Differential Equations

DIFFERENTIAL EQUATIONS
Paul Dawkins
Differential Equations
Table of Contents
Preface ...................................................................................................................................... 3 Outline ..................................................................................................................................... iv Basic Concepts .......................................................................................................................... 1
Introduction ......................................................................................................................................... 1 Definitions ........................................................................................................................................... 2 Direction Fields.................................................................................................................................... 8 Final Thoughts ................................................................................................................................... 19
First Order Differential Equations .........................................................................................20

Introduction ....................................................................................................................................... 20 Linear Differential Equations.............................................................................................................. 21 Separable Differential Equations ........................................................................................................ 34 Exact Differential Equations ............................................................................................................... 45 Bernoulli Differential Equations ......................................................................................................... 56 Substitutions ...................................................................................................................................... 63 Intervals of Validity ........................................................................................................................... 72 Modeling with First Order Differential Equations ............................................................................... 77 Equilibrium Solutions......................................................................................................................... 90 Eulers Method .................................................................................................................................. 94
Second Order Differential Equations ................................................................................... 102
Introduction ...................................................................................................................................... 102 Basic Concepts.................................................................................................................................. 104 Real, Distinct Roots .......................................................................................................................... 109 Complex Roots ................................................................................................................................. 113 Repeated Roots ................................................................................................................................. 118 Reduction of Order............................................................................................................................ 122 Fundamental Sets of Solutions........................................................................................................... 126 More on the Wronskian ..................................................................................................................... 131 Nonhomogeneous Differential Equations ........................................................................................... 137 Undetermined Coefficients ................................................................................................................ 139 Variation of Parameters ..................................................................................................................... 156 Mechanical Vibrations ...................................................................................................................... 162 Introduction ...................................................................................................................................... 181 The Definition ................................................................................................................................... 183 Laplace Transforms........................................................................................................................... 187 Inverse Laplace Transforms............................................................................................................... 191 Step Functions .................................................................................................................................. 202 Solving IVPs with Laplace Transforms............................................................................................. 215 Nonconstant Coefficient IVPs .......................................................................................................... 222 IVPs With Step Functions ................................................................................................................ 226 Dirac Delta Function ......................................................................................................................... 233 Convolution Integrals ........................................................................................................................ 236
Laplace Transforms .............................................................................................................. 181
Systems of Differential Equations ......................................................................................... 241

Introduction ...................................................................................................................................... 241 Review : Systems of Equations .......................................................................................................... 243 Review : Matrices and Vectors .......................................................................................................... 249 Review : Eigenvalues and Eigenvectors ............................................................................................. 259 Systems of Differential Equations...................................................................................................... 269 Solutions to Systems ......................................................................................................................... 273 Phase Plane ....................................................................................................................................... 275 Real, Distinct Eigenvalues ................................................................................................................. 280 Complex Eigenvalues ........................................................................................................................ 290 Repeated Eigenvalues........................................................................................................................ 296 2007 Paul Dawkins i http://tutorial.math.lamar.edu/terms.aspx
Differential Equations Nonhomogeneous Systems ................................................................................................................ 303 Laplace Transforms........................................................................................................................... 307 Modeling .......................................................................................................................................... 309
Series Solutions to Differential Equations ............................................................................ 318
Introduction ...................................................................................................................................... 318 Review : Power Series....................................................................................................................... 319 Review : Taylor Series ...................................................................................................................... 327 Series Solutions to Differential Equations .......................................................................................... 330 Euler Equations ................................................................................................................................. 340 Introduction ...................................................................................................................................... 346 Basic Concepts for nth Order Linear Equations ................................................................................... 347 Linear Homogeneous Differential Equations...................................................................................... 350 Undetermined Coefficients ................................................................................................................ 355 Variation of Parameters ..................................................................................................................... 357 Laplace Transforms........................................................................................................................... 363 Systems of Differential Equations...................................................................................................... 365 Series Solutions................................................................................................................................. 370
Higher Order Differential Equations .................................................................................... 346
Boundary Value Problems & Fourier Series ........................................................................ 374

Introduction ...................................................................................................................................... 374 Boundary Value Problems ................................................................................................................ 375 Eigenvalues and Eigenfunctions ........................................................................................................ 381 Periodic Functions, Even/Odd Functions and Orthogonal Functions ................................................... 398 Fourier Sine Series ............................................................................................................................ 406 Fourier Cosine Series ........................................................................................................................ 417 Fourier Series .................................................................................................................................... 426 Convergence of Fourier Series ........................................................................................................... 434
Partial Differential Equations ............................................................................................... 440

Introduction ...................................................................................................................................... 440 The Heat Equation ............................................................................................................................ 442 The Wave Equation ........................................................................................................................... 449 Terminology ..................................................................................................................................... 451 Separation of Variables ..................................................................................................................... 454 Solving the Heat Equation ................................................................................................................. 465 Heat Equation with Non-Zero Temperature Boundaries ..................................................................... 478 Laplaces Equation ............................................................................................................................ 481 Vibrating String ................................................................................................................................ 492 Summary of Separation of Variables.................................................................................................. 495
2007 Paul Dawkins
ii
http://tutorial.math.lamar.edu/terms.aspx
Preface
Here are my online notes for my differential equations course that I teach here at Lamar University. Despite the fact that these are my class notes, they should be accessible to anyone wanting to learn how to solve differential equations or needing a refresher on differential equations. Ive tried to make these notes as self contained as possible and so all the information needed to read through them is either from a Calculus or Algebra class or contained in other sections of the notes. A couple of warnings to my students who may be here to get a copy of what happened on a day that you missed. 1. Because I wanted to make this a fairly complete set of notes for anyone wanting to learn differential equations I have included some material that I do not usually have time to cover in class and because this changes from semester to semester it is not noted here. You will need to find one of your fellow class mates to see if there is something in these notes that wasnt covered in class. 2. In general I try to work problems in class that are different from my notes. However, with Differential Equation many of the problems are difficult to make up on the spur of the moment and so in this class my class work will follow these notes fairly close as far as worked problems go. With that being said I will, on occasion, work problems off the top of my head when I can to provide more examples than just those in my notes. Also, I often dont have time in class to work all of the problems in the notes and so you will find that some sections contain problems that werent worked in class due to time restrictions. 3. Sometimes questions in class will lead down paths that are not covered here. I try to anticipate as many of the questions as possible in writing these up, but the reality is that I cant anticipate all the questions. Sometimes a very good question gets asked in class that leads to insights that Ive not included here. You should always talk to someone who was in class on the day you missed and compare these notes to their notes and see what the differences are. 4. This is somewhat related to the previous three items, but is important enough to merit its own item. THESE NOTES ARE NOT A SUBSTITUTE FOR ATTENDING CLASS!! Using these notes as a substitute for class is liable to get you in trouble. As already noted not everything in these notes is covered in class and often material or insights not in these notes is covered in class.
2007 Paul Dawkins
iii
Outline
Here is a listing and brief description of the material in this set of notes. Basic Concepts Definitions Some of the common definitions and concepts in a differential equations course Direction Fields An introduction to direction fields and what they can tell us about the solution to a differential equation. Final Thoughts A couple of final thoughts on what we will be looking at throughout this course. First Order Differential Equations Linear Equations Identifying and solving linear first order differential equations. Separable Equations Identifying and solving separable first order differential equations. Well also start looking at finding the interval of validity from the solution to a differential equation. Exact Equations Identifying and solving exact differential equations. Well do a few more interval of validity problems here as well. Bernoulli Differential Equations In this section well see how to solve the Bernoulli Differential Equation. This section will also introduce the idea of using a substitution to help us solve differential equations. Substitutions Well pick up where the last section left off and take a look at a couple of other substitutions that can be used to solve some differential equations that we couldnt otherwise solve. Intervals of Validity Here we will give an in-depth look at intervals of validity as well as an answer to the existence and uniqueness question for first order differential equations. Modeling with First Order Differential Equations Using first order differential equations to model physical situations. The section will show some very real applications of first order differential equations. Equilibrium Solutions We will look at the behavior of equilibrium solutions and autonomous differential equations. Eulers Method In this section well take a brief look at a method for approximating solutions to differential equations. Second Order Differential Equations Basic Concepts Some of the basic concepts and ideas that are involved in solving second order differential equations. Real Roots Solving differential equations whose characteristic equation has real roots. Complex Roots Solving differential equations whose characteristic equation complex real roots.
2007 Paul Dawkins
iv
Repeated Roots Solving differential equations whose characteristic equation has repeated roots. Reduction of Order A brief look at the topic of reduction of order. This will be one of the few times in this chapter that non-constant coefficient differential equation will be looked at. Fundamental Sets of Solutions A look at some of the theory behind the solution to second order differential equations, including looks at the Wronskian and fundamental sets of solutions. More on the Wronskian An application of the Wronskian and an alternate method for finding it. Nonhomogeneous Differential Equations A quick look into how to solve nonhomogeneous differential equations in general. Undetermined Coefficients The first method for solving nonhomogeneous differential equations that well be looking at in this section. Variation of Parameters Another method for solving nonhomogeneous differential equations. Mechanical Vibrations An application of second order differential equations. This section focuses on mechanical vibrations, yet a simple change of notation can move this into almost any other engineering field. Laplace Transforms The Definition The definition of the Laplace transform. We will also compute a couple Laplace transforms using the definition. Laplace Transforms As the previous section will demonstrate, computing Laplace transforms directly from the definition can be a fairly painful process. In this section we introduce the way we usually compute Laplace transforms. Inverse Laplace Transforms In this section we ask the opposite question. Heres a Laplace transform, what function did we originally have? Step Functions This is one of the more important functions in the use of Laplace transforms. With the introduction of this function the reason for doing Laplace transforms starts to become apparent. Solving IVPs with Laplace Transforms Heres how we used Laplace transforms to solve IVPs. Nonconstant Coefficient IVPs We will see how Laplace transforms can be used to solve some nonconstant coefficient IVPs IVPs with Step Functions Solving IVPs that contain step functions. This is the section where the reason for using Laplace transforms really becomes apparent. Dirac Delta Function One last function that often shows up in Laplace transform problems. Convolution Integral A brief introduction to the convolution integral and an application for Laplace transforms. Table of Laplace Transforms This is a small table of Laplace Transforms that well be using here. Systems of Differential Equations Review : Systems of Equations The traditional starting point for a linear algebra class. We will use linear algebra techniques to solve a system of equations. Review : Matrices and Vectors A brief introduction to matrices and vectors. We will look at arithmetic involving matrices and vectors, inverse of a matrix,
2007 Paul Dawkins v http://tutorial.math.lamar.edu/terms.aspx
determinant of a matrix, linearly independent vectors and systems of equations revisited. Review : Eigenvalues and Eigenvectors Finding the eigenvalues and eigenvectors of a matrix. This topic will be key to solving systems of differential equations. Systems of Differential Equations Here we will look at some of the basics of systems of differential equations. Solutions to Systems We will take a look at what is involved in solving a system of differential equations. Phase Plane A brief introduction to the phase plane and phase portraits. Real Eigenvalues Solving systems of differential equations with real eigenvalues. Complex Eigenvalues Solving systems of differential equations with complex eigenvalues. Repeated Eigenvalues Solving systems of differential equations with repeated eigenvalues. Nonhomogeneous Systems Solving nonhomogeneous systems of differential equations using undetermined coefficients and variation of parameters. Laplace Transforms A very brief look at how Laplace transforms can be used to solve a system of differential equations. Modeling In this section well take a quick look at some extensions of some of the modeling we did in previous sections that lead to systems of equations. Series Solutions Review : Power Series A brief review of some of the basics of power series. Review : Taylor Series A reminder on how to construct the Taylor series for a function. Series Solutions In this section we will construct a series solution for a differential equation about an ordinary point. Euler Equations We will look at solutions to Eulers differential equation in this section. Higher Order Differential Equations Basic Concepts for nth Order Linear Equations Well start the chapter off with a quick look at some of the basic ideas behind solving higher order linear differential equations. Linear Homogeneous Differential Equations In this section well take a look at extending the ideas behind solving 2nd order differential equations to higher order. Undetermined Coefficients Here well look at undetermined coefficients for higher order differential equations. Variation of Parameters Well look at variation of parameters for higher order differential equations in this section. Laplace Transforms In this section were just going to work an example of using Laplace transforms to solve a differential equation on a 3rd order differential equation just so say that we looked at one with order higher than 2nd. Systems of Differential Equations Here well take a quick look at extending the ideas we discussed when solving 2 x 2 systems of differential equations to systems of size 3 x 3.
2007 Paul Dawkins
vi
Series Solutions This section serves the same purpose as the Laplace Transform section. It is just here so we can say weve worked an example using series solutions for a differential equations of order higher than 2nd. Boundary Value Problems & Fourier Series Boundary Value Problems In this section well define the boundary value problems as well as work some basic examples. Eigenvalues and Eigenfunctions Here well take a look at the eigenvalues and eigenfunctions for boundary value problems. Periodic Functions and Orthogonal Functions Well take a look at periodic functions and orthogonal functions in section. Fourier Sine Series In this section well start looking at Fourier Series by looking at a special case : Fourier Sine Series. Fourier Cosine Series Well continue looking at Fourier Series by taking a look at another special case : Fourier Cosine Series. Fourier Series Here we will look at the full Fourier series. Convergence of Fourier Series Here well take a look at some ideas involved in the just what functions the Fourier series converge to as well as differentiation and integration of a Fourier series. Partial Differential Equations The Heat Equation We do a partial derivation of the heat equation in this section as well as a discussion of possible boundary values. The Wave Equation Here we do a partial derivation of the wave equation. Terminology In this section we take a quick look at some of the terminology used in the method of separation of variables. Separation of Variables We take a look at the first step in the method of separation of variables in this section. This first step is really the step motivates the whole process. Solving the Heat Equation In this section we go through the complete separation of variables process and along the way solve the heat equation with three different sets of boundary conditions. Heat Equation with Non-Zero Temperature Boundaries Here we take a quick look at solving the heat equation in which the boundary conditions are fixed, non-zero temperature conditions. Laplaces Equation We discuss solving Laplaces equation on both a rectangle and a disk in this section. Vibrating String Here we solve the wave equation for a vibrating string. Summary of Separation of Variables In this final section we give a quick summary of the method of separation of variables.
2007 Paul Dawkins
vii
Basic Concepts
Introduction
There isnt really a whole lot to this chapter it is mainly here so we can get some basic definitions and concepts out of the way. Most of the definitions and concepts introduced here can be introduced without any real knowledge of how to solve differential equations. Most of them are terms that well use throughout a class so getting them out of the way right at the beginning is a good idea. During an actual class I tend to hold off on a couple of the definitions and introduce them at a later point when we actually start solving differential equations. The reason for this is mostly a time issue. In this class time is usually at a premium and some of the definitions/concepts require a differential equation and/or its solution so I use the first couple differential equations that we will solve to introduce the definition or concept. Here is a quick list of the topics in this Chapter. Definitions Some of the common definitions and concepts in a differential equations course Direction Fields An introduction to direction fields and what they can tell us about the solution to a differential equation. Final Thoughts A couple of final thoughts on what we will be looking at throughout this course.
2007 Paul Dawkins
Definitions
Differential Equation The first definition that we should cover should be that of differential equation. A differential equation is any equation which contains derivatives, either ordinary derivatives or partial derivatives. There is one differential equation that everybody probably knows, that is Newtons Second Law of Motion. If an object of mass m is moving with acceleration a and being acted on with force F then Newtons Second Law tells us.
F = ma
(1)
To see that this is in fact a differential equation we need to rewrite it a little. First, remember that we can rewrite the acceleration, a, in one of two ways.
a=
dv dt
OR
a=
d 2u dt 2
(2)
Where v is the velocity of the object and u is the position function of the object at any time t. We should also remember at this point that the force, F may also be a function of time, velocity, and/or position. So, with all these things in mind Newtons Second Law can now be written as a differential equation in terms of either the velocity, v, or the position, u, of the object as follows.
dv = F (t, v ) dt d 2u du m 2 = F t , u, dt dt m
So, here is our first differential equation. We will see both forms of this in later chapters. Here are a few more examples of differential equations.
(3) (4)
ay + by + cy = g ( t )
(5) (6) (7) (8) (9) (10)
sin ( y )
d2y dy = (1 - y ) + y 2e -5 y 2 dx dx ( 4) y + 10 y - 4 y + 2 y = cos ( t )
a2
2u u = x 2 t a 2uxx = utt
3u u = 1+ 2 xt y
Order The order of a differential equation is the largest derivative present in the differential equation. In the differential equations listed above (3) is a first order differential equation, (4), (5), (6), (8),
2007 Paul Dawkins 2 http://tutorial.math.lamar.edu/terms.aspx
and (9) are second order differential equations, (10) is a third order differential equation and (7) is a fourth order differential equation. Note that the order does not depend on whether or not youve got ordinary or partial derivatives in the differential equation. We will be looking almost exclusively at first and second order differential equations in these notes. As you will see most of the solution techniques for second order differential equations can be easily (and naturally) extended to higher order differential equations and well discuss that idea later on. Ordinary and Partial Differential Equations A differential equation is called an ordinary differential equation, abbreviated by ode, if it has ordinary derivatives in it. Likewise, a differential equation is called a partial differential equation, abbreviated by pde, if it has differential derivatives in it. In the differential equations above (3) - (7) are odes and (8) - (10) are pdes. The vast majority of these notes will deal with odes. The only exception to this will be the last chapter in which well take a brief look at a common and basic solution technique for solving pdes. Linear Differential Equations A linear differential equation is any differential equation that can be written in the following form.
an ( t ) y ( n ) ( t ) + an -1 ( t ) y ( n -1) ( t ) + L + a1 ( t ) y ( t ) + a0 ( t ) y ( t ) = g ( t )
(11)
The important thing to note about linear differential equations is that there are no products of the function, y ( t ) , and its derivatives and neither the function or its derivatives occur to any power other than the first power. The coefficients a0 ( t ) , K , an ( t ) and g ( t ) can be zero or non-zero functions, constant or nonused in determining if a differential equation is linear. If a differential equation cannot be written in the form, (11) then it is called a non-linear differential equation. In (5) - (7) above only (6) is non-linear, all the other are linear differential equations. We cant classify (3) and (4) since we do not know what form the function F has. These could be either linear or non-linear depending on F. Solution A solution to a differential equation on an interval a < t < b is any function y ( t ) which satisfies the differential equation in question on the interval a < t < b . It is important to note that solutions are often accompanied by intervals and these intervals can impart some important information about the solution. Consider the following example.
constant functions, linear or non-linear functions. Only the function, y ( t ) , and its derivatives are
2007 Paul Dawkins
Example 1 Show that y ( x ) = x
3 2
is a solution to 4 x 2 y + 12 xy + 3 y = 0 for x > 0 .
Solution Well need the first and second derivative to do this.
3 -5 y ( x ) = - x 2 2
y ( x ) =
15 - 7 x 2 4
Plug these as well as the function into the differential equation.
15 - 7 3 -5 -3 4 x 2 x 2 + 12 x - x 2 + 3 x 2 = 0 4 2 15 x
So, y ( x ) = x
3 2 3 2
- 18 x
3 2
+ 3x
3 2
=0
0=0
does satisfy the differential equation and hence is a solution. Why then did I include the condition that x > 0 ? I did not use this condition anywhere in the work showing that the function would satisfy the differential equation. To see why recall that
3 2
y ( x) = x
1 x3
In this form it is clear that well need to avoid x = 0 at the least as this would give division by zero. Also, there is a general rule of thumb that were going to run with in this class. This rule of thumb is : Start with real numbers, end with real numbers. In other words, if our differential equation only contains real numbers then we dont want solutions that give complex numbers. So, in order to avoid complex numbers we will also need to avoid negative values of x. So, we saw in the last example that even though a function may symbolically satisfy a differential equation, because of certain restrictions brought about by the solution we cannot use all values of the independent variable and hence, must make a restriction on the independent variable. This will be the case with many solutions to differential equations. In the last example, note that there are in fact many more possible solutions to the differential equation given. For instance all of the following are also solutions
y ( x) = x
1 2 3 2
y ( x ) = -9 x y ( x) = 7x
-
1 2 3 2 1 2
y ( x ) = -9 x
+ 7x
2007 Paul Dawkins
Ill leave the details to you to check that these are in fact solutions. Given these examples can you come up with any other solutions to the differential equation? There are in fact an infinite number of solutions to this differential equation. So, given that there are an infinite number of solutions to the differential equation in the last example (provided you believe me when I say that anyway.) we can ask a natural question. Which is the solution that we want or does it matter which solution we use? This question leads us to the next definition in this section. Initial Condition(s) Initial Condition(s) are a condition, or set of conditions, on the solution that will allow us to determine which solution that we are after. Initial conditions (often abbreviated i.c.s when Im feeling lazy) are of the form,
y ( t0 ) = y0
and/or
y ( k ) ( t0 ) = y k
So, in other words, initial conditions are values of the solution and/or its derivative(s) at specific points. As we will see eventually, solutions to nice enough differential equations are unique and hence only one solution will meet the given conditions. The number of initial conditions that are required for a given differential equation will depend upon the order of the differential equation as we will see.
Example 2 y ( x ) = x y ( 4 ) = 3 . 64
3 2
is a solution to 4 x 2 y + 12 xy + 3 y = 0 , y ( 4 ) =
1 , and 8
Solution As we saw in previous example the function is a solution and we can then note that
y ( 4) = 4
3 2
( 4)
1 8 1 =1 8
3 -5 3 y ( 4) = - 4 2 = 2 2
( 4)
3 64
3 and y ( 4 ) = - 64 . In fact,
and so this solution also meets the initial conditions of y ( 4 ) =
y ( x) = x
3 2
is the only solution to this differential equation that satisfies these two initial
conditions. Initial Value Problem An Initial Value Problem (or IVP) is a differential equation along with an appropriate number of initial conditions.
Example 3 The following is an IVP. 4 x 2 y + 12 xy + 3 y = 0

2007 Paul Dawkins 5
1 3 y ( 4 ) = , y ( 4 ) = 8 64
Example 4 Heres another IVP. 2t y + 4 y = 3
y (1) = -4
As I noted earlier the number of initial condition required will depend on the order of the differential equation. Interval of Validity The interval of validity for an IVP with initial condition(s)
y ( t0 ) = y0
and/or
y ( k ) ( t0 ) = y k
is the largest possible interval on which the solution is valid and contains t0 . These are easy to define, but can be difficult to find, so Im going to put off saying anything more about these until we get into actually solving differential equations and need the interval of validity. General Solution The general solution to a differential equation is the most general form that the solution can take and doesnt take any initial conditions into account.
Example 5 y ( t ) =
3 c + is the general solution to 4 t2 2t y + 4 y = 3
Ill leave it to you to check that this function is in fact a solution to the given differential equation. In fact, all solutions to this differential equation will be in this form. This is one of the first differential equations that you will learn how to solve and you will be able to verify this shortly for yourself. Actual Solution The actual solution to a differential equation is the specific solution that not only satisfies the differential equation, but also satisfies the given initial condition(s).
Example 6 What is the actual solution to the following IVP? 2t y + 4 y = 3 y (1) = -4

Solution This is actually easier to do that it might at first appear. From the previous example we already know (well that is provided you believe my solution to this example) that all solutions to the differential equation are of the form.
y (t ) =
3 c + 4 t2 3 19 =4 4
All that we need to do is determine the value of c that will give us the solution that were after. To find this all we need do is use our initial condition as follows.
-4 = y (1) =
So, the actual solution to the IVP is.
3 c + 4 12
c = -4 -
y (t ) =
2007 Paul Dawkins 6
3 19 4 4t 2
From this last example we can see that once we have the general solution to a differential equation finding the actual solution is nothing more than applying the initial condition(s) and solving for the constant(s) that are in the general solution. Implicit/Explicit Solution In this case its easier to define an explicit solution, then tell you what an implicit solution isnt, and then give you an example to show you the difference. So, thats what Ill do. An explicit solution is any solution that is given in the form y = y ( t ) . In other words, the only place that y actually shows up is once on the left side and only raised to the first power. An implicit solution is any solution that isnt in explicit form. Note that it is possible to have either general implicit/explicit solutions and actual implicit/explicit solutions.
Example 7 y 2 = t 2 - 3 is the actual implicit solution to y =
t , y ( 2 ) = -1 y
At this point I will ask that you trust me that this is in fact a solution to the differential equation. You will learn how to get this solution in a later section. The point of this example is that since there is a y 2 on the left side instead of a single y ( t ) this is not an explicit solution!
Example 8 Find an actual explicit solution to y =
t , y ( 2 ) = -1 . y
Solution We already know from the previous example that an implicit solution to this IVP is y 2 = t 2 - 3 . To find the explicit solution all we need to do is solve for y ( t ) .
y (t ) = t 2 - 3
Now, weve got a problem here. There are two functions here and we only want one and in fact only one will be correct! We can determine the correct function by reapplying the initial condition. Only one of them will satisfy the initial condition. In this case we can see that the - solution will be the correct one. The actual explicit solution is then
y (t ) = - t 2 - 3
In this case we were able to find an explicit solution to the differential equation. It should be noted however that it will not always be possible do find an explicit solution. Also, note that in this case we were only able to get the explicit actual solution because we had the initial condition to help us determine which of the two functions would be the correct solution. Weve now gotten most of the basic definitions out of the way and so we can move onto other topics.
2007 Paul Dawkins
Direction Fields
This topic is given its own section for a couple of reasons. First, understanding direction fields and what they tell us about a differential equation and its solution is important and can be introduced without any knowledge of how to solve a differential equation and so can be done here before we get into solving them. So, having some information about the solution to a differential equation without actually having the solution is a nice idea that needs some investigation. Next, since we need a differential equation to work with this is a good section to show you that differential equations occur naturally in many cases and how we get them. Almost every physical situation that occurs in nature can be described with an appropriate differential equation. The differential equation may be easy or difficult to arrive at depending on the situation and the assumptions that are made about the situation and we may not ever be able to solve it, however it will exist. The process of describing a physical situation with a differential equation is called modeling. We will be looking at modeling several times throughout this class. One of the simplest physical situations to think of is a falling object. So lets consider a falling object with mass m and derive a differential equation that, when solved, will give us the velocity of the object at any time, t. We will assume that only gravity and air resistance will act upon the object as it falls. Below is a figure showing the forces that will act upon the object.
Before defining all the terms in this problem we need to set some conventions. We will assume that forces acting in the downward direction are positive forces while forces that act in the upward direction are negative. Likewise, we will assume that an object moving downward (i.e. a falling object) will have a positive velocity. Now, lets take a look at the forces shown in the diagram above. FG is the force due to gravity and is given by FG = mg where g is the acceleration due to gravity. In this class I use g = 9.8 m/s2 or g = 32 ft/s2 depending on whether we will use the metric or British system. FA is the force due to air resistance and for this example we will assume that it is proportional to the velocity, v, of the mass. Therefore the force due to air resistance is then given by FA = -g v , where g > 0 . Note that the is required to get the correct sign on the force. Both g and v are positive and the force is acting upward and hence must be negative. The will give us the correct sign and hence direction for this force. Recall from the previous section that Newtons Second Law of motion can be written as
dv = F (t, v ) dt
where F(t,v) is the sum of forces that act on the object and may be a function of the time t and the velocity of the object, v. For our situation we will have two forces acting on the object gravity,
FG = mg . acting in the downward direction and hence will be positive, and air resistance, FA = -g v , acting in the upward direction and hence will be negative. Putting all of this together
into Newtons Second Law gives the following.
dv = mg - g v dt dv gv =gdt m
To simplify the differential equation lets divide out the mass, m.
(1)
This then is a first order linear differential equation that, when solved, will give the velocity, v (in m/s), of a falling object of mass m that has both gravity and air resistance acting upon it. In order to look at direction fields (that is after all the topic of this section....) it would be helpful to have some numbers for the various quantities in the differential equation. So, lets assume that we have a mass of 2 kg and that g = 0.392. Plugging this into (1) gives the following differential equation.
dv = 9.8 - 0.196v dt
(2)
Let's take a geometric view of this differential equation. Let's suppose that for some time, t, the velocity just happens to be v = 30 m/s. Note that were not saying that the velocity ever will be 30 m/s. All that were saying is that lets suppose that by some chance the velocity does happen to be 30 m/s at some time t. So, if the velocity does happen to be 30 m/s at some time t we can plug v = 30 into (2) to get.
dv = 3.92 dt
Recall from your Calculus I course that a positive derivative means that the function in question, the velocity in this case, is increasing, so if the velocity of this object is ever 30m/s for any time t the velocity must be increasing at that time. Also, recall that the value of the derivative at a particular value of t gives the slope of the tangent line to the graph of the function at that time, t. So, if for some time t the velocity happens to be 30 m/s the slope of the tangent line to the graph of the velocity is 3.92. We could continue in this fashion and pick different values of v and compute the slope of the tangent line for those values of the velocity. However, let's take a slightly more organized approach to this. Let's first identify the values of the velocity that will have zero slope or horizontal tangent lines. These are easy enough to find. All we need to do is set the derivative equal to zero and solve for v. In the case of our example we will have only one value of the velocity which will have horizontal tangent lines, v = 50 m/s. What this means is that IF (again, theres that word if), for some time t, the velocity happens to be 50 m/s then the tangent line at that point will be horizontal. What the slope of the tangent line is at times before and after this point is not known yet and has no bearing on the slope at this particular time, t.
2007 Paul Dawkins
So, if we have v = 50, we know that the tangent lines will be horizontal. We denote this on an axis system with horizontal arrows pointing in the direction of increasing t at the level of v = 50 as shown in the following figure.
Now, let's get some tangent lines and hence arrows for our graph for some other values of v. At this point the only exact slope that is useful to us is where the slope horizontal. So instead of going after exact slopes for the rest of the graph we are only going to go after general trends in the slope. Is the slope increasing or decreasing? How fast is the slope increasing or decreasing? For this example those types of trends are very easy to get. First, notice that the right hand side of (2) is a polynomial and hence continuous. This means that it can only change sign if it first goes through zero. So, if the derivative will change signs (no guarantees that it will) it will do so at v = 50 and the only place that it may change sign is v = 50. This means that for v > 50 the slope of the tangent lines to the velocity will have the same sign. Likewise, for v < 50 the slopes will also have the same sign. The slopes in these ranges may have (and probably will) have different values, but we do know what their signs must be. Let's start by looking at v < 50. We saw earlier that if v = 30 the slope of the tangent line will be 3.92, or positive. Therefore, for all values of v < 50 we will have positive slopes for the tangent lines. Also, by looking at (2) we can see that as v approaches 50, always staying less than 50, the slopes of the tangent lines will approach zero and hence flatten out. If we move v away from 50, staying less than 50, the slopes of the tangent lines will become steeper. If you want to get an idea of just how steep the tangent lines become you can always pick specific values of v and compute values of the derivative. For instance, we know that at v = 30 the derivative is 3.92 and so arrows at this point should have a slope of around 4. Using this information we can now add in some arrows for the region below v = 50 as shown in the graph below.
2007 Paul Dawkins
10
Now, lets look at v > 50. The first thing to do is to find out if the slopes are positive or negative. We will do this the same way that we did in the last bit, i.e. pick a value of v, plug this into (2) and see if the derivative is positive or negative. Note, that you should NEVER assume that the derivative will change signs where the derivative is zero. It is easy enough to check so you should always do so. We need to check the derivative so let's use v = 60. Plugging this into (2) gives the slope of the tangent line as -1.96, or negative. Therefore, for all values of v > 50 we will have negative slopes for the tangent lines. As with v < 50, by looking at (2) we can see that as v approaches 50, always staying greater than 50, the slopes of the tangent lines will approach zero and flatten out. While moving v away from 50 again, staying greater than 50, the slopes of the tangent lines will become steeper. We can now add in some arrows for the region above v = 50 as shown in the graph below.
This graph above is called the direction field for the differential equation. So, just why do we care about direction fields? There are two nice pieces of information that can be readily found from the direction field for a differential equation.
2007 Paul Dawkins
11
1. Sketch of solutions. Since the arrows in the direction fields are in fact tangents to the actual solutions to the differential equations we can use these as guides to sketch the graphs of solutions to the differential equation. 2. Long Term Behavior. In many cases we are less interested in the actual solutions to the differential equations as we are in how the solutions behave as t increases. Direction fields, if we can get our hands on them, can be used to find information about this long term behavior of the solution. So, back to the direction field for our differential equation. Suppose that we want to know what the solution that has the value v(0) = 30 looks like. We can go to our direction field and start at 30 on the vertical axis. At this point we know that the solution is increasing and that as it increases the solution should flatten out because the velocity will be approaching the value of v = 50. So we start drawing an increasing solution and when we hit an arrow we just make sure that we stay parallel to that arrow. This gives us the figure below.
To get a better idea of how all the solutions are behaving, let's put a few more solutions in. Adding some more solutions gives the figure below. The set of solutions that we've graphed below is often called the family of solution curves or the set of integral curves. The number of solutions that is plotted when plotting the integral curves varies. You should graph enough solution curves to illustrate how solutions in all portions of the direction field are behaving.
2007 Paul Dawkins
12
Now, from either the direction field, or the direction field with the solution curves sketched in we can see the behavior of the solution as t increases. For our falling object, it looks like all of the solutions will approach v = 50 as t increases. We will often want to know if the behavior of the solution will depend on the value of v(0). In this case the behavior of the solution will not depend on the value of v(0), but that is probably more of the exception than the rule so dont expect that. Lets take a look at a more complicated example.
Example 1 Sketch the direction field for the following differential equation. Sketch the set of integral curves for this differential equation. Determine how the solutions behave as t and
if this behavior depends on the value of y(0) describe this dependency.
y = ( y 2 - y - 2 ) (1 - y )
Solution
First, do not worry about where this differential equation came from. To be honest, I just make it up. It may, or may not describe an actual physical situation. This differential equation looks somewhat more complicated than the falling object example from above. However, with the exception of a little more work, it is not much more complicated. The first step is to determine where the derivative is zero.
0 = ( y 2 - y - 2 ) (1 - y )
2 2
0 = ( y - 2 )( y + 1)(1 - y )
We can now see that we have three values of y in which the derivative, and hence the slope of tangent lines, will be zero. The derivative will be zero at y = -1, 1, and 2. So, let's start our direction field with drawing horizontal tangents for these values. This is shown in the figure below.
Now, we need to add arrows to the four regions that the graph is now divided into. For each of these regions I will pick a value of y in that region and plug it into the right hand side of the differential equation to see if the derivative is positive or negative in that region. Again, to get an
accurate direction fields you should pick a few more over values over the whole range to see how the arrows are behaving over the whole range. y < -1 In this region we can use y = -2 as the test point. At this point we have y = 36 . So, tangent lines in this region will have very steep and positive slopes. Also as y -1 the slopes will flatten out while staying positive. The figure below shows the direction fields with arrows in this region.
-1 < y < 1 In this region we can use y = 0 as the test point. At this point we have y = -2 . Therefore, tangent lines in this region will have negative slopes and apparently not be very steep. So what do the arrows look like in this region? As y 1 staying less that 1 of course, the slopes should be negative and approach zero. As we move away from 1 and towards -1 the slopes will start to get steeper (and stay negative), but eventually flatten back out, again staying negative, as y -1 since the derivative must approach zero at that point. The figure below shows the direction fields with arrows added to this region.
2007 Paul Dawkins
14
1<y<2 In this region we will use y = 1.5 as the test point. At this point we have y = -0.3125 . Tangent lines in this region will also have negative slopes and apparently not be as steep as the previous region. Arrows in this region will behave essentially the same as those in the previous region. Near y = 1 and y = 2 the slopes will flatten out and as we move from one to the other the slopes will get somewhat steeper before flattening back out. The figure below shows the direction fields with arrows added to this region.
y>2 In this last region we will use y = 3 as the test point. At this point we have y = 16 . So, as we saw in the first region tangent lines will start out fairly flat near y = 2 and then as we move way from y = 2 they will get fairly steep. The complete direction field for this differential equation is shown below.
Here is the set of integral curves for this differential equation. Note that due to the steepness of the solutions in the lowest region and the software used to generate these images I was unable to include more than one solution curve in this region.
2007 Paul Dawkins
15
Finally, let's take a look at long term behavior of all solutions. Unlike the first example, the long term behavior in this case will depend on the value of y at t = 0. By examining either of the previous two figures we can arrive at the following behavior of solutions as t . Value of y(0) Behavior as t
y (0) < 1 1 y ( 0) < 2 y (0) = 2 y (0) > 2
y -1 y 1 y2 y
Do, not forget to acknowledge what the horizontal solutions are doing. This is often the most missed portion of this kind of problem. In both of the examples that we've worked to this point the right hand side of the derivative has only contained the function and NOT the independent variable. When the right hand side of the differential equation contains both the function and the independent variable the behavior can be much more complicated and sketching the direction fields by hand can be very difficult. Computer software is very handy in these cases. In some cases they arent too difficult to do by hand however. Lets take a look at the following example.
Example 2 Sketch the direction field for the following differential equation. Sketch the set of
integral curves for this differential equation.
y = y - x
Solution To sketch direction fields for this kind of differential equations we first identify places where the derivative will be constant. To do this we set the derivative in the differential equation equal to a constant, say c. This gives us a family of equations, called isoclines, that we can plot and on each of these curves the derivative will be a constant value of c. Notice that in the previous examples we looked at the isocline for c = 0 to get the direction field
started. For our case the family of isoclines is.
c= y-x
The graph of these curves for several values of c is shown below.
Now, on each of these lines, or isoclines, the derivative will be constant and will have a value of c. On the c = 0 isocline the derivative will always have a value of zero and hence the tangents will all be horizontal. On the c = 1 isocline the tangents will always have a slope of 1, on the c = -2 isocline the tangents will always have a slope of -2, etc. Below is a few tangents put in for each of these isoclines.
2007 Paul Dawkins
17
To add more arrows for those areas between the isoclines start at say, c = 0 and move up to c = 1 and as we do that we increase the slope of the arrows (tangents) from 0 to 1. This is shown in the figure below.
We can then add in integral curves as we did in the previous examples. This is shown in the figure below.
2007 Paul Dawkins
18
Final Thoughts
Before moving on to learning how to solve differential equations I want give a few final thoughts. Any differential equations course will concern itself with answering one or more of the following questions. 1. Given a differential equation will a solution exist? Not all differential equations will have solutions so its useful to know ahead of time if there is a solution or not. If there isnt a solution why waste our time trying to find something that doesnt exist? This question is usually called the existence question in a differential equations course. 2. If a differential equation does have a solution how many solutions are there? As we will see eventually, it is possible for a differential equation to have more than one solution. We would like to know how many solutions there will be for a given differential equation. There is a sub question here as well. What condition(s) on a differential equation are required to obtain a single unique solution to the differential equation? Both this question and the sub question are more important than you might realize. Suppose that we derive a differential equation that will give the temperature distribution in a bar of iron at any time t. If we solve the differential equation and end up with two (or more) completely separate solutions we will have problems. Consider the following situation to see this. If we subject 10 identical iron bars to identical conditions they should all exhibit the same temperature distribution. So only one of our solutions will be accurate, but we will have no way of knowing which one is the correct solution. It would be nice if, during the derivation of our differential equation, we could make sure that our assumptions would give us a differential equation that upon solving will yield a single unique solution. This question is usually called the uniqueness question in a differential equations course. 3. If a differential equation does have a solution can we find it? This may seem like an odd question to ask and yet the answer is not always yes. Just because we know that a solution to a differential equations exists does not mean that we will be able to find it. In a first course in differential equations (such as this one) the third question is the question that we will concentrate on. We will answer the first two equations for special, and fairly simple, cases, but most of our efforts will be concentrated on answering the third question for as wide a variety of differential equations as possible.
2007 Paul Dawkins
19
First Order Differential Equations

Introduction
In this chapter we will look at solving first order differential equations. The most general first order differential equation can be written as,
dy = f ( y, t ) dt
(1)
As we will see in this chapter there is no general formula for the solution to (1). What we will do instead is look at several special cases and see how to solve those. We will also look at some of the theory behind first order differential equations as well as some applications of first order differential equations. Below is a list of the topics discussed in this chapter. Linear Equations Identifying and solving linear first order differential equations. Separable Equations Identifying and solving separable first order differential equations. Well also start looking at finding the interval of validity from the solution to a differential equation. Exact Equations Identifying and solving exact differential equations. Well do a few more interval of validity problems here as well. Bernoulli Differential Equations In this section well see how to solve the Bernoulli Differential Equation. This section will also introduce the idea of using a substitution to help us solve differential equations. Substitutions Well pick up where the last section left off and take a look at a couple of other substitutions that can be used to solve some differential equations that we couldnt otherwise solve. Intervals of Validity Here we will give an in-depth look at intervals of validity as well as an answer to the existence and uniqueness question for first order differential equations. Modeling with First Order Differential Equations Using first order differential equations to model physical situations. The section will show some very real applications of first order differential equations. Equilibrium Solutions We will look at the behavior of equilibrium solutions and autonomous differential equations. Eulers Method In this section well take a brief look at a method for approximating solutions to differential equations.
2007 Paul Dawkins
20
Linear Differential Equations

The first special case of first order differential equations that we will look is the linear first order differential equation. In this case, unlike most of the first order cases that we will look at, we can actually derive a formula for the general solution. The general solution is derived below. However, I would suggest that you do not memorize the formula itself. Instead of memorizing the formula you should memorize and understand the process that I'm going to use to derive the formula. Most problems are actually easier to work by using the process instead of using the formula. So, let's see how to solve a linear first order differential equation. Remember as we go through this process that the goal is to arrive at a solution that is in the form y = y ( t ) . It's sometimes easy to lose sight of the goal as we go through this process for the first time. In order to solve a linear first order differential equation we MUST start with the differential equation in the form shown below. If the differential equation is not in this form then the process were going to use will not work.
dy + p (t ) y = g (t ) dt
(1)
Where both p(t) and g(t) are continuous functions. Recall that a quick and dirty definition of a continuous function is that a function will be continuous provided you can draw the graph from left to right without ever picking up your pencil/pen. In other words, a function is continuous if there are no holes or breaks in it. Now, we are going to assume that there is some magical function somewhere out there in the world, m ( t ) , called an integrating factor. Do not, at this point, worry about what this function is or where it came from. We will figure out what m ( t ) is once we have the formula for the general solution in hand. So, now that we have assumed the existence of m ( t ) multiply everything in (1) by m ( t ) . This will give.
m (t )
dy + m (t ) p (t ) y = m (t ) g (t ) dt
(2)
Now, this is where the magic of m ( t ) comes into play. We are going to assume that whatever
m ( t ) is, it will satisfy the following. m (t ) p (t ) = m (t )

(3)
Again do not worry about how we can find a m ( t ) that will satisfy (3). As we will see, provided p(t) is continuous we can find it. So substituting (3) into (2) we now arrive at.
m (t )
dy + m ( t ) y = m (t ) g (t ) dt
(4)
2007 Paul Dawkins
21
At this point we need to recognize that the left side of (4) is nothing more than the following product rule.
m (t )
dy + m ( t ) y = ( m ( t ) y ( t ) ) dt
So we can replace the left side of (4) with this product rule. Upon doing this (4) becomes
( m ( t ) y ( t ) ) = m ( t ) g ( t )
(5)
Now, recall that we are after y(t). We can now do something about that. All we need to do is integrate both sides then use a little algebra and we'll have the solution. So, integrate both sides of (5) to get.
( m ( t ) y ( t ) ) dt = m ( t ) g ( t ) dt m ( t ) y ( t ) + c = m ( t ) g ( t ) dt
Note the constant of integration, c, from the left side integration is included here. It is vitally important that this be included. If it is left out you will get the wrong answer every time. The final step is then some algebra to solve for the solution, y(t).
(6)
m ( t ) y ( t ) = m ( t ) g ( t ) dt - c
y (t ) =
m ( t ) g ( t ) dt - c
m (t )
Now, from a notational standpoint we know that the constant of integration, c, is an unknown constant and so to make our life easier we will absorb the minus sign in front of it into the constant and use a plus instead. This will NOT affect the final answer for the solution. So with this change we have.
y (t ) =
m ( t ) g ( t ) dt + c
m (t )
(7)
Again, changing the sign on the constant will not affect our answer. If you choose to keep the minus sign you will get the same value of c as I do except it will have the opposite sign. Upon plugging in c we will get exactly the same answer. There is a lot of playing fast and loose with constants of integration in this section, so you will need to get used to it. When we do this we will always to try to make it very clear what is going on and try to justify why we did what we did. So, now that weve got a general solution to (1) we need to go back and determine just what this magical function m ( t ) is. This is actually an easier process that you might think. Well start with (3).
m (t ) p (t ) = m (t )
2007 Paul Dawkins
22
Divide both sides by m ( t ) ,
m (t ) = p (t ) m (t )
Now, hopefully you will recognize the left side of this from your Calculus I class as nothing more than the following derivative.
( ln m ( t ) ) = p ( t )
As with the process above all we need to do is integrate both sides to get.
ln m ( t ) + k = p ( t ) dt
ln m ( t ) = p ( t ) dt + k
You will notice that the constant of integration from the left side, k, had been moved to the right side and had the minus sign absorbed into it again as we did earlier. Also note that were using k here because weve already used c and in a little bit well have both of them in the same equation. So, to avoid confusion we used different letters to represent the fact that they will, in all probability, have different values. Exponentiate both sides to get m ( t ) out of the natural logarithm.
m ( t ) = e
p ( t ) dt + k
Now, its time to play fast and loose with constants again. It is inconvenient to have the k in the exponent so were going to get it out of the exponent in the following way.
m ( t ) = e
p( t ) dt + k p (t ) dt
= ek e
Recall x a +b = x a xb !
Now, lets make use of the fact that k is an unknown constant. If k is an unknown constant then so is e k so we might as well just rename it k and make our life easier. This will give us the following.
m ( t ) = k e
p( t ) dt
(8)
So, we now have a formula for the general solution, (7), and a formula for the integrating factor, (8). We do have a problem however. Weve got two unknown constants and the more unknown constants we have the more trouble well have later on. Therefore, it would be nice if we could find a way to eliminate one of them (well not be able to eliminate both.). This is actually quite easy to do. First, substitute (8) into (7) and rearrange the constants.
2007 Paul Dawkins
23
ke y (t ) =
p (t ) dt g ( t ) dt + c k e k e
p ( t ) dt
p ( t ) dt k e g ( t ) dt + c p ( t ) dt
p(t ) dt g ( t ) dt + c k p t dt ( ) e
So, (7) can be written in such a way that the only place the two unknown constants show up is a ratio of the two. Then since both c and k are unknown constants so is the ratio of the two constants. Therefore well just call the ratio c and then drop k out of (8) since it will just get absorbed into c eventually. The solution to a linear first order differential equation is then
y (t ) =
where,
m ( t ) g ( t ) dt + c
m (t )
p (t ) dt
(9)
m ( t ) = e
(10)
Now, the reality is that (9) is not as useful as it may seem. It is often easier to just run through the process that got us to (9) rather than using the formula. We will not use this formula in any of my examples. We will need to use (10) regularly, as that formula is easier to use than the process to derive it. Solution Process The solution process for a first order linear differential equation is as follows. 1. Put the differential equation in the correct initial form, (1). 2. Find the integrating factor, m ( t ) , using (10). 3. Multiply everything in the differential equation by m ( t ) and verify that the left side becomes the product rule m ( t ) y ( t ) ' and write it as such. 4. Integrate both sides, make sure you properly deal with the constant of integration. 5. Solve for the solution y(t). Lets work a couple of examples. Lets start by solving the differential equation that we derived back in the Direction Field section.
2007 Paul Dawkins
24
Example 1 Find the solution to the following differential equation. dv = 9.8 - 0.196v dt
Solution First we need to get the differential equation in the correct form.
dv + 0.196v = 9.8 dt
From this we can see that p(t)=0.196 and so m ( t ) is then.
m ( t ) = e
0.196 dt
= e0.196t
Note that officially there should be a constant of integration in the exponent from the integration. However, we can drop that for exactly the same reason that we dropped the k from (8). Now multiply all the terms in the differential equation by the integrating factor and do some simplification.
e0.196t
dv + 0.196e 0.196t v = 9.8e 0.196t dt e 0.196t v = 9.8e 0.196t
Integrate both sides and don't forget the constants of integration that will arise from both integrals.
( e0.196t v ) dt = 9.8e 0.196t dt e0.196t v + k = 50e 0.196t + c

Okay. Its time to play with constants again. We can subtract k from both sides to get.
e0.196t v = 50e 0.196t + c - k

Both c and k are unknown constants and so the difference is also an unknown constant. We will therefore write the difference as c. So, we now have
e0.196t v = 50e0.196t + c
From this point on we will only put one constant of integration down when we integrate both sides knowing that if we had written down one for each integral, as we should, the two would just end up getting absorbed into each other. The final step in the solution process is then to divide both sides by e 0.196 t or to multiply both sides by e -0.196 t . Either will work, but I usually prefer the multiplication route. Doing this gives the general solution to the differential equation.
v ( t ) = 50 + ce -0.196t
2007 Paul Dawkins
25
From the solution to this example we can now see why the constant of integration is so important in this process. Without it, in this case, we would get a single, constant solution, v(t)=50. With the constant of integration we get infinitely many solutions, one for each value of c. Back in the direction field section where we first derived the differential equation used in the last example we used the direction field to help us sketch some solutions. Let's see if we got them correct. To sketch some solutions all we need to do is to pick different values of c to get a solution. Several of these are shown in the graph below.
So, it looks like we did pretty good sketching the graphs back in the direction field section. Now, recall from the Definitions section that the Initial Condition(s) will allow us to zero in on a particular solution. Solutions to first order differential equations (not just linear as we will see) will have a single unknown constant in them and so we will need exactly one initial condition to find the value of that constant and hence find the solution that we were after. The initial condition for first order differential equations will be of the form
y ( t0 ) = y0
Recall as well that a differential equation along with a sufficient number of initial conditions is called an Initial Value Problem (IVP).
Example 2 Solve the following IVP. dv = 9.8 - 0.196v dt
v ( 0 ) = 48
Solution To find the solution to an IVP we must first find the general solution to the differential equation and then use the initial condition to identify the exact solution that we are after. So, since this is the same differential equation as we looked at in Example 1, we already have its general solution.
v = 50 + ce -0.196t
Now, to find the solution we are after we need to identify the value of c that will give us the solution we are after. To do this we simply plug in the initial condition which will give us an equation we can solve for c. So let's do this
48 = v ( 0 ) = 50 + c
So, the actual solution to the IVP is.
c = -2
v = 50 - 2e -0.196t
A graph of this solution can be seen in the figure above. Lets do a couple of examples that are a little more involved.
Example 3 Solve the following IVP. cos ( x ) y + sin ( x ) y = 2cos 3 ( x ) sin ( x ) - 1 p y = 3 2, 4 0 x<
p 2
Solution : Rewrite the differential equation to get the coefficient of the derivative a one.
y +
sin ( x ) 1 y = 2cos 2 ( x ) sin ( x ) cos ( x ) cos ( x )
y + tan ( x ) y = 2cos 2 ( x ) sin ( x ) - sec ( x )

Now find the integrating factor.
m ( t ) = e
tan x dx
=e
ln sec x
= eln sec x = sec x
Can you do the integral? If not rewrite tangent back into sines and cosines and then use a simple substitution. Note that we could drop the absolute value bars on the secant because of the limits on x. In fact, this is the reason for the limits on x. Also note that we made use of the following fact.
ln f ( x )
= f ( x)
(11)
This is an important fact that you should always remember for these problems. We will want to simplify the integrating factor as much as possible in all cases and this fact will help with that simplification. Now back to the example. Multiply the integrating factor through the differential equation and verify the left side is a product rule. Note as well that we multiply the integrating factor through the rewritten differential equation and NOT the original differential equation. Make sure that you do this. If you multiply the integrating factor through the original differential equation you will get the wrong solution!
sec ( x ) y + sec ( x ) tan ( x ) y = 2sec ( x ) cos 2 ( x ) sin ( x ) - sec 2 ( x )
(sec ( x ) y ) = 2 cos ( x ) sin ( x ) - sec ( x )

2
Integrate both sides.
2007 Paul Dawkins
27
(sec ( x ) y ( x ) ) dx = 2cos ( x ) sin ( x ) - sec ( x ) dx sec ( x ) y ( x ) = sin ( 2 x ) - sec ( x ) dx

2 2
1 sec ( x ) y ( x ) = - cos ( 2 x ) - tan ( x ) + c 2

Note the use of the trig formula sin ( 2q ) = 2sin q cos q that made the integral easier. Next, solve for the solution.
1 y ( x ) = - cos ( x ) cos ( 2 x ) - cos ( x ) tan ( x ) + c cos ( x ) 2 1 = - cos ( x ) cos ( 2 x ) - sin ( x ) + c cos ( x ) 2

Finally, apply the initial condition to find the value of c.
1 p p p p p 3 2 = y = - cos cos - sin + c cos 2 4 4 2 4 4 3 2=c=7

The solution is then.
2 2 +c 2 2
1 y ( x ) = - cos ( x ) cos ( 2 x ) - sin ( x ) + 7 cos ( x ) 2

Below is a plot of the solution.
2007 Paul Dawkins
28
Example 4 Find the solution to the following IVP. t y + 2 y = t 2 - t + 1 y (1) = 1 2
Solution First, divide through by the t to get the differential equation into the correct form.
y +
Now lets get the integrating factor, m ( t ) .
2 1 y = t -1 + t t
m ( t ) = e t = e 2ln t
Now, we need to simplify m ( t ) . However, we cant use (11) yet as that requires a coefficient of one in front of the logarithm. So, recall that
2 dt
ln x r = r ln x
and rewrite the integrating factor in a form that will allow us to simplify it.
m ( t ) = e 2ln t = eln t = t = t 2
2
We were able to drop the absolute value bars here because we were squaring the t, but often they cant be dropped so be careful with them and dont drop them unless you know that you can. Often the absolute value bars must remain. Now, multiply the rewritten differential equation (remember we cant use the original differential equation here) by the integrating factor.
(t y ) = t
2
- t2 + t
Integrate both sides and solve for the solution.
t 2 y = t 3 - t 2 + t dt 1 1 1 = t4 - t3 + t2 + c 4 3 2 1 2 1 1 c y (t ) = t - t + + 2 4 3 2 t
Finally, apply the initial condition to get the value of c.
1 1 1 1 = y (1) = - + + c 2 4 3 2
The solution is then,
c=
1 12
1 1 1 1 y (t ) = t 2 - t + + 4 3 2 12t 2
Here is a plot of the solution.
Example 5 Find the solution to the following IVP. t y - 2 y = t 5 sin ( 2t ) - t 3 + 4t 4 3 y (p ) = p 4 2
Solution First, divide through by t to get the differential equation in the correct form.
y -
2 y = t 4 sin ( 2t ) - t 2 + 4t 3 t
Now that we have done this we can find the integrating factor, m ( t ) .
m ( t ) = e
2 - dt t
= e -2ln t
Do not forget that the - is part of p(t). Forgetting this minus sign can take a problem that is very easy to do and turn it into a very difficult, if not impossible problem so be careful! Now, we just need to simplify this as we did in the previous example.
-2
m ( t ) = e -2ln t = e
ln t
=t
-2
= t -2
Again, we can drop the absolute value bars since we are squaring the term. Now multiply the differential equation by the integrating factor (again, make sure its the rewritten one and not the original differential equation).
(t y ) = t
-2
sin ( 2t ) - 1 + 4t
2007 Paul Dawkins
30
t -2 y ( t ) = t 2 sin ( 2t ) dt + -1 + 4t dt 1 1 1 t -2 y ( t ) = - t 2 cos ( 2t ) + t sin ( 2t ) + cos ( 2t ) - t + 2t 2 + c 2 2 4 1 1 1 y ( t ) = - t 4 cos ( 2t ) + t 3 sin ( 2t ) + t 2 cos ( 2t ) - t 3 + 2t 4 + ct 2 2 2 4

