, y ( −1) = 3

In each of Problems 12 through 23, find all singular points of the given equation and determine whether each one is regular or irregular.

12. x y + ( 1 − x) y + x y = 0 13. x 2 ( 1 − x) 2 y + 2x y + 4y = 0

Boyce 9131 Ch05 2

September 29, 2016

17:30

219

5.5 Series Solutions Near a Regular Singular Point, Part I

14. x 2 ( 1 − x) y + ( x − 2) y − 3x y = 0 2 15. x 2 ( 1 − x 2 ) y + y + 4y = 0

form

16. 17. 18. 19. 20. 21. 22. 23. 24.

one nonzero solution of this form in Problem 30 and that there are no nonzero solutions of this form in Problem 31. Thus in neither case can the general solution be found in this manner. This is typical of equations with singular points.

( 1 − x ) y + x( 1 − x) y + ( 1 + x) y = 0 x 2 y + x y + ( x 2 − ν 2 ) y = 0

(Bessel equation)

( x + 2) ( x − 1) y + 3( x − 1) y − 2( x + 2) y = 0 2

30. 2x y + 3y + x y = 0 31. 2x 2 y + 3x y − ( 1 + x) y = 0 32. Singularities at Infinity. The definitions of an ordinary point

x( 3 − x) y + ( x + 1) y − 2y = 0 x y + e x y + ( 3 cos x) y = 0 y + ( ln |x|) y + 3x y = 0

an x n . Show that (except for constant multiples) there is only

n=0

x

2 2

∞

and a regular singular point given in the preceding sections apply only if the point x0 is finite. In more advanced work in differential equations, it is often necessary to consider the point at infinity. This is done by making the change of variable ξ = 1/ x and studying the resulting equation at ξ = 0. Show that, for the differential equation

( sin x) y + x y + 4y = 0 ( x sin x) y + 3y + x y = 0

Find all values of α for which all solutions of 5 y = 0 approach zero as x → 0. 2 25. Find all values of β for which all solutions of x 2 y + β y = 0 approach zero as x → 0. x 2 y + α x y +

P( x) y + Q( x) y + R( x) y = 0, the point at infinity is an ordinary point if 1 P( 1/ξ )

26. Find γ so that the solution of the initial-value problem 2

x y − 2y = 0, y( 1) = 1, y ( 1) = γ is bounded as x → 0. conditions on α and β so that: a. All solutions approach zero as x → 0. b. All solutions are bounded as x → 0. c. All solutions approach zero as x → ∞. d. All solutions are bounded as x → ∞. e. All solutions are bounded both as x → 0 and as x → ∞.

ξ P( 1/ξ )

r (r − 1) + αr + β = 0, and x ln x are solutions of x 2 y +α x y +β y = 0 for x > 0. r1

29. Verify that W [x cos( μ ln x) , x sin( μ ln x) ] = μ x λ

2λ−1

33. 34. 35. 36. 37.

.

In each of Problems 30 and 31, show that the point x = 0 is a regular singular point. In each problem try to find solutions of the

2P( 1/ξ ) Q( 1/ξ ) − ξ ξ2

P( x) y + Q( x) y + R( x) y = 0

y − 2x y + λ y = 0

n=0

∞ n=0

and

R( 1/ξ ) ξ 2 P( 1/ξ )

y − x y = 0

qn x n ,

(Legendre equation)

(Hermite equation)

(Airy equation)

(1)

in the neighborhood of a regular singular point x = x0 . For convenience we assume that x0 = 0. If x0 = 0, the equation can be transformed into one for which the regular singular point is at the origin by letting x − x0 equal t. The assumption that x = 0 is a regular singular point of equation (1) means that x Q( x) / P( x) = x p( x) and x 2 R( x) / P( x) = x 2 q( x) have finite limits as x → 0 and are analytic at x = 0. Thus they have convergent power series expansions of the form x 2 q( x) =

( 1 − x 2 ) y − 2x y + α ( α + 1) y = 0

We now consider the question of solving the general second-order linear differential equation

pn x n ,

R( 1/ξ ) ξ 4 P( 1/ξ )

x 2 y + x y − 4y = 0

Series Solutions Near a Regular Singular Point, Part I

∞

and

y + y = 0

5.5

x p( x) =

In each of Problems 33 through 37, use the results of Problem 32 to determine whether the point at infinity is an ordinary point, a regular singular point, or an irregular singular point of the given differential equation.

repeated root of

λ

Q( 1/ξ ) 2P( 1/ξ ) − ξ ξ2

do have such expansions.

28. Using the method of reduction of order, show that if r1 is a

then x

have Taylor series expansions about ξ = 0. Show also that the point at infinity is a regular singular point if at least one of the above functions does not have a Taylor series expansion, but both

27. Consider the Euler equation x 2 y + α x y + β y = 0. Find

r1

219

(2)

Boyce 9131 Ch05 2

220

September 29, 2016

17:30

220

CHAPTER 5 Series Solutions of Second-Order Linear Equations

on some interval |x| < ρ about the origin, where ρ > 0. To make the quantities x p( x) and x 2 q( x) appear in equation (1), it is convenient to divide equation (1) by P( x) and then to multiply by x 2 , obtaining x 2 y + x( x p( x) ) y + x 2 q( x) y = 0, (3) or x 2 y + x( p0 + p1 x + · · · + pn x n + · · ·) y + ( q0 + q1 x + · · · + qn x n + · · ·) y = 0.

(4)

Notice that the first terms of x p( x) and of x 2 q( x) are p0 = lim x→0

x Q( x) x 2 R( x) and q0 = lim . P( x) x→0 P( x)

(5)

If all other coefficients pn and qn for n ≥ 1 in equation (2) are zero, then equation (4) reduces to the Euler equation x 2 y + p0 x y + q0 y = 0,

(6)

which was discussed in the preceding section. In general, of course, some of the coefficients pn and qn , n ≥ 1, are not zero. However, the essential character of solutions of equation (4) in the neighborhood of the singular point is identical to that of solutions of the Euler equation (6). The presence of the terms p1 x + · · · + pn x n + · · · and q1 x + · · · + qn x n + · · · merely complicates the calculations. We restrict our discussion primarily to the interval x > 0. The interval x < 0 can be treated, just as for the Euler equation, by making the change of variable x = −ξ and then solving the resulting equation for ξ > 0. The coefficients in equation (4) can be viewed as “Euler coefficients” times power series. To see this, you can write the coefficient of y in equation (4) as p1 p2 2 pn n p0 x 1 + x+ x + ··· + x + ··· , p0 p0 p0 and similarly for the coefficient of y. Thus it may seem natural to seek solutions of equation (4) in the form of “Euler solutions” times power series. Hence we assume that y = x r ( a0 + a1 x + · · · + an x n + · · ·) = x r

∞ n=0

an x n =

∞

an x r +n ,

(7)

n=0

where a0 = 0. In other words, r is the exponent of the first nonzero term in the series, and a0 is its coefficient. As part of the solution, we have to determine: 1. The values of r for which equation (1) has a solution of the form (7) 2. The recurrence relation for the coefficients an ∞ 3. The radius of convergence of the series an x n n=0

The general theory was constructed by Frobenius12 and is fairly complicated. Rather than trying to present this theory, we simply assume, in this and the next two sections, that there does exist a solution of the stated form. In particular, we assume that any power series in an expression for a solution has a nonzero radius of convergence and concentrate on showing how to determine the coefficients in such a series. To illustrate the method of Frobenius, we first consider an example. ......................................................................................................................................................................... 12 Ferdinand Georg Frobenius (1849--1917) grew up in the suburbs of Berlin, received his doctorate in 1870 from the University of Berlin, and returned as professor in 1892. For most of the intervening years he was professor at the Eidgenössische Polytechnikum at Zürich. He showed how to construct series solutions about regular singular points in 1874. His most distinguished work, however, was in algebra, where he was one of the foremost early developers of group theory.

Boyce 9131 Ch05 2

September 29, 2016

17:30

221

5.5 Series Solutions Near a Regular Singular Point, Part I

EXAMPLE 1 Solve the differential equation 2x 2 y − x y + ( 1 + x) y = 0.

(8)

Solution: It is easy to show that x = 0 is a regular singular point of equation (8). Further, x p( x) = −1/2 and x 2 q( x) = ( 1 + x) /2. Thus p0 = −1/2, q0 = 1/2, q1 = 1/2, and all other pn ’s and qn ’s are zero. Then, from equation (6), the Euler equation corresponding to equation (8) is 2x 2 y − x y + y = 0.

(9)

To solve equation (8), we assume that there is a solution of the form (7). Then y and y are given by

∞

an (r + n) x r +n−1

(10)

an (r + n) (r + n − 1) x r +n−2 .

(11)

y =

n=0

and

y =

∞ n=0

By substituting the expressions for y, y , and y in equation (8), we obtain 2x 2 y − x y + ( 1 + x) y =

∞

2an (r + n) (r + n − 1) x r +n

n=0

−

∞

an (r + n) x r +n +

∞

n=0

an x r +n +

n=0

The last term in equation (12) can be written as

∞

∞

an x r +n+1 .

(12)

n=0

an−1 x r +n , so by combining the terms in

n=1

equation (12), we obtain 2x 2 y − x y + ( 1 + x) y = a0 [2r (r − 1) − r + 1]x r +

∞

( 2(r + n) (r + n − 1) − (r + n) + 1) an + an−1 x r +n = 0.

(13)

n=1

If equation (13) is to be satisfied for all x, the coefficient of each power of x in equation (13) must be zero. From the coefficient of x r we obtain, since a0 = 0, 2r (r − 1) − r + 1 = 2r 2 − 3r + 1 = (r − 1) ( 2r − 1) = 0.

(14)

Equation (14) is called the indicial equation for equation (8). Note that it is exactly the polynomial equation we would obtain for the Euler equation (9) associated with equation (8). The roots of the indicial equation are 1 r1 = 1, r2 = . (15) 2 These values of r are called the exponents at the singularity for the regular singular point x = 0. They determine the qualitative behavior of the solution (7) in the neighborhood of the singular point. Now we return to equation (13) and set the coefficient of x r +n equal to zero. This gives the relation ( 2(r + n) (r + n − 1) − (r + n) + 1) an + an−1 = 0,

n ≥ 1,

(16)

or an = − =−

▼

an−1 2(r +

n) 2

− 3(r + n) + 1

an−1 , ( (r + n) − 1) ( 2(r + n) − 1)

n ≥ 1.

(17)

221

Boyce 9131 Ch05 2

222

September 29, 2016

17:30

222

CHAPTER 5 Series Solutions of Second-Order Linear Equations

▼ For each root r1 and r2 of the indicial equation, we use the recurrence relation (17) to determine a set of coefficients a1 , a2 , . . . . For r = r1 = 1, equation (17) becomes an−1 an = − , n ≥ 1. ( 2n + 1) n

Thus a1 = −

a0 , 3·1

a2 = −

a1 a0 = , 5·2 ( 3 · 5) ( 1 · 2)

and a3 = −

a2 a0 =− . 7·3 ( 3 · 5 · 7) ( 1 · 2 · 3)

In general, we have an =

( −1) n a0 , ( 3 · 5 · 7 · · · ( 2n + 1) ) n!

n ≥ 4.

(18)

If we multiply both the numerator and denominator of the right-hand side of equation (18) by 2 · 4 · 6 · · · · · 2n = 2n n!, we can rewrite an as an =

( −1) n 2n a0 , ( 2n + 1) !

n ≥ 1.

Hence, if we omit the constant multiplier a0 , one solution of equation (8) is

⎛

y1 ( x) = x ⎝1 +

⎞

∞ ( −1) n 2n n=1

( 2n + 1) !

x n ⎠,

x > 0.

(19)

To determine the radius of convergence of the series in equation (19), we use the ratio test: an+1 x n+1 2|x| = lim =0 lim an x n n→∞ ( 2n + 2) ( 2n + 3) n→∞ for all x. Thus the series converges for all x. Corresponding to the second root r = r2 =

1 , we proceed similarly. From equation (17) we 2

have an = −

a

n−1 = −

2n n −

1 2

an−1 , n( 2n − 1)

n ≥ 1.

Hence a0 , 1·1 a1 a0 a2 = − = , 2·3 ( 1 · 2) ( 1 · 3) a2 a0 a3 = − =− , 3·5 ( 1 · 2 · 3) ( 1 · 3 · 5) a1 = −

and, in general, an =

( −1) n a0 , n!( 1 · 3 · 5 · · · ( 2n − 1) )

n ≥ 4.

(20)

Just as in the case of the first root r1 , we multiply the numerator and denominator by 2 · 4 · 6 · · · · · 2n = 2n n!. Then we have an =

( −1) n 2n a0 , ( 2n) !

n ≥ 1.

Again omitting the constant multiplier a0 , we obtain the second solution

⎛

y2 ( x) = x 1/2 ⎝1 +

∞ ( −1) n 2n n=1

▼

( 2n) !

⎞

x n ⎠,

x > 0.

(21)

Boyce 9131 Ch05 2

September 29, 2016

17:30

223

5.5 Series Solutions Near a Regular Singular Point, Part I

▼ As before, we can show that the series in equation (21) converges for all x. Since y1 and y2 behave like

x and x 1/2 , respectively, near x = 0, they are linearly independent and so they form a fundamental set of solutions. Hence the general solution of equation (8) is y = c1 y1 ( x) + c2 y2 ( x) ,

x > 0.

The preceding example illustrates that if x = 0 is a regular singular point, then sometimes there are two solutions of the form (7) in the neighborhood of this point. Similarly, if there is a regular singular point at x = x0 , then there may be two solutions of the form y = ( x − x0 ) r

∞

an ( x − x 0 ) n

(22)

n=0

that are valid near x = x0 . However, just as an Euler equation may not have two solutions of the form y = x r , so a more general equation with a regular singular point may not have two solutions of the form (7) or (22). In particular, we show in the next section that if the roots r1 and r2 of the indicial equation are equal or differ by an integer, then the second solution normally has a more complicated structure. In all cases, though, it is possible to find at least one solution of the form (7) or (22); if r1 and r2 differ by an integer, this solution corresponds to the larger value of r . If there is only one such solution, then the second solution involves a logarithmic term, just as for the Euler equation when the roots of the characteristic equation are equal. The method of reduction of order or some other procedure can be invoked to determine the second solution in such cases. This is discussed in Sections 5.6 and 5.7. If the roots of the indicial equation are complex, then they cannot be equal or differ by an integer, so there are always two solutions of the form (7) or (22). Of course, these solutions are complex-valued functions of x. However, as for the Euler equation, it is possible to obtain real-valued solutions by taking the real and imaginary parts of the complex solutions. Finally, we mention a practical point. If P, Q, and R are polynomials, it is often much better to work directly with equation (1) than with equation (3). This avoids the necessity of expressing x Q( x) / P( x) and x 2 R( x) / P( x) as power series. For example, it is more convenient to consider the equation x( 1 + x) y + 2y + x y = 0 than to write it in the form x 2 y +

which would entail expanding

2x x2 y + y = 0, 1+x 1+x

2x x2 and in power series. 1+x 1+x

Problems In each of Problems 1 through 6: a. Show that the given differential equation has a regular singular point at x = 0. b. Determine the indicial equation, the recurrence relation, and the roots of the indicial equation. c. Find the series solution ( x > 0) corresponding to the larger root. d. If the roots are unequal and do not differ by an integer, find the series solution corresponding to the smaller root also.

1. 2x y + y + x y = 0 1 2 2 2. x y + x y + x − y=0 9

3. 4. 5. 6.

xy + y = 0 x y + y − y = 0 x 2 y + x y + ( x − 2) y = 0 x y + ( 1 − x) y − y = 0

223

Boyce 9131 Ch05 2

224

September 29, 2016

17:30

224

CHAPTER 5 Series Solutions of Second-Order Linear Equations

a. Show that x = 0 is a regular singular point. b. Show that the roots of the indicial equation are r1 = r2 = 0. c. Show that one solution for x > 0 is

7. The Legendre equation of order α is ( 1 − x 2 ) y − 2x y + α ( α + 1) y = 0. The solution of this equation near the ordinary point x = 0 was discussed in Problems 17 and 18 of Section 5.3. In Example 4 of Section 5.4, it was shown that x = ±1 are regular singular points. a. Determine the indicial equation and its roots for the point x = 1. b. Find a series solution in powers of x − 1 for x − 1 > 0. Hint: Write 1 + x = 2 + ( x − 1) and x = 1 + ( x − 1) . Alternatively, make the change of variable x − 1 = t and determine a series solution in powers of t.

8. The Chebyshev equation is

J0 ( x) = 1 +

∞ ( −1) n x 2n n=1

.

The function J0 is known as the Bessel function of the first kind of order zero. d. Show that the series for J0 ( x) converges for all x.

11. Referring to Problem 10, use the method of reduction of order to show that the second solution of the Bessel equation of order zero contains a logarithmic term. Hint: If y2 ( x) = J0 ( x) v( x) , then

( 1 − x 2 ) y − x y + α 2 y = 0,

y2 ( x) = J0 ( x)

where α is a constant; see Problem 8 of Section 5.3. a. Show that x = 1 and x = −1 are regular singular points, and find the exponents at each of these singularities. b. Find two solutions about x = 1.

22n ( n!) 2

dx

x J0 ( x)

2 .

1 Find the first term in the series expansion of 2 . x J0 ( x)

12. The Bessel equation of order one is

9. The Laguerre13 differential equation is

x 2 y + x y + ( x 2 − 1) y = 0.

x y + ( 1 − x) y + λ y = 0.

a. Show that x = 0 is a regular singular point. b. Show that the roots of the indicial equation are r1 = 1 and

a. Show that x = 0 is a regular singular point. b. Determine the indicial equation, its roots, and the recurrence

r2 = −1. c. Show that one solution for x > 0 is

relation. c. Find one solution (for x > 0). Show that if λ = m, a positive integer, this solution reduces to a polynomial. When properly normalized, this polynomial is known as the Laguerre polynomial, L m ( x) .

∞ x ( −1) n x 2n . 2 ( n + 1) ! n! 22n

J1 ( x) =

n=0

10. The Bessel equation of order zero is

The function J1 is known as the Bessel function of the first kind of order one. d. Show that the series for J1 ( x) converges for all x. e. Show that it is impossible to determine a second solution of the form

x 2 y + x y + x 2 y = 0. .............................................................................................................................. 13 Edmond

Nicolas Laguerre (1834--1886), a French geometer and analyst, studied the polynomials named for him about 1879. He is also known for an algorithm for calculating roots of polynomial equations.

x −1

∞

bn x n ,

x > 0.

n=0

Series Solutions Near a Regular Singular Point, Part II 5.6

Now let us consider the general problem of determining a solution of the equation L[y] = x 2 y + x( x p( x) ) y + x 2 q( x) y = 0,

(1)

where x p( x) =

∞ n=0

pn x n ,

x 2 q( x) =

∞

qn x n ,

(2)

n=0

and both series converge in an interval |x| < ρ for some ρ > 0. The point x = 0 is a regular singular point, and the corresponding Euler equation is x 2 y + p0 x y + q0 y = 0.

(3)

Boyce 9131 Ch05 2

September 29, 2016

17:30

225

5.6 Series Solutions Near a Regular Singular Point, Part II

We seek a solution of equation (1) for x > 0 and assume that it has the form y = φ (r, x) = x r

∞

an x n =

n=0

∞

an x r +n ,

(4)

n=0

where a0 = 0, and we have written y = φ (r, x) to emphasize that φ depends on r as well as x. It follows that y =

∞

(r + n) an x r +n−1 ,

y =

n=0

∞

(r + n) (r + n − 1) an x r +n−2 .

(5)

n=0

Then, substituting from equations (2), (4), and (5) in equation (1) gives L[φ ](r, x) = a0r (r − 1) x r + a1 (r + 1)r x r +1 + · · · + an (r + n) (r + n − 1) x r +n + · · ·

+ p0 + p1 x + · · · + pn x n + · · · a0r x r + a1 (r + 1) x r +1 + · · · + an (r + n) x r +n + · · ·

+ q0 + q1 x + · · · + qn x n + · · · a0 x r + a1 x r +1 + · · · + an x r +n + · · · = 0. Multiplying the infinite series together and then collecting terms, we obtain L[φ ](r, x) = a0 F(r ) x r + a1 F(r + 1) + a0 ( p1r + q1 ) x r +1

+ a2 F(r + 2) + a0 ( p2r + q2 ) + a1 p1 (r + 1) + q1 x r +2

+ · · · + an F(r + n) + a0 ( pn r + qn ) + a1 pn−1 (r + 1) + qn−1

+ · · · + an−1 p1 (r + n − 1) + q1 x r +n + · · · = 0, or, in a more compact form, L[φ ] = a0 F(r ) x r ⎛ ⎞ n−1 ∞ ⎝ F(r + n) an + + ak ( (r + k) pn−k + qn−k ) ⎠x r +n = 0, n=1

(6)

k=0

where F(r ) = r (r − 1) + p0r + q0 .

(7)

For equation (6) to be satisfied for all x > 0, the coefficient of each power of x must be zero. 0, the term involving x r yields the equation F(r ) = 0. This equation is Since a0 = called the indicial equation; note that it is exactly the equation we would obtain in looking for solutions y = x r of the Euler equation (3). Let us denote the roots of the indicial equation by r1 and r2 with r1 ≥ r2 if the roots are real. If the roots are complex, the designation of the roots is immaterial. Only for these values of r can we expect to find solutions of equation (1) of the form (4). The roots r1 and r2 are called the exponents at the singularity; they determine the qualitative nature of the solution in the neighborhood of the singular point. Setting the coefficient of x r +n in equation (6) equal to zero gives the recurrence relation F(r + n) an +

n−1

ak ( (r + k) pn−k + qn−k ) = 0,

n ≥ 1.