Apply the initial condition to find the value of c.
The solution is then
3 4 1 1 3 1 p = y (p ) = - p 4 + p 2 - p 3 + 2p 4 + cp 2 = p 4 - p 3 + p 2 + cp 2 2 2 4 2 4 1 p 3 - p 2 = cp 2 4 1 c =p 4 1 1 1 1 y ( t ) = - t 4 cos ( 2t ) + t 3 sin ( 2t ) + t 2 cos ( 2t ) - t 3 + 2t 4 + p - t 2 2 2 4 4
Below is a plot of the solution.
Lets work one final example that looks more at interpreting a solution rather than finding a solution.
Example 6 Find the solution to the following IVP and determine all possible behaviors of the solution as t . If this behavior depends on the value of y0 give this dependence. 2 y - y = 4sin ( 3t ) y ( 0 ) = y0
Solution First, divide through by a 2 to get the differential equation in the correct form.
y Now find m ( t ) .
2007 Paul Dawkins 31
1 y = 2sin ( 3t ) 2
m ( t ) = e
- 1 dt
2
=e
t 2
Multiply m ( t ) through the differential equation and rewrite the left side as a product rule.
t t -2 2 e y = 2e sin ( 3t )
e 2 y = 2e 2 sin ( 3t ) dt + c
t t 24 - 2 4 -2 e y = - e cos ( 3t ) - e sin ( 3t ) + c 37 37 t 24 4 y ( t ) = - cos ( 3t ) - sin ( 3t ) + ce 2 37 37 t 2
Apply the initial condition to find the value of c and note that it will contain y0 as we dont have a value for that.
y0 = y ( 0 ) = So the solution is
24 +c 37
c = y0 +
24 37
y (t ) = -
24 4 24 t cos ( 3t ) - sin ( 3t ) + y0 + e 2 37 37 37
Now that we have the solution, lets look at the long term behavior (i.e. t ) of the solution. The first two terms of the solution will remain finite for all values of t. It is the last term that will determine the behavior of the solution. The exponential will always go to infinity as t , however depending on the sign of the coefficient c (yes weve already found it, but for ease of this discussion well continue to call it c). The following table gives the long term behavior of the solution for all values of c. Range of c Behavior of solution as t c<0 c=0 c>0
y ( t ) - y ( t ) remains finite y (t )
This behavior can also be seen in the following graph of several of the solutions.
2007 Paul Dawkins
32
Now, because we know how c relates to y0 we can relate the behavior of the solution to y0. The following table give the behavior of the solution in terms of y0 instead of c. Range of y0 Behavior of solution as t
24 37 24 y0 = 37 24 y0 > 37 y0 < -
y ( t ) - y ( t ) remains finite y (t )
24 Note that for y0 = - 37 the solution will remain finite. That will not always happen.
Investigating the long term behavior of solutions is sometimes more important than the solution itself. Suppose that the solution above gave the temperature in a bar of metal. In this case we would want the solution(s) that remains finite in the long term. With this investigation we would now have the value of the initial condition that will give us that solution and more importantly values of the initial condition that we would need to avoid so that we didnt melt the bar.
2007 Paul Dawkins
33
Separable Differential Equations

We are now going to start looking at nonlinear first order differential equations. The first type of nonlinear first order differential equations that we will look at is separable differential equations. A separable differential equation is any differential equation that we can write in the following form.
N ( y)
dy = M ( x) dx
(1)
Note that in order for a differential equation to be separable all the y's in the differential equation must be multiplied by the derivative and all the x's in the differential equation must be on the other side of the equal sign. Solving separable differential equation is fairly easy. We first rewrite the differential equation as the following
N ( y ) dy = M ( x ) dx
Then you integrate both sides.
N ( y ) dy = M ( x ) dx
(2)
So, after doing the integrations in (2) you will have an implicit solution that you can hopefully solve for the explicit solution, y(x). Note that it won't always be possible to solve for an explicit solution. Recall from the Definitions section that an implicit solution is a solution that is not in the form y = y ( x ) while an explicit solution has been written in that form. We will also have to worry about the interval of validity for many of these solutions. Recall that the interval of validity was the range of the independent variable, x in this case, on which the solution is valid. In other words, we need to avoid division by zero, complex numbers, logarithms of negative numbers or zero, etc. Most of the solutions that we will get from separable differential equations will not be valid for all values of x. Lets start things off with a fairly simple example so we can see the process without getting lost in details of the other issues that often arise with these problems.
Example 1 Solve the following differential equation and determine the interval of validity for
the solution.
dy = 6 y2x dx
y (1) =
1 25
Solution It is clear, hopefully, that this differential equation is separable. So, lets separate the differential equation and integrate both sides. As with the linear first order officially we will pick up a constant of integration on both sides from the integrals on each side of the equal sign. The two can be moved to the same side an absorbed into each other. We will use the convention that puts the single constant on the side with the xs.
y -2 dy = 6 x dx
-2
dy = 6 x dx 1 = 3x 2 + c y
So, we now have an implicit solution. This solution is easy enough to get an explicit solution, however before getting that it is usually easier to find the value of the constant at this point. So apply the initial condition and find the value of c.
1 2 = 3 (1) + c 1 25 1 = 3 x 2 - 28 y 1 y ( x) = 28 - 3x 2
c = -28
Plug this into the general solution and then solve to get an explicit solution.
Now, as far as solutions go weve got the solution. We do need to start worrying about intervals of validity however. Recall that there are two conditions that define an interval of validity. First, it must be a continuous interval with no breaks or holes in it. Second it must contain the value of the independent variable in the initial condition, x = 1 in this case. So, for our case weve got to avoid two values of x. Namely, x
28 3
3.05505 since these
will give us division by zero. This gives us three possible intervals of validity.
- < x < -
28 3
28 28 < x< 3 3
28 <x< 3
However, only one of these will contain the value of x from the initial condition and so we can see that
28 28 < x< 3 3
must be the interval of validity for this solution. Here is a graph of the solution.
2007 Paul Dawkins
35
Note that this does not say that either of the other two intervals listed above cant be the interval of validity for any solution. With the proper initial condition either of these could have been the interval of validity. Well leave it to you to verify the details of the following claims. If we use an initial condition of
y ( -4 ) = -
1 20 28 3
we will get exactly the same solution however in this case the interval of validity would be the first one.
- < x < Likewise, if we use
y (6) = -
1 80
as the initial condition we again get exactly the same solution and in this case the third interval becomes the interval of validity.
28 <x< 3
So, simply changing the initial condition a little can give any of the possible intervals.
Example 2 Solve the following IVP and find the interval of validity for the solution. 3x 2 + 4 x - 4 y = y (1) = 3 2y - 4
Solution This differential equation is clearly separable, so let's put it in the proper form and then integrate both sides.
2007 Paul Dawkins
36
( 2 y - 4 ) dy = ( 3x 2 + 4 x - 4 ) dx
( 2 y - 4) dy = (3x
+ 4 x - 4 ) dx
y 2 - 4 y = x3 + 2 x 2 - 4 x + c
We now have our implicit solution, so as with the first example lets apply the initial condition at this point to determine the value of c.
( 3)
- 4 ( 3) = (1) + 2 (1) - 4 (1) + c

3 2
c = -2
The implicit solution is then
y 2 - 4 y = x3 + 2 x 2 - 4 x - 2
We now need to find the explicit solution. This is actually easier than it might look and you already know how to do it. First we need to rewrite the solution a little
y 2 - 4 y - ( x3 + 2 x2 - 4 x - 2) = 0
To solve this all we need to recognize is that this is quadratic in y and so we can use the quadratic formula to solve it. However, unlike quadratics you are used to, at least some of the constants will not actually be constant, but will in fact involve xs. So, upon using the quadratic formula on this we get.
y ( x) = =
4 16 - 4 (1) - ( x3 + 2 x 2 - 4 x - 2 ) 4 16 + 4 ( x + 2 x 2 - 4 x - 2 )
3
Next, notice that we can factor a 4 out from under the square root (it will come out as a 2) and then simplify a little.
y ( x) =
4 2 4 + ( x3 + 2 x 2 - 4 x - 2 ) 2
= 2 x3 + 2 x 2 - 4 x + 2
We are almost there. Notice that weve actually got two solutions here (the ) and we only want a single solution. In fact, only one of the signs can be correct. So, to figure out which one is correct we can reapply the initial condition to this. Only one of the signs will give the correct value so we can use this to figure out which one of the signs is correct. Plugging x = 1 into the solution gives.
3 = y (1) = 2 1 + 2 - 4 + 2 = 2 1 = 3,1
In this case it looks like the + is the correct sign for our solution. Note that it is completely possible that the could be the solution so dont always expect it to be one or the other. The explicit solution for our differential equation is.
y ( x ) = 2 + x3 + 2 x 2 - 4 x + 2
To finish the example out we need to determine the interval of validity for the solution. If we were to put a large negative value of x in the solution we would end up with complex values in our solution and we want to avoid complex numbers in our solutions here. So, we will need to determine which values of x will give real solutions. To do this we will need to solve the following inequality.
x3 + 2 x 2 - 4 x + 2 0
In other words, we need to make sure that the quantity under the radical stays positive. Using a computer algebra system like Maple or Mathematica we see that the left side is zero at x = 3.36523 as well as two complex values, but we can ignore complex values for interval of validity computations. Finally a graph of the quantity under the radical is shown below.
So, in order to get real solutions we will need to require x -3.36523 because this is the range of xs for which the quantity is positive. Notice as well that this interval also contains the value of x that is in the initial condition as it should. Therefore, the interval of validity of the solution is x -3.36523 . Here is graph of the solution.
2007 Paul Dawkins
38
Example 3 Solve the following IVP and find the interval of validity of the solution. xy 3 y = y ( 0 ) = -1 1 + x2
Solution First separate and then integrate both sides.
y -3dy = x (1 + x 2 ) 2 dx
2 -2 = + y dy x x 1 ( ) dx 1 - 2 = 1 + x2 + c 2y -3 1
Apply the initial condition to get the value of c.
The implicit solution is then,
1 = 1+c 2
c=-
3 2
Now lets solve for y(x).
1 3 = 1 + x2 2 2y 2
1 = 3 - 2 1 + x2 y2 1 y2 = 3 - 2 1 + x2 1 y ( x) = 3 - 2 1 + x2
2007 Paul Dawkins
39
Reapplying the initial condition shows us that the is the correct sign. The explicit solution is then,
y ( x) = -
1 3 - 2 1 + x2
Lets get the interval of validity. Thats easier than it might look for this problem. First, since 1 + x 2 0 the inner root will not be a problem. Therefore all we need to worry about is division by zero and negatives under the outer root. We can take care of both be requiring
3 - 2 1 + x2 > 0 3 > 2 1 + x2 9 > 4 (1 + x 2 ) 9 > 1 + x2 4 5 > x2 4

Note that we were able to square both sides of the inequality because both sides of the inequality are guaranteed to be positive in this case. Finally solving for x we see that the only possible range of xs that will not give division by zero or square roots of negative numbers will be,
5 5 <x< 2 2
and nicely enough this also contains the initial condition x=0. This interval is therefore our interval of validity. Here is a graph of the solution.
2007 Paul Dawkins
40
Example 4 Solve the following IVP and find the interval of validity of the solution. y = e- y ( 2 x - 4 ) y (5) = 0
Solution This differential equation is easy enough to separate, so let's do that and then integrate both sides.
e y dy = ( 2 x - 4 ) dx
y
e
Applying the initial condition gives
dy = ( 2 x - 4 ) dx e y = x2 - 4x + c
1 = 25 - 20 + c e y = x2 - 4 x - 4
c = -4
This then gives an implicit solution of.
We can easily find the explicit solution to this differential equation by simply taking the natural log of both sides.
y ( x ) = ln ( x 2 - 4 x - 4 )
Finding the interval of validity is the last step that we need to take. Recall that we can't plug negative values or zero into a logarithm, so we need to solve the following inequality
x2 - 4x - 4 > 0
The quadratic will be zero at the two points x = 2 2 2 . A graph of the quadratic (shown below) shows that there are in fact two intervals in which we will get positive values of the polynomial and hence can be possible intervals of validity.
So, possible intervals of validity are
- < x < 2-2 2 2+ 2 2 < x <

From the graph of the quadratic we can see that the second one contains x = 5, the value of the
independent variable from the initial condition. Therefore the interval of validity for this solution is.
2+ 2 2 < x <
Here is a graph of the solution.
Example 5 Solve the following IVP and find the interval of validity for the solution. dr r 2 = r (1) = 2 dq q
Solution This is actually a fairly simple differential equation to solve. Im doing this one mostly because of the interval of validity. So, get things separated out and then integrate.
1 1 dr = dq 2 q r 1 1 2 dr = dq r q 1 - = ln q + c r
Now, apply the initial condition to find c.
So, the implicit solution is then,
1 = ln (1) + c 2 1 1 - = ln q r 2
c=-
1 2
Solving for r gets us our explicit solution.
r=
1 2
1 - ln q
2007 Paul Dawkins
42
Now, there are two problems for our solution here. First we need to avoid q = 0 because of the natural log. Notice that because of the absolute value on the q we dont need to worry about q being negative. We will also need to avoid division by zero. In other words, we need to avoid the following points.
1 - ln q = 0 2 1 ln q = 2
exponentiate both sides
q = e2 q = e
So, these three points break the number line up into four portions, each of which could be an interval of validity.
- <q < - e - e <q < 0 0 <q < e e <q <

The interval that will be the actual interval of validity is the one that contains q = 1. Therefore, the interval of validity is 0 < q < e . Here is a graph of the solution.
2007 Paul Dawkins
43
Example 6 Solve the following IVP. dy = e y - t sec ( y ) (1 + t 2 ) dt
y (0) = 0
Solution This problem will require a little work to get it separated and in a form that we can integrate, so let's do that first.
e - y cos ( y ) dy = e- t (1 + t 2 ) dt
dy e y e- t 1+ t2 ) = ( dt cos ( y )
Now, with a little integration by parts on both sides we can get an implicit solution.
-y
cos ( y ) dy = e -t (1 + t 2 ) dt
e- y (sin ( y ) - cos ( y ) ) = -e-t (t 2 + 2t + 3) + c 2

Applying the initial condition gives.
1 ( -1) = - ( 3) + c 2
Therefore, the implicit solution is.
c=
5 2
ey 5 sin ( y ) - cos ( y ) ) = -e -t ( t 2 + 2t + 3) + ( 2 2
It is not possible to find an explicit solution for this problem and so we will have to leave the solution in its implicit form. Finding intervals of validity from implicit solutions can often be very difficult so we will also not bother with that for this problem. As this last example showed it is not always possible to find explicit solutions so be on the lookout for those cases.
2007 Paul Dawkins
44
Exact Differential Equations

The next type of first order differential equations that well be looking at is exact differential equations. Before we get into the full details behind solving exact differential equations its probably best to work an example that will help to show us just what an exact differential equation is. It will also show some of the behind the scenes details that we usually dont bother with in the solution process. The vast majority of the following example will not be done in any of the remaining examples and the work that we will put into the remaining examples will not be shown in this example. The whole point behind this example is to show you just what an exact differential equation is, how we use this fact to arrive at a solution and why the process works as it does. The majority of the actual solution details will be shown in a later example.
Example 1 Solve the following differential equation. 2 xy - 9 x 2 + ( 2 y + x 2 + 1) dy =0 dx
Solution Lets start off by supposing that somewhere out there in the world is a function Y(x,y) that we can find. For this example the function that we need is
Y ( x, y ) = y 2 + ( x 2 + 1) y - 3x 3
Do not worry at this point about where this function came from and how we found it. Finding the function, Y(x,y), that is needed for any particular differential equation is where the vast majority of the work for these problems lies. As stated earlier however, the point of this example is to show you why the solution process works rather than showing you the actual solution process. We will see how to find this function in the next example, so at this point do not worry about how to find it, simply accept that it can be found and that weve done that for this particular differential equation. Now, take some partial derivatives of the function.
Y x = 2 xy - 9 x 2 Y y = 2 y + x2 +1
Now, compare these partial derivatives to the differential equation and youll notice that with these we can now write the differential equation as.
Yx + Y y
dy =0 dx
(1)
Now, recall from your multi-variable calculus class (probably Calculus III) that (1) is nothing more that the following derivative (youll need the multi-variable chain rule for this).
d Y ( x, y ( x ) ) dx
So, the differential equation can now be written as
2007 Paul Dawkins
45
d Y ( x, y ( x ) ) = 0 dx
Now, if the ordinary (not partial) derivative of something is zero, that something must have been a constant to start with. In other words, weve got to have Y ( x, y ) = c . Or,
y 2 + ( x 2 + 1) y - 3x 3 = c
This then is an implicit solution for our differential equation! If we had an initial condition we could solve for c. We could also find an explicit solution if we wanted to, but well hold off on that until the next example. Okay, so what did we learn from the last example? Lets look at things a little more generally. Suppose that we have the following differential equation.
M ( x, y ) + N ( x, y )
dy =0 dx
(2)
Note that its important that it be in this form! There must be an = 0 on one side and the sign separating the two terms must be a +. Now, if there is a function somewhere out there in the world, Y(x,y), so that,
Y x = M ( x, y )
and
Y y = N ( x, y )
then we call the differential equation exact. In these cases we can write the differential equation as
Yx + Y y
dy =0 dx
(3)
Then using the chain rule from Calculus III we can further reduce the differential equation to the following derivative,
d Y ( x, y ( x ) ) = 0 dx
The (implicit) solution to an exact differential equation is then
Y ( x, y ) = c
(4)
Well, its the solution provided we can find Y(x,y) anyway. Therefore, once we have the function we can always just jump straight to (4) to get an implicit solution to our differential equation. Finding the function Y(x,y) is clearly the central task in determining if a differential equation is exact and in finding it's solution. As we will see, finding Y(x,y) can be a somewhat lengthy process in which there is the chance of mistakes. Therefore, it would be nice if there was some simple test that we could use before even starting to see if a differential equation is exact or not. This will be especially useful if it turns out that the differential equation is not exact, since in this case Y(x,y) will not exist. It would be a waste of time to try and find a nonexistent function!
2007 Paul Dawkins
46
So, let's see if we can find a test for exact differential equations. Let's start with (2) and assume that the differential equation is in fact exact. Since its exact we know that somewhere out there is a function Y(x,y) that satisfies
Yx = M Yy = N
Now, provided Y(x,y) is continuous and its first order derivatives are also continuous we know that
Yx y = Y yx
However, we also have the following.
Yx y = (Yx )y = (M )y = M y Y y x = ( Y y ) = ( N )x = N x
x
Therefore, if a differential equation is exact and Y(x,y) meets all of its continuity conditions we must have.
M y = Nx
Likewise if (5) is not true there is no way for the differential equation to be exact.
(5)
Therefore, we will use (5) as a test for exact differential equations. If (5) is true we will assume that the differential equation is exact and that Y(x,y) meets all of its continuity conditions and proceed with finding it. Note that for all the examples here the continuity conditions will be met and so this wont be an issue. Okay, lets go back and rework the first example. This time we will use the example to show how to find Y(x,y). Well also add in an initial condition to the problem.
Example 2 Solve the following IVP and find the interval of validity for the solution. dy 2 xy - 9 x 2 + ( 2 y + x 2 + 1) = 0, y ( 0 ) = -3 dx
Solution First identify M and N and check that the differential equation is exact.
M = 2 xy - 9 x 2 N = 2 y + x2 + 1
M y = 2x Nx = 2x
So, the differential equation is exact according to the test. However, we already knew that as we have given you Y(x,y). Its not a bad thing to verify it however and to run through the test at least once however. Now, how do we actually find Y(x,y)? Well recall that
Yx = M Yy = N
2007 Paul Dawkins
47
We can use either of these to get a start on finding Y(x,y) by integrating as follows.
Y = M dx
OR
Y = N dy
However, we will need to be careful as this wont give us the exact function that we need. Often it doesnt matter which one you choose to work with while in other problems one will be significantly easier than the other. In this case it doesnt matter which one we use as either will be just as easy. So, Ill use the first one.
Y ( x, y ) = 2 xy - 9 x 2 dx = x 2 y - 3 x3 + h ( y )
Note that in this case the constant of integration is not really a constant at all, but instead it will be a function of the remaining variable(s), y in this case. Recall that in integration we are asking what function we differentiated to get the function we are integrating. Since we are working with two variables here and talking about partial differentiation with respect to x, this means that any term that contained only constants or ys would have differentiated away to zero, therefore we need to acknowledge that fact by adding on a function of y instead of the standard c. Okay, weve got most of Y(x,y) we just need to determine h(y) and well be done. This is actually easy to do. We used Y x = M to find most of Y(x,y) so well use Y y = N to find h(y). Differentiate our Y(x,y) with respect to y and set this equal to N (since they must be equal after all). Dont forget to differentiate h(y) ! Doing this gives,
Y y = x 2 + h ( y ) = 2 y + x 2 + 1 = N h ( y ) = 2 y + 1
From this we can see that
Note that at this stage h(y) must be only a function of y and so if there are any xs in the equation at this stage we have made a mistake somewhere and its time to go look for it. We can now find h(y) by integrating.
h ( y ) = 2 y + 1 dy = y 2 + y + k
Youll note that we included the constant of integration, k, here. It will turn out however that this will end up getting absorbed into another constant so we can drop it in general. So, we can now write down Y(x,y).
Y ( x, y ) = x 2 y - 3x 3 + y 2 + y + k = y 2 + ( x 2 + 1) y - 3x 3 + k
With the exception of the k this is identical to the function that we used in the first example. We can now go straight to the implicit solution using (4).
y 2 + ( x 2 + 1) y - 3x 3 + k = c
2007 Paul Dawkins
48
Well now take care of the k. Since both k and c are unknown constants all we need to do is subtract one from both sides and combine and we still have an unknown constant.
y 2 + ( x 2 + 1) y - 3x 3 = c - k y 2 + ( x 2 + 1) y - 3x 3 = c
Therefore, well not include the k in anymore problems. This is where we left off in the first example. Lets now apply the initial condition to find c.
( -3) + ( 0 + 1)( -3) - 3 ( 0 )

2
=c
c=6
The implicit solution is then.
y 2 + ( x 2 + 1) y - 3x 3 - 6 = 0
Now, as we saw in the separable differential equation section, this is quadratic in y and so we can solve for y(x) by using the quadratic formula.
y ( x) = =
- ( x 2 + 1)
(x
+ 1) - 4 (1) ( -3 x3 - 6 )
2
2 (1) - ( x 2 + 1) x 4 + 12 x 3 + 2 x 2 + 25 2 -1 25 -1 5 = = -3, 2 2 2
Now, reapply the initial condition to figure out which of the two signs in the that we need.
-3 = y ( 0 ) =
So, it looks like the - is the one that we need. The explicit solution is then.
y ( x) =
- ( x 2 + 1) - x 4 + 12 x 3 + 2 x 2 + 25 2
Now, for the interval of validity. It looks like we might well have problems with square roots of negative numbers. So, we need to solve
x 4 + 12 x3 + 2 x 2 + 25 = 0
Upon solving this equation is zero at x = 11.81557624 and x = 1.396911133. Note that youll need to use some form of computational aid in solving this equation. Here is a graph of the polynomial under the radical.
2007 Paul Dawkins
49
So, it looks like there are two intervals where the polynomial will be positive.
- < x -11.81557624 -1.396911133 x <
However, recall that intervals of validity need to be continuous intervals and contain the value of x that is used in the initial condition. Therefore the interval of validity must be.
-1.396911133 x <
Here is a quick graph of the solution.
That was a long example, but mostly because of the initial explanation of how to find Y(x,y). The remaining examples will not be as long.
2007 Paul Dawkins
50
Example 3 Find the solution and interval of validity for the following IVP. 2 xy 2 + 4 = 2 ( 3 - x 2 y ) y y ( -1) = 8
Solution Here, we first need to put the differential equation into proper form before proceeding. Recall that it needs to be = 0 and the sign separating the two terms must be a plus!
2 xy 2 + 4 - 2 ( 3 - x 2 y ) y = 0
2 xy 2 + 4 + 2 ( x 2 y - 3) y = 0
So we have the following
M = 2 xy 2 + 4 N = 2x2 y - 6
M y = 4 xy N x = 4 xy
and so the differential equation is exact. We can either integrate M with respect to x or integrate N with respect to y. In this case either would be just as easy so well integrate N this time so we can say that weve got an example of both down here.
Y ( x, y ) = 2 x 2 y - 6 dy = x 2 y 2 - 6 y + h ( x )
This time, as opposed to the previous example, our constant of integration must be a function of x since we integrated with respect to y. Now differentiate with respect to x and compare this to M.
Y x = 2 xy 2 + h ( x ) = 2 xy 2 + 4 = M h ( x ) = 4
So, it looks like
h ( x ) = 4x
Again, well drop the constant of integration that technically should be present in h(x) since it will just get absorbed into the constant we pick up in the next step. Also note that, h(x) should only involve xs at this point. If there are any ys left at this point a mistake has been made so go back and look for it. Writing everything down gives us the following for Y(x,y).
Y ( x, y ) = x 2 y 2 - 6 y + 4 x x2 y 2 - 6 y + 4x = c
So, the implicit solution to the differential equation is
Applying the initial condition gives, The solution is then
64 - 48 - 4 = c
c = 12
x 2 y 2 - 6 y + 4 x - 12 = 0
Using the quadratic formula gives us
y ( x) = = = =
6 36 - 4 x 2 ( 4 x - 12 ) 2x2 6 36 + 48 x 2 - 16 x3 2 x2 6 2 9 + 12 x 2 - 4 x 3 2x2 3 9 + 12 x 2 - 4 x 3 x2
Reapplying the initial condition shows that this time we need the + (well leave those details to you to check). Therefore, the explicit solution is
y ( x) =
3 + 9 + 12 x 2 - 4 x 3 x2
Now lets find the interval of validity. Well need to avoid x = 0 so we dont get division by zero. Well also have to watch out for square roots of negative numbers so solve the following equation.
-4 x 3 + 12 x 2 + 9 = 0
The only real solution here is x = 3.217361577. Below is a graph of the polynomial.
So, it looks like the polynomial will be positive, and hence okay under the square root on
- < x < 3.217361577
Now, this interval cant be the interval of validity because it contains x = 0 and we need to avoid that point. Therefore, this interval actually breaks up into two different possible intervals of validity.
- < x < 0 0 < x < 3.217361577
The first one contains x = -1, the x value from the initial condition. Therefore, the interval of
validity for this problem is - < x < 0 . Here is a graph of the solution.
Example 4 Find the solution and interval of validity to the following IVP. 2ty - 2t - 2 - ln ( t 2 + 1) y = 0 y ( 5) = 0 2 t +1
Solution So, first deal with that minus sign separating the two terms.
2ty - 2t + ln ( t 2 + 1) - 2 y = 0 t +1
2
Now, find M and N and check that its exact.
M=
2ty - 2t t2 +1
My =
N = ln ( t 2 + 1) - 2
So, its exact. Well integrate the first one in this case.
2t t +1 2t Nt = 2 t +1
2
2ty Y ( t, y ) = - 2t dt = y ln ( t 2 + 1) - t 2 + h ( y ) 2 t +1
Differentiate with respect to y and compare to N.
Y y = ln ( t 2 + 1) + h ( y ) = ln ( t 2 + 1) - 2 = N h ( y ) = -2 h ( y ) = -2 y
So, it looks like weve got.
This gives us
Y ( t , y ) = y ln ( t 2 + 1) - t 2 - 2 y
53 http://tutorial.math.lamar.edu/terms.aspx
2007 Paul Dawkins
The implicit solution is then,
y ln ( t 2 + 1) - t 2 - 2 y = c -25 = c
Applying the initial condition gives, The implicit solution is now,
y ln ( t 2 + 1) - 2 - t 2 = -25
This solution is much easier to solve than the previous ones. No quadratic formula is needed this time, all we need to do is solve for y. Heres what we get for an explicit solution.
y (t ) =
t 2 - 25 ln ( t 2 + 1) - 2
Alright, lets get the interval of validity. The term in the logarithm is always positive so we dont need to worry about negative numbers in that. We do need to worry about division by zero however. We will need to avoid the following point(s).
ln ( t 2 + 1) - 2 = 0
ln ( t 2 + 1) = 2 t 2 + 1 = e2 t = e2 - 1
We now have three possible intervals of validity.
- < t < - e2 - 1 - e2 - 1 < t < e2 - 1 e2 - 1 < t <

The last one contains t = 5 and so is the interval of validity for this problem is Heres a graph of the solution.
e2 - 1 < t < .
2007 Paul Dawkins
54
Example 5 Find the solution and interval of validity for the following IVP. 3 y 3e3 xy - 1 + ( 2 ye3 xy + 3 xy 2e3 xy ) y = 0 y ( 0) = 1
Solution Lets identify M and N and check that its exact.
M = 3 y 3e3 xy - 1 N = 2 ye3 xy + 3 xy 2e3 xy
M y = 9 y 2e3 xy + 9 xy 3e3 xy N x = 9 y 2e3 xy + 9 xy 3e3 xy
So, its exact. With the proper simplification integrating the second one isnt too bad. However, the first is already set up for easy integration so lets do that one.
Y ( x, y ) = 3 y 3e3xy - 1 dx = y 2e3 xy - x + h ( y )
Differentiate with respect to y and compare to N.
Y y = 2 ye3 xy + 3 xy 2 e3 xy + h ( y ) = 2 ye3 xy + 3 xy 2e3 xy = N h ( y ) = 0 h ( y) = 0
So, it looks like weve got
Recall that actually h(y) = k, but we drop the k because it will get absorbed in the next step. That gives us h(y) = 0. Therefore, we get.
Y ( x, y ) = y 2e3 xy - x y 2e3 xy - x = c
The implicit solution is then
Apply the initial condition. The implicit solution is then
1= c y 2e3 xy - x = 1
This is as far as we can go. There is no way to solve this for y and get an explicit solution.
2007 Paul Dawkins
55
Bernoulli Differential Equations

In this section we are going to take a look at differential equations in the form,
y + p ( x ) y = q ( x ) y n
where p(x) and q(x) are continuous functions on the interval were working on and n is a real number. Differential equations in this form are called Bernoulli Equations. First notice that if n = 0 or n = 1 then the equation is linear and we already know how to solve it in these cases. Therefore, in this section were going to be looking at solutions for values of n other than these two. In order to solve these well first divide the differential equation by y n to get,
y - n y + p ( x ) y1- n = q ( x )
We are now going to use the substitution v = y1- n to convert this into a differential equation in terms of v. As well see this will lead to a differential equation that we can solve. We are going to have to be careful with this however when it comes to dealing with the derivative, y . We need to determine just what y is in terms of our substitution. This is easier to do than it might at first look to be. All that we need to do is differentiate both sides of our substitution with respect to x. Remember that both v and y are functions of x and so well need to use the chain rule on the right side. If you remember your Calculus I youll recall this is just implicit differentiation. So, taking the derivative gives us,
v = (1 - n ) y - n y
Now, plugging this as well as our substitution into the differential equation gives,
1 v + p ( x ) v = q ( x ) 1- n
This is a linear differential equation that we can solve for v and once we have this in hand we can also get the solution to the original differential equation by plugging v back into our substitution and solving for y. Lets take a look at an example.
Example 1 Solve the following IVP and find the interval of validity for the solution. 4 y + y = x3 y 2 y ( 2 ) = -1, x>0 x
Solution So, the first thing that we need to do is get this into the proper form and that means dividing everything by y 2 . Doing this gives,
y -2 y +
4 -1 y = x3 x
2007 Paul Dawkins
56
The substitution and derivative that well need here is,
v = y -1
With this substitution the differential equation becomes,
v = - y -2 y 4 -v + v = x 3 x
So, as noted above this is a linear differential equation that we know how to solve. Well do the details on this one and then for the rest of the examples in this section well leave the details for you to fill in. If you need a refresher on solving linear differential equations then go back to that section for a quick review. Heres the solution to this differential equation.
4
4 v - v = - x3 x ( x -4 v ) dx = - x -1 dx -4 x v = - ln x + c
- dx m ( x ) = e x = e -4 ln x = x -4
v ( x ) = cx 4 - x 4 ln x
Note that we dropped the absolute value bars on the x in the logarithm because of the assumption that x > 0 . Now we need to determine the constant of integration. This can be done in one of two ways. We can can convert the solution above into a solution in terms of y and then use the original initial condition or we can convert the initial condition to an initial condition in terms of v and use that. Because well need to convert the solution to ys eventually anyway and it wont add that much work in well do it that way. So, to get the solution in terms of y all we need to do is plug the substitution back in. Doing this gives,
y -1 = x 4 ( c - ln x )
At this point we can solve for y and then apply the initial condition or apply the initial condition and then solve for y. Well generally do this with the later approach so lets apply the initial condition to get,
( -1)
y ( x) =
-1
= c 2 4 - 2 4 ln 2
c = ln 2 -
1 16
Plugging in for c and solving for y gives,

4
1 -16 -16 = 4 = 4 1 x x ( ln 2 - 16 - ln x ) x (1 + 16 ln x - 16 ln 2 ) x (1 + 16ln 2 )
Note that we did a little simplification in the solution. This will help with finding the interval of validity. Before finding the interval of validity however, we mentioned above that we could convert the
original initial condition into an initial condition for v. Lets briefly talk about how to do that. To do that all we need to do is plug x = 2 into the substitution and then use the original initial condition. Doing this gives,
v ( 2 ) = y -1 ( 2 ) = ( -1) = -1
-1
So, in this case we got the same value for v that we had for y. Dont expect that to happen in general if you chose to do the problems in this manner. Okay, lets now find the interval of validity for the solution. First we already know that x > 0 and that means well avoid the problems of having logarithms of negative numbers and division by zero at x = 0 . So, all that we need to worry about then is division by zero in the second term and this will happen where,
1 + 16 ln x =0 2
x 1 ln 2 = - 16 x
1 16 1 16
=e
1 16 1 16
x = 2e
1.8788
The two possible intervals of validity are then,
0 < x < 2e
then 2e
1 16
2e
< x<
and since the second one contains the initial condition we know that the interval of validity is
< x<.
Lets do a couple more examples and as noted above were going to leave it to you to solve the linear differential equation when we get to that stage.
2007 Paul Dawkins
58
Example 2 Solve the following IVP and find the interval of validity for the solution. y = 5 y + e -2 x y -2 y (0) = 2
Solution The first thing well need to do here is multiply through by y 2 and well also do a little rearranging to get things into the form well need for the linear differential equation. This gives,
y 2 y - 5 y 3 = e -2 x
The substitution here and its derivative is,
v = y3 v - 5v = e -2 x
v = 3 y 2 y v - 15v = 3e -2 x
Plugging the substitution into the differential equation gives,

1 3
m ( x ) = e-15 x
We rearranged a little and gave the integrating factor for the linear differential equation solution. Upon solving we get,
3 -2 x v ( x ) = ce15 x - 17 e
Now go back to ys.

3 -2 x y 3 = ce15 x - 17 e
Applying the initial condition and solving for c gives,

3 8 = c - 17
c = 139 17
Plugging in c and solving for y gives,
139e15 x - 3e -2 x 3 y ( x) = 17
There are no problem values of x for this solution and so the interval of validity is all real numbers. Heres a graph of the solution.
2007 Paul Dawkins
59
Example 3 Solve the following IVP and find the interval of validity for the solution. 6 y - 2 y = x y 4 y ( 0 ) = -2
Solution First get the differential equation in the proper form and then write down the substitution.
6 y -4 y - 2 y -3 = x -2v - 2v = x
v = y -3 v + v = - 1 2 x
v = -3 y -4 y
Plugging the substitution into the differential equation gives,
m ( x) = e x
Again, weve rearranged a little and given the integrating factor needed to solve the linear differential equation. Upon solving the linear differential equation we have,
-x v ( x) = - 1 2 ( x - 1) + ce
Now back substitute to get back into ys.
-x y -3 = - 1 2 ( x - 1) + ce
Now we need to apply the initial condition and solve for c.

1 -1 8 = 2 +c
c=-5 8
Plugging in c and solving for y gives,
y ( x) = -
( 4 x - 4 + 5e- x )3
Next, we need to think about the interval of validity. In this case all we need to worry about it is division by zero issues and using some form of computational aid (such as Maple or Mathematica) we will see that the denominator of our solution is never zero and so this solution will be valid for all real numbers. Here is a graph of the solution.
2007 Paul Dawkins
60
To this point weve only worked examples in which n was an integer (positive and negative) and so we should work a quick example where n is not an integer.
Example 4 Solve the following IVP and find the interval of validity for the solution. y y + - y = 0 y (1) = 0 x
Solution Lets first get the differential equation into proper form.
1 1 y = y2 x 1
y +
y 2 y +
1 1 y2 = 1 x
The substitution is then,
v= y
1 2
v =
1 -1 y 2 y 2
1
Now plug the substitution into the differential equation to get,
1 2v + v = 1 x
v +
1 1 v= 2x 2
m ( x) = x2
As weve done with the previous examples weve done some rearranging and given the integrating factor needed for solving the linear differential equation. Solving this gives us,
v ( x) = 1 3 x + cx
In terms of y this is,
1
1 2
y2 = 1 3 x + cx
1 2
0= 1 3 +c
c=-1 3
3
Plugging in for c and solving for y gives us the solution.

1 - x3 - 2 x 2 + 1 1 2 y ( x) = 1 x x = 3 3 9x 2
Note that we multiplied everything out and converted all the negative exponents to positive exponents to make the interval of validity clear here. Because of the root (in the second term in the numerator) and the x in the denominator we can see that we need to require x > 0 in order for the solution to exist and it will exist for all positive xs and so this is also the interval of validity. Here is the graph of the solution.
2007 Paul Dawkins
61
2007 Paul Dawkins
62
Substitutions
In the previous section we looked at Bernoulli Equations and saw that in order to solve them we needed to use the substitution v = y1- n . Upon using this substitution we were able to convert the differential equation into a form that we could deal with (linear in this case). In this section we want to take a look at a couple of other substitutions that can be used to reduce some differential equations down to a solvable form. The first substitution well take a look at will require the differential equation to be in the form,
y y = F x
First order differential equations that can be written in this form are called homogeneous differential equations. Note that we will usually have to do some rewriting in order to put the differential equation into the proper form. Once we have verified that the differential equation is a homogeneous differential equation and weve gotten it written in the proper form we will use the following substitution.
v ( x) =
We can then rewrite this as,
y x
y = xv y = v + xv
and then remembering that both y and v are functions of x we can use the chain rule (recall that is implicit differentiation from Calculus I) to compute,
Under this substitution the differential equation is then,
v + xv = F ( v )
xv = F ( v ) - v
dv dx = F (v) - v x
As we can see with a small rewrite of the new differential equation we will have a separable differential equation after the substitution. Lets take a quick look at a couple of examples of this kind of substitution.
Example 1 Solve the following IVP and find the interval of validity for the solution. x y y + 4 x 2 + y 2 = 0 y ( 2 ) = -7, x>0
Solution Lets first divide both sides by x 2 to rewrite the differential equation as follows,
y y2 y y = -4 - 2 = -4 - x x x
Now, this is not in the officially proper form as we have listed above, but we can see that everywhere the variables are listed they show up as the ratio, y/x and so this is really all the farther that we need to go. So, lets plug the substitution into this form of the differential equation to get,
v ( v + xv ) = -4 - v 2
Next, rewrite the differential equation to get everything separated out.
v x v = -4 - 2v 2 4 + 2v 2 v v 1 dv = - dx 2 4 + 2v x x v = Integrating both sides gives,

1 4
ln ( 4 + 2v 2 ) = - ln ( x ) + c
We need to do a little rewriting using basic logarithm properties in order to be able to easily solve this for v.
ln ( 4 + 2v 2 ) 4 = ln ( x ) + c
-1
Now exponentiate both sides and do a little rewriting
( 4 + 2v )
1 2 4
=e
ln ( x ) + c
-1
= ec e
ln ( x )
-1
c x
Note that because c is an unknown constant then so is e c and so we may as well just call this c as we did above. Finally, lets solve for v and then plug the substitution back in and well play a little fast and loose with constants again.
c4 c 4 + 2v = 4 = 4 x x c v2 = 1 - 4 2 4 x 2 4 c - 4x y =1 2 2 4 x x
2 4 c - 4x4 2 c - 4x y2 = 1 x = 2 4 2x2 x
At this point it would probably be best to go ahead and apply the initial condition. Doing that gives,
2007 Paul Dawkins
64
49 =
c - 4 (16 ) 2 ( 4)
c = 456
Note that we could have also converted the original initial condition into one for in terms of v and then applied it upon solving the separable differential equation. In this case however, it was probably a little easier to do it in terms of y given all the logarithms in the solution to the separable differential equation. Finally, plug in c and solve for y to get,
y2 =
228 - 2 x 4 x2
y ( x) =
228 - 2 x 4 x2
The initial condition tells us that the must be the correct sign and so the actual solution is,
y ( x) = -
228 - 2 x 4 x2
For the interval of validity we can see that we need to avoid x = 0 and because we cant allow negative numbers under the square root we also need to require that,
228 - 2 x 4 0 x 4 114 -3.2676 x < 0 -3.2676 x 3.2676
So, we have two possible intervals of validity,
0 < x 3.2676 and the initial condition tells us that it must be 0 < x 3.2676 .
The graph of the solution is,
2007 Paul Dawkins
65
Example 2 Solve the following IVP and find the interval of validity for the solution. x y = y ( ln x - ln y ) y (1) = 4, x>0
Solution On the surface this differential equation looks like it wont be homogeneous. However, with a quick logarithm property we can rewrite this as,
y =
y x ln x y
In this form the differential equation is clearly homogeneous. Applying the substitution and separating gives,
1 v + xv = v ln v 1 xv = v ln - 1 v v ( ln ( v ) - 1)
1
dv
dx x
Integrate both sides and do a little rewrite to get,
- ln ( ln ( 1 v ) - 1) = ln x + c ln ( ln ( 1 v ) - 1) = c - ln x
You were able to do the integral on the left right? It used the substitution u = ln ( 1 v ) -1 . Now, solve for v and note that well need to exponentiate both sides a couple of times and play fast and loose with constants again.
1 ln ( v ) -1 = e ln ( x ) + c
-1
= ec e
ln ( x )
-1
c x
ln ( 1 v)=
c +1 x v=e
-
c +1 1 =ex v
c -1 x
Plugging the substitution back in and solving for y gives,

c - -1 y =e x x
y ( x ) = xe
c -1 x
4 = e - c -1
2007 Paul Dawkins
c = - (1 + ln 4 )
66
y ( x ) = xe
1+ ln 4
-1
We clearly need to avoid x = 0 to avoid division by zero and so with the initial condition we can see that the interval of validity is x > 0 . The graph of the solution is,
For the next substitution well take a look at well need the differential equation in the form,
y = G ( ax + by )
In these cases well use the substitution,
v = ax + by
v = a + by
Plugging this into the differential equation gives,

b
1
( v - a ) = G ( v )
v = a + bG ( v )
dv = dx a + bG ( v )
So, with this substitution well be able to rewrite the original differential equation as a new separable differential equation that we can solve. Lets take a look at a couple of examples.
Example 3 Solve the following IVP and find the interval of validity for the solution. 2 y - ( 4 x - y + 1) = 0 y (0) = 2
Solution In this case well use the substitution.
v = 4x - y
v = 4 - y
Note that we didnt include the +1 in our substitution. Usually only the ax + by part gets included in the substitution. There are times where including the extra constant may change the
difficulty of the solution process, either easier or harder, however in this case it doesnt really make much difference so we wont include it in our substitution. So, plugging this into the differential equation gives,
4 - v - ( v + 1) = 0
2
v = 4 - ( v + 1) dv
( v + 1)
-4
= - dx
As weve shown above we definitely have a separable differential equation. Also note that to help with the solution process we left a minus sign on the right side. Well need to integrate both sides and in order to do the integral on the left well need to use partial fractions. Well leave it to you to fill in the missing details and given that well be doing quite a bit of partial fraction work in a few chapters you should really make sure that you can do the missing details.
dv dv = = - dx 2 v + 2v - 3 ( v + 3)( v - 1) 1 1 1 dv = - dx 4 v -1 v + 3 1 4 ( ln ( v - 1) - ln ( v + 3 ) ) = - x + c v -1 ln = c - 4x v+3
Note that we played a little fast and loose with constants above. The next step is fairly messy but needs to be done and that is to solve for v and note that well be playing fast and loose with constants again where we can get away with it and well be skipping a few steps that you shouldnt have any problem verifying.
v - 1 c -4 x =e = c e -4 x v+3 v - 1 = c e -4 x ( v + 3 )
v (1 - c e -4 x ) = 1 + 3c e -4 x
At this stage we should back away a bit and note that we cant play fast and loose with constants anymore. We were able to do that in first step because the c appeared only once in the equation. At this point however the c appears twice and so weve got to keep them around. If we absorbed the 3 into the c on the right the new c would be different from the c on the left because the c on the left didnt have the 3 as well. So, lets solve for v and then go ahead and go back into terms of y.
2007 Paul Dawkins
68
v= 4x - y =
1 + 3c e -4 x 1 - c e -4 x 1 + 3c e -4 x 1 - c e -4 x 1 + 3c e-4 x 1 - c e -4 x
y ( x) = 4x -
The last step is to then apply the initial condition and solve for c.
2 = y (0) = The solution is then,
1 + 3c 1- c
c = -3
y ( x) = 4x -
1 - 9 e -4 x 1 + 3 e -4 x
Note that because exponentials exist everywhere and the denominator of the second term is always positive (because exponentials are always positive and adding a positive one onto that wont change the fact that its positive) the interval of validity for this solution will be all real numbers. Here is a graph of the solution.
Example 4 Solve the following IVP and find the interval of validity for the solution. y = e9 y - x y (0) = 0
Solution Here is the substitution that well need for this example.
v = 9y - x
v = 9 y - 1
Plugging this into our differential equation gives,
2007 Paul Dawkins
69
1 9
( v + 1) = ev
v = 9ev - 1 dv = dx 9ev - 1 e- v dv = dx 9 - e- v
Note that we did a little rewrite on the separated portion to make the integrals go a little easier. By multiplying the numerator and denominator by e - v we can turn this into a fairly simply substitution integration problem. So, upon integrating both sides we get,
ln ( 9 - e - v ) = x + c
Solving for v gives,
9 - e - v = e ce x = ce x e - v = 9 - ce x v = - ln ( 9 - ce x )
Plugging the substitution back in and solving for y gives us,
x y ( x) = 1 9 x - ln ( 9 - ce )
)
c =8
Next, apply the initial condition and solve for c.
0 = y (0) = - 1 9 ln ( 9 - c )
x y ( x) = 1 9 x - ln ( 9 - 8e )
)
x < ln 9 8 = 0.1178
Now, for the interval of validity we need to make sure that we only take logarithms of positive numbers as well need to require that,
9 - 8e x > 0
ex <
9 8
2007 Paul Dawkins
70
In both this section and the previous section weve seen that sometimes a substitution will take a differential equation that we cant solve and turn it into one that we can solve. This idea of substitutions is an important idea and should not be forgotten. Not every differential equation can be made easier with a substitution and there is no way to show every possible substitution but remembering that a substitution may work is a good thing to do. If you get stuck on a differential equation you may try to see if a substitution if some kind will work for you.
2007 Paul Dawkins
71
Intervals of Validity
Ive called this section Intervals of Validity because all of the examples will involve them. However, there is a lot more to this section. We will see a couple of theorems that will tell us when we can solve a differential equation. We will also see some of the differences between linear and nonlinear differential equations. First let's take a look at a theorem about linear first order differential equations. This is a very important theorem although were not going to really use it for its most important aspect. Theorem 1 Consider the following IVP.
y ( t0 ) = y0 If p(t) and g(t) are continuous functions on an open interval a < t < b and the interval contains
to, then there is a unique solution to the IVP on that interval. So, just what does this theorem tell us? First, it tells us that for nice enough linear first order differential equations solutions are guaranteed to exist and more importantly the solution will be unique. We may not be able to find the solution, but do know that it exists and that there will only be one of them. This is the very important aspect of this theorem. Knowing that a differential equation has a unique solution is sometimes more important than actually having the solution itself! Next, if the interval in the theorem is the largest possible interval on which p(t) and g(t) are continuous then the interval is the interval of validity for the solution. This means, that for linear first order differential equations, we won't need to actually solve the differential equation in order to find the interval of validity. Notice as well that the interval of validity will depend only partially on the initial condition. The interval must contain to, but the value of yo, has no effect on the interval of validity. Lets take a look at an example.
y + p ( t ) y = g ( t )
Example 1 Without solving, determine the interval of validity for the following initial value
problem.
(t
- 9 ) y + 2 y = ln 20 - 4t
y ( 4 ) = -3
Solution First, in order to use the theorem to find the interval of validity we must write the differential equation in the proper form given in the theorem. So we will need to divide out by the coefficient of the derivative.
y +
ln 20 - 4t 2 y= t -9 t2 - 9
2
Next, we need to identify where the two functions are not continuous. This will allow us to find all possible intervals of validity for the differential equation. So, p(t) will be discontinuous at t = 3 since these points will give a division by zero. Likewise, g(t) will also be discontinuous at t = 3 as well as t = 5 since at this point we will have the natural logarithm of zero. Note that in
this case we won't have to worry about natural log of negative numbers because of the absolute values. Now, with these points in hand we can break up the real number line into four intervals where both p(t) and g(t) will be continuous. These four intervals are,
- < t < -3
-3 < t < 3
3< t < 5
5<t <
The endpoints of each of the intervals are points where at least one of the two functions is discontinuous. This will guarantee that both functions are continuous everywhere in each interval. Finally, let's identify the actual interval of validity for the initial value problem. The actual interval of validity is the interval that will contain to = 4. So, the interval of validity for the initial value problem is.
3<t <5
In this last example we need to be careful to not jump to the conclusion that the other three intervals cannot be intervals of validity. By changing the initial condition, in particular the value of to, we can make any of the four intervals the interval of validity. The first theorem required a linear differential equation. There is a similar theorem for non-linear first order differential equations. This theorem is not as useful for finding intervals of validity as the first theorem was so we wont be doing all that much with it. Here is the theorem. Theorem 2 Consider the following IVP.
y = f ( t , y )
y ( t0 ) = y0
If f(t,y) and
f are continuous functions in some rectangle a < t < b , g < y < d containing the y
point (to, yo) then there is a unique solution to the IVP in some interval to h < t < to + h that is contained in a < t < b . Thats it. Unlike the first theorem, this one cannot really be used to find an interval of validity. So, we will know that a unique solution exists if the conditions of the theorem are met, but we will actually need the solution in order to determine its interval of validity. Note as well that for non-linear differential equations it appears that the value of y0 may affect the interval of validity. Here is an example of the problems that can arise when the conditions of this theorem arent met.
Example 2 Determine all possible solutions to the following IVP. 1 y = y 3 y (0) = 0

Solution First, notice that this differential equation does NOT satisfy the conditions of the theorem.
f ( y) = y3
1
df 1 = 2 dy 3 y 3
2007 Paul Dawkins
So, the function is continuous on any interval, but the derivative is not continuous at y = 0 and so will not be continuous at any interval containing y = 0. In order to use the theorem both must be continuous on an interval that contains yo = 0 and this is problem for us since we do have yo = 0. Now, lets actually work the problem. This differential equation is separable and is fairly simple to solve.
y 3 dy = dt
3 2 y3 = t + c 2
Applying the initial condition gives c = 0 and so the solution is.
-1
3 2 y3 = t 2 2 2 y3 = t 3 2 y = t 3
2 3
2 2 y (t ) = t 3
So, weve got two possible solutions here, both of which satisfy the differential equation and the initial condition. There is also a third solution to the IVP. y(t) = 0 is also a solution to the differential equation and satisfies the initial condition. In this last example we had a very simple IVP and it only violated one of the conditions of the theorem, yet it had three different solutions. All the examples weve worked in the previous sections satisfied the conditions of this theorem and had a single unique solution to the IVP. This example is a useful reminder of the fact that, in the field of differential equations, things dont always behave nicely. Its easy to forget this as most of the problems that are worked in a differential equations class are nice and behave in a nice, predictable manner. Lets work one final example that will illustrate one of the differences between linear and nonlinear differential equations.
Example 3 Determine the interval of validity for the initial value problem below and give its
dependence on the value of yo
y = y 2
y ( 0 ) = y0
Solution Before proceeding in this problem, we should note that the differential equation is non-linear and meets both conditions of the Theorem 2 and so there will be a unique solution to the IVP for each possible value of yo. Also, note that the problem asks for any dependence of the interval of validity on the value of yo. This immediately illustrates a difference between linear and non-linear differential equations.
Intervals of validity for linear differential equations do not depend on the value of yo. Intervals of validity for non-linear differential can depend on the value of yo as we pointed out after the second theorem. So, lets solve the IVP and get some intervals of validity. First note that if yo = 0 then y(t) = 0 is the solution and this has an interval of validity of
- < t <
So for the rest of the problem let's assume that y0 0 . Now, the differential equation is separable so let's solve it and get a general solution.
y
-2
dy = dt
1 =t+c y 1 y0
c=The solution is then.
1 1 =ty y0
1 y0
y (t ) = y (t ) =
1 -t
y0 1 - y 0t 1 y0 1 y0
Now that we have a solution to the initial value problem we can start finding intervals of validity. From the solution we can see that the only problem that well have is division by zero at
t=
This leads to two possible intervals of validity.
- < t <
1 <t < y0
That actual interval of validity will be the interval that contains to = 0. This however, depends on the value of yo. If yo < 0 then y10 < 0 and so the second interval will contain to = 0. Likewise if yo > 0 then
1 y0
> 0 and in this case the first interval will contain to = 0.
This leads to the following possible intervals of validity, depending on the value of yo.
If y0 > 0 If y0 = 0 If y0 < 0
- < t <
1 is the interval of validity. y0
- < t < is the interval of validity. 1 < t < is the interval of validity. y0
On a side note, notice that the solution, in its final form, will also work if yo = 0. So what did this example show us about the difference between linear and non-linear differential equations? First, as pointed out in the solution to the example, intervals of validity for non-linear differential equations can depend on the value of yo, whereas intervals of validity for linear differential equations dont. Second, intervals of validity for linear differential equations can be found from the differential equation with no knowledge of the solution. This is definitely not the case with non-linear differential equations. It would be very difficult to see how any of these intervals in the last example could be found from the differential equation. Knowledge of the solution was required in order for us to find the interval of validity.
2007 Paul Dawkins
76
Modeling with First Order Differential Equations

We now move into one of the main applications of differential equations both in this class and in general. Modeling is the process of writing a differential equation to describe a physical situation. Almost all of the differential equations that you will use in your job (for the engineers out there in the audience) are there because somebody, at some time, modeled a situation to come up with the differential equation that you are using. This section is not intended to completely teach you how to go about modeling all physical situations. A whole course could be devoted to the subject of modeling and still not cover everything! This section is designed to introduce you to the process of modeling and show you what is involved in modeling. We will look at three different situations in this section : Mixing Problems, Population Problems, and Falling Bodies. In all of these situations we will be forced to make assumptions that do not accurately depict reality in most cases, but without them the problems would be very difficult and beyond the scope of this discussion (and the course in most cases to be honest). So lets get started. Mixing Problems In these problems we will start with a substance that is dissolved in a liquid. Liquid will be entering and leaving a holding tank. The liquid entering the tank may or may not contain more of the substance dissolved in it. Liquid leaving the tank will of course contain the substance dissolved in it. If Q(t) gives the amount of the substance dissolved in the liquid in the tank at any time t we want to develop a differential equation that, when solved, will give us an expression for Q(t). Note as well that in many situations we can think of air as a liquid for the purposes of these kinds of discussions and so we dont actually need to have an actual liquid, but could instead use air as the liquid. The main assumption that well be using here is that the concentration of the substance in the liquid is uniform throughout the tank. Clearly this will not be the case, but if we allow the concentration to vary depending on the location in the tank the problem becomes very difficult and will involve partial differential equations, which is not the focus of this course. The main equation that well be using to model this situation is : Rate of change of Q(t) where, Rate of change of Q(t) = Rate at which Q(t) = enters the tank Rate at which Q(t) exits the tank
dQ = Q ( t ) dt
Rate at which Q(t) enters the tank = (flow rate of liquid entering) x (concentration of substance in liquid entering) Rate at which Q(t) exits the tank = (flow rate of liquid exiting) x (concentration of substance in liquid exiting) Lets take a look at the first problem.
Example 1 A 1500 gallon tank initially contains 600 gallons of water with 5 lbs of salt
dissolved in it. Water enters the tank at a rate of 9 gal/hr and the water entering the tank has a salt concentration of 1 5 1 + cos ( t ) lbs/gal. If a well mixed solution leaves the tank at a rate of 6
gal/hr, how much salt is in the tank when it overflows? Solution First off, lets address the well mixed solution bit. This is the assumption that was mentioned earlier. We are going to assume that the instant the water enters the tank it somehow instantly disperses evenly throughout the tank to give a uniform concentration of salt in the tank at every point. Again, this will clearly not be the case in reality, but it will allow us to do the problem. Now, to set up the IVP that well need to solve to get Q(t) well need the flow rate of the water entering (weve got that), the concentration of the salt in the water entering (weve got that), the flow rate of the water leaving (weve got that) and the concentration of the salt in the water exiting (we dont have this yet). So, we first need to determine the concentration of the salt in the water exiting the tank. Since we are assuming a uniform concentration of salt in the tank the concentration at any point in the tank and hence in the water exiting is given by,
Concentration =
Amount of salt in the tank at any time, t Volume of water in the tank at any time, t
The amount at any time t is easy its just Q(t). The volume is also pretty easy. We start with 600 gallons and every hour 9 gallons enters and 6 gallons leave. So, if we use t in hours, every hour 3 gallons enters the tank, or at any time t there is 600 + 3t gallons of water in the tank. So, the IVP for this situation is,
Q (t ) 1 Q ( t ) = ( 9 ) (1 + cos ( t ) ) - ( 6 ) 5 600 + 3t Q ( t ) = 2Q ( t ) 9 1 + cos ( t ) ) ( 5 200 + t
Q (0) = 5 Q (0) = 5
This is a linear differential equation and it isnt too difficult to solve (hopefully). We will show most of the details, but leave the description of the solution process out. If you need a refresher on solving linear first order differential equations go back and take a look at that section.
Q ( t ) +
2Q ( t ) 9 = (1 + cos ( t ) ) 200 + t 5
m (t ) = e
2 dt 200 + t
= e 2ln ( 200+t ) = ( 200 + t )
9 2 2 ( 200 + t ) Q ( t ) dt = ( 200 + t ) (1 + cos ( t ) ) dt 5
2007 Paul Dawkins
78
( 200 + t )
91 3 2 Q ( t ) = ( 200 + t ) + ( 200 + t ) sin ( t ) + 2 ( 200 + t ) cos ( t ) - 2sin ( t ) + c 53 2cos ( t ) 2sin ( t ) c 91 Q ( t ) = ( 200 + t ) + sin ( t ) + + 2 53 200 + t ( 200 + t ) ( 200 + t )2
So, heres the general solution. Now, apply the initial condition to get the value of the constant, c.
c 91 2 5 = Q ( 0 ) = ( 200 ) + + 53 200 ( 200 )2

So, the amount of salt in the tank at any time t is.
c = -4600720
2cos ( t ) 2sin ( t ) 4600720 91 Q ( t ) = ( 200 + t ) + sin ( t ) + 2 ( 200 + t )2 + 5 3 200 t + 200 t ( )

Now, the tank will overflow at t = 300 hrs. The amount of salt in the tank at that time is.
Q ( 300 ) = 279.797 lbs
Heres a graph of the salt in the tank before it overflows.
Note that the whole graph should have small oscillations in it as you can see in the range from 200 to 250. The scale of the oscillations however was small enough that the program used to generate the image had trouble showing all of them. The work was a little messy with that one, but they will often be that way so dont get excited about it. This first example also assumed that nothing would change throughout the life of the process. That, of course will usually not be the case. Lets take a look at an example where something changes in the process.
2007 Paul Dawkins
79
Example 2 A 1000 gallon holding tank that catches runoff from some chemical process initially
has 800 gallons of water with 2 ounces of pollution dissolved in it. Polluted water flows into the tank at a rate of 3 gal/hr and contains 5 ounces/gal of pollution in it. A well mixed solution leaves the tank at 3 gal/hr as well. When the amount of pollution in the holding tank reaches 500 ounces the inflow of polluted water is cut off and fresh water will enter the tank at a decreased rate of 2 gallons while the outflow is increased to 4 gal/hr. Determine the amount of pollution in the tank at any time t. Solution Okay, so clearly the pollution in the tank will increase as time passes. If the amount of pollution ever reaches the maximum allowed there will be a change in the situation. This will necessitate a change in the differential equation describing the process as well. In other words, well need two IVPs for this problem. One will describe the initial situation when polluted runoff is entering the tank and one for after the maximum allowed pollution is reached and fresh water is entering the tank. Here are the two IVPs for this problem.
Q (t ) Q1 ( t ) = ( 3)( 5 ) - ( 3) 1 800 Q2 ( t ) Q2 ( t ) = ( 2 )( 0 ) - ( 4 ) 800 - 2 ( t - tm )
Q1 ( 0 ) = 2
0 t tm
Q2 ( tm ) = 500 tm t te
The first one is fairly straight forward and will be valid until the maximum amount of pollution is reached. Well call that time tm. Also, the volume in the tank remains constant during this time so we dont need to do anything fancy with that this time in the second term as we did in the previous example. Well need a little explanation for the second one. First notice that we dont start over at t = 0. We start this one at tm, the time at which the new process starts. Next, fresh water is flowing into the tank and so the concentration of pollution in the incoming water is zero. This will drop out the first term, and thats okay so dont worry about that. Now, notice that the volume at any time looks a little funny. During this time frame we are losing two gallons of water every hour of the process so we need the -2 in there to account for that. However, we cant just use t as we did in the previous example. When this new process starts up there needs to be 800 gallons of water in the tank and if we just use t there we wont have the required 800 gallons that we need in the equation. So, to make sure that we have the proper volume we need to put in the difference in times. In this way once we are one hour into the new process (i.e t - tm = 1) we will have 798 gallons in the tank as required. Finally, the second process cant continue forever as eventually the tank will empty. This is denoted in the time restrictions as te. We can also note that te = tm + 400 since the tank will empty 400 hours after this new process starts up. Well, it will end provided something doesnt come along and start changing the situation again. Okay, now that weve got all the explanations taken care of heres the simplified version of the IVPs that well be solving.
2007 Paul Dawkins
80
3Q1 ( t ) 800 2Q2 ( t ) Q2 ( t ) = 400 - ( t - tm ) Q1 ( t ) = 15 -
Q1 ( 0 ) = 2
0 t tm
Q2 ( tm ) = 500 tm t te
The first IVP is a fairly simple linear differential equation so well leave the details of the solution to you to check. Upon solving you get.
Q1 ( t ) = 4000 - 3998e
- 3t 800
Now, we need to find tm. This isnt too bad all we need to do is determine when the amount of pollution reaches 500. So we need to solve.
Q1 ( t ) = 4000 - 3998e
- 3t 800
= 500
tm = 35.475
So, the second process will pick up at 35.475 hours. For completeness sake here is the IVP with this information inserted.
Q2 ( t ) = -
2Q2 ( t ) 435.475 - t
Q2 ( 35.475) = 500
35.475 t 435.475
This differential equation is both linear and separable and again isnt terribly difficult to solve so Ill leave the details to you again to check that we should get.
Q2 ( t )
( 435.476 - t ) =
320
So, a solution that encompasses the complete running time of the process is
3t 4000 - 3998e - 800 2 Q (t ) = ( 435.476 - t ) 320
0 t 35.475 35.475 t 435.4758
Here is a graph of the amount of pollution in the tank at any time t.
2007 Paul Dawkins
81
As you can surely see, these problems can get quite complicated if you want them to. Take the last example. A more realistic situation would be that once the pollution dropped below some predetermined point the polluted runoff would, in all likelihood, be allowed to flow back in and then the whole process would repeat itself. So, realistically, there should be at least one more IVP in the process. Lets move on to another type of problem now. Population These are somewhat easier than the mixing problems although, in some ways, they are very similar to mixing problems. So, if P(t) represents a population in a given region at any time t the basic equation that well use is identical to the one that we used for mixing. Namely, Rate at Rate at Rate of which P(t) which P(t) change of = enters the exits the P(t) region region Here the rate of change of P(t) is still the derivative. Whats different this time is the rate at which the population enters and exits the region. For population problems all the ways for a population to enter the region are included in the entering rate. Birth rate and migration into the region are examples of terms that would go into the rate at which the population enters the region. Likewise, all the ways for a population to leave an area will be included in the exiting rate. Therefore things like death rate, migration out and predation are examples of terms that would go into the rate at which the population exits the area. Heres an example.
Example 3 A population of insects in a region will grow at a rate that is proportional to their
current population. In the absence of any outside factors the population will triple in two weeks time. On any given day there is a net migration into the area of 15 insects and 16 are eaten by the local bird population and 7 die of natural causes. If there are initially 100 insects in the area will the population survive? If not, when do they die out? Solution Lets start out by looking at the birth rate. We are told that the insects will be born at a rate that is proportional to the current population. This means that the birth rate can be written as where r is a positive constant that will need to be determined. Now, lets take everything into account and get the IVP for this problem.
rP
P = ( rP + 15 ) - (16 + 7 ) P = rP - 8
P ( 0 ) = 100
P ( 0 ) = 100
Note that in the first line we used parenthesis to note which terms went into which part of the differential equation. Also note that we dont make use of the fact that the population will triple in two weeks time in the absence of outside factors here. In the absence of outside factors means that the ONLY thing that we can consider is birth rate. Nothing else can enter into the picture and clearly we have other influences in the differential equation.
2007 Paul Dawkins
82
So, just how does this tripling come into play? Well, we should also note that without knowing r we will have a difficult time solving the IVP completely. We will use the fact that the population triples in two week time to help us find r. In the absence of outside factors the differential equation would become.
P = rP
P ( 0 ) = 100
P (14 ) = 300
Note that since we used days as the time frame in the actual IVP I needed to convert the two weeks to 14 days. We could have just as easily converted the original IVP to weeks as the time frame, in which case there would have been a net change of 56 per week instead of the 8 per week that we are currently using in the original differential equation. Okay back to the differential equation that ignores all the outside factors. This differential equation is separable and linear and is a simple differential equation to solve. Ill leave the detail to you to get the general solution.
P ( t ) = ce rt
Applying the initial condition gives c = 100. Now apply the second condition.
300 = P (14 ) = 100e14 r
300 = 100e14r
We need to solve this for r. First divide both sides by 100, then take the natural log of both sides.
3 = e14 r ln 3 = ln e14 r ln 3 = 14r ln 3 r= 14

g x We made use of the fact that ln e ( ) = g ( x ) here to simplify the problem. Now, that we have r
we can go back and solve the original differential equation. Well rewrite it a little for the solution process.
P -
ln 3 P = -8 14
P ( 0 ) = 100
This is a fairly simple linear differential equation, but that coefficient of P always get people bent out of shape, so well go through at least some of the details here.
m (t ) = e
- ln3 dt 14
=e
- ln 3 t 14
Now, dont get excited about the integrating factor here. Its just like e 2 t only this time the constant is a little more complicated than just a 2, but it is a constant! Now, solve the differential equation.
2007 Paul Dawkins
83
- ln 3 t - ln 3 t 14 dt = -8e 14 dt Pe 3t - ln 3 t 14 - ln 14 Pe 14 = -8 +c e ln 3 ln 3 t 112 P (t ) = + ce 14 ln 3
Again, do not get excited about doing the right hand integral, its just like integrating e 2 t ! Applying the initial condition gives the following.
P (t ) =
3t ln 3 t 112 112 ln 112 14 + 100 - 1.94679e 14 e = ln 3 ln 3 ln 3
Now, the exponential has a positive exponent and so will go to plus infinity as t increases. Its coefficient, however, is negative and so the whole population will go negative eventually. Clearly, population cant be negative, but in order for the population to go negative it must pass through zero. In other words, eventually all the insects must die. So, they dont survive and we can solve the following to determine when they die out.
0=
ln 3 t 112 - 1.94679e 14 ln 3
t = 50.4415days
So, the insects will survive for around 7.2 weeks. Here is a graph of the population during the time in which they survive.
As with the mixing problems, we could make the population problems more complicated by changing the circumstances at some point in time. For instance, if at some point in time the local bird population saw a decrease due to disease they wouldnt eat as much after that point and a second differential equation to govern the time after this point. Lets now take a look at the final type of problem that well be modeling in this section.
2007 Paul Dawkins
84
Falling Body This will not be the first time that weve looked into falling bodies. If you recall, we looked at one of these when we were looking at Direction Fields. In that section we saw that the basic equation that well use is Newtons Second Law of Motion.
mv = F ( t , v )
The two forces that well be looking at here are gravity and air resistance. The main issue with these problems is to correctly define conventions and then remember to keep those conventions. By this we mean define which direction will be termed the positive direction and then make sure that all your forces match that convention. This is especially important for air resistance as this is usually dependent on the velocity and so the sign of the velocity can and does affect the sign of the air resistance force. Lets take a look at an example.
Example 4 A 50 kg mass is shot from a cannon straight up with an initial velocity of 10m/s off
a bridge that is 100 meters above the ground. If air resistance is given by 5v determine the velocity of the mass when it hits the ground. Solution First, notice that when we say straight up, we really mean straight up, but in such a way that it will miss the bridge on the way back down. Here is a sketch of the situation.
Notice the conventions that we set up for this problem. Since the vast majority of the motion will be in the downward direction we decided to assume that everything acting in the downward direction should be positive. Note that we also defined the zero position as the bridge, which makes the ground have a position of 100. Okay, if you think about it we actually have two situations here. The initial phase in which the mass is rising in the air and the second phase when the mass is on its way down. We will need to examine both situations and set up an IVP for each. We will do this simultaneously. Here are the forces that are acting on the object on the way up and on the way down.
2007 Paul Dawkins
85
Notice that the air resistance force needs a negative in both cases in order to get the correct sign or direction on the force. When the mass is moving upwards the velocity (and hence v) is negative, yet the force must be acting in a downward direction. Therefore, the - must be part of the force to make sure that, overall, the force is positive and hence acting in the downward direction. Likewise, when the mass is moving downward the velocity (and so v) is positive. Therefore, the air resistance must also have a - in order to make sure that its negative and hence acting in the upward direction. So, the IVP for each of these situations are.
Up mv = mg - 5v v ( 0 ) = -10
Down mv = mg - 5v v ( t0 ) = 0
In the second IVP, the t0 is the time when the object is at the highest point and is ready to start on the way down. Note that at this time the velocity would be zero. Also note that the initial condition of the first differential equation will have to be negative since the initial velocity is upward. In this case, the differential equation for both of the situations is identical. This wont always happen, but in those cases where it does, we can ignore the second IVP and just let the first govern the whole process. So, lets actually plug in for the mass and gravity (well be using g = 9.8 m/s2 here). Well go ahead and divide out the mass while were at it since well need to do that eventually anyway.
v = 9.8 -
5v v = 9.8 50 10
v ( 0 ) = -10
This is a simple linear differential equation to solve so well leave the details to you. Upon solving we arrive at the following equation for the velocity of the object at any time t.
v ( t ) = 98 - 108e
t - 10
Okay, we want the velocity of the ball when it hits the ground. Of course we need to know when it hits the ground before we can ask this. In order to find this we will need to find the position function. This is easy enough to do.
s ( t ) = v ( t ) dt = 98 - 108e
t - 10
dt = 98t + 1080e
t - 10
+c
We can now use the fact that I took the convention that s(0) = 0 to find that c = -1080. The position at any time is then.
s ( t ) = 98t + 1080e
t - 10
t - 10
- 1080
To determine when the mass hits the ground we just need to solve.
100 = 98t + 1080e
- 1080
t = -3.32203, 5.98147
Weve got two solutions here, but since we are starting things at t = 0, the negative is clearly the incorrect value. Therefore the mass hits the ground at t = 5.98147. The velocity of the object upon hitting the ground is then.
v ( 5.98147 ) = 38.61841
This last example gave us an example of a situation where the two differential equations needed for the problem ended up being identical and so we didnt need the second one after all. Be careful however to not always expect this. We could very easily change this problem so that it required two different differential equations. For instance we could have had a parachute on the mass open at the top of its arc changing its air resistance. This would have completely changed the second differential equation and forced us to use it as well. Or, we could have put a river under the bridge so that before it actually hit the ground it would have first had to go through some water which would have a different air resistance for that phase necessitating a new differential equation for that portion. Or, we could be really crazy and have both the parachute and the river which would then require three IVPs to be solved before we determined the velocity of the mass before it actually hits the solid ground. Before leaving this section lets work a couple examples illustrating the importance of remembering the conventions that you set up for the positive direction in these problems. Awhile back I gave my students a problem in which a sky diver jumps out of a plane. Most of my students are engineering majors and following the standard convention from most of their engineering classes they defined the positive direction as upward, despite the fact that all the motion in the problem was downward. There is nothing wrong with this assumption, however, because they forgot the convention that up was positive they did not correctly deal with the air resistance which caused them to get the incorrect answer. So, lets take a look at the problem and set up the IVP that will give the sky divers velocity at any time t.
Example 5 Set up the IVP that will give the velocity of a 60 kg sky diver that jumps out of a plane with no initial velocity and an air resistance of 0.8 v . For this example assume that the
positive direction is upward. Solution Here are the forces that are acting on the sky diver
2007 Paul Dawkins
87
Because of the conventions the force due to gravity is negative and the force due to air resistance is positive. As set up, these forces have the correct sign and so the IVP is
mv = - mg + 0.8 v
v (0) = 0
The problem arises when you go to remove the absolute value bars. In order to do the problem they do need to be removed. This is where most of the students made their mistake. Because they had forgotten about the convention and the direction of motion they just dropped the absolute value bars to get.
mv = - mg + 0.8v
v ( 0) = 0
( incorrect IVP!!)
So, why is this incorrect? Well remember that the convention is that positive is upward. However in this case the object is moving downward and so v is negative! Upon dropping the absolute value bars the air resistance became a negative force and hence was acting in the downward direction! To get the correct IVP recall that because v is negative then |v| = -v. Using this, the air resistance becomes FA = -0.8v and despite appearances this is a positive forces since the - cancels out against the velocity (which is negative) to get a positive force. The correct IVP is then
mv = - mg - 0.8v
Plugging in the mass gives
v ( 0) = 0
v = -9.8 -
v 75
v (0) = 0
For the sake of completeness the velocity of the sky diver, at least until the parachute opens, which I didnt include in this problem is.
v ( t ) = -735 + 735e
t - 75
This mistake was made in part because the students were in a hurry and werent paying attention, but also because they simply forgot about their convention and the direction of motion! Dont fall into this mistake. Always pay attention to your conventions and what is happening in the problems. Just to show you the difference here is the problem worked by assuming that down is positive.
2007 Paul Dawkins
88
Example 6 Set up the IVP that will give the velocity of a 60 kg sky diver that jumps out of a plane with no initial velocity and an air resistance of 0.8 v . For this example assume that the
positive direction is downward. Solution Here are the forces that are acting on the sky diver
In this case the force due to gravity is positive since its a downward force and air resistance is an upward force and so needs to be negative. In this case since the motion is downward the velocity is positive so |v| = v. The air resistance is then FA = -0.8v. The IVP for this case is
mv = mg - 0.8v
Plugging in the mass gives
v (0) = 0
v = 9.8 Solving this gives
v 75
v ( 0) = 0
v ( t ) = 735 - 735e
t - 75
This is the same solution as the previous example, except that its got the opposite sign. This is to be expected since the conventions have been switched between the two examples.
2007 Paul Dawkins
89
Equilibrium Solutions
In the previous section we modeled a population based on the assumption that the growth rate would be a constant. However, in reality this doesnt make much sense. Clearly a population cannot be allowed to grow forever at the same rate. The growth rate of a population needs to depend on the population itself. Once a population reaches a certain point the growth rate will start reduce, often drastically. A much more realistic model of a population growth is given by the logistic growth equation. Here is the logistic growth equation.
P P = r 1 - P K
In the logistic growth equation r is the intrinsic growth rate and is the same r as in the last section. In other words, it is the growth rate that will occur in the absence of any limiting factors. K is called either the saturation level or the carrying capacity. Now, we claimed that this was a more realistic model for a population. Lets see if that in fact is correct. To allow us to sketch a direction field lets pick a couple of numbers for r and K. Well use r = 1 2 and K = 10. For these values the logistics equation is.
1 P P = 1 - P 2 10
If you need a refresher on sketching direction fields go back and take a look at that section. First notice that the derivative will be zero at P = 0 and P = 10. Also notice that these are in fact solutions to the differential equation. These two values are called equilibrium solutions since they are constant solutions to the differential equation. Well leave the rest of the details on sketching the direction field to you. Here is the direction field as well as a couple of solutions sketched in as well.
Note, that we included a small portion of negative Ps in here even though they really dont make any sense for a population problem. The reason for this will be apparent down the road. Also, notice that a population of say 8 doesnt make all that much sense so lets assume that population is in thousands or millions so that 8 actually represents 8,000 or 8,000,000 individuals in a population.
Notice that if we start with a population of zero, there is no growth and the population stays at zero. So, the logistic equation will correctly figure out that. Next, notice that if we start with a population in the range 0 < P(0) < 10 then the population will grow, but start to level off once we get close to a population of 10. If we start with a population of 10, the population will stay at 10. Finally if we start with a population that is greater than 10, then the population will actually die off until we start nearing a population of 10, at which point the population decline will start to slow down. Now, from a realistic standpoint this should make some sense. Populations cant just grow forever without bound. Eventually the population will reach such a size that the resources of an area are no longer able to sustain the population and the population growth will start to slow as it comes closer to this threshold. Also, if you start off with a population greater than what an area can sustain there will actually be a die off until we get near to this threshold. In this case that threshold appears to be 10, which is also the value of K for our problem. That should explain the name that we gave K initially. The carrying capacity or saturation level of an area is the maximum sustainable population for that area. So, the logistics equation, while still quite simplistic, does a much better job of modeling what will happen to a population. Now, lets move on to the point of this section. The logistics equation is an example of an autonomous differential equation. Autonomous differential equations are differential equations that are of the form.
dy = f ( y) dt
The only place that the independent variable, t in this case, appears is in the derivative. Notice that if f ( y0 ) = 0 for some value y = y0 then this will also be a solution to the differential equation. These values are called equilibrium solutions or equilibrium points. What we would like to do is classify these solutions. By classify we mean the following. If solutions start near an equilibrium solution will they move away from the equilibrium solution or towards the equilibrium solution? Upon classifying the equilibrium solutions we can then know what all the other solutions to the differential equation will do in the long term simply by looking at which equilibrium solutions they start near. So, just what do I mean by near? Go back to our logistics equation.
1 P P = 1 - P 2 10
As we pointed out there are two equilibrium solutions to this equation P = 0 and P = 10. If we ignore the fact that were dealing with population these point break up the P number line into three distinct regions.
- < P < 0
0 < P < 10
10 < P <
We will say that a solution starts near an equilibrium solution if it starts in a region that is on either side of that equilibrium solution. So solutions that start near the equilibrium solution P = 10 will start in either
0 < P < 10 - < P < 0
OR OR
10 < P < 0 < P < 10
and solutions that start near P = 0 will start in either
For regions that lie between two equilibrium solutions we can think of any solutions starting in that region as starting near either of the two equilibrium solutions as we need to. Now, solutions that start near P = 0 all move away from the solution as t increases. Note that moving away does not necessarily mean that they grow without bound as they move away. It only means that they move away. Solutions that start out greater than P = 0 move away, but do stay bounded as t grows. In fact, they move in towards P = 10. Equilibrium solutions in which solutions that start near them move away from the equilibrium solution are called unstable equilibriums points or unstable equilibrium solutions. So, for our logistics equation, P = 0 is an unstable equilibrium solution. Next, solutions that start near P = 10 all move in toward P = 10 as t increases. Equilibrium solutions in which solutions that start near them move toward the equilibrium solution are called asymptotically stable equilibrium points or asymptotically stable equilibrium solutions. So, P = 10 is an asymptotically stable equilibrium solution. There is one more classification, but Ill wait until we get an example in which this occurs to introduce it. So, lets take a look at a couple of examples.
Example 1 Find and classify all the equilibrium solutions to the following differential equation. y = y 2 - y - 6
Solution First, find the equilibrium solutions. This is generally easy enough to do.
y 2 - y - 6 = ( y - 3)( y + 2 ) = 0
So, it looks like weve got two equilibrium solutions. Both y = -2 and y = 3 are equilibrium solutions. Below is the sketch of some integral curves for this differential equation. A sketch of the integral curves or direction fields can simplify the process of classifying the equilibrium solutions.
From this sketch it appears that solutions that start near y = -2 all move towards it as t increases and so y = -2 is an asymptotically stable equilibrium solution and solutions that start near y = 3
all move away from it as t increases and so y = 3 is an unstable equilibrium solution. This next example will introduce the third classification that we can give to equilibrium solutions.
Example 2 Find and classify the equilibrium solutions of the following differential equation. 2 y = ( y 2 - 4 ) ( y + 1)
Solution The equilibrium solutions are to this differential equation are y = -2, y = 2, and y = -1. Below is the sketch of the integral curves. We included the direction field this time as the integral curves where a little hard to interpret without it.
From this it is clear (hopefully) that y = 2 is an unstable equilibrium solution and y = -2 is an asymptotically stable equilibrium solution. However, y = -1 behaves differently from either of these two. Solutions that start above it move towards y = -1 while solutions that start below y = -1 move away as t increases. In cases where solutions on one side of an equilibrium solution move towards the equilibrium solution and on the other side of the equilibrium solution move away from it we call the equilibrium solution semi-stable. So, y = -1 is a semi-stable equilibrium solution.
2007 Paul Dawkins
93
Eulers Method
Up to this point practically every differential equation that weve been presented with could be solved. The problem with this is that these are the exceptions rather than the rule. The vast majority of first order differential equations cant be solved. In order to teach you something about solving first order differential equations weve had to restrict ourselves down to the fairly restrictive cases of linear, separable, or exact differential equations or differential equations that could be solve with a set of very specific substitutions. Most first order differential equations however fall into none of these categories. In fact even those that are separable or exact cannot always be solved for an explicit solution. Without explicit solutions to these it would be hard to get any information about the solution. So what do we do when faced with a differential equation that we cant solve? The answer depends on what you are looking for. If you are only looking for long term behavior of a solution you can always sketch a direction field. This can be done without too much difficulty for some fairly complex differential equations that we cant solve to get exact solutions. The problem with this approach is that its only really good for getting general trends in solutions and for long term behavior of solutions. There are times when we will need something more. For instance, maybe we need to determine how a specific solution behaves, including some values that the solution will take. There are also a fairly large set of differential equations that are not easy to sketch good direction fields for. In these cases we resort to numerical methods that will allow us to approximate solutions to differential equations. There are many different methods that can be used to approximate solutions to a differential equation and in fact whole classes can be taught just dealing with the various methods. We are going to look at one of the oldest and easiest to use here. This method was originally devised by Euler and is called, oddly enough, Eulers Method. Lets start with a general first order IVP
dy = f (t, y ) dt
y ( t0 ) = y0
(1)
where f(t,y) is a known function and the values in the initial condition are also known numbers. From the second theorem in the Intervals of Validity section we know that if f and fy are continuous function then there is a unique solution to the IVP in some interval surrounding t = t0 . So, lets assume that everything is nice and continuous so that we know that a solution will in fact exist. We want to approximate the solution to (1) near t = t0 . Well start with the two pieces of information that we do know about the solution. First, we know the value of the solution at t = t0 from the initial condition. Second, we also know the value of the derivative at t = t0 . We can get this by plugging the initial condition into f(t,y) into the differential equation itself. So, the derivative at this point is.
dy = f ( t0 , y0 ) dt t =t0
2007 Paul Dawkins
94
Now, recall from your Calculus I class that these two pieces of information are enough for us to write down the equation of the tangent line to the solution at t = t0 . The tangent line is
y = y0 + f ( t0 , y0 )( t - t0 )
Take a look at the figure below
If t1 is close enough to t0 then the point y1 on the tangent line should be fairly close to the actual value of the solution at t1, or y(t1). Finding y1 is easy enough. All we need to do is plug t1 in the equation for the tangent line.
y1 = y0 + f ( t0 , y0 )( t1 - t0 )
Now, we would like to proceed in a similar manner, but we dont have the value of the solution at t1 and so we wont know the slope of the tangent line to the solution at this point. This is a problem. We can partially solve it however, by recalling that y1 is an approximation to the solution at t1. If y1 is very good approximation to the actual value of the solution then we can use that to estimate the slope of the tangent line at t1. So, lets hope that y1 is a good approximation to the solution and construct a line through the point (t1, y1) that has slope f (t1, y1). This gives
y = y1 + f ( t1 , y1 )( t - t1 )
Now, to get an approximation to the solution at t = t2 we will hope that this new line will be fairly close the actual solution at t2 and use the value of the line at t2 as an approximation to the actual solution. This gives.
y2 = y1 + f ( t1 , y1 )( t2 - t1 )
We can continue in this fashion. Use the previously computed approximation to get the next approximation. So,
y3 = y2 + f ( t2 , y2 )( t3 - t2 ) y4 = y3 + f ( t3 , y3 )( t4 - t3 ) etc.
2007 Paul Dawkins
95
In general, if we have tn and the approximation to the solution at this point, yn, and we want to find the approximation at tn+1 all we need to do is use the following.
yn +1 = yn + f ( tn , yn ) ( tn +1 - tn )
If we define f n = f ( tn , yn ) we can simplify the formula to
yn+1 = yn + f n ( tn+1 - tn )
(2)
Often, we will assume that the step sizes between the points t0 , t1 , t2 , to a uniform size of h. In other words, we will often assume that
tn +1 - tn = h
This doesnt have to be done and there are times when its best that we not do this. However, if we do the formula for the next approximation becomes.
yn+1 = yn + h f n
(3)
So, how do we use Eulers Method? Its fairly simple. We start with (1) and then decide if we want to use a uniform step size or not. Then starting with (t0, y0) we repeatedly evaluate (2) or (3) depending on whether we chose to use a uniform set size or not. We continue until weve gone the desired number of steps or reached the desired time. This will give us a sequence of numbers y1 , y2 , y3 , yn that will approximate the value of the actual solution at t1 , t2 , t3 , tn. What do we do if we want value of the solution at some other point than those used here? One possibility is to go back and redefine our set of points to a new set that will include the points we are after and redo Eulers Method using this new set of points. However this is cumbersome and could take a lot of time especially if we had to make changes to the set of points more than once. Another possibility is to remember how we arrived at the approximations in the first place. Recall that we used the tangent line
y = y0 + f ( t0 , y0 )( t - t0 ) y = y1 + f ( t1 , y1 )( t - t1 )
to get the value of y1. We could use this tangent line as an approximation for the solution on the interval [t0, t1]. Likewise, we used the tangent line to get the value of y2. We could use this tangent line as an approximation for the solution on the interval [t1, t2]. Continuing in this manner we would get a set of lines that, when strung together, should be an approximation to the solution as a whole. In practice you would need to write a computer program to do these computations for you. In most cases the function f(t,y) would be too large and/or complicated to use by hand and in most serious uses of Eulers Method you would want to use hundreds of steps which would make doing this by hand prohibitive. So, here is a bit of pseudo-code that you can use to write a program for Eulers Method that uses a uniform step size, h.
2007 Paul Dawkins
96
1. define f ( t , y ) . 2. input t0 and y0. 3. input step size, h and the number of steps, n. 4. for j from 1 to n do a. m = f (t0, y0) b. y1 = y0 + h*m c. t1 = t0 + h d. Print t1 and y1 e. t0 = t1 f. y0 = y1 5. end The pseudo-code for a non-uniform step size would be a little more complicated, but it would essentially be the same. So, lets take a look at a couple of examples. Well use Eulers Method to approximate solutions to a couple of first order differential equations. The differential equations that well be using are linear first order differential equations that can be easily solved for an exact solution. Of course, in practice we wouldnt use Eulers Method on these kinds of differential equations, but by using easily solvable differential equations we will be able to check the accuracy of the method. Knowing the accuracy of any approximation method is a good thing. It is important to know if the method is liable to give a good approximation or not.
Example 1 For the IVP y + 2 y = 2 - e -4t
y ( 0) = 1
Use Eulers Method with a step size of h = 0.1 to find approximate values of the solution at t = 0.1, 0.2, 0.3, 0.4, and 0.5. Compare them to the exact values of the solution as these points. Solution This is a fairly simple linear differential equation so well leave it to you to check that the solution is
1 1 y ( t ) = 1 + e -4 t - e -2 t 2 2
In order to use Eulers Method we first need to rewrite the differential equation into the form given in (1).
y = 2 - e -4t - 2 y
From this we can see that f ( t , y ) = 2 - e -4t - 2 y . Also note that t0 = 0 and y0 = 1. We can now start doing some computations.
f 0 = f ( 0,1) = 2 - e -4( 0) - 2 (1) = -1 y1 = y0 + h f 0 = 1 + ( 0.1)( -1) = 0.9

So, the approximation to the solution at t1 = 0.1 is y1 = 0.9. At the next step we have
f1 = f ( 0.1, 0.9 ) = 2 - e -4( 0.1) - 2 ( 0.9 ) = - 0.470320046 y2 = y1 + h f1 = 0.9 + ( 0.1)( - 0.470320046 ) = 0.852967995
Therefore, the approximation to the solution at t2 = 0.2 is y2 = 0.852967995. Ill leave it to you to check the remainder of these computations.
f 2 = -0.155264954 f3 = 0.023922788 f 4 = 0.1184359245
y3 = 0.837441500 y4 = 0.839833779 y5 = 0.851677371
Heres a quick table that gives the approximations as well as the exact value of the solutions at the given points. Time, tn t0 = 0 t1 = 0.1 t2 = 0.2 t3 = 0.3 t4 = 0.4 t5 = 0.5 Approximation y0 =1 y1 =0.9 y2 =0.852967995 y3 =0.837441500 y4 =0.839833779 y5 =0.851677371 Exact y(0) = 1 y(0.1) = 0.925794646 y(0.2) = 0.889504459 y(0.3) = 0.876191288 y(0.4) = 0.876283777 y(0.5) = 0.883727921 Error 0% 2.79 % 4.11 % 4.42 % 4.16 % 3.63 %
Weve also included the error as a percentage. Its often easier to see how well an approximation does if you look at percentages. The formula for this is,
percent error =
exact - approximate 100 exact
We used absolute value in the numerator because we really dont care at this point if the approximation is larger or smaller than the exact. Were only interested in how close the two are. The maximum error in the approximations from the last example was 4.42%, which isnt too bad, but also isnt all the great of an approximation. So, provided we arent after very accurate approximations this didnt do too badly. This kind of error is generally unacceptable in almost all real applications however. So, how can we get better approximations? Recall that we are getting the approximations by using a tangent line to approximate the value of the solution and that we are moving forward in time by steps of h. So, if we want a more accurate approximation, then it seems like one way to get a better approximation is to not move forward as much with each step. In other words, take smaller hs.
Example 2 Repeat the previous example only this time give the approximations at t = 1, t = 2,
t = 3, t = 4, and t = 5. Use h = 0.1, h = 0.05, h = 0.01, h = 0.005, and h = 0.001 for the approximations. Solution Below are two tables, one gives approximations to the solution and the other gives the errors for each approximation. Well leave the computational details to you to check.
Time t=1 t=2 t=3 t=4 t=5
Exact 0.9414902 0.9910099 0.9987637 0.9998323 0.9999773
h = 0.1 0.9313244 0.9913681 0.9990501 0.9998976 0.9999890
Approximations h = 0.05 h = 0.01 0.9364698 0.9404994 0.9911126 0.9910193 0.9988982 0.9987890 0.9998657 0.9998390 0.9999837 0.9999786
h = 0.005 0.9409957 0.9910139 0.9987763 0.9998357 0.9999780
h = 0.001 0.9413914 0.9910106 0.9987662 0.9998330 0.9999774
Percentage Errors Time h = 0.1 h = 0.05 h = 0.01 h = 0.005 h = 0.001 t=1 1.08 % 0.53 % 0.105 % 0.053 % 0.0105 % t = 2 0.036 % 0.010 % 0.00094 % 0.00041 % 0.0000703 % t = 3 0.029 % 0.013 % 0.0025 % 0.0013 % 0.00025 % t = 4 0.0065 % 0.0033 % 0.00067 % 0.00034 % 0.000067 % t = 5 0.0012 % 0.00064 % 0.00013 % 0.000068 % 0.000014 % We can see from these tables that decreasing h does in fact improve the accuracy of the approximation as we expected. There are a couple of other interesting things to note from the data. First, notice that in general, decreasing the step size, h, by a factor of 10 also decreased the error by about a factor of 10 as well. Also, notice that as t increases the approximation actually tends to get better. This isnt the case completely as we can see that in all but the first case the t = 3 error is worse than the error at t = 2, but after that point, it only get better. This should not be expected in general. In this case this is more a function of the shape of the solution. Below is a graph of the solution (the line) as well as the approximations (the dots) for h = 0.1.
Notice that the approximation is worst where the function is changing rapidly. This should not be too surprising. Recall that were using tangent lines to get the approximations and so the value of the tangent line at a given t will often be significantly different than the function due to the rapidly changing function at that point.
Also, in this case, because the function ends up fairly flat as t increases, the tangents start looking like the function itself and so the approximations are very accurate. This wont always be the case of course. Lets take a look at one more example.
Example 3 For the IVP

t 1 t y - y = - e 2 sin ( 5t ) + 5e 2 cos ( 5t ) 2
y (0) = 0
Use Eulers Method to find the approximation to the solution at t = 1, t = 2, t = 3, t = 4, and t = 5. Use h = 0.1, h = 0.05, h = 0.01, h = 0.005, and h = 0.001 for the approximations. Solution Ill leave it to you to check the details of the solution process. The solution to this linear first order differential equation is.
y ( t ) = e 2 sin ( 5t )
t
Here are two tables giving the approximations and the percentage error for each approximation. Time Exact h = 0.1 t = 1 -1.58100 -0.97167 t = 2 -1.47880 0.65270 t = 3 2.91439 7.30209 t = 4 6.74580 15.56128 t = 5 -1.61237 21.95465 Approximations h = 0.05 h = 0.01 h = 0.005 h = 0.001 -1.26512 -1.51580 -1.54826 -1.57443 -0.34327 -2.18657 -1.35810 -1.45453 5.34682 3.44488 3.18259 2.96851 11.84839 7.89808 7.33093 6.86429 12.24018 1.56056 0.0018864 -1.28498
Percentage Errors Time h = 0.1 h = 0.05 h = 0.01 h = 0.005 h = 0.001 t=1 38.54 % 19.98 % 4.12 % 2.07 % 0.42 % t = 2 144.14 % 76.79 % 16.21 % 8.16 % 1.64 % t = 3 150.55 % 83.46 % 18.20 % 9.20 % 1.86 % t = 4 130.68 % 75.64 % 17.08 % 8.67 % 1.76 % t = 5 1461.63 % 859.14 % 196.79 % 100.12 % 20.30 % So, with this example Eulers Method does not do nearly as well as it did on the first IVP. Some of the observations we made in Example 2 are still true however. Decreasing the size of h decreases the error as we saw with the last example and would expect to happen. Also, as we saw in the last example, decreasing h by a factor of 10 also decreases the error by about a factor of 10. However, unlike the last example increasing t sees an increasing error. This behavior is fairly common in the approximations. We shouldnt expect the error to decrease as t increases as we saw in the last example. Each successive approximation is found using a previous approximation. Therefore, at each step we introduce error and so approximations should, in general, get worse as t increases. Below is a graph of the solution (the line) as well as the approximations (the dots) for h = 0.05.
As we can see the approximations do follow the general shape of the solution, however, the error is clearly getting much worse as t increases. So, Eulers method is a nice method for approximating fairly nice solutions that dont change rapidly. However, not all solutions will be this nicely behaved. There are other approximation methods that do a much better job of approximating solutions. These are not the focus of this course however, so Ill leave it to you to look further into this field if you are interested. Also notice that we dont generally have the actual solution around to check the accuracy of the approximation. We generally try to find bounds on the error for each method that will tell us well an approximation should do. These error bounds are again not really the focus of this course, so Ill leave these to you as well if youre interested in looking into them.
2007 Paul Dawkins
101
Second Order Differential Equations

Introduction
In the previous chapter we looked at first order differential equations. In this chapter we will move on to second order differential equations. Just as we did in the last chapter we will look at some special cases of second order differential equations that we can solve. Unlike the previous chapter however, we are going to have to be even more restrictive as to the kinds of differential equations that well look at. This will be required in order for us to actually be able to solve them. Here is a list of topics that will be covered in this chapter. Basic Concepts Some of the basic concepts and ideas that are involved in solving second order differential equations. Real Roots Solving differential equations whose characteristic equation has real roots. Complex Roots Solving differential equations whose characteristic equation complex real roots. Repeated Roots Solving differential equations whose characteristic equation has repeated roots. Reduction of Order A brief look at the topic of reduction of order. This will be one of the few times in this chapter that non-constant coefficient differential equation will be looked at. Fundamental Sets of Solutions A look at some of the theory behind the solution to second order differential equations, including looks at the Wronskian and fundamental sets of solutions. More on the Wronskian An application of the Wronskian and an alternate method for finding it. Nonhomogeneous Differential Equations A quick look into how to solve nonhomogeneous differential equations in general. Undetermined Coefficients The first method for solving nonhomogeneous differential equations that well be looking at in this section. Variation of Parameters Another method for solving nonhomogeneous differential equations.
2007 Paul Dawkins
102
Mechanical Vibrations An application of second order differential equations. This section focuses on mechanical vibrations, yet a simple change of notation can move this into almost any other engineering field.
2007 Paul Dawkins
103
Basic Concepts
In this chapter we will be looking exclusively at linear second order differential equations. The most general linear second order differential equation is in the form.
p ( t ) y + q ( t ) y + r ( t ) y = g ( t )
(1)
In fact, we will rarely look at non-constant coefficient linear second order differential equations. In the case where we assume constant coefficients we will use the following differential equation.
ay + by + cy = g ( t )
(2)
Where possible we will use (1) just to make the point that certain facts, theorems, properties, and/or techniques can be used with the non-constant form. However, most of the time we will be using (2) as it can be fairly difficult to solve second order non-constant coefficient differential equations. Initially we will make our life easier by looking at differential equations with g(t) = 0. When g(t) = 0 we call the differential equation homogeneous and when g ( t ) 0 we call the differential equation nonhomogeneous. So, lets start thinking about how to go about solving a constant coefficient, homogeneous, linear, second order differential equation. Here is the general constant coefficient, homogeneous, linear, second order differential equation.
ay + by + cy = 0
Its probably best to start off with an example. This example will lead us to a very important fact that we will use in every problem from this point on. The example will also give us clues into how to go about solving these in general.
Example 1 Determine some solutions to
y - 9 y = 0
Solution We can get some solutions here simply by inspection. We need functions whose second derivative is 9 times the original function. One of the first functions that I can think of that comes back to itself after two derivatives is an exponential function and with proper exponents the 9 will get taken care of as well. So, it looks like the following two functions are solutions.
y ( t ) = e 3t
and
y ( t ) = e -3t
Well leave it to you to verify that these are in fact solutions. These two functions are not the only solutions to the differential equation however. Any of the following are also solutions to the differential equation.
2007 Paul Dawkins
104
y ( t ) = -9e3t y ( t ) = 56e -3t y ( t ) = 7e3t - 6e -3t
y ( t ) = 123e3t 14 -3t e 9 y ( t ) = -92e3t - 16e -3t y (t ) =
In fact if you think about it any function that is in the form will be a solution to the differential equation.
y ( t ) = c1e3t + c2e -3t
This example leads us to a very important fact that we will use in practically every problem in this chapter. Principle of Superposition If y1 ( t ) and y2 ( t ) are two solutions to a linear, homogeneous differential equation then so is
y ( t ) = c1 y1 ( t ) + c2 y2 ( t )
Note that we didnt include the restriction of constant coefficient or second order in this. This will work for any linear homogeneous differential equation.
(3)
If we further assume second order and one other condition (which well give in a second) we can go a step further. If y1 ( t ) and y2 ( t ) are two solutions to a linear, second order homogeneous differential equation and they are nice enough then the general solution to the linear, second order differential equation is given by (3). So, just what do we mean by nice enough? Well hold off on that until a later section. At this point youll hopefully believe it when we say that specific functions are nice enough. So, if we now make the assumption that we are dealing with a linear, second order differential equation, we now know that (3) will be its general solution. The next question that we can ask is how to find the constants c1 and c2. Since we have two constants it makes sense, hopefully, that we will need two equations, or conditions, to find them. One way to do this is to specify the value of the solution at two distinct points, or,
y ( t0 ) = y0
y ( t1 ) = y1
These are typically called boundary values and are not really the focus of this course so we wont be working with them. Another way to find the constants would be to specify the value of the solution and its derivative at a particular point. Or,
y ( t0 ) = y0
y ( t0 ) = y0
2007 Paul Dawkins
105
These are the two conditions that well be using here. As with the first order differential equations these will be called initial conditions.
Example 2 Solve the following IVP. y - 9 y = 0

Solution First, the two functions
y (0) = 2
y ( 0 ) = -1 y ( t ) = e -3t
y ( t ) = e 3t
and
are nice enough for us to form the general solution to the differential equation. At this point, please just believe this. You will be able to verify this for yourself in a couple of sections. The general solution to our differential equation is then
y ( t ) = c1e -3t + c2e3t
Now all we need to do is apply the initial conditions. This means that we need the derivative of the solution.
y ( t ) = -3c1e -3t + 3c2e3t 2 = y ( 0 ) = c1 + c2
Plug in the initial conditions
-1 = y ( 0 ) = -3c1 + 3c2
This gives us a system of two equations and two unknowns that can be solved. Doing this yields
c1 =
The solution to the IVP is then,
7 6
c2 =
5 6
7 5 y ( t ) = e -3 t + e 3 t 6 6
Up to this point weve only looked at a single differential equation and we got its solution by inspection. For a rare few differential equations we can do this. However, for the vast majority of the second order differential equations out there we will be unable to do this. So, we would like a method for arriving at the two solutions we will need in order to form a general solution that will work for any linear, constant coefficient, second order differential equation. This is easier than it might initially look. We will use the solutions we found in the first example as a guide. All of the solutions in this example were in the form
y (t ) = er t
Note, that we didnt include a constant in front of it since we can literally include any constant that we want and still get a solution. The important idea here is to get the exponential function. Once we have that we can add on constants to our hearts content.
So, lets assume that all solutions to will be of the form
ay + by + cy = 0 y (t ) = er t
(4) (5)
To see if we are correct all we need to do is plug this into the differential equation and see what happens. So, lets get some derivatives and then plug in.
y ( t ) = re rt
a ( r 2 e r t ) + b ( re r t ) + c ( e r t ) = 0 e r t ( ar 2 + br + c ) = 0
y ( t ) = r 2e r t
So, if (5) is to be a solution to (4) then the following must be true
e r t ( ar 2 + br + c ) = 0
This can be reduced further by noting that exponentials are never zero. Therefore, (5) will be a solution to (4) provided r is a solution to
ar 2 + br + c = 0
This equation is typically called the characteristic equation for (4). Okay, so how do we use this to find solutions to a linear, constant coefficient, second order differential equation? First write down the characteristic equation, (6), for the differential equation, (4). This will be a quadratic equation and so we should expect two roots, r1 and r2. Once we have these two roots we have two solutions to the differential equation.
(6)
y1 ( t ) = e 1
rt
and
y2 ( t ) = e
r2 t
(7)
Lets take a look at a quick example.
Example 3 Find two solutions to
y - 9 y = 0
Solution This is the same differential equation that we looked at in the first example. This time however, lets not just guess. Lets go through the process as out lined above to see the functions that we guess above are the same as the functions the process gives us. First write down the characteristic equation for this differential equation and solve it.
r2 - 9 = 0 y1 ( t ) = e3t
r = 3 y 2 ( t ) = e -3 t
The two roots are 3 and -3. Therefore, two solutions are
and
These match up with the first guesses that we made in the first example.
2007 Paul Dawkins
107
Youll notice that we neglected to mention whether or not the two solutions listed in (7) are in fact nice enough to form the general solution to (4). This was intentional. We have three cases that we need to look at and this will be addressed differently in each of these cases. So, what are the cases? As we previously noted the characteristic equation is quadratic and so will have two roots, r1 and r2. The roots will have three possible forms. These are 1. Real, distinct roots, r1 r2 . 2. Complex root, r1,2 = l m i . 3. Double roots, r1 = r2 = r . The next three sections will look at each of these in some more depth, including giving forms for the solution that will be nice enough to get a general solution.
2007 Paul Dawkins
108
Real, Distinct Roots

Its time to start solving constant coefficient, homogeneous, linear, second order differential equations. So, lets recap how we do this from the last section. We start with the differential equation.
ay + by + cy = 0 ar 2 + br + c = 0
Write down the characteristic equation.
Solve the characteristic equation for the two roots, r1 and r2. This gives the two solutions
y1 ( t ) = e 1
rt
and
y2 ( t ) = e
r2 t
Now, if the two roots are real and distinct (i.e. r1 r2 ) it will turn out that these two solutions are nice enough to form the general solution
y ( t ) = c1e 1 + c2e
r t
r2 t
As with the last section, well ask that you believe us when we say that these are nice enough. You will be able to prove this easily enough once we reach a later section. With real, distinct roots there really isnt a whole lot to do other than work a couple of examples so lets do that.
Example 1 Solve the following IVP. y + 11y + 24 y = 0

Solution The characteristic equation is
y ( 0) = 0
y ( 0 ) = -7
r 2 + 11r + 24 = 0
( r + 8)( r + 3) = 0
Its roots are r1 = - 8 and r2 = -3 and so the general solution and its derivative is.
y ( t ) = c1e -8 t + c2e -3t
y ( t ) = -8c1e -8t - 3c2e -3t

Now, plug in the initial conditions to get the following system of equations.
0 = y ( 0 ) = c1 + c2
-7 = y ( 0 ) = -8c1 - 3c2
Solving this system gives c1 = then
7 5
and c2 = - 7 5 . The actual solution to the differential equation is
7 7 y ( t ) = e -8 t - e -3 t 5 5
Example 2 Solve the following IVP y + 3 y - 10 y = 0

y (0) = 4 r 2 + 3r - 10 = 0
y ( 0 ) = -2
( r + 5 )( r - 2 ) = 0
Its roots are r1 = - 5 and r2 = 2 and so the general solution and its derivative is.
y ( t ) = c1e-5t + c2 e 2t
y ( t ) = -5c1e -5t + 2c2e 2t

4 = y ( 0 ) = c1 + c2
-2 = y ( 0 ) = -5c1 + 2c2
18 Solving this system gives c1 = 10 7 and c2 = 7 . The actual solution to the differential equation is then
y (t ) = Example 3 Solve the following IVP. 3 y + 2 y - 8 y = 0

10 -5t 18 2 t e + e 7 7 y ( 0 ) = -6 y ( 0 ) = -18
3r 2 + 2r - 8 = 0
( 3r - 4 )( r + 2 ) = 0
Its roots are r1 =
4 3
and r2 = -2 and so the general solution and its derivative is.
y ( t ) = c1e 3 + c2e -2 t 4 4t y ( t ) = c1e 3 - 2c2e -2 t 3

4t
-6 = y ( 0 ) = c1 + c2
4 -18 = y ( 0 ) = c1 - 2c2 3
Solving this system gives c1 = -9 and c2 = 3. The actual solution to the differential equation is then.
y ( t ) = -9e 3 + 3e -2t
4t
2007 Paul Dawkins
110
Example 4 Solve the following IVP 4 y - 5 y = 0

y ( -2 ) = 0 4r 2 - 5r = 0 r ( 4r - 5 ) = 0
y ( -2 ) = 7
The roots of this equation are r1 = 0 and r2 = derivative.
5 4
. Here is the general solution as well as its

5t 5t
y ( t ) = c1e0 + c2e 4 = c1 + c2e 4 5 5t y ( t ) = c2e 4 4
Up to this point all of the initial conditions have been at t = 0 and this one isnt. Dont get to locked into initial conditions always being at t = 0 and you just automatically use that instead of the actual value for a given problem. So, plugging in the initial conditions gives the following system of equations to solve.
0 = y ( -2 ) = c1 + c2e 7 = y ( -2 ) =
Solving this gives.
-5 2
5 -5 c2e 2 4 28 5 e2 5
c1 = -
28 5
c2 =
The solution to the differential equation is then.
y (t ) = -
5t 28 28 5 28 28 5 t + 5 + e 2e 4 = - + e 4 2 5 5 5 5
In a differential equations class most instructors (including me.) tend to use initial conditions at t = 0 because it makes the work a little easier for the students as they are trying to learn the subject. However, there is no reason to always expect that this will be the case, so do not start to always expect initial conditions at t = 0! Lets do one final example to make another point that you need to be made aware of.
2007 Paul Dawkins
111
Example 5 Find the general solution to the following differential equation. y - 6 y - 2 y = 0

Solution The characteristic equation is.
r 2 - 6r - 2 = 0
The roots of this equation are.
r1,2 = 3 11
Now, do NOT get excited about these roots they are just two real numbers.
r1 = 3 + 11
and
r1 = 3 - 11
Admittedly they are not as nice looking as we may be used to, but they are just real numbers. Therefore, the general solution is
y ( t ) = c1e
( 3+ 11) t
+ c2e
( 3- 11) t
If we had initial conditions we could proceed as we did in the previous two examples although the work would be somewhat messy and so we arent going to do that for this example. The point of the last example is make sure that you dont get to used to nice, simple roots. In practice roots of the characteristic equation will generally not be nice, simple integers or fractions so dont get too used to them!
2007 Paul Dawkins
112
Complex Roots
In this section we will be looking at solutions to the differential equation
ay + by + cy = 0
in which roots of the characteristic equation,
ar 2 + br + c = 0 are complex roots in the form r1,2 = l m i .