(8)

k=0

Equation (8) shows that, in general, an depends on the value of r and all the preceding coefficients a0 , a1 , . . . , an−1 . It also shows that we can successively compute a1 , a2 , . . . , an , . . . in terms of a0 and the coefficients in the series for x p( x) and x 2 q( x) , provided that F(r + 1) , F(r + 2) , . . . , F(r + n) , . . . are not zero. The only values of r for which F(r ) = 0 are r = r1 and r = r2 ; since r1 ≥ r2 , it follows that r1 + n is not equal to r1 or r2 for n ≥ 1. Consequently, F(r1 + n) = 0 for n ≥ 1. Hence we can always determine one solution of equation (1) in the form (4), namely, ⎛ ⎞ ∞ y1 ( x) = x r1 ⎝1 + an (r1 ) x n ⎠, x > 0. (9) n=1

225

Boyce 9131 Ch05 2

226

September 29, 2016

17:30

226

CHAPTER 5 Series Solutions of Second-Order Linear Equations

Here we have introduced the notation an (r1 ) to indicate that an has been determined from equation (8) with r = r1 . The solution involves an arbitrary constant; the solution in equation (9) is obtained by assigning a0 the value 1. If r2 is not equal to r1 , and r1 − r2 is not a positive integer, then r2 + n is not equal to r1 for any value of n ≥ 1; hence F(r2 + n) = 0, and we can also obtain a second solution ⎛ ⎞ ∞ y2 ( x) = x r2 ⎝1 + an (r2 ) x n ⎠, x > 0. (10) n=1

Just as for the series solutions about ordinary points discussed in Section 5.3, the series in equations (9) and (10) converge at least in the interval |x| < ρ where the series for both x p( x) ∞ and x 2 q( x) converge. Within their radii of convergence, the power series 1 + an (r1 ) x n n=1 ∞ and 1 + an (r2 ) x n define functions that are analytic at x = 0. Thus the singular behavior, n=1

if there is any, of the solutions y1 and y2 is due to the factors x r1 and x r2 that multiply these two analytic functions. Next, to obtain real-valued solutions for x < 0, we can make the substitution x = −ξ with ξ > 0. As we might expect from our discussion of the Euler equation, it turns out that we need only replace x r1 in equation (9) and x r2 in equation (10) by |x|r1 and |x|r2 , respectively. Finally, note that if r1 and r2 are complex numbers, then they are necessarily complex conjugates and r2 = r1 + N for any positive integer N . Thus, in this case we can always find two series solutions of the form (4); however, they are complex-valued functions of x. Realvalued solutions can be obtained by taking the real and imaginary parts of the complex-valued solutions. The exceptional cases in which r1 = r2 or r1 − r2 = N , where N is a positive integer, require more discussion and will be considered later in this section. It is important to realize that r1 and r2 , the exponents at the singular point, are easy to find and that they determine the qualitative behavior of the solutions. To calculate r1 and r2 , it is only necessary to solve the quadratic indicial equation r (r − 1) + p0r + q0 = 0, whose coefficients are given by p0 = lim x p( x) ,

(11)

q0 = lim x 2 q( x) .

x→0

(12)

x→0

Note that these are exactly the limits that must be evaluated in order to classify the singularity as a regular singular point; thus they have usually been determined at an earlier stage of the investigation. Further, if x = 0 is a regular singular point of the equation P( x) y + Q( x) y + R( x) y = 0,

(13)

where the functions P, Q, and R are polynomials, then x p( x) = x Q( x) / P( x) and x 2 q( x) = x 2 R( x) / P( x) . Thus p0 = lim x x→0

Q( x) , P( x)

q0 = lim x 2 x→0

R( x) . P( x)

(14)

Finally, the radii of convergence for the series in equations (9) and (10) are at least equal to the distance from the origin to the nearest zero of P other than the regular singular point x = 0 itself.

EXAMPLE 1 Discuss the nature of the solutions of the equation 2x( 1 + x) y + ( 3 + x) y − x y = 0 near the singular points.

▼

Boyce 9131 Ch05 2

September 29, 2016

17:30

227

5.6 Series Solutions Near a Regular Singular Point, Part II

▼ Solution: This equation is of the form (13) with P( x) = 2x( 1 + x) , Q( x) = 3 + x, and R( x) = −x. The points x = 0 and x = −1 are the only singular points. The point x = 0 is a regular singular point, since lim x x→0

lim x 2 x→0

Further, from equation (14), p0 =

Q( x) 3+x 3 = lim x = , P( x) 2 x→0 2x( 1 + x) R( x) −x = lim x 2 = 0. P( x) 2x( 1 + x) x→0

3 3 and q0 = 0. Thus the indicial equation is r (r − 1) + r = 0, 2 2

1 and the roots are r1 = 0, r2 = − . Since these roots are not equal and do not differ by an integer, 2 there are two solutions of the form y1 ( x) = 1 +

∞

⎛

an ( 0) x n and y2 ( x) = |x|−1/2 ⎝1 +

n=1

∞

⎞

an − 12 x n ⎠

n=1

for 0 < |x| < ρ. A lower bound for the radius of convergence of each series is 1, the distance from x = 0 to x = −1, the other zero of P( x) . Note that the solution y1 is bounded as x → 0, indeed is analytic there, and that the second solution y2 is unbounded as x → 0. The point x = −1 is also a regular singular point, since lim ( x + 1) x→−1

lim ( x + 1) 2 x→−1

Q( x) ( x + 1) ( 3 + x) = lim = −1, P( x) 2x( 1 + x) x→−1 R( x) ( x + 1) 2 ( −x) = lim = 0. P( x) 2x( 1 + x) x→−1

In this case p0 = −1, q0 = 0, so the indicial equation is r (r − 1) − r = 0. The roots of the indicial equation are r1 = 2 and r2 = 0. Corresponding to the larger root there is a solution of the form

⎛

y1 ( x) = ( x + 1) 2 ⎝1 +

∞

⎞

an ( 2) ( x + 1) n ⎠.

n=1

The series converges at least for |x + 1| < 1, and y1 is an analytic function there. Since the two roots differ by a positive integer, there may or may not be a second solution of the form y2 ( x) = 1 +

∞

an ( 0) ( x + 1) n .

n=1

We cannot say more without further analysis. Observe that no complicated calculations were required to discover the information about the solutions presented in this example. All that was needed was to evaluate a few limits and solve two quadratic equations.

We now consider the cases in which the roots of the indicial equation are equal or differ by a positive integer, r1 − r2 = N . As we have shown earlier, there is always one solution of the form (9) corresponding to the larger root r1 of the indicial equation. By analogy with the Euler equation, we might expect that if r1 = r2 , then the second solution contains a logarithmic term. This may also be true if the roots differ by an integer. Equal Roots. The method of finding the second solution is essentially the same as the one we used in finding the second solution of the Euler equation (see Section 5.4) when the roots of the indicial equation were equal. We consider r to be a continuous variable and determine an as a function of r by solving the recurrence relation (8). For this choice of an (r ) for n ≥ 1, the terms in equation (6) involving x r +1 , x r +2 , x r +3 , . . . all have coefficients equal to zero. Therefore, since r1 is a repeated root of F(r ) , equation (6) reduces to L[φ ](r, x) = a0 F(r ) x r = a0 (r − r1 ) 2 x r .

(15)

227

Boyce 9131 Ch05 2

228

September 29, 2016

17:30

228

CHAPTER 5 Series Solutions of Second-Order Linear Equations

Setting r = r1 in equation (15), we find that L[φ ](r1 , x) = 0; hence, as we already know, y1 ( x) given by equation (9) is one solution of equation (1). But more important, it also follows from equation (15), just as for the Euler equation, that ∂φ ∂ r L (r1 , x) = a0 x (r − r1 ) 2 r =r1 ∂r ∂r = a0 (r − r1 ) 2 x r ln x + 2(r − r1 ) x r = 0. (16) r =r1

Hence, a second solution of equation (1) is ⎞⎞ ⎛ ⎛ ∞ ∂ φ (r, x) ∂ ⎝ r⎝ n ⎠⎠ y2 ( x) = = an (r ) x x a0 + ∂r ∂r n=1

r =r1

⎛ = ( x r1 ln x) ⎝a0 +

∞

r =r1

⎞ an (r1 ) x n ⎠ + x r1

n=1

= y1 ( x) ln x + x r1

∞

∞

an (r1 ) x n

n=1

an (r1 ) x n ,

x > 0,

(17)

n=1

dan evaluated at r = r1 . where an (r1 ) denotes dr Although equation (17) provides an explicit expression for a second solution y2 ( x) , it may turn out that it is difficult to determine an (r ) as a function of r from the recurrence relation (8) and then to differentiate the resulting expression with respect to r . An alternative is simply to assume that y has the form of equation (17). That is, assume that y = y1 ( x) ln x + x r1

∞

bn x n ,

x > 0,

(18)

n=1

where y1 ( x) has already been found. The coefficients bn are calculated, as usual, by substituting into the differential equation, collecting terms, and setting the coefficient of each power of x equal to zero. A third possibility is to use the method of reduction of order to find y2 ( x) once y1 ( x) is known. Roots r1 and r2 Differing by an Integer N. For this case the derivation of the second solution is considerably more complicated and will not be given here. The form of this solution is stated in equation (24) in the following theorem. The coefficients cn (r2 ) in equation (24) are given by d cn (r2 ) = (19) [(r − r2 ) an (r ) ] , n = 1, 2, . . . , dr r =r2 where an (r ) is determined from the recurrence relation (8) with a0 = 1. Further, the coefficient a in equation (24) is a = lim (r − r2 ) a N (r ) . (20) r →r2

If a N (r2 ) is finite, then a = 0 and there is no logarithmic term in y2 . A full derivation of formulas (19) and (20) may be found in Coddington (Chapter 4). In practice, the best way to determine whether a is zero in the second solution is simply to try to compute the an corresponding to the root r2 and to see whether it is possible to determine 0. a N (r2 ) . If so, there is no further problem. If not, we must use the form (24) with a = When r1 − r2 = N , there are again three ways to find a second solution. First, we can calculate a and cn (r2 ) directly by substituting the expression (24) for y in equation (1). Second, we can calculate cn (r2 ) and a of equation (24) using the formulas (19) and (20). If this is the planned procedure, then in calculating the solution corresponding to r = r1 , be sure to obtain the general formula for an (r ) rather than just an (r1 ) . The third alternative is to use the method of reduction of order. The following theorem summarizes the results that we have obtained in this section.

Boyce 9131 Ch05 2

September 29, 2016

17:30

229

5.6 Series Solutions Near a Regular Singular Point, Part II

229

Theorem 5.6.1 Consider the differential equation (1)

x 2 y + x( x p( x) ) y + x 2 q( x) y = 0, where x = 0 is a regular singular point. Then x p( x) and x 2 q( x) are analytic at x = 0 with convergent power series expansions x p( x) =

∞

pn x n ,

x 2 q( x) =

n=0

∞

qn x n

n=0

for |x| < ρ, where ρ > 0 is the minimum of the radii of convergence of the power series for x p( x) and x 2 q( x) . Let r1 and r2 be the roots of the indicial equation F(r ) = r (r − 1) + p0 r + q0 = 0, with r1 ≥ r2 if r1 and r2 are real. Then in either the interval −ρ < x < 0 or the interval 0 < x < ρ, there exists a solution of the form ⎛ ⎞ y1 ( x) = |x| 1 ⎝1 + r

∞

an (r1 ) x n ⎠,

(21)

n=1

where the an (r1 ) are given by the recurrence relation (8) with a0 = 1 and r = r1 . CASE 1 If r1 − r2 is not zero or a positive integer, then in either the interval −ρ < x < 0 or the interval 0 < x < ρ, there exists a second solution of the form

⎛

y2 ( x) = |x| 2 ⎝1 + r

∞

⎞

an (r2 ) x n ⎠.

(22)

n=1

The an (r2 ) are also determined by the recurrence relation (8) with a0 = 1 and r = r2 . The power series in equations (21) and (22) converge at least for |x| < ρ. CASE 2 If r1 = r2 , then the second solution is r

y2 ( x) = y1 ( x) ln |x| + |x| 1

∞

bn (r1 ) x n .

(23)

n=1

CASE 3 If r1 − r2 = N , a positive integer, then

⎛ r2 ⎝

y2 ( x) = ay1 ( x) ln |x| + |x|

1+

∞

⎞ cn (r2 ) x n ⎠.

(24)

n=1

The coefficients an (r1 ) , bn (r1 ) , and cn (r2 ) and the constant a can be determined by substituting the form of the series solutions for y in equation (1). The constant a may turn out to be zero, in which case there is no logarithmic term in the solution (24). Each of the series in equations (23) and (24) converges at least for |x| < ρ and defines a function that is analytic in some neighborhood of x = 0. In all three cases, the two solutions y1 ( x) and y2 ( x) form a fundamental set of solutions of the given differential equation.

Problems In each of Problems 1 through 8: a. Find all the regular singular points of the given differential equation. b. Determine the indicial equation and the exponents at the singularity for each regular singular point. x y + 2x y + 6e x y = 0 x 2 y − x( 2 + x) y + ( 2 + x 2 ) y = 0 y + 4x y + 6y = 0 2x( x + 2) y + y − x y = 0 1 5. x 2 y + ( x + sin x) y + y = 0 2

1. 2. 3. 4.

6. x 2 ( 1 − x) y − ( 1 + x) y + 2x y = 0 7. ( x − 2) 2 ( x + 2) y + 2x y + 3( x − 2) y = 0 8. ( 4 − x 2 ) y + 2x y + 3y = 0 In each of Problems 9 through 12: a. Show that x = 0 is a regular singular point of the given differential equation. b. Find the exponents at the singular point x = 0. c. Find the first three nonzero terms in each of two solutions (not multiples of each other) about x = 0.

9. x y + y − y = 0 10. x y + 2x y + 6e x y = 0 ( see Problem 1)

Boyce 9131 Ch05 2

230

September 29, 2016

17:30

230

CHAPTER 5 Series Solutions of Second-Order Linear Equations

11. x y + y = 0 12. x 2 y + ( sin x) y − ( cos x) y = 0 13. a. Show that ( ln x) y +

15. Consider the differential equation x 3 y + α x y + β y = 0, where α and β are real constants and α = 0. a. Show that x = 0 is an irregular singular point.

1 y +y=0 2

has a regular singular point at x = 1. b. Determine the roots of the indicial equation at x = 1. c. Determine the first three nonzero terms in the series ∞

an ( x − 1) r +n corresponding to the larger root.

n=0

You can assume x − 1 > 0. d. What would you expect the radius of convergence of the series to be?

14. In several problems in mathematical physics, it is necessary to

x( 1 − x) y + ( γ − ( 1 + α + β ) x) y − αβ y = 0,

an x r +n ,

n=0

show that the indicial equation for r is linear and that, consequently, there is only one formal solution of the assumed form. c. Show that if β /α = −1, 0, 1, 2, . . . , then the formal series solution terminates and therefore is an actual solution. For other values of β /α , show that the formal series solution has a zero radius of convergence and so does not represent an actual solution in any interval.

(25)

where α , β , and γ are constants. This equation is known as the hypergeometric equation. a. Show that x = 0 is a regular singular point and that the roots of the indicial equation are 0 and 1 − γ . b. Show that x = 1 is a regular singular point and that the roots of the indicial equation are 0 and γ − α − β . c. Assuming that 1 − γ is not a positive integer, show that, in the neighborhood of x = 0, one solution of equation (25) is α ( α + 1) β ( β + 1) 2 αβ y1 ( x) = 1 + x+ x + ···. γ · 1! γ ( γ + 1) 2!

( α − γ + 1) ( β − γ + 1) x+ ( 2 − γ ) 1!

( α − γ + 1) ( α − γ + 2) ( β − γ + 1) ( β − γ + 2) 2 x +· · · . ( 2 − γ ) ( 3 − γ ) 2!

e. Show that the point at infinity is a regular singular point and that the roots of the indicial equation are α and β . See Problem 32 of Section 5.4.

5.7

y +

α β y + t y = 0, xs x

(26)

where α = 0 and β = 0 are real numbers, and s and t are positive integers that for the moment are arbitrary. a. Show that if s > 1 or t > 2, then the point x = 0 is an irregular singular point. b. Try to find a solution of equation (26) of the form y=

∞

an x r +n ,

x > 0.

(27)

n=0

What would you expect the radius of convergence of this series to be? d. Assuming that 1 − γ is not an integer or zero, show that a second solution for 0 < x < 1 is

∞

16. Consider the differential equation

study the differential equation

y2 ( x) = x 1−γ 1 +

b. By attempting to determine a solution of the form

Show that if s = 2 and t = 2, then there is only one possible value of r for which there is a formal solution of equation (26) of the form (27). c. Show that if s = 1 and t = 3, then there are no solutions of equation (26) of the form (27). d. Show that the maximum values of s and t for which the indicial equation is quadratic in r [and hence we can hope to find two solutions of the form (27)] are s = 1 and t = 2. These are precisely the conditions that distinguish a “weak singularity,” or a regular singular point, from an irregular singular point, as we defined them in Section 5.4. As a note of caution, we point out that although it is sometimes possible to obtain a formal series solution of the form (27) at an irregular singular point, the series may not have a positive radius of convergence. See Problem 15 for an example.

Bessel’s Equation

In this section we illustrate the discussion in Section 5.6 by considering three special cases of Bessel’s14 equation, x 2 y + x y + ( x 2 − ν 2 ) y = 0,

(1)

......................................................................................................................................................................... 14 Friedrich Wilhelm Bessel (1784--1846) left school at the age of 14 to embark on a career in the import-export business but soon became interested in astronomy and mathematics. He was appointed director of the observatory at Königsberg in 1810 and held this position until his death. His study of planetary perturbations led him in 1824 to make the first systematic analysis of the solutions, known as Bessel functions, of equation (1). He is also famous for making, in 1838, the first accurate determination of the distance from the earth to a star.

Boyce 9131 Ch05 2

September 29, 2016

17:30

231

5.7 Bessel’s Equation

where ν is a constant. It is easy to show that x = 0 is a regular singular point of equation (1). We have Q( x) 1 p0 = lim x = lim x = 1, x→0 P( x) x→0 x q0 = lim x 2 x→0

R( x) x2 − ν 2 = −ν 2 . = lim x 2 P( x) x2 x→0

Thus the indicial equation is F(r ) = r (r − 1) + p0r + q0 = r (r − 1) + r − ν 2 = r 2 − ν 2 = 0, 1 , and ν = 1 for the 2 interval x > 0. Bessel functions will reappear in Sections 11.4 and 11.5. with the roots r = ±ν . We will consider the three cases ν = 0, ν =

Bessel Equation of Order Zero. In this case ν = 0, so differential equation (1) reduces to L[y] = x 2 y + x y + x 2 y = 0,

(2)

and the roots of the indicial equation are equal: r1 = r2 = 0. Substituting y = φ (r, x) = a0 x r +

∞

an x r +n

(3)

n=1

in equation (2), we obtain ∞ ∞ L[φ ](r, x) = an ( (r + n) (r + n − 1) + (r + n) ) x r +n + an x r +n+2 n=0

n=0

= a0 (r (r − 1) + r ) x r + a1 ( (r + 1)r + (r + 1) ) x r +1 +

∞

an ( (r + n) (r + n − 1) + (r + n) ) + an−2 x r +n = 0.

(4)

n=2

As we have already noted, the roots of the indicial equation F(r ) = r (r − 1) + r = 0 are r1 = 0 and r2 = 0. The recurrence relation is an (r ) = −

an−2 (r ) an−2 (r ) , =− (r + n) (r + n − 1) + (r + n) (r + n) 2

n ≥ 2.

(5)

To determine y1 ( x) , we set r equal to 0. Then, from equation (4), it follows that for the coefficient of x r +1 to be zero we must choose a1 = 0. Hence, from equation (5), a3 = a5 = a7 = · · · = 0. Further, an ( 0) = −

an−2 ( 0) , n2

n = 2, 4, 6, 8, . . . ,

or, letting n = 2m, we obtain a2m ( 0) = −

a2m−2 ( 0) , ( 2m) 2

m = 1, 2, 3, . . . .

Thus a2 ( 0) = −

a0 , 22

a4 ( 0) =

a0 , 24 22

a6 ( 0) = −

a0 , 26 ( 3 · 2) 2

and, in general, a2m ( 0) = Hence

( −1) m a0 , 22m ( m!) 2

⎛ y1 ( x) = a0 ⎝1 +

m = 1, 2, 3, . . . .

∞ ( −1) m x 2m m=1

22m ( m!) 2

(6)

⎞ ⎠,

x > 0.

(7)

231

Boyce 9131 Ch05 2

232

September 29, 2016

17:30

232

CHAPTER 5 Series Solutions of Second-Order Linear Equations

The function in brackets is known as the Bessel function of the first kind of order zero and is denoted by J0 ( x) . It follows from Theorem 5.6.1 that the series converges for all x and that J0 is analytic at x = 0. Some of the important properties of J0 are discussed in the problems. Figure 5.7.1 shows the graphs of y = J0 ( x) and some of the partial sums of the series (7). y n = 4 n = 8 n = 12

2

n = 16 n = 20

1

2

4

6

8

10

x

y = J0(x) –1 n=2

n=6

n = 10

n = 14

n = 18

FIGURE 5.7.1 Polynomial approximations to J0 ( x) , the Bessel function of the first kind of order zero. The value of n is the degree of the approximating polynomial.

To determine y2 ( x) we will use equation (17) in Section 5.6. This requires that we calculate15 an ( 0) . First we note from the coefficient of x r +1 in differential equation (4) that (r + 1) 2 a1 (r ) = 0. Thus a1 (r ) = 0 for all r near r = 0. So not only does a1 ( 0) = 0 but also a1 ( 0) = 0. From the recurrence relation (5) it follows that a3 ( 0) = a5 ( 0) = · · · = a2n+1 ( 0) = · · · = 0; hence we need only compute a2m ( 0) , m = 1, 2, 3, . . . . From equation (5) we have

a2m (r ) = −

a2m−2 (r ) (r + 2m) 2

m = 1, 2, 3, . . . .

By solving this recurrence relation, we obtain a0 a0 a2 (r ) = − , a4 (r ) = , 2 (r + 2) (r + 2) 2 (r + 4) 2 and, in general, a2m (r ) =

( −1) m a0 , (r + 2) 2 · · · (r + 2m) 2

m ≥ 3.

(8)

The computation of a2m (r ) can be carried out most conveniently by noting that if

f ( x) = ( x − α 1 ) β 1 ( x − α 2 ) β 2 ( x − α 3 ) β 3 · · · ( x − α n ) β n , and if x is not equal to α 1 , α 2 , . . . , α n , then β2 β1 βn f ( x) + + ··· + . = f ( x) x − α1 x − α2 x − αn Applying this result to a2m (r ) from equation (8), we find that a2m (r ) 1 1 1 = −2 + + ··· + , a2m (r ) r +2 r +4 r + 2m and setting r equal to 0, we obtain a2m ( 0)

1 1 1 = −2 + + ··· + a2m ( 0) . 2 4 2m

......................................................................................................................................................................... 15 Problem 9 outlines an alternative procedure, in which we simply substitute the form (23) of Section 5.6 in equation (2) and then determine the bn .

Boyce 9131 Ch05 2

September 29, 2016

17:30

233

5.7 Bessel’s Equation

Substituting for a2m ( 0) from equation (6), and letting Hm = 1 +

1 1 1 + + ··· + , 2 3 m

(9)

we obtain, finally, a2m ( 0) = −Hm

( −1) m a0 , 22m ( m!) 2

m = 1, 2, 3, . . . .

The second solution of the Bessel equation of order zero is found by setting a0 = 1 and ( 0) = b2m ( 0) in equation (23) of Section 5.6. We obtain substituting for y1 ( x) and a2m y2 ( x) = J0 ( x) ln x +

∞ ( −1) m+1 Hm m=1

22m ( m!) 2

x 2m ,

x > 0.