Now, recall that we arrived at the characteristic equation by assuming that all solutions to the differential equation will be of the form
y ( t ) = e rt
Plugging our two roots into the general form of the solution gives the following solutions to the differential equation. (l + m i) t (l - m i ) t
y1 ( t ) = e
and
y2 ( t ) = e
Now, these two functions are nice enough (theres those words again well get around to defining them eventually) to form the general solution. We do have a problem however. Since we started with only real numbers in our differential equation we would like our solution to only involve real numbers. The two solutions above are complex and so we would like to get our hands on a couple of solutions (nice enough of course) that are real. To do this well need Eulers Formula.
eiq = cos q + i sin q

A nice variant of Eulers Formula that well need is.
e -iq = cos ( -q ) + i sin ( -q ) = cos q - i sin q
Now, split up our two solutions into exponentials that only have real exponents and exponentials that only have imaginary exponents. Then use Eulers formula, or its variant, to rewrite the second exponential.
y1 ( t ) = el t eim t = e l t ( cos ( m t ) + i sin ( m t ) )
y2 ( t ) = el t e -i m t = el t ( cos ( m t ) - i sin ( m t ) )
This doesnt eliminate the complex nature of the solutions, but it does put the two solutions into a form that we can eliminate the complex parts. Recall from the basics section that if two solutions are nice enough then any solution can be written as a combination of the two solutions. In other words,
y ( t ) = c1 y1 ( t ) + c2 y2 ( t )
will also be a solution. Using this lets notice that if we add the two solutions together we will arrive at.
y1 ( t ) + y2 ( t ) = 2el t cos ( m t )
113
2007 Paul Dawkins
This is a real solution and just to eliminate the extraneous 2 lets divide everything by a 2. This gives the first real solution that were after.
u (t ) =
1 1 y1 ( t ) + y2 ( t ) = e l t cos ( m t ) 2 2 1 2
Note that this is just equivalent to taking
c1 = c2 =
Now, we can arrive at a second solution in a similar manner. This time lets subtract the two original solutions to arrive at.
y1 ( t ) - y2 ( t ) = 2i el t sin ( m t )
On the surface this doesnt appear to fix the problem as the solution is still complex. However, upon learning that the two constants, c1 and c2 can be complex numbers we can arrive at a real solution by dividing this by 2i. This is equivalent to taking
c1 =
Our second solution will then be
1 2i
and
c2 = -
1 2i
v (t ) =
1 1 y1 ( t ) - y2 ( t ) = elt sin ( m t ) 2i 2i
We now have two solutions (well leave it to you to check that they are in fact solutions) to the differential equation.
u ( t ) = elt cos ( m t )
and
v ( t ) = elt sin ( m t )
It also turns out that these two solutions are nice enough to form a general solution. So, if the roots of the characteristic equation happen to be r1,2 = l m i the general solution to the differential equation is.
y ( t ) = c1elt cos ( m t ) + c2e lt sin ( m t )
Lets take a look at a couple of examples now.
Example 1 Solve the following IVP. y - 4 y + 9 y = 0
y (0) = 0
y ( 0 ) = -8
Solution The characteristic equation for this differential equation is.
r 2 - 4r + 9 = 0 The roots of this equation are r1,2 = 2 5 i . The general solution to the differential equation is
then.
y ( t ) = c1e2t cos
( 5t ) + c e
2
2t
sin
( 5t )
2007 Paul Dawkins
114
Now, youll note that we didnt differentiate this right away as we did in the last section. The reason for this is simple. While the differentiation is not terribly difficult, it can get a little messy. So, first looking at the initial conditions we can see from the first one that if we just applied it we would get the following.
0 = y ( 0 ) = c1
In other words, the first term will drop out in order to meet the first condition. This makes the solution, along with its derivative
y ( t ) = c2e 2t sin y ( t ) = 2c2 e 2t
( 5t ) sin ( 5t ) +
5c2e2t cos
( 5t )
8 5
A much nicer derivative than if wed done the original solution. Now, apply the second initial condition to the derivative to get.
-8 = y ( 0 ) = 5c2
The actual solution is then.
c2 = -
y (t ) = Example 2 Solve the following IVP. y - 8 y + 17 y = 0

Solution The characteristic equation this time is.
8 2t e sin 5
( 5t )
y ( 0 ) = -1
y ( 0 ) = -4
r 2 - 8r + 17 = 0
The roots of this are r1,2 = 4 i . The general solution as well as its derivative is
y ( t ) = c1e4 t cos ( t ) + c2e 4 t sin ( t ) y ( t ) = 4c1e 4t cos ( t ) - c1e 4t sin ( t ) + 4c2e 4t sin ( t ) + c2e 4t cos ( t )
Notice that this time we will need the derivative from the start as we wont be having one of the terms drop out. Applying the initial conditions gives the following system.
-4 = y ( 0 ) = c1
-1 = y ( 0 ) = 4c1 + c2
Solving this system gives c1 = -4 and c2 = 15 . The actual solution to the IVP is then.
y ( t ) = -4e 4t cos ( t ) + 15e 4t sin ( t )
2007 Paul Dawkins
115
Example 3 Solve the following IVP. 4 y + 24 y + 37 y = 0

Solution The characteristic equation this time is.
y (p ) = 1
y (p ) = 0
4r 2 + 24r + 37 = 0
The roots of this are r1,2 = -3 1 2 i . The general solution as well as its derivative is
t t y ( t ) = c1e-3t cos + c2e -3t sin 2 2 t c t t c t y ( t ) = -3c1e -3t cos - 1 e-3t sin - 3c2 e -3t sin + 2 e -3t cos 2 2 2 2 2 2
Applying the initial conditions gives the following system.
p p 1 = y (p ) = c1e -3p cos + c2 e-3p sin = c2e -3p 2 2 c 0 = y (p ) = - 1 e -3p - 3c2e -3p 2
Do not forget to plug the t = p into the exponential! This is one of the more common mistakes that students make on these problems. Also, make sure that you evaluate the trig functions as much as possible in these cases. It will only make your life simpler. Solving this system gives
c1 = -6e3p
The actual solution to the IVP is then.
c2 = e3p
t t y ( t ) = -6e3p e -3t cos + e3p e-3t sin 2 2 t t y ( t ) = -6e -3( t -p ) cos + e -3( t -p ) sin 2 2
Lets do one final example before moving on to the next topic.
Example 4 Solve the following IVP. y + 16 y = 0 p y = -10 2 p y = 3 2 r = 4 i
Solution The characteristic equation for this differential equation and its roots are.
r 2 + 16 = 0
The general solution to this differential equation and its derivative is.
y ( t ) = c1 cos ( 4t ) + c2 sin ( 4t )
y ( t ) = -4c1 sin ( 4t ) + 4c2 cos ( 4t )

Plugging in the initial conditions gives the following system.
p -10 = y = c1 2 p 3 = y = 4c2 2
c1 = -10 c2 = 3 4
So, the constants drop right out with this system and the actual solution is.
3 y ( t ) = -10 cos ( 4t ) + sin ( 4t ) 4
2007 Paul Dawkins
117
Repeated Roots
In this section we will be looking at the last case for the constant coefficient, linear, homogeneous second order differential equations. In this case we want solutions to
ay + by + cy = 0 ar 2 + br + c = 0
where solutions to the characteristic equation are double roots r1 = r2 = r. This leads to a problem however. Recall that the solutions are
y1 ( t ) = e r1 t = er t
y2 ( t ) = e r2 t = e r t
These are the same solution and will NOT be nice enough to form a general solution. I do promise that Ill define nice enough eventually! So, we can use the first solution, but were going to need a second solution. Before finding this second solution lets take a little side trip. The reason for the side trip will be clear eventually. From the quadratic formula we know that the roots to the characteristic equation are,
r1,2 =
-b b 2 - 4ac 2a
In this case, since we have double roots we must have
b 2 - 4ac = 0
This is the only way that we can get double roots and in this case the roots will be
r1,2 =
So, the one solution that weve got is
-b 2a
y1 ( t ) = e
- bt 2a
To find a second solution we will use the fact that a constant times a solution to a linear homogeneous differential equation is also a solution. If this is true then maybe well get lucky and the following will also be a solution
y2 ( t ) = v ( t ) y1 ( t ) = v ( t ) e
bt -2 a
(1)
with a proper choice of v(t). To determine if this in fact can be done, lets plug this back into the differential equation and see what we get. Well first need a couple of derivatives.
( t ) = v e y2
bt -2 a
( t ) = v e y2 = v e
bt -2 a
bt -2 a
b - bt v e 2a 2a b b b2 - bt - bt - bt - v e 2 a - v e 2 a + 2 v e 2 a 2a 2a 4a 2 b - bt b - bt - v e 2 a + 2 v e 2 a a 4a
2007 Paul Dawkins
We dropped the (t) part on the v to simplify things a little for the writing out of the derivatives. Now, plug these into the differential equation.
bt bt bt - bt b - b t b 2 - bt b a v e 2 a - v e 2a + 2 v e 2 a + b v e 2 a - v e 2 a + c v e 2 a = 0 a 4a 2a
We can factor an exponential out of all the terms so lets do that. Well also collect all the coefficients of v and its derivatives.
bt -2 a
b2 b2 + cv = 0 av + ( -b + b ) v + 4a 2a 2 b - bt e 2 a av + - + c v = 0 4a 1 - bt e 2 a av - ( b 2 - 4ac ) v = 0 4a
Now, because we are working with a double root we know that that the second term will be zero. Also exponentials are never zero. Therefore, (1) will be a solution to the differential equation provided v(t) is a function that satisfies the following differential equation.
av = 0
OR
v = 0
We can drop the a because we know that it cant be zero. If it were we wouldnt have a second order differential equation! So, we can now determine the most general possible form that is allowable for v(t).
v = v dt = c
v ( t ) = v dt = ct + k
This is actually more complicated than we need and in fact we can drop both of the constants from this. To see why this is lets go ahead and use this to get the second solution. The two solutions are then
y1 ( t ) = e
bt -2 a
y2 ( t ) = ( ct + k ) e
bt -2 a
Eventually you will be able to show that these two solutions are nice enough to form a general solution. The general solution would then be the following.
y ( t ) = c1e = c1e
bt -2 a bt -2 a
+ c2 ( ct + k ) e
bt -2 a
bt -2 a bt -2 a bt -2 a
+ ( c2 ct + c2 k ) e
= ( c1 + c2 k ) e
+ c2c t e
Notice that we rearranged things a little. Now, c, k, c1, and c2 are all unknown constants so any combination of them will also be unknown constants. In particular, c1+c2 k and c2 c are unknown constants so well just rewrite them as follows.
y ( t ) = c1e
bt -2 a
+ c2 t e
bt -2 a
2007 Paul Dawkins
119
So, if we go back to the most general form for v(t) we can take c=1 and k=0 and we will arrive at the same general solution. Lets recap. If the roots of the characteristic equation are r1 = r2 = r, then the general solution is then
y ( t ) = c1e r t + c2te r t
Now, lets work a couple of examples.
Example 1 Solve the following IVP. y - 4 y + 4 y = 0

Solution The characteristic equation and its roots are.
y ( 0 ) = 12
2
y ( 0 ) = -3
r 2 - 4r + 4 = ( r - 2 ) = 0
The general solution and its derivative are
r1,2 = 2
y ( t ) = c1e 2t + c2te 2t
y ( t ) = 2c1e 2 t + c2e 2t + 2c2te 2 t

Dont forget to product rule the second term! Plugging in the initial conditions gives the following system.
12 = y ( 0 ) = c1
-3 = y ( 0 ) = 2c1 + c2
This system is easily solve to get c1 = 12 and c2 = -27. The actual solution to the IVP is then.
y ( t ) = 12e2 t - 27te2 t
Example 2 Solve the following IVP. 16 y - 40 y + 25 y = 0

y (0) = 3
y ( 0 ) = -
9 4
16r 2 - 40r + 25 = ( 4r - 5 ) = 0
2
r1,2 =
5 4
y ( t ) = c1e 4 + c2te 4
5t
5t
5t 5t 5 5t 5 y ( t ) = c1e 4 + c2e 4 + c2te 4 4 4
Dont forget to product rule the second term! Plugging in the initial conditions gives the following system.
2007 Paul Dawkins
120
3 = y ( 0 ) = c1 9 5 = y ( 0 ) = c1 + c2 4 4 y ( t ) = 3e 4 - 6te 4 Example 3 Solve the following IVP y + 14 y + 49 y = 0

5t 5t
This system is easily solve to get c1 = 3 and c2 = -6. The actual solution to the IVP is then.
y ( -4 ) = -1
2
y ( -4 ) = 5
r 2 + 14r + 49 = ( r + 7 ) = 0
r1,2 = -7
y ( t ) = c1e -7 t + c2 te-7 t
y ( t ) = -7c1e -7 t + c2e -7 t - 7c2te -7 t

Plugging in the initial conditions gives the following system of equations.
-1 = y ( -4 ) = c1e 28 - 4c2e28
5 = y ( -4 ) = -7c1e28 + c2e 28 + 28c2e 28 = -7c1e 28 + 29c2e28

Solving this system gives the following constants.
c1 = -9e -28
The actual solution to the IVP is then.
c2 = -2e -28
y ( t ) = -9e -28e -7t - 2te -28e -7t y ( t ) = -9e -7( t + 4) - 2te-7(t + 4)
2007 Paul Dawkins
121
Reduction of Order
Were now going to take a brief detour and look at solutions to non-constant coefficient, second order differential equations of the form.
p ( t ) y + q ( t ) y + r ( t ) y = 0
In general, finding solutions to these kinds of differential equations can be much more difficult than finding solutions to constant coefficient differential equations. However, if we already know one solution to the differential equation we can use the method that we used in the last section to find a second solution. This method is called reduction of order. Lets take a quick look at an example to see how this is done.
Example 1 Find the general solution to 2t 2 y + ty - 3 y = 0

given that y1 ( t ) = t -1 is a solution. Solution Reduction of order requires that a solution already be known. Without this known solution we wont be able to do reduction order. Once we have this first solution we will then assume that a second solution will have the form
y2 ( t ) = v ( t ) y1 ( t )
(1)
for a proper choice of v(t). To determine the proper choice, we plug the guess into the differential equation and get a new differential equation that can be solved for v(t). So, lets do that for this problem. Here is the form of the second solution as well as the derivatives that well need.
y2 ( t ) = t -1v
( t ) = -t -2v + t -1v y2
( t ) = 2t -3v - 2t -2v + t -1v y2
Plugging these into the differential equation gives
2t 2 ( 2t -3v - 2t -2v + t -1v ) + t ( -t -2v + t -1v ) - 3 ( t -1v ) = 0 2tv + ( -4 + 1) v + ( 4t -1 - t -1 - 3t -1 ) v = 0 2tv - 3v = 0
Rearranging and simplifying gives
Note that upon simplifying the only terms remaining are those involving the derivatives of v. The term involving v drops out. If youve done all of your work correctly this should always happen. Sometimes, as in the repeated roots case, the first derivative term will also drop out. So, in order for (1) to be a solution then v must satisfy
2tv - 3v = 0
(2)
This appears to be a problem. In order to find a solution to a second order non-constant

coefficient differential equation we need to solve a different second order non-constant coefficient differential equation. However, this isnt the problem that it appears to be. Because the term involving the v drops out we can actually solve (2) and we can do it with the knowledge that we already have at this point. We will solve this by making the following change of variable.
w = v
w = v
With this change of variable (2) becomes
2tw - 3w = 0
and this is a linear, first order differential equation that we can solve. This also explains the name of this method. Weve managed to reduce a second order differential equation down to a first order differential equation. This is a fairly simple first order differential equation so Ill leave the details of the solving to you. If you need a refresher on solving linear, first order differential equations go back to the second chapter and check out that section. The solution to this differential equation is
w (t ) = c t 2
3
Now, this is not quite what we were after. We are after a solution to (2). However, we can now find this. Recall our change of variable.
v = w
With this we can easily solve for v(t).

3 2 5 v ( t ) = w dt = ct 2 dt = ct 2 + k 5
This is the most general possible v(t) that we can use to get a second solution. So, just as we did in the repeated roots section, we can choose the constants to be anything we want so choose them to clear out all the extraneous constants. In this case we can use
c=
5 2
k =0
Using these gives the following for v(t) and for the second solution.
v (t ) = t 2
5
y 2 ( t ) = t -1 t 2 = t 2
5 3
( )
Then general solution will then be,
y ( t ) = c1t -1 + c2t 2
3
If we had been given initial conditions we could then differentiate, apply the initial conditions and solve for the constants. Reduction of order, the method used in the previous example can be used to find second solutions to differential equations. However, this does require that we already have a solution and often finding that first solution is a very difficult task and often in the process of finding the first solution you will also get the second solution without needing to resort to reduction of order. So,
for those cases when we do have a first solution this is a nice method for getting a second solution. Lets do one more example.
Example 2 Find the general solution to t 2 y + 2ty - 2 y = 0 given that y1 ( t ) = t is a solution.

Solution The form for the second solution as well as its derivatives are,
y2 ( t ) = tv
y 2 ( t ) = v + tv
( t ) = 2v + tv y2
Plugging these into the differential equation gives,
t 2 ( 2v + tv ) + 2t ( v + tv ) - 2 ( tv ) = 0 = 0
Rearranging and simplifying gives the differential equation that well need to solve in order to determine the correct v that well need for the second solution.
t 3v + 4t 2 v = 0
Next use the variable transformation as we did in the previous example.
w = v
w = v
With this change of variable the differential equation becomes
t 3 w + 4t 2 w = 0
and this is a linear, first order differential equation that we can solve. Well leave the details of the solution process to you.
w ( t ) = c t -4
Now solve for v(t).
1 v ( t ) = w dt = ct -4 dt = - ct -3 + k 3
As with the first example well drop the constants and use the following v(t)
v ( t ) = t -3
y2 ( t ) = t ( t -3 ) = t -2 c2 t2
Then general solution will then be,
y ( t ) = c1t +
On a side note, both of the differential equations in this section were of the form,
t 2 y + a t y + b y = 0
2007 Paul Dawkins
124
These are called Euler differential equations and are fairly simple to solve directly for both solutions. To see how to solve these directly take a look at the Euler Differential Equation section.
2007 Paul Dawkins
125
Fundamental Sets of Solutions

The time has finally come to define nice enough. Weve been using this term throughout the last few sections to describe those solutions that could be used to form a general solution and it is now time to officially define it. First, because everything that were going to be doing here only requires linear and homogeneous we wont require constant coefficients in our differential equation. So, lets start with the following IVP.
p ( t ) y + q ( t ) y + r ( t ) y = 0 y ( t0 ) = y0 y ( t0 ) = y0
(1)
Lets also suppose that we have already found two solutions to this differential equation, y1(t) and y2(t). We know from the Principle of Superposition that
y ( t ) = c1 y1 ( t ) + c2 y2 ( t )
(2)
Will also be a solution to the differential equation. What we want to know is whether or not it will be a general solution. In order for (2) to be considered a general solution it must satisfy the general initial conditions in (1).
y ( t0 ) = y0
y ( t0 ) = y0
This will also imply that any solution to the differential equation can be written in this form. So, lets see if we can find constants that will satisfy these conditions. First differentiate (2) and plug in the initial conditions.
y0 = y ( t0 ) = c1 y1 ( t0 ) + c2 y2 ( t0 )
= y ( t0 ) = c1 y1 ( t0 ) + c2 y2 ( t0 ) y0
Since we are assuming that weve already got the two solutions everything in this system is technically known and so this is a system that can be solved for c1 and c2. This can be done in general using Cramers Rule. Using Cramers Rule gives the following solution.
(3)
c1 =
y1 ( t0 ) ( t0 ) y1
y0 y0
y2 ( t0 ) ( t0 ) y2
y 2 ( t0 ) ( t0 ) y2 a b = ad - bc c d
c2 =
y1 ( t0 ) ( t0 ) y1
y1 ( t0 ) ( t0 ) y1
y 2 ( t0 ) ( t0 ) y2
y0 y0
(4)
where,
is the determinant of a 2x2 matrix. If you dont know about determinants that is okay, just use the formula that weve provided above.
2007 Paul Dawkins
126
Now, (4) will give the solution to the system (3). Note that in practice we generally dont use Cramers Rule to solve systems, we just proceed in a straightforward manner and solve the system using basic algebra techniques. So, why did we use Cramers Rule here then? We used Cramers Rule because we can use (4) to develop a condition that will allow us to determine when we can solve for the constants. All three (yes three, the denominators are the same!) of the quantities in (4) are just numbers and the only thing that will prevent us from actually getting a solution will be when the denominator is zero. The quantity in the denominator is called the Wronskian and is denoted as
W ( f , g )( t ) =
f (t ) g (t ) = f (t ) g ( t ) - g (t ) f (t ) f (t ) g (t )
When it is clear what the functions and/or t are we often just denote the Wronskian by W. Lets recall what we were after for here. We wanted to determine when two solutions to (1) would be nice enough to form a general solution. The two solutions will form a general solution to (1) if they satisfy the general initial conditions given in (1) and we can see from Cramers Rule we can see that they will satisfy the initial conditions provided the Wronskian isnt zero. Or,
W ( y1 , y2 )( t0 ) =
y1 ( t0 ) ( t0 ) y1
y 2 ( t0 ) ( t0 ) - y2 ( t0 ) y1 ( t0 ) 0 = y1 ( t0 ) y2 ( t0 ) y2
So, suppose that y1(t) and y2(t) are two solutions to (1) and that W ( y1 , y2 )( t ) 0 . Then the two solutions are called a fundamental set of solutions and the general solution to (1) is
y ( t ) = c1 y1 ( t ) + c2 y2 ( t )
We know now what nice enough means. Two solutions are nice enough if they are a fundamental set of solutions. So, lets check one of the claims that we made in a previous section. Well leave the other two to you to check if youd like to.
Example 1 Back in the complex root section we made the claim that y1 ( t ) = e lt cos ( m t ) and y2 ( t ) = elt sin ( m t )
where a fundamental set of solutions. Prove that they in fact are. Solution So, to prove this we will need to take find the Wronskian for these two solutions and show that it isnt zero.
2007 Paul Dawkins
127
W=
elt cos ( m t ) elt sin ( m t ) l e lt cos ( m t ) - m elt sin ( m t ) l e lt sin ( m t ) + me lt cos ( m t ) elt sin ( m t ) ( l elt cos ( m t ) - me lt sin ( m t ) )
= elt cos ( m t ) ( le lt sin ( m t ) + m elt cos ( m t ) ) = me 2 lt cos 2 ( m t ) + me 2 lt sin 2 ( m t ) = me 2 lt ( cos 2 ( m t ) + sin 2 ( m t ) ) = me 2 lt

Now, the exponential will never be zero and m 0 (if it were we wouldnt have complex roots!) and so W 0 . Therefore, these two solutions are in fact a fundamental set of solutions and so the general solution in this case is.
y ( t ) = c1elt cos ( m t ) + c2e lt sin ( m t )
Example 2 In the first example that we worked in the Reduction of Order section we found a
second solution to
2t 2 y + ty - 3 y = 0
Show that this second solution, along with the given solution, form a fundamental set of solutions for the differential equation. Solution The two solutions from that example are
y1 ( t ) = t -1
Lets compute the Wronskian of these two solutions.
3
y2 ( t ) = t 2
3
W=
t -1 -t -2
t2 3 -1 5 -1 5 -1 = t 2 - -t 2 = t 2 = 3 1 2 2 t t2 2 2
( )
So, the Wronskian will never be zero. Note that we cant plug t = 0 into the Wronskian. This would be a problem in finding the constants in the general solution, except that we also cant plug t = 0 into the solution either and so this isnt the problem that it might appear to be. So, since the Wronskian isnt zero for any t the two solutions form a fundamental set of solutions and the general solution is
y ( t ) = c1t -1 + c2t 2
3
as we claimed in that example. To this point weve found a set of solutions then weve claimed that they are in fact a fundamental set of solutions. Of course, you can now verify all those claims that weve made,
2007 Paul Dawkins
128
however this does bring up a question. How do we know that for a given differential equation a set of fundamental solutions will exist? The following theorem answers this question. Theorem Consider the differential equation
y + p ( t ) y + q ( t ) y = 0 y ( t0 ) = 1 y ( t0 ) = 0 y ( t0 ) = 1
where p(t) and q(t) are continuous functions on some interval I. Choose t0 to be any point in the interval I. Let y1(t) be a solution to the differential equation that satisfies the initial conditions.
Let y2(t) be a solution to the differential equation that satisfies the initial conditions.
y ( t0 ) = 0
Then y1(t) and y2(t) form a fundamental set of solutions for the differential equation. It is easy enough to show that these two solutions form a fundamental set of solutions. Just compute the Wronskian.
W ( y1 , y2 )( t0 ) =
y1 ( t0 ) ( t0 ) y1
y 2 ( t0 ) 1 0 = = 1- 0 = 1 0 ( t0 ) 0 1 y2
So, fundamental sets of solutions will exist provided we can solve the two IVPs given in the theorem.
Example 3 Use the theorem to find a fundamental set of solutions for y + 4 y + 3 y = 0

using t0 = 0. Solution Using the techniques from the first part of this chapter we can find the two solutions that weve been using to this point.
y ( t ) = e -3t
y ( t ) = e -t
These do form a fundamental set of solutions as we can easily verify. However, they are NOT the set that will be given by the theorem. Neither of these solutions will satisfy either of the two sets of initial conditions given in the theorem. We will have to use these to find the fundamental set of solutions that is given by the theorem. We know that the following is also solution to the differential equation.
y ( t ) = c1e -3t + c2 e -t
So, lets apply the first set of initial conditions and see if we can find constants that will work.
y (0) = 1
y ( 0) = 0
Well leave it to you to verify that we get the following solution upon doing this.
2007 Paul Dawkins
129
1 3 y1 ( t ) = - e -3t + e- t 2 2
Likewise, if we apply the second set of initial conditions,
y (0) = 0
y ( 0 ) = 1
we will get
1 1 y2 ( t ) = - e -3t + e -t 2 2
According to the theorem these should form a fundament set of solutions. This is easy enough to check.
1 3 1 1 - e -3t + e -t - e-3t + e - t 2 2 2 2 W= 3 -3 t 3 - t 3 -3t 1 -t e - e e - e 2 2 2 2 3 3 1 1 1 3 3 1 = - e-3t + e - t e -3t - e -t - - e -3t + e - t e -3t - e -t 2 2 2 2 2 2 2 2 = e -4 t 0

So, we got a completely different set of fundamental solutions from the theorem that what weve been using up to this point. This is not a problem. There are an infinite number of pairs of functions that we could use as a fundamental set of solutions for this problem. So, which set of fundamental solutions should we use? Well, if we use the ones that we originally found, the general solution would be,
y ( t ) = c1e -3t + c2 e -t
Whereas, if we used the set from the theorem the general solution would be,
3 1 1 1 y ( t ) = c1 - e -3t + e -t + c2 - e -3t + e -t 2 2 2 2
This would not be very fun to work with when it came to determining the coefficients to satisfy a general set of initial conditions. So, which set of fundamental solutions should we use? We should always try to use the set that is the most convenient to use for a given problem.
2007 Paul Dawkins
130
More on the Wronskian

In the previous section we introduced the Wronskian to help us determine whether two solutions were a fundamental set of solutions. In this section we will look at another application of the Wronskian as well as an alternate method of computing the Wronskian. Lets start with the application. We need to introduce a couple of new concepts first. Given two non-zero functions f(x) and g(x) write down the following equation.
c f ( x) + k g ( x) = 0
(1)
Notice that c = 0 and k = 0 will make (1) true for all x regardless of the functions that we use. Now, if we can find non-zero constants c and k for which (1) will also be true for all x then we call the two functions linearly dependent. On the other hand if the only two constants for which (1) is true are c = 0 and k = 0 then we call the functions linearly independent.
Example 1 Determine if the following sets of functions are linearly dependent or linearly
independent. (a) f ( x ) = 9cos ( 2 x ) (b) f ( t ) = 2t 2 Solution (a) f ( x ) = 9cos ( 2 x )
g ( x ) = 2 cos 2 ( x ) - 2sin 2 ( x ) [Solution] g ( t ) = t 4 [Solution]
g ( x ) = 2 cos 2 ( x ) - 2sin 2 ( x ) c ( 9cos ( 2 x ) ) + k ( 2cos 2 ( x ) - 2sin 2 ( x ) ) = 0
Well start by writing down (1) for these two functions.
We need to determine if we can find non-zero constants c and k that will make this true for all x or if c = 0 and k = 0 are the only constants that will make this true for all x. This is often a fairly difficult process. The process can be simplified with a good intuition for this kind of thing, but thats hard to come by, especially if you havent done many of these kinds of problems. In this case the problem can be simplified by recalling
cos 2 ( x ) - sin 2 ( x ) = cos ( 2 x ) 9c cos ( 2 x ) + 2k cos ( 2 x ) = 0
Using this fact our equation becomes.
( 9c + 2k ) cos ( 2 x ) = 0
With this simplification we can see that this will be zero for any pair of constants c and k that satisfy
9c + 2k = 0
Among the possible pairs on constants that we could use are the following pairs.
c = 1, 2 c= , 9 c = -2 7 c=6 etc.
k=-
9 2
k = -1 k =9 21 k= 4
As Im sure you can see there are literally thousands of possible pairs and they can be made as simple or as complicated as you want them to be. So, weve managed to find a pair of non-zero constants that will make the equation true for all x and so the two functions are linearly dependent.
[Return to Problems]
(b) f ( t ) = 2t 2
g (t ) = t 4 2ct 2 + kt 4 = 0
As with the last part, well start by writing down (1) for these functions.
In this case there isnt any quick and simple formula to write one of the functions in terms of the other as we did in the first part. So, were just going to have to see if we can find constants. Well start by noticing that if the original equation is true, then if we differentiate everything we get a new equation that must also be true. In other words, weve got the following system of two equations in two unknowns.
2ct 2 + kt 4 = 0 4ct + 4kt 3 = 0

We can solve this system for c and k and see what we get. Well start by solving the second equation for c.
c = - kt 2
Now, plug this into the first equation.
2 ( - kt 2 ) t 2 + kt 4 = 0 - kt 4 = 0
Recall that we are after constants that will make this true for all t. The only way that this will ever be zero for all t is if k = 0! So, if k = 0 we must also have c = 0. Therefore, weve shown that the only way that
2ct 2 + kt 4 = 0
will be true for all t is to require that c = 0 and k = 0. The two functions therefore, are linearly independent.
2007 Paul Dawkins
132
As we saw in the previous examples determining whether two functions are linearly independent or linearly dependent can be a fairly involved process. This is where the Wronskian can help. Fact Given two functions f(x) and g(x) that are differentiable on some interval I. (1) If W ( f , g )( x0 ) 0 for some x0 in I, then f(x) and g(x) are linearly independent on the interval I. (2) If f(x) and g(x) are linearly dependent on I then W(f,g)(x) = 0 for all x in the interval I. Be very careful with this fact. It DOES NOT say that if W(f,g)(x) = 0 then f(x) and g(x) are linearly dependent! In fact it is possible for two linearly independent functions to have a zero Wronskian! This fact is used to quickly identify linearly independent functions and functions that are liable to be linearly dependent.
Example 2 Verify the fact using the functions from the previous example.
Solution (a) f ( x ) = 9cos ( 2 x )
g ( x ) = 2 cos 2 ( x ) - 2sin 2 ( x )
In this case if we compute the Wronskian of the two functions we should get zero since we have already determined that these functions are linearly dependent.
W= = =
9cos ( 2 x ) 2 cos 2 ( x ) - 2sin 2 ( x ) -18sin ( 2 x ) -4cos ( x ) sin ( x ) - 4sin ( x ) cos ( x ) 9cos ( 2 x ) 2cos ( 2 x ) -18sin ( 2 x ) -2sin ( 2 x ) - 2sin ( 2 x ) 9cos ( 2 x ) 2cos ( 2 x ) -18sin ( 2 x ) -4sin ( 2 x )
= -36 cos ( 2 x ) sin ( 2 x ) - ( -36 cos ( 2 x ) sin ( 2 x ) ) = 0

So, we get zero as we should have. Notice the heavy use of trig formulas to simplify the work! (b) f ( t ) = 2t 2
g (t ) = t 4
Here we know that the two functions are linearly independent and so we should get a non-zero Wronskian.
t4 = 8t 5 - 4t 5 = 4t 5 3 4t The Wronskian is non-zero as we expected provided t 0 . This is not a problem. As long as the
Wronskian is not identically zero for all t we are okay.
2t 2 W= 4t
2007 Paul Dawkins
133
Example 3 Determine if the following functions are linearly dependent or linearly independent. (a) f ( t ) = cos t g ( t ) = sin t [Solution]
(b) f ( x ) = 6 x
g ( x ) = 6 x + 2 [Solution]
Solution (a) Now that we have the Wronskian to use here lets first check that. If its non-zero then we will know that the two functions are linearly independent and if its zero then we can be pretty sure that they are linearly dependent.
W=
cos t sin t = cos 2 t + sin 2 t = 1 0 - sin t cos t

So, by the fact these two functions are linearly independent. Much easier this time around! (b) Well do the same thing here as we did in the first part. Recall that
( a ) = a
x
ln a
Now compute the Wronskian.
W=
6x 6 ln 6 6
x
6x+2
x+2
ln 6
= 6 x 6 x + 2 ln 6 - 6 x + 2 6 x ln 6 = 0
Now, this does not say that the two functions are linearly dependent! However, we can guess that they probably are linearly dependent. To prove that they are in fact linearly dependent well need to write down (1) and see if we can find non-zero c and k that will make it true for all x.
c 6 x + k 6x+2 = 0 c 6 x + k 6 x 62 = 0 c 6 x + 36k 6 x = 0
( c + 36k ) 6 x = 0
So, it looks like we could use any constants that satisfy
c + 36k = 0
to make this zero for all x. In particular we could use
c = 36 c = -36 c=9 etc.
k = -1 k =1 1 k=4
We have non-zero constants that will make the equation true for all x. Therefore, the functions are linearly dependent.
2007 Paul Dawkins
134
Before proceeding to the next topic in this section lets talk a little more about linearly independent and linearly dependent functions. Lets start off by assuming that f(x) and g(x) are linearly dependent. So, that means there are non-zero constants c and k so that
c f ( x) + k g ( x) = 0
is true for all x. Now, we can solve this in either of the following two ways.
k f ( x) = - g ( x) c
OR
g ( x) = -
c f ( x) k
Note that this can be done because we know that c and k are non-zero and hence the divisions can be done without worrying about division by zero. So, this means that two linearly dependent functions can be written in such a way that one is nothing more than a constant time the other. Go back and look at both of the sets of linearly dependent functions that we wrote down and you will see that this is true for both of them. Two functions that are linearly independent cant be written in this manner and so we cant get from one to the other simply by multiplying by a constant. Next, we dont want to leave you with the impression that linear independence and linear dependence is only for two functions. We can easily extend the idea to as many functions as wed like. Lets suppose that we have n non-zero functions, f1(x), f2(x),, fn(x). Write down the following equation.
c1 f1 ( x ) + c2 f 2 ( x ) + L + cn f n ( x ) = 0
(2)
If we can find constants c1, c2, , cn with at least two non-zero so that (2) is true for all x then we call the functions linearly dependent. If, on the other hand, the only constants that make (2) true for x are c1 = 0, c2 = 0, , cn = 0 then we call the functions linearly independent. Note that unlike the two function case we can have some of the constants be zero and still have the functions be linearly dependent. In this case just what does it mean for the functions to be linearly dependent? Well, lets suppose that they are. So, this means that we can find constants, with at least two non-zero so that (2) is true for all x. For the sake of argument lets suppose that c1 is one of the non-zero constants. This means that we can do the following.
c1 f1 ( x ) + c2 f 2 ( x ) + L + cn f n ( x ) = 0
c1 f1 ( x ) = - ( c2 f 2 ( x ) + L + cn f n ( x ) ) f1 ( x ) = 1 ( c2 f2 ( x ) + L + cn f n ( x ) ) c1
In other words, if the functions are linearly dependent then we can write at least one of them in terms of the other functions.
2007 Paul Dawkins
135
Okay, lets move on the other topic of this section. There is an alternate method of computing the Wronskian. The following theorem gives this alternate method. Abels Theorem If y1(t) and y2(t) are two solutions to
y + p ( t ) y + q ( t ) y = 0
-
then the Wronskian of the two solutions is
W ( y1 , y2 )( t ) = W ( y1 , y2 )( t0 ) e
for some t0.
t0 p( x ) dx
Because we dont know the Wronskian and we dont know t0 this wont do us a lot of good apparently. However, we can rewrite this as
- p ( t ) dt W ( y1 , y2 )( t ) = c e
(3)
where the original Wronskian sitting in front of the exponential is absorbed into the c and the evaluation of the integral at t0 will put a constant in the exponential that can also be brought out and absorbed into the constant c. If you dont recall how to do this go back and take a look at the linear, first order differential equation section as we did something similar there. With this rewrite we can compute the Wronskian up to a multiplicative constant, which isnt too bad. Notice as well that we dont actually need the two solutions to do this. All we need is the coefficient of the first derivative from the differential equation (provided the coefficient of the second derivative is one of course). Lets take a look at a quick example of this.
Example 4 Without solving, determine the Wronskian of two solutions to the following
differential equation.
t 4 y - 2t 3 y - t 8 y = 0
Solution The first thing that we need to do is divide the differential equation by the coefficient of the second derivative as that needs to be a one. This gives us
y Now, using (3) the Wronskian is

2 - - dt t
2 y - t 4 y = 0 t = ce 2ln t = celn t = ct 2
2
W = ce
2007 Paul Dawkins
136
Nonhomogeneous Differential Equations

Its now time to start thinking about how to solve nonhomogeneous differential equations. A second order, linear nonhomogeneous differential equation is
y + p ( t ) y + q ( t ) y = g ( t )
(1)
where g(t) is a non-zero function. Note that we didnt go with constant coefficients here because everything that were going to do in this section doesnt require it. Also, were using a coefficient of 1 on the second derivative just to make some of the work a little easier to write down. It is not required to be a 1. Before talking about how to solve one of these we need to get some basics out of the way, which is the point of this section. First, we will call
y + p ( t ) y + q ( t ) y = 0
(2)
the associated homogeneous differential equation to (1). Now, lets take a look at the following theorem. Theorem Suppose that Y1(t) and Y2(t) are two solutions to (1) and that y1(t) and y2(t) are a fundamental set of solutions to the associated homogeneous differential equation (2) then,
Y1 ( t ) - Y2 ( t )
is a solution to (2) and it can be written as
Y1 ( t ) - Y2 ( t ) = c1 y1 ( t ) + c2 y2 ( t )
Note the notation used here. Capital letters referred to solutions to (1) while lower case letters referred to solutions to (2). This is a fairly common convention when dealing with nonhomogeneous differential equations. This theorem is easy enough to prove so lets do that. To prove that Y1(t) - Y2(t) is a solution to (2) all we need to do is plug this into the differential equation and check it.
(Y1 - Y2 ) + p ( t )(Y1 - Y2 ) + q ( t )(Y1 - Y2 ) = 0

Y1 + p ( t ) Y1 + q ( t ) Y1 - Y2 + p ( t ) Y2 + q ( t ) Y2 = 0 g (t ) - g (t ) = 0
We used the fact that Y1(t) and Y2(t) are two solutions to (1) in the third step. Because they are solutions to (1) we know that
0=0
Y1 + p ( t ) Y1 + q ( t ) Y1 = g ( t ) Y2 + p ( t ) Y2 + q ( t ) Y2 = g ( t )
So, we were able to prove that the difference of the two solutions is a solution to (2).
2007 Paul Dawkins
137
Proving that
Y1 ( t ) - Y2 ( t ) = c1 y1 ( t ) + c2 y2 ( t ) y ( t ) = c1 y1 ( t ) + c2 y2 ( t )
is even easier. Since y1(t) and y2(t) are a fundamental set of solutions to (2) we know that they form a general solution and so any solution to (2) can be written in the form
Well, Y1(t) - Y2(t) is a solution to (2), as weve shown above, therefore it can be written as
Y1 ( t ) - Y2 ( t ) = c1 y1 ( t ) + c2 y2 ( t )
So, what does this theorem do for us? We can use this theorem to write down the form of the general solution to (1). Lets suppose that y(t) is the general solution to (1) and that YP(t) is any solution to (1) that we can get our hands on. Then using the second part of our theorem we know that
y ( t ) - YP ( t ) = c1 y1 ( t ) + c2 y2 ( t )
where y1(t) and y2(t) are a fundamental set of solutions for (2). Solving for y(t) gives,
y ( t ) = c1 y1 ( t ) + c2 y2 ( t ) + YP ( t ) yc ( t ) = c1 y1 ( t ) + c2 y2 ( t ) y ( t ) = yc ( t ) + YP ( t )
We will call
the complimentary solution and YP(t) a particular solution. The general solution to a differential equation can then be written as.
So, to solve a nonhomogeneous differential equation, we will need to solve the homogeneous differential equation, (2), which for constant coefficient differential equations is pretty easy to do, and well need a solution to (1). This seems to be a circular argument. In order to write down a solution to (1) we need a solution. However, this isnt the problem that it seems to be. There are ways to find a solution to (1). They just wont, in general, be the general solution. In fact, the next two sections are devoted to exactly that, finding a particular solution to a nonhomogeneous differential equation. There are two common methods for finding particular solutions : Undetermined Coefficients and Variation of Parameters. Both have their advantages and disadvantages as you will see in the next couple of sections.
2007 Paul Dawkins
138
Undetermined Coefficients
In this section we will take a look at the first method that can be used to find a particular solution to a nonhomogeneous differential equation.
y + p ( t ) y + q ( t ) y = g ( t )
One of the main advantages of this method is that it reduces the problem down to an algebra problem. The algebra can get messy on occasion, but for most of the problems it will not be terribly difficult. Another nice thing about this method is that the complimentary solution will not be explicitly required, although as we will see knowledge of the complimentary solution will be needed in some cases and so well generally find that as well. There are two disadvantages to this method. First, it will only work for a fairly small class of g(t)s. The class of g(t)s for which the method works, does include some of the more common functions, however, there are many functions out there for which undetermined coefficients simply wont work. Second, it is generally only useful for constant coefficient differential equations. The method is quite simple. All that we need to do is look at g(t) and make a guess as to the form of YP(t) leaving the coefficient(s) undetermined (and hence the name of the method). Plug the guess into the differential equation and see if we can determine values of the coefficients. If we can determine values for the coefficients then we guessed correctly, if we cant find values for the coefficients then we guessed incorrectly. Its usually easier to see this method in action rather than to try and describe it, so lets jump into some examples.
Example 1 Determine a particular solution to y - 4 y - 12 y = 3e5t

Solution The point here is to find a particular solution, however the first thing that were going to do is find the complimentary solution to this differential equation. Recall that the complimentary solution comes from solving,
y - 4 y - 12 y = 0
The characteristic equation for this differential equation and its roots are.
r 2 - 4r - 12 = ( r - 6 )( r + 2 ) = 0
r1 = -2, r2 = 6
The complimentary solution is then,
yc ( t ) = c1e-2t + c2e 6t
At this point the reason for doing this first will not be apparent, however we want you in the habit of finding it before we start the work to find a particular solution. Eventually, as well see, having the complimentary solution in hand will be helpful and so its best to be in the habit of finding it first prior to doing the work for undetermined coefficients. Now, lets proceed with finding a particular solution. As mentioned prior to the start of this
example we need to make a guess as to the form of a particular solution to this differential equation. Since g(t) is an exponential and we know that exponentials never just appear or disappear in the differentiation process it seems that a likely form of the particular solution would be
YP ( t ) = Ae5t
Now, all that we need to do is do a couple of derivatives, plug this into the differential equation and see if we can determine what A needs to be. Plugging into the differential equation gives
25 Ae5t - 4 ( 5 Ae5 t ) - 12 ( Ae5t ) = 3e5t -7 Ae5t = 3e5t
So, in order for our guess to be a solution we will need to choose A so that the coefficients of the exponentials on either side of the equal sign are the same. In other words we need to choose A so that,
-7 A = 3
A=-
3 7
Okay, we found a value for the coefficient. This means that we guessed correctly. A particular solution to the differential equation is then,
3 YP ( t ) = - e5t 7
Before proceeding any further lets again note that we started off the solution above by finding the complimentary solution. This is not technically part the method of Undetermined Coefficients however, as well eventually see, having this in hand before we make our guess for the particular solution can save us a lot of work and/or headache. Finding the complimentary solution first is simply a good habit to have so well try to get you in the habit over the course of the next few examples. At this point do not worry about why it is a good habit. Well eventually see why it is a good habit. Now, back to the work at hand. Notice in the last example that we kept saying a particular solution, not the particular solution. This is because there are other possibilities out there for the particular solution weve just managed to find one of them. Any of them will work when it comes to writing down the general solution to the differential equation. Speaking of which This section is devoted to finding particular solutions and most of the examples will be finding only the particular solution. However, we should do at least one full blown IVP to make sure that we can say that weve done one.
Example 2 Solve the following IVP y - 4 y - 12 y = 3e5t y (0) = 18 7 y ( 0 ) = 1 7
Solution We know that the general solution will be of the form,
y ( t ) = yc ( t ) + YP ( t )
2007 Paul Dawkins
and we already have both the complimentary and particular solution from the first example so we dont really need to do any extra work for this problem. One of the more common mistakes in these problems is to find the complimentary solution and then, because were probably in the habit of doing it, apply the initial conditions to the complimentary solution to find the constants. This however, is incorrect. The complimentary solution is only the solution to the homogeneous differential equation and we are after a solution to the nonhomogeneous differential equation and the initial conditions must satisfy that solution instead of the complimentary solution. So, we need the general solution to the nonhomogeneous differential equation. Taking the complimentary solution and the particular solution that we found in the previous example we get the following for a general solution and its derivative.
3 y ( t ) = c1e-2 t + c2e6 t - e5t 7 15 y ( t ) = -2c1e-2t + 6c2e 6t - e5t 7

Now, apply the initial conditions to these.
18 3 = y ( 0 ) = c1 + c2 7 7 1 15 - = y ( 0 ) = -2c1 + 6c2 7 7
Solving this system gives c1 = 2 and c2 = 1. The actual solution is then.
3 y ( t ) = 2e -2t + e 6t - e5t 7
This will be the only IVP in this section so dont forget how these are done for nonhomogeneous differential equations! Lets take a look at another example that will give the second type of g(t) for which undetermined coefficients will work.
Example 3 Find a particular solution for the following differential equation. y - 4 y - 12 y = sin ( 2t )
Solution
Again, lets note that we should probably find the complimentary solution before we proceed onto the guess for a particular solution. However, because the homogeneous differential equation for this example is the same as that for the first example we wont bother with that here. Now, lets take our experience from the first example and apply that here. The first example had an exponential function in the g(t) and our guess was an exponential. This differential equation has a sine so lets try the following guess for the particular solution. YP ( t ) = A sin ( 2t )
Differentiating and plugging into the differential equation gives,
-4 A sin ( 2t ) - 4 ( 2 A cos ( 2t ) ) - 12 ( A sin ( 2t ) ) = sin ( 2t ) -16 A sin ( 2t ) - 8 A cos ( 2t ) = sin ( 2t )
Collecting like terms yields
We need to pick A so that we get the same function on both sides of the equal sign. This means that the coefficients of the sines and cosines must be equal. Or,
cos ( 2t ) :
- 8A = 0 -16 A = 1
A=0 A=1 16
sin ( 2t ) :
Notice two things. First, since there is no cosine on the right hand side this means that the coefficient must be zero on that side. More importantly we have a serious problem here. In order for the cosine to drop out, as it must in order for the guess to satisfy the differential equation, we need to set A = 0, but if A = 0, the sine will also drop out and that cant happen. Likewise, choosing A to keep the sine around will also keep the cosine around. What this means is that our initial guess was wrong. If we get multiple values of the same constant or are unable to find the value of a constant then we have guessed wrong. One of the nicer aspects of this method is that when we guess wrong our work will often suggest a fix. In this case the problem was the cosine that cropped up. So, to counter this lets add a cosine to our guess. Our new guess is
YP ( t ) = A cos ( 2t ) + B sin ( 2t )
Plugging this into the differential equation and collecting like terms gives,
-4 A cos ( 2t ) - 4 B sin ( 2t ) - 4 ( -2 A sin ( 2t ) + 2 B cos ( 2t ) ) -
12 ( A cos ( 2t ) + B sin ( 2t ) ) = sin ( 2t )
( -4 A - 8B - 12 A ) cos ( 2t ) + ( -4 B + 8 A - 12 B ) sin ( 2t ) = sin ( 2t ) ( -16 A - 8B ) cos ( 2t ) + (8 A - 16 B ) sin ( 2t ) = sin ( 2t )

Now, set the coefficients equal
cos ( 2t ) : sin ( 2t ) :
-16 A - 8B = 0 8 A - 16 B = 1 B=1 20
Solving this system gives us
A=
1 40
We found constants and this time we guessed correctly. A particular solution to the differential equation is then,
YP ( t ) =
2007 Paul Dawkins
1 1 cos ( 2t ) - sin ( 2t ) 40 20
Notice that if we had had a cosine instead of a sine in the last example then our guess would have been the same. In fact, if both a sine and a cosine had shown up we will see that the same guess will also work. Lets take a look at the third and final type of basic g(t) that we can have. There are other types of g(t) that we can have, but as we will see they will all come back to two types that weve already done as well as the next one.
Example 4 Find a particular solution for the following differential equation. y - 4 y - 12 y = 2t 3 - t + 3

Solution Once, again we will generally want the complimentary solution in hand first, but again were working with the same homogeneous differential equation (youll eventually see why we keep working with the same homogeneous problem) so well again just refer to the first example. For this example g(t) is a cubic polynomial. For this we will need the following guess for the particular solution.
YP ( t ) = At 3 + Bt 2 + Ct + D
Notice that even though g(t) doesnt have a t2 in it our guess will still need one! So, differentiate and plug into the differential equation.
6 At + 2 B - 4 ( 3 At 2 + 2 Bt + C ) - 12 ( At 3 + Bt 2 + Ct + D ) = 2t 3 - t + 3
-12 At 3 + ( -12 A - 12 B ) t 2 + ( 6 A - 8B - 12C ) t + 2 B - 4C - 12 D = 2t 3 - t + 3

Now, as weve done in the previous examples we will need the coefficients of the terms on both sides of the equal sign to be the same so set coefficients equal and solve.
t3 : t2 : t1 : t0 :
- 12 A = 2 - 12 A - 12 B = 0 6 A - 8 B - 12C = -1 2 B - 4C - 12 D = 3
A=B= 1 6
1 6
C=-
1 9 5 D=27
Notice that in this case it was very easy to solve for the constants. The first equation gave A. Then once we knew A the second equation gave B, etc. A particular solution for this differential equation is then
1 1 1 5 YP ( t ) = - t 3 + t 2 - t 6 6 9 27
Now that weve gone over the three basic kinds of functions that we can use undetermined coefficients on lets summarize.
2007 Paul Dawkins
143
g(t)
YP(t) guess
ae a cos ( b t ) b sin ( b t ) a cos ( b t ) + b sin ( b t )

n degree polynomial
th
bt
Ae b t A cos ( b t ) + B sin ( b t ) A cos ( b t ) + B sin ( b t ) A cos ( b t ) + B sin ( b t ) An t n + An -1t n-1 + L A1t + A0
Notice that there are really only three kinds of functions given above. If you think about it the single cosine and single sine functions are really special cases of the case where both the sine and cosine are present. Also, we have not yet justified the guess for the case where both a sine and a cosine show up. We will justify this later. We now need move on to some more complicated functions. The more complicated functions arise by taking products and sums of the basic kinds of functions. Lets first look at products.
Example 5 Find a particular solution for the following differential equation. y - 4 y - 12 y = te4t
Solution Youre probably getting tired of the opening comment, but again find the complimentary solution first really a good idea but again weve already done the work in the first example so we wont do it again here. We promise that eventually youll see why we keep using the same homogeneous problem and why we say its a good idea to have the complimentary solution in hand first. At this point all were trying to do is reinforce the habit of finding the complimentary solution first. Okay, lets start off by writing down the guesses for the individual pieces of the function. The guess for the t would be
At + B C e 4t
while the guess for the exponential would be
Now, since weve got a product of two functions it seems like taking a product of the guesses for the individual pieces might work. Doing this would give
Ce4t ( At + B )
However, we will have problems with this. As we will see, when we plug our guess into the differential equation we will only get two equations out of this. The problem is that with this guess weve got three unknown constants. With only two equations we wont be able to solve for all the constants. This is easy to fix however. Lets notice that we could do the following
Ce4 t ( At + B ) = e 4 t ( ACt + BC )
If we multiply the C through, we can see that the guess can be written in such a way that there are
really only two constants. So, we will use the following for our guess.
YP ( t ) = e 4t ( At + B )
Notice that this is nothing more than the guess for the t with an exponential tacked on for good measure. Now that weve got our guess, lets differentiate, plug into the differential equation and collect like terms.
e 4t (16 At + 16 B + 8 A) - 4 ( e 4t ( 4 At + 4 B + A) ) - 12 ( e4 t ( At + B ) ) = te 4 t
(16 A - 16 A - 12 A) te4t + (16 B + 8 A - 16 B - 4 A - 12 B ) e4t = te4t -12 Ate 4 t + ( 4 A - 12 B ) e 4t = te 4 t

Note that when were collecting like terms we want the coefficient of each term to have only constants in it. Following this rule we will get two terms when we collect like terms. Now, set coefficients equal.
te 4 t : e4t :
- 12 A = 1 4 A - 12 B = 0
1 12 1 B=36 A=-
A particular solution for this differential equation is then
1 1 t YP ( t ) = e 4t - - = - ( 3t + 1) e 4t 36 12 36
This last example illustrated the general rule that we will follow when products involve an exponential. When a product involves an exponential we will first strip out the exponential and write down the guess for the portion of the function without the exponential, then we will go back and tack on the exponential without any leading coefficient. Lets take a look at some more products. In the interest of brevity we will just write down the guess for a particular solution and not go through all the details of finding the constants. Also, because we arent going to give an actual differential equation we cant deal with finding the complimentary solution first.
Example 6 Write down the form of the particular solution to y + p ( t ) y + q ( t ) y = g ( t )

for the following g(t)s. (a) g ( t ) = 16e7 t sin (10t ) [Solution] (b) g ( t ) = 9t 2 - 103t cos t (c) g ( t ) = -e
2007 Paul Dawkins
-2 t
[Solution] [Solution]
( 3 - 5t ) cos ( 9t )
145
Solution (a) g ( t ) = 16e7 t sin (10t ) So, we have an exponential in the function. Remember the rule. We will ignore the exponential and write down a guess for 16 sin(10t) then put the exponential back in. The guess for the sine is
A cos (10t ) + B sin (10t )
Now, for the actual guess for the particular solution well take the above guess and tack an exponential onto it. This gives,
YP ( t ) = e7 t ( A cos (10t ) + B sin (10t ) )
One final note before we move onto the next part. The 16 in front of the function has absolutely no bearing on our guess. Any constants multiplying the whole function are ignored.
2 (b) g ( t ) = 9t - 103t cos t
We will start this one the same way that we initially started the previous example. The guess for the polynomial is
At 2 + Bt + C
and the guess for the cosine is If we multiply the two guesses we get.
D cos t + E sin t
( At
+ Bt + C ) ( D cos t + E sin t )
Lets simplify things up a little. First multiply the polynomial through as follows.
( At + Bt + C ) ( D cos t ) + ( At + Bt + C ) ( E sin t ) ( ADt + BDt + CD ) cos t + ( AEt + BEt + CE ) sin t

2 2 2 2
Notice that everywhere one of the unknown constants occurs it is in a product of unknown constants. This means that if we went through and used this as our guess the system of equations that we would need to solve for the unknown constants would have products of the unknowns in them. These types of systems are generally very difficult to solve. So, to avoid this we will do the same thing that we did in the previous example. Everywhere we see a product of constants we will rename it and call it a single constant. The guess that well use for this function will be.
YP ( t ) = ( At 2 + Bt + C ) cos t + ( Dt 2 + Et + F ) sin t
This is a general rule that we will use when faced with a product of a polynomial and a trig function. We write down the guess for the polynomial and then multiply that by a cosine. We then write down the guess for the polynomial again, using different coefficients, and multiply this by a sine.
[Return to Problems] 2007 Paul Dawkins 146 http://tutorial.math.lamar.edu/terms.aspx
(c) g ( t ) = -e -2 t ( 3 - 5t ) cos ( 9t ) This final part has all three parts to it. First we will ignore the exponential and write down a guess for.
- ( 3 - 5t ) cos ( 9t )
The minus sign can also be ignored. The guess for this is
( At + B ) cos ( 9t ) + ( Ct + D ) sin ( 9t )
Now, tack an exponential back on and were done.
YP ( t ) = e -2t ( At + B ) cos ( 9t ) + e -2t ( Ct + D ) sin ( 9t )

Notice that we put the exponential on both terms. There a couple of general rules that you need to remember for products. 1. If g(t) contains an exponential, ignore it and write down the guess for the remainder. Then tack the exponential back on without any leading coefficient. 2. For products of polynomials and trig functions you first write down the guess for just the polynomial and multiply that by the appropriate cosine. Then add on a new guess for the polynomial with different coefficients and multiply that by the appropriate sine. If you can remember these two rules you cant go wrong with products. Writing down the guesses for products is usually not that difficult. The difficulty arises when you need to actually find the constants. Now, lets take a look at sums of the basic components and/or products of the basic components. To do this well need the following fact. Fact If YP1(t) is a particular solution for
y + p ( t ) y + q ( t ) y = g1 ( t )
and if YP2(t) is a particular solution for
y + p ( t ) y + q ( t ) y = g 2 ( t )
then YP1(t)+ YP2(t) is a particular solution for
y + p ( t ) y + q ( t ) y = g1 ( t ) + g 2 ( t )
This fact can be used to both find particular solutions to differential equations that have sums in then and to write down guess for functions that have sums in them.
2007 Paul Dawkins
147
Example 7 Find a particular solution for the following differential equation. y - 4 y - 12 y = 3e5t + sin ( 2t ) + te 4t
Solution This example is the reason that weve been using the same homogeneous differential equation for all the previous examples. There is nothing to do with this problem. All that we need to do it go back to the appropriate examples above and get the particular solution from that example and add them all together. Doing this gives
3 1 1 1 YP ( t ) = - e5t + cos ( 2t ) - sin ( 2t ) - ( 3t + 1) e 4 t 7 40 20 36

Lets take a look at a couple of other examples. As with the products well just get guesses here and not worry about actually finding the coefficients.
Example 8 Write down the form of the particular solution to y + p ( t ) y + q ( t ) y = g ( t )

for the following g(t)s. (a) g ( t ) = 4cos ( 6t ) - 9sin ( 6t ) [Solution] (b) g ( t ) = -2sin t + sin (14t ) - 5cos (14t ) [Solution] (c) g ( t ) = e 7 t + 6 [Solution] (d) g ( t ) = 6t 2 - 7sin ( 3t ) + 9 [Solution] (e) g ( t ) = 10et - 5te -8 t + 2e -8 t (f) g ( t ) = t cos t - 5t sin t
2
[Solution]
[Solution]
(g) g ( t ) = 5e
-3t
+e
-3 t
cos ( 6t ) - sin ( 6t ) [Solution]
Solution (a) g ( t ) = 4cos ( 6t ) - 9sin ( 6t ) This first one weve actually already told you how to do. This is in the table of the basic functions. However we wanted to justify the guess that we put down there. Using the fact on sums of function we would be tempted to write down a guess for the cosine and a guess for the sine. This would give.
A cos ( 6t ) + B sin ( 6t ) + C cos ( 6t ) + D sin ( 6t ) 1444 2444 3 144424443

guess for the cosine guess for the sine
So, we would get a cosine from each guess and a sine from each guess. The problem with this as a guess is that we are only going to get two equations to solve after plugging into the differential equation and yet we have 4 unknowns. We will never be able to solve for each of the constants. To fix this notice that we can combine some terms as follows.
( A + C ) cos ( 6t ) + ( B + D ) sin ( 6t )
2007 Paul Dawkins
Upon doing this we can see that weve really got a single cosine with a coefficient and a single sine with a coefficient and so we may as well just use
YP ( t ) = A cos ( 6t ) + B sin ( 6t )
The general rule of thumb for writing down guesses for functions that involve sums is to always combine like terms into single terms with single coefficients. This will greatly simplify the work required to find the coefficients.
(b) g ( t ) = -2sin t + sin (14t ) - 5cos (14t ) For this one we will get two sets of sines and cosines. This will arise because we have two different arguments in them. We will get on set for the sine with just a t as its argument and well get another set for the sine and cosine with the 14t as their arguments. The guess for this function is
YP ( t ) = A cos t + B sin t + C cos (14t ) + D sin (14t )

(c) g ( t ) = e 7 t + 6 The main point of this problem is dealing with the constant. But that isnt too bad. We just wanted to make sure that an example of that is somewhere in the notes. If you recall that a constant is nothing more than a zeroth degree polynomial the guess becomes clear. The guess for this function is
Y p ( t ) = Ae 7 t + B
(d) g ( t ) = 6t 2 - 7sin ( 3t ) + 9 This one can be a little tricky if you arent paying attention. Lets first rewrite the function
g ( t ) = 6t 2 - 7 sin ( 3t ) + 9
as
g ( t ) = 6t 2 + 9 - 7 sin ( 3t )
All we did was move the 9. However upon doing that we see that the function is really a sum of a quadratic polynomial and a sine. The guess for this is then
YP ( t ) = At 2 + Bt + C + D cos ( 3t ) + E sin ( 3t ) At 2 + Bt + C + D cos ( 3t ) + E sin ( 3t ) + G
If we dont do this and treat the function as the sum of three terms we would get and as with the first part in this example we would end up with two terms that are essentially the same (the C and the G) and so would need to be combined. An added step that isnt really necessary if we first rewrite the function. Look for problems where rearranging the function can simplify the initial guess.
(e) g ( t ) = 10et - 5te -8 t + 2e -8 t So, this look like weve got a sum of three terms here. Lets write down a guess for that.
Aet + ( Bt + C ) e -8 t + De -8t
Notice however that if we were to multiply the exponential in the second term through we would end up with two terms that are essentially the same and would need to be combined. This is a case where the guess for one term is completely contained in the guess for a different term. When this happens we just drop the guess thats already included in the other term. So, the guess here is actually.
YP ( t ) = Aet + ( Bt + C ) e-8t
Notice that this arose because we had two terms in our g(t) whose only difference was the polynomial that sat in front of them. When this happens we look at the term that contains the largest degree polynomial, write down the guess for that and dont bother writing down the guess for the other term as that guess will be completely contained in the first guess.
(f) g ( t ) = t 2 cos t - 5t sin t In this case weve got two terms whose guess without the polynomials in front of them would be the same. Therefore, we will take the one with the largest degree polynomial in front of it and write down the guess for that one and ignore the other term. So, the guess for the function is
YP ( t ) = ( At 2 + Bt + C ) cos t + ( Dt 2 + Et + F ) sin t
(g) g ( t ) = 5e -3t + e -3t cos ( 6t ) - sin ( 6t ) This last part is designed to make sure you understand the general rule that we used in the last two parts. This time there really are three terms and we will need a guess for each term. The guess here is
YP ( t ) = Ae -3t + e-3t ( B cos ( 6t ) + C sin ( 6t ) ) + D cos ( 6t ) + E sin ( 6t )
We can only combine guesses if they are identical up to the constant. So we cant combine the first exponential with the second because the second is really multiplied by a cosine and a sine and so the two exponentials are in fact different functions. Likewise, the last sine and cosine cant be combined with those in the middle term because the sine and cosine in the middle term are in fact multiplied by an exponential and so are different.
So, when dealing with sums of functions make sure that you look for identical guesses that may or may not be contained in other guesses and combine them. This will simplify your work later on.
2007 Paul Dawkins
150
We have one last topic in this section that needs to be dealt with. In the first few examples we were constantly harping on the usefulness of having the complimentary solution in hand before making the guess for a particular solution. We never gave any reason for this other that trust us. It is now time to see why having the complimentary solution in hand first it useful. This is best shown with an example so lets jump into one.
Example 9 Find a particular solution for the following differential equation. y - 4 y - 12 y = e6t
Solution This problem seems almost too simple to be given this late in the section. This is especially true given the ease of finding a particular solution for g(t)s that are just exponential functions. Also, because the point of this example is to illustrate why it is generally a good idea to have the complimentary solution in hand first well lets go ahead and recall the complimentary solution first. Here it is,
yc ( t ) = c1e-2t + c2e 6t
Now, without worrying about the complimentary solution for a couple more seconds lets go ahead and get to work on the particular solution. There is not much to the guess here. From our previous work we know that the guess for the particular solution should be,
YP ( t ) = Ae6 t
Plugging this into the differential equation gives,
36 Ae 6t - 24 Ae6t - 12 Ae6t = e 6t 0 = e 6t
Hmmmm. Something seems wrong here. Clearly an exponential cant be zero. So, what went wrong? We finally need the complimentary solution. Notice that the second term in the complimentary solution (listed above) is exactly our guess for the form of the particular solution and now recall that both portions of the complimentary solution are solutions to the homogeneous differential equation,
y - 4 y - 12 y = 0
In other words, we had better have gotten zero by plugging our guess into the differential equation, it is a solution to the homogeneous differential equation! So, how do we fix this? The way that we fix this is to add a t to our guess as follows.
YP ( t ) = Ate 6t
Plugging this into our differential equation gives,
(12 Ae
6t
+ 36 Ate6 t ) - 4 ( Ae6 t + 6 Ate 6t ) - 12 Ate 6 t = e 6t
( 36 A - 24 A - 12 A) te6t + (12 A - 4 A) e6t = e6t

8 Ae 6t = e 6t
Now, we can set coefficients equal.
8A = 1
A=
1 8
So, the particular solution in this case is,
t YP ( t ) = e6 t 8
So, what did we learn from this last example. While technically we dont need the complimentary solution to do undetermined coefficients, you can go through a lot of work only to figure out at the end that you needed to add in a t to the guess because it appeared in the complimentary solution. This work is avoidable if we first find the complimentary solution and comparing our guess to the complimentary solution and seeing if any portion of your guess shows up in the complimentary solution. If a portion of your guess does show up in the complimentary solution then well need to modify that portion of the guess by adding in a t to the portion of the guess that is causing the problems. We do need to be a little careful and make sure that we add the t in the correct place however. The following set of examples will show you how to do this.
Example 10 Write down the guess for the particular solution to the given differential equation.
Do not find the coefficients. (a) y + 3 y - 28 y = 7t + e -7 t - 1 [Solution] (b) y - 100 y = 9t 2e10 t + cos t - t sin t
[Solution]
t 2 t t (d) 4 y + 16 y + 17 y = e -2t sin + 6t cos [Solution] 2 2 (e) y + 8 y + 16 y = e -4t + ( t 2 + 5 ) e -4 t [Solution]

(c) 4 y + y = e -2t sin + 6t cos [Solution] Solution In these solutions well leave the details of checking the complimentary solution to you. (a) y + 3 y - 28 y = 7t + e -7 t - 1 The complimentary solution is
t 2
yc ( t ) = c1e4t + c2e -7t YP ( t ) = At + B + Ce -7 t
Remembering to put the -1 with the 7t gives a first guess for the particular solution.
Notice that the last term in the guess is the last term in the complimentary solution. The first two terms however arent a problem and dont appear in the complimentary solution. Therefore, we will only add a t onto the last term. The correct guess for the form of the particular solution is.
YP ( t ) = At + B + Cte-7 t
2007 Paul Dawkins
152
(b) y - 100 y = 9t 2e10 t + cos t - t sin t The complimentary solution is
yc ( t ) = c1e10t + c2 e -10t
A first guess for the particular solution is
YP ( t ) = ( At 2 + Bt + C ) e10t + ( Et + F ) cos t + ( Gt + H ) sin t
Notice that if we multiplied the exponential term through the parenthesis that we would end up getting part of the complimentary solution showing up. Since the problem part arises from the first term the whole first term will get multiplied by t. The second and third terms are okay as they are. The correct guess for the form of the particular solution in this case is.
YP ( t ) = t ( At 2 + Bt + C ) e10t + ( Et + F ) cos t + ( Gt + H ) sin t
So, in general, if you were to multiply out a guess and any part of a that shows up in the complimentary solution, then the whole term will get a t not just the problem portion of the term.
-2 t (c) 4 y + y = e sin + 6t cos
t 2
t 2
The complimentary solution is
t t yc ( t ) = c1 cos + c2 sin 2 2
A first guess for the particular solution is
t t t t YP ( t ) = e -2t A cos + B sin + ( Ct + D ) cos + ( Et + F ) sin 2 2 2 2

In this case both the second and third terms contain portions of the complimentary solution. The first term doesnt however, since upon multiplying out, both the sine and the cosine would have an exponential with them and that isnt part of the complimentary solution. We only need to worry about terms showing up in the complimentary solution if the only difference between the complimentary solution term and the particular guess term is the constant in front of them. So, in this case the second and third terms will get a t while the first wont The correct guess for the form of the particular solution is.
t t t t YP ( t ) = e -2t A cos + B sin + t ( Ct + D ) cos + t ( Et + F ) sin 2 2 2 2

2007 Paul Dawkins
153
(d) 4 y + 16 y + 17 y = e -2t sin + 6t cos To get this problem we changed the differential equation from the last example and left the g(t) alone. The complimentary solution this time is
t 2
t 2
t t yc ( t ) = c1e -2t cos + c2e -2t sin 2 2

As with the last part, a first guess for the particular solution is
t t t t YP ( t ) = e -2t A cos + B sin + ( Ct + D ) cos + ( Et + F ) sin 2 2 2 2

This time however it is the first term that causes problems and not the second or third. In fact, the first term is exactly the complimentary solution and so it will need a t. Recall that we will only have a problem with a term in our guess if it only differs from the complimentary solution by a constant. The second and third terms in our guess dont have the exponential in them and so they dont differ from the complimentary solution by only a constant. The correct guess for the form of the particular solution is.
t t t t YP ( t ) = te -2t A cos + B sin + ( Ct + D ) cos + ( Et + F ) sin 2 2 2 2

-4 t 2 -4 t (e) y + 8 y + 16 y = e + t + 5 e
The complimentary solution is
yc ( t ) = c1e-4 t + c2te -4t
The two terms in g(t) are identical with the exception of a polynomial in front of them. So this means that we only need to look at the term with the highest degree polynomial in front of it. A first guess for the particular solution is
YP ( t ) = ( At 2 + Bt + C ) e -4t
Notice that if we multiplied the exponential term through the parenthesis the last two terms would be the complimentary solution. Therefore, we will need to multiply this whole thing by a t. The next guess for the particular solution is then.
YP ( t ) = t ( At 2 + Bt + C ) e -4 t
This still causes problems however. If we multiplied the t and the exponential through, the last term will still be in the complimentary solution. In this case, unlike the previous ones, a t wasnt sufficient to fix the problem. So, we will add in another t to our guess. The correct guess for the form of the particular solution is.
YP ( t ) = t 2 ( At 2 + Bt + C ) e-4t
Upon multiplying this out none of the terms are in the complimentary solution and so it will be okay.
As this last set of examples has shown, we really should have the complimentary solution in hand before even writing down the first guess for the particular solution. By doing this we can compare our guess to the complimentary solution and if any of the terms from your particular solution show up we will know that well have problems. Once the problem is identified we can add a t to the problem term(s) and compare our new guess to the complimentary solution. If there are no problems we can proceed with the problem, if there are problems add in another t and compare again. Can you see a general rule as to when a t will be needed and when a t2 will be needed for second order differential equations?
2007 Paul Dawkins
155
Variation of Parameters
In the last section we looked at the method of undetermined coefficients for finding a particular solution to
p ( t ) y + q ( t ) y + r ( t ) y = g ( t )
(1)
and we saw that while it reduced things down to just an algebra problem, the algebra could become quite messy. On top of that undetermined coefficients will only work for a fairly small class of functions. The method of Variation of Parameters is a much more general method that can be used in many more cases. However, there are two disadvantages to the method. First, the complimentary solution is absolutely required to do the problem. This is in contrast to the method of undetermined coefficients where it was advisable to have the complimentary solution on hand, but was not required. Second, as we will see, in order to complete the method we will be doing a couple of integrals and there is no guarantee that we will be able to do the integrals. So, while it will always be possible to write down a formula to get the particular solution, we may not be able to actually find it if the integrals are too difficult or if we are unable to find the complimentary solution. Were going to derive the formula for variation of parameters. Well start off by acknowledging that the complimentary solution to (1) is
yc ( t ) = c1 y1 ( t ) + c2 y2 ( t )
Remember as well that this is the general solution to the homogeneous differential equation.
p ( t ) y + q ( t ) y + r ( t ) y = 0
(2)
Also recall that in order to write down the complimentary solution we know that y1(t) and y2(t) are a fundamental set of solutions. What were going to do is see if we can find a pair of functions, u1(t) and u2(t) so that
YP ( t ) = u 1 ( t ) y1 ( t ) + u 2 ( t ) y2 ( t )
will be a solution to (1). We have two unknowns here and so well need two equations eventually. One equation is easy. Our proposed solution must satisfy the differential equation, so well get the first equation by plugging our proposed solution into (1). The second equation can come from a variety of places. We are going to get our second equation simply by making an assumption that will make our work easier. Well say more about this shortly. So, lets start. If were going to plug our proposed solution into the differential equation were going to need some derivatives so lets get those. The first derivative is
y1 + u1 y1 + u2 y2 + u2 y2 YP ( t ) = u1
Heres the assumption. Simply to make the first derivative easier to deal with we are going to assume that whatever u1(t) and u2(t) are they will satisfy the following.
y1 + u2 y2 = 0 u1
(3)
2007 Paul Dawkins
156
Now, there is no reason ahead of time to believe that this can be done. However, we will see that this will work out. We simply make this assumption on the hope that it wont cause problems down the road and to make the first derivative easier so dont get excited about it. With this assumption the first derivative becomes.
+ u2 y2 YP ( t ) = u1 y1
The second derivative is then,
y1 + u1 y1 + u2 y2 + u2 y2 YP ( t ) = u1
Plug the solution and its derivatives into (1).
y1 + u1 y1 + u2 y2 + u2 y2 ) + q ( t )( u1 y1 + u2 y p ( t )( u1 2 ) + r ( t )( u 1 y1 + u 2 y2 ) = g ( t ) y1 + u2 y2 ) + u1 ( t ) ( p ( t ) y1 + q ( t ) y1 + r ( t ) y1 ) + p ( t )( u1 u2 ( t ) ( p ( t ) y 2 + q ( t ) y2 + r ( t ) y2 ) = g ( t )
Rearranging a little gives the following.
Now, both y1(t) and y2(t) are solutions to (2) and so the second and third terms are zero. Acknowledging this and rearranging a little gives us,
y1 + u2 y2 ) + u1 ( t )( 0 ) + u2 ( t )( 0 ) = g ( t ) p ( t )( u1 y1 + u2 y2 = u1 g (t ) p (t )
(4)
Weve almost got the two equations that we need. Before proceeding were going to go back and make a further assumption. The last equation, (4), is actually the one that we want, however, in order to make things simpler for us we are going to assume that the function p(t) = 1. In other words, we are going to go back and start working with the differential equation,
y + q ( t ) y + r ( t ) y = g ( t )
If the coefficient of the second derivative isnt one divide it out so that it becomes a one. The formula that were going to be getting will assume this! Upon doing this the two equations that we want so solve for the unknown functions are
y1 + u2 y2 = g (t ) u1 Note that in this system we know the two solutions and so the only two unknowns here are u1
y1 + u2 y2 = 0 u1
(5) (6)
and u 2 . Solving this system is actually quite simple. First, solve (5) for u1 and plug this into (6) and do some simplification. =u1 u 2 y2 y1 (7)
2007 Paul Dawkins
157
u2 y2 y2 = g (t ) y1 + u2 y1 y y y2 - 2 1 = g (t ) u2 y1 y y - y2 y1 1 2 u2 = g (t ) y 1 y1 g ( t ) = u2 - y2 y1 y1 y2 y2 g ( t ) - y2 y1 y1 y2
(8)
So, we now have an expression for u 2 . Plugging this into (7) will give us an expression for u1 . =u1
Next, lets notice that
(9)
- y2 y1 0 W ( y1 , y2 ) = y1 y2
Recall that y1(t) and y2(t) are a fundamental set of solutions and so we know that the Wronskian wont be zero! Finally, all that we need to do is integrate (8) and (9) in order to determine what u1(t) and u2(t) are. Doing this gives,
y g (t ) u1 ( t ) = - 2 dt W ( y1 , y2 )
y g (t ) u2 ( t ) = 1 dt W ( y1 , y2 )
So, provided we can do these integrals, a particular solution to the differential equation is
YP ( t ) = y1u1 + y2u2
y g (t ) y g (t ) = - y1 2 dt + y2 1 dt W ( y1 , y2 ) W ( y1 , y2 )
So, lets summarize up what weve determined here. Variation of Parameters Consider the differential equation,
y + q ( t ) y + r ( t ) y = g ( t ) y + q ( t ) y + r ( t ) y = 0
Assume that y1(t) and y2(t) are a fundamental set of solutions for Then a particular solution to the nonhomogeneous differential equation is,
y g (t ) y g (t ) YP ( t ) = - y1 2 dt + y2 1 dt W ( y1 , y2 ) W ( y1 , y2 )
158
2007 Paul Dawkins
Depending on the person and the problem, some will find the formula easier to memorize and use, while others will find the process used to get the formula easier. The examples in this section will be done using the formula. Before proceeding with a couple of examples lets first address the issues involving the constants of integration that will arise out of the integrals. Putting in the constants of integration will give the following.
y2 g ( t ) y1 g ( t ) YP ( t ) = - y1 dt + c + y dt + k 2 W ( y , y ) W ( y , y ) 1 2 1 2 y g (t ) y g (t ) = - y1 2 dt + y2 1 dt + ( -cy1 + ky2 ) W ( y1 , y2 ) W ( y1 , y2 )
The final quantity in the parenthesis is nothing more than the complimentary solution with c1 = -c and c2 = k and we know that if we plug this into the differential equation it will simplify out to zero since it is the solution to the homogeneous differential equation. In other words, these terms add nothing to the particular solution and so we will go ahead and assume that c = 0 and k = 0 in all the examples. One final note before we proceed with examples. Do not worry about which of your two solutions in the complimentary solution is y1(t) and which one is y2(t). It doesnt matter. You will get the same answer no matter which one you choose to be y1(t) and which one you choose to be y2(t). Lets work a couple of examples now.
Example 1 Find a general solution to the following differential equation. 2 y + 18 y = 6 tan ( 3t )

Solution First, since the formula for variation of parameters requires a coefficient of a one in front of the second derivative lets take care of that before we forget. The differential equation that well actually be solving is
y + 9 y = 3tan ( 3t )
Well leave it to you to verify that the complimentary solution for this differential equation is
yc ( t ) = c1 cos ( 3t ) + c2 sin ( 3t )
So, we have
y1 ( t ) = cos ( 3t ) cos ( 3t ) sin ( 3t )
y2 ( t ) = sin ( 3t )
The Wronskian of these two functions is
W=
-3sin ( 3t ) 3cos ( 3t )
= 3cos 2 ( 3t ) + 3sin 2 ( 3t ) = 3
The particular solution is then,
2007 Paul Dawkins
159
3sin ( 3t ) tan ( 3t ) 3cos ( 3t ) tan ( 3t ) YP ( t ) = - cos ( 3t ) dt + sin ( 3t ) dt 3 3 sin 2 ( 3t ) dt + sin ( 3t ) sin ( 3t ) dt = - cos ( 3t ) cos ( 3t ) 1 - cos 2 ( 3t ) dt + sin ( 3t ) sin ( 3t ) dt = - cos ( 3t ) cos ( 3t ) = - cos ( 3t ) sec ( 3t ) - cos ( 3t ) dt + sin ( 3t ) sin ( 3t ) dt =cos ( 3t ) sin ( 3t ) ln sec ( 3t ) + tan ( 3t ) - sin ( 3t ) + ( - cos ( 3t ) ) 3 3 cos ( 3t ) =ln sec ( 3t ) + tan ( 3t ) 3
The general solution is,
y ( t ) = c1 cos ( 3t ) + c2 sin ( 3t ) -
cos ( 3t ) ln sec ( 3t ) + tan ( 3t ) 3
Example 2 Find a general solution to the following differential equation. et y - 2 y + y = 2 t +1

Solution We first need the complimentary solution for this differential equation. Well leave it to you to verify that the complimentary solution is,
yc ( t ) = c1et + c2tet
So, we have
y1 ( t ) = et
y 2 ( t ) = te t
The Wronskian of these two functions is
et W= t e
The particular solution is then,
te t = e t ( e t + te t ) - e t ( te t ) = e 2 t t t e + te
t
tet et et et t YP ( t ) = -e 2t 2 dt + te 2t 2 dt e ( t + 1) e ( t + 1) t 1 = -e t dt + tet dt 2 2 t +1 t +1 1 = - et ln (1 + t 2 ) + tet tan -1 ( t ) 2

1 y ( t ) = c1et + c2tet - et ln (1 + t 2 ) + tet tan -1 ( t ) 2

This method can also be used on non-constant coefficient differential equations, provided we know a fundamental set of solutions for the associated homogeneous differential equation.
Example 3 Find the general solution to ty - ( t + 1) y + y = t 2

given that
y1 ( t ) = et
y2 ( t ) = t + 1
form a fundamental set of solutions for the homogeneous differential equation. Solution As with the first example, we first need to divide out by a t.
1 1 y - 1 + y + y = t t t
The Wronskian for the fundamental set of solutions is
et W= t e
The particular solution is.
t +1 1
= et - et ( t + 1) = -tet
t ( t + 1) t + + 1 e ( t ) YP ( t ) = -et dt t ( ) t dt t -te - te
= et ( t + 1) e - t dt - ( t + 1) dt = et ( -e -t ( t + 2 ) ) - ( t + 1) t = -t 2 - 2t - 2
The general solution for this differential equation is.
y ( t ) = c1et + c2 ( t + 1) - t 2 - 2t - 2
We need to address one more topic about the solution to the previous example. The solution can be simplified down somewhat if we do the following.
y ( t ) = c1et + c2 ( t + 1) - t 2 - 2t - 2 = c1et + ( c2 - 2 )( t + 1) - t 2
= c1et + c2 ( t + 1) - t 2 - 2 ( t + 1)
Now, since c2 is an unknown constant subtracting 2 from it wont change that fact. So we can just write the c2 - 2 as c2 and be done with it. Here is a simplified version of the solution for this example.
y ( t ) = c1et + c2 ( t + 1) - t 2
This isnt always possible to do, but when it is you can simplify future work.
2007 Paul Dawkins
161
Mechanical Vibrations
Its now time to take a look at an application of second order differential equations. Were going to take a look at mechanical vibrations. In particular we are going to look at a mass that is hanging from a spring. Vibrations can occur in pretty much all branches of engineering and so what were going to be doing here can be easily adapted to other situations, usually with just a change in notation. Lets get the situation setup. We are going to start with a spring of length l, called the natural length, and were going to hook an object with mass m up to it. When the object is attached to the spring the spring will stretch a length of L. We will call the equilibrium position the position of the center of gravity for the object as it hangs on the spring with no movement. Below is sketch of the spring with and without the object attached to it.
As denoted in the sketch we are going to assume that all forces, velocities, and displacements in the downward direction will be positive. All forces, velocities, and displacements in the upward direction will be negative. Also, as shown in the sketch above, we will measure all displacement of the mass from its equilibrium position. Therefore, the u = 0 position will correspond to the center of gravity for the mass as it hangs on the spring and is at rest (i.e. no movement). Now, we need to develop a differential equation that will give the displacement of the object at any time t. First, recall Newtons Second Law of Motion.
ma = F
In this case we will use the second derivative of the displacement, u, for the acceleration and so Newtons Second Law becomes,
mu = F ( t , u , u )
162
2007 Paul Dawkins
We now need to determine all the forces that will act upon the object. There are four forces that we will assume act upon the object. Two that will always act on the object and two that may or may not act upon the object. Here is a list of the forces that will act upon the object. 1. Gravity, Fg The force due to gravity will always act upon the object of course. This force is
Fg = mg
2. Spring, Fs We are going to assume that Hookes Law will govern the force that the spring exerts on the object. This force will always be present as well and is
Fs = - k ( L + u )
Hookes Law tells us that the force exerted by a spring will be the spring constant, k > 0, times the displacement of the spring from its natural length. For our set up the displacement from the springs natural length is L + u and the minus sign is in there to make sure that the force always has the correct direction. Lets make sure that this force does what we expect it to. If the object is at rest in its equilibrium position the displacement is L and the force is simply Fs = kL which will act in the upward position as it should since the spring has been stretched from its natural length. If the spring has been stretched further down from the equilibrium position then L + u will be positive and Fs will be negative acting to pull the object back up as it should be. Next, if the object has been moved up past its equilibrium point, but not yet to its natural length then u will be negative, but still less than L and so L + u will be positive and once again Fs will be negative acting to pull the object up. Finally, if the object has been moved upwards so that the spring is now compressed, then u will be negative and greater than L. Therefore, L + u will be negative and now Fs will be positive acting to push the object down. So, it looks like this force will act as we expect that it should. 3. Damping, Fd The next force that we need to consider is damping. This force may or may not be present for any given problem Dampers work to counteract any movement. There are several ways to define a damping force. The one that well use is the following.
Fd = -g u
where, g > 0 is the damping coefficient. Lets think for a minute about how this force will act. If the object is moving downward, then the velocity ( u ) will be positive and so Fd will be negative and acting to pull the object back up. Likewise, if the object is
moving upward, the velocity ( u ) will be negative and so Fd will be positive and acting to push the object back down. In other words, the damping force as weve defined it will always act to counter the current motion of the object and so will act to damp out any motion in the object. 4. External Forces, F(t) This is the catch all force. If there are any other forces that we decide we want to act on our object we lump them in here and call it good. We typically call F(t) the forcing function. Putting all of these together gives us the following for Newtons Second Law.
mu = mg - k ( L + u ) - g u + F ( t ) mu + g u + ku = mg - kL + F ( t )
Or, upon rewriting, we get,
Now, when the object is at rest in its equilibrium position there are exactly two forces acting on the object, the force due to gravity and the force due to the spring. Also, since the object is at rest (i.e. not moving) these two forces must be canceling each other out. This means that we must have,
mg = kL
(1)
Using this in Newtons Second Law gives us the final version of the differential equation that well work with.
mu + g u + ku = F ( t )
(2)
Along with this differential equation we will have the following initial conditions.
u ( 0 ) = u0 u ( 0 ) = u0
Initial displacement from the equilibrium position. Initial velocity.
(3)
Note that well also be using (1) to determine the spring constant, k. Okay. Lets start looking at some specific cases. Free, Undamped Vibrations This is the simplest case that we can consider. Free or unforced vibrations means that F(t) = 0 and undamped vibrations means that g = 0. In this case the differential equation becomes,
mu + ku = 0 k m
This is easy enough to solve in general. The characteristic equation has the roots,
r= i
This is usually reduced to,
r = w0 i
where,
w0 =
k m (4)
and w0 is called the natural frequency. Recall as well that m > 0 and k > 0 and so we can guarantee that this quantity will be complex. The solution in this case is then
u ( t ) = c1 cos (w0 t ) + c2 sin (w0 t ) u ( t ) = R cos (w0t - d )
We can write (4) in the following form,
(5)
where R is the amplitude of the displacement and d is the phase shift or phase angle of the displacement. When the displacement is in the form of (5) it is usually easier to work with. However, its easier to find the constants in (4) from the initial conditions than it is to find the amplitude and phase shift in (5) from the initial conditions. So, in order to get the equation into the form in (5) we will first put the equation in the form in (4), find the constants, c1 and c2 and then convert this into the form in (5). So, assuming that we have c1 and c2 how do we determine R and d ? Lets start with (5) and use a trig identity to write it as
u ( t ) = R cos (d ) cos (w0 t ) + R sin (d ) sin (w0t ) c1 = R cos d c2 = R sin d
(6)
Now, R and d are constants and so if we compare (6) to (4) we can see that
We can find R in the following way.

2 c12 + c2 = R 2 cos 2 d + R 2 sin 2 d = R 2
Taking the square root of both sides and assuming that R is positive will give
2 R = c12 + c2
(7)
Finding d is just as easy. Well start with
c2 R sin d = = tan d c1 R cos d c d = tan -1 2 c1
Taking the inverse tangent of both sides gives,
(8)
Before we work any examples lets talk a little bit about units of mass and the British vs. metric system differences. Recall that the weight of the object is given by
W = mg
where m is the mass of the object and g is the gravitational acceleration. For the examples in this problem well be using the following values for g.
British : g = 32 ft / s 2 Metric : g = 9.8 m / s 2

This is not the standard 32.2 ft/s2 or 9.81 m/s2, but using these will make some of the numbers come out a little nicer. In the metric system the mass of objects is given in kilograms (kg) and there is nothing for us to do. However, in the British system we tend to be given the weight of an object in pounds (yes, pounds are the units of weight not mass) and so well need to compute the mass for these problems. At this point we should probably work an example of all this to see how this stuff works.
Example 1 A 16 lb object stretches a spring
ft by itself. There is no damping and no external forces acting on the system. The spring is initially displaced 6 inches upwards from its equilibrium position and given an initial velocity of 1 ft/sec downward. Find the displacement at any time t, u(t).
8 9
Solution We first need to set up the IVP for the problem. This requires us to get our hands on m and k. This is the British system so well need to compute the mass.
m=
W 16 1 = = g 32 2
Now, lets get k. We can use the fact that mg = kL to find k. Dont forget that well need all of our length units the same. Well use feet for the unit of measurement for this problem.
k=
mg 16 = = 18 8 L 9 u (0) = 1 2 u ( 0 ) = 1
We can now set up the IVP.
1 u + 18u = 0 2
For the initial conditions recall that upward displacement/motion is negative while downward displacement/motion is positive. Also, since we decided to do everything in feet we had to convert the initial displacement to feet. Now, to solve this we can either go through the characteristic equation or we can just jump straight to the formula that we derived above. Well do it that way. First, we need the natural frequency,
2007 Paul Dawkins
166
w0 =
18 = 36 = 6 1 2
The general solution, along with its derivative, is then,
u ( t ) = c1 cos ( 6t ) + c2 sin ( 6t ) u ( t ) = -6c1 sin ( 6t ) + 6c2 cos ( 6t )
Applying the initial conditions gives
1 = u ( 0 ) = c1 2 1 = u ( 0 ) = 6c2 cos ( 6t )
c1 = c2 = 1 6
1 2
The displacement at any time t is then
1 1 u ( t ) = - cos ( 6t ) + sin ( 6t ) 2 6
Now, lets convert this to a single cosine. First lets get the amplitude, R.
10 1 1 R = - + = = 0.52705 6 2 6
You can use either the exact value here or a decimal approximation. Often the decimal approximation will be easier. Now lets get the phase shift.
1 d = tan 6 = -0.32175 -1 2
-1
We need to be careful with this part. The phase angle found above is in Quadrant IV, but there is also an angle in Quadrant II that would work as well. We get this second angle by adding p onto the first angle. So, we actually have two angles. They are
d1 = -0.32175 d 2 = d1 + p = 2.81984
We need to decide which of these phase shifts is correct, because only one will be correct. To do this recall that
c1 = R cos d c2 = R sin d
Now, since we are assuming that R is positive this means that the sign of cosd will be the same as the sign of c1 and the sign of sind will be the same as the sign of c2. So, for this particular case
we must have cosd < 0 and sind > 0. This means that the phase shift must be in Quadrant II and so the second angle is the one that we need. So, after all of this the displacement at any time t is.
u ( t ) = 0.52705cos ( 6t - 2.81984 )
Here is a sketch of the displacement for the first 5 seconds.
Now, lets take a look at a slightly more realistic situation. No vibration will go on forever. So lets add in a damper and see what happens now. Free, Damped Vibrations We are still going to assume that there will be no external forces acting on the system, with the exception of damping of course. In this case the differential equation will be.
mu + g u + ku = 0
where m, d, and k are all positive constants. Upon solving for the roots of the characteristic equation we get the following.
r1,2 =
We will have three cases here.
-g g 2 - 4mk 2m
1. g 2 - 4mk = 0 In this case we will get a double root out of the characteristic equation and the displacement at any time t will be.
u ( t ) = c1e
gt 2m
+ c1te
gt 2m
Notice that as t the displacement will approach zero and so the damping in this case will do what its supposed to do. This case is called critical damping and will happen when the damping coefficient is,
2007 Paul Dawkins
168
g 2 - 4mk = 0 g 2 = 4mk g = 2 mk = g CR
The value of the damping coefficient that gives critical damping is called the critical damping coefficient and denoted by gCR. 2. g 2 - 4mk > 0 In this case lets rewrite the roots a little.
-g g 2 - 4mk r1,2 = 2m 4mk -g g 1 - 2 g = 2m g 4mk =1 1- 2 g 2m

Also notice that from our initial assumption that we have,
g 2 > 4mk 4mk 1> 2 g

Using this we can see that the fraction under the square root above is less than one. Then if the quantity under the square root is less than one, this means that the square root of this quantity is also going to be less than one. In other words,
1-
4mk <1 g2
Why is this important? Well, the quantity in the parenthesis is now one plus/minus a number that is less than one. This means that the quantity in the parenthesis is guaranteed to be positive and so the two roots in this case are guaranteed to be negative. Therefore the displacement at any time t is,
u ( t ) = c1er1t + c2 er2t
and will approach zero as t . So, once again the damper does what it is supposed to do. This case will occur when
g 2 > 4mk g > 2 mk g > g CR

and is called over damping.
2007 Paul Dawkins
169
3. g 2 - 4mk < 0 In this case we will get complex roots out of the characteristic equation.
r1,2 =
g 2 - 4mk -g =lmi 2m 2m
where the real part is guaranteed to be negative and so the displacement is
u ( t ) = c1elt cos ( m t ) + c2elt sin ( mt ) = elt ( c1 cos ( mt ) + c2 sin ( mt ) ) = Relt cos ( m t - d )
Notice that we reduced the sine and cosine down to a single cosine in this case as we did in the undamped case. Also, since l < 0 the displacement will approach zero as t and the damper will also work as its supposed to in this case. We will get this case will occur when
g 2 < 4mk g < 2 mk g < g CR

and is called under damping. Lets take a look at a couple of examples here with damping.
Example 2 Take the spring and mass system from the first example and attach a damper to it
that will exert a force of 12 lbs when the velocity is 2 ft/s. Find the displacement at any time t, u(t). Solution The mass and spring constant were already found in the first example so we wont do the work here. We do need to find the damping coefficient however. To do this we will use the formula for the damping force given above with one modification. The original damping force formula is,
Fd = -g u
However, remember that the force and the velocity are always acting in opposite directions. So, if the velocity is upward (i.e. negative) the force will be downward (i.e. positive) and so the minus in the formula will cancel against the minus in the velocity. Likewise, if the velocity is downward (i.e. positive) the force will be upwards (i.e. negative) and in this case the minus sign in the formula will cancel against the minus in the force. In other words, we can drop the minus sign in the formula and use
Fd = g u
and then just ignore any signs for the force and velocity. Doing this gives us the following for the damping coefficient
12 = g ( 2 )
g =6
The IVP for this example is then,

1 u + 6u + 18u = 0 2
u (0) = -
1 2
u ( 0 ) = 1
Before solving lets check to see what kind of damping weve got. To do this all we need is the critical damping coefficient.
1 g CR = 2 km = 2 (18 ) = 2 9 = 6 2
So, it looks like weve got critical damping. Note that this means that when we go to solve the differential equation we should get a double root. Speaking of solving, lets do that. Ill leave the details to you to check that the displacement at any time t is.
1 u ( t ) = - e-6t - 2te -6 t 2
Here is a sketch of the displacement during the first 3 seconds.
Notice that the vibration in the system is not really a true vibration as we tend to think of them. In the critical damping case there isnt going to be a real oscillation about the equilibrium point that we tend to associate with vibrations. The damping in this system is strong enough to force the vibration to die out before it ever really gets a chance to do much in the way of oscillation.
Example 3 Take the spring and mass system from the first example and this time lets attach a
damper to it that will exert a force of 17 lbs when the velocity is 2 ft/s. Find the displacement at any time t, u(t). Solution So, the only difference between this example and the previous example is damping force. So lets find the damping coefficient
17 = g ( 2 )
g =
17 = 8.5 > g CR 2
So it looks like weve got over damping this time around so we should expect to get two real
distinct roots from the characteristic equation and they should both be negative. The IVP for this example is,
1 17 u + u + 18u = 0 2 2
u (0) = -
1 2
u ( 0 ) = 1
This ones a little messier than the previous example so well do a couple of the steps, leaving it to you to fill in the blanks. The roots of the characteristic equation are
r1,2 =
-17 145 = -2.4792, - 14, 4208 2
In this case it will be easier to just convert to decimals and go that route. Note that, as predicted we got two real, distinct and negative roots. The general and actual solution for this example are then,
u ( t ) = c1e-2.4792 t + c2 e -14.5208 t
u ( t ) = -0.5198e -2.4792 t + 0.0199e -14.5208 t

Heres a sketch of the displacement for this example.
Notice an interesting thing here about the displacement here. Even though we are over damped in this case, it actually takes longer for the vibration to die out than in the critical damping case. Sometimes this happens, although it will not always be the case that over damping will allow the vibration to continue longer than the critical damping case. Also notice that, as with the critical damping case, we dont get a vibration in the sense that we usually think of them. Again, the damping is strong enough to force the vibration do die out quick enough so that we dont see much, if any, of the oscillation that we typically associate with vibrations. Lets take a look at one more example before moving on the next type of vibrations.
2007 Paul Dawkins
172
Example 4 Take the spring and mass system from the first example and for this example lets
attach a damper to it that will exert a force of 5 lbs when the velocity is 2 ft/s. Find the displacement at any time t, u(t). Solution So, lets get the damping coefficient.
5 = g ( 2)
g =
5 = 2.5 < g CR 2
So its under damping this time. That shouldnt be too surprising given the first two examples. The IVP for this example is,
1 5 u + u + 18u = 0 2 2
u (0) = -
1 2
u ( 0 ) = 1
In this case the roots of the characteristic equation are
r1,2 =
-5 119 i 2
They are complex as we expected to get since we are in the under damped case. The general solution and actual solution are
119 119 c1 cos t c sin + 2 2 2 t 5t - 119 119 u ( t ) = e 2 -0.5 cos t 0.04583sin 2 2 t u (t ) = e

5t 2
Lets convert this to a single cosine as we did in the undamped case.
R=
( -0.5 ) + ( -0.04583)
2
= 0.502096 OR
-0.04583 d1 = tan -1 = 0.09051 -0.5
d 2 = d1 + p = 3.2321
As with the undamped case we can use the coefficients of the cosine and the sine to determine which phase shift that we should use. The coefficient of the cosine (c1) is negative and so cosd must also be negative. Likewise, the coefficient of the sine (c2) is also negative and so sind must also be negative. This means that d must be in the Quadrant III and so the second angle is the one that we want. The displacement is then
u ( t ) = 0.502096e
Here is a sketch of this displacement.
5t 2
119 cos t 3.2321 2
2007 Paul Dawkins
173
In this case we finally got want we usually consider to be a true vibration. In fact that is the point of critical damping. As we increase the damping coefficient, the critical damping coefficient will be the first one in which a true oscillation in the displacement will not occur. For all values of the damping coefficient larger than this (i.e. over damping) we will also not see a true oscillation in the displacement. From a physical standpoint critical (and over) damping is usually preferred to under damping. Think of the shock absorbers in your car. When you hit a bump you dont want to spend the next few minutes bouncing up and down while the vibration set up by the bump die out. You would like there to be as little movement as possible. In other words, you will want to set up the shock absorbers in your car so get at the least critical damping so that you can avoid the oscillations that will arise from an under damped case. Its now time to look at systems in which we allow other external forces to act on the object in the system. Undamped, Forced Vibrations We will first take a look at the undamped case. The differential equation in this case is
mu + ku = F ( t )
This is just a nonhomogeneous differential equation and we know how to solve these. The general solution will be
u ( t ) = uc ( t ) + U P ( t )
where the complimentary solution is the solution to the free, undamped vibration case. To get the particular solution we can use either undetermined coefficients or variation of parameters depending on which we find easier for a given forcing function. There is a particular type of forcing function that we should take a look at since it leads to some interesting results. Lets suppose that the forcing function is a simple periodic function of the form
F ( t ) = F0 cos (w t )
OR
F ( t ) = F0 sin (w t )
For the purposes of this discussion well use the first one. Using this, the IVP becomes,
mu + ku = F0 cos (w t )
174
2007 Paul Dawkins
The complimentary solution, as pointed out above, is just where w0 is the natural frequency.
uc ( t ) = c1 cos (w0t ) + c2 sin (w0 t )
We will need to be careful in finding a particular solution. The reason for this will be clear if we use undetermined coefficients. With undetermined coefficients our guess for the form of the particular solution would be,
U P ( t ) = A cos (w t ) + B sin (w t )
Now, this guess will be problems if w0 = w . If this were to happen the guess for the particular solution is exactly the complimentary solution and so wed need to add in a t. Of course if we dont have w0 = w then there will be nothing wrong with the guess. So, we will need to look at this in two cases. 1. w0 w In this case our initial guess is okay since it wont be the complimentary solution. Upon differentiating the guess and plugging it into the differential equation and simplifying we get,
( -mw
A + kA) cos ( wt ) + ( - mw 2 B + kB ) sin ( wt ) = F0 cos ( wt )
Setting coefficients equal gives us,
cos (w t ) : sin (w t ) :
( -mw ( -mw
+ k ) A = F0 + k)B = 0
A=
F0 k - mw 2
B=0
The particular solution is then
U P (t ) =
F0 cos (w t ) k - mw 2 F0 = cos (w t ) k 2 m -w m F0 = cos (wt ) 2 m (w0 -w2 )
Note that we rearranged things a little. Depending on the form that youd like the displacement to be in we can have either of the following.
2007 Paul Dawkins
175
u ( t ) = c1 cos ( w0 t ) + c2 sin (w0t ) + u ( t ) = R cos ( w0 t - d ) + F0
2 m (w0 -w2 )
F0
cos ( wt )
2 m (w0 -w2 )
cos ( wt )
If we used the sine form of the forcing function we could get a similar formula. 2. w0 = w In this case we will need to add in a t to the guess for the particular solution.
U P ( t ) = At cos (w0t ) + Bt sin (w0t )
Note that we went ahead and acknowledge that w0 = w in our guess. Acknowledging this will help with some simplification that well need to do later on. Differentiating our guess, plugging it into the differential equation and simplifying gives us the following.
( -mw
2 0
+ k ) At cos ( wt ) + ( - mw02 + k ) Bt sin ( wt ) +
2mw0 B cos (wt ) - 2mw0 A sin ( wt ) = F0 cos ( wt )

Before setting coefficients equal, lets remember the definition of the natural frequency and note that
k k - mw + k = - m +k = 0 m + k = -m m
2 0
So, the first two terms actually drop out (which is a very good thing) and this gives us,
2mw0 B cos (wt ) - 2mw0 A sin ( wt ) = F0 cos ( wt )
Now lets set coefficient equal.
cos (wt ) : sin (wt ) :
2mw0 B = F0 2mw0 A = 0
B=
F0 2mw0
A=0
In this case the particular will be,
U P (t ) =
F0 t sin (w0t ) 2mw0 F0 t sin (w0t ) 2mw0
The displacement for this case is then
u ( t ) = c1 cos (w0t ) + c2 sin ( w0 t ) + u ( t ) = R cos (w0t - d ) +
F0 t sin ( w0 t ) 2mw0
depending on the form that you prefer for the displacement.

So, what was the point of the two cases here? Well in the first case, w0 w our displacement function consists of two cosines and is nice and well behaved for all time. In contrast, the second case, w0 = w will have some serious issues at t increases. The addition of the t in the particular solution will mean that we are going to see an oscillation that grows in amplitude as t increases. This case is called resonance and we would generally like to avoid this at all costs. In this case resonance arose by assuming that the forcing function was,
F ( t ) = F0 cos (w t ) F ( t ) = F0 sin (w t )
We would also have the possibility of resonance if we assumed a forcing function of the form.
We should also take care to not assume that a forcing function will be in one of these two forms. Forcing functions can come in a wide variety of forms. If we do run into a forcing function different from the one that used here you will have to go through undetermined coefficients or variation of parameters to determine the particular solution.
Example 5 A 3 kg object is attached to spring and will stretch the spring 392 mm by itself.
There is no damping in the system and a forcing function of the form
F ( t ) = 10cos (w t )
is attached to the object and the system will experience resonance. If the object is initially displaced 20 cm downward from its equilibrium position and given a velocity of 10 cm/sec upward find the displacement at any time t. Solution Since we are in the metric system we wont need to find mass as its been given to us. Also, for all calculations well be converting all lengths over to meters. The first thing we need to do is find k.
k=
mg ( 3)( 9.8) = = 75 L 0.392
Now, we are told that the system experiences resonance so lets go ahead and get the natural frequency so we can completely set up the IVP.
w0 =
The IVP for this is then
75 k = =5 m 3 u ( 0 ) = 0.2 u ( 0 ) = -0.1
3u + 75u = 10 cos ( 5t )
Solution wise there isnt a whole lot to do here. The complimentary solution is the free undamped solution which is easy to get and for the particular solution we can just use the formula that we derived above.
2007 Paul Dawkins
177
The general solution is then,
u ( t ) = c1 cos ( 5t ) + c2 sin ( 5t ) +
10 t sin ( 5t ) 2 ( 3)( 5)
1 u ( t ) = c1 cos ( 5t ) + c2 sin ( 5t ) + t sin ( 5t ) 3

Applying the initial conditions gives the displacement at any time t. Well leave the details to you to check.
1 1 1 u ( t ) = cos ( 5t ) - sin ( 5t ) + t sin ( 5t ) 5 50 3

The last thing that well do is combine the first two terms into a single cosine.
2 2
1 1 R = + - = 0.200998 5 50 -1 50 = -0.099669 d1 = tan -1 1 5
d 2 = d1 + p = 3.041924
In this case the coefficient of the cosine is positive and the coefficient of the sine is negative. This forces cosd to be positive and sind to be negative. This means that the phase shift needs to be in Quadrant IV and so the first one is the correct phase shift this time. The displacement then becomes,
1 u ( t ) = 0.200998cos ( 5t + 0.099669 ) + t sin ( 5t ) 3

Here is a sketch of the displacement for this example.
Its now time to look at the final vibration case. Forced, Damped Vibrations This is the full blown case where we consider every last possible force that can act upon the system. The differential equation for this case is,
mu + g u + ku = F ( t )
The displacement function this time will be,
u ( t ) = uc ( t ) + U P ( t )
where the complimentary solution will be the solution to the free, damped case and the particular solution will be found using undetermined coefficients or variation of parameter, whichever is most convenient to use. There are a couple of things to note here about this case. First, from our work back in the free, damped case we know that the complimentary solution will approach zero as t increases. Because of this the complimentary solution is often called the transient solution in this case. Also, because of this behavior the displacement will start too look more and more like the particular solution as t increases and so the particular solution is often called the steady state solution or forced response. Lets work one final example before leaving this section. As with the previous examples, were going to leave most of the details out for you to check.
Example 6 Take the system from the last example and add in a damper that will exert a force of
45 Newtons when then velocity is 50 cm/sec. Solution So, all we need to do is compute the damping coefficient for this problem then pull everything else down from the previous problem. The damping coefficient is
Fd = g u
45 = g ( 0.5 )
g = 90
The IVP for this problem is.
3u + 90u + 75u = 10cos ( 5t )
u ( 0 ) = 0.2
u ( 0 ) = -0.1
The complimentary solution for this example is
( -15+10 2 )t ( -15-10 2 )t uc ( t ) = c1e + c2e

uc ( t ) = c1e-0.8579t + c2e -29.1421t
For the particular solution we the form will be,
U P ( t ) = A cos ( 5t ) + B sin ( 5t )
Plugging this into the differential equation and simplifying gives us,
450 B cos ( 5t ) - 450 A sin ( 5t ) = 10cos ( 5t ) 1 sin ( 5t ) 45
Setting coefficient equal gives,
U P (t ) =
The general solution is then
u ( t ) = c1e-0.8579t + c2e -29.1421t +

1 sin ( 5t ) 45 1 sin ( 5t ) 45
u ( t ) = 0.1986e -0.8579t + 0.001398e -29.1421t +

Here is a sketch of the displacement for this example.
2007 Paul Dawkins
180
Laplace Transforms
Introduction
In this chapter we will be looking at how to use Laplace transforms to solve differential equations. There are many kinds of transforms out there in the world. Laplace transforms and Fourier transforms are probably the main two kinds of transforms that are used. As we will see in later sections we can use Laplace transforms to reduce a differential equation to an algebra problem. The algebra can be messy on occasion, but it will be simpler than actually solving the differential equation directly in many cases. Laplace transforms can also be used to solve IVPs that we cant use any previous method on. For simple differential equations such as those in the first few sections of the last chapter Laplace transforms will be more complicated than we need. In fact, for most homogeneous differential equations such as those in the last chapter Laplace transforms is significantly longer and not so useful. Also, many of the simple nonhomogeneous differential equations that we saw in the Undetermined Coefficients and Variation of Parameters are still simpler (or at the least no more difficult than Laplace transforms) to do as we did them there. However, at this point, the amount of work required for Laplace transforms is starting to equal the amount of work we did in those sections. Laplace transforms comes into its own when the forcing function in the differential equation starts getting more complicated. In the previous chapter we looked only at nonhomogeneous differential equations in which g(t) was a fairly simple continuous function. In this chapter we will start looking at g(t)s that are not continuous. It is these problems were the reasons for using Laplace transforms starts to become clear. We will also see that, for some of the more complicated nonhomogeneous differential equations from the last chapter, Laplace transforms are actually easier on those problems as well. Here is a brief rundown of the sections in this chapter. The Definition The definition of the Laplace transform. We will also compute a couple Laplace transforms using the definition. Laplace Transforms As the previous section will demonstrate, computing Laplace transforms directly from the definition can be a fairly painful process. In this section we introduce the way we usually compute Laplace transforms. Inverse Laplace Transforms In this section we ask the opposite question. Heres a Laplace transform, what function did we originally have? Step Functions This is one of the more important functions in the use of Laplace transforms. With the introduction of this function the reason for doing Laplace transforms starts to become apparent.
Solving IVPs with Laplace Transforms Heres how we used Laplace transforms to solve IVPs. Nonconstant Coefficient IVPs We will see how Laplace transforms can be used to solve some nonconstant coefficient IVPs IVPs with Step Functions Solving IVPs that contain step functions. This is the section where the reason for using Laplace transforms really becomes apparent. Dirac Delta Function One last function that often shows up in Laplace transform problems. Convolution Integral A brief introduction to the convolution integral and an application for Laplace transforms. Table of Laplace Transforms This is a small table of Laplace Transforms that well be using here.
2007 Paul Dawkins
182
The Definition
You know, its always a little scary when we devote a whole section just to the definition of something. Laplace transforms (or just transforms) can seem scary when we first start looking at them. However, as we will see, they arent as bad as they may appear at first. Before we start with the definition of the Laplace transform we need to get another definition out of the way. A function is called piecewise continuous on an interval if the interval can be broken into a finite number of subintervals on which the function is continuous on each open subinterval (i.e. the subinterval without its endpoints) and has a finite limit at the endpoints of each subinterval. Below is a sketch of a piecewise continuous function.
In other words, a piecewise continuous function is a function that has a finite number of breaks in it and doesnt blow up to infinity anywhere. Now, lets take a look at the definition of the Laplace transform. Definition Suppose that f(t) is a piecewise continuous function. The Laplace transform of f(t) is denoted L f ( t ) and defined as
L { f ( t )} = e - st f ( t ) dt
0
(1)
There is an alternate notation for Laplace transforms. For the sake of convenience we will often denote Laplace transforms as,
L { f ( t )} = F ( s )
With this alternate notation, note that the transform is really a function of a new variable, s, and that all the ts will drop out in the integration process. Now, the integral in the definition of the transform is called an improper integral and it would probably be best to recall how these kinds of integrals work before we actually jump into computing some transforms.
Example 1 If c 0 , evaluate the following integral.

0 0
ec t dt
Solution Remember that you need to convert improper integrals to limits as follows,
e ct dt = lim e ct dt
n 0 n
Now, do the integral, then evaluate the limit.
e ct dt = lim ect dt
n 0
1 = lim ect n c 0
1 1 = lim ec n - n c c
Now, at this point, weve got to be careful. The value of c will affect our answer. Weve already assumed that c was non-zero, now we need to worry about the sign of c. If c is positive the exponential will go to infinity. On the other hand, if c is negative the exponential will go to zero. So, the integral is only convergent (i.e. the limit exists and is finite) provided c<0. In this case we get,
e ct dt = -
1 c
provided c < 0
(2)
Now that we remember how to do these, lets compute some Laplace transforms. Well start off with probably the simplest Laplace transform to compute.
Example 2 Compute L{1}.

Solution Theres not really a whole lot do here other than plug the function f(t) = 1 into (1)
L {1} = e - st dt
0
Now, at this point notice that this is nothing more than the integral in the previous example with c = - s . Therefore, all we need to do is reuse (2) with the appropriate substitution. Doing this gives,
L {1} = e- st dt = 0
1 -s
provided - s < 0
Or, with some simplification we have,
L {1} =
1 s
184
provided s > 0
2007 Paul Dawkins
Notice that we had to put a restriction on s in order to actually compute the transform. All Laplace transforms will have restrictions on s. At this stage of the game, this restriction is something that we tend to ignore, but we really shouldnt ever forget that its there. Lets do another example.
Example 3 Compute L {ea t }

Solution Plug the function into the definition of the transform and do a little simplification.
L {e at } = e - st ea t dt = e(
0 0
a - s )t
dt
Once again, notice that we can use (2) provided c = a - s . So lets do this.
L {ea t } = e(
0
a - s )t
dt provided a - s < 0 provided s > a
=-
1 a-s 1 = s-a
Lets do one more example that doesnt come down to an application of (2).
Example 4 Compute L{sin(at)}.