(10)

Instead of y2 , the second solution is usually taken to be a certain linear combination of J0 and y2 . It is known as the Bessel function of the second kind of order zero and is denoted by Y0 . Following Copson (Chapter 12), we define16 Y0 ( x) =

2 [y2 ( x) + ( γ − ln 2) J0 ( x) ]. π

(11)

Here γ is a constant known as the Euler--Máscheroni17 constant; it is defined by the equation γ = lim ( Hn − ln n) ∼ = 0.5772.

(12)

n→∞

Substituting for y2 ( x) in equation (11), we obtain ⎤ ⎡ ∞ ( −1) m+1 Hm 2m ⎦ 2⎣ x Y0 ( x) = , x γ + ln J0 ( x) + π 2 22m ( m!) 2

x > 0.

(13)

m=1

The general solution of the Bessel equation of order zero for x > 0 is y = c1 J0 ( x) + c2 Y0 ( x) . Note that J0 ( x) → 1 as x → 0 and that Y0 ( x) has a logarithmic singularity at x = 0; that is, Y0 ( x) behaves as ( 2/π ) ln x when x → 0 through positive values. Thus, if we are interested in solutions of Bessel’s equation of order zero that are finite at the origin, which is often the case, we must discard Y0 . The graphs of the functions J0 and Y0 are shown in Figure 5.7.2. y 1

y = J0(x)

0.5

y = Y0(x)

2

4

6

8

10

12

14

x

–0.5

FIGURE 5.7.2 The Bessel functions of order zero: y = J0 ( x) (blue) and y = Y0 ( x) (red).

......................................................................................................................................................................... 16 Other

authors use other definitions for Y0 . The present choice for Y0 is also known as the Weber function, after Heinrich Weber (1842--1913), who taught at several German universities.

17 The

Euler--Máscheroni constant first appeared in 1734 in a paper by Euler. Lorenzo Máscheroni (1750--1800) was an Italian priest and professor at the University of Pavia. He correctly calculated the first 19 decimal places of γ in 1790.

233

Boyce 9131 Ch05 2

234

September 29, 2016

17:30

234

CHAPTER 5 Series Solutions of Second-Order Linear Equations

It is interesting to note from Figure 5.7.2 that for x large, both J0 ( x) and Y0 ( x) are oscillatory. Such a behavior might be anticipated from the original equation; indeed it is true for the solutions of the Bessel equation of order ν . If we divide equation (1) by x 2 , we obtain ! 1 ν2 y + y + 1 − 2 y = 0. x x For x very large, it is reasonable to conjecture that the terms ( 1/ x) y and ( ν 2 / x 2 ) y are small and hence can be neglected. If this is true, then the Bessel equation of order ν can be approximated by y + y = 0. The solutions of this equation are sin x and cos x; thus we might anticipate that the solutions of Bessel’s equation for large x are similar to linear combinations of sin x and cos x. This is correct insofar as the Bessel functions are oscillatory; however, it is only partly correct. For x large the functions J0 and Y0 also decay as x increases; thus the equation y + y = 0 does not provide an adequate approximation to the Bessel equation for large x, and a more delicate analysis is required. In fact, it is possible to show that J0 ( x) ∼ =

2 πx

1/2

π cos x − 4

as x → ∞

(14)

π sin x − 4

as x → ∞.

(15)

and that Y0 ( x) ∼ =

2 πx

1/2

These asymptotic approximations, as x → ∞, are actually very good. For example, Figure 5.7.3 shows that the asymptotic approximation (14) to J0 ( x) is reasonably accurate for all x ≥ 1. Thus to approximate J0 ( x) over the entire range from zero to infinity, you can use two or three terms of the series (7) for x ≤ 1 and the asymptotic approximation (14) for x ≥ 1. y 2 1/2

Asymptotic approximation: y = (2/π x)

cos(x – π /4)

1 y = J0(x)

x

–1 FIGURE 5.7.3 Asymptotic approximation to J0 ( x) .

Bessel Equation of Order One-Half. This case illustrates the situation in which the roots of the indicial equation differ by a positive integer but there is no logarithmic term in the second solution. Setting ν = 12 in equation (1) gives 1 L[y] = x 2 y + x y + x 2 − (16) y = 0. 4

Boyce 9131 Ch05 2

September 29, 2016

17:30

235

5.7 Bessel’s Equation

When we substitute the series (3) for y = φ (r, x) , we obtain ∞ ∞ 1 L[φ ](r, x) = (r + n) (r + n − 1) + (r + n) − an x r +n+2 an x r +n + 4 n=0 n=0 1 1 = r2 − a0 x r + (r + 1) 2 − a1 x r +1 4 4 ∞ 1 (r + n) 2 − an + an−2 x r +n = 0. + (17) 4 n=2

1 1 1 = 0 are r1 = and r2 = − ; hence the roots differ 4 2 2 by an integer. The recurrence relation is 1 2 (r + n) − (18) an = −an−2 , n ≥ 2. 4

The roots of the indicial equation r 2 −

1 , we find, from the coefficient of x r +1 in equation (17), 2 1 that a1 = 0. Hence, from equation (18), a3 = a5 = · · · = a2n+1 = · · · = 0. Further, for r = , 2 an−2 an = − , n = 2, 4, 6 . . . , n( n + 1)

Corresponding to the larger root r1 =

or, letting n = 2m, we obtain a2m = −

a2m−2 , 2m( 2m + 1)

m = 1, 2, 3, . . . .

By solving this recurrence relation, we find that a0 a0 a2 = − , a4 = , . . . 3! 5! and, in general, a2m =

( −1) m a0 , ( 2m + 1) !

m = 1, 2, 3, . . . .

Hence, taking a0 = 1, we obtain ⎛ ⎞ ∞ ∞ m 2m ( −1) x ( −1) m x 2m+1 ⎠ = x −1/2 y1 ( x) = x 1/2 ⎝1 + , ( 2m + 1) ! ( 2m + 1) ! m=1

x > 0.

(19)

m=0

The second power series in equation (19) is precisely the Taylor series for sin x; hence one solution of the Bessel equation of order one-half is x −1/2 sin x. The Bessel function of the first kind of order one-half, J1/2 , is defined as ( 2/π ) 1/2 y1 . Thus 2 1/2 J1/2 ( x) = sin x, x > 0. (20) πx 1 Corresponding to the root r2 = − , it is possible that we may have difficulty in computing 2 1 a1 since N = r1 − r2 = 1. However, from equation (17) for r = − , the coefficients of x r 2 and x r +1 are both zero regardless of the choice of a0 and a1 . Hence a0 and a1 can be chosen arbitrarily. From the recurrence relation (18), we obtain a set of even-numbered coefficients corresponding to a0 and a set of odd-numbered coefficients corresponding to a1 . Thus no logarithmic term is needed to obtain a second solution in this case. It is left as an exercise to 1 show that, for r = − , 2 a2n =

( −1) n a0 , ( 2n) !

a2n+1 =

( −1) n a1 , ( 2n + 1) !

n = 1, 2, . . . .

235

Boyce 9131 Ch05 2

236

September 29, 2016

17:30

236

CHAPTER 5 Series Solutions of Second-Order Linear Equations

Hence

⎛ y2 ( x) = x −1/2 ⎝a0

∞ ( −1) n x 2n n=0

= a0

( 2n) !

cos x sin x + a1 1/2 , x 1/2 x

+ a1

∞ ( −1) n x 2n+1 n=0

( 2n + 1) !

⎞ ⎠

x > 0.

(21)

The constant a1 simply introduces a multiple of y1 ( x) . The second solution of the Bessel equation of order one-half is usually taken to be the solution for which a0 = ( 2/π ) 1/2 and a1 = 0. It is denoted by J−1/2 . Then 2 1/2 J−1/2 ( x) = cos x, x > 0. (22) πx The general solution of equation (16) is y = c1 J1/2 ( x) + c2 J−1/2 ( x) . By comparing equations (20) and (22) with equations (14) and (15), we see that, except for a phase shift of π/4, the functions J−1/2 and J1/2 resemble J0 and Y0 , respectively, for large x. The graphs of J1/2 and J−1/2 are shown in Figure 5.7.4. y 1 y = J–1/2(x) 0.5

y = J1/2(x)

2

4

6

8

10

12

14

x

–0.5

FIGURE 5.7.4 The Bessel functions of order one-half: y = J1/2 ( x) ( blue) and y = J−1/2 ( x) ( red) .

Bessel Equation of Order One. This case illustrates the situation in which the roots of the indicial equation differ by a positive integer and the second solution involves a logarithmic term. Setting ν = 1 in equation (1) gives L[y] = x 2 y + x y + ( x 2 − 1) y = 0.

(23)

If we substitute the series (3) for y = φ (r, x) and collect terms as in the preceding cases, we obtain L[φ ](r, x) = a0 (r 2 − 1) x r + a1 (r + 1) 2 − 1 x r +1 +

∞

(r + n) 2 − 1 an + an−2 x r +n = 0.

(24)

n=2

The roots of the indicial equation r 2 − 1 = 0 are r1 = 1 and r2 = −1. The recurrence relation is (r + n) 2 − 1 an (r ) = −an−2 (r ) , n ≥ 2. (25) Corresponding to the larger root r = 1, the recurrence relation becomes an−2 an = − , n = 2, 3, 4, . . . . ( n + 2) n We also find, from the coefficient of x r +1 in equation (24), that a1 = 0; hence, from the recurrence relation, a3 = a5 = · · · = 0. For even values of n, we can write n = 2m, where m

Boyce 9131 Ch05 2

September 29, 2016

17:30

237

5.7 Bessel’s Equation

is a positive integer; then a2m = −

a2m−2 a2m−2 , =− 2 ( 2m + 2) ( 2m) 2 ( m + 1) m

m = 1, 2, 3, . . . .

By solving this recurrence relation, we obtain a2m =

( −1) m a0 , 22m ( m + 1) !m!

m = 1, 2, 3, . . . .

(26)

The Bessel function of the first kind of order one, denoted by J1 , is obtained by choosing a0 = 1/2. Hence J1 ( x) =

∞ x ( −1) m x 2m . 2 22m ( m + 1) !m!

(27)

m=0

The series converges absolutely for all x, so the function J1 is analytic everywhere. In determining a second solution of Bessel’s equation of order one, we illustrate the method of direct substitution. The calculation of the general term in equation (28) below is rather complicated, but the first few coefficients can be found fairly easily. According to Theorem 5.6.1, we assume that ⎛ ⎞ ∞ y2 ( x) = a J1 ( x) ln x + x −1 ⎝1 + cn x n ⎠, x > 0. (28) n=1

Computing y2 ( x) and y2 ( x) , substituting in equation (23), and making use of the fact that J1 is a solution of equation (23), we obtain 2ax J1 ( x) +

∞

( ( n − 1) ( n − 2) cn + ( n − 1) cn − cn ) x n−1 +

n=0

∞

cn x n+1 = 0,

(29)

n=0

where c0 = 1. Substituting for J1 ( x) from equation (27), shifting the indices of summation in the two series, and carrying out several steps of algebra, we arrive at ∞

−c1 + 0 · c2 + c0 x + ( n 2 − 1) cn+1 + cn−1 x n n=2

⎛

= −a ⎝x +

∞ ( −1) m ( 2m + 1) x 2m+1 m=1

22m ( m + 1) ! m!

⎞ ⎠.

(30)

From equation (30) we observe first that c1 = 0, and a = −c0 = −1. Further, since there are only odd powers of x on the right, the coefficient of each even power of x on the left must be zero. Thus, since c1 = 0, we have c3 = c5 = · · · = 0. Corresponding to the odd powers of x, writing n = 2m + 1 on the left-hand side of equation (30), we obtain the following recurrence relation: ( −1) m ( 2m + 1) ( 2m + 1) 2 − 1 c2m+2 + c2m = 2m (31) , m = 1, 2, 3, . . . . 2 ( m + 1) ! m! When we set m = 1 in equation (31), we obtain ( 32 − 1) c4 + c2 =

( −1) 3 . 22 · 2!

Notice that c2 can be selected arbitrarily, and then this equation determines c4 . Also notice that in the equation for the coefficient of x, c2 appeared multiplied by 0, and that equation was used to determine a. That c2 is!arbitrary is not surprising, since c2 is the coefficient of x in the ∞ expression x −1 1 + cn x n . Consequently, c2 simply generates a multiple of J1 , and y2 n=1

is determined only up to an additive multiple of J1 . In accordance with the usual practice, we

237

Boyce 9131 Ch05 2

238

September 29, 2016

17:30

238

CHAPTER 5 Series Solutions of Second-Order Linear Equations

choose c2 = 1/22 . Then we obtain 1 −1 3 −1 c4 = 4 1+ +1 = 4 +1 2 2 ·2 2 2 2! =

( −1) ( H2 + H1 ) . 24 · 2!

It is possible to show that the solution of the recurrence relation (31) is c2m =

( −1) m+1 ( Hm + Hm−1 ) , 22m m!( m − 1) !

m = 1, 2, . . .

with the understanding that H0 = 0. Thus ⎞ ⎛ ∞ ( −1) m ( Hm + Hm−1 ) 2m 1⎝ y2 ( x) = −J1 ( x) ln x + 1− x ⎠, x 22m m!( m − 1) !

x > 0.

(32)

m=1

The calculation of y2 ( x) using the alternative procedure (see equations (19) and (20) of Section 5.6) in which we determine the cn (r2 ) is slightly easier. In particular, the latter procedure yields the general formula for c2m without the necessity of solving a recurrence relation of the form (31) (see Problem 10). In this regard, you may also wish to compare the calculations of the second solution of Bessel’s equation of order zero in the text and in Problem 9. The second solution of equation (23), the Bessel function of the second kind of order one, Y1 , is usually taken to be a certain linear combination of J1 and y2 . Following Copson (Chapter 12), Y1 is defined as Y1 ( x) =

2 −y2 ( x) + ( γ − ln 2) J1 ( x) , π

(33)

where γ is defined in equation (12). The general solution of equation (23) for x > 0 is y = c1 J1 ( x) + c2 Y1 ( x) . Notice that although J1 is analytic at x = 0, the second solution Y1 becomes unbounded in the same manner as 1/ x as x → 0. The graphs of J1 and Y1 are shown in Figure 5.7.5. y 1 y = J1(x) 0.5

y = Y1(x)

2

4

6

8

10

12

–0.5

FIGURE 5.7.5 The Bessel functions of order one: y = J1 ( x) (blue) and y = Y1 ( x) (red).

14

x

Boyce 9131 Ch05 2

September 29, 2016

17:30

239

239

5.7 Bessel’s Equation

Problems In each of Problems 1 through 3, show that the given differential equation has a regular singular point at x = 0, and determine two solutions for x > 0.

1. 2. 3. 4.

x 2 y + 2x y + x y = 0

9. In this section we showed that one solution of Bessel’s equation of order zero L[y] = x 2 y + x y + x 2 y = 0 is J0 , where J0 ( x) is given by equation (7) with a0 = 1. According to Theorem 5.6.1, a second solution has the form ( x > 0)

x 2 y + 3x y + ( 1 + x) y = 0 x 2 y + x y + 2x y = 0

Find two solutions (not multiples of each other) of the Bessel 3 equation of order 2

2

x y + xy +

9 x − 4

y = 0,

2

y2 ( x) = J0 ( x) ln x +

1 4

x2 −

a. Show that L[y2 ]( x) =

x > 0.

y = 0,

∞

n( n − 1) bn x n +

n=2 ∞

+

x>0

b1 x + 22 b2 x 2 +

6. Show directly that the series for J0 ( x) , equation (7), converges absolutely for all x. 7. Show directly that the series for J1 ( x) , equation (27), converges

absolutely for all x and that J0 ( x) = −J1 ( x) .

= −2

x 2 y + x y + ( x 2 − ν 2 ) y = 0,

+

m=3

2 +

1 2!( 1 + ν ) ( 2 + ν )

( −1) m!( 1 + ν ) · · · ( m + ν )

y2 ( x) = x

1 1− 1!( 1 − ν )

+

∞ m=3

2 x 2

( n 2 bn + bn−2 ) x n

2m x 2

4 x 2

( −1) m m!( 1 − ν ) · · · ( m − ν )

⎠.

2m x 2

(35)

b4 = −

1 22 42

1+

1 2

( −1) n ( 2n) , n = 2, 3, 4, . . . . 22n ( n!) 2

and b6 =

1 22 42 62

1+

1 1 + . 2 3

The

general solution of the recurrence relation is ( −1) n+1 Hn b2n = . Substituting for bn in the expression for 22n ( n!) 2 y2 ( x) , we obtain the solution given in equation (10).

10. Find a second solution of Bessel’s equation of order one by

⎞

1 + 2!( 1 − ν ) ( 2 − ν )

.

side of equation (35). Show that b1 = b3 = b5 = · · · = 0, 1 , and that b2 = 2 2 ( 1!) 2

x > 0,

c. If 2ν is not an integer, show that a second solution is −ν

(34)

c. Note that only even powers of x appear on the right-hand

Deduce that

m

∞

22n ( n!) 2

( 2n) 2 b2n + b2n−2 = −2

where ν is real and positive. a. Show that x = 0 is a regular singular point and that the roots of the indicial equation are ν and −ν . b. Corresponding to the larger root ν , show that one solution is

∞

bn x n+2 + 2x J0 ( x) .

∞ ( −1) n 2nx 2n n=1

8. Consider the Bessel equation of order ν

x 2

n=1

n=3

by the change of dependent variable y = x −1/2 v( x) . From this, conclude that y1 ( x) = x −1/2 cos x and y2 ( x) = x −1/2 sin x are solutions of the Bessel equation of order one-half.

1 1!( 1 + ν )

nbn x n

b. Substituting the series representation for J0 ( x) in equation (34), show that

v + v = 0

1−

∞

n=1

can be reduced to the equation

y1 ( x) = x ν

bn x n .

n=1

5. Show that the Bessel equation of order one-half x 2 y + x y +

∞

4 x 2

⎞ ⎠.

Note that y1 ( x) → 0 as x → 0, and that y2 ( x) is unbounded as x → 0. d. Verify by direct methods that the power series in the expressions for y1 ( x) and y2 ( x) converge absolutely for all x. Also verify that y2 is a solution, provided only that ν is not an integer.

computing the cn (r2 ) and a of equation (24) of Section 5.6 according to the formulas (19) and (20) of that section. Some guidelines along the way of this calculation are the following. First, use equation (24) of this section to show that a1 ( −1) and a1 ( −1) are 0. Then show that c1 ( −1) = 0 and, from the recurrence relation, that cn ( −1) = 0 for n = 3, 5, . . . . Finally, use equation (25) to show that a0 a2 (r ) = − , (r + 1) (r + 3) a0 , a4 (r ) = (r + 1) (r + 3) (r + 3) (r + 5) and that a2m (r ) =

( −1) m a0 , m ≥ 3. (r + 1) · · · (r + 2m − 1) (r + 3) · · · (r + 2m + 1)

Then show that c2m ( −1) =

( −1) m+1 Hm + Hm−1 22m m!( m − 1) !

,

m ≥ 1.

Boyce 9131 Ch05 2

240

September 29, 2016

17:30

240

CHAPTER 5 Series Solutions of Second-Order Linear Equations

11. By a suitable change of variables it is sometimes possible to transform another differential equation into a Bessel equation. For example, show that a solution of

x 2 y +

8.653 (see Figure 5.7.1). Let λ j , j = 1, 2, 3, . . . , denote the zeros of J0 ; it follows that

"

J0 ( λ j x) =

α 2 β 2 x 2β

1 + − ν 2 β 2 y = 0, 4

x>0

of the Airy equation

y +

1 2 y + λ j y = 0, x

1

if λ i = λ j.

0

x>0

x > 0.

Hence show that x J0 ( λ i x) J0 ( λ j x) d x = 0

y − x y = 0,

x = 0, x = 1.

Verify that y = J0 ( λ j x) satisfies the differential equation

is given by y = x 1/2 f α x β , where f ( ξ ) is a solution of the Bessel equation of order ν .

12. Using the result of Problem 11, show that the general solution

1, 0,

2 3/2 2 3/2 ix ix c1 f 1 + c2 f 2 , where f 1 ( ξ ) and is y = x 3 3 f 2 ( ξ ) are a fundamental set of solutions of the Bessel equation of order one-third. 1/2

13. It can be shown that J0 has infinitely many zeros for x > 0. In particular, the first three zeros are approximately 2.405, 5.520, and

This important property of J0 ( λ i x) , which is known as the orthogonality property, is useful in solving boundary value problems. Hint: Write the differential equation for J0 ( λ i x) . Multiply it by x J0 ( λ j x) and subtract that result from x J0 ( λ i x) times the differential equation for J0 ( λ j x) . Then integrate from 0 to 1.

References Coddington, E. A., An Introduction to Ordinary Differential Equations (Englewood Cliffs, NJ: Prentice-Hall, 1961; New York: Dover, 1989). Coddington, E. A., and Carlson, R., Linear Ordinary Differential Equations (Philadelphia, PA: Society for Industrial and Applied Mathematics, 1997). Copson, E. T., An Introduction to the Theory of Functions of a Complex Variable (Oxford: Oxford University Press, 1935). K. Knopp, Theory and Applications of Infinite Series (NewYork: Hafner, 1951). Proofs of Theorems 5.3.1 and 5.6.1 can be found in intermediate or advanced books; for example, see Chapters 3 and 4 of Coddington, Chapters 5 and 6 of Coddington and Carlson, or Chapters 3 and 4 of Rainville, E. D., Intermediate Differential Equations (2nd ed.) (New York: Macmillan, 1964). Also see these texts for a discussion of the point at infinity, which was mentioned in Problem 32 of Section 5.4. The behavior of solutions near an irregular singular point is an even more advanced topic; a brief discussion can be found in Chapter 5 of Coddington, E. A., and Levinson, N., Theory of Ordinary Differential Equations (New York: McGraw-Hill, 1955; Malabar, FL: Krieger, 1984).

Fuller discussions of the Bessel equation, the Legendre equation, and many of the other named equations can be found in advanced books on differential equations, methods of applied mathematics, and special functions. One text dealing with special functions such as the Legendre polynomials and the Bessel functions is Hochstadt, H., Special Functions of Mathematical Physics (New York: Holt, 1961). An excellent compilation of formulas, graphs, and tables of Bessel functions, Legendre functions, and other special functions of mathematical physics may be found in Abramowitz, M., and Stegun, I. A. (eds.), Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables (New York: Dover, 1965); originally published by the National Bureau of Standards, Washington, DC, 1964. The digital successor to Abramowitz and Stegun is Digital Library of Mathematical Functions. Released August 29, 2011. National Institute of Standards and Technology from http://dlmf.nist.gov/.

Boyce 9131 Ch06 2

September 29, 2016

17:34

241

CHAPTER 6 The Laplace Transform Many practical engineering problems involve mechanical or electrical systems acted on by discontinuous or impulsive forcing terms. For such problems the methods described in Chapter 3 are often rather awkward to use. Another method that is especially well suited to these problems, although useful much more generally, is based on the Laplace transform. In this chapter we describe how this important method works, emphasizing problems typical of those that arise in engineering applications.

6.1

Definition of the Laplace Transform

Improper Integrals. Since the Laplace transform involves an integral from zero to infinity, a knowledge of improper integrals of this type is necessary to appreciate the subsequent development of the properties of the transform. We provide a brief review of such improper integrals here. If you are already familiar with improper integrals, you may wish to skip over this review. On the other hand, if improper integrals are new to you, then you should probably consult a calculus book, where you will find many more details and examples. An improper integral over an unbounded interval is defined as a limit of integrals over finite intervals; thus ∞ A f ( t) dt = lim f ( t) dt, (1) A→∞

a

a

where A is a positive real number. If the definite integral from a to A exists for each A > a, and if the limit of these values as A → ∞ exists, then the improper integral is said to converge to that limiting value. Otherwise the integral is said to diverge, or to fail to exist. The following examples illustrate both possibilities.

EXAMPLE 1

∞

Does the improper integral 1

dt diverge or converge? t

Solution: From equation (1) we have

1

∞

dt = lim t A→∞

1

A

dt = lim ln A. t A→∞

Since lim ln A = ∞, the improper integral diverges. A→∞

241

Boyce 9131 Ch06 2

242

September 29, 2016

17:34

242

CHAPTER 6 The Laplace Transform

EXAMPLE 2

∞

ect dt. For what values of c does this improper integral converge?

Evaluate the improper integral 0

Solution: Suppose c is a real nonzero constant. Then

∞

ect dt = lim

A→∞

0

A

ect A→∞ c 0

A

ect dt = lim = lim A→∞

0

1 cA ( e − 1) . c

It follows that the improper integral converges to the value −1/c if c < 0 and diverges if c > 0. If c = 0, the integrand ect is the constant function with value 1. In this case

A

1dt = lim ( A − 0) = ∞,

lim A→∞

A→∞

0

so the integral again diverges.

EXAMPLE 3

∞

Find all real numbers p for which the improper integral

t − p dt converges. For what values of p

1

does it diverge? Solution: Suppose that p is a real constant and p = 1; the case p = 1 was considered in Example 1. Then

∞

t 1

−p

A

dt = lim A→∞

t − p dt = lim A→∞

1

1 ( A1− p − 1) . 1− p

As A → ∞, A

1− p

→ 0 if p > 1, but A

1− p

→ ∞ if p < 1. Hence

∞

t − p dt converges to the

1

value 1/( p − 1) for p > 1 but (incorporating the result of Example 1) diverges for p ≤ 1. These results are analogous to those for the infinite series

∞

n− p .

n=1

∞ Before discussing the possible existence of a f ( t) dt, it is helpful to define certain terms. A function f is said to be piecewise continuous on an interval α ≤ t ≤ β if the interval1 can be partitioned by a finite number of points α = t0 < t1 < · · · < tn = β so that 1. f is continuous on each open subinterval ti−1 < t < ti . 2. f approaches a finite limit as the endpoints of each subinterval are approached from within the subinterval. In other words, f is piecewise continuous on α ≤ t ≤ β if it is continuous there except for a finite number of jump discontinuities. If f is piecewise continuous on α ≤ t ≤ β for every β > α , then f is said to be piecewise continuous on t ≥ α . An example of a piecewise continuous function is shown in Figure 6.1.1. The integral of a piecewise continuous function on a finite interval is just the sum of the integrals on the subintervals created by the partition points. For instance, for the function f ( t) ......................................................................................................................................................................... 1 It

is not essential that the interval be closed; the same definition applies if the interval is open at one or both ends.

Boyce 9131 Ch06 2

September 29, 2016

17:34

243

6.1 Definition of the Laplace Transform

y

α

t1

β

t2

t

FIGURE 6.1.1 A piecewise continuous function y = f ( t) .

shown in Figure 6.1.1, we have β f ( t) dt = α

t1

f ( t) dt +

α

t2

β

f ( t) dt +

t1

f ( t) dt.

(2)

t2

For the function shown in Figure 6.1.1, we have assigned values to the function at the endpoints α and β and at the partition points t1 and t2 . However, as far as the integrals in equation (2) are concerned, it does not matter whether f ( t) is defined at these points, or what values may be assigned to f ( t) at them. The values of the integrals in equation (2) remain the same regardless. A Thus, if f is piecewise continuous on the interval a ≤ t ≤ A, then a f ( t) dt exists. A Hence, if f is piecewise continuous for t ≥ a, then a f ( t) dt exists for each A > a. However, piecewise continuity is not enough to ensure convergence of the improper integral ∞ f ( t) dt, as the preceding examples show. a If f cannot be integrated easily in terms of elementary functions, the definition of ∞ convergence of a f ( t) dt may be difficult to apply. Frequently, the most convenient way to test the convergence or divergence of an improper integral is by the following comparison theorem, which is analogous to a similar theorem for infinite series.

Theorem 6.1.1 If f is piecewise continuous for t ≥ a, if | f ( t) | ≤ g( t) when t ≥ M for some positive constant M, ∞ ∞ and if M g( t) dt converges, then a f ( t) dt also converges. On the other hand, if f ( t) ≥ g( t) ≥ 0 for t ≥ M, and if also diverges.

∞ M

g( t) dt diverges, then

∞ a

f ( t) dt

The proof of these results from calculus will not be given here. ∞They are made plausible, ∞ however, by comparing the areas represented by M g( t) dt and M | f ( t) |dt. The functions most useful for comparison purposes are ect and t − p , which we considered in Examples 1, 2, and 3. The Laplace Transform. Among the tools that are very useful for solving linear differential equations are integral transforms. An integral transform is a relation of the form β F( s) = K ( s, t) f ( t) dt, (3) α

where K ( s, t) is a given function, called the kernel of the transformation, and the limits of integration α and β are also given. It is possible that α = −∞, or β = ∞, or both. The relation (3) transforms the function f into another function F, which is called the transform of f . There are several integral transforms that are useful in applied mathematics, but in this chapter we consider only the Laplace2 transform. This transform is defined in the following ......................................................................................................................................................................... 2 The Laplace transform is named for the eminent French mathematician P. S. Laplace, who studied the relation (3) in

1782. However, the techniques described in this chapter were not developed until a century or more later. We owe them mainly to Oliver Heaviside (1850--1925), an innovative self-taught English electrical engineer, who made significant contributions to the development and application of electromagnetic theory. He was also one of the developers of vector calculus.

243

Boyce 9131 Ch06 2

244

September 29, 2016

17:34

244

CHAPTER 6 The Laplace Transform

way. Let f ( t) be given for t ≥ 0, and suppose that f satisfies certain conditions to be stated a little later. Then the Laplace transform of f , which we will denote by L{ f ( t) } or by F( s) , is defined by the equation ∞ L{ f ( t) } = F( s) = e−st f ( t) dt, (4) 0

whenever this improper integral converges. The Laplace transform makes use of the kernel K ( s, t) = e−st . Since the solutions of linear differential equations with constant coefficients are based on the exponential function, the Laplace transform is particularly useful for such equations. The general idea in using the Laplace transform to solve a differential equation is as follows: 1. Use the relation (4) to transform an initial value problem for an unknown function f in the t-domain into a simpler problem (indeed, an algebraic problem) for F in the s-domain. 2. Solve this algebraic problem to find F. 3. Recover the desired function f from its transform F. This last step is known as “inverting the transform.” In general, the parameter s may be complex, and the full power of the Laplace transform becomes available only when we regard F( s) as a function of a complex variable. However, for the problems discussed here, it is sufficient to consider only real values of s. The Laplace transform F of a function f exists if f satisfies certain conditions, such as those stated in the following theorem.

Theorem 6.1.2 Suppose that

(i) f is piecewise continuous on the interval 0 ≤ t ≤ A for any positive A and (ii) there exist real constants K , a, and M, with K and M positive, such that f ( t) ≤ K eat when t ≥ M. Then the Laplace transform L{ f ( t) } = F( s) , defined by equation (4), exists for s > a.

To establish this theorem, we must show that the integral in equation (4) converges for s > a. Splitting the improper integral into two parts, we have ∞ M ∞ e−st f ( t) dt = e−st f ( t) dt + e−st f ( t) dt. (5) 0

0

M

The first integral on the right-hand side of equation (5) exists by hypothesis (i) of the theorem; hence the existence of F( s) depends on the convergence of the second integral. By hypothesis (ii) we have, for t ≥ M, |e−st f ( t) | ≤ K e−st eat = K e( a−s) t , ∞ and thus, by Theorem 6.1.1, F( s) exists provided that M e( a−s) t dt converges. Referring to Example 1 with c replaced by a − s, we see that this latter integral converges when a − s < 0, which establishes Theorem 6.1.2. In this chapter (except in Section 6.5), we deal almost exclusively with functions that satisfy the conditions of Theorem 6.1.2. Such functions are described as piecewise continuous and of exponential order as t → ∞. Note that there are functions that are not of exponential 2 order as t → ∞. One such function is f ( t) = et . As t → ∞, this function increases faster than K eat regardless of how large the constants K and a may be. The Laplace transforms of some important elementary functions are given in the following examples.

Boyce 9131 Ch06 2

September 29, 2016

17:34

245

6.1 Definition of the Laplace Transform

EXAMPLE 4 Find L{1}. Solution: Let f ( t) = 1, t ≥ 0. Then, as in Example 2,

∞

L{1} =

e

−st

0

A

e−st 1 dt = − lim = , s A→∞ s 0

s > 0.

EXAMPLE 5

Find L eat . Solution: Let f ( t) = eat , t ≥ 0. Then, again referring to Example 2,

∞

L{e } = at

e

−st at

e dt =

0

=

∞

e−( s−a) t dt

0

1 , s−a

s > a.

EXAMPLE 6 Find the Laplace transform of the function graphed in Figure 6.1.2. y 1 k t

1

FIGURE 6.1.2 Graph of the piecewise-defined function in Example 6.

Solution: Let f ( t) =

⎧ ⎨1, 0 ≤ t < 1, k, 0,

⎩

t = 1, t > 1,

where k is a constant. In engineering contexts f ( t) often represents a unit pulse, perhaps of force or voltage. Note that f is a piecewise continuous function. Then

L{ f ( t) } =

∞

e 0

−st

f ( t) dt =

1

1

e 0

−st

e−st 1 − e−s dt = − = , s 0 s

s > 0.

Observe that L{ f ( t) } does not depend on k, the function value at the point of discontinuity. Even if f ( t) is not defined at this point, the Laplace transform of f remains the same. Thus there are many functions, differing only in their value at a single point, that have the same Laplace transform.

245

Boyce 9131 Ch06 2

246

September 29, 2016

17:34

246

CHAPTER 6 The Laplace Transform

EXAMPLE 7 Find L{sin( at) }. For what values of s is this transform defined? Solution: Let f ( t) = sin( at) , t ≥ 0. Then

∞

L{sin( at) } = F( s) =

e−st sin( at) dt,

s > 0.

0

Since

A

F( s) = lim A→∞

e−st sin( at) dt,

0

upon integrating by parts, we obtain

F( s) = lim A→∞

=

1 s − a a

A

e−st cos( at) s − −a a 0

∞

A

e

−st

cos( at) dt

0

e−st cos( at) dt.

0

A second integration by parts then yields s2 1 − 2 a a

F( s) =

∞

e−st sin( at) dt

0

s2 1 − 2 F( s) . a a

= Now, solving for F( s) , we have

F( s) =

a , s2 + a2

s > 0.

s for s > 0. Now + a2 let us suppose that f 1 and f 2 are two functions whose Laplace transforms exist for s > a1 and s > a2 , respectively. Then, for s greater than the maximum of a1 and a2 , ∞

L{c1 f 1 ( t) + c2 f 2 ( t) } = e−st c1 f 1 ( t) + c2 f 2 ( t) dt In Problem 5 you will use a similar process to find L{cos( at) } =

s2

0

= c1

∞

e

−st

f 1 ( t) dt + c2

0

∞

e−st f 2 ( t) dt;

0

hence L{c1 f 1 ( t) + c2 f 2 ( t) } = c1 L{ f 1 ( t) } + c2 L{ f 2 ( t) }.

(6)

Equation (6) states that the Laplace transform is a linear operator, and we make frequent use of this property later. The sum in equation (6) can be readily extended to an arbitrary number of terms.

EXAMPLE 8 Find the Laplace transform of f ( t) = 5e−2t − 3 sin( 4t) , t ≥ 0.

▼

Boyce 9131 Ch06 2

September 29, 2016

17:34

247

6.1 Definition of the Laplace Transform

247

▼ Solution: Using equation (6), we write

L f ( t)

= 5L e−2t − 3L sin( 4t) .

Then, from Examples 5 and 7, we obtain L{ f ( t) } =

5 12 , − 2 s + 2 s + 16

s > 0.

Problems In each of Problems 1 through 3, sketch the graph of the given function. In each case determine whether f is continuous, piecewise continuous, or neither on the interval 0 ≤ t ≤ 3.

1.

2.

3.

f ( t) =

f ( t) =

f ( t) =

⎧ ⎨t 2 ,

2 + t, ⎩6 − t,

⎧ ⎨t 2 ,

0≤t ≤1 1< t ≤2 2< t ≤3

( t − 1) −1 , ⎩1,

⎧ ⎨t 2 ,

1, ⎩3 − t,

0≤t ≤1 1< t ≤2 2< t ≤3

0≤t ≤1 1< t ≤2 2< t ≤3

4. Find the Laplace transform of each of the following functions: a. f ( t) = t b. f ( t) = t 2 c. f ( t) = t n , where n is a positive integer 5. Find the Laplace transform of f ( t) = cos( at) , where a is a real constant. 1 1 bt ( e + e−bt ) and sinh( bt) = ( ebt − e−bt ) . 2 2 In each of Problems 6 through 7, use the linearity of the Laplace transform to find the Laplace transform of the given function; a and b are real constants. cosh( bt) =

f ( t) = cosh( bt) f ( t) = sinh( bt)

Recall that 1 1 ibt ( e + e−ibt ) and sin( bt) = ( eibt − e−ibt ) . 2 2i In each of Problems 8 through 11, use the linearity of the Laplace transform to find the Laplace transform of the given function; a and b are real constants. Assume that the necessary elementary integration formulas extend to this case. cos( bt) =

8. 9. 10. 11.

f ( t) = sin( bt) f ( t) = cos( bt) f ( t) = eat sin( bt) f ( t) = e cos( bt) at

In each of Problems 12 through 15, use integration by parts to find the Laplace transform of the given function; n is a positive integer and a is a real constant.

12. 13. 14. 15.

17.

f ( t) =

t, 1,

⎧ ⎨t,

0≤t < 1 1≤t < ∞

0≤t < 1 2 − t, 1 ≤ t < 2 ⎩ 0, 2≤t < ∞ In each of Problems 19 through 21, determine whether the given integral converges or diverges.

18.

f ( t) =

∞

19. 0 ∞ 20. 0 21.

∞

( t 2 + 1) −1 dt te−t dt t −2 et dt

1

22. Suppose that f and f are continuous for t ≥ 0 and of

Recall that

6. 7.

In each of Problems 16 through 18, find the Laplace transform of the given function. 1, 0 ≤ t < π 16. f ( t) = 0, π ≤ t < ∞

f ( t) = teat f ( t) = t sin( at) f ( t) = t n eat f ( t) = t 2 sin( at)

exponential order as t → ∞. Use integration by parts to show that if F( s) = L{ f ( t) }, then lim F( s) = 0. The result is actually true s→∞

under less restrictive conditions, such as those of Theorem 6.1.2.

23. The Gamma Function. The gamma function is denoted by Γ( p) and is defined by the integral

Γ( p + 1) =

∞

e−x x p d x.

(7)

0

The integral converges as x → ∞ for all p. For p < 0 it is also improper at x = 0, because the integrand becomes unbounded as x → 0. However, the integral can be shown to converge at x = 0 for p > −1. a. Show that, for p > 0, Γ( p + 1) = pΓ( p) .

b. Show that Γ( 1) = 1. c. If p is a positive integer n, show that Γ( n + 1) = n!. Since Γ( p) is also defined when p is not an integer, this function provides an extension of the factorial function to nonintegral values of the independent variable. Note that it is also consistent to define 0! = 1. d. Show that, for p > 0, Γ( p + n) p( p + 1) ( p + 2) · · · ( p + n − 1) = . Γ( p) Thus Γ( p) can be determined for all positive values of p if Γ( p) is known in a single interval 0 < p ≤1. Itis of unit length---say, √ 1 3 11 = π . Find Γ and Γ . possible to show that Γ 2 2 2

Boyce 9131 Ch06 2

248

September 29, 2016

17:34

248

CHAPTER 6 The Laplace Transform

24. Consider the Laplace transform of t p , where p > −1. a. Referring to Problem 23, show that ∞ ∞ e−st t p dt =

L{t p } = 0

Γ( p) = p+1 , s

1

It is possible to show that

e

e−x x p d x

s p+1

hence

0

c. Show that L{t

−1/2

2 }= √ s

s

, n+1

∞

2

−1/2

}=

d. Show that

√ L{t 1/2 } =

s > 0.

e−x d x,

√ π dx = ; 2

L{t

b. Let p be a positive integer n in part a; show that n!

−x 2

0

s > 0.

L{t n } =

∞

π , s

π

2s 3/2

,

s > 0.

s > 0.

s > 0.

0

Solution of Initial Value Problems

6.2

In this section we show how the Laplace transform can be used to solve initial value problems for linear differential equations with constant coefficients. The usefulness of the Laplace transform for this purpose rests primarily on the fact that the transform of f is related in a simple way to the transform of f . The relationship is expressed in the following theorem.

Theorem 6.2.1 Suppose that f is continuous and f is piecewise continuous on any interval 0 ≤ t ≤ A. Suppose further that there exist constants K , a, and M such that | f ( t) | ≤ K eat for t ≥ M. Then L{ f ( t) } exists for s > a, and moreover, L{ f ( t) } = sL{ f ( t) } − f ( 0) .

(1)

To prove this theorem, we consider the integral A e−st f ( t) dt, 0

whose limit as A → ∞, if it exists, is the Laplace transform of f . To calculate this limit we first need to write the integral in a suitable form. If f has points of discontinuity in the interval 0 ≤ t ≤ A, let them be denoted by t1 , t2 , . . . , tk . Then we can write the integral as

A

e 0

−st

f ( t) dt =

t1

e

−st

f ( t) dt +

t2

e

−st

f ( t) dt + · · · +

e−st f ( t) dt.

tk

t1

0

A

Integrating each term on the right by parts yields 0

A

t t A e−st f ( t) dt = e−st f ( t) 01 + e−st f ( t) t2 + · · · + e−st f ( t) tk 1

+s

t1

e 0

−st

f ( t) dt +

t2

e t1

−st

f ( t) dt + · · · +

A

e tk

−st

f ( t) dt .

Boyce 9131 Ch06 2

September 29, 2016

17:34

249

6.2 Solution of Initial Value Problems

Since f is continuous, the contributions of the integrated terms at t1 , t2 , . . . , tk cancel. Further, the integrals on the right-hand side can be combined into a single integral so that we obtain A A e−st f ( t) dt = e−s A f ( A) − f ( 0) + s e−st f ( t) dt. (2) 0

0

Now we let A → ∞ in equation (2). The integral on the right-hand side of this equation approaches L{ f ( t) }. Further, for A ≥ M, we have | f ( A) | ≤ K ea A ; consequently, |e−s A f ( A) | ≤ K e−( s−a) A . Hence e−s A f ( A) → 0 as A → ∞ whenever s > a. Thus the right-hand side of equation (2) has the limit sL{ f ( t) } − f ( 0) . Consequently, the left-hand side of equation (2) also has a limit, and as noted above, this limit is L{ f ( t) }. Therefore, for s > a, we conclude that L{ f ( t) } = sL{ f ( t) } − f ( 0) , which completes the proof of Theorem 6.2.1. If f and f satisfy the same conditions that are imposed on f and f , respectively, in Theorem 6.2.1, then it follows that the Laplace transform of f also exists for s > a and is given by L{ f ( t) } = sL{ f ( t) } − f ( 0) = s( sL{ f ( t) } − f ( 0) ) − f ( 0) = s 2 L{ f ( t) } − s f ( 0) − f ( 0) .

(3)

Indeed, provided the function f and its derivatives satisfy suitable conditions, an expression for the transform of the n th derivative f ( n) can be derived by n successive applications of this theorem. The result is given in the following corollary.

Corollary 6.2.2 Suppose that the functions f, f , . . . , f ( n−1) are continuous and that f ( n) is piecewise continuous on any interval 0 ≤ t ≤ A. Suppose further that there exist constants K , a, and M such that | f ( t) | ≤ K eat , | f ( t) | ≤ K eat , . . . , | f ( n−1) ( t) | ≤ K eat for t ≥ M. Then L{ f ( n) ( t) } exists for s > a and is given by

L f ( n) ( t)

= s n L f ( t) − s n−1 f ( 0) − · · · − s f ( n−2) ( 0) − f ( n−1) ( 0) .

(4)

We now show how the Laplace transform can be used to solve initial value problems. It is most useful for problems involving nonhomogeneous differential equations, as we will demonstrate in later sections of this chapter. However, we begin by looking at some homogeneous equations, which are a bit simpler.

EXAMPLE 1 Find the solution of the differential equation y − y − 2y = 0

(5)

that satisfies the initial conditions y( 0) = 1,

y ( 0) = 0.

(6)

Solution: This initial value problem is easily solved by the methods of Section 3.1. The characteristic equation is r 2 − r − 2 = (r − 2) (r + 1) = 0,

▼

249

Boyce 9131 Ch06 2

250

September 29, 2016

17:34

250

CHAPTER 6 The Laplace Transform

▼ and consequently, the general solution of equation (5) is y = c1 e−t + c2 e2t .