Solution Note that were going to leave it to you to check most of the integration here. Plug the function into the definition. This time lets also use the alternate notation.
L {sin ( at )} = F ( s )
0
= e - st sin ( at ) dt = lim e - st sin ( at ) dt

n 0 n
Now, if we integrate by parts we will arrive at,

n 1 - st s n F ( s ) = lim - e cos ( at ) - e - st cos ( at ) dt n 0 a 0 a
Now, evaluate the first term to simplify it a little and integrate by parts again on the integral. Doing this arrives at,
n 1 s 1 - st s n - sn F ( s ) = lim (1 - e cos ( an ) ) - e sin ( at ) + e - st sin ( at ) dt n a a 0 a 0 a
Now, evaluate the second term, take the limit and simplify.
s1 s n 1 F ( s ) = lim (1 - e- sn cos ( an ) ) - e - sn sin ( an ) + e - st sin ( at ) dt 0 n a aa a 1 ss = - e - st sin ( at ) dt a aa 0 = 1 s2 a a2
e - st sin ( at ) dt
Now, notice that in the limits we had to assume that s>0 in order to do the following two limits.
lim e - sn cos ( an ) = 0
n n
lim e - sn sin ( an ) = 0
Without this assumption, we get a divergent integral again. Also, note that when we got back to the integral we just converted the upper limit back to infinity. The reason for this is that, if you think about it, this integral is nothing more than the integral that we started with. Therefore, we now get,
F (s) =
Now, simply solve for F(s) to get,
1 s2 - F ( s) a a2
L {sin ( at )} = F ( s ) =
a s + a2
2
provided s > 0
As this example shows, computing Laplace transforms is often messy. Before moving on to the next section, we need to do a little side note. On occasion you will see the following as the definition of the Laplace transform.
L { f ( t )} = e - st f ( t ) dt
-
Note the change in the lower limit from zero to negative infinity. In these cases there is almost always the assumption that the function f(t) is in fact defined as follows,
if t < 0 0 f (t ) = f ( t ) if t 0
In other words, it is assumed that the function is zero if t<0. In this case the first half of the integral will drop out since the function is zero and we will get back to the definition given in (1). A Heaviside function is usually used to make the function zero for t<0. We will be looking at these in a later section.
2007 Paul Dawkins
186
Laplace Transforms
As we saw in the last section computing Laplace transforms directly can be fairly complicated. Usually we just use a table of transforms when actually computing Laplace transforms. The table that is provided here is not an inclusive table, but does include most of the commonly used Laplace transforms and most of the commonly needed formulas pertaining to Laplace transforms. Before doing a couple of examples to illustrate the use of the table lets get a quick fact out of the way. Fact Given f(t) and g(t) then, for any constants a and b. In other words, we dont worry about constants and we dont worry about sums or differences of functions in taking Laplace transforms. All that we need to do is take the transform of the individual functions, then put any constants back in and add or subtract the results back up. So, lets do a couple of quick examples.
L {af ( t ) + bg ( t )} = a F ( s ) + bG ( s )
Example 1 Find the Laplace transforms of the given functions. (a) f ( t ) = 6e -5t + e3t + 5t 3 - 9 [Solution]
(b) g ( t ) = 4cos ( 4t ) - 9sin ( 4t ) + 2cos (10t ) [Solution] (c) h ( t ) = 3sinh ( 2t ) + 3sin ( 2t ) [Solution] (d) g ( t ) = e3t + cos ( 6t ) - e3t cos ( 6t ) [Solution] Solution Okay, theres not really a whole lot to do here other than go to the table, transform the individual functions up, put any constants back in and then add or subtract the results. Well do these examples in a little more detail than is typically used since this is the first time were using the tables. (a) f ( t ) = 6e -5t + e3t + 5t 3 - 9
F (s) = 6 =
1 1 3! 1 + + 5 3+1 - 9 s - ( -5 ) s - 3 s s
6 1 30 9 + + 4 s+5 s-3 s s
2007 Paul Dawkins
187
(b) g ( t ) = 4cos ( 4t ) - 9sin ( 4t ) + 2cos (10t )
G (s) = 4 =
s s + ( 4)
2 2 2
-9
4 s + ( 4)
2 2
+2
s s + (10 )
2 2
4s 36 2s - 2 + 2 s + 16 s + 16 s + 100
(c) h ( t ) = 3sinh ( 2t ) + 3sin ( 2t )
H (s) = 3 =
2 s - ( 2)
2 2 2
+3
2 s + ( 2)
2 2
6 6 + 2 s -4 s +4
(d) g ( t ) = e3t + cos ( 6t ) - e3t cos ( 6t )
G ( s) = =
1 s s-3 + 2 2 s - 3 s + ( 6 ) ( s - 3) 2 + ( 6 ) 2 1 s s-3 + 2 s - 3 s + 36 ( s - 3)2 + 36

Make sure that you pay attention to the difference between a normal trig function and hyperbolic functions. The only difference between them is the + a2 for the normal trig functions becomes a - a2 in the hyperbolic function! Its very easy to get in a hurry and not pay attention and grab the wrong formula. If you dont recall the definition of the hyperbolic functions see the notes for the table. Lets do one final set of examples.
Example 2 Find the transform of each of the following functions. (a) f ( t ) = t cosh ( 3t ) [Solution]
(b) h ( t ) = t 2 sin ( 2t ) [Solution] (c) g ( t ) = t 2
3
[Solution]
3
(d) f ( t ) = (10t ) 2 Solution (a) f ( t ) = t cosh ( 3t )
[Solution]
(e) f ( t ) = tg ( t ) [Solution]
This function is not in the table of Laplace transforms. However we can use #30 in the table to compute its transform. This will correspond to #30 if we take n=1.
F ( s ) = L {tg ( t )} = -G ( s ) ,
So, we then have,
where g ( t ) = cosh ( 3t ) s2 + 9
2
G (s) =
Using #30 we then have,
s 2 s -9 F (s) =
G ( s ) = -
(s
- 9)
(s
s2 + 9
2
- 9)
(b) h ( t ) = t sin ( 2t )
2
This part will also use #30 in the table. In fact we could use #30 in one of two ways. We could use it with n = 1 .
H ( s ) = L {tf ( t )} = - F ( s ) ,
where f ( t ) = t sin ( 2t ) where f ( t ) = sin ( 2t )
Or we could use it with n = 2 .
H ( s ) = L {t 2 f ( t )} = F ( s ) ,
Since its less work to do one derivative, lets do it the first way. So using #9 we have,
F (s) =
The transform is then,
(s
4s
2
+ 4)
F(s) = -
12 s 2 - 16
(s
+ 4)
H (s) =
(c) g ( t ) = t
3 2
12 s 2 - 16
(s
+ 4)
This part can be done using either #6 (with n = 2 ) or #32 (along with #5). We will use #32 so we can see an example of this. In order to use #32 well need to notice that
Now, using #5,
t 0
2 3 v dv = t 2 3 f (t ) = t
t2 = F (s) = 1 3 p 5 = s 4s 2
3 t v dv 2 0
p
2s 2
3
we get the following.
3 p G ( s) = 3 2 2s 2
This is what we would have gotten had we used #6.

(d) f ( t ) = (10t ) 2
3
For this part we will use #24 along with the answer from the previous part. To see this note that if
g (t ) = t 2
3
then
f ( t ) = g (10t ) 1 s G 10 10
Therefore, the transform is.
F (s) =
1 3 p = 5 10 s 2 4 10 = 10 2
3
3 p 4s 2
5
(e) f ( t ) = tg ( t ) This final part will again use #30 from the table as well as #35.
L {tg ( t )} = -
d L { g } ds d = - {sG ( s ) - g ( 0 )} ds = - ( G ( s ) + sG ( s ) - 0 ) = -G ( s ) - sG ( s )
Remember that g(0) is just a constant so when we differentiate it we will get zero!
As this set of examples has shown us we cant forget to use some of the general formulas in the table to derive new Laplace transforms for functions that arent explicitly listed in the table!
2007 Paul Dawkins
190
Inverse Laplace Transforms

Finding the Laplace transform of a function is not terribly difficult if weve got a table of transforms in front of us to use as we saw in the last section. What we would like to do now is go the other way. We are going to be given a transform, F(s), and ask what function (or functions) did we have originally. As you will see this can be a more complicated and lengthy process than taking transforms. In these cases we say that we are finding the Inverse Laplace Transform of F(s) and use the following notation.
f ( t ) = L-1 {F ( s )}
As with Laplace transforms, weve got the following fact to help us take the inverse transform. Fact Given the two Laplace transforms F(s) and G(s) then for any constants a and b. So, we take the inverse transform of the individual transforms, put any constants back in and then add or subtract the results back up. Lets take a look at a couple of fairly simple inverse transforms.
L-1 {aF ( s ) + bG ( s )} = aL-1 {F ( s )} + bL-1 {G ( s )}
Example 1 Find the inverse transform of each of the following. 6 1 4 (a) F ( s ) = + [Solution] s s -8 s -3 19 1 7 (b) H ( s ) = + 5 [Solution] s + 2 3s - 5 s 6s 3 (c) F ( s ) = 2 + 2 [Solution] s + 25 s + 25 8 3 (d) G ( s ) = 2 + 2 [Solution] 3s + 12 s - 49
Solution Ive always felt that the key to doing inverse transforms is to look at the denominator and try to identify what youve got based on that. If there is only one entry in the table that has that particular denominator, the next step is to make sure the numerator is correctly set up for the inverse transform process. If it isnt, correct it (this is always easy to do) and then take the inverse transform. If there is more than one entry in the table has a particular denominator, then the numerators of each will be different, so go up to the numerator and see which one youve got. If you need to correct the numerator to get it into the correct form and then take the inverse transform. So, with this advice in mind lets see if we can take some inverse transforms.
2007 Paul Dawkins
191
(a) F ( s ) =
6 1 4 + s s -8 s -3
From the denominator of the first term it looks like the first term is just a constant. The correct numerator for this term is a 1 so well just factor the 6 out before taking the inverse transform. The second term appears to be an exponential with a = 8 and the numerator is exactly what it needs to be. The third term also appears to be an exponential, only this time a = 3 and well need to factor the 4 out before taking the inverse transforms. So, with a little more detail than well usually put into these,
F (s) = 6
1 1 1 +4 s s -8 s-3 8t f ( t ) = 6 (1) - e + 4 ( e3t ) = 6 - e8t + 4e3t

(b) H ( s ) =
19 1 7 + 5 s + 2 3s - 5 s
The first term in this case looks like an exponential with a = -2 and well need to factor out the 19. Be careful with negative signs in these problems, its very easy to lose track of them. The second term almost looks like an exponential, except that its got a 3s instead of just an s in the denominator. It is an exponential, but in this case well need to factor a 3 out of the denominator before taking the inverse transform. The denominator of the third term appears to be #3 in the table with n = 4 . The numerator however, is not correct for this. There is currently a 7 in the numerator and we need a 4! = 24 in the numerator. This is very easy to fix. Whenever a numerator is off by a multiplicative constant, as in this case, all we need to do is put the constant that we need in the numerator. We will just need to remember to take it back out by dividing by the same constant. So, lets first rewrite the transform.
H (s) =
7 4! 19 1 4! + 4 +1 5 s - ( -2 ) 3 ( s - 3 ) s 1 1 1 7 4! + 5 s - ( -2 ) 3 s - 3 4! s 4+1
= 19
So, what did we do here? We factored the 19 out of the first term. We factored the 3 out of the denominator of the second term since it cant be there for the inverse transform and in the third term we factored everything out of the numerator except the 4! since that is the portion that we need in the numerator for the inverse transform process. Lets now take the inverse transform.
1 5t 7 h ( t ) = 19e -2t - e 3 + t 4 3 24
(c) F ( s ) =
6s 3 + 2 s + 25 s + 25
2
In this part weve got the same denominator in both terms and our table tells us that weve either got #7 or #8. The numerators will tell us which weve actually got. The first one has an s in the numerator and so this means that the first term must be #8 and well need to factor the 6 out of the numerator in this case. The second term has only a constant in the numerator and so this term must be #7, however, in order for this to be exactly #8 well need to multiply/divide a 5 in the numerator to get it correct for the table. The transform becomes,
F (s) = 6 =6
Taking the inverse transform gives,
s s2 + (5) s s2 + (5)
2 2
+ +
s 2 + ( 5)
35 5
3 5 5 s 2 + ( 5 )2
3 f ( t ) = 6 cos ( 5t ) + sin ( 5t ) 5
(d) G ( s ) =
8 3 + 2 3s + 12 s - 49
2
In this case the first term will be a sine once we factor a 3 out of the denominator, while the second term appears to be a hyperbolic sine (#17). Again, be careful with the difference between these two. Both of the terms will also need to have their numerators fixed up. Here is the transform once were done rewriting it.
G (s) =
1 8 3 + 2 2 3 s + 4 s - 49 37 1 ( 4 )( 2 ) 7 = 2 + 3 s + ( 2 )2 s 2 - ( 7 ) 2
Notice that in the first term we took advantage of the fact that we could get the 2 in the numerator that we needed by factoring the 8. The inverse transform is then,
4 3 g ( t ) = sin ( 2t ) + sinh ( 7t ) 3 7
So, probably the best way to identify the transform is by looking at the denominator. If there is more than one possibility use the numerator to identify the correct one. Fix up the numerator if needed to get it into the form needed for the inverse transform process. Finally, take the inverse transform.
2007 Paul Dawkins
193
Lets do some slightly harder problems. These are a little more involved than the first set.
Example 2 Find the inverse transform of each of the following. 6s - 5 [Solution] (a) F ( s ) = 2 s +7 1 - 3s (b) F ( s ) = 2 [Solution] s + 8s + 21 3s - 2 (c) G ( s ) = 2 [Solution] 2s - 6 s - 2 s+7 (d) H ( s ) = 2 [Solution] s - 3s - 10
Solution (a) F ( s ) =
6s - 5 s2 + 7
From the denominator of this one it appears that it is either a sine or a cosine. However, the numerator doesnt match up to either of these in the table. A cosine wants just an s in the numerator with at most a multiplicative constant, while a sine wants only a constant and no s in the numerator. Weve got both in the numerator. This is easy to fix however. We will just split up the transform into two terms and then do inverse transforms.
5 7 6s F (s) = 2 - 2 7 s +7 s +7 5 f ( t ) = 6 cos 7t sin 7
( )
( 7t )
Do not get too used to always getting the perfect squares in sines and cosines that we saw in the first set of examples. More often than not (at least in my class) they wont be perfect squares!
(b) F ( s ) =
1 - 3s s + 8s + 21
2
In this case there are no denominators in our table that look like this. We can however make the denominator look like one of the denominators in the table by completing the square on the denominator. So, lets do that first.
s 2 + 8s + 21 = s 2 + 8s + 16 - 16 + 21 = s 2 + 8s + 16 + 5 = ( s + 4) + 5
2
Recall that in completing the square you take half the coefficient of the s, square this, and then add and subtract the result to the polynomial. After doing this the first three terms should factor as a perfect square.
So, the transform can be written as the following.
F (s) =
1 - 3s
2
( s + 4)
+5
Okay, with this rewrite it looks like weve got #19 and/or #20s from our table of transforms. However, note that in order for it to be a #19 we want just a constant in the numerator and in order to be a #20 we need an s a in the numerator. Weve got neither of these so well have to correct the numerator to get it into proper form. In correcting the numerator always get the s a first. This is the important part. We will also need to be careful of the 3 that sits in front of the s. One way to take care of this is to break the term into two pieces, factor the 3 out of the second and then fix up the numerator of this term. This will work, however it will put three terms into our answer and there are really only two terms. So, we will leave the transform as a single term and correct it as follows,
F (s) =
1 - 3 ( s + 4 - 4)
2
( s + 4) + 5 1 - 3 ( s + 4 ) + 12 = 2 ( s + 4) + 5 -3 ( s + 4 ) + 13 = 2 ( s + 4) + 5
We needed an s + 4 in the numerator, so we put that in. We just needed to make sure and take the 4 back out by subtracting it back out. Also, because of the 3 multiplying the s we needed to do all this inside a set of parenthesis. Then we partially multiplied the 3 through the second term and combined the constants. With the transform in this form, we can break it up into two transforms each of which are in the tables and so we can do inverse transforms on them,
F ( s ) = -3
s+4
( s + 4)
+5
13
( s + 4)
5 5 2
+5 sin
f ( t ) = -3e -4t cos
e ( 5t ) + 13 5
-4 t
( 5t )
(c) G ( s ) =
3s - 2 2s - 6 s - 2
2
This one is similar to the last one. We just need to be careful with the completing the square however. The first thing that we should do is factor a 2 out of the denominator, then complete the square. Remember that when completing the square a coefficient of 1 on the s2 term is needed! So, heres the work for this transform.
2007 Paul Dawkins
195
G ( s) = = =
3s - 2 2 ( s 2 - 3s - 1) 1 3s - 2 2 9 2 s - 3s + 9 4 - 4 -1 1 3s - 2 2 13 2 (s - 3 2) - 4
So, it looks like weve got #21 and #22 with a corrected numerator. Heres the work for that and the inverse transform.
G (s) = =
3 1 3( s - 3 2 + 2)-2 2 13 2 (s - 3 2) - 4 5 1 3(s - 3 2)+ 2 2 13 2 (s - 3 2) - 4
5 13 1 3( s - 3 2 13 2) = + 3 2 13 3 2 13 2 s s ( ) ( ) 2 4 2 4
13 13 1 3t 5 32t g ( t ) = 3e 2 cosh t + e sinh 2 2 t 2 13

In correcting the numerator of the second term, notice that I only put in the square root since we already had the over 2 part of the fraction that we needed in the numerator.
(d) H ( s ) =
s+7 s - 3s - 10
2
This one appears to be similar to the previous two, but it actually isnt. The denominators in the previous two couldnt be easily factored. In this case the denominator does factor and so we need to deal with it differently. Here is the transform with the factored denominator.
H (s) =
s+7 ( s + 2 )( s - 5 )
The denominator of this transform seems to suggest that weve got a couple of exponentials, however in order to be exponentials there can only be a single term in the denominator and no ss in the numerator. To fix this we will need to do partial fractions on this transform. In this case the partial fraction decomposition will be
H (s) =
A B + s+2 s-5
Dont remember how to do partial fractions? In this example well show you one way of getting
the values of the constants and after this example well review how to get the correct form of the partial fraction decomposition. Okay, so lets get the constants. There is a method for finding the constants that will always work, however it can lead to more work than is sometimes required. Eventually, we will need that method, however in this case there is an easier way to find the constants. Regardless of the method used, the first step is to actually add the two terms back up. This gives the following.
A ( s - 5) + B ( s + 2 ) s+7 = ( s + 2 )( s - 5) ( s + 2 )( s - 5 )
Now, this needs to be true for any s that we should choose to put in. So, since the denominators are the same we just need to get the numerators equal. Therefore, set the numerators equal.
s + 7 = A ( s - 5) + B ( s + 2)
Again, this must be true for ANY value of s that we want to put in. So, lets take advantage of that. If it must be true for any value of s then it must be true for s = -2 , to pick a value at random. In this case we get,
5 = A ( -7 ) + B ( 0 )
A=-
5 7 12 7
We found A by appropriately picking s. We can B in the same way if we chose s = 5 .
12 = A ( 0 ) + B ( 7 )
B=
This will not always work, but when it does it will usually simplify the work considerably. So, with these constants the transform becomes,
12 -5 7 + 7 s+2 s-5
H (s) =
We can now easily do the inverse transform to get,
5 12 h ( t ) = - e -2 t + e 5 t 7 7
The last part of this example needed partial fractions to get the inverse transform. When we finally get back to differential equations and we start using Laplace transforms to solve them, you will quickly come to understand that partial fractions are a fact of life in these problems. Almost every problem will require partial fractions to one degree or another. Note that we could have done the last part of this example as we had done the previous two parts. If we had we would have gotten hyperbolic functions. However, recalling the definition of the hyperbolic functions we could have written the result in the form we got from the way we worked our problem. However, most students have a better feel for exponentials than they do for hyperbolic functions and so its usually best to just use partial fractions and get the answer in
terms of exponentials. It may be a little more work, but it will give a nicer (and easier to work with) form of the answer. Be warned that in my class Ive got a rule that if the denominator can be factored with integer coefficients then it must be. So, lets remind you how to get the correct partial fraction decomposition. The first step is to factor the denominator as much as possible. Then for each term in the denominator we will use the following table to get a term or terms for our partial fraction decomposition. Factor in denominator Term in partial fraction decomposition
ax + b
A ax + b Ak A1 A2 + +L + 2 k ax + b ( ax + b ) ( ax + b ) Ax + B ax 2 + bx + c Ak x + Bk A1 x + B1 A2 x + B2 + +L+ k 2 2 ax + bx + c ( ax 2 + bx + c ) ( ax2 + bx + c )
( ax + b )k
ax 2 + bx + c
( ax
+ bx + c )
Notice that the first and third cases are really special cases of the second and fourth cases respectively. So, lets do a couple more examples to remind you how to do partial fractions.
Example 3 Find the inverse transform of each of the following. 86s - 78 (a) G ( s ) = [Solution] ( s + 3)( s - 4 )( 5s - 1)
(b) F ( s ) = (c) G ( s ) = Solution (a) G ( s ) =
2 - 5s [Solution] ( s - 6 ) ( s 2 + 11) 25 [Solution] s ( s + 4s + 5)

3 2
86s - 78 ( s + 3)( s - 4 )( 5s - 1) A B C + + s + 3 s - 4 5s - 1
Heres the partial fraction decomposition for this part.
G ( s) =
Now, this time we wont go into quite the detail as we did in the last example. We are after the
numerator of the partial fraction decomposition and this is usually easy enough to do in our heads. Therefore, we will go straight to setting numerators equal.
86 s - 78 = A ( s - 4 )( 5s - 1) + B ( s + 3)( 5s - 1) + C ( s + 3)( s - 4 ) s = -3 s= 1 5 - 336 = A ( -7 )( -16 ) A = -3 C=5 B=2
As with the last example, we can easily get the constants by correctly picking values of s.
304 16 19 = C - 5 5 5 266 = B ( 7 )(19 )
s=4
So, the partial fraction decomposition for this transform is,
G (s) = -
3 2 5 + + s + 3 s - 4 5s - 1
Now, in order to actually take the inverse transform we will need to factor a 5 out of the denominator of the last term. The corrected transform as well as its inverse transform is.
G (s) = -
3 2 1 + + s+3 s-4 s- 1 5
t
g ( t ) = -3e -3t + 2e 4t + e 5
(b) F ( s ) =
2 - 5s ( s - 6 ) ( s 2 + 11)
So, for the first time weve got a quadratic in the denominator. Heres the decomposition for this part.
F (s) =
Setting numerators equal gives,
A Bs + C + 2 s - 6 s + 11
2 - 5s = A ( s 2 + 11) + ( Bs + C )( s - 6 )
Okay, in this case we could use s = 6 to quickly find A, but thats all it would give. In this case we will need to go the long way around to getting the constants. Note that this way will always work, but is sometimes more work than is required. The long way is to completely multiply out the right side and collect like terms.
2 - 5s = A ( s 2 + 11) + ( Bs + C )( s - 6 )
= As 2 + 11A + Bs 2 - 6 B + Cs - 6C = ( A + B ) s 2 + ( -6 B + C ) s + 11A - 6C
In order for these two to be equal the coefficients of the s2, s and the constants must all be equal.
So, setting coefficients equal gives the following system of equations that can be solved.
A+ B = 0 s : - 6 B + C = -5 s 0 : 11A - 6C = 2 s2 :
1
A=-
28 28 67 , B= , C=47 47 47
Notice that I used s0 to denote the constants. This is habit on my part and isnt really required, its just what Im used to doing. Also, the coefficients are fairly messy fractions in this case. Get used to that. They will often be like this when we get back into solving differential equations. There is a way to make our life a little easier as well with this. Since all of the fractions have a denominator of 47 well factor that out as we plug them back into the decomposition. This will make dealing with them much easier. The partial fraction decomposition is then,
F (s) =
1 28 28s - 67 + 47 s - 6 s 2 + 11
67 11 1 28 28s = + 2 - 2 11 47 s - 6 s + 11 s + 11
The inverse transform is then.
f (t ) =
1 -28e 6t + 28cos 47
11t -
67 sin 11
11t
(c) G ( s ) =
25 s ( s + 4s + 5)
3 2
With this last part do not get excited about the s3. We can think of this term as
s3 = ( s - 0) G (s) =
and it becomes a linear term to a power. So, the partial fraction decomposition is
A B C Ds + E + 2+ 3+ 2 s s s s + 4s + 5
Setting numerators equal and multiplying out gives.
25 = As 2 ( s 2 + 4 s + 5 ) + Bs ( s 2 + 4s + 5) + C ( s 2 + 4 s + 5 ) + ( Ds + E ) s 3 = ( A + D ) s 4 + ( 4 A + B + E ) s 3 + ( 5 A + 4 B + C ) s 2 + ( 5 B + 4C ) s + 5C
Setting coefficients equal gives the following system.
2007 Paul Dawkins
200
s4 : A+ D = 0 3 s : 4A+ B + E = 0 s 2 : 5 A + 4B + C = 0 s1 : 5 B + 4C = 0 s0 : 5C = 25
A=
11 11 24 , B = -4, C = 5, D = - , E = 5 5 5
This system looks messy, but its easier to solve than it might look. First we get C for free from the last equation. We can then use the fourth equation to find B. The third equation will then give A, etc. When plugging into the decomposition well get everything with a denominator of 5, then factor that out as we did in the previous part in order to make things easier to deal with.
1 11 20 25 11s + 24 G (s) = - 2 + 3 - 2 5 s s s s + 4s + 5
Note that we also factored a minus sign out of the last two terms. To complete this part well need to complete the square on the later term and fix up a couple of numerators. Heres that work.
1 11 20 25 11s + 24 G ( s) = - 2 + 3 - 2 5 s s s s + 4s + 5 1 11 20 25 11( s + 2 - 2 ) + 24 = - 2 + 3 2 5 s s s s + 2 + 1 ( ) 1 11 20 25 2! 11( s + 2 ) 2 = - 2 + 32! 2 2 5 s s s + 2 + 1 + 2 + 1 s s ( ) ( )

The inverse transform is then.
1 25 g ( t ) = 11 - 20t + t 2 - 11e -2t cos ( t ) - 2e -2t sin ( t ) 5 2

So, one final time. Partial fractions are a fact of life when using Laplace transforms to solve differential equations. Make sure that you can deal with them.
2007 Paul Dawkins
201
Step Functions
Before proceeding into solving differential equations we should take a look at one more function. Without Laplace transforms it would be much more difficult to solve differential equations that involve this function in g(t). The function is the Heaviside function and is defined as,
0 if t < c uc ( t ) = 1 if t c
Here is a graph of the Heaviside function.
Heaviside functions are often called step functions. Here is some alternate notation for Heaviside functions.
uc ( t ) = u ( t - c ) = H ( t - c )
We can think of the Heaviside function as a switch that is off until t = c at which point it turns on and takes a value of 1. So what if we want a switch that will turn on and takes some other value, say 4, or -7? Heaviside functions can only take values of 0 or 1, but we can use them to get other kinds of switches. For instance 4uc(t) is a switch that is off until t = c and then turns on and takes a value of 4. Likewise, -7uc(t) will be a switch that will take a value of -7 when it turns on. Now, suppose that we want a switch that is on (with a value of 1) and then turns off at t = c. We can use Heaviside functions to represent this as well. The following function will exhibit this kind of behavior.
1 - 0 = 1 if t < c 1 - uc ( t ) = 1 - 1 = 0 if t c
Prior to t = c the Heaviside is off and so has a value of zero. The function as whole then for t < c has a value of 1. When we hit t = c the Heaviside function will turn on and the function will now take a value of 0. We can also modify this so that it has values other than 1 when it is one. For instance,
3 - 3uc ( t )
will be a switch that has a value of 3 until it turns off at t = c. We can also use Heaviside functions to represent much more complicated switches.
Example 1 Write the following function (or switch) in terms of Heaviside functions. if t < 6 -4 25 if 6 t < 8 f (t ) = if 8 t < 30 16 if t 30 10
Solution There are three sudden shifts in this function and so (hopefully) its clear that were going to need three Heaviside functions here, one for each shift in the function. Heres the function in terms of Heaviside functions.
f ( t ) = -4 + 29u6 ( t ) - 9u8 ( t ) - 6u30 ( t )
Its fairly easy to verify this. In the first interval, t < 6 all three Heaviside functions are off and the function has the value
f ( t ) = -4
Notice that when we know that Heaviside functions are on or off we tend to not write them at all as we did in this case. In the next interval, 6 t < 8 the first Heaviside function is now on while the remaining two are still off. So, in this case the function has the value.
f ( t ) = -4 + 29 = 25
In the third interval, 8 t < 30 the first two Heaviside functions are one while the last remains off. Here the function has the value.
f ( t ) = -4 + 29 - 9 = 16
In the last interval, t 30 all three Heaviside function are one and the function has the value.
f ( t ) = -4 + 29 - 9 - 6 = 10
So, the function has the correct value in all the intervals. All of this is fine, but if we continue the idea of using Heaviside function to represent switches, we really need to acknowledge that most switches will not turn on and take constant values. Most switches will turn on and vary continually with the value of t. So, lets consider the following function.
2007 Paul Dawkins
203
We would like a switch that is off until t = c and then turns on and takes the values above. By this we mean that when t = c we want the switch to turn on and take the value of f(0) and when t = c + 4 we want the switch to turn on and take the value of f(4), etc. In other words, we want the switch to look like the following,
Notice that in order to take the values that we want the switch to take it needs to turn on and take the values of f ( t - c ) ! We can use Heaviside functions to help us represent this switch as well. Using Heaviside functions this switch can be wrote as
g ( t ) = uc ( t ) f ( t - c )
(1)
Okay, weve talked a lot about Heaviside functions to this point, but we havent even touched on Laplace transforms yet. So, lets start thinking about that. Lets determine the Laplace transform of (1). This is actually easy enough to derive so lets do that. Plugging (1) into the definition of the Laplace transform gives,
L {uc ( t ) f ( t - c )} = e - st uc ( t ) f ( t - c ) dt
0
= e - st f ( t - c ) dt
c
2007 Paul Dawkins
204
Notice that we took advantage of the fact that the Heaviside function will be zero if t < c and 1 otherwise. This means that we can drop the Heaviside function and start the integral at c instead of 0. Now use the substitution u = t c and the integral becomes,
- s u +c L {uc ( t ) f ( t - c )} = e ( ) f ( u ) du 0
= e - su e - c s f ( u ) du
0
The second exponential has no us in it and so it can be factored out of the integral. Note as well that in the substitution process the lower limit of integration went back to 0.
L {uc ( t ) f ( t - c )} = e - c s e - su f ( u ) du
0
Now, the integral left is nothing more than the integral that we would need to compute if we were going to find the Laplace transform of f(t). Therefore, we get the following formula
L {uc ( t ) f ( t - c )} = e- c s F ( s )
(2)
In order to use (2) the function f(t) must be shifted by c, the same value that is used in the Heaviside function. Also note that we only take the transform of f(t) and not f(t-c)! We can also turn this around to get a useful formula for inverse Laplace transforms.
L -1 {e - c s F ( s )} = uc ( t ) f ( t - c )
(3)
We can use (2) to get the Laplace transform of a Heaviside function by itself. To do this we will consider the function in (2) to by f(t) = 1. Doing this gives us
1 e-c s L {uc ( t )} = L {uc ( t ) g 1} = e - c s L {1} = e - c s = s s

Putting all of this together leads to the following two formulas.
L {uc ( t )} =
Let do some examples.
e-c s s
e- c s L -1 = uc ( t ) s
(4)
Example 2 Find the Laplace transform of each of the following. 3 (a) g ( t ) = 10u12 ( t ) + 2 ( t - 6 ) u6 ( t ) - ( 7 - e12 -3t ) u4 ( t ) [Solution]
(b) f ( t ) = -t 2 u3 ( t ) + cos ( t ) u5 ( t ) [Solution]
t 4 if t < 5 (c) h ( t ) = 4 [Solution] t 1 t + 3sin 10 - 2 if t 5 if t < 6 t (d) f ( t ) = [Solution] 2 8 + t 6 if t 6 ( )

Solution In all of these problems remember that the function MUST be in the form
uc ( t ) f ( t - c )
before we start taking transforms. If it isnt in that form we will have to put it into that form!
12 3t (a) g ( t ) = 10u12 ( t ) + 2 ( t - 6 ) u6 ( t ) - 7 - e - u4 ( t ) 3
So there are three terms in this function. The first is simply a Heaviside function and so we can use (4) on this term. The second and third terms however have functions with them and we need to identify the functions that are shifted for each of these. In the second term it appears that we are using the following function,
f ( t ) = 2t 3
The third term uses the following function,
f (t - 6) = 2 ( t - 6)
and this has been shifted by the correct amount.
f ( t ) = 7 - e -3 t
f ( t - 4 ) = 7 - e -3( t - 4) = 7 - e12-3t
which has also been shifted by the correct amount. With these functions identified we can now take the transform of the function.
G (s) = =
2 ( 3!) 7 10e -12 s 1 -4 s + e -6 s 3+1 - e s s s s+3 10e -12 s 12e -6 s 7 1 -4 s + 3+1 - e s s s s +3
(b) f ( t ) = -t 2 u3 ( t ) + cos ( t ) u5 ( t ) This part is going to cause some problems. There are two terms and neither has been shifted by the proper amount. The first term needs to be shifted by 3 and the second needs to be shifted by 5. So, since they havent been shifted, we will need to force the issue. We will need to add in the shifts, and then take them back out of course. Here they are.
f ( t ) = - ( t - 3 + 3) u3 ( t ) + cos ( t - 5 + 5 ) u5 ( t )
2
Now we still have some potential problems here. The first function is still not really shifted correctly, so well need to use
( a + b )2 = a 2 + 2ab + b 2
to get this shifted correctly. The second term can be dealt with in one of two ways. The first would be to use the formula
cos ( a + b ) = cos ( a ) cos ( b ) - sin ( a ) sin ( b )
to break it up into cosines and sines with arguments of t-5 which will be shifted as we expect. There is an easier way to do this one however. From out table of Laplace transforms we have #16 and using that we can see that if
g ( t ) = cos ( t + 5 )
g ( t - 5) = cos ( t - 5 + 5 )
This will make our life a little easier so well do it this way. Now, breaking up the first term and leaving the second term along gives us,
f ( t ) = - ( t - 3) + 6 ( t - 3) + 9 u3 ( t ) + cos ( t - 5 + 5 ) u5 ( t )
2
Okay, so it looks like the two functions that have been shifted here are
g ( t ) = t 2 + 6t + 9
g ( t ) = cos ( t + 5 )
Taking the transform then gives,
s cos ( 5 ) - sin ( 5) -5 s 2 6 9 F ( s ) = - 3 + 2 + e -3 s + e s s2 + 1 s s
Its messy, especially the second term, but there it is. Also, do not get excited about the cos ( 5 ) and sin ( 5 ) . They are just numbers.
t 4 if t < 5 (c) h ( t ) = 4 t 1 t + 3sin 10 - 2 if t 5

This one isnt as bad as it might look on the surface. The first thing that we need to do is write it in terms of Heaviside functions.
t 1 h ( t ) = t 4 + 3u5 ( t ) sin - 10 2 1 = t 4 + 3u5 ( t ) sin ( t - 5 ) 10

Since the t4 is in both terms there isnt anything to do when we add in the Heaviside function. The only thing that gets added in is the sine term. Notice as well that the sine has been shifted by the proper amount. All we need to do now is to take the transform.
1 ) e -5 s 4! 3 ( 10 + 1 2 s 5 s 2 + ( 10 )
H (s) = =
3 -5 s e 24 10 + 5 2 1 s s + 100
2007 Paul Dawkins
207
(d) f ( t ) =
t 2 -8 + ( t - 6 )
if t < 6 if t 6
Again, the first thing that we need to do is write the function in terms of Heaviside functions.
f ( t ) = t + -8 - t + ( t - 6 ) u6 ( t )
2
We had to add in a -8 in the second term since that appears in the second part and we also had to subtract a t in the second term since the t in the first portion is no longer there. This subtraction of the t adds a problem because the second function is no longer correctly shifted. This is easier to fix than the previous example however. Here is the corrected function.
f ( t ) = t + - 8 - ( t - 6 + 6 ) + ( t - 6 ) u6 ( t )
2 2 6
( ) = t + ( -8 - ( t - 6 ) - 6 + ( t - 6 ) ) u ( t ) = t + ( -14 - ( t - 6 ) + ( t - 6 ) ) u ( t )
2 6
So, in the second term it looks like we are shifting
g ( t ) = t 2 - t - 14 1 2 1 14 -6 s + - - e s 2 s3 s 2 s
The transform is then,
F (s) =
Without the Heaviside function taking Laplace transforms is not a terribly difficult process provided we have our trusty table of transforms. However, with the advent of Heaviside functions, taking transforms can become a fairly messy process on occasion. So, lets do some inverse Laplace transforms to see how they are done.
Example 3 Find the inverse Laplace transform of each of the following. s e -4 s (a) H ( s ) = [Solution] ( 3s + 2 )( s - 2 )
(b) G ( s ) = (c) F ( s ) = (d) G ( s ) =
5e -6 s - 3e -11s [Solution] ( s + 2) ( s2 + 9) 4s + e - s [Solution] ( s - 1)( s + 2 ) 3s + 8e -20 s - 2 se -3s + 6e -7 s s 2 ( s + 3)

208 [Solution]
2007 Paul Dawkins
Solution All of these will use (3) somewhere in the process. Notice that in order to use this formula the exponential doesnt really enter into the mix until the very end. The vast majority of the process is finding the inverse transform of the stuff without the exponential. In these problems we are not going to go into detail on many of the inverse transforms. If you need a refresher on some of the basics of inverse transforms go back and take a look at the previous section. (a) H ( s ) =
s e -4 s ( 3s + 2 )( s - 2 ) s
In light of the comments above lets first rewrite the transform in the following way.
H ( s ) = e -4 s
( 3s + 2 )( s - 2 )
= e -4 s F ( s )
Now, this problem really comes down to needing f(t). So, lets do that. Well need to partial fraction F(s) up. Heres the partial fraction decomposition.
F (s) =
A B + 3s + 2 s - 2
s = A ( s - 2 ) + B ( 3s + 2 ) 1 4 1 A= 4
Well find the constants here by selecting values of s. Doing this gives,
s=2 s=2 3 -
2 = 8B 2 8 =- A 3 3
1 4
B=
So, the partial fraction decomposition becomes,
F (s) =
3( s + 2 3)
1 4
s-2
Notice that we factored a 3 out of the denominator in order to actually do the inverse transform. The inverse transform of this is then,
f (t ) =
1 - 23t 1 2 t e + e 12 4
Now, lets go back and do the actual problem. The original transform was,
H ( s ) = e -4 s F ( s )
Note that we didnt bother to plug in F(s). There really isnt a reason to plug it back in. Lets just use (3) to write down the inverse transform in terms of symbols. The inverse transform is,
2007 Paul Dawkins
209
h ( t ) = u4 ( t ) f ( t - 4 )
where, f(t) is,
f (t ) =
1 - 23t 1 2 t e + e 12 4
This is all the farther that well go with the answer. There really isnt any reason to plug in f(t) at this point. It would make the function longer and definitely messier. We will give almost all of our answers to these types of inverse transforms in this form.
(b) G ( s ) =
5e -6 s - 3e -11s ( s + 2) ( s2 + 9)
This problem is not as difficult as it might at first appear to be. Because there are two exponentials we will need to deal with them separately eventually. Now, this might lead us to conclude that the best way to deal with this function is to split it up as follows,
G ( s ) = e -6 s
( s + 2 ) ( s2 + 9)
- e -11s
( s + 2 ) ( s2 + 9)
Notice that we factored out the exponential, as we did in the last example, since we would need to do that eventually anyway. This is where a fairly common complication arises. Many people will call the first function F(s) and the second function H(s) and the partial fraction both of them. However, if instead of just factoring out the exponential we would also factor out the coefficient we would get,
G ( s ) = 5e -6 s
( s + 2) ( s
+ 9)
- 3e -11s
( s + 2) ( s2 + 9 )
Upon doing this we can see that the two functions are in fact the same function. The only difference is the constant that was in the numerator. So, the way that well do these problems is to first notice that both of the exponentials have only constants as coefficients. Instead of breaking things up then, we will simply factor out the whole numerator and get,
G ( s ) = ( 5e-6 s - 3e -11s )
( s + 2 ) ( s 2 + 9)
= ( 5e -6 s - 3e-11s ) F ( s )
and now we will just partial fraction F(s). Here is the partial fraction decomposition.
F (s) =
A Bs + C + 2 s+2 s +9
Setting numerators equal and combining gives us,
1 = A ( s 2 + 9 ) + ( s + 2 )( Bs + C ) = ( A + B ) s 2 + ( 2 B + C ) s + 9 A + 2C
2007 Paul Dawkins
210
Setting coefficient equal and solving gives,
A+ B = 0 s : 2 B + C = 0 s 0 : 9 A + 2C = 1 s2 :
1
A=
1 1 2 , B=- , C= 13 13 13
Substituting back into the transform gives and fixing up the numerators as needed gives,
F (s) = =
1 1 -s + 2 + 2 13 s + 2 s + 9 23 1 1 s - 2 + 2 3 13 s + 2 s + 9 s + 9
As we did in the previous section we factored out the common denominator to make our work a little simpler. Taking the inverse transform then gives,
f (t ) =
1 -2t 2 e - cos ( 3t ) + sin ( 3t ) 13 3
At this point we can go back and start thinking about the original problem.
G ( s ) = ( 5e -6 s - 3e -11s ) F ( s )
= 5e -6 s F ( s ) - 3e -11s F ( s )
Well also need to distribute the F(s) through as well in order to get the correct inverse transform. Recall that in order to use (3) to take the inverse transform you must have a single exponential times a single transform. This means that we must multiply the F(s) through the parenthesis. We can now take the inverse transform,
g ( t ) = 5u6 ( t ) f ( t - 6 ) - 3u11 ( t ) f ( t - 11) f (t ) = 1 -2t 2 e - cos ( 3t ) + sin ( 3t ) 13 3
where,
(c) F ( s ) =
4s + e - s ( s - 1)( s + 2 )
In this case, unlike the previous part, we will need to break up the transform since one term has a constant in it and the other has an s. Note as well that we dont consider the exponential in this, only its coefficient. Breaking up the transform gives,
F (s) =
4s 1 + e-s = G ( s ) + e- s H ( s ) s 1 s + 2 s 1 s + 2 ( )( ) ( )( ) A B + s -1 s + 2
We will need to partial fraction both of these terms up. Well start with G(s).
G (s) =
4s = A ( s + 2 ) + B ( s - 1) 8 3 4 A= 3
Now, pick values of s to find the constants.
s = -2 s =1
So G(s) and its inverse transform is,
-8 = -3B 4 = 3A
B=
G (s) =
4 3
s -1 s + 2 4 8 g ( t ) = et + e -2 t 3 3 A B + s -1 s + 2
8 3
Now, repeat the process for H(s).
H (s) =
1 = A ( s + 2 ) + B ( s - 1) 1 3
s = -2 s =1
So H(s) and its inverse transform is,
1 = -3B 1 = 3A
B=A= 1 3
s -1 s + 2 1 1 h ( t ) = e t - e -2 t 3 3
Putting all of this together gives the following,
H (s) =
1 3
1 3
F ( s ) = G ( s ) + e-s H ( s ) f ( t ) = g ( t ) + u1 ( t ) h ( t - 1)
where,
4 8 g ( t ) = et + e -2 t 3 3
and
1 1 h ( t ) = e t - e -2 t 3 3
2007 Paul Dawkins
212
(d) G ( s ) =
3s + 8e -20 s - 2 se -3s + 6e -7 s s 2 ( s + 3) s ( 3 - 2e -3 s ) + (8e -20 s + 6e -7 s ) s 2 ( s + 3)
This one looks messier than it actually is. Lets first rearrange the numerator a little.
G ( s) =
In this form it looks like we can break this up into two pieces that will require partial fractions. When we break these up we should always try and break things up into as few pieces as possible for the partial fractioning. Doing this can save you a great deal of unnecessary work. Breaking up the transform as suggested above gives,
G ( s ) = ( 3 - 2e -3 s )
= ( 3 - 2e -3 s ) F ( s ) + ( 8e -20 s + 6e -7 s ) H ( s )
1 1 + (8e -20 s + 6e-7 s ) 2 s ( s + 3) s ( s + 3)
Note that we canceled an s in F(s). You should always simplify as much a possible before doing the partial fractions. Lets partial fraction up F(s) first.
F (s) =
A B + s s+3
1 = A ( s + 3) + Bs 1 3
s = -3 s=0
So F(s) and its inverse transform is,
1 = -3B 1 = 3A
B=A= 1 3
F (s) = 3 - 3 s s+3 1 1 f ( t ) = - e -3t 3 3

Now partial fraction H(s).
H (s) =
A B C + 2+ s s s+3
1 = As ( s + 3) + B ( s + 3) + Cs 2
2007 Paul Dawkins
Pick values of s to find the constants.
s = -3 s=0 s =1
1 = 9C 1 = 3B 1 = 4 A + 4B + C = 4 A + 13 9
1
C=
1 9 1 B= 3 A=1 9
So H(s) and its inverse transform is,
H ( s ) = - 9 + 32 + 9 s s s+3 1 1 1 -3 t h (t ) = - + t + e 9 3 9
Now, lets go back to the original problem, remembering to multiply the transform through the parenthesis.
G ( s ) = 3F ( s ) - 2e -3 s F ( s ) + 8e -20 s H ( s ) + 6e -7 s H ( s )
Taking the inverse transform gives,
g ( t ) = 3 f ( t ) - 2u3 ( t ) f ( t - 3) + 8u20 ( t ) h ( t - 20 ) + 6u7 ( t ) h ( t - 7 )

So, as this example has shown, these can be a somewhat messy. However, the mess is really only that of notation and amount of work. The actual partial fraction work was identical to the previous sections work. The main difference in this section is we had to do more of it. As far as the inverse transform process goes. Again, the vast majority of that was identical to the previous section as well. So, dont let the apparent messiness of these problems get you to decide that you cant do them. Generally they arent as bad as they seem initially.
2007 Paul Dawkins
214
Solving IVPs with Laplace Transforms

Its now time to get back to differential equations. Weve spent the last three sections learning how to take Laplace transforms and how to take inverse Laplace transforms. These are going to be invaluable skills for the next couple of sections so dont forget what we learned there. Before proceeding into differential equations we will need one more formula. We will need to know how to take the Laplace transform of a derivative. First recall that f(n) denotes the nth derivative of the function f. We now have the following fact. Fact Suppose that f, f, f,f(n-1) are all continuous functions and f(n) is a piecewise continuous function. Then,
L f ( n ) = s n F ( s ) - s n-1 f ( 0 ) - s n- 2 f ( 0 ) - L - sf ( n -2) ( 0 ) - f ( n-1) ( 0 )

Since we are going to be dealing with second order differential equations it will be convenient to have the Laplace transform of the first two derivatives.
{ }
L { y } = s 2Y ( s ) - sy ( 0 ) - y ( 0 )
Notice that the two function evaluations that appear in these formulas, y ( 0 ) and y ( 0 ) , are often what weve been using for initial condition in out IVPs. So, this means that if we are to use these formulas to solve an IVP we will need initial conditions at t = 0. While Laplace transforms are particularly useful for nonhomogeneous differential equations which have Heaviside functions in the forcing function well start off with a couple of fairly simple problems to illustrate how the process works.
L { y } = sY ( s ) - y ( 0 )
Example 1 Solve the following IVP. y - 10 y + 9 y = 5t ,
y ( 0 ) = -1
y ( 0 ) = 2
Solution The first step in using Laplace transforms to solve an IVP is to take the transform of every term in the differential equation.
L { y} - 10L { y} + 9L { y} = L {5t}
Using the appropriate formulas from our table of Laplace transforms gives us the following.
s 2Y ( s ) - sy ( 0 ) - y ( 0 ) - 10 ( sY ( s ) - y ( 0 ) ) + 9Y ( s ) =
5 s2
Plug in the initial conditions and collect all the terms that have a Y(s) in them.
(s
Solve for Y(s).
- 10s + 9 ) Y ( s ) + s - 12 =
5 s2
2007 Paul Dawkins
215
Y (s) =
5 12 - s + s ( s - 9 )( s - 1) ( s - 9 )( s - 1)
2
At this point its convenient to recall just what were trying to do. We are trying to find the solution, y(t), to an IVP. What weve managed to find at this point is not the solution, but its Laplace transform. So, in order to find the solution all that we need to do is to take the inverse transform. Before doing that lets notice that in its present form we will have to do partial fractions twice. However, if we combine the two terms up we will only be doing partial fractions once. Not only that, but the denominator for the combined term will be identical to the denominator of the first term. This means that we are going to partial fraction up a term with that denominator no matter what so we might as well make the numerator slightly messier and then just partial fraction once. This is one of those things where we are apparently making the problem messier, but in the process we are going to save ourselves a fair amount of work! Combining the two terms gives,
Y (s) =
5 + 12s 2 - s 3 s 2 ( s - 9 )( s - 1)
The partial fraction decomposition for this transform is,
Y (s) =
A B C D + 2+ + s s s - 9 s -1
5 + 12 s 2 - s 3 = As ( s - 9 )( s - 1) + B ( s - 9 )( s - 1) + Cs 2 ( s - 1) + Ds 2 ( s - 9 ) 5 9 D = -2 31 C= 81 50 A= 81 B=
Picking appropriate values of s and solving for the constants gives,
s=0 s =1 s=9 s=2

Plugging in the constants gives,
5 = 9B 16 = -8D 248 = 648C 45 = -14 A + 4345 81

31 81
Y (s) =
50 81
5 9 2
s -9
2 s -1
Finally taking the inverse transform gives us the solution to the IVP.
y (t ) =
50 5 31 9t + t + e - 2et 81 9 81
2007 Paul Dawkins
That was a fair amount of work for a problem that probably could have been solved much quicker using the techniques for the previous chapter. The point of this problem however, was to show how we would use Laplace transforms to solve an IVP. There are a couple of things to note here about using Laplace transforms to solve an IVP. First, using Laplace transforms reduces a differential equation down to an algebra problem. In the case of the last example the algebra was probably more complicated than the straight forward approach from the last chapter. However, in later problems this will be reversed. The algebra, while still very messy, will often be easier than a straight forward approach. Second, unlike the approach in the last chapter, we did not need to first find a general solution, differentiate this, plug in the initial conditions and then solve for the constants to get the solution. With Laplace transforms, the initial conditions are applied during the first step and at the end we get the actual solution instead of a general solution. In many of the later problems Laplace transforms will make the problems significantly easier to work than if we had done the straight forward approach of the last chapter. Also, as we will see, there are some differential equations that simply cant be done using the techniques from the last chapter and so, in those cases, Laplace transforms will be our only solution. Lets take a look at another fairly simple problem.
Example 2 Solve the following IVP. 2 y + 3 y - 2 y = te-2 t ,
y (0) = 0
y ( 0 ) = -2
Solution As with the first example, lets first take the Laplace transform of all the terms in the differential equation. Well the plug in the initial conditions to get,
2 ( s 2Y ( s ) - sy ( 0 ) - y ( 0 ) ) + 3 ( sY ( s ) - y ( 0 ) ) - 2Y ( s ) =
( s + 2)
1
( 2s
Now solve for Y(s).
+ 3s - 2 ) Y ( s ) + 4 =
( s + 2)
Y (s) =
( 2 s - 1)( s + 2 )
4 ( 2 s - 1)( s + 2 )
Now, as we did in the last example well go ahead and combine the two terms together as we will have to partial fraction up the first denominator anyway, so we may as well make the numerator a little more complex and just do a single partial fraction. This will give,
Y (s) = =
The partial fraction decomposition is then,
2007 Paul Dawkins
1 - 4 ( s + 2)
2 3
( 2 s - 1)( s + 2 ) ( 2 s - 1)( s + 2 )
-4s 2 - 16s - 15
3
217
Y (s) =
Setting numerator equal gives,
A B C D + + + 2 2 s - 1 s + 2 ( s + 2 ) ( s + 2 )3
-4s 2 - 16 s - 15 = A ( s + 2 ) + B ( 2s - 1)( s + 2 ) + C ( 2 s - 1)( s + 2 ) + D ( 2 s - 1)

3 2
= ( A + 2 B ) s 3 + ( 6 A + 7 B + 2C ) s 2 + (12 A + 4 B + 3C + 2 D ) s + 8 A - 4 B - 2C - D
In this case its probably easier to just set coefficients equal and solve the resulting system of equation rather than pick values of s. So, here is the system and its solution.
s3 : A + 2B = 0 s2 : 6 A + 7 B + 2C = -4 s1 :12 A + 4 B + 3C + 2 D = -16 s 0 : 8 A - 4 B - 2C - D = -15
192 125 2 C=25 A= -
96 125 1 D=5 B=
We will get a common denominator of 125 on all these coefficients and factor that out when we go to plug them back into the transform. Doing this gives,
Y (s) =
25 2! 1 -192 96 10 2! + 2 3 1 + 125 2 s s 2 ( ) + 2 + 2 s s ( ) ( ) 2
Notice that we also had to factor a 2 out of the denominator of the first term and fix up the numerator of the last term in order to get them to match up to the correct entries in our table of transforms. Taking the inverse transform then gives,
t 1 25 2 -2t -2 t -2 t 2 -96e + 96e - 10te - t e 125 2
y (t ) =
Example 3 Solve the following IVP. y - 6 y + 15 y = 2sin ( 3t ) ,
y ( 0 ) = -1
y ( 0 ) = -4
Solution Take the Laplace transform of everything and plug in the initial conditions.
s 2Y ( s ) - sy ( 0 ) - y ( 0 ) - 6 ( sY ( s ) - y ( 0 ) ) + 15Y ( s ) = 2
3 s +9 ( s 2 - 6s + 15)Y ( s ) + s - 2 = s 2 6 +9
2
Now solve for Y(s) and combine into a single term as we did in the previous two examples.
- s 3 + 2s 2 - 9 s + 24 Y (s) = 2 ( s + 9 )( s 2 - 6s + 15)
2007 Paul Dawkins
218
Now, do the partial fractions on this. First lets get the partial fraction decomposition.
Y (s) =
Now, setting numerators equal gives,
As + B Cs + D + 2 2 s + 9 s - 6 s + 15
- s 3 + 2 s 2 - 9 s + 24 = ( As + B ) ( s 2 - 6 s + 15 ) + ( Cs + D ) ( s 2 + 9 ) = ( A + C ) s 3 + ( -6 A + B + D ) s 2 + (15 A - 6 B + 9C ) s + 15 B + 9 D
Setting coefficients equal and solving for the constants gives,
s3 : A + C = -1 2 s : - 6A + B + D = 2 s1 :15 A - 6 B + 9C = -9 s0 : 15 B + 9 D = 24
1 10 11 C=10 A=
1 10 5 D= 2 B=
Now, plug these into the decomposition, complete the square on the denominator of the second term and then fix up the numerators for the inverse transform process.
Y (s) = =
-11s + 25 1 s +1 + 2 2 10 s + 9 s - 6s + 15 1 s + 1 -11( s - 3 + 3) + 25 + 2 2 10 ( s - 3) + 6 s +9
8 6 11( s - 3) 13 1 s 6 3 = 2 + 2 2 2 10 s + 9 s + 9 ( s - 3) + 6 ( s - 3) + 6
Finally, take the inverse transform.
y (t ) =
1 1 cos ( 3t ) + sin ( 3t ) - 11e3t cos 10 3
( 6t ) -
8 3t e sin 6
( 6t )
To this point weve only looked at IVPs in which the initial values were at t = 0. This is because we need the initial values to be at this point in order to take the Laplace transform of the derivatives. The problem with all of this is that there are IVPs out there in the world that have initial values at places other than t = 0. Laplace transforms would not be as useful as it is if we couldnt use it on these types of IVPs. So, we need to take a look at an example in which the initial conditions are not at t = 0 in order to see how to handle these kinds of problems.
Example 4 Solve the following IVP. y + 4 y = cos ( t - 3) + 4t ,
y ( 3) = 0
y ( 3) = 7
Solution The first thing that we will need to do here it to take care of the fact that initial conditions are not at t = 0. The only way that we can take the Laplace transform of the derivatives is to have the initial conditions at t = 0. This means that we will need to formulate the IVP in such a way that the initial conditions are at t
= 0. This is actually fairly simple to do, however we will need to do a change of variable to make it work. We are going to define
h = t -3
t =h + 3
Lets start with the original differential equation.
y ( t ) + 4 y ( t ) = cos ( t - 3) + 4t
Notice that we put in the (t) part on the derivatives to make sure that we get things correct here. We will next substitute in for t.
y (h + 3) + 4 y (h + 3 ) = cos (h ) + 4 (h + 3) u (h ) = y (h + 3) du dy dt = = y (h + 3) dh dt dh
Now, to simplify life a little lets define,
Then, by the chain rule, we get that the following for the first derivative.
u (h ) =
By a similar argument we get the following for the second derivative.
u (h ) = y (h + 3)
The initial conditions for u(h) are,
u ( 0 ) = y ( 0 + 3) = y ( 3) = 0
u ( 0 ) = y ( 0 + 3) = y ( 3) = 7
The IVP under these new variables is then,
u + 4u = cos (h ) + 4h + 12,
u (0) = 0
u ( 0 ) = 7
This is an IVP that we can use Laplace transforms on provided we replace all the ts in our table with hs. So, taking the Laplace transform of this new differential equation and plugging in the new initial conditions gives,
s 2U ( s ) - su ( 0 ) - u ( 0 ) + 4 ( sU ( s ) - u ( 0 ) ) =
s 4 12 + 2+ s +1 s s s 4 12 + ( s 2 + 4s )U ( s ) - 7 = s 2 + 1 + s 2 s
2
Solving for U(s) gives,
(s
+ 4 s )U ( s ) =
s 4 + 12s + 7 s 2 + s2 + 1 s2 1 4 + 12s + 7 s 2 U ( s) = + ( s + 4 ) ( s 2 + 1) s3 ( s + 4 )
Note that unlike the previous examples we did not completely combine all the terms this time. In all the previous examples we did this because the denominator of one of the terms was the common denominator for all the terms. Therefore, upon combining, all we did was make the
numerator a little messier, and reduced the number of partial fractions required down from two to one. Note that all the terms in this transform that had only powers of s in denominator where combined for exactly this reason. In this transform however, if we combined both of the remaining terms into a single term we would be left with a fairly involved partial fraction problem. Therefore, in this case, it would probably be easier to just do partial fractions twice. Weve done several partial fractions problems in this section and many partial fraction problems in the previous couple of sections so were going to leave the details of the partial fractioning to you to check. Partial fractioning each of the terms in our transform gives us the following.
( s + 4) ( s
+ 1)
1 -s + 4 s + 4 17 s 2 + 1
1 17
17 17 4 + 12s + 7 s 2 1 11 4 16 16 = + + 3 3 2 s ( s + 4) s s s s+4
Plugging these into our transform and combining like terms gives us
U (s) = =
17 273 1 11 1 -s + 4 4 16 272 + + + 2 3 2 s s s s + 4 17 s + 1 17 273 11 1 2! 1 -s 4 2! + 42 + 16 - 272 + 2 + 2 3 s s s s + 4 17 s + 1 s + 1
Now, taking the inverse transform will give the solution to our new IVP. Dont forget to use hs instead of ts!
1 11 17 273 -4h 1 e + ( 4sin (h ) - cos (h ) ) u (h ) = h 2 + h + 2 4 16 272 17

This is not the solution that we are after of course. We are after y(t). However, we can get this by noticing that
y ( t ) = y (h + 3) = u (h ) = u ( t - 3)
So the solution to the original IVP is,
1 11 17 273 1 2 ( t - 3) + ( t - 3) + - e-4(t -3) + ( 4sin ( t - 3) - cos ( t - 3) ) 2 4 16 272 17 1 2 1 43 273 -4( t -3) 1 y (t ) = t - t - e + ( 4sin ( t - 3) - cos ( t - 3) ) 2 4 16 272 17 y (t ) =
So, we can now do IVPs that dont have initial conditions that are at t = 0. We also saw in the last example that it isnt always the best to combine all the terms into a single partial fraction problem as we have been doing prior to this example. The examples worked in this section would have just as easy, if not easier, if we had used techniques from the previous chapter. They were worked here using Laplace transforms to illustrate the technique and method.
2007 Paul Dawkins
221
Nonconstant Coefficient IVPs

In this section we are going to see how Laplace transforms can be used to solve some differential equations that do not have constant coefficients. This is not always an easy thing to do. However, there are some simple cases that can be done. To do this we will need a quick fact. Fact If f(t) is a piecewise continuous function on [ 0, ) of exponential order then,
lim F ( s ) = 0
s
(1)
A function f(t) is said to be of exponential order a if there is exist positive constants T and M such that
f ( t ) Mea t
for all t T
Put in other words, a function that is of exponential order will grow no faster than
Mea t
for some M and a and all sufficiently large t. One way to check whether a function is of exponential order or not is to compute the following limit.
lim
t
f (t ) ea t
If this limit is finite for some a then the function will be of exponential order a. Likewise, if the limit is infinite for every a then the function is not of exponential order. Almost all of the functions that you are liable to deal with in a first course in differential equations are of exponential order. A good example of a function that is not of exponential order is
f ( t ) = et
We can check this by computing the above limit.
3
This is true for any value of a and so the function is not of exponential order. Do not worry too much about this exponential order stuff. This fact is occasionally needed in using Laplace transforms with non constant coefficients. So, lets take a look at an example.
3 t (t 2 -a ) et lim a t = lim et -a t = lim e = t e t t
2007 Paul Dawkins
222
Example 1 Solve the following IVP. y + 3ty - 6 y = 2,
y (0) = 0
y ( 0 ) = 0
Solution So, for this one we will need to recall that #30 in our table of Laplace transforms tells us that,
L {ty} = -
d ( L { y}) ds d = - ( sY ( s ) - y ( 0 ) ) ds = - sY ( s ) - Y ( s )
So, upon taking the Laplace transforms of everything and plugging in the initial conditions we get,
s 2Y ( s ) - sy ( 0 ) - y ( 0 ) + 3 ( - sY ( s ) - Y ( s ) ) - 6Y ( s ) = 2 s 2 3 s Y (s) + - Y ( s) = - 2 3s s 3 -3sY ( s ) + ( s 2 - 9 ) Y ( s ) =
2 s
Unlike the examples in the previous section where we ended up with a transform for the solution, here we get a linear first order differential equation that must be solved in order to get a transform for the solution. The integrating factor for this differential equation is,
s ln ( s ) - s (3 s - 3 ) ds 6 =e
3
m ( s ) = e
= s 3e
-s 6
Multiplying through, integrating and solving for Y(s) gives,

2 3 - s62 - 2 se - s6 ds s Y s ds e = ( ) 3
s 3e
-s 6
Y ( s ) = 2e
-s 6
+c
s2
2 e6 Y (s) = 3 + c 3 s s
Now, we have a transform for the solution. However that second term looks unlike anything weve seen to this point. This is where the fact about the transforms of exponential order functions comes into play. We are going to assume that whatever our solution is, it is of exponential order. This means that
2 ce s6 lim 3 + 3 = 0 s s s
2
2007 Paul Dawkins
223
The first term does go to zero in the limit. The second term however, will only go to zero if c = 0. Therefore, we must have c = 0 in order for this to be the transform of our solution. So, the transform of our solution, as well as the solution is,
Y (s) =
2 s3
y (t ) = t 2
Ill leave it to you to verify that this is in fact a solution if youd like to. Now, not all nonconstant differential equations need to use (1). So, lets take a look at one more example.
Example 2 Solve the following IVP. ty - ty + y = 2,

Solution From the first example we have,
y (0) = 2
y ( 0 ) = -4
L {ty} = - sY ( s ) - Y ( s ) d ( L { y}) ds d = - ( s 2Y ( s ) - sy ( 0 ) - y ( 0 ) ) ds = - s 2Y ( s ) - 2 sY ( s ) + y ( 0 ) 2 s
Well also need,
L {ty} = -
Taking the Laplace transform of everything and plugging in the initial conditions gives,
- s 2Y ( s ) - 2sY ( s ) + y ( 0 ) - ( - sY ( s ) - Y ( s ) ) + Y ( s ) =
( s - s ) Y ( s ) + ( 2 - 2s ) Y ( s ) + 2 = 2 s
2
s (1 - s ) Y ( s ) + 2 (1 - s ) Y ( s ) = 2 2 Y(s) + Y (s) = 2 s s
2 (1 - s ) s
Once again we have a linear first order differential equation that we must solve in order to get a transform for the solution. Notice as well that we never used the second initial condition in this work. That is okay, we will use it eventually. Since this linear differential equation is much easier to solve compared to the first one, well leave the details to you. Upon solving the differential equation we get,
Y (s) =
2 c + s s2
Now, this transform goes to zero for all values of c and we can take the inverse transform of the second term. Therefore, we wont need to use (1) to get rid of the second term as did in the previous example. Taking the inverse transform gives,
y ( t ) = 2 + ct
Now, is where we will use the second initial condition. Upon differentiating and plugging in the second initial condition we can see that c = -4. So, the solution to this IVP is,
y ( t ) = 2 - 4t
So, weve seen how to use Laplace transforms to solve some nonconstant coefficient differential equations. Notice however that all we did was add in an occasional t to the coefficients. We couldnt get too complicated with the coefficients. If we had we would not have been able to easily use Laplace transforms to solve them. Sometimes Laplace transforms can be used to solve nonconstant differential equations, however, in general, nonconstant differential equations are still very difficult to solve.
2007 Paul Dawkins
225
IVPs With Step Functions

In this section we will use Laplace transforms to solve IVPs which contain Heaviside functions in the forcing function. This is where Laplace transform really starts to come into its own as a solution method. To work these problems well just need to remember the following two formulas,
L {uc ( t ) f ( t - c )} = e - c s F ( s )
L-1 {e- c s F ( s )} = uc ( t ) f ( t - c )
where F ( s ) = L { f ( t )}
where f ( t ) = L-1 { F ( s )}
In other words, we will always need to remember that in order to take the transform of a function that involves a Heaviside weve got to make sure the function has been properly shifted. Lets work an example.
Example 1 Solve the following IVP. y - y + 5 y = 4 + u2 ( t ) e 4- 2 t ,
y (0) = 2
y ( 0 ) = -1
Solution First lets rewrite the forcing function to make sure that its being shifted correctly and to identify the function that is actually being shifted.
y - y + 5 y = 4 + u2 ( t ) e -2( t -2)
So, it is being shifted correctly and the function that is being shifted is e -2t . Taking the Laplace transform of everything and plugging in the initial conditions gives,
s 2Y ( s ) - sy ( 0 ) - y ( 0 ) - ( sY ( s ) - y ( 0 ) ) + 5Y ( s ) =
4 e -2 s + s s+2
(s
Now solve for Y(s).
- s + 5 ) Y ( s ) - 2s + 3 =
4 e -2 s + s s+2
4 e -2 s + + 2s - 3 s s+2 2 -2 s ( s 2 - s + 5) Y ( s ) = 2s -s3s + 4 + se+ 2 2 s 2 - 3s + 4 1 -2 s e Y (s) = + s ( s 2 - s + 5) ( s + 2) ( s2 - s + 5)
( s 2 - s + 5) Y ( s ) =
Y ( s ) = F ( s ) + e -2 s G ( s )
Notice that we combined a couple of terms to simplify things a little. Now we need to partial fraction F(s) and G(s). Well leave it to you to check the details of the partial fractions.
2007 Paul Dawkins
226
F (s) = G (s) =
2 s 2 - 3s + 4 1 4 6 s - 11 = + 2 s ( s 2 - s + 5) 5 s s - s + 5 1 1 1 s -3 = - 2 2 ( s + 2 ) ( s - s + 5 ) 11 s + 2 s - s + 5
We now need to do the inverse transforms on each of these. Well start with F(s).
1 1 4 6(s - 1 2 + 2 ) - 11 + F (s) = 1 2 19 5 s + s ( 2) 4 2 8 19 6(s - 1 14 2 2) 19 = + s ( s - 1 )2 + 19 ( s - 1 ) 2 + 19 5 2 4 2 4
19 16 2 19 t t 1 f ( t ) = 4 + 6e 2 cos t e sin 2 2 t 5 19
Now G(s).
1 s- 1 1 1 2 + 2 -3 2 1 19 + 11 2 s + s ( ) 2 4
G ( s) =
5 19 1 1 1 s- 2 2 19 = + 2 2 19 19 1 1 + 11 2 s + + s s ( ) ( ) 2 4 2 4
g (t ) =
19 19 t t 1 -2 t 2 5 2 e - e cos t + e sin 2 2 t 11 19
Okay, we can now get the solution to the differential equation. Starting with the transform we get,
Y ( s ) = F ( s ) + e -2 s G ( s )
y ( t ) = f ( t ) + u2 ( t ) g ( t - 2 )
where f(t) and g(t) are the functions shown above. There is can be a fair amount of work involved in solving differential equations that involve Heaviside functions. Lets take a look at another example or two.
2007 Paul Dawkins
227
Example 2 Solve the following IVP. y - y = cos ( 2t ) + cos ( 2t - 12 ) u6 ( t )
y ( 0 ) = -4, y ( 0 ) = 0
Solution Lets rewrite the differential equation so we can identify the function that is actually being shifted.
y - y = cos ( 2t ) + cos ( 2 ( t - 6 ) ) u6 ( t )
So, the function that is being shifted is cos ( 2t ) and it is being shifted correctly. Taking the Laplace transform of everything and plugging in the initial conditions gives,
s 2Y ( s ) - sy ( 0 ) - y ( 0 ) - ( sY ( s ) - y ( 0 ) ) =
Now solve for Y(s).
s se -6 s + s2 + 4 s2 + 4 -6 s ( s 2 - s )Y ( s ) + 4s - 4 = s 2 s+ 4 + ss2e+ 4
(s
- s )Y ( s ) =
s + se -6 s - 4s + 4 s2 + 4 s (1 + e -6 s ) s -1 Y (s) = -4 2 s ( s - 1) s ( s - 1) ( s + 4 ) = 1 + e -6 s 4 2 ( s - 1) ( s + 4 ) s 4 s
Y ( s ) = (1 + e -6 s ) F ( s ) +
Notice that we combined the first two terms to simplify things a little. Also there was some canceling going on in this one. Do not expect that to happen on a regular basis. We now need to partial fraction F(s). Well leave the details to you to check.
F (s) =
1 1 1 s +1 = - 2 2 ( s - 1) ( s + 4 ) 5 s - 1 s + 4
1 1 f ( t ) = et - cos ( 2t ) - sin ( 2t ) 5 2
Y ( s ) = F ( s ) + F ( s ) e -6 s +
where f(t) is given above.
4 s y ( t ) = f ( t ) + u6 ( t ) f ( t - 6 ) + 4
2007 Paul Dawkins
228
Example 3 Solve the following IVP. y - 5 y - 14 y = 9 - u3 ( t ) + 4 ( t - 1) u1 ( t )
y ( 0 ) = 0, y ( 0 ) = 10
Solution Lets take the Laplace transform of everything and note that in the third term we are shifting 4t.
s 2Y ( s ) - sy ( 0 ) - y ( 0 ) - 5 ( sY ( s ) - y ( 0 ) ) - 14Y ( s ) =
9 e -3 s e- s + +4 2 s s s -3 s -s ( s 2 - 5s - 14 ) Y ( s ) - 10 = 9 +se + 4 es 2
Now solve for Y(s).
( s 2 - 5s - 14 )Y ( s ) - 10 =
9 + e -3 s e- s +4 2 s s -3 s 9+e 4e - s 10 Y (s) = + 2 + s ( s - 7 )( s + 2 ) s ( s - 7 )( s + 2 ) ( s - 7 )( s + 2 ) Y ( s ) = ( 9 + e-3s ) F ( s ) + 4e - s G ( s ) + H ( s )
So, we have three functions that well need to partial fraction for this problem. Ill leave it to you to check the details.
F (s) =
1 1 1 1 1 1 1 =+ + 14 s 63 s - 7 18 s + 2 s ( s - 7 )( s + 2 )
1 1 7t 1 -2 t + e + e 14 63 18 1 5 1 1 1 1 1 1 1 G ( s) = 2 = + 2 441 s - 7 36 s + 2 s ( s - 7 )( s + 2 ) 196 s 14 s f (t ) = g (t ) = 5 1 1 7 t 1 -2 t - t+ e - e 196 14 441 36 10 10 1 10 1 H (s) = = ( s - 7 )( s + 2 ) 9 s - 7 9 s + 2 h (t ) = 10 7 t 10 -2t e - e 9 9
Y ( s ) = 9 F ( s ) + e -3s F ( s ) + 4e - s G ( s ) + H ( s )
y ( t ) = 9 f ( t ) + u3 ( t ) f ( t - 3) + 4u1 ( t ) g ( t - 1) + h ( t )
where f(t), g(t) and h(t) are given above. Lets work one more example.
2007 Paul Dawkins
229
Example 4 Solve the following IVP. y + 3 y + 2 y = g ( t ) ,

where,
y (0) = 0 t <6 6 t < 10 t 10
y ( 0 ) = -2
2 g ( t ) = t 4
Solution The first step is to get g(t) written in terms of Heaviside functions so that we can take the transform.
g ( t ) = 2 + ( t - 2 ) u6 ( t ) + ( 4 - t ) u10 ( t )
Now, while this is g(t) written in terms of Heaviside functions it is not yet in proper form for us to take the transform. Remember that each function must be shifted by a proper amount. So, getting things set up for the proper shifts gives us,
g ( t ) = 2 + ( t - 6 + 6 - 2 ) u6 ( t ) + ( 4 - ( t - 10 + 10 ) ) u10 ( t ) g ( t ) = 2 + ( t - 6 + 4 ) u6 ( t ) + ( -6 - ( t - 10 ) ) u10 ( t )
So, for the first Heaviside it looks like f ( t ) = t + 4 is the function that is being shifted and for the second Heaviside it looks like f ( t ) = -6 - t is being shifted. Now take the Laplace transform of everything and plug in the initial conditions.
s 2Y ( s ) - sy ( 0 ) - y ( 0 ) + 3 ( sY ( s ) - y ( 0 ) ) + 2Y ( s ) =
2 -6 s 1 4 -10 s 1 6 +e 2 + -e 2 + s s s s s 1 4 1 6 + e -6 s 2 + - e -10 s 2 + ( s 2 + 3s + 2) Y ( s ) + 2 = 2 s s s s s
Solve for Y(s).
2 -6 s 1 4 -10 s 1 6 +e 2 + -e 2 + -2 s s s s s -6 s -10 s -6 s -10 s -e -2 ( s 2 + 3s + 2 )Y ( s ) = 2 + 4e s - 6e + e s 2 2 + 4e -6 s - 6e -10 s 2 e -6 s - e -10 s Y (s) = + 2 s ( s + 1)( s + 2 ) s ( s + 1)( s + 2 ) ( s + 1)( s + 2 )
(s
+ 3s + 2 ) Y ( s ) =
Y ( s ) = ( 2 + 4e -6 s - 6e -10 s ) F ( s ) + ( e -6 s - e -10 s ) G ( s ) - H ( s )
Now, in the solving process we simplified things into as few terms as possible. Even doing this, it looks like well still need to do three partial fractions. Ill leave the details of the partial fractioning to you to verify. The partial fraction form and
inverse transform of each of these are.
F (s) = f (t ) =
1 1 1 1 =2+ 2 s ( s + 1)( s + 2 ) s s + 1 s + 2
1 -t 1 -2 t -e + e 2 2 3 1 1 1 1 G (s) = 2 = - 4 + 22 + - 4 s ( s + 1)( s + 2 ) s s s +1 s + 2 3 1 1 g ( t ) = - + t + e - t - e -2 t 4 2 4 2 2 2 H (s) = = ( s + 1)( s + 2 ) s + 1 s + 2 h ( t ) = 2e -t - 2e -2 t

Putting this all back together is going to be a little messy. First rewrite the transform a little to make the inverse transform process possible.
Y ( s ) = 2 F ( s ) + e -6 s ( 4 F ( s ) + G ( s ) ) - e -10 s ( 6 F ( s ) + G ( s ) ) - H ( s )
Now, taking the inverse transform of all the pieces gives us the final solution to the IVP. where f(t), g(t), and h(t) are defined above.
y ( t ) = 2 f ( t ) - h ( t ) + u6 ( t ) ( 4 f ( t - 6 ) + g ( t - 6 ) ) - u10 ( t ) ( 6 f ( t - 10 ) + g ( t - 10 ) )
So, the answer to this example is a little messy to write down, but overall the work here wasnt too terribly bad. Before proceeding with the next section lets see how we would have had to solve this IVP if we hadnt had Laplace transforms. To solve this IVP we would have had to solve three separate IVPs. One for each portion of g(t). Here is a list of the IVPs that we would have had to solve. 1. 0 < t < 6
y + 3 y + 2 y = 2,
y (0) = 0
y ( 0 ) = -2
The solution to this IVP, with some work, can be made to look like,
y1 ( t ) = 2 f ( t ) - h ( t )
2.
6 t < 10 y + 3 y + 2 y = t ,
y ( 6 ) = y1 ( 6 )
y ( 6 ) = y1 ( 6 )
where, y1(t) is the solution to the first IVP. The solution to this IVP, with some work, can be made to look like,
y2 ( t ) = 2 f ( t ) - h ( t ) + 4 f ( t - 6 ) + g ( t - 6 )
2007 Paul Dawkins
231
3.
t 10 y + 3 y + 2 y = 4,
y (10 ) = y2 (10 )
y (10 ) = y2 (10 )
where, y2(t) is the solution to the second IVP. The solution to this IVP, with some work, can be made to look like,
y3 ( t ) = 2 f ( t ) - h ( t ) + 4 f ( t - 6 ) + g ( t - 6 ) - 6 f ( t - 10 ) - g ( t - 10 )
There is a considerable amount of work required to solve all three of these and in each of these the forcing function is not that complicated. Using Laplace transforms saved us a fair amount of work.
2007 Paul Dawkins
232
Dirac Delta Function

When we first introduced Heaviside functions we noted that we could think of them as switches changing the forcing function, g(t), at specified times. However, Heaviside functions are really not suited to forcing functions that exert a large force over a small time frame. Examples of this kind of forcing function would be a hammer striking an object or a short in an electrical system. In both of these cases a large force (or voltage) would be exerted on the system over a very short time frame. The Dirac Delta function is used to deal with these kinds of forcing function. Dirac Delta Function There are many ways to actually define the Dirac Delta function. To see some of these definitions visit Wolframs MathWorld. There three main properties of the Dirac Delta function that we need to be aware of. These are, 1. d ( t - a ) = 0, t a 2.
a -e d ( t - a ) dt = 1,
a +e
a +e
e >0 e >0
3.
a -e f ( t ) d ( t - a ) dt = f ( a ) ,
At t = a the Dirac Delta function is sometimes thought of has having an infinite value. So, the Dirac Delta function is a function that is zero everywhere except one point and at that point it can be thought of as either undefined or as having an infinite value. Note that the integrals in the second and third property are actually true for any interval containing t = a , provided its not one of the endpoints. The limits given here are needed to prove the properties and so they are also given in the properties. We will however use the fact that they are true provided we are integrating over an interval containing t = a . This is a very strange function. It is zero everywhere except one point and yet the integral of any interval containing that one point has a value of 1. The Dirac Delta function is not a real function as we think of them. It is instead an example of something called a generalized function or distribution. Despite the strangeness of this function it does a very nice job of modeling sudden shocks or large forces to a system. Before solving an IVP we will need the transform of the Dirac Delta function. We can use the third property above to get this.
L {d ( t - a )} = e - st d ( t - a ) dt = e - a s
0
provided a > 0
Note that often the second and third properties are given with limits of infinity and negative infinity, but they are valid for any interval in which t = a is in the interior of the interval.
With this we can now solve an IVP that involves a Dirac Delta function.
Example 1 Solve the following IVP. y + 2 y - 15 y = 6d ( t - 9 ) ,
y ( 0 ) = -5
y ( 0 ) = 7
Solution As with all previous problems well first take the Laplace transform of everything in the differential equation and apply the initial conditions.
s 2Y ( s ) - sy ( 0 ) - y ( 0 ) + 2 ( sY ( s ) - y ( 0 ) ) - 15Y ( s ) = 6e-9 s
(s
+ 2 s - 15 ) Y ( s ) + 5s + 3 = 6e -9 s
Now solve for Y(s).
Y (s) =
6e-9 s 5s + 3 ( s + 5 )( s - 3) ( s + 5 )( s - 3)
= 6e -9 s F ( s ) - G ( s )
Well leave it to you to verify the partial fractions and their inverse transforms are,
F (s) =
( s + 5 )( s - 3)
1 8
s-3
1 8
s+5
1 1 f ( t ) = e 3 t - e -5 t 8 8 9 11 5s + 3 G (s) = = 4 + 4 ( s + 5)( s - 3 ) s - 3 s + 5 9 11 g ( t ) = e3t + e -5t 4 4

Y ( s ) = 6e -9 s F ( s ) - G ( s ) y ( t ) = 6u9 ( t ) f ( t - 9 ) - g ( t )
where, f(t) and g(t) are defined above.
Example 2 Solve the following IVP. 2 y + 10 y = 3u12 ( t ) - 5d ( t - 4 ) ,
y ( 0 ) = -1
y ( 0 ) = -2
Solution Take the Laplace transform of everything in the differential equation and apply the initial conditions.
2 ( s 2Y ( s ) - sy ( 0 ) - y ( 0 ) ) + 10Y ( s ) =
3e -12s - 5e -4 s s
( 2s 2 + 10) Y ( s ) + 2s + 4 =
Now solve for Y(s).
3e -12 s - 5e -4 s s
Y (s) =
3e -12 s 5e -4 s 2s + 4 - 2 2 2 s ( 2 s + 10 ) 2 s + 10 2s + 10
= 3e -12 s F ( s ) - 5e -4 s G ( s ) - H ( s )
Well need to partial fraction the first function. The remaining two will just need a little work and theyll be ready. Ill leave the details to you to check.
F (s) = f (t ) =
1 1 1 1 s = 2 s ( 2 s + 10 ) 10 s 10 s + 5
2
1 1 - cos 5 t 10 10 1 g (t ) = sin 5 t 2 5 2 h ( t ) = cos 5 t + sin 5
( ) ( ) ( 5 t)
( )
Y ( s ) = 3e -12 s F ( s ) - 5e-4 s G ( s ) - H ( s ) y ( t ) = 3u12 ( t ) f ( t - 12 ) - 5u4 ( t ) g ( t - 4 ) - h ( t )
where, f(t),g(t) and h(t) are defined above. So, with the exception of the new function these work the same way that all the problems that weve seen to this point work. Note as well that the exponential was introduced into the transform by the Dirac Delta function, but once in the transform it doesnt matter where it came from. In other words, when we went to the inverse transforms it came back out as a Heaviside function. Before proceeding to the next section lets take a quick side trip and note that we can relate the Heaviside function and the Dirac Delta function. Start with the following integral.
0 d u a du = ( ) - 1
t
if t < a if t > a
However, this is precisely the definition of the Heaviside function. So,
(t ) = ua
t -
d ( u - a ) du = ua ( t )
Now, recalling the Fundamental Theorem of Calculus, we get,
d dt
t -
d ( u - a ) du = d ( t - a )
So, the derivative of the Heaviside function is the Dirac Delta function.
2007 Paul Dawkins
235
Convolution Integrals
On occasion we will run across transforms of the form,
H (s) = F (s)G (s)
that cant be dealt with easily using partial fractions. We would like a way to take the inverse transform of such a transform. We can use a convolution integral to do this. Convolution Integral If f(t) and g(t) are piecewise continuous function on [ 0, ) then the convolution integral of f(t) and g(t) is,
( f * g )( t ) = 0 f ( t - t ) g (t ) dt
A nice property of convolution integrals is. Or,
( f * g )( t ) = ( g * f )( t )
t 0
f ( t - t ) g (t ) dt = f (t ) g ( t - t ) dt
0
The following fact will allow us to take the inverse transforms of a product of transforms. Fact
L { f * g} = F ( s ) G ( s )
L-1 { F ( s ) G ( s )} = ( f * g )( t )
Lets work a quick example to see how this can be used.
Example 1 Use a convolution integral to find the inverse transform of the following transform. 1 H (s) = 2 2 ( s + a2 )
Solution First note that we could use #11 from out table to do this one so that will be a nice check against our work here. Now, since we are going to use a convolution integral here we will need to write it as a product whose terms are easy to find the inverse transforms of. This is easy to do in this case.
1 1 H (s) = 2 2 2 2 s + a s + a
So, in this case we have,
F (s) = G (s) =
1 s + a2
2
f (t ) = g (t ) =
1 sin ( at ) a
Using a convolution integral h(t) is,
2007 Paul Dawkins
236
h ( t ) = ( f * g )( t ) 1 t sin ( at - at ) sin ( at ) dt a2 0 1 = 3 ( sin ( at ) - at cos ( at ) ) 2a =
This is exactly what we would have gotten by using #11 from the table. Convolution integrals are very useful in the following kinds of problems.
Example 2 Solve the following IVP 4 y + y = g ( t ) ,
y ( 0) = 3
y ( 0 ) = -7
Solution First, notice that the forcing function in this case has not been specified. Prior to this section we would not have been able to get a solution to this IVP. With convolution integrals we will be able to get a solution to this kind of IVP. The solution will be in terms of g(t) but it will be a solution. Take the Laplace transform of all the terms and plug in the initial conditions.
4 ( s 2Y ( s ) - sy ( 0 ) - y ( 0 ) ) + Y ( s ) = G ( s )
2
( 4s
+ 1) Y ( s ) - 12s + 28 = G ( s )
Notice here that all we could do for the forcing function was to write down G(s) for its transform. Now, solve for Y(s).
( 4s
+ 1) Y ( s ) = G ( s ) + 12s - 28 Y (s) =
G ( s) 12 s - 28 + 2 4(s + 1 4 ( s2 + 1 4) 4)
We factored out a 4 from the denominator in preparation for the inverse transform process. To take inverse transforms well need to split up the first term and well also rewrite the second term a little.
Y (s) = =
G (s) 12s - 28 + 4 ( s2 + 1 4 ( s2 + 1 4) 4)
2 72 3s 1 2 2 + + G s ( ) 2 2 1 1 s2 + 1 s + 4 s +4 4 4
Now, the first two terms are easy to inverse transform. Well need to use a convolution integral on the last term. The two functions that we will be using are,
g (t )
t f ( t ) = 2sin 2
We can shift either of the two functions in the convolution integral. Well shift g(t) in our solution. Taking the inverse transform gives us,
t t 1 t y ( t ) = 3cos + 14sin + sin g ( t - t ) dt 2 2 2 0 2

So, once we decide on a g(t) all we need to do is to an integral and well have the solution. As this last example has shown, using convolution integrals will allow us to solve IVPs with general forcing functions. This could be very convenient in cases where we have a variety of possible forcing functions and dont which one were going to use. With a convolution integral all that we need to do in these cases is solve the IVP once then go back and evaluate an integral for each possible g(t). This will save us the work of having to solve the IVP for each and every g(t).
2007 Paul Dawkins
238
f (t ) = L
1. 3. 5. 7. 9. 1
-1
{F ( s )}
F ( s ) = L { f ( t )} 1 s n! s n +1
Table of Laplace Transforms
f ( t ) = L -1 { F ( s )}
F ( s ) = L { f ( t )} 1 s-a G ( p + 1) s p +1 1 3 5L ( 2n - 1) p 2n s 2 s 2 s + a2 s2 - a2
2 n+ 1
2. 4. 6. 8.
2
e at t p , p > -1 t
n-1 2
t n , n = 1, 2,3,K t sin ( at ) t sin ( at ) sin ( at ) - at cos ( at ) cos ( at ) - at sin ( at ) sin ( at + b ) sinh ( at ) e at sin ( bt ) e at sinh ( bt ) t ne at , n = 1, 2, 3, K uc ( t ) = u ( t - c )
Heaviside Function
p
2s a 2 s + a2 2as
2
3 2
, n = 1, 2,3,K
cos ( at ) t cos ( at ) sin ( at ) + at cos ( at ) cos ( at ) + at sin ( at ) cos ( at + b ) cosh ( at ) e at cos ( bt ) e at cosh ( bt ) f ( ct )
(s
+ a2 )
10.
(s
+ a2 )
11.
13. 15. 17. 19. 21. 23. 25. 27. 29. 31. 33. 35. 37.
(s + a ) s (s - a ) (s + a )
2 2 2 2 2 2 2 2
2a 3
12.
14. 16. 18. 20. 22. 24. 26. 28. 30. 32. 34. 36.
(s + a ) s ( s + 3a ) (s + a )
2 2 2 2 2 2 2 2
2as 2
s sin ( b ) + a cos ( b ) s2 + a2 a 2 s - a2 b
s cos ( b ) - a sin ( b ) s2 + a2 s 2 s - a2 s-a
(s - a)
+ b2 -b
n +1 2
(s - a)
+ b2 - b2
s-a
(s - a)
(s - a)
n!
(s - a)
1 s F c c e - cs e - cs L { g ( t + c )}
uc ( t ) f ( t - c ) ect f ( t ) 1 f (t ) t
e - cs s - cs e F (s) F (s - c)
s
d (t - c )
Dirac Delta Function
uc ( t ) g ( t )
t 0
t n f ( t ) , n = 1, 2,3,K
F ( u ) du
f ( v ) dv
( -1)n F ( n) ( s ) F (s)
s
f ( t - t ) g (t ) dt
0
F (s)G ( s) sF ( s ) - f ( 0 )
f ( t + T ) = f (t ) f ( t )
T 0
- st
f ( t ) dt
f (t ) f ( n) ( t )
2007 Paul Dawkins
1 - e - sT s 2 F ( s ) - sf ( 0 ) - f ( 0 )
s n F ( s ) - s n-1 f ( 0 ) - s n - 2 f ( 0 ) L - sf ( n -2) ( 0 ) - f ( n-1) ( 0 )

Table Notes
1. This list is not inclusive and only contains some of the more commonly used Laplace transforms and formulas. 2. Recall the definition of hyperbolic functions.
et + e - t cosh ( t ) = 2
et - e- t sinh ( t ) = 2
3. Be careful when using normal trig function vs. hyperbolic functions. The only difference in the formulas is the + a2 for the normal trig functions becomes a - a2 for the hyperbolic functions! 4. Formula #4 uses the Gamma function which is defined as
G ( t ) = e - x xt -1 dx
0
If n is a positive integer then,
G ( n + 1) = n !
The Gamma function is an extension of the normal factorial function. Here are a couple of quick facts for the Gamma function
G ( p + 1) = pG ( p ) p ( p + 1)( p + 2 )L ( p + n - 1) = 1 G = p 2 G ( p + n) G( p)
2007 Paul Dawkins
240
Systems of Differential Equations

Introduction
To this point weve only looked as solving single differential equations. However, many real life situations are governed by a system of differential equations. Consider the population problems that we looked at back in the modeling section of the first order differential equations chapter. In these problems we looked only at a population of one species, yet the problem also contained some information about predators of the species. We assumed that any predation would be constant in these cases. However, in most cases the level of predation would also be dependent upon the population of the predator. So, to be more realistic we should also have a second differential equation that would give the population of the predators. Also note that the population of the predator would be, in some way, dependent upon the population of the prey as well. In other words, we would need to know something about one population to find the other population. So to find the population of either the prey or the predator we would need to solve a system of at least two differential equations. The next topic of discussion is then how to solve systems of differential equations. However, before doing this we will first need to do a quick review of Linear Algebra. Much of what we will be doing in this chapter will be dependent upon topics from linear algebra. This review is not intended to completely teach you the subject of linear algebra, as that is a topic for a complete class. The quick review is intended to get you familiar enough with some of the basic topics that you will be able to do the work required once we get around to solving systems of differential equations. Here is a brief listing of the topics covered in this chapter. Review : Systems of Equations The traditional starting point for a linear algebra class. We will use linear algebra techniques to solve a system of equations. Review : Matrices and Vectors A brief introduction to matrices and vectors. We will look at arithmetic involving matrices and vectors, inverse of a matrix, determinant of a matrix, linearly independent vectors and systems of equations revisited. Review : Eigenvalues and Eigenvectors Finding the eigenvalues and eigenvectors of a matrix. This topic will be key to solving systems of differential equations. Systems of Differential Equations Here we will look at some of the basics of systems of differential equations. Solutions to Systems We will take a look at what is involved in solving a system of differential equations. Phase Plane A brief introduction to the phase plane and phase portraits. Real Eigenvalues Solving systems of differential equations with real eigenvalues.
Complex Eigenvalues Solving systems of differential equations with complex eigenvalues. Repeated Eigenvalues Solving systems of differential equations with repeated eigenvalues. Nonhomogeneous Systems Solving nonhomogeneous systems of differential equations using undetermined coefficients and variation of parameters. Laplace Transforms A very brief look at how Laplace transforms can be used to solve a system of differential equations. Modeling In this section well take a quick look at some extensions of some of the modeling we did in previous chapters that lead to systems of equations.
2007 Paul Dawkins
242
Review : Systems of Equations

Because we are going to be working almost exclusively with systems of equations in which the number of unknowns equals the number of equations we will restrict our review to these kinds of systems. All of what we will be doing here can be easily extended to systems with more unknowns than equations or more equations than unknowns if need be. Lets start with the following system of n equations with the n unknowns, x1, x2,, xn.
a21 x1 + a22 x2 + L + a2 n xn = b2 M an1 x1 + an 2 x2 + L + ann xn = bn
a11 x1 + a12 x2 + L + a1n xn = b1
(1)
Note that in the subscripts on the coefficients in this system, aij, the i corresponds to the equation that the coefficient is in and the j corresponds to the unknown that is multiplied by the coefficient. To use linear algebra to solve this system we will first write down the augmented matrix for this system. An augmented matrix is really just the all the coefficients of the system and the numbers for the right side of the system written in matrix form. Here is the augmented matrix for this system.
a11 a 21 M an1
a12 a22 M an 2
L a1n L a2 n O M L ann
b1 b2 M bn
To solve this system we will use elementary row operations (which well define these in a bit) to rewrite the augmented matrix in triangular form. The matrix will be in triangular form if all the entries below the main diagonal (the diagonal containing a11, a22, ,ann) are zeroes. Once this is done we can recall that each row in the augmented matrix corresponds to an equation. We will then convert our new augmented matrix back to equations and at this point solving the system will become very easy. Before working an example lets first define the elementary row operations. There are three of them. 1. Interchange two rows. This is exactly what it says. We will interchange row i with row j. The notation that well use to denote this operation is : Ri R j 2. Multiply row i by a constant, c. This means that every entry in row i will get multiplied by the constant c. The notation for this operation is : cRi 3. Add a multiply of row i to row j. In our heads we will multiply row i by an appropriate constant and then add the results to row j and put the new row back into row j leaving row i in the matrix unchanged. The notation for this operation is : cRi + R j
Its always a little easier to understand these operations if we see them in action. So, lets solve a couple of systems.
Example 1 Solve the following system of equations. -2 x1 + x2 - x3 = 4 x1 + 2 x2 + 3x3 = 13 3 x1 + x3 = -1

Solution The first step is to write down the augmented matrix for this system. Dont forget that coefficients of terms that arent present are zero.
-2 1 -1 4 1 2 3 13 3 0 1 -1
Now, we want the entries below the main diagonal to be zero. The main diagonal has been colored red so we can keep track of it during this first example. For reasons that will be apparent eventually we would prefer to get the main diagonal entries to all be ones as well. We can get a one in the upper most spot by noticing that if we interchange the first and second row we will get a one in the uppermost spot for free. So lets do that.
-2 1 -1 4 1 2 3 13 R1 R2 1 -2 1 -1 4 2 3 13 3 0 1 -1 3 0 1 -1
Now we need to get the last two entries (the -2 and 3) in the first column to be zero. We can do this using the third row operation. Note that if we take 2 times the first row and add it to the second row we will get a zero in the second entry in the first column and if we take -3 times the first row to the third row we will get the 3 to be a zero. We can do both of these operations at the same time so lets do that.
1 2 3 13 2 R1 + R2 1 2 3 13 -2 1 -1 4 -3R + R 0 5 5 30 1 3 3 0 1 -1 0 -6 -8 -40
Before proceeding with the next step, lets make sure that you followed what we just did. Lets take a look at the first operation that we performed. This operation says to multiply an entry in row 1 by 2 and add this to the corresponding entry in row 2 then replace the old entry in row 2 with this new entry. The following are the four individual operations that we performed to do this.
2 (1) + ( -2 ) = 0 2 ( 2) +1 = 5
2 ( 3 ) + ( -1) = 5 2 (13) + 4 = 30
Okay, the next step optional, but again is convenient to do. Technically, the 5 in the second column is okay to leave. However, it will make our life easier down the road if it is a 1. We can use the second row operation to take care of this. We can divide the whole row by 5. Doing this gives,
1 2 3 13 1 1 2 3 13 0 5 5 30 5 R2 0 1 1 6 0 -6 -8 -40 0 -6 -8 -40
The next step is to then use the third row operation to make the -6 in the second column into a zero.
1 2 3 13 1 2 3 13 6 R2 + R3 0 1 1 6 0 1 1 6 0 -6 -8 -40 0 0 -2 -4
Now, officially we are done, but again its somewhat convenient to get all ones on the main diagonal so well do one last step.
1 2 3 13 1 1 2 3 13 - 2 R3 0 1 1 6 0 1 1 6 0 0 -2 -4 0 0 1 2
We can now convert back to equations.
1 2 3 13 0 1 1 6 0 0 1 2
x1 + 2 x2 + 3x3 = 13 x2 + x3 = 6 x3 = 2
At this point the solving is quite easy. We get x3 for free and once we get that we can plug this into the second equation and get x2. We can then use the first equation to get x1. Note as well that having 1s along the main diagonal helped somewhat with this process. The solution to this system of equation is
x1 = -1
x2 = 4
x3 = 2
The process used in this example is called Gaussian Elimination. Lets take a look at another example.
2007 Paul Dawkins
245
Example 2 Solve the following system of equations. x1 - 2 x2 + 3x3 = -2 - x1 + x2 - 2 x3 = 3 2 x1 - x2 + 3x3 = 1

Solution First write down the augmented matrix.
1 -2 3 -2 -1 1 -2 3 2 -1 3 1
We wont put down as many words in working this example. Heres the work for this augmented matrix.
1 -2 3 -2 R1 + R2 1 -2 3 -2 -1 1 -2 3 -2 R + R 0 -1 1 1 1 3 2 -1 3 1 0 3 -3 5 1 -2 3 -2 1 -2 3 -2 - R2 -3R2 + R3 0 1 -1 -1 0 1 -1 -1 0 3 -3 5 0 0 0 8
We wont go any farther in this example. Lets go back to equations to see why.
1 -2 3 -2 x1 - 2 x2 + 3x3 = -2 0 1 -1 -1 x2 - x3 = -1 0 0 0 8 0=8
The last equation should cause some concern. Theres one of three options here. First, weve somehow managed to prove that 0 equals 8 and we know thats not possible. Second, weve made a mistake, but after going back over our work it doesnt appear that we have made a mistake. This leaves the third option. When we get something like the third equation that simply doesnt make sense we immediately know that there is no solution. In other words, there is no set of three numbers that will make all three of the equations true at the same time. Lets work another example. We are going to get the system for this new example by making a very small change to the system from the previous example.
2007 Paul Dawkins
246
Example 3 Solve the following system of equations. x1 - 2 x2 + 3x3 = -2 - x1 + x2 - 2 x3 = 3 2 x1 - x2 + 3x3 = -7

Solution So, the only difference between this system and the system from the second example is we changed the 1 on the right side of the equal sign in the third equation to a -7. Now write down the augmented matrix for this system.
1 -2 3 -2 -1 1 -2 3 2 -1 3 -7
The steps for this problem are identical to the steps for the second problem so we wont write them all down. Upon performing the same steps we arrive at the following matrix.
1 -2 3 -2 0 1 -1 -1 0 0 0 0
This time the last equation reduces to
0=0
and unlike the second example this is not a problem. Zero does in fact equal zero! We could stop here and go back to equations to get a solution and there is a solution in this case. However, if we go one more step and get a zero above the one in the second column as well as below it our life will be a little simpler. Doing this gives,
1 -2 3 -2 1 0 1 -4 0 1 -1 -1 2 R2 + R1 0 1 -1 -1 0 0 0 0 0 0 0 0
If we now go back to equation we get the following two equations.
1 0 1 -4 0 1 -1 -1 0 0 0 0
x1 + x3 = -4 x2 - x3 = -1
We have two equations and three unknowns. This means that we can solve for two of the variables in terms of the remaining variable. Since x3 is in both equations we will solve in terms of that.
x1 = - x3 - 4 x2 = x3 - 1
What this solution means is that we can pick the value of x3 to be anything that wed like and then find values of x1 and x2. In these cases we typically write the solution as follows,
x1 = -t - 4 x2 = t - 1 x3 = t
In this way we get an infinite number of solutions, one for each and every value of t. These three examples lead us to a nice fact about systems of equations. Fact Given a system of equations, (1), we will have one of the three possibilities for the number of solutions. 1. No solution. 2. Exactly one solution. 3. Infinitely many solutions. Before moving on to the next section we need to take a look at one more situation. The system of equations in (1) is called a nonhomogeneous system if at least one of the bis is not zero. If however all of the bis are zero we call the system homogeneous and the system will be,
t = any real number
a21 x1 + a22 x2 + L + a2n xn = 0 M an1 x1 + an 2 x2 + L + ann xn = 0

Now, notice that in the homogeneous case we are guaranteed to have the following solution. This solution is often called the trivial solution.
a11 x1 + a12 x2 + L + a1n xn = 0
(2)
x1 = x2 = L = xn = 0
For homogeneous systems the fact above can be modified to the following. Fact Given a homogeneous system of equations, (2), we will have one of the two possibilities for the number of solutions. 1. Exactly one solution, the trivial solution 2. Infinitely many non-zero solutions in addition to the trivial solution. In the second possibility we can say non-zero solution because if there are going to be infinitely many solutions and we know that one of them is the trivial solution then all the rest must have at least one of the xis be non-zero and hence we get a non-zero solution.
2007 Paul Dawkins
248
Review : Matrices and Vectors

This section is intended to be a catch all for many of the basic concepts that are used occasionally in working with systems of differential equations. There will not be a lot of details in this section, nor will we be working large numbers of examples. Also, in many cases we will not be looking at the general case since we wont need the general cases in our differential equations work. Lets start with some of the basic notation for matrices. An n x m (this is often called the size or dimension of the matrix) matrix is a matrix with n rows and m columns and the entry in the ith row and jth column is denoted by aij. A short hand method of writing a general n x m matrix is the following.
a11 a A = 21 M an1
a12 L a1m a22 L a2 m = ( aij )n x m M M an 2 L anm n x m
The size or dimension of a matrix is subscripted as shown if required. If its not required or clear from the problem the subscripted size is often dropped from the matrix. Special Matrices There are a few special matrices out there that we may use on occasion. The first special matrix is the square matrix. A square matrix is any matrix whose size (or dimension) is n x n. In other words it has the same number of rows as columns. In a square matrix the diagonal that starts in the upper left and ends in the lower right is often called the main diagonal. The next two special matrices that we want to look at are the zero matrix and the identity matrix. The zero matrix, denoted 0n x m , is a matrix all of whose entries are zeroes. The identity matrix is a square n x n matrix, denoted In, whose main diagonals are all 1s and all the other elements are zero. Here are the general zero and identity matrices.
0n x m
0 0 L 0 = M M M 0 0 L 0 n x m
1 0 In = M 0
0 1 M 0
L L O L
0 0 M 1 n x n
In matrix arithmetic these two matrices will act in matrix work like zero and one act in the real number system. The last two special matrices that well look at here are the column matrix and the row matrix. These are matrices that consist of a single column or a single row. In general they are,
x1 x 2 x= M xn n x 1
We will often refer to these as vectors.
y = ( y1
y2 L ym )1 x m
Arithmetic We next need to take a look at arithmetic involving matrices. Well start with addition and subtraction of two matrices. So, suppose that we have two n x m matrices, A and B. The sum (or difference) of these two matrices is then,
An x m Bn x m = ( aij )
nxm
( bij )
nxm
= ( aij bij )
nxm
The sum or difference of two matrices of the same size is a new matrix of identical size whose entries are the sum or difference of the corresponding entries from the original two matrices. Note that we cant add or subtract entries with different sizes. Next, lets look at scalar multiplication. In scalar multiplication we are going to multiply a matrix A by a constant (sometimes called a scalar) a. In this case we get a new matrix whose entries have all been multiplied by the constant, a.
a An x m = a ( aij )
n xm
= (a aij )
n xm
Example 1 Given the following two matrices, 3 -2 A= -9 1

compute A-5B. Solution There isnt much to do here other than the work.
-4 1 B = 0 -5
3 -2 -4 1 A - 5B = - 5 -9 1 0 -5 3 -2 -20 5 = - -9 1 0 -25 23 -7 = -9 26
We first multiplied all the entries of B by 5 then subtracted corresponding entries to get the entries in the new matrix. The final matrix operation that well take a look at is matrix multiplication. Here we will start with two matrices, An x p and Bp x m . Note that A must have the same number of columns as B has rows. If this isnt true then we cant perform the multiplication. If it is true then we can perform the following multiplication.
An x p Bp x m = ( cij )
n xm
The new matrix will have size n x m and the entry in the ith row and jth column, cij, is found by multiplying row i of matrix A by column j of matrix B. This doesnt always make sense in words so lets look at an example.
2007 Paul Dawkins
250
Example 2 Given 2 -1 0 A= -3 6 1 2 x 3
compute AB. Solution The new matrix will have size 2 x 4. The entry in row 1 and column 1 of the new matrix will be found by multiplying row 1 of A by column 1 of B. This means that we multiply corresponding entries from the row of A and the column of B and then add the results up. Here are a couple of the entries computed all the way out.
1 0 -1 2 B = -4 3 1 0 0 3 0 -2 3 x 4
c11 = ( 2 )(1) + ( -1)( -4 ) + ( 0 )( 0 ) = 6
c13 = ( 2 )( -1) + ( -1)(1) + ( 0 )( 0 ) = -3 c24 = ( -3)( 2 ) + ( 6 )( 0 ) + (1)( -2 ) = -8

Heres the complete solution.
6 -3 -3 4 C = -27 21 9 -8
In this last example notice that we could not have done the product BA since the number of columns of B does not match the number of row of A. It is important to note that just because we can compute AB doesnt mean that we can compute BA. Likewise, even if we can compute both AB and BA they may or may not be the same matrix. Determinant The next topic that we need to take a look at is the determinant of a matrix. The determinant is actually a function that takes a square matrix and converts it into a number. The actual formula for the function is somewhat complex and definitely beyond the scope of this review. The main method for computing determinants of any square matrix is called the method of cofactors. Since we are going to be dealing almost exclusively with 2 x 2 matrices and the occasional 3 x 3 matrix we wont go into the method here. We can give simple formulas for each of these cases. The standard notation for the determinant of the matrix A is.
det ( A ) = A
Here are the formulas for the determinant of 2 x 2 and 3 x 3 matrices.
a c = ad - cb b d a11 a21 a31 a12 a22 a32 a13 a a23 = a11 22 a32 a33 a23 a a a - a12 21 23 + a13 21 a33 a31 a33 a31 a22 a32
2007 Paul Dawkins
251
Example 3 Find the determinant of each of the following matrices. 2 3 1 -9 -18 A= B= -1 -6 7 4 2 4 5 -1

Solution For the 2 x 2 there isnt much to do other than to plug it into the formula.
det ( A ) =
-9 -18 = ( -9 )( 4 ) - ( -18 )( 2 ) = 0 2 4
For the 3 x 3 we could plug it into the formula, however unlike the 2 x 2 case this is not an easy formula to remember. There is an easier way to get the same result. A quicker way of getting the same result is to do the following. First write down the matrix and tack a copy of the first two columns onto the end as follows.
2 3 1 2 3 det ( B ) = -1 -6 7 -1 -6 4 5 -1 4 5
Now, notice that there are three diagonals that run from left to right and three diagonals that run from right to left. What we do is multiply the entries on each diagonal up and the if the diagonal runs from left to right we add them up and if the diagonal runs from right to left we subtract them. Here is the work for this matrix.
2 3 1 2 3 det ( B ) = -1 -6 7 -1 -6 4 5 -1 4 5 = ( 2 )( -6 )( -1) + ( 3)( 7 )( 4 ) + (1)( -1)( 5) -
( 3)( -1)( -1) - ( 2 )( 7 )( 5 ) - (1)( -6 )( 4 )

= 42
You can either use the formula or the short cut to get the determinant of a 3 x 3. If the determinant of a matrix is zero we call that matrix singular and if the determinant of a matrix isnt zero we call the matrix nonsingular. The 2 x 2 matrix in the above example was singular while the 3 x 3 matrix is nonsingular. Matrix Inverse Next we need to take a look at the inverse of a matrix. Given a square matrix, A, of size n x n if we can find another matrix of the same size, B such that, then we call B the inverse of A and denote it by B=A-1.
AB = BA = I n
Computing the inverse of a matrix, A, is fairly simple. First we form a new matrix,
(A
In )
2007 Paul Dawkins
252
and then use the row operations from the previous section and try to convert this matrix into the form,
( In
B)
If we can then B is the inverse of A. If we cant then there is no inverse of the matrix A.
Example 4 Find the inverse of the following matrix, if it exists. 2 1 1 A= -5 -3 0 1 1 -1

Solution We first form the new matrix by tacking on the 3 x 3 identity matrix to this matrix. This is
2 1 1 1 0 0 -5 -3 0 0 1 0 1 1 -1 0 0 1
We will now use row operations to try and convert the first three columns to the 3 x 3 identity. In other words we want a 1 on the diagonal that starts at the upper left corner and zeroes in all the other entries in the first three columns. If you think about it, this process is very similar to the process we used in the last section to solve systems, it just goes a little farther. Here is the work for this problem.
2 1 1 1 0 0 1 1 -1 0 0 1 R2 + 5 R1 -5 -3 0 0 1 0 R1 R3 -5 -3 0 0 1 0 R - 2 R 1 3 1 1 -1 0 0 1 2 1 1 1 0 0 1 1 -1 0 0 1 1 1 1 -1 0 0 1 0 2 -5 0 1 5 2 R2 0 1 -5 0 1 5 R3 + R2 2 2 2 0 -1 3 1 0 -2 0 -1 3 1 0 -2 1 1 -1 0 0 1 1 1 2 R3 -5 5 1 0 1 2 0 2 2 0 1 0 0 0 0 1 1 1 1 2 2 2 1 1 0 2 1 2 1 0 1 0 5 3 5 R1 - R2 0 0 0 1 2 1 1 0 -1 0 0 1 R2 + 5 R 2 3 -5 5 1 0 2 2 R1 + R3 2 1 2 1 1 0 0 -3 -2 -3 1 0 5 3 5 0 1 2 1 1
So, we were able to convert the first three columns into the 3 x 3 identity matrix therefore the inverse exists and it is,
-3 -2 -3 A = 5 3 5 2 1 1
-1
2007 Paul Dawkins
253
So, there was an example in which the inverse did exist. Lets take a look at an example in which the inverse doesnt exist.
Example 5 Find the inverse of the following matrix, provided it exists. 1 -3 B= -2 6

Solution In this case we will tack on the 2 x 2 identity to get the new matrix and then try to convert the first two columns to the 2 x 2 identity matrix.
1 -3 1 0 2 R1 + R2 1 -3 1 0 -2 6 0 1 0 0 2 1
And we dont need to go any farther. In order for the 2 x 2 identity to be in the first two columns we must have a 1 in the second entry of the second column and a 0 in the second entry of the first column. However, there is no way to get a 1 in the second entry of the second column that will keep a 0 in the second entry in the first column. Therefore, we cant get the 2 x 2 identity in the first two columns and hence the inverse of B doesnt exist. We will leave off this discussion of inverses with the following fact. Fact Given a matrix A. 1. If A is nonsingular then A-1 will exist. 2. If A is singular then A-1 will NOT exist. Ill leave it to you to verify this fact for the previous two examples. Systems of Equations Revisited We need to do a quick revisit of systems of equations. Lets start with a general system of equations.
a21 x1 + a22 x2 + L + a2 n xn = b2 M an1 x1 + an 2 x2 + L + ann xn = bn