(7)

To satisfy the initial conditions (6), we must have c1 + c2 = 1 and −c1 + 2c2 = 0; hence c1 = and c2 =

1 , so the solution of the initial value problem (4) and (5) is 3 2 1 y = φ ( t) = e−t + e2t . 3 3

2 3

(8)

Now let us solve the same problem using the Laplace transform. To do this, we must assume that the problem has a solution y = y( t) , which with its first two derivatives satisfies the conditions of Corollary 6.2.2. Then, taking the Laplace transform of the differential equation (5), we obtain

L y − y − 2y = L{y } − L{y } − 2L{y} = 0,

(9)

where we have used the linearity of the Laplace transform to write the transform of a sum as the sum of the separate transforms. Upon using Corollary 6.2.2 to express L{y } and L{y } in terms of L{y}, we find that equation (9) becomes s 2 L{y} − sy( 0) − y ( 0) − ( sL{y} − y( 0) ) − 2L{y} = 0, or ( s 2 − s − 2) Y ( s) + ( 1 − s) y( 0) − y ( 0) = 0,

(10)

where Y ( s) = L{y}. Substituting for y( 0) and y ( 0) in equation (10) from the initial conditions (6), and then solving for Y ( s) , we obtain Y ( s) =

s2

s−1 s−1 = . ( s − 2) ( s + 1) −s−2

(11)

We have thus obtained an expression for the Laplace transform Y ( s) of the solution y( t) of the given initial value problem. To determine the solution, y( t) , we must find the function whose Laplace transform is Y ( s) , as given in equation (11). This can be done most easily by expanding the right-hand side of equation (11) in partial fractions. Thus we write Y ( s) =

s−1 a b a( s + 1) + b( s − 2) = + = , ( s − 2) ( s + 1) s−2 s+1 ( s − 2) ( s + 1)

(12)

where the coefficients a and b are to be determined. By equating numerators of the second and fourth members of equation (12), we obtain s − 1 = a( s + 1) + b( s − 2) , an equation that must hold for all s. In particular, if we set s = 2 then it follows that a = if we set s = −1 then we find that b =

1 . Similarly, 3

2 . Substituting these values for a and b into equation (12), 3

we have Y ( s) =

2/3 1/3 + . s−2 s+1

Finally, if we use the result of Example 5 of Section 6.1, it follows that

(13) 1 2t e has the transform 3

2 2 1 ( s − 2) −1 ; similarly, e−t has the transform ( s + 1) −1 . Hence, by the linearity of the Laplace 3 3 3 transform, y( t) =

1 2t 2 −t e + e 3 3

has the transform (13) and is therefore the solution of the initial value problem (4), (5). Observe that it does satisfy the conditions of Corollary 6.2.2, as we assumed initially. Of course, this is the same solution that we obtained earlier.

The same procedure can be applied to the general second-order linear equation with constant coefficients ay + by + cy = f ( t) .

(14)

Boyce 9131 Ch06 2

September 29, 2016

17:34

251

6.2 Solution of Initial Value Problems

Assuming that the solution y( t) satisfies the conditions of Corollary 6.2.2 for n = 2, we can take the transform of equation (14) and thereby obtain a s 2 Y ( s) − sy( 0) − y ( 0) + b( sY ( s) − y( 0) ) + cY ( s) = F( s) , (15) where F( s) is the transform of f ( t) . By solving equation (15) for Y ( s) , we find that Y ( s) =

( as + b) y( 0) + ay ( 0) F( s) + 2 . 2 as + bs + c as + bs + c

(16)

The problem is then solved, provided that we can find the function y( t) whose transform is Y ( s) . Even at this early stage of our discussion we can point out four essential features of the Laplace transform method. In the first place, the transform Y ( s) of the unknown function y( t) is found by solving an algebraic equation rather than a differential equation, algebraic equation (10) rather than differential equation (5) in Example 1, or in general algebraic equation (15) rather than differential equation (14). This is the key to the usefulness of Laplace transforms for solving linear, constant coefficient, ordinary differential equations---the problem is reduced from a differential equation to an algebraic one. Next, the solution satisfying given initial conditions is automatically found, so that the task of determining appropriate values for the arbitrary constants in the general solution does not arise. Further, as indicated in equation (15), nonhomogeneous equations are handled in exactly the same way as homogeneous ones; it is not necessary to solve the corresponding homogeneous equation first. Finally, the method can be applied in the same way to higher-order equations, as long as we assume that the solution satisfies the conditions of Corollary 6.2.2 for the appropriate value of n. Observe that the polynomial as 2 + bs + c in the denominator on the right-hand side of equation (16) is precisely the characteristic polynomial associated with equation (14). Since the use of a partial fraction expansion of Y ( s) to determine y( t) requires us to factor this polynomial, the use of Laplace transforms does not avoid the necessity of finding roots of the characteristic equation. For equations of higher than second order, this may require a numerical approximation, particularly if the roots are irrational or complex. The main difficulty that occurs in solving initial value problems by Laplace transforms lies in the problem of determining the function y( t) corresponding to the transform Y ( s) . This problem is known as the inversion problem for the Laplace transform; y( t) is called the inverse Laplace transform corresponding to Y ( s) , and the process of finding y( t) from Y ( s) is known as inverting the Laplace transform. We also use the notation L−1 {Y ( s) } to denote the inverse transform of Y ( s) . There is a general formula for the inverse Laplace transform, but its use requires a familiarity with functions of a complex variable, and we do not consider it in this book. However, it is still possible to develop many important properties of the Laplace transform, and to solve many interesting problems, without the use of complex variables. In solving the initial value problem (4), (5), we did not consider the question of whether there may be functions other than the one given by equation (8) that also have the transform (13). By Theorem 3.2.1 we know that the initial value problem has no other solutions. We also know that the unique solution (8) of the initial value problem is continuous. Consistent with this fact, it can be shown that if f and g are continuous functions with the same Laplace transform, then f and g must be identical. On the other hand, if f and g are only piecewise continuous, then they may differ at one or more points of discontinuity and yet have the same Laplace transform; see Example 6 in Section 6.1. This lack of uniqueness of the inverse Laplace transform for piecewise continuous functions is of no practical significance in applications. Thus there is essentially a one-to-one correspondence between functions and their Laplace transforms. This fact suggests the compilation of a table, such as Table 6.2.1, giving the Laplace transforms of functions frequently encountered, and vice versa. The entries in the second column of Table 6.2.1 are the Laplace transforms of those in the first column. Perhaps more important, the functions in the first column are the inverse Laplace transforms of those in the second column. Thus, for example, if the transform of the solution of a differential equation is known, the solution itself can often be found merely by looking it up in the table. Some of the entries in Table 6.2.1 have been used as examples, or appear as problems in Section 6.1, while others will be developed later in the chapter. The third column of the table indicates where the derivation of the given transforms may be found. Although Table 6.2.1 is sufficient for the examples and problems in this book, much larger tables are also available (see the list

251

Boyce 9131 Ch06 2

252

September 29, 2016

17:34

252

CHAPTER 6 The Laplace Transform

Elementary Laplace Transforms

T A B L E 6.2.1

f (t) = L−1 {F(s)}

F(s) = L { f (t)}

Notes

1. 1

1 , s

2. eat

1 , s−a

s>a

Sec. 6.1; Ex. 5

3. t n , n a positive integer

n! , s n+1

s>0

Sec. 6.1; Prob. 24

p > −1

Γ( p + 1) , s p+1

4. t p ,

s>0

Sec. 6.1; Ex. 4

s>0

Sec. 6.1; Prob. 24

5. sin( at)

a , s2 + a2

s>0

Sec. 6.1; Ex. 7

6. cos( at)

s , s2 + a2

s>0

Sec. 6.1; Prob. 5

7. sinh( at)

a , s2 − a2

s > |a|

Sec. 6.1; Prob. 7

8. cosh( at)

s , s2 − a2

s > |a|

Sec. 6.1; Prob. 6

9. eat sin( bt)

b , ( s − a) 2 + b2

s>a

Sec. 6.1; Prob. 10

10. eat cos( bt)

s−a , ( s − a) 2 + b2

s>a

Sec. 6.1; Prob. 11

11. t n eat , n a positive integer

12. u c ( t) =

n! , ( s − a) n+1 e−cs , s

0 t0

Sec. 6.1; Prob. 14 Sec. 6.3

e−cs F( s)

Sec. 6.3

F( s − c)

Sec. 6.3

s 1

15. f ( ct)

s>a

c

,

c> 0

Sec. 6.3; Prob. 17

F( s) G( s)

Sec. 6.6

e−cs

Sec. 6.5

s n F( s) − s n−1 f ( 0) − · · · − f ( n−1) ( 0) Sec. 6.2; Cor. 6.2.2 F ( n) ( s)

Sec. 6.2; Prob. 21

of references at the end of the chapter). Transforms and inverse transforms can also be readily obtained electronically by using a computer algebra system. Frequently, a Laplace transform F( s) is expressible as a sum of several terms F( s) = F1 ( s) + F2 ( s) + · · · + Fn ( s) .

(17)

Boyce 9131 Ch06 2

September 29, 2016

17:34

253

6.2 Solution of Initial Value Problems

Suppose that f 1 ( t) = L−1 {F1 ( s) }, . . . , f n ( t) = L−1 {Fn ( s) }. Then the function f ( t) = f 1 ( t) + · · · + f n ( t) has the Laplace transform F( s) . By the uniqueness property stated previously, no other continuous function f has the same transform. Thus L−1 {F( s) } = L−1 {F1 ( s) } + · · · + L−1 {Fn ( s) };

(18)

that is, the inverse Laplace transform is also a linear operator. In many problems it is convenient to make use of the linearity property by decomposing a given transform into a sum of functions whose inverse transforms are already known or can be found in the table. Partial fraction expansions are particularly useful for this purpose, and a general result covering many cases is given in Problem 29. Other useful properties of Laplace transforms are derived later in this chapter. As further illustrations of the technique of solving initial value problems by means of the Laplace transform and partial fraction expansions, consider the following examples.

EXAMPLE 2 Find the solution of the differential equation y + y = sin( 2t)

(19)

satisfying the initial conditions y( 0) = 2,

y ( 0) = 1.

(20)

Solution: We assume that this initial value problem has a solution y( t) , which with its first two derivatives satisfies the conditions of Corollary 6.2.2. Then, taking the Laplace transform of the differential equation (19), we have s 2 Y ( s) − sy( 0) − y ( 0) + Y ( s) =

2 , s2 + 4

where the transform of sin( 2t) has been obtained from line 5 of Table 6.2.1. Substituting for y( 0) and y ( 0) from the initial conditions (20) and solving for Y ( s) , we obtain Y ( s) =

2s 3 + s 2 + 8s + 6 . ( s 2 + 1) ( s 2 + 4)

(21)

Using partial fractions, we can write Y ( s) in the form Y ( s) =

as + b cs + d ( as + b) ( s 2 + 4) + ( cs + d) ( s 2 + 1) + 2 = . 2 s +1 s +4 ( s 2 + 1) ( s 2 + 4)

(22)

By expanding the numerator on the right-hand side of equation (22) and equating it to the numerator in equation (21), we find that 2s 3 + s 2 + 8s + 6 = ( a + c) s 3 + ( b + d) s 2 + ( 4a + c) s + ( 4b + d)

(23)

for all s. Then, comparing coefficients of like powers of s, we have3 a + c = 2,

b + d = 1,

4a + c = 8,

4b + d = 6.

..................................................................................................................................................................................... 3 We could find the value of the four coefficients by evaluating equation (23) for four different values of s, but, unlike Example 1, it is not obvious what four values of s will give trivial equations to solve for a, b, c, and d.

▼

253

Boyce 9131 Ch06 2

254

September 29, 2016

17:34

254

CHAPTER 6 The Laplace Transform

▼

Consequently, a = 2, c = 0, b =

2 5 , and d = − , from which it follows that 3 3

Y ( s) =

5/3 2/3 2s + − . s2 + 1 s2 + 1 s2 + 4

(24)

From lines 5 and 6 of Table 6.2.1, the solution of the given initial value problem is y = 2 cos t +

5 1 sin t − sin( 2t) . 3 3

(25)

EXAMPLE 3 Find the solution of the initial value problem y ( 4) − y = 0, y( 0) = 0,

y ( 0) = 1,

y ( 0) = 0,

(26) y ( 0) = 0.

(27)

Solution: In this problem we need to assume that the solution y( t) satisfies the conditions of Corollary 6.2.2 for n = 4. The Laplace transform of the differential equation (26) is s 4 Y ( s) − s 3 y( 0) − s 2 y ( 0) − sy ( 0) − y ( 0) − Y ( s) = 0. Then, using the initial conditions (27) and solving for Y ( s) , we have Y ( s) =

s4

s2 . −1

(28)

A partial fraction expansion of Y ( s) is Y ( s) =

as + b cs + d + 2 , s2 − 1 s +1

(29)

and it follows that ( as + b) ( s 2 + 1) + ( cs + d) ( s 2 − 1) = s 2

(30)

for all s. In this problem we use a combination of substituting values of s and equating coefficients of like powers of s. First, setting s = 1 and s = −1, respectively, in equation (30), we obtain the pair of equations 2( a + b) = 1,

2( −a + b) = 1,

1 1 . If we set s = 0 in equation (30), then b − d = 0, so d = . Finally, 2 2 equating the coefficients of the cubic terms on each side of equation (30), we find that a + c = 0, so c = 0. Thus and therefore a = 0 and b =

Y ( s) =

1/2 1/2 + , s2 − 1 s2 + 1

(31)

and from lines 7 and 5 of Table 6.2.1, the solution of the initial value problem (26), (27) is y( t) =

1 ( sinh t + sin t) . 2

(32)

We conclude by noting that we could have looked for a partial fraction expansion of Y ( s) in the form Y ( s) =

a b cs + d . + + 2 s−1 s+1 s +1

We used the form in equation (29) because Table 6.2.1 includes entries for both 1/( s 2 ± 1) and s/( s 2 ± 1) .

Boyce 9131 Ch06 2

September 29, 2016

17:34

255

6.2 Solution of Initial Value Problems

255

The most important elementary applications of the Laplace transform are in the study of mechanical vibrations and in the analysis of electric circuits; the governing equations were derived in Section 3.7. A vibrating spring-mass system has the equation of motion m

d 2u du + ku = F( t) , +γ dt dt 2

(33)

where m is the mass, γ the damping coefficient, k the spring constant, and F( t) the applied external force. The equation that describes an electric circuit containing an inductance L, a resistance R, and a capacitance C (an LRC circuit) is L

d2 Q dQ 1 +R + Q = E( t) , dt C dt 2

(34)

where Q( t) is the charge on the capacitor and E( t) is the applied voltage. In terms of the current I ( t) = d Q( t) /dt, we can differentiate equation (34) and write L

d2 I dI 1 dE +R + I = ( t) . 2 dt C dt dt

(35)

Suitable initial conditions on u, Q, or I must also be prescribed. We have noted previously, in Section 3.7, that equation (33) for the spring-mass system and equations (34) or (35) for the electric circuit are identical mathematically, differing only in the interpretation of the constants and variables appearing in them. There are other physical problems that also lead to the same differential equation. Thus, once the mathematical problem is solved, its solution can be interpreted in terms of whichever corresponding physical problem is of immediate interest. In the problem lists following this and other sections in this chapter are numerous initialvalue problems for second-order linear differential equations with constant coefficients. Many can be interpreted as models of particular physical systems, but usually we do not point this out explicitly.

Problems In each of Problems 1 through 7, find the inverse Laplace transform of the given function. 3 1. F( s) = 2 s +4 4 2. F( s) = ( s − 1) 3 2 3. F( s) = 2 s + 3s − 4 2s + 2 4. F( s) = 2 s + 2s + 5 2s − 3 5. F( s) = 2 s −4 8s 2 − 4s + 12 s( s 2 + 4) 1 − 2s 7. F( s) = 2 s + 4s + 5 In each of Problems 8 through 16, use the Laplace transform to solve the given initial value problem.

6. F( s) =

8. y − y − 6y = 0; y( 0) = 1, y ( 0) = −1 9. y + 3y + 2y = 0; y( 0) = 1, y ( 0) = 0

10. y − 2y + 2y = 0; 11. y − 2y + 4y = 0; 12. y + 2y + 5y = 0;

y( 0) = 0, y ( 0) = 1 y( 0) = 2, y ( 0) = 0 y( 0) = 2, y ( 0) = −1

13. y ( 4) − 4y + 6y − 4y + y = 0; y ( 0) = 1, y ( 0) = 0, y ( 0) = 1

14. y ( 4) − y = 0;

y ( 0) = 0

y( 0) = 0,

y( 0) = 1, y ( 0) = 0, y ( 0) = 1,

15. y + ω 2 y = cos( 2t) , ω 2 = 4; y( 0) = 1, y ( 0) = 0 16. y − 2y + 2y = e−t ; y( 0) = 0, y ( 0) = 1 In each of Problems 17 through 19, find the Laplace transform Y ( s) = L{y} of the solution of the given initial value problem. A method of determining the inverse transform is developed in Section 6.3. You may wish to refer to Problems 16 through 18 in Section 6.1.

17. y + 4y =

1, 0,

0 ≤ t < π, y( 0) = 1, y ( 0) = 0 π ≤ t < ∞;

t, 0 ≤ t < 1, y( 0) = 0, y ( 0) = 0 18. y + 4y = 1, 1 ≤ t < ∞; ⎧ 0 ≤ t < 1, ⎨t, 19. y + y = 2 − t, 1 ≤ t < 2, y( 0) = 0, y ( 0) = 0 ⎩ 0,

2 ≤ t < ∞;

Boyce 9131 Ch06 2

256

September 29, 2016

17:34

256

CHAPTER 6 The Laplace Transform

20. The Laplace transforms of certain functions can be found

26. Consider Bessel’s equation of order zero

conveniently from their Taylor series expansions.

a. Using the Taylor series for sin t ∞ ( −1) n t 2n+1

sin t =

n=0

( 2n + 1) !

,

and assuming that the Laplace transform of this series can be computed term-by-term, verify that 1 , L{sin t} = 2 s +1

b. Let

s > 1.

1 , s

s > 1.

c. The Bessel function of the first kind of order zero, J0 , has the Taylor series (see Section 5.7) J0 ( t) =

n=0

( −1) n t 2n . 22n ( n!) 2

and

= ( s 2 + 1) −1/2 ,

s>1

√ L J0 ( t) = s −1 e−1/( 4s) ,

s > 0.

Problems 21 through 27 are concerned with differentiation of the Laplace transform.

21. Let

F( s) =

∞

e−st f ( t) dt.

It is possible to show that as long as f satisfies the conditions of Theorem 6.1.2, it is legitimate to differentiate under the integral sign with respect to the parameter s when s > a.

a. Show that F ( s) = L{−t f ( t) }. b. Show that F ( n) ( s) = L{( −t) n f ( t) }; hence differentiating the Laplace transform corresponds to multiplying the original function by −t. In each of Problems 22 through 25, use the result of Problem 21 to find the Laplace transform of the given function; a and b are real numbers and n is a positive integer. f ( t) = teat

23.

f ( t) = t 2 sin( bt)

24. 25.

f ( t) = t e

n at

f ( t) = teat sin( bt)

n=0

22n ( n!) 2

= c J0 ( t) ,

where J0 is the Bessel function of the first kind of order zero. Note that J0 ( 0) = 1 and that J0 has finite derivatives of all orders at t = 0. It was shown in Section 5.7 that the second solution of this equation becomes unbounded as t → 0.

27. For each of the following initial value problems, use the results of Problem 21 to find the differential equation satisfied by Y ( s) = L{y( t) }, where y( t) is the solution of the given initial value problem.

a. y − t y = 0; y( 0) = 1, y ( 0) = 0 (Airy’s equation) b. ( 1−t 2 ) y −2t y +α ( α +1) y = 0; y( 0) = 0, y ( 0) = 1 Note that the differential equation for Y ( s) is of first-order in part a, but of second-order in part b. This is due to the fact that t appears at most to the first power in the equation of part a, whereas it appears to the second power in that of part b. This illustrates that the Laplace transform is not often useful in solving differential equations with variable coefficients, unless all the coefficients are at most linear functions of the independent variable.

28. Suppose that

t

g( t) =

f ( τ ) dτ . 0

0

22.

∞ ( −1) n t 2n

(Legendre’s equation)

Assuming that the following Laplace transforms can be computed term-by-term, verify that L J0 ( t)

b. Show that Y ( s) = c( 1 + s 2 ) −1/2 , where c is an arbitrary constant. c. Writing ( 1 + s 2 ) −1/2 = s −1 ( 1 + s −2 ) −1/2 , expanding in a binomial series valid for s > 1, and assuming that it is permissible to take the inverse transform term-by-term, show that y=c

Show that f ( t) is continuous for all real values of t. Find the Taylor series for f about t = 0. Assuming that the Laplace transform of this function can be computed term-by-term, verify that

∞

Recall from Section 5.7 that t = 0 is a regular singular point for this equation, and therefore solutions may become unbounded as t → 0. However, let us try to determine whether there are any solutions that remain finite at t = 0 and have finite derivatives there. Assuming that there is such a solution y = φ ( t) , let Y ( s) = L{φ ( t) }. a. Show that Y ( s) satisfies ( 1 + s 2 ) Y ( s) + sY ( s) = 0.

⎧ ⎨ sin t , t = 0, f ( t) = t ⎩1, t = 0.

L{ f ( t) } = arctan

t y + y + t y = 0.

If G( s) and F( s) are the Laplace transforms of g( t) and f ( t) , respectively, show that G( s) =

F( s) . s

29. In this problem we show how a general partial fraction expansion can be used to calculate many inverse Laplace transforms. Suppose that F( s) =

P( s) , Q( s)

where Q( s) is a polynomial of degree n with n distinct zeros r1 , . . . , rn , and P( s) is a polynomial of degree less than n. In this case it is possible to show that P( s) / Q( s) has a partial fraction expansion of the form A1 P( s) An = + ··· + , Q( s) s − r1 s − rn where the coefficients A1 , . . . , An must be determined.

(36)

Boyce 9131 Ch06 2

September 29, 2016

17:34

257

6.3 Step Functions

by s − rk because equation (36) is not defined at each root of Q( s) . b. Show that

a. Show that Ak =

P(rk ) , Q (rk )

k = 1, . . . , n.

L−1 {F( s) } =

Hint: One way to do this is to multiply equation (36) by s −rk and then to take the limit as s → rk . Note that limits are used because it is not appropriate to simply evaluate equation (36) multiplied

6.3

In Section 6.2 we outlined the general procedure involved in solving initial value problems by means of the Laplace transform. Some of the most interesting elementary applications of the transform method occur in the solution of linear differential equations with discontinuous or impulsive forcing functions. Equations of this type frequently arise in the analysis of the flow of current in electric circuits or the vibrations of mechanical systems. In this section and the remaining sections in Chapter 6, we develop some additional properties of the Laplace transform that are useful in the solution of such problems. Unless a specific statement is made to the contrary, all functions appearing below will be assumed to be piecewise continuous and of exponential order, so that their Laplace transforms exist, at least for s sufficiently large. To deal effectively with functions having jump discontinuities, it is very helpful to introduce a function known as the unit step function or Heaviside function. This function will be denoted by u c and is defined by 0, t < c, u c ( t) = (1) 1, t ≥ c. Since the Laplace transform involves values of t in the interval [0, ∞) , we are also interested only in nonnegative values of c. The graph of y = u c ( t) is shown in Figure 6.3.1. We have somewhat arbitrarily assigned the value one to u c at t = c. However, for a piecewise continuous function such as u c , the value at a discontinuity point is usually irrelevant. The step can also be negative. For instance, Figure 6.3.2 shows the graph of y = 1 − u c ( t) . y

y

1

1

t

FIGURE 6.3.1 Graph of y = u c ( t) .

c

t

FIGURE 6.3.2 Graph of y = 1 − u c ( t) .

If we associate the value 1 with “on” and 0 with “off,” then the function u c ( t) represents a switch that is turned on at time c. Likewise, 1 − u c ( t) represents a switch being turned off at time c.

EXAMPLE 1 Sketch the graph of y = h( t) , where h( t) = u π ( t) − u 2π ( t) ,

▼

t ≥ 0.

n P(rk ) k=1

Step Functions

c

257

Q (rk )

er k t .

Boyce 9131 Ch06 2

258

September 29, 2016

17:34

258

CHAPTER 6 The Laplace Transform

▼ Solution: From the definition of u c ( t) in equation (1), we have

h( t) =

=

0, 1,

t < π, − t ≥π

1, 0,

0 ≤ t < π, π ≤ t < 2π, 2π ≤ t < ∞.

⎧ ⎨ 0, ⎩

0, 1,

⎧ ⎨0 − 0,

t < 2π, = 1 − 0, t ≥ 2π ⎩ 1 − 1,

0 ≤ t < π, π ≤ t < 2π, 2π ≤ t < ∞,

Thus the equation y = h( t) has the graph shown in Figure 6.3.3. This function can be thought of as a switch that is initially off, turned on at t = π and then turned off at t = 2π ; this is also often referred to as a rectangular pulse. y 1

π

2π

t

3π

FIGURE 6.3.3 Graph of y = u π ( t) − u 2π ( t) .

EXAMPLE 2 Consider the function

⎧ 2, 0 ≤ t < 4, ⎪ ⎪ ⎨ 5, 4 ≤ t < 7, f ( t) = −1, 7 ≤ t < 9, ⎪ ⎪ ⎩ 1,

(2)

t ≥ 9.

Sketch the graph of y = f ( t) . Express f ( t) in terms of u c ( t) . Solution: The graph of y = f ( t) is piecewise constant. Paying attention to include the left endpoint of each horizontal segment, we arrive at Figure 6.3.4.

y 5 4 3 2 1

2

4

6

8

10

–1 FIGURE 6.3.4 Graph of the function in equation (2).

▼

12 t

Boyce 9131 Ch06 2

September 29, 2016

17:34

259

6.3 Step Functions

▼

We start with the function f 1 ( t) = 2, which agrees with f ( t) on [0, 4) . To produce the jump of three units at t = 4, we add 3u 4 ( t) to f 1 ( t) , obtaining f 2 ( t) = 2 + 3u 4 ( t) , which agrees with f ( t) on [0, 7) . The negative jump of six units at t = 7 corresponds to adding −6u 7 ( t) , which gives f 3 ( t) = 2 + 3u 4 ( t) − 6u 7 ( t) . Finally, we must add 2u 9 ( t) to match the jump of two units at t = 9. Thus we obtain f ( t) = 2 + 3u 4 ( t) − 6u 7 ( t) + 2u 9 ( t) .

(3)

The Laplace transform of u c for c ≥ 0 is easily determined: ∞ ∞ L{u c ( t) } = e−st u c ( t) dt = e−st dt c

0

=

e−cs , s

s > 0.

(4)

Notice that L{u 0 ( t) } =

1 e0 = = L{1}. s s

This is true because u 0 ( t) = 1 for all t ≥ 0. For a given function f defined for t ≥ 0, we will often want to consider the related function g defined by 0, t < c, g( t) = f ( t − c) , t ≥ c, which represents a translation of f a distance c in the positive t direction and is zero for t < c; see Figure 6.3.5. Making use of the unit step function, we can write g( t) in the convenient form g( t) = u c ( t) f ( t − c) . y

y

f (0)

f (0)

t

c

(a)

t (b)

FIGURE 6.3.5 A translation of the given function. (a) y = f ( t) ; (b) y = u c ( t) f ( t − c) .

The unit step function is particularly important in Laplace transform use because of the following relation between the transform of f ( t) and that of its translation u c ( t) f ( t − c) .

Theorem 6.3.1 If the Laplace transform of f ( t) , F( s) = L{ f ( t) }, exists for s > a ≥ 0, and if c is a positive constant, then L{u c ( t) f ( t − c) } = e−cs L{ f ( t) } = e−cs F( s) ,

s > a.

(5)

Conversely, if f ( t) is the inverse Laplace transform of F( s) , f ( t) = L−1 {F( s) }, then u c ( t) f ( t − c) = L−1 {e−cs F( s) }.

(6)

259

Boyce 9131 Ch06 2

260

September 29, 2016

17:34

260

CHAPTER 6 The Laplace Transform

Theorem 6.3.1 simply states that the translation of f ( t) a distance c in the positive t direction corresponds to the multiplication of F( s) by e−cs . To prove Theorem 6.3.1, it is sufficient to compute the transform of u c ( t) f ( t − c) : ∞ L{u c ( t) f ( t − c) } = e−st u c ( t) f ( t − c) dt 0

=

∞

e−st f ( t − c) dt.

c

Introducing a new integration variable σ = t − c, we have ∞ −( σ +c) s −cs L{u c ( t) f ( t − c) } = e f ( σ ) dσ = e 0

∞

e−sσ f ( σ ) dσ

0

= e−cs F( s) . Thus equation (5) is established; equation (6) follows by taking the inverse transform of both sides of equation (5). A simple example of this theorem occurs if we take f ( t) = 1. Recalling that L{1} = 1/s, we immediately have from equation (5) that L{u c ( t) } = e−cs /s. This result agrees with that of equation (4). Examples 3 and 4 illustrate further how Theorem 6.3.1 can be used in the calculation of Laplace transforms and inverse Laplace transforms.

EXAMPLE 3 Given the function f defined by

f ( t) =

⎧ ⎪ ⎨sin t,

0≤t

a ≥ 0, and if c is a constant, then L{ect f ( t) } = F( s − c) ,

s > a + c.

(7)

−1

Conversely, if f ( t) = L {F( s) }, then ect f ( t) = L−1 {F( s − c) }.

(8)

According to Theorem 6.3.2, multiplication of f ( t) by ect results in a translation of the transform F( s) a distance c in the positive s direction, and conversely. To prove this theorem, we evaluate L{ect f ( t) }. Thus ∞ ∞ ct −st ct L{e f ( t) } = e e f ( t) dt = e−( s−c) t f ( t) dt 0

0

= F( s − c) , which is equation (7). The restriction s > a + c follows from the observation that, according to hypothesis (ii) of Theorem 6.1.2, | f ( t) | ≤ K eat ; hence |ect f ( t) | ≤ K e( a+c) t . Equation (8) is obtained by taking the inverse transform of equation (7), and the proof is complete. The principal application of Theorem 6.3.2 is in the evaluation of certain inverse transforms, as illustrated by Example 5.

EXAMPLE 5 Find the inverse Laplace transform of G( s) =

1 . s 2 − 4s + 5

Solution: First, to avoid dealing with the complex-valued roots of the denominator s 2 − 4s + 5, we complete the square in the denominator: G( s) =

1 = F( s − 2) , ( s − 2) 2 + 1

where F( s) = ( s 2 + 1) −1 . Since L−1 {F( s) } = sin t, it follows from Theorem 6.3.2 that g( t) = L−1 {G( s) } = e2t sin t.

The results of this section are often useful in solving differential equations, particularly those that have discontinuous forcing functions. The next section is devoted to examples illustrating this point.

Problems In each of Problems 1 through 4, sketch the graph of the given function on the interval t ≥ 0.

1. g( t) = u 1 ( t) + 2u 3 ( t) − 6u 4 ( t)

2. g( t) = f ( t − π ) u π ( t) , where f ( t) = t 2 3. g( t) = f ( t − 3) u 3 ( t) , where f ( t) = sin t 4. g( t) = ( t − 1) u 1 ( t) − 2( t − 2) u 2 ( t) + ( t − 3) u 3 ( t)

Boyce 9131 Ch06 2

September 29, 2016

17:34

263

6.3 Step Functions

In each of Problems 5 through 8: a. Sketch the graph of the given function. b. Express f ( t) in terms of the unit step function u c ( t) . ⎧ 0, 0 ≤ t < 3, ⎪ ⎪ ⎨−2, 3 ≤ t < 5, 5. f ( t) = ⎪ 2, 5 ≤ t < 7, ⎪ ⎩ 1, t ≥ 7.

6.

f ( t) =

⎧ 1, 0 ≤ t < 1, ⎪ ⎪ ⎪ ⎪ ⎪ ⎨−1, 1 ≤ t < 2,

7.

8.

f ( t) =

f ( t) =

1,

2 ≤ t < 3,

0,

t ≥ 4.

⎪ ⎪ ⎪ ⎪ −1, 3 ≤ t < 4, ⎪ ⎩ 1,

0 ≤ t < 2,

e−( t−2) ,

t ≥ 2.

⎧ t, ⎪ ⎪ ⎪ ⎨

22.

⎪ 7 − t, 5 ≤ t < 7, ⎪ ⎪ ⎩

23.

11. 12.

1

1

t − π,

π ≤ t < 2π

0,

t ≥ 2π

2

3

4

5

t

FIGURE 6.3.8 The function f ( t) in Problem 23;

a square wave.

24. Let f satisfy f ( t + T ) = f ( t) for all t ≥ 0 and for some

f ( t) = ( t − 3) u 2 ( t) − ( t − 2) u 3 ( t)

fixed positive number T ; f is said to be periodic with period T on 0 ≤ t < ∞. Show that

In each of Problems 13 through 16, find the inverse Laplace transform of the given function. 3! 13. F( s) = ( s − 2) 4 e−2s +s−2 2( s − 1) e−2s 15. F( s) = 2 s − 2s + 2 e−s + e−2s − e−3s − e−4s 16. F( s) = s 17. Suppose that F( s) = L{ f ( t) } exists for s > a ≥ 0. a. Show that if c is a positive constant, then s2

L{ f ( ct) } =

( −1) k u k ( t) . See Figure 6.3.8.

y

t 0, then t 1 −1

e−st f ( t) dt

0

In each of Problems 25 through 28, use the result of Problem 24 to find the Laplace transform of the given function. 1, 0 ≤ t < 1, f ( t + 2) = f ( t) . 25. f ( t) = 0, 1 ≤ t < 2; Compare with Problem 23. 1, 0 ≤ t < 1, f ( t + 2) = f ( t) . 26. f ( t) = −1, 1 ≤ t < 2; See Figure 6.3.9.

b. Show that if k is a positive constant, then t 1 L−1 {F( ks) } =

T

a

–1

.

FIGURE 6.3.9 The function f ( t) in Problem 26; a

square wave.

Boyce 9131 Ch06 2

264

27.

September 29, 2016

17:34

264

CHAPTER 6 The Laplace Transform

f ( t) = t,

0 ≤ t < 1;

29. a. If f ( t) = 1 − u 1 ( t) , find L{ f ( t) }. Sketch the graph of y = f ( t) . Compare t with Problem 21. b. Let g( t) = f ( ξ ) dξ , where the function f is defined in

f ( t +1) = f ( t) .

See Figure 6.3.10. y

0

1

1

2

3

4

t

30. Consider the function p defined by

FIGURE 6.3.10 The function f ( t) in Problem 27; a sawtooth wave.

28.

f ( t) = sin t,

0 ≤ t < π;

part a. Sketch the graph of y = g( t) and find L{g( t) }. Use your expression for L{g( t) } to find an explicit formula for g( t) . Hint: See Problem 28 in Section 6.2. c. Let h( t) = g( t) − u 1 ( t) g( t − 1) , where g is defined in part b. Sketch the graph of y = h( t) and find L{h( t) }. Use your expression for L{h( t) } to find an explicit formula for h( t) .

p( t) =

f ( t + π ) = f ( t) .

t, 2 − t,

0 ≤ t < 1, 1 ≤ t < 2;

p( t + 2) = p( t) .

a. Sketch the graph of y = p( t) . b. Find L{ p( t) } by noting that p is the periodic extension of the

See Figure 6.3.11.

function h in Problem 29c; then use the result of Problem 24. c. Find L{ p( t) } by noting that

y

p( t) =

1

t

f ( t) dt, 0

π

2π

3π

t

where f is the function in Problem 26; then use Theorem 6.2.1.

FIGURE 6.3.11 The function f ( t) in Problem 28; a rectified sine wave.

Differential Equations with Discontinuous Forcing Functions 6.4

In this section we turn our attention to some examples in which the nonhomogeneous term, or forcing function, is discontinuous.

EXAMPLE 1 Find the solution of the differential equation 2y + y + 2y = g( t) , where

g( t) = u 5 ( t) − u 20 ( t) =

1, 0,

5 ≤ t < 20, 0 ≤ t < 5 or t ≥ 20.

(1)

(2)

Assume that the initial conditions are y( 0) = 0,

▼

y ( 0) = 0.

(3)

Boyce 9131 Ch06 2

September 29, 2016

17:34

265

6.4 Differential Equations with Discontinuous Forcing Functions

▼ This problem governs the charge on the capacitor in a simple electric circuit with a unit voltage pulse

for 5 ≤ t < 20. Alternatively, y may represent the response of a damped oscillator subject to the applied force g( t) . Solution: The Laplace transform of equation (1) is 2s 2 Y ( s) − 2sy( 0) − 2y ( 0) + sY ( s) − y( 0) + 2Y ( s) = L{u 5 ( t) } − L{u 20 ( t) } =

1 −5s e − e−20s . s

Introducing the initial values (3) and solving for Y ( s) , we obtain e−5s − e−20s . s( 2s 2 + s + 2)

Y ( s) =

(4)

To find y( t) , it is convenient to write Y ( s) as Y ( s) = ( e−5s − e−20s ) H ( s) ,

(5)

where H ( s) =

s( 2s 2

1 . + s + 2)

(6)

Then, if h( t) = L−1 {H ( s) }, we have y( t) = u 5 ( t) h( t − 5) − u 20 ( t) h( t − 20) .

(7) −5s

Observe that we have used Theorem 6.3.1 to write the inverse transforms of e H ( s) and e−20s H ( s) , respectively. Finally, to determine h( t) , we use the partial fraction expansion of H ( s) : bs + c a . (8) H ( s) = + 2 s 2s + s + 2 Upon determining the coefficients, we find that a =

1 1 , b = −1, and c = − . Thus 2 2

1 1 s+ + s+ 4 1 1 1 2 = s− H ( s) = s − 2 2 2 2 2 2s + s + 2 1 s+ + 4

⎛

1 = s− 2

⎜ ⎜

1⎜ ⎜ 2⎜

⎝

1 s+ 4

2 +

15 16

1 4

s+

1 4

1

2 +

15

15 4

1 s+ 4

2

⎞ ⎟ ⎟ ⎟ . 2 ⎟ ⎟ 15 ⎠

15 4

+

(9)

4

Then, by referring to lines 9 and 10 of Table 6.2.1, we obtain

1 1 h( t) = − 2 2

e

−t/4

cos

15 t 4

1

+

15

e

−t/4

sin

15 t 4

.

(10)

In Figure 6.4.1 the graph of y( t) from equations (7) and (10) shows that the solution consists of three distinct parts. For 0 < t < 5, the differential equation is 2y + y + 2y = 0,

(11)

and the initial conditions are given by equation (3). Since the initial conditions impart no energy to the system, and since there is no external forcing, the system remains at rest; that is, y = 0 for 0 < t < 5. This can be confirmed by solving equation (11) subject to the initial conditions (3). In particular, evaluating the solution and its derivative at t = 5, or, more precisely, as t approaches 5 from below, we have y( 5) = 0,

y ( 5) = 0.

(12)

Once t > 5, the differential equation becomes 2y + y + 2y = 1,

▼

(13)

265

Boyce 9131 Ch06 2

266

September 29, 2016

17:34

266

CHAPTER 6 The Laplace Transform

▼ whose solution is the sum of a constant (the response to the constant forcing function) and a damped oscillation (the solution of the corresponding homogeneous equation). The plot in Figure 6.4.1 shows this behavior clearly for the interval 5 ≤ t ≤ 20. An expression for this portion of the solution can be found by solving the differential equation (13) subject to the initial conditions (12). Alternatively, since u 5 ( t) = 1 and u 20 ( t) = 0 for 5 ≤ t < 20, equations (7) and (10) reduce to y( t) = h( t − 5)

15( t − 5) 4

1 1 = − e−( t−5) /4 cos 2 2

1

+ e 2 15

−( t−5) /4

sin

15( t − 5) 4

. (14)

Finally, for t > 20 the differential equation becomes equation (11) again, and the initial conditions are obtained by evaluating the solution of equations (13), (12), that is, equation (14) and its derivative, at t = 20. These values are, approximately, ∼ 0.01125. ∼ 0.50162, y ( 20) = (15) y( 20) = The initial value problem (7), (10) contains no external forcing, so its solution is a damped oscillation about y = 0, as can be seen in Figure 6.4.1. y 0.8

0.6

0.4

0.2 10

20

30

40 t

– 0.2

FIGURE 6.4.1 Solution of the initial value problem (1), (2), (3): 2y + y + 2y = u 5 ( t) − u 20 ( t) , y( 0) = 0, y ( 0) = 0.

Although it may be helpful to visualize the solution shown in Figure 6.4.1 as composed of solutions of three separate initial value problems in three separate intervals, it is somewhat tedious to find the solution by solving these separate problems. Laplace transform methods provide a much more convenient and elegant approach to the problem in Example 1 and to others that have discontinuous forcing functions. The effect of the discontinuity in the forcing function can be seen if we examine the solution y( t) of Example 1 more closely. According to the existence and uniqueness theorem (Theorem 3.2.1), the solution y( t) and its first two derivatives are continuous except possibly at the points t = 5 and t = 20, where g is discontinuous. This can also be seen at once from equation (7). One can also show by direct computation from equation (7) that y( t) and y ( t) are continuous even at t = 5 and t = 20. However, if we calculate y ( t) , we find that 1 lim y ( t) = 0, lim y ( t) = . 2 t→5− t→5+ Consequently, y ( t) has a jump of

1 at t = 5. In a similar way, we can show that y ( t) has a 2

1 at t = 20. Thus the jump in the forcing term g( t) at these points is balanced by 2 a corresponding jump in the highest order term 2y on the left-hand side of the equation. Consider now the general second-order linear equation jump of −

y + p( t) y + q( t) y = g( t) ,

(16)

where p and q are continuous on some interval α < t < β , but g is only piecewise continuous there. If y( t) is a solution of equation (16), then y( t) and y ( t) are continuous on α < t < β ,

Boyce 9131 Ch06 2

September 29, 2016

17:34

267

6.4 Differential Equations with Discontinuous Forcing Functions

but y ( t) has jump discontinuities at the same points as g. Similar remarks apply to higherorder equations; the highest derivative of the solution appearing in the differential equation has jump discontinuities at the same points as the jump discontinuities in the forcing function, but the solution itself and its lower derivatives are continuous even at those points.

EXAMPLE 2 Describe the qualitative nature of the solution of the initial value problem y + 4y = g( t) , y ( 0) = 0,

y( 0) = 0, where

(17)

⎧ 0, ⎪ ⎨

(18)

0 ≤ t < 5,

1 g( t) = ( t − 5) , ⎪ 5 ⎩ 1,

5 ≤ t < 10,

(19)

t ≥ 10,

and then find the solution. Solution: In this example the forcing function has the graph shown in Figure 6.4.2 and is known as ramp loading. It is relatively easy to identify the general form of the solution. For t < 5, the solution is simply y = 0. On the other hand, for t > 10, the solution has the form 1 y = c1 cos( 2t) + c2 sin( 2t) + . 4

(20)

The constant 1/4 is a particular solution of the nonhomogeneous equation, while the other two terms are the general solution of the corresponding homogeneous equation. Thus the solution (20) is a simple harmonic oscillation about y = 1/4. Similarly, in the intermediate range 5 < t < 10, the solution is an oscillation about a certain linear function. In an engineering context, for example, we might be interested in knowing the amplitude of the eventual steady oscillation. y 1 y = g(t) 0.5

5

10

15

20 t

FIGURE 6.4.2 Ramp loading; y = g( t) from equation (19) or equation (21).

To solve the problem, it is convenient to write g( t) =

1 u 5 ( t) ( t − 5) − u 10 ( t) ( t − 10) , 5

(21)

as you may verify. Then we take the Laplace transform of the differential equation and use the initial conditions, thereby obtaining ( s 2 + 4) Y ( s) =

e−5s − e−10 5s 2

or Y ( s) =

1 −5s e − e−10s H ( s) , 5

(22)

1 . s 2 ( s 2 + 4)

(23)

where H ( s) =

▼

267

Boyce 9131 Ch06 2

268

September 29, 2016

17:34

268

CHAPTER 6 The Laplace Transform

▼ Thus the solution of the initial value problem (17), (18), (19) is 1 y( t) =

5

u 5 ( t) h( t − 5) − u 10 ( t) h( t − 10) ,

(24)

where h( t) is the inverse transform of H ( s) . The partial fraction expansion of H ( s) is H ( s) =

1/4 1/4 , − 2 s2 s +4

(25)

and it then follows from lines 3 and 5 of Table 6.2.1 that h( t) =

1 1 t − sin( 2t) . 4 8

(26)

The graph of y( t) is shown in Figure 6.4.3. Observe that it has the qualitative form that we indicated earlier. To find the amplitude of the eventual steady oscillation, it is sufficient to locate one of the maximum or minimum points for t > 10. Setting the derivative of the solution (24) equal to zero, we find that the first maximum is located approximately at ( 10.642, 0.2979) , so the amplitude of the oscillation is approximately 0.2979 − 0.25 = 0.0479. y 0.30

0.20

0.10

5

10

15

20

t

FIGURE 6.4.3 Solution of the initial value problem (12), (13), (14).

Note that in this example, the forcing function g is continuous, but g is discontinuous at t = 5 and t = 10. It follows that the solution y( t) and its first two derivatives are continuous everywhere, but y ( t) has discontinuities at t = 5 and at t = 10 that match the discontinuities in g at those points.

Problems In each of Problems 1 through 8: a. Sketch the graph of the forcing function on an appropriate interval. b. Find the solution of the given initial value problem. G c. Plot the graph of the solution. d. Explain how the graphs of the forcing function and the solution are related.

1. y + y = f ( t) ; f ( t) =

y( 0) = 0, y ( 0) = 1; 0 ≤ t < 3π 3π ≤ t < ∞

1, 0,

2. y + 2y + 2y = h( t) ; h( t) =

1, 0,

y( 0) = 0, y ( 0) = 1; π ≤ t < 2π 0 ≤ t < π or t ≥ 2π

3. y + 4y = sin t − u 2π ( t) sin( t − 2π ) ; y( 0) = 0, y ( 0) = 0 4. y + 3y + 2y = f ( t) ; y( 0) = 0, y ( 0) = 0; f ( t) =

1, 0,

0 ≤ t < 10 t ≥ 10

5 π 5. y + y + y = t −u π/2 ( t) t − ; 4

2

5

6. y + y + y = g( t) ; 4 g( t) =

sin t, 0,

y( 0) = 0, y ( 0) = 0

y( 0) = 0, y ( 0) = 0;

0≤t < π t ≥π

7. y + 4y = u π ( t) − u 3π ( t) ;

y( 0) = 0, y ( 0) = 0

Boyce 9131 Ch06 2

September 29, 2016

17:34

269

6.4 Differential Equations with Discontinuous Forcing Functions

8. y ( 4) + 5y + 4y = 1 − u π ( t) ;

y( 0) = 0, y ( 0) = 0,

y ( 0) = 0, y ( 0) = 0

9. Find an expression involving u c ( t) for a function f that ramps up from zero at t = t0 to the value h at t = t0 + k. 10. Find an expression involving u c ( t) for a function g that ramps up from zero at t = t0 to the value h at t = t0 + k and then ramps back down to zero at t = t0 + 2k.

11. A certain spring-mass system satisfies the initial value problem 1 u + u + u = kg( t) , 4

the given forcing function g( t) by

1 u 5 ( t) ( t − 5) − u 5+k ( t) ( t − 5 − k) / k. k

a. Sketch the graph of f ( t) and describe how it depends on k. For what value of k is f ( t) identical to g( t) in the example? b. Solve the initial value problem y + 4y = f ( t) ,

y( 0) = 0, y ( 0) = 0.

G c. The solution in part b depends on k, but for sufficiently large t, the solution is always a simple harmonic oscillation about y = 1/4. Try to decide how the amplitude of this eventual oscillation depends on k. Then confirm your conclusion by plotting the solution for a few different values of k.

Resonance and Beats. In Section 3.8 we observed that an undamped harmonic oscillator (such as a spring-mass system) with a sinusoidal forcing term experiences resonance if the frequency of the forcing term is the same as the natural frequency. If the forcing frequency is slightly different from the natural frequency, then the system exhibits a beat. In Problems 13 through 17 we explore the effect of some nonsinusoidal periodic forcing functions.

13. Consider the initial value problem y + y = f ( t) ,

y( 0) = 0, y ( 0) = 0,

where f ( t) = u 0 ( t) + 2

n

( −1) k u kπ ( t) .

k=1

a. Draw the graph of f ( t) on an interval such as 0 ≤ t ≤ 6π . b. Find the solution of the initial value problem. G c. Let n = 15. Plot the graph of the solution for 0 ≤ t ≤ 60. Describe the solution and explain why it behaves as it does.

d. Investigate how the solution changes as n increases. What happens as n → ∞?

14. Consider the initial value problem y + 0.1y + y = f ( t) ,

15. Consider the initial value problem y + y = g( t) ,

u( 0) = 0, u ( 0) = 0,

12. Modify the problem in Example 2 of this section by replacing f ( t) =

G a. Plot the graph of the solution. Use a large enough value of n and a long enough t-interval so that the transient part of the solution has become negligible and the steady state is clearly shown. b. Estimate the amplitude and frequency of the steady-state part of the solution. c. Compare the results of part b with those from Section 3.8 for a sinusoidally forced oscillator.

where g( t) = u 3/2 ( t) − u 5/2 ( t) and k > 0 is a parameter. a. Sketch the graph of g( t) . Observe that it is a pulse of unit magnitude extending over one time unit. b. Solve the initial value problem. G c. Plot the solution for k = 1/2, k = 1, and k = 2. Describe the principal features of the solution and how they depend on k. N d. Find, to two decimal places, the smallest value of k for which the solution u( t) reaches the value 2. N e. Suppose k = 2. Find the time τ after which |u( t) | < 0.1 for all t > τ .

y( 0) = 0, y ( 0) = 0,

where f ( t) is the same as in Problem 13.

269

y( 0) = 0, y ( 0) = 0,

where n

g( t) = u 0 ( t) +

( −1) k u kπ ( t) .

k=1

a. Draw the graph of g( t) on an interval such as 0 ≤ t ≤ 6π . Compare the graph with that of f ( t) in Problem 13a.

b. Find the solution of the initial value problem. G c. Let n = 15. Plot the graph of the solution for 0 ≤ t ≤ 60. Describe the solution and explain why it behaves as it does. Compare it with the solution of Problem 13. d. Investigate how the solution changes as n increases. What happens as n → ∞?

16. Consider the initial value problem y + 0.1y + y = g( t) ,

y( 0) = 0, y ( 0) = 0,

where g( t) is the same as in Problem 15. G a. Plot the graph of the solution. Use a large enough value of n and a long enough t-interval so that the transient part of the solution has become negligible and the steady state is clearly shown. N b. Estimate the amplitude and frequency of the steady-state part of the solution. c. Compare the results of part b with those from Problem 15 and from Section 3.8 for a sinusoidally forced oscillator.

17. Consider the initial value problem y + y = h( t) ,

y( 0) = 0, y ( 0) = 0,

where h( t) = u 0 ( t) + 2

n

( −1) k u 11k/4 ( t) .

k=1

Observe that this problem is identical to Problem 15, except that the frequency of the forcing term has been increased somewhat. a. Find the solution of this initial value problem. G b. Let n ≥ 33 and plot the solution for 0 ≤ t ≤ 90 or longer. Your plot should show a clearly recognizable beat. N c. From the graph in part b, estimate the “slow period” and the “fast period” for this oscillator. d. For a sinusoidally forced oscillator, it was shown in Section 1 3.8 that the “slow frequency” is given by |ω − ω 0 |, where 2 ω 0 is the natural frequency of the system and ω is the forcing 1 frequency. Similarly, the “fast frequency” is ( ω + ω 0 ) . Use 2 these expressions to calculate the “fast period” and the “slow period” for the oscillator in this problem. How well do the results compare with your estimates from part c?

Boyce 9131 Ch06 2

270

September 29, 2016

17:34

270

CHAPTER 6 The Laplace Transform

Impulse Functions

6.5

In some applications it is necessary to deal with phenomena of an impulsive nature---for example, voltages or forces of large magnitude that act over very short time intervals. Such problems often lead to differential equations of the form ay + by + cy = g( t) ,

(1)

where g( t) is large during a short interval t0 − τ < t < t0 + τ for some τ > 0, and is otherwise zero. The integral I ( τ ) , defined by t0 +τ I (τ ) = g( t) dt, (2) t0 −τ

or, since g( t) = 0 outside of the interval ( t0 − τ , t0 + τ ) , by ∞ I (τ ) = g( t) dt,

(3)

−∞

is a measure of the strength of the forcing function. In a mechanical system, where g( t) is a force, I ( τ ) is the total impulse of the force g( t) over the time interval ( t0 − τ , t0 + τ ) . Similarly, if y is the current in an electric circuit and g( t) is the time derivative of the voltage, then I ( τ ) represents the total voltage impressed on the circuit during the interval ( t0 − τ , t0 + τ ) . In particular, let us suppose that t0 is zero and that g( t) is given by ⎧ ⎨ 1 , −τ < t < τ , g( t) = dτ ( t) = 2τ (4) ⎩ 0, t ≤ −τ or t ≥ τ , where τ is a small positive constant (see Figure 6.5.1). According to equation (2) or (3), it follows immediately that in this case, I ( τ ) = 1 independent of the value of τ , as long as τ = 0. Now let us idealize the forcing function dτ by prescribing it to act over shorter and shorter time intervals; that is, we consider the functions dτ ( t) as τ → 0+ (see Figure 6.5.2). As a result of this limiting operation, we obtain lim dτ ( t) = 0,

τ →0+

t= 0.

(5)

y τ =1

τ = 1/2

y 1 2τ

τ = 1/4 τ = 1/8

τ

τ

t

–8

–4

–2 –1 0 1 2

4

τ = 1/16

t

8 +

FIGURE 6.5.1 Graph of y = dτ ( t) .

FIGURE 6.5.2 Graphs of y = dτ ( t) as τ → 0 .

Further, since I ( τ ) = 1 for each τ = 0, it follows that lim I ( τ ) = 1.

τ →0+

(6)

Boyce 9131 Ch06 2

September 29, 2016

17:34

271

6.5 Impulse Functions

Equations (5) and (6) are used to define an idealized unit impulse function δ , which imparts an impulse of magnitude one at t = 0 but is zero for all values of t other than zero. That is, the “function” δ is defined to have the properties δ ( t) = 0, t = 0; ∞ δ ( t) dt = 1.

(7) (8)

−∞

There is no ordinary function of the kind studied in elementary calculus that satisfies both equations (7) and (8). The “function” δ , defined by those equations, is an example of what are known as generalized functions; it is usually called the Dirac4 delta function. Since δ ( t) corresponds to a unit impulse at t = 0, a unit impulse at an arbitrary point t = t0 is given by δ ( t − t0 ) . From equations (7) and (8), it follows that δ ( t − t0 ) = 0, t = t0 ; ∞ δ ( t − t0 ) dt = 1.

(9) (10)

−∞

The Dirac delta function does not satisfy the conditions of Theorem 6.1.2, but its Laplace transform can nevertheless be formally defined. Since δ ( t) is defined as the limit of dτ ( t) as τ → 0+ , it is natural to define the Laplace transform of δ as a similar limit of the transform of dτ . In particular, we will assume that t0 > 0 and will define L{δ ( t − t0 ) } by the equation L{δ ( t − t0 ) } = lim L{dτ ( t − t0 ) }.

(11)

τ →0+

To evaluate the limit in equation (11), we first observe that if τ < t0 , which must eventually be the case as τ → 0+ , then t0 − τ > 0. Since dτ ( t − t0 ) is nonzero only in the interval from t0 − τ to t0 + τ , we have ∞ L{dτ ( t − t0 ) } = e−st dτ ( t − t0 ) dt 0

=

t0 +τ

t0 −τ

e−st dτ ( t − t0 ) dt.

Substituting for dτ ( t − t0 ) from equation (4), we obtain t0 +τ 1 1 −st t=t0 +τ −st L{dτ ( t − t0 ) } = e dt = − e 2τ t0 −τ 2sτ t=t0 −τ =

1 −st0 sτ e ( e − e−sτ ) 2sτ

or sinh( sτ ) −st0 (12) e . sτ The quotient sinh( sτ ) /( sτ ) is indeterminate as τ → 0+ , but its limit can be evaluated by l’Hôpital’s5 rule. We obtain L{dτ ( t − t0 ) } =

lim τ →0+

sinh( sτ ) s cosh( sτ ) = lim = 1. sτ s τ →0+

......................................................................................................................................................................... 4 Paul

A. M. Dirac (1902--1984), English mathematical physicist, received his Ph.D. from Cambridge in 1926 and was professor of mathematics there until 1969. He was awarded the Nobel Prize for Physics in 1933 (with Erwin Schrödinger) for fundamental work in quantum mechanics. His most celebrated result was the relativistic equation for the electron, published in 1928. From this equation he predicted the existence of an “anti-electron,” or positron, which was first observed in 1932. Following his retirement from Cambridge, Dirac moved to the United States and held a research professorship at Florida State University.

5 Marquis Guillaume de l’Hôpital (1661--1704) was a French nobleman with deep interest in mathematics. For a time he employed Johann Bernoulli as his private tutor in calculus. L’Hôpital published the first textbook on differential calculus in 1696; in it appears the property of limits that is named for him.

271

Boyce 9131 Ch06 2

272

September 29, 2016

17:34

272

CHAPTER 6 The Laplace Transform

Then from equation (11) it follows that L{δ ( t − t0 ) } = e−st0 .

(13)

Equation (13) defines L{δ ( t − t0 ) } for any t0 > 0. We extend this result, to allow t0 to be zero, by letting t0 → 0+ on the right-hand side of equation (13); thus L{δ ( t) } = lim e−st0 = 1. (14) t0 →0+

It is reassuring to see that the Laplace transform formulas derived in equations (13) and (14) are consistent with the Laplace transform of a horizontally shifted function: L{δ ( t − t0 ) } = e−st0 L{δ ( t) } = e−st0 . In a similar way, it is possible to define the integral of the product of the delta function and any continuous function f . We have ∞ ∞ δ ( t − t0 ) f ( t) dt = lim dτ ( t − t0 ) f ( t) dt. (15) −∞

τ →0+

−∞

Using the definition (4) of dτ ( t) and the mean value theorem for integrals, we find that ∞ t0 +τ 1 dτ ( t − t0 ) f ( t) dt = f ( t) dt 2τ t0 −τ −∞ =

1 · 2τ · f ( t ∗ ) = f ( t ∗ ) , 2τ

where t0 − τ < t ∗ < t0 + τ . Hence t ∗ → t0 as τ → 0+ , and it follows from equation (15) that ∞ δ ( t − t0 ) f ( t) dt = f ( t0 ) . (16) −∞

The following example illustrates the use of the delta function in solving an initial value problem with an impulsive forcing function.

EXAMPLE 1 Find the solution of the initial value problem 2y + y + 2y = δ ( t − 5) ,

(17)

y ( 0) = 0.

y( 0) = 0,

(18)

Solution: This initial value problem arises from the study of the same electric circuit or mechanical oscillator as in Example 1 of Section 6.4. The only difference is in the forcing term. To solve the given problem, we first take the Laplace transform of the differential equation and use the initial conditions, obtaining ( 2s 2 + s + 2) Y ( s) = e−5s . Thus Y ( s) =

2s 2

e−5s e−5s = 2 +s+2

1 s+

1 4

2

. +

(19)

15 16

By Theorem 6.3.2, or from line 9 of Table 6.2.1, −1

L

▼

⎧ ⎪ ⎨

⎫ ⎪ ⎬

1

2 ⎪ ⎩ s+1 + 4

15 16

⎪ ⎭

4

= e 15

−t/4

sin

15 t 4

(20)

Boyce 9131 Ch06 2

September 29, 2016

17:34

273

6.5 Impulse Functions

273

▼ Hence, by Theorem 6.3.1, we have 2

−1

y( t) = L {Y ( s) } =

15

u 5 ( t) e

−( t−5) /4

sin

15 ( t − 5) 4

(21)

which is the formal solution of the given problem. It is also possible to write y( t) in the form

y=

⎧ 0, ⎪ ⎨

2

⎪ ⎩

15

15 ( t − 5) , 4

e−( t−5) /4 sin

t < 5, (22)

t ≥ 5.

The graph of equation (22) is shown in Figure 6.5.3. Since the initial conditions at t = 0 are homogeneous and there is no external excitation until t = 5, there is no response in the interval 0 < t < 5. The impulse at t = 5 produces a decaying oscillation that persists indefinitely. The response is continuous at t = 5 despite the singularity in the forcing function at that point. However, the first derivative of the solution has a jump discontinuity at t = 5, and the second derivative has an infinite discontinuity there. This is required by the differential equation (17), since a singularity on one side of the equation must be balanced by a corresponding singularity on the other side. y

0.3

0.2

0.1

5

10

15

20

t

–0.1

FIGURE 6.5.3 Solution of the initial value problem (17), (18): 2y + y + 2y = δ ( t − 5) , y( 0) = 0, y ( 0) = 0.

In dealing with problems that involve impulsive forcing, the use of the delta function usually simplifies the mathematical calculations, often quite significantly. However, if the actual excitation extends over a short, but nonzero, time interval, then an error will be introduced by modeling the excitation as taking place instantaneously. This error may be negligible, but in a practical problem it should not be dismissed without consideration. In Problem 12 you are asked to investigate this issue for a simple harmonic oscillator.

Problems In each of Problems 1 through 8: a. Find the solution of the given initial value problem. G b. Plot a graph of the solution.

1. y + 2y + 2y = δ ( t − π ) ; y( 0) = 1, y ( 0) = 0 2. y + 4y = δ ( t − π ) − δ ( t − 2π ) ; y( 0) = 0, y ( 0) = 0 3. y + 3y + 2y = δ ( t − 5) + u 10 ( t) ; y( 0) = 0, y ( 0) = 1/2

4. 5. 6. 7. 8.

y + 2y + 3y = sin t + δ ( t − 3π ) ; y + y = δ ( t − 2π ) cos t;

y + 4y = 2δ ( t − π/4) ;

y( 0) = 0, y ( 0) = 0

y( 0) = 0, y ( 0) = 1 y( 0) = 0, y ( 0) = 0

y + 2y + 2y = cos t + δ ( t − π/2) ;

y − y = δ ( t − 1) ; y ( 0) = 0, y ( 0) = 0 ( 4)

y( 0) = 0, y ( 0) = 0

y( 0) = 0, y ( 0) = 0,

Boyce 9131 Ch06 2

274

September 29, 2016

17:34

274

CHAPTER 6 The Laplace Transform

9. Consider again the system in Example 1 of this section, in which an oscillation is excited by a unit impulse at t = 5. Suppose that it is desired to bring the system to rest again after exactly one cycle---that is, when the response first returns to equilibrium moving in the positive direction. N a. Determine the impulse kδ ( t − t0 ) that should be applied to the system in order to accomplish this objective. Note that k is the magnitude of the impulse and t0 is the time of its application. G b. Solve the resulting initial value problem, and plot its solution to confirm that it behaves in the specified manner. N

10. Consider the initial value problem y + γ y + y = δ ( t − 1) ,

For each of the following choices for f ( t) : a. Try to predict the nature of the solution without solving the problem. G b. Test your prediction by finding the solution and plotting its graph. c. Determine what happens after the sequence of impulses ends.

13.

f ( t) =

20

δ ( t − kπ )

k=1

y( 0) = 0, y ( 0) = 0,

where γ is the damping coefficient (or resistance). G a. Let γ = 1 . Find the solution of the initial value problem 2 and plot its graph. b. Find the time t1 at which the solution attains its maximum value. Also find the maximum value y1 of the solution. G c. Let γ = 1 and repeat parts a and b. 4 d. Determine how t1 and y1 vary as γ decreases. What are the values of t1 and y1 when γ = 0?

11. Consider the initial value problem y + γ y + y = kδ ( t − 1) ,

Problems 13 through 16 deal with the effect of a sequence of impulses on an undamped oscillator. Suppose that y + y = f ( t) , y( 0) = 0, y ( 0) = 0.

y( 0) = 0, y ( 0) = 0,

where k is the magnitude of an impulse at t = 1, and γ is the damping coefficient (or resistance). 1 G a. Let γ = . Find the value of k for which the response has 2 a peak value of 2; call this value k1 . 1 G b. Repeat part (a) for γ = . 4 c. Determine how k1 varies as γ decreases. What is the value of k1 when γ = 0?

14.

f ( t) =

20

( −1) k+1 δ ( t − kπ )

k=1

15.

f ( t) =

15

δ ( t − ( 2k − 1) π )

k=1

16.

f ( t) =

40

k=1

y + y = f k ( t) ,

y( 0) = 0, y ( 0) = 0,

1 u 4−k ( t) − u 4+k ( t) with 0 < k ≤ 1. 2k a. Find the solution y = φ ( t, k) of the initial value problem. b. Calculate lim φ ( t, k) from the solution found in part a.

where f k ( t) =

k→0+

c. Observe that lim f k ( t) = δ ( t − 4) . Find the solution φ 0 ( t)

k→0 +

1 1 G d. Plot φ t, , φ t, , and φ 0 ( t) on the same axes. 2 4 Describe the relation between φ ( t, k) and φ 0 ( t) .

11 k 4

initial value problem y + 0.1y + y =

20

( −1) k+1 δ ( t − kπ ) ,

y( 0) = 0, y ( 0) = 0.

k=1

Observe that, except for the damping term, this problem is the same as Problem 14. a. Try to predict the nature of the solution without solving the problem. G b. Test your prediction by finding the solution and drawing its graph. c. Determine what happens after the sequence of impulses ends. G

18. Proceed as in Problem 17 for the oscillator satisfying

y + 0.1y + y =

15

δ ( t − ( 2k − 1) π ) ,

y( 0) = 0, y ( 0) = 0.

k=1

Observe that, except for the damping term, this problem is the same as Problem 15.

19. a. By the method of variation of parameters, show that the solution of the initial value problem y + 2y + 2y = f ( t) ;

k→0+

of the given initial value problem with f k ( t) replaced by δ ( t −4) . Is it true that φ 0 ( t) = lim φ ( t, k) ?

t−

17. The position of a certain lightly damped oscillator satisfies the

12. Consider the initial value problem

( −1) k+1 δ

is

y=

t

y( 0) = 0, y ( 0) = 0

e−( t−τ ) f ( τ ) sin( t − τ ) dτ .

0

b. Show that if f ( t) = δ ( t − π ) , then the solution of part a reduces to y = u π ( t) e−( t−π ) sin( t − π ) .

c. Use a Laplace transform to solve the given initial value problem with f ( t) = δ ( t − π ) , and confirm that the solution agrees with the result of part b.

Boyce 9131 Ch06 2

September 29, 2016

17:34

275

6.6 The Convolution Integral

6.6

The Convolution Integral

Sometimes it is possible to identify a Laplace transform H ( s) as the product of two other Laplace transforms F( s) and G( s) , the latter transforms corresponding to known functions f and g, respectively. In this event, we might anticipate that H ( s) would be the transform of the product of f and g. However, this is not the case; in other words, the Laplace transform cannot be commuted with ordinary multiplication. On the other hand, if an appropriately defined “generalized product” is introduced, then the situation changes, as stated in the following theorem.

Theorem 6.6.1 | Convolution Theorem If F( s) = L{ f ( t) } and G( s) = L{g( t) } both exist for s > a ≥ 0, then H ( s) = F( s) G( s) = L{h( t) },

s > a,

(1)

where

h( t) =

t

f ( t − τ ) g( τ ) dτ = 0

t

f ( τ ) g( t − τ ) dτ .

(2)

0

The function h is known as the convolution of f and g; the integrals in equation (2) are called convolution integrals.

The equality of the two integrals in equation (2) follows by making the change of variable t − τ = ξ in the first integral. Before giving the proof of this theorem, let us make some observations about the convolution integral. According to this theorem, the transform of the convolution of two functions, rather than the transform of their ordinary product, is given by the product of the separate transforms. It is conventional to emphasize that the convolution integral can be thought of as a “generalized product” by writing h( t) = ( f ∗ g) ( t) .

(3)

In particular, the notation ( f ∗ g) ( t) serves to indicate the first integral appearing in equation (2); the second integral in equation (2) is denoted as ( g ∗ f ) ( t) . The convolution f ∗ g has many of the properties of ordinary multiplication. For example, it is relatively simple to show that f ∗g = g∗ f

(commutative law)

f ∗ ( g1 + g2 ) = f ∗ g1 + f ∗ g2 (distributive law) ( f ∗ g) ∗ h = f ∗ ( g ∗ h) f ∗ 0 = 0 ∗ f = 0.

(4) (5)

(associative law)

(6)

(zero property)

(7)

In equation (7) the zeros denote not the number 0 but the function that has the value 0 for each value of t. The proofs of these properties are left to you as exercises. However, there are other properties of ordinary multiplication that the convolution integral does not have. For example, it is not true in general that f ∗ 1 is equal to f . To see this, note that t t ( f ∗ 1) ( t) = f ( t − τ ) · 1 dτ = f ( t − τ ) dτ . 0

0

If, for example, f ( t) = cos t, then τ =t t ( f ∗ 1) ( t) = cos( t − τ ) dτ = − sin( t − τ ) 0

τ =0

= − sin 0 + sin t = sin t. Clearly, ( f ∗ 1) ( t) = f ( t) in this case. Similarly, it may not be true that f ∗ f is nonnegative. See Problem 3 for an example.

275

Boyce 9131 Ch06 2

276

September 29, 2016

17:34

276

CHAPTER 6 The Laplace Transform

Convolution integrals arise in various applications in which the behavior of the system at time t depends not only on its state at time t but also on its past history. Systems of this kind are sometimes called hereditary systems and occur in such diverse fields as neutron transport, viscoelasticity, and population dynamics, among others. Turning now to the proof of Theorem 6.6.1, we note first that if ∞ ∞ F( s) = e−sξ f ( ξ ) dξ and G( s) = e−sτ g( τ ) dτ , 0

0

then

F( s) G( s) =

∞

e−sξ f ( ξ ) dξ

0

∞

e−sτ g( τ ) dτ .

(8)

0

Since the integrand of the first integral does not depend on the integration variable of the second, we can write F( s) G( s) as an iterated integral ∞ ∞ −sτ −sξ F( s) G( s) = e g( τ ) e f ( ξ ) dξ dτ 0 0 ∞ ∞ −s( ξ +τ ) = g( τ ) e f ( ξ ) dξ dτ . (9) 0

0

The latter integral can be put into a more convenient form by introducing a change of variable. Let ξ = t − τ , for fixed τ , so that dξ = dt. Further, ξ = 0 corresponds to t = τ , and ξ = ∞ corresponds to t = ∞; then the integral with respect to ξ in equation (9) is transformed into one with respect to t: ∞ ∞ −st F( s) G( s) = g( τ ) e f ( t − τ ) dt dτ . (10) τ

0

The iterated integral on the right-hand side of equation (10) is carried out over the shaded triangular region extending to infinity in the tτ -plane shown in Figure 6.6.1. Assuming that the order of integration can be reversed, we rewrite equation (10) so that the integration with respect to τ is executed first. In this way we obtain t ∞ F( s) G( s) = e−st f ( t − τ ) g( τ ) dτ dt (11) 0

0

or

F( s) G( s) =

∞

e−st h( t) dt = L{h( t) },

(12)

0

where h( t) is defined by equation (2). This completes the proof of Theorem 6.6.1. τ

τ=t

t =τ

t→

τ=0

t

FIGURE 6.6.1 Region of integration in F( s) G( s) .

EXAMPLE 1 Find the inverse Laplace transform of H ( s) =

▼

a . s2( s2 + a2)

(13)

Boyce 9131 Ch06 2

September 29, 2016

17:34

277

6.6 The Convolution Integral

▼ Solution: It is convenient to think of H ( s) as the product of s −2 and a/( s 2 + a 2 ) , which, according to lines 3 and 5 of Table 6.2.1, are the transforms of t and sin( at) , respectively. Hence, by Theorem 6.6.1, the inverse Laplace transform of H ( s) is

t

h( t) =

( t − τ ) sin( aτ ) dτ = 0

at − sin( at) . a2

(14)

You can verify that the same result is obtained if h( t) is written in the alternative form

t

τ sin( a( t − τ ) ) dτ ,

h( t) = 0

which confirms equation (2) in this case. Of course, h( t) can also be found by expanding H ( s) in partial fractions.

EXAMPLE 2 Find the solution of the initial value problem y + 4y = g( t) , y( 0) = 3,

y ( 0) = −1.

(15) (16)

Solution: By taking the Laplace transform of the differential equation and using the initial conditions, we obtain s 2 Y ( s) − 3s + 1 + 4Y ( s) = G( s) or Y ( s) =

G( s) 3s − 1 + 2 . s2 + 4 s +4

(17)

Observe that the first and second terms on the right-hand side of equation (17) contain the dependence of Y ( s) on the initial conditions and forcing function, respectively. It is convenient to write Y ( s) in the form Y ( s) = 3

s2

s 1 2 1 2 − + G( s) . 2 + 4 2 s + 4 2 s2 + 4

(18)

Then, using lines 5 and 6 of Table 6.2.1 and Theorem 6.6.1, we obtain y = 3 cos( 2t) −

1 1 sin( 2t) + 2 2

t

sin( 2( t − τ ) ) g( τ ) dτ .

(19)

0

If a specific forcing function g is given, then the integral in equation (19) can be evaluated (by numerical means, if necessary).

Example 2 illustrates the power of the convolution integral as a tool for writing the solution of an initial value problem in terms of an integral. In fact, it is possible to proceed in much the same way in more general problems. Consider the problem consisting of the differential equation ay + by + cy = g( t) ,

(20)

where a, b, and c are real constants and g is a given function, together with the initial conditions y( 0) = y0 ,

y ( 0) = y0 .

(21)

The Laplace transform approach yields some important insights concerning the structure of the solution of any problem of this type.

277

Boyce 9131 Ch06 2

278

September 29, 2016

17:34

278

CHAPTER 6 The Laplace Transform

The initial value problem (15), (16) is often referred to as an input-output problem. The coefficients a, b, and c describe the properties of some physical system, and g( t) is the input to the system. The values y0 and y0 describe the initial state, and the solution y is the output at time t. Taking the Laplace transform of equation (20) and using initial conditions (21), we obtain ( as 2 + bs + c) Y ( s) − ( as + b) y0 − ay0 = G( s) . If we let Φ ( s) =

( as + b) y0 + ay0 as 2 + bs + c

and Ψ ( s) =

G( s) , as 2 + bs + c

(22)

then we can write Y ( s) = Φ ( s) + Ψ ( s) .

(23)

y( t) = φ ( t) + ψ ( t) ,

(24)

Consequently, where φ ( t) = L−1 {Φ ( s) } and ψ ( t) = L−1 {Ψ ( s) }. Observe that φ ( t) is the solution of the initial value problem ay + by + cy = 0,

y( 0) = y0 , y ( 0) = y0 ,

(25)

obtained from equations (20) and (21) by setting g( t) equal to zero. Similarly, ψ ( t) is the solution of ay + by + cy = g( t) ,

y( 0) = 0, y ( 0) = 0,

(26)

in which the initial values y0 and y0 are each replaced by zero. Once specific values of a, b, and c are given, we can use Table 6.2.1 to find φ ( t) = L−1 {Φ ( s) }, possibly in conjunction with a translation or a partial fraction expansion. To find ψ ( t) = L−1 {Ψ ( s) }, it is convenient to write Ψ ( s) as Ψ ( s) = H ( s) G( s) ,

(27)

where H ( s) = ( as 2 + bs + c) −1 . The function H is known as the transfer function6 and depends only on the properties of the system under consideration; that is, H ( s) is determined entirely by the coefficients a, b, and c. On the other hand, G( s) depends only on the external excitation g( t) that is applied to the system. By the Convolution Theorem (Theorem 6.6.1) we can write t −1 ψ ( t) = L {H ( s) G( s) } = h( t − τ ) g( τ ) dτ , (28) 0

where h( t) = L−1 {H ( s) }, and g( t) is the given forcing function. To obtain a better understanding of the significance of h( t) , we consider the case in which G( s) = 1; consequently, g( t) = δ ( t) and Ψ ( s) = H ( s) . This means that y = h( t) is the solution of the initial value problem ay + by + cy = δ ( t) ,

y( 0) = 0, y ( 0) = 0,

(29)

obtained from equation (26) by replacing g( t) by δ ( t) . Thus h( t) is the response of the system to a unit impulse applied at t = 0, and it is natural to call h( t) the impulse response of the system. Equation (28) then says that ψ ( t) is the convolution of the impulse response and the forcing function. Referring to Example 2, we note that the transfer function is H ( s) = 1/( s 2 + 4) and the impulse response is h( t) = 12 sin( 2t) . Also, the first two terms on the right-hand side of equation (19) constitute the function φ ( t) , the solution of the corresponding homogeneous equation that satisfies the given initial conditions. ......................................................................................................................................................................... 6 This terminology arises from the fact that H ( s) is the ratio of the transforms of the output and the input of the problem (20).

Boyce 9131 Ch06 2

September 29, 2016

17:34

279

6.6 The Convolution Integral

279

Problems 1. Prove the commutative, distributive, and associative properties of the convolution integral. a. f ∗ g = g ∗ f b. f ∗ ( g1 + g2 ) = f ∗ g1 + f ∗ g2 c. f ∗ ( g ∗ h) = ( f ∗ g) ∗ h 2. Find an example different from the one in the text showing that ( f ∗ 1) ( t) need not be equal to f ( t) .

3. Show, by means of the example f ( t) = sin t, that f ∗ f is not necessarily nonnegative. In each of Problems 4 through 6, find the Laplace transform of the given function.

4.

t

f ( t) =

( t − τ ) 2 cos( 2τ ) dτ

t

f ( t) =

k( t − ξ ) φ ( ξ ) dξ = f ( t) ,

φ ( t) + 0

in which f and k are known functions, and φ is to be determined. Since the unknown function φ appears under an integral sign, the given equation is called an integral equation; in particular, it belongs to a class of integral equations known as Volterra integral equations7 . Take the Laplace transform of the given integral equation and obtain an expression for L{φ ( t) } in terms of the transforms L{ f ( t) } and L{k( t) } of the given functions f and k. The inverse transform of L{φ ( t) } is the solution of the original integral equation.

17. Consider the Volterra integral equation (see Problem 16) t

0

5.

16. Consider the equation t

( t − ξ ) φ ( ξ ) dξ = sin( 2t) .

φ ( t) +

e−( t−τ ) sin τ dτ

(30)

0

0

6.

t

f ( t) =

a. Solve the integral equation (30) by using the Laplace transform. b. By differentiating equation (30) twice, show that φ ( t) satisfies the differential equation

sin( t − τ ) cos τ dτ 0

In each of Problems 7 through 9, find the inverse Laplace transform of the given function by using the convolution theorem. 1 7. F( s) = 4 2 s ( s + 1) s 8. F( s) = ( s + 1) ( s 2 + 4)

9. F( s) =

Show also that the initial conditions are φ ( 0) = 0,

1 ( s + 1) 2 ( s 2 + 4)

integers, show that

1

u m ( 1 − u) n du.

f ∗ g = t m+n+1 0

b. Use the convolution theorem to show that 1 u m ( 1 − u) n du = 0

m! n! . ( m + n + 1) !

c. Extend the result of part b to the case where m and n are positive numbers but not necessarily integers. In each of Problems 11 through 15, express the solution of the given initial value problem in terms of a convolution integral.

11. y + ω 2 y = g( t) ;

y( 0) = 0, y ( 0) = 1

12. 4y + 4y + 17y = g( t) ; 5 4

y( 0) = 0, y ( 0) = 0

13. y + y + y = 1 − u π ( t) ; 14. y + 3y + 2y = cos( α t) ; 15. y ( 4) + 5y + 4y = g( t) ;

y ( 0) = 0, y ( 0) = 0

φ ( 0) = 2.

c. Solve the initial value problem in part b, and verify that the solution is the same as the one in part a.

10. a. If f ( t) = t m and g( t) = t n , where m and n are positive

φ ( t) + φ ( t) = −4 sin( 2t) .

y( 0) = 1, y ( 0) = −1 y( 0) = 1, y ( 0) = 0

y( 0) = 1, y ( 0) = 0,

In each of Problems 18 and 19: a. Solve the given Volterra integral equation by using the Laplace transform. b. Convert the integral equation into an initial value problem, as in Problem 17b. c. Solve the initial value problem in part b, and verify that the solution is the same as the one in part a.

t

18. φ ( t) +

( t − ξ ) φ ( ξ ) dξ = 1 0

t

19. φ ( t) + 2

cos( t − ξ ) φ ( ξ ) dξ = e−t

0

There are also equations, known as integro-differential equations, in which both derivatives and integrals of the unknown function appear. In each of Problems 20 and 21: a. Solve the given integro-differential equation by using the Laplace transform. b. By differentiating the integro-differential equation a sufficient number of times, convert it into an initial value problem. c. Solve the initial value problem in part b, and verify that the solution is the same as the one in part a.

t

20. φ ( t) +

( t − ξ ) φ ( ξ ) dξ = t,

φ ( 0) = 0

0

21. φ ( t) −

1 2

t

( t − ξ ) 2 φ ( ξ ) dξ = −t,

φ ( 0) = 1

0

............................................................................................................................. 7 See

the footnote about Vito Volterra in Section 9.5.

Boyce 9131 Ch06 2

280

September 29, 2016

17:34

280

CHAPTER 6 The Laplace Transform

22. The Tautochrone. A problem of interest in the history of mathematics is that of finding the tautochrone8 ---the curve down which a particle will slide freely under gravity alone, reaching the bottom in the same time regardless of its starting point on the curve. This problem arose in the construction of a clock pendulum whose period is independent of the amplitude of its motion. The tautochrone was found by Christian Huygens (1629--1695) in 1673 by geometric methods, and later by Leibniz and Jakob Bernoulli using analytic arguments. Bernoulli’s solution (in 1690) was one of the first occasions on which a differential equation was explicitly solved. The geometric configuration is shown in Figure 6.6.2. The starting point P( a, b) is joined to the terminal point ( 0, 0) by the arc C. Arc length s is

measured from the origin, and f ( y) denotes the rate of change of s with respect to y:

ds = f ( y) = dy

1+

2 1/2

.

(31)

Then it follows from the principle of conservation of energy that the time T ( b) required for a particle to slide from P to the origin is 1

b

T ( b) =

2g

0

f ( y) √ dy. b−y

(32)

a. Assume that T ( b) = T0 , a constant, for each b. By taking the Laplace transform of equation (32) in this case, and using the convolution theorem, Theorem 6.6.1, show that

y

F( s) = P(a, b)

dx dy

2g T0 √ ; π s

(33)

then show that

C

2g T0 √ . π y Hint: See Problem 24 of Section 6.1. b. Combining equations (32) and (34), show that f ( y) =

s

dx = dy

x FIGURE 6.6.2 The tautochrone.

2α − y , y

(34)

(35)

2

where α = gT0 /π 2 .

c. Use the substitution y = 2α sin2 ( θ /2) to solve equation (35), and show that x = α ( θ + sin θ ) , .............................................................................................................................. 8 The

word “tautochrone” comes from the Greek words tauto, which means “same,” and chronos, which means “time.”

y = α ( 1 − cos θ ) .

(36)

Equations (36) can be identified as parametric equations of a cycloid. Thus the tautochrone is an arc of a cycloid.

References The books listed below contain additional information on the Laplace transform and its applications. Churchill, R. V., Operational Mathematics (3rd ed.) (New York: McGraw-Hill, 1971). Doetsch, G., Introduction to the Theory and Application of the Laplace Transform (trans. W. Nader) (New York: Springer, 1974). Kaplan, W., Operational Methods for Linear Systems (Reading, MA: Addison-Wesley, 1962). Kuhfittig, P. K. F., Introduction to the Laplace Transform (New York: Plenum, 1978). Miles, J. W., Integral Transforms in Applied Mathematics (Oxford: Cambridge University Press, 2008).

Rainville, E. D., The Laplace Transform: An Introduction (New York: Macmillan, 1963). Each of the books just mentioned contains a table of transforms. Extensive tables are also available. See, for example, Erdelyi, A. (ed.), Tables of Integral Transforms (Vol. 1) (New York: McGraw-Hill, 1954). Roberts, G. E., and Kaufman, H., Table of Laplace Transforms (Philadelphia: Saunders, 1966). A further discussion of generalized functions can be found in Lighthill, M. J., An Introduction to Fourier Analysis and Generalized Functions (Cambridge, UK: Cambridge University Press, 1958).

Boyce 9131 Ch07 2

September 29, 2016

17:36

281

CHAPTER 7 Systems of First-Order Linear Equations Many physical problems involve a number of separate but interconnected components. For example, the current and voltage in an electrical network, each mass in a mechanical system, each element (or compound) in a chemical system, or each species in a biological system have this character. In these and similar cases, the corresponding mathematical problem consists of a system of two or more differential equations, which can always be written as first-order differential equations. In this chapter we focus on systems of first-order linear differential equations and, in particular, differential equations having constant coefficients, utilizing some of the elementary aspects of linear algebra to unify the presentation. In many respects this chapter follows the same lines as the treatment of second-order linear differential equations in Chapter 3.

7.1

Introduction

Systems of simultaneous ordinary differential equations arise naturally in problems involving several dependent variables, each of which is a function of the same single independent variable. We will denote the independent variable by t and will let x1 , x2 , x3 , . . . represent dependent variables that are functions of t. Differentiation1 with respect to t will be denoted d x1 by, for example, or x1 . dt Let us begin by considering the spring--mass system in Figure 7.1.1. The two masses move on a frictionless surface under the influence of external forces F1 ( t) and F2 ( t) , and they are also constrained by the three springs whose constants are k1 , k2 , and k3 , respectively. We regard motion and displacement to the right as being positive. F1(t)

F2(t) k2

k1 m1

k3 m2

x1

x2

FIGURE 7.1.1 A two-mass, three-spring system.

Using arguments similar to those in Section 3.7, we find the following equations for the coordinates x1 and x2 of the two masses: d 2 x1 m 1 2 = k2 ( x2 − x1 ) − k1 x1 + F1 ( t) = −( k1 + k2 ) x1 + k2 x2 + F1 ( t) , dt (1) d 2 x2 m 2 2 = −k3 x2 − k2 ( x2 − x1 ) + F2 ( t) = k2 x1 − ( k2 + k3 ) x2 + F2 ( t) . dt See Problem 14 for a full derivation of the system of differential equations (1). Next, consider the parallel LRC circuit shown in Figure 7.1.2. Let V be the voltage drop across the capacitor and I the current through the inductor. Then, referring to Section 3.7 and ......................................................................................................................................................................... 1 In

some treatments you will see differentiation with respect to time represented with a dot over the function, as in d x1 d 2 x1 . We reserve this notation for a specific purpose, which will be introduced in Section 9.6. and x˙˙1 = x˙1 = dt dt 2

281

Boyce 9131 Ch07 2

282

September 29, 2016

17:36

282

CHAPTER 7 Systems of First-Order Linear Equations

to Problem 16 of this section, we can show that the voltage and current are described by the system of equations dI V = , dt L (2) I V dV =− − , dt C RC where L is the inductance, C is the capacitance, and R is the resistance. C R L FIGURE 7.1.2 A parallel LRC circuit.

One reason why systems of first-order equations are particularly important is that equations of higher order can always be transformed into such systems. This is usually required if a numerical approach is planned, because, as we will see in Chapter 8, almost all codes for generating numerical approximations to solutions of differential equations are written for systems of first-order equations. The following example illustrates how easy it is to make the transformation from a second-order differential equation to a system of two first-order differential equations.

EXAMPLE 1 The motion of a certain spring--mass system (see Example 3 of Section 3.7) is described by the second-order differential equation 1 u + u + u = 0. (3) 8 Rewrite this equation as a system of first-order equations. Solution: Let x1 = u and x2 = u . Then it follows that x1 = x2 . Further, u = x2 . Then, by substituting for u, u , and u in equation (3), we obtain 1 x2 + x2 + x1 = 0. 8 Thus x1 and x2 satisfy the following system of two first-order differential equations: x1 = x2 , x2 = −x1 −

1 x2 . 8

(4)

The general equation of motion of a spring--mass system mu + γ u + ku = F( t)

(5)

can be transformed into a system of first-order differential equations in the same manner. If we let x1 = u and x2 = u , and proceed as in Example 1, we quickly obtain the system x1 = x2 , x2 = −

k γ 1 x1 − x2 + F( t) m m m

(6)

To transform an arbitrary n th order equation y ( n) = F( t, y, y , . . . , y ( n−1) )

(7)

Boyce 9131 Ch07 2

September 29, 2016

17:36

283

7.1 Introduction

into a system of n first-order differential equations, we extend the method of Example 1 by introducing the variables x1 , x2 , . . . , xn defined by x1 = y, x2 = y , x3 = y , . . . , xn = y ( n−1) .

(8)

It then follows immediately that x1 = x2 , x2 = x3 , . . . = xn , xn−1

(9)

and, from equation (7), xn = F( t, x1 , x2 , . . . , xn ) .

(10)

Equations (9) and (10) are a special case of the more general system x1 = F1 ( t, x1 , x2 , . . . , xn ) , x2 = F2 ( t, x1 , x2 , . . . , xn ) , . . . xn = Fn ( t, x1 , x2 , . . . , xn ) .

(11)

In a similar way, the system (1) can be reduced to a system of four first-order equations of the form (11), and the system (2) is already in this form. In fact, systems of the form (11) include almost all cases of interest. Much of the more advanced theory of differential equations is devoted to such systems. A solution of the system (11) on the interval I : α < t < β consists of n functions x1 = φ 1 ( t) , x2 = φ 2 ( t) , . . . , xn = φ n ( t)

(12)

where each function is differentiable at all points in interval I and the system of equations (11) is satisfied at all points in interval I . In addition to the given system of differential equations, there may also be given n initial conditions of the form 0

0

x1 ( t0 ) = x1 , x2 ( t0 ) = x2 , . . . , xn ( t0 ) = xn0 , 0

(13)

0

where t0 is a specified value of t in I , and x1 , . . . , xn are prescribed numbers. The differential equations (11) and the initial conditions (13) together form an initial value problem. A solution (12) can be viewed as a set of parametric equations in an n-dimensional space. For a given value of t, equations (12) give values for the coordinates x1 , . . . , xn of a point in the space. As t changes, the coordinates in general also change. The collection of points corresponding to α < t < β forms a curve in the space. It is often helpful to think of the curve as the trajectory, or path, of a particle moving in accordance with the system of differential equations (11). The initial conditions (13) determine the starting point of the moving particle. This curve is most easily visualized when n = 2 and the curve lies in the x1 x2 -plane. The following conditions on F1 , F2 , . . . , Fn , which are easily checked in specific problems, are sufficient to ensure that the initial value problem (11), (13) has a unique solution. Theorem 7.1.1 is analogous to Theorem 2.4.2, the existence and uniqueness theorem for a single first-order equation.

Theorem 7.1.1 ∂ F1 ∂ F1 , ... , , ∂ x1 ∂ xn

Let each of the n functions F1 , . . . , Fn and the n 2 first partial derivatives ... ,

∂ Fn ∂ Fn , ... , ∂ x1 ∂ x1

be continuous in a region R of t x1 x2 · · · xn -space defined by

α < t < β , α 1 < x1 < β 1 , . . . , α n < xn < β n , and let the point

0

0

0

t0 , x1 , x2 , . . . , xn

be

in R. Then there is an interval |t − t0 | < h in which there exists a unique solution x1 = φ 1 ( t) , . . . , xn = φ n ( t) of the system of differential equations (11) that also satisfies the initial conditions (13).

283

Boyce 9131 Ch07 2

284

September 29, 2016

17:36

284

CHAPTER 7 Systems of First-Order Linear