Now, covert each side into a vector to get,
a11 x1 + a12 x2 + L + a1n xn = b1
(1)
a11 x1 + a12 x2 + L + a1n xn b1 a x + a x +L + a x b 2n n 21 1 22 2 = 2 M M an1 x1 + an 2 x2 + L + ann xn bn
The left side of this equation can be thought of as a matrix multiplication.
2007 Paul Dawkins
254
a11 a 21 M an1
a12 a22 M an 2
L a1n x1 b1 L a2 n x2 = b2 O M M M L ann xn bn r r Ax = b (2)
Simplifying up the notation a little gives,
where, x is a vector whose components are the unknowns in the original system of equations. We call (2) the matrix form of the system of equations (1) and solving (2) is equivalent to solving (1). The solving process is identical. The augmented matrix for (2) is
(A b)
Once we have the augmented matrix we proceed as we did with a system that hasnt been wrote in matrix form. We also have the following fact about solutions to (2). Fact Given the system of equation (2) we have one of the following three possibilities for solutions. 1. There will be no solutions. 2. There will be exactly one solution. 3. There will be infinitely many solutions. In fact we can go a little farther now. Since we are assuming that weve got the same number of equations as unknowns the matrix A in (2) is a square matrix and so we can compute its determinant. This gives the following fact. Fact Given the system of equations in (2) we have the following. 1. If A is nonsingular then there will be exactly one solution to the system. 2. If A is singular then there will either be no solution or infinitely many solutions to the system. The matrix form of a homogeneous system is
r where 0 is the vector of all zeroes. In the homogeneous system we are guaranteed to have a r r solution, x = 0 . The fact above for homogeneous systems is then,
Fact Given the homogeneous system (3) we have the following. r r 1. If A is nonsingular then the only solution will be x = 0 . 2. If A is singular then there will be infinitely many nonzero solutions to the system.
r r Ax = 0
(3)
Linear Independence/Linear Dependence This is not the first time that weve seen this topic. We also saw linear independence and linear dependence back when we were looking at second order differential equations. In that section we
were dealing with functions, but the concept is essentially the same here. If we start with n vectors, If we can find constants, c1,c2,,cn with at least two nonzero such that then we call the vectors linearly dependent. If the only constants that work in (4) are c1=0, c2=0, , cn=0 then we call the vectors linearly independent. If we further make the assumption that each of the n vectors has n components, i.e. each of the vectors look like,
r r r x1 , x2 , K , xn
r r r r c1 x1 + c2 x2 + K + cn xn = 0
(4)
x1 x r x = 2 M xn
we can get a very simple test for linear independence and linear dependence. Note that this does not have to be the case, but in all of our work we will be working with n vectors each of which has n components. Fact Given the n vectors each with n components, form the matrix,
r r r x1 , x2 , K , xn r r x2 L xn ) r
r X = ( x1
So, the matrix X is a matrix whose ith column is the ith vector, xi . Then, 1. If X is nonsingular (i.e. det(X) is not zero) then the n vectors are linearly independent, and 2. if X is singular (i.e. det(X) = 0) then the n vectors are linearly dependent and the constants that make (4) true can be found by solving the system where c is a vector containing the constants in (4).
r r Xc =0
Example 6 Determine if the following set of vectors are linearly independent or linearly
dependent. If they are linearly dependent find the relationship between them.
1 -2 6 r (1) r (2) r (3) x = -3 , x = 1 , x = -2 5 4 1

Solution So, the first thing to do is to form X and compute its determinant.
1 -2 6 -3 1 -2 X = 5 4 1
det ( X ) = -79
This matrix is non singular and so the vectors are linearly independent.
2007 Paul Dawkins
256
Example 7 Determine if the following set of vectors are linearly independent or linearly
dependent. If they are linearly dependent find the relationship between them.
1 -4 2 r (1) r (2) r (3) x = -1 , x = 1 , x = -1 3 -6 4

Solution As with the last example first form X and compute its determinant.
1 -4 2 X = -1 1 -1 3 -6 4
det ( X ) = 0
So, these vectors are linearly dependent. We now need to find the relationship between the vectors. This means that we need to find constants that will make (4) true. So we need to solve the system
r r Xc =0 2 0 1 -4 2 R3 + 2 R2 1 0 0 -3 1 -2 0 0 0 0 0 c1 + 2 c =0 3 3 1 0 c2 - 3 c3 = 0 0 0=0 0 1 - R2 3 0 0 c1 = - 2 c 3 3 1 c2 = 3 c3
Here is the augmented matrix and the solution work for this system.
1 -4 2 0 R2 + R1 1 -4 -1 1 -1 0 R - 3R 0 -3 1 3 3 -6 4 0 0 6 1 0 2 1 -4 2 0 3 R1 + 4 R2 1 1 0 1 -3 0 0 1 -3 0 0 0 0 0 0 0
Now, we would like actual values for the constants so, if use c3 = 3 we get the following solution c1 = -2 , c2 = 1 , and c3 = 3 . The relationship is then.
0 r r r -2 x (1) + x (2) + 3x (3) = 0 0

Calculus with Matrices There really isnt a whole lot to this other than to just make sure that we can deal with calculus with matrices. First, to this point weve only looked at matrices with numbers as entries, but the entries in a matrix can be functions as well. So we can look at matrices in the following form,
a11 ( t ) a12 ( t ) L a1n ( t ) a21 ( t ) a22 ( t ) L a2 n ( t ) A (t ) = M M M a ( t ) a (t ) L a (t ) m2 mn m1
2007 Paul Dawkins
257
Now we can talk about differentiating and integrating a matrix of this form. To differentiate or integrate a matrix of this form all we do is differentiate or integrate the individual entries.
( t ) a12 ( t ) L a1 n ( t ) a11 ( t ) a22 ( t ) L a2 n (t ) a21 A ( t ) = M M M a ( t ) a ( t ) L a ( t ) m2 mn m1
a11 ( t ) dt a ( t ) dt 21 A t dt = ( ) M am1 ( t ) dt
am 2 ( t ) dt L amn ( t ) dt
a ( t ) dt a ( t ) dt
12 22
L L
a ( t ) dt a ( t ) dt
1n 2n
So when we run across this kind of thing dont get excited about it. Just differentiate or integrate as we normally would. In this section we saw a very condensed set of topics from linear algebra. When we get back to differential equations many of these topics will show up occasionally and you will at least need to know what the words mean. The main topic from linear algebra that you must know however if you are going to be able to solve systems of differential equations is the topic of the next section.
2007 Paul Dawkins
258
Review : Eigenvalues and Eigenvectors

If you get nothing out of this quick review of linear algebra you must get this section. Without this section you will not be able to do any of the differential equations work that is in this chapter. So lets start with the following. If we multiply an n x n matrix by an n x 1 vector we will get a new n x 1 vector back. In other words,
r r Ah = y
What we want to know is if it is possible for the following to happen. Instead of just getting a brand new vector out of the multiplication is it possible instead to get the following,
r r Ah = lh
(1) r
In other words is it possible, at least for certain l andh , to have matrix multiplication be the same as must multiplying the vector by a constant? Of course, we probably wouldnt be talking about this if the answer was no. So, it is possible for this to happen, however, it wont happen for r r just any value of l orh . If we do happen to have a l andh for which this works (and they will r always come in pairs) then we call l an eigenvalue of A and h an eigenvector of A. So, how do we go about find the eigenvalues and eigenvectors for a matrix? Well first notice that that if h = 0 then (1) is going to be true for any value of l and so we are going to make the assumption that h 0 . With that out of the way lets rewrite (1) a little.
r r r Ah - lh = 0 r r r Ah - l I nh = 0 r r ( A - l I n )h = 0
Notice that before we factored out theh we added in the appropriately sized identity matrix. This is equivalent to multiplying things by a one and so doesnt change the value of anything. We needed to do this because without it we would have had the difference of a matrix, A, and a constant, l, and this cant be done. We now have the difference of two matrices of the same size which can be done. So, with this rewrite we see that
( A - l I n )h = 0
r r
(2)
is equivalent to (1). In order to find the eigenvectors for a matrix we will need to solve a homogeneous system. Recall the fact from the previous section that we know that we will either r r have exactly one solution ( h = 0 ) or we will have infinitely many nonzero solutions. Since weve already said that dont want h = 0 this means that we want the second case. Knowing this will allow us to find the eigenvalues for a matrix. Recall from this fact that we will get the second case only if the matrix in the system is singular. Therefore we will need to determine the values of l for which we get,
det ( A - l I ) = 0
259
2007 Paul Dawkins
Once we have the eigenvalues we can then go back and determine the eigenvectors for each eigenvalue. Lets take a look at a couple of quick facts about eigenvalues and eigenvectors. Fact If A is an n x n matrix then det ( A - l I ) = 0 is an nth degree polynomial. This polynomial is called the characteristic polynomial. To find eigenvalues of a matrix all we need to do is solve a polynomial. Thats generally not too bad provided we keep n small. Likewise this fact also tells us that for an n x n matrix, A, we will have n eigenvalues if we include all repeated eigenvalues. Fact If l1 , l 2 ,K , l n is the complete list of eigenvalues for A (including all repeated eigenvalues) then, 1. If l occurs only once in the list then we call l simple. 2. If l occurs k>1 times in the list then we say that l has multiplicity k. 3. If l1 , l 2 ,K , l n ( k n ) are the simple eigenvalues in the list with corresponding
eigenvectors h ( ) , h ( ) , , h ( ) then the eigenvectors are all linearly independent. 4. If l is an eigenvalue of k > 1 then l will have anywhere from 1 to k linearly independent eigenvectors.
r1
r2
rk
The usefulness of these facts will become apparent when we get back into differential equations since in that work we will want linearly independent solutions. Lets work a couple of examples now to see how we actually go about finding eigenvalues and eigenvectors.
Example 1 Find the eigenvalues and eigenvectors of the following matrix. 2 7 A= -1 -6

Solution The first thing that we need to do is find the eigenvalues. That means we need the following matrix,
2 7 1 0 2 - l A - lI = -l = -1 -6 0 1 -1
7 -6 - l
In particular we need to determine where the determinant of this matrix is zero.
det ( A - l I ) = ( 2 - l )( -6 - l ) + 7 = l 2 + 4l - 5 = ( l + 5)( l - 1)
So, it looks like we will have two simple eigenvalues for this matrix, l1 = -5 and l 2 = 1 . We will now need to find the eigenvectors for each of these. Also note that according to the fact above, the two eigenvectors should be linearly independent. To find the eigenvectors we simply plug in each eigenvalue into (2) and solve. So, lets do that.
l1 = -5 :
In this case we need to solve the following system.
7 7 r 0 -1 -1h = 0
Recall that officially to solve this system we use the following augmented matrix.
R + R2 7 7 0 7 7 0 1 7 1 -1 -1 0 0 0 0
Upon reducing down we see that we get a single equation
7h1 + 7h 2 = 0
h1 = -h 2
that will yield an infinite number of solutions. This is expected behavior. Recall that we picked the eigenvalues so that the matrix would be singular and so we would get infinitely many solutions. Notice as well that we could have identified this from the original system. This wont always be the case, but in the 2 x 2 case we can see from the system that one row will be a multiple of the other and so we will get infinite solutions. From this point on we wont be actually solving systems in these cases. We will just go straight to the equation and we can use either of the two rows for this equation. Now, lets get back to the eigenvector, since that is what we were after. In general then the eigenvector will be any vector that satisfies the following,
r h -h h = 1= 2 h 2 h 2
,h 2 0
To get this we used the solution to the equation that we found above. We really dont want a general eigenvector however so we will pick a value for h 2 to get a specific eigenvector. We can choose anything (except h 2 = 0 ), so pick something that will make the eigenvector nice. Note as well that since weve already assumed that the eigenvector is not zero we must choose a value that will not give us zero, which is why we want to avoid except h 2 = 0 in this case. Heres the eigenvector for this eigenvalue.
-1 r h (1) = , 1
using h 2 = 1
Now we get to do this all over again for the second eigenvalue.
l2 = 1 :
Well do much less work with this part than we did with the previous part. We will need to solve the following system.
1 7 r 0 -1 -7 h = 0
261
2007 Paul Dawkins
Clearly both rows are multiples of each other and so we will get infinitely many solutions. We can choose to work with either row. Well run with the first because to avoid having too many minus signs floating around. Doing this gives us,
h1 + 7h2 = 0
h1 = -7h 2
Note that we can solve this for either of the two variables. However, with an eye towards working with these later on lets try to avoid as many fractions as possible. The eigenvector is then,
r h -7h2 h = 1= ,h 2 0 h 2 h2 -7 r h (2) = , using h 2 = 1 1

Summarizing we have,
l1 = -5 l2 = 1
-1 r h (1) = 1 r -7 h (1) = 1
Note that the two eigenvectors are linearly independent as predicted.
Example 2 Find the eigenvalues and eigenvectors of the following matrix. 1 -1 A=4 1 9 -3
Solution This matrix has fractions in it. Thats life so dont get excited about it. First we need the eigenvalues.
det ( A - l I ) =
1- l
4 9
-1 - -l
1 3
1 4 = (1 - l ) - - l + 3 9 2 1 = l2 - l + 3 9 1 = l - 3
2
l1,2 =
1 3
So, it looks like weve got an eigenvalue of multiplicity 2 here. Remember that the power on the term will be the multiplicity. Now, lets find the eigenvector(s). This one is going to be a little different from the first example. There is only one eigenvalue so lets do the work for that one. We will need to solve the following system,
2 3 4 9
-1 h1 0 = -2 0 3 h 2
R1 =
3 R2 2
So, the rows are multiples of each other. Well work with the first equation in this example to find the eigenvector.
2 h1 - h 2 = 0 3
2 h2 = h1 3
Recall in the last example we decided that we wanted to make these as nice as possible and so should avoid fractions if we can. Sometimes, as in this case, we simply cant so well have to deal with it. In this case the eigenvector will be,
r h h1 h = 1 = 2 , h 2 3 h1 3 r h (1) = , 2
h1 0 h1 = 3
Note that by careful choice of the variable in this case we were able to get rid of the fraction that we had. This is something that in general doesnt much matter if we do or not. However, when we get back to differential equations it will be easier on us if we dont have any fractions so we will usually try to eliminate them at this step. Also in this case we are only going to get a single (linearly independent) eigenvector. We can get other eigenvectors, by choosing different values of h 1 . However, each of these will be linearly dependent with the first eigenvector. If youre not convinced of this try it. Pick some values for h 1 and get a different vector and check to see if the two are linearly dependent. Recall from the fact above that an eigenvalue of multiplicity k will have anywhere from 1 to k linearly independent eigenvectors. In this case we got one. For most of the 2 x 2 matrices that well be working with this will be the case, although it doesnt have to be. We can, on occasion, get two.
Example 3 Find the eigenvalues and eigenvectors of the following matrix. -4 -17 A= 2 2
Solution So, well start with the eigenvalues.
det ( A - l I ) =
-4 - l 2
-17 2-l
= ( -4 - l )( 2 - l ) + 34 = l 2 + 2l + 26
This doesnt factor, so upon using the quadratic formula we arrive at,
l1,2 = -1 5i
In this case we get complex eigenvalues which are definitely a fact of life with eigenvalue/eigenvector problems so get used to them. Finding eigenvectors for complex eigenvalues is identical to the previous two examples, but it will be somewhat messier. So, lets do that.
l1 = -1 + 5 i :
The system that we need to solve this time is
-4 - ( -1 + 5i ) -17 h1 0 = 2 2 - ( -1 + 5i ) h2 0 -3 - 5i -17 h1 0 = 2 3 - 5i h 2 0
Now, its not super clear that the rows are multiples of each other, but they are. In this case we have,
R1 = -
1 ( 3 + 5i ) R2 2
This is not something that you need to worry about, we just wanted to make the point. For the work that well be doing later on with differential equations we will just assume that weve done everything correctly and weve got two rows are multiples of each other. Therefore, all that we need to do here is pick one of the rows and work with it. Well work with the second row this time.
2h1 + ( 3 - 5i )h2 = 0
Now we can solve for either of the two variables. However, again looking forward to differential equations, we are going to need the i in the numerator so solve the equation in such a way as this will happen. Doing this gives,
2h1 = - ( 3 - 5i )h2
h1 = So, the eigenvector in this case is
1 ( 3 - 5i )h2 2
1 r h1 - ( 3 - 5i )h 2 h = = 2 , h2 h2 -3 + 5i r h (1) = , 2
h2 0 h2 = 2
As with the previous example we choose the value of the variable to clear out the fraction. Now, the work for the second eigenvector is almost identical and so well not dwell on that too much.
l 2 = -1 - 5 i :
The system that we need to solve here is
-4 - ( -1 - 5i ) -17 h1 0 = 2 2 - ( -1 - 5i ) h 2 0 -3 + 5i -17 h1 0 = 2 3 + 5i h 2 0 1 ( 3 + 5i )h2 2
Working with the second row again gives,
2h1 + ( 3 + 5i )h 2 = 0
The eigenvector in this case is
h1 = -
1 r h1 - ( 3 + 5i )h 2 h = = 2 , h2 h2 -3 - 5i r h (2) = , 2
Summarizing,
h2 0 h2 = 2
l1 = -1 + 5i l 2 = -1 - 5i
-3 + 5i r h (1) = 2 r -3 - 5i h ( 2) = 2
There is a nice fact that we can use to simplify the work when we get complex eigenvalues. We need a bit of terminology first however. If we start with a complex number, then the complex conjugate of z is
z = a + bi z = a - bi
To compute the complex conjugate of a complex number we simply change the sign on the term that contains the i. The complex conjugate of a vector is just the conjugate of each of the vectors components. We now have the following fact about complex eigenvalues and eigenvectors. Fact If A is an n x n matrix with only real numbers and if l1 = a + bi is an eigenvalue with eigenvector h ( ) . Then l 2 = l 1 = a - bi is also an eigenvalue and its eigenvector is the conjugate of h ( ) .
r1
r1
This is a fact is something that you should feel free to use as you need to in our work. Now, we need to work one final eigenvalue/eigenvector problem. To this point weve only worked with 2 x 2 matrices and we should work at least one that isnt 2 x 2. Also, we need to work one in which we get an eigenvalue of multiplicity greater than one that has more than one linearly independent eigenvector.
Example 4 Find the eigenvalues and eigenvectors of the following matrix. 0 1 1 A= 1 0 1 1 1 0

Solution Despite the fact that this is a 3 x 3 matrix, it still works the same as the 2 x 2 matrices that weve been working with. So, start with the eigenvalues
-l det ( A - l I ) = 1 1
1 -l 1
1 1 -l
2
= -l 3 + 3l + 2 = ( l - 2 )( l + 1)
l1 = 2, l2,3 = -1
So, weve got a simple eigenvalue and an eigenvalue of multiplicity 2. Note that we used the same method of computing the determinant of a 3 x 3 matrix that we used in the previous section. We just didnt show the work. Lets now get the eigenvectors. Well start with the simple eigenvector.
l1 = 2 :
Here well need to solve,
1 h1 0 -2 1 1 -2 1 h = 0 2 1 1 -2 h 0 3
This time, unlike the 2 x 2 cases we worked earlier, we actually need to solve the system. So lets do that.
-2 1 1 0 1 -2 1 0 R2 + 2 R1 1 -2 1 0 1 -2 1 0 R1 R2 -2 1 1 0 R - R 0 -3 3 0 1 3 1 1 -2 0 1 1 -2 0 0 3 -3 0 1 -2 1 0 R3 - 3R2 1 0 -1 0 -1 3 R2 0 1 -1 0 R1 + 2 R2 0 1 -1 0 0 0 0 0 0 3 3 0
Going back to equations gives,
h1 - h3 = 0 h 2 - h3 = 0
h1 = h3 h 2 = h3
So, again we get infinitely many solutions as we should for eigenvectors. The eigenvector is then,
h1 h3 r h = h2 = h3 , h h 3 3 1 r (1) h = 1 , 1
Now, lets do the other eigenvalue.
h3 0
h3 = 1
l 2 = -1 :
Here well need to solve,
1 1 1 h1 0 1 1 1h = 0 2 1 1 1h 0 3
Okay, in this case is clear that all three rows are the same and so there isnt any reason to actually solve the system since we can clear out the bottom two rows to all zeroes in one step. The equation that we get then is,
h1 + h 2 + h3 = 0
h1 = -h2 - h3
So, in this case we get to pick two of the values for free and will still get infinitely many solutions. Here is the general eigenvector for this case,
h1 -h2 - h3 r h = h 2 = h 2 , h h 3 3
h 2 0 and h3 0 at the same time
Notice the restriction this time. Recall that we only require that the eigenvector not be the zero vector. This means that we can allow one or the other of the two variables to be zero, we just cant allow both of them to be zero at the same time! What this means for us is that we are going to get two linearly independent eigenvectors this time. Here they are.
2007 Paul Dawkins
267
-1 r (2) h = 0 1 -1 r (3) h = 1 0
h 2 = 0 and h3 = 1
h 2 = 1 and h3 = 0
Now when we talked about linear independent vectors in the last section we only looked at n vectors each with n components. We can still talk about linear independence in this case however. Recall back with we did linear independence for functions we saw at the time that if two functions were linearly dependent then they were multiples of each other. Well the same thing holds true for vectors. Two vectors will be linearly dependent if they are multiples of each
r2 r3 other. In this case there is no way to get h ( ) by multiplying h ( ) by a constant. Therefore, these two vectors must be linearly independent.
So, summarizing up, here are the eigenvalues and eigenvectors for this matrix
l1 = 2
l2 = -1
1 r (1) h = 1 1 -1 r ( 2) h = 0 1 -1 r ( 3) h = 1 0
l3 = -1
2007 Paul Dawkins
268

In the introduction to this section we briefly discussed how a system of differential equations can arise from a population problem in which we keep track of the population of both the prey and the predator. It makes sense that the number of prey present will affect the number of the predator present. Likewise, the number of predator present will affect the number of prey present. Therefore the differential equation that governs the population of either the prey or the predator should in some way depend on the population of the other. This will lead to two differential equations that must be solved simultaneously in order to determine the population of the prey and the predator. The whole point of this is to notice that systems of differential equations can arise quite easily from naturally occurring situations. Developing an effective predator-prey system of differential equations is not the subject of this chapter. However, systems can arise from nth order linear differential equations as well. Before we get into this however, lets write down a system and get some terminology out of the way. We are going to be looking at first order, linear systems of differential equations. These terms mean the same thing that they have meant up to this point. The largest derivative anywhere in the system will be a first derivative and all unknown functions and their derivatives will only occur to the first power and will not be multiplied by other unknown functions. Here is an example of a system of first order, linear differential equations.
= x1 + 2 x2 x1 = 3 x1 + 2 x2 x2
We call this kind of system a coupled system since knowledge of x2 is required in order to find x1 and likewise knowledge of x1 is required to find x2. We will worry about how to go about solving these later. At this point we are only interested in becoming familiar with some of the basics of systems. Now, as mentioned earlier, we can write an nth order linear differential equation as a system. Lets see how that can be done.
Example 1 Write the following 2nd order differential equations as a system of first order, linear
differential equations.
2 y - 5 y + y = 0
y ( 3) = 6
y ( 3) = -1
Solution We can write higher order differential equations as a system with a very simple change of variable. Well start by defining the following two new functions.
x1 ( t ) = y ( t )
x2 ( t ) = y ( t )
Now notice that if we differentiate both sides of these we get,
= y = x2 x1
= y = x2
2007 Paul Dawkins
1 5 1 5 y + y = - x1 + x2 2 2 2 2
Note the use of the differential equation in the second equation. We can also convert the initial conditions over to the new functions.
x1 ( 3) = y ( 3) = 6
x2 ( 3) = y ( 3) = -1
Putting all of this together gives the following system of differential equations.
= x2 x1 1 5 = - x1 + x2 x2 2 2
x1 ( 3) = 6
x2 ( 3) = -1
We will call the system in the above example an Initial Value Problem just as we did for differential equations with initial conditions. Lets take a look at another example.
Example 2 Write the following 4th order differential equations as a system of first order, linear
differential equations.
y ( 4) + 3 y - sin ( t ) y + 8 y = t 2
y ( 0 ) = 1 y ( 0 ) = 2 y ( 0 ) = 3 y ( 0 ) = 4
Solution Just as we did in the last example well need to define some new functions. This time well need 4 new functions.
x1 = y
x2 = y x3 = y x4 = y
= y = x2 x1 = y = x3 x2
= y = x4 x3 = y ( 4) = -8 y + sin ( t ) y - 3 y + t 2 = -8 x1 + sin ( t ) x2 - 3 x3 + t 2 x4 = x2 x1 x1 ( 0 ) = 1 x2 ( 0 ) = 2 x3 ( 0 ) = 3 x4 ( 0 ) = 4
The system along with the initial conditions is then,
= x3 x2 = x4 x3 = -8 x1 + sin ( t ) x2 - 3 x3 + t 2 x4
Now, when we finally get around to solving these we will see that we generally dont solve systems in the form that weve given them in this section. Systems of differential equations can be converted to matrix form and this is the form that we usually use in solving systems.
2007 Paul Dawkins
270
Example 3 Convert the system the following system to matrix from. = 4 x1 + 7 x2 x1 = -2 x1 - 5 x2 x2

Solution First write the system so that each side is a vector.
4 x1 + 7 x2 x1 x = 2 x 5 x 2 - 1 - 2
Now the right side can be written as a matrix multiplication,
4 7 x1 x1 x = 2 -2 -5 x2
Now, if we define,
r x x = 1 x2
then,
r x x = 1 x2
The system can then be wrote in the matrix form,
r 4 7 r x = x -2 -5
Example 4 Convert the systems from Examples 1 and 2 into matrix form.
Solution Well start with the system from Example 1.
= x2 x1 1 5 = - x1 + x2 x2 2 2
First define,
x1 ( 3) = 6 x2 ( 3) = -1
r x x = 1 x2
The system is then,
0 r x = 1 2
1 r x 5 2
x ( 3) 6 r x ( 3) = 1 = x2 ( 3) -1
Now, lets do the system from Example 2.

= x2 x1 = x3 x2 = x4 x3 = -8 x1 + sin ( t ) x2 - 3 x3 + t 2 x4
x1 ( 0 ) = 1 x2 ( 0 ) = 2 x3 ( 0 ) = 3 x4 ( 0 ) = 4
In this case we need to be careful with the t2 in the last equation. Well start by writing the system as a vector again and then break it up into two vectors, one vector that contains the unknown functions and the other that contains any known functions.
x2 x2 0 x1 x x3 x3 = + 0 2= 0 x3 x4 x4 2 -8 x1 + sin ( t ) x2 - 3 x3 + t 2 x4 -8 x1 + sin ( t ) x2 - 3x3 t

Now, the first vector can now be written as a matrix multiplication and well leave the second vector alone.
1 0 0 0 0 1 r x = 0 0 0 -8 sin ( t ) -3
where,
0 0 0 r 0 x + 0 1 2 0 t
1 2 r x ( 0) = 3 4
x1 ( t ) x2 ( t ) r x (t ) = x3 ( t ) x (t ) 4
Note that occasionally for large systems such as this we will one step farther and write the system as,
r r r x = Ax + g ( t )
r r r x = Ax + g ( t ) r r we say that the system is homogeneous if g ( t ) = 0 and we say the system is nonhomogeneous r r if g ( t ) 0 .
The last thing that we need to do in this section is get a bit of terminology out of the way. Starting with
2007 Paul Dawkins
272
Solutions to Systems
Now that weve got some of the basic out of the way for systems of differential equations its time to start thinking about how to solve a system of differential equations. We will start with the homogeneous system written in matrix form,
(1) r where, A is an n x n matrix and x is a vector whose components are the unknown functions in the
system. Now, if we start with n = 1 then the system reduces to a fairly simple linear (or separable) first order differential equation.
r r x = A x
x = ax
and this has the following solution,
x ( t ) = ce a t r r x ( t ) = h e rt
So, lets use this as a guide and for a general n lets see if
(2)
will be a solution. Note that the only real difference here is that we let the constant in front of the exponential be a vector. All we need to do then is plug this into the differential equation and see what we get. First notice that the derivative is,
r r x ( t ) = rh e rt
So upon plugging the guess into the differential equation we get,
r r rh er t = Ah e r t r r r ( Ah - rh ) er t = 0 r r ( A - rI )h er t = 0 r
Now, since we know that exponentials are not zero we can drop that portion and we then see that in order for (2) to be a solution to (1) then we must have
( A - rI )h = 0
r
Or, in order for (2) to be a solution to (1), r and h must be an eigenvalue and eigenvector for the matrix A. Therefore, in order to solve (1) we first find the eigenvalues and eigenvectors of the matrix A and then we can form solutions using (2). There are going to be three cases that well need to look at. The cases are real, distinct eigenvalues, complex eigenvalues and repeated eigenvalues. None of this tells us how to completely solve a system of differential equations. Well need the following couple of facts to do this.
2007 Paul Dawkins
273
Fact r r 1. If x1 ( t ) and x2 ( t ) are two solutions to a homogeneous system, (1), then
r r c1 x1 ( t ) + c2 x2 ( t )
is also a solution to the system. r r r 2. Suppose that A is an n x n matrix and suppose that x1 ( t ) , x2 ( t ) , , xn ( t ) are solutions to a homogeneous system, (1). Define,
In other words, X is a matrix whose ith column is the ith solution. Now define,
r r r X = ( x1 x2 L xn ) W = det ( X )
We call W the Wronskian. If W 0 then the solutions form a fundamental set of solutions and the general solution to the system is,
r r r r x ( t ) = c1 x1 ( t ) + c2 x2 ( t ) + L + cn xn ( t )
Note that if we have a fundamental set of solutions then the solutions are also going to be linearly independent. Likewise, if we have a set of linearly independent solutions then they will also be a fundamental set of solutions since the Wronskian will not be zero.
2007 Paul Dawkins
274
Phase Plane
Before proceeding with actually solving systems of differential equations theres one topic that we need to take a look at. This is a topic thats not always taught in a differential equations class but in case youre in a course where it is taught we should cover it so that you are prepared for it. Lets start with a general homogeneous system, Notice that
r r x = Ax r r x=0
(1)
is a solution to the system of differential equations. What wed like to ask is, do the other solutions to the system approach this solution as t increases or do they move away from this solution? We did something similar to this when we classified equilibrium solutions in a previous section. In fact, what were doing here is simply an extension of this idea to systems of differential equations. The solution x = 0 is called an equilibrium solution for the system. As with the single differential equations case, equilibrium solutions are those solutions for which
r r Ax = 0 r r x=0
We are going to assume that A is a nonsingular matrix and hence will have only one solution, and so we will have only one equilibrium solution. Back in the single differential equation case recall that we started by choosing values of y and plugging these into the function f(y) to determine values of y . We then used these values to sketch tangents to the solution at that particular value of y. From this we could sketch in some solutions and use this information to classify the equilibrium solutions. We are going to do something similar here, but it will be slightly different as well. First we are going to restrict ourselves down to the 2 x 2 case. So, well be looking at systems of the form,
= ax1 + bx2 x1 = cx1 + dx2 x2
r a b r x = x c d
Solutions to this system will be of the form,
r x (t ) x = 1 x2 ( t ) r 0 x = 0
and our single equilibrium solution will be,
In the single differential equation case we were able to sketch the solution, y(t) in the y-t plane and see actual solutions. However, this would somewhat difficult in this case since our solutions are actually vectors. What were going to do here is think of the solutions to the system as points
in the x1-x2 plane and plot these points. Our equilibrium solution will correspond to the origin of x1-x2 plane and the x1-x2 plane is called the phase plane. To sketch a solution in the phase plane we can pick values of t and plug these into the solution. This gives us a point in the x1-x2 or phase plane that we can plot. Doing this for many values of t will then give us a sketch of what the solution will be doing in the phase plane. A sketch of a particular solution in the phase plane is called the trajectory of the solution. Once we have the trajectory of a solution sketched we can then ask whether or not the solution will approach the equilibrium solution as t increases. We would like to be able to sketch trajectories without actually having solutions in hand. There are a couple of ways to do this. Well look at one of those here and well look at the other in the next couple of sections. One way to get a sketch of trajectories is to do something similar to what we did the first time we r looked at equilibrium solutions. We can choose values of x (note that these will be points in the r r phase plane) and compute Ax . This will give a vector that represents x at that particular solution. As with the single differential equation case this vector will be tangent to the trajectory at that point. We can sketch a bunch of the tangent vectors and then sketch in the trajectories. This is a fairly work intensive way of doing these and isnt the way to do them in general. However, it is a way to get trajectories without doing any solution work. All we need is the system of differential equations. Lets take a quick look at an example.
Example 1 Sketch some trajectories for the system, = x1 + 2 x2 x1 = 3 x1 + 2 x2 x2
r 1 2 r x = x 3 2
Solution So, what we need to do is pick some points in the phase plane, plug them into the right side of the system. Well do this for a couple of points.
r -1 x = 1 r 2 x = 0 r -3 x = -2
r 1 x = 3 r 1 x = 3
2 -1 1 = 2 1 -1 2 2 2 = 2 0 6
r 1 2 -3 -7 x = = 3 2 -2 -13
So, what does this tell us? Well at the point (-1, 1) in the phase plane there will be a vector pointing in the direction 1, -1 . At the point (2,0) there will be a vector pointing in the direction
2, 6 . At the point (-3,-2) there will be a vector pointing in the direction -7, -13 .
Doing this for a large number of points in the phase plane will give the following sketch of vectors.
2007 Paul Dawkins
276
Now all we need to do is sketch in some trajectories. To do this all we need to do is remember that the vectors in the sketch above are tangent to the trajectories. Also the direction of the vectors give the direction of the trajectory as t increases so we can show the time dependence of the solution by adding in arrows to the trajectories. Doing this gives the following sketch.
This sketch is called the phase portrait. Usually phase portraits only include the trajectories of the solutions and not any vectors. All of our phase portraits form this point on will only include the trajectories. In this case it looks like most of the solutions will start away from the equilibrium solution then as t starts to increase they move in towards the equilibrium solution and then eventually start moving away from the equilibrium solution again. There seem to be four solutions that have slightly different behaviors. It looks like two of the solutions will start at (or near at least) the equilibrium solution and them move straight away from
it while two other solution start away from the equilibrium solution and then move straight in towards the equilibrium solution. In these kinds of cases we call the equilibrium point a saddle point and we call the equilibrium point in this case unstable since all but two of the solutions are moving away from it as t increases. As we noted earlier this is not generally the way that we will sketch trajectories. All we really need to get the trajectories are the eigenvalues and eigenvectors of the matrix A. We will see how to do this over the next couple of sections as we solve the systems. Here are a few more phase portraits so you can see some more possible examples. Well actually be generating several of these throughout the course of the next couple of sections.
2007 Paul Dawkins
278
Not all possible phase portraits have been shown here. These are here to show you some of the possibilities. Make sure to notice that several kinds can be either asymptotically stable or unstable depending upon the direction of the arrows. Notice the difference between stable and asymptotically stable. In an asymptotically stable node or spiral all the trajectories will move in towards the equilibrium point as t increases whereas, a center (which is always stable) trajectories will just move around the equilibrium point but never actually move in towards it.
2007 Paul Dawkins
279
Real, Distinct Eigenvalues

Its now time to start solving systems of differential equations. Weve seen that solutions to the system,
r r x = Ax
will be of the form
where l and h are eigenvalues and eigenvectors of the matrix A. We will be working with 2 x 2 systems so this means that we are going to be looking for two solutions, x1 ( t ) and x2 ( t ) , where the determinant of the matrix, is nonzero. We are going to start by looking at the case where our two eigenvalues, l1 and l 2 are real and distinct. In other words they will be real, simple eigenvalues. Recall as well that the eigenvectors for simple eigenvalues are linearly independent. This means that the solutions we get from these will also be linearly independent. If the solutions are linearly independent the matrix X must be nonsingular and hence these two solutions will be a fundamental set of solutions. The general solution in this case will then be,
r r x = h el t r r
r r X = ( x1 x2 )
r1 r2 r x ( t ) = c1e l1th ( ) + c2el2th ( )
Note that each of our examples will actually be broken into two examples. The first example will be solving the system and the second example will be sketching the phase portrait for the system. Phase portraits are not always taught in a differential equations course and so well strip those out of the solution process so that if you havent covered them in your class you can ignore the phase portrait example for the system.
Example 1 Solve the following IVP. r 1 2 r x = x, 3 2 1- l 3 2 2-l
0 r x ( 0) = -4
Solution So, the first thing that we need to do is find the eigenvalues for the matrix.
det ( A - l I ) =
= l 2 - 3l - 4 = ( l + 1)( l - 4 )
Now lets find the eigenvectors for each of these.
l1 = -1, l2 = 4
l1 = -1 :
Well need to solve,
2 2 h1 0 = 3 3 h 2 0
2007 Paul Dawkins
280
2h1 + 2h 2 = 0
h1 = -h 2
The eigenvector in this case is,
r -h h = 2 h2
-1 r h (1) = , 1
h2 = 1
l2 = 4 :
Well need to solve,
-3 2 h1 0 = 3 -2 h 2 0
2 h r 3 h = 2 h2
-3h1 + 2h2 = 0
2 h1 = h 2 3
2 r h ( 2) = , 3
h2 = 3
Then general solution is then,
-1 2 r x ( t ) = c1e - t + c2e 4t 1 3
Now, we need to find the constants. To do this we simply need to apply the initial conditions.
0 r -1 2 -4 = x ( 0 ) = c1 1 + c2 3
All we need to do now is multiply the constants through and we then get two equations (one for each row) that we can solve for the constants. This gives,
-c1 + 2c2 = 0 c1 + 3c2 = -4

8 4 c1 = - , c2 = 5 5
r 8 -1 4 2 x ( t ) = - e -t - e 4 t 5 1 5 3
Now, lets take a look at the phase portrait for the system.
Example 2 Sketch the phase portrait for the following system. r 1 2 r x = x 3 2

Solution From the last example we know that the eigenvalues and eigenvectors for this system are,
l1 = -1 l2 = 4
-1 r h (1) = 1 r 2 h ( 2) = 3
It turns out that this is all the information that we will need to sketch the direction field. We will relate things back to our solution however so that we can see that things are going correctly. Well start by sketching lines that follow the direction of the two eigenvectors. This gives,
Now, from the first example our general solution is
-1 2 r x ( t ) = c1e - t + c2e 4t 1 3
If we have c2 = 0 then the solution is an exponential times a vector and that all the exponential does is affect the magnitude of the vector and the constant c1 will affect both the sign and the magnitude of the vector. In other words, the trajectory in this case will be a straight line that is r1 parallel to the vector, h ( ) . Also notice that as t increases the exponential will get smaller and smaller and hence the trajectory will be moving in towards the origin. If c1>0 the trajectory will be in Quadrant II and if c1<0 the trajectory will be in Quadrant IV.
() So the line in the graph above marked with h will be a sketch of the trajectory corresponding to c2 = 0 and this trajectory will approach the origin as t increases.
r1
If we now turn things around and look at the solution corresponding to have c1 = 0 we will have a
trajectory that is parallel to h ( ) . Also, since the exponential will increase as t increases and so in this case the trajectory will now move away from the origin as t increases. We will denote this with arrows on the lines in the graph above.
r2
2007 Paul Dawkins
282
Notice that we could have gotten this information with actually going to the solution. All we really need to do is look at the eigenvalues. Eigenvalues that are negative will correspond to solutions that will move towards the origin as t increase in a direction that is parallel to its eigenvector. Likewise, eigenvalues that are positive move away from the origin as t increases in a direction that will be parallel to its eigenvector. If both constants are in the solution we will have a combination of these behaviors. For large negative ts the solution will be dominated by the portion that has the negative eigenvalue since in these cases the exponent will be large and positive. Trajectories for large negative ts will be r (1) parallel to h and moving in the same direction. Solutions for large positive ts will be dominated by the portion with the positive eigenvalue. Trajectories in this case will be parallel to h
r( 2)
and moving in the same direction.
In general, it looks like trajectories will start near h ( ) , move in towards the origin and then as they get closer to the origin they will start moving towards h ( ) and then continue up along this vector. Sketching some of these in will give the following phase portrait. Here is a sketch of this with the trajectories corresponding to the eigenvectors marked in blue.
r1
r2
2007 Paul Dawkins
283
In this case the equilibrium solution (0,0) is called a saddle point and is unstable. In this case unstable means that solution move away from it as t increases. So, weve solved a system in matrix form, but remember that we started out without the systems in matrix form. Now lets take a quick look at an example of a system that isnt in matrix form initially.
Example 3 Find the solution to the following system. = x1 + 2 x2 x1 x1 ( 0 ) = 0 = 3 x1 + 2 x2 x2 x2 ( 0 ) = -4

Solution We first need to convert this into matrix form. This is easy enough. Here is the matrix form of the system.
r 1 2 r x = x, 3 2
0 r x ( 0) = -4
This is just the system from the first example and so weve already got the solution to this system. Here it is.
r 8 -1 4 2 x ( t ) = - e -t - e 4 t 5 1 5 3
Now, since we want to solution to the system not in matrix form lets go one step farther here. Lets multiply the constants and exponentials into the vectors and then add up the two vectors.
4t -t 8 4t 8 e -t 8 8 r 5e 5e - 5e x ( t ) = 5 8 -t - 12 = 8 - t 12 4t 4t - 5e 5 e - 5 e - 5 e
Now, recall,
2007 Paul Dawkins
284
x (t ) r x (t ) = 1 x2 ( t )
So, the solution to the system is then,
8 8 x1 ( t ) = e -t - e4 t 5 5 8 12 x2 ( t ) = - e - t - e 4t 5 5
Lets work another example.
Example 4 Solve the following IVP. r -5 1 r x = x, 4 -2 -5 - l 4 1 -2 - l
1 r x ( 0) = 2
Solution So, the first thing that we need to do is find the eigenvalues for the matrix.
det ( A - l I ) =
= l 2 + 7l + 6 = ( l + 1)( l + 6 )
Now lets find the eigenvectors for each of these.
l1 = -1, l2 = -6
l1 = -1 :
Well need to solve,
-4 1 h1 0 = 4 -1 h 2 0
-4h1 + h 2 = 0
h 2 = 4h1
r h h = 1 4h1
1 r h (1) = , 4
h1 = 1
l 2 = -6 :
Well need to solve,
1 1 h1 0 = 4 4 h 2 0
h1 + h2 = 0
h1 = -h 2
r -h h = 2 h2
-1 r h ( 2) = , 1
h2 = 1
2007 Paul Dawkins
285
Then general solution is then,
1 -1 r x ( t ) = c1e -t + c2e -6t 4 1

Now, we need to find the constants. To do this we simply need to apply the initial conditions.
1 r 1 -1 2 = x ( 0 ) = c1 4 + c2 1
Now solve the system for the constants.
c1 - c2 = 1 4c1 + c2 = 2
3 2 c1 = , c2 = 5 5
-1 r 3 1 2 x ( t ) = e - t - e -6 t 5 4 5 1
Now lets find the phase portrait for this system.
Example 5 Sketch the phase portrait for the following system. r -5 1 r x = x 4 -2

Solution From the last example we know that the eigenvalues and eigenvectors for this system are,
l1 = -1 l2 = -6
1 r h (1) = 4 r -1 h ( 2) = 1
This one is a little different from the first one. However it starts in the same way. Well first sketch the trajectories corresponding to the eigenvectors. Notice as well that both of the eigenvalues are negative and so trajectories for these will move in towards the origin as t increases. When we sketch the trajectories well add in arrows to denote the direction they take as t increases. Here is the sketch of these trajectories.
2007 Paul Dawkins
286
Now, here is where the slight difference from the first phase portrait comes up. All of the trajectories will move in towards the origin as t increases since both of the eigenvalues are negative. The issue that we need to decide upon is just how they do this. This is actually easier than it might appear to be at first. The second eigenvalue is larger than the first. For large and positive ts this means that the solution for this eigenvalue will be smaller than the solution for the first eigenvalue. Therefore, r (1) as t increases the trajectory will move in towards the origin and do so parallel to h . Likewise, since the second eigenvalue is larger than the first this solution will dominate for large and negative ts. Therefore, as we decrease t the trajectory will move away from the origin and do so parallel to h
r( 2)
Adding in some trajectories gives the following sketch.
In these cases we call the equilibrium solution (0,0) a node and it is asymptotically stable. Equilibrium solutions are asymptotically stable if all the trajectories move in towards it as t increases.
Note that nodes can also be unstable. In the last example if both of the eigenvalues had been positive all the trajectories would have moved away from the origin and in this case the equilibrium solution would have been unstable. Before moving on to the next section we need to do one more example. When we first started talking about systems it was mentioned that we can convert a higher order differential equation into a system. We need to do an example like this so we can see how to solve higher order differential equations using systems.
Example 6 Convert the following differential equation into a system, solve the system and use
this solution to get the solution to the original differential equation.
2 y + 5 y - 3 y = 0, x1 = y x2 = y
The system is then,
y ( 0 ) = -4 y ( 0 ) = 9
Solution So, we first need to convert this into a system. Heres the change of variables,
= y = x2 x1 x 2 = y =
3 5 3 5 y - y = x1 - x2 2 2 2 2 -4 r x ( 0) = 9
r 0 1 r x = 3 x 5 2 2
where,
x (t ) y (t ) r x (t ) = 1 = x2 ( t ) y ( t ) -l
3 2
Now we need to find the eigenvalues for the matrix.
det ( A - l I ) =
1 -5 2 -l 1 2
3 = l2 + 5 2l- 2
=
Now lets find the eigenvectors.
1 2
( l + 3)( 2l - 1)
l1 = -3, l2 =
l1 = -3 :
Well need to solve,
3 1 h1 0 3 1 = 2 2 h 2 0
3h1 + h2 = 0
h 2 = -3h1
2007 Paul Dawkins
288
r h h = 1 -3h1
1 r h (1) = , -3
h1 = 1
l2 = 1 2 :
Well need to solve,
- 1 2 3 2
1 h1 0 = -3 h2 0 r h h =1 1 2 h1
1 - h1 + h2 = 0 2
1 h 2 = h1 2
2 r h ( 2) = , 1
h1 = 2
t 2 1 r x ( t ) = c1e -3t + c2e 2 -3 1
Apply the initial condition.
-4 r 1 2 9 = x ( 0 ) = c1 -3 + c2 1
This gives the system of equations that we can solve for the constants.
c1 + 2c2 = -4 -3c1 + c2 = 9
The actual solution to the system is then,
c1 = -
22 3 , c2 = 7 7
1 3 t 2 r 22 x ( t ) = - e -3t - e 2 7 -3 7 1
Now recalling that,
y (t ) r x (t ) = y ( t ) 22 -3t 6 2t e - e 7 7
we can see that the solution to the original differential equation is just the top row of the solution to the matrix system. The solution to the original differential equation is then,
y (t ) = -
Notice that as a check, in this case, the bottom row should be the derivative of the top row.
2007 Paul Dawkins
289
Complex Eigenvalues
In this section we will look at solutions to
r r x = Ax
where the eigenvalues of the matrix A are complex. With complex eigenvalues we are going to have the same problem that we had back when we were looking at second order differential equations. We want our solutions to only have real numbers in them, however since our solutions to systems are of the form,
r r x = h el t
we are going to have complex numbers come into our solution from both the eigenvalue and the eigenvector. Getting rid of the complex numbers here will be similar to how we did it back in the second order differential equation case, but will involve a little more work this time around. Its easiest to see how to do this in an example.
Example 1 Solve the following IVP. r 3 -9 r x = x 4 -3 3-l 4 -9 -3 - l
2 r x ( 0) = -4
Solution We first need the eigenvalues and eigenvectors for the matrix.
det ( A - l I ) =
= l 2 + 27
l1,2 = 3 3 i
So, now that we have the eigenvalues recall that we only need to get the eigenvector for one of the eigenvalues since we can get the second eigenvector for free from the first eigenvector.
l1 = 3 3 i :
We need to solve the following system.
3-3 3i h1 0 -9 = 4 -3 - 3 3 i h 2 0
Using the first equation we get,
(3 - 3 3 i )h - 9h
1
=0 1 1 - 3 i h1 3
h2 =
So, the first eigenvector is,
h1 r h= 1 1 - 3 i h1 3 3 r h (1) = 1 - 3 i
h1 = 3
2007 Paul Dawkins
290
When finding the eigenvectors in these cases make sure that the complex number appears in the numerator of any fractions since well need it in the numerator later on. Also try to clear out any fractions by appropriately picking the constant. This will make our life easier down the road. Now, the second eigenvector is,
3 r h (2) = 1 + 3 i
However, as we will see we wont need this eigenvector. The solution that we get from the first eigenvalue and eigenvector is,
r x1 ( t ) = e3
3it
3 1 - 3 i
So, as we can see there are complex numbers in both the exponential and vector that we will need to get rid of in order to use this as a solution. Recall from the complex roots section of the second order differential equation chapter that we can use Eulers formula to get the complex number out of the exponential. Doing this gives us,
3 r x1 ( t ) = cos 3 3t + i sin 3 3t 1 - 3 i
( (
))
The next step is to multiply the cosines and signs into the vector.
3cos 3 3t + 3i sin 3 3t r x1 ( t ) = cos 3 3t + i sin 3 3t - 3 i cos 3 3t + 3 sin 3 3t
( ) ( )
Now combine the terms with an i in them and split these terms off from those terms that dont contain an i. Also factor the i out of this vector.
3cos 3 3t 3sin 3 3t r x1 ( t ) = +i cos 3 3t + 3 sin 3 3t sin 3 3t - 3 cos 3 3t r r = u (t ) + i v (t )
Now, it can be shown (well leave the details to you) that u ( t ) and v ( t ) are two linearly independent solutions to the system of differential equations. This means that we can use them to form a general solution and they are both real solutions. So, the general solution to a system with complex roots is
where u ( t ) and v ( t ) are found by writing the first solution as
r r r x ( t ) = c1u ( t ) + c2v ( t ) r r r x (t ) = u (t ) + i v (t )
2007 Paul Dawkins
291
For our system then, the general solution is,
3cos 3 3t 3sin 3 3t r + c2 x ( t ) = c1 cos 3 3t + 3 sin 3 3t sin 3 3t - 3 cos 3 3t
We now need to apply the initial condition to this to find the constants.
0 2 r 3 = x = c + c 0 ( ) -4 1 2 1 - 3
This leads to the following system of equations to be solved,
c1 - 3c2 = -4
The actual solution is then,
3c1 = 2
2 14 c1 = , c2 = 3 3 3
3cos 3 3t 3sin 3 3t 2 14 r x (t ) = + 3 cos 3 3t + 3 sin 3 3t 3 3 sin 3 3t - 3 cos 3 3t
As we did in the last section well do the phase portraits separately from the solution of the system in case phase portraits havent been taught in your class.
Example 2 Sketch the phase portrait for the system. r 3 -9 r x = x 4 -3

Solution When the eigenvalues of a matrix A are purely complex, as they are in this case, the trajectories of the solutions will be circles or ellipses that are centered at the origin. The only thing that we really need to concern ourselves with here are whether they are rotating in a clockwise or counterclockwise direction. This is easy enough to do. Recall when we first looked at these phase portraits a couple of r sections ago that if we pick a value of x ( t ) and plug it into our system we will get a vector that will be tangent to the trajectory at that point and pointing in the direction that the trajectory is traveling.. So, lets pick the following point and see what we get.
r 1 x = 0
r 3 -9 1 3 x = = 4 -3 0 4
Therefore at the point (1,0) in the phase plane the trajectory will be point in a upwards direction. The only way that this can be is if the trajectories are traveling in a counterclockwise direction. Here is the sketch of some of the trajectories for this problem.
2007 Paul Dawkins
292
The equilibrium solution in the case is called a center and is stable. Note in this last example that the equilibrium solution is stable and not asymptotically stable. Asymptotically stable refers to the fact that the trajectories are moving in toward the equilibrium solution as t increases. In this example the trajectories are simply revolving around the equilibrium solution and not moving in towards it. The trajectories are also not moving away from the equilibrium solution and so they arent unstable. Therefore we call the equilibrium solution stable. Not all complex eigenvalues will result in centers so lets take a look at an example where we get something different.
Example 3 Solve the following IVP. r 3 -13 r x = x 5 1

Solution Lets get the eigenvalues and eigenvectors for the matrix.
3 r x ( 0) = -10
det ( A - l I ) =
3 - l -13 5 1- l
= l 2 - 4l + 68
Now get the eigenvector for the first eigenvalue.
l1,2 = 2 8 i
l1 = 2 + 8i :
We need to solve the following system.
-13 h1 0 1 - 8i = -1 - 8i h2 0 5
Using the second equation we get,
5h1 + ( -1 - 8 i )h 2 = 0
h1 =
So, the first eigenvector is,
1 (1 + 8 i )h2 5
1 r (1 + 8 i )h 2 h= 5 h2 r 1 + 8 i h (1) = 5 1 + 8 i r x1 ( t ) = e( 2 +8i )t 5 1 + 8 i = e 2t e 8i t 5
h2 = 5
The solution corresponding the this eigenvalue and eigenvector is
1 + 8 i = e 2t ( cos ( 8t ) + i sin ( 8t ) ) 5
As with the first example multiply cosines and signs into the vector and split it up. Dont forget about the exponential that is in the solution this time.
cos (8t ) - 8sin (8t ) 8cos (8t ) + sin ( 8t ) r 2t x1 ( t ) = e 2t + ie 5cos ( 8t ) 5sin (8t ) r r = u (t ) + i v (t )
The general solution to this system then,
cos ( 8t ) - 8sin ( 8t ) 8cos ( 8t ) + sin ( 8t ) r 2t x ( t ) = c1e 2 t + c2 e 5cos ( 8t ) 5sin ( 8t ) 3 r 1 2t 8 -10 = x ( 0 ) = c1 5 + c2e 0 c1 + 8c2 = 3 5 c1 = -2, c2 = 5c1 = -10 8
Now apply the initial condition and find the constants.
cos ( 8t ) - 8sin ( 8t ) 5 2t 8cos ( 8t ) + sin ( 8t ) r x ( t ) = -2e 2t + e 5cos 8 t 5sin ( 8t ) ( ) 8
Lets take a look at the phase portrait for this problem.

Example 4 Sketch the phase portrait for the system. r 3 -13 r x = x 5 1

Solution When the eigenvalues of a system are complex with a real part the trajectories will spiral into or out of the origin. We can determine which one it will be by looking at the real portion. Since the real portion will end up being the exponent of an exponential function (as we saw in the solution to this system) if the real part is positive the solution will grow very large as t increases. Likewise, if the real part is negative the solution will die out as t increases. So, if the real part is positive the trajectories will spiral out from the origin and if the real part is negative they will spiral into the origin. We determine the direction of rotation (clockwise vs. counterclockwise) in the same way that we did for the center. In our case the trajectories will spiral out from the origin since the real part is positive and
r 3 -13 1 3 x = = 5 1 0 5
will rotate in the counterclockwise direction as the last example did. Here is a sketch of some of the trajectories for this system.
Here we call the equilibrium solution a spiral (oddly enough) and in this case its unstable since the trajectories move away from the origin. If the real part of the eigenvalue is negative the trajectories will spiral into the origin and in this case the equilibrium solution will be asymptotically stable.
2007 Paul Dawkins
295
Repeated Eigenvalues
This is the final case that we need to take a look at. In this section we are going to look at solutions to the system,
r r x = Ax
where the eigenvalues are repeated eigenvalues. Since we are going to be working with systems in which A is a 2 x 2 matrix we will make that assumption from the start. So the system will have a double eigenvalue, l. This presents us with a problem. We want two linearly independent solutions so that we can form a general solution. However, with a double eigenvalue we will have only one,
r r x1 = h elt
So, we need to come up with a second solution. Recall that when we looked at the double root case with the second order differential equations we ran into a similar problem. In that section we simply added a t to the solution and were able to get a second solution. Lets see if the same thing will work in this case as well. Well see if
r r x = t e lth
will also be a solution. To check all we need to do is plug into the system. Dont forget to product rule the proposed solution when you differentiate!
r r r h elt + lh te lt = Ah te lt
Now, we got two functions here on the left side, an exponential by itself and an exponential times a t. So, in order for our guess to be a solution we will need to require,
r r Ah = lh r r h =0
( A - l I )h = 0
The first requirement isnt a problem since we this just says that l is an eigenvalue and its r eigenvector is h . We already knew this however so theres nothing new there. The second r however is a problem. Since h is an eigenvector we know that it cant be zero, yet in order to satisfy the second condition it would have to be. So, our guess was incorrect. The problem seems to be that there is a lone term with just an exponential in it so lets see if we cant fix up our guess to correct that. Lets try the following guess. where r is an unknown vector that well need to determine. As with the first guess lets plug this into the system and see what we get.
r r r x = t e lth + elt r
r r r r r h elt + lh te lt + lr elt = A (h te lt + r elt ) r r r r r (h + lr ) elt + lh telt = Ah telt + Ar elt
Now set coefficients equal again,

r r lh = Ah r r r h + lr = Ar
( A - l I )h = 0 r r ( A - lI ) r =h
As with our first guess the first equation tells us nothing that we didnt already know. This time r the second equation is not a problem. All the second equation tells us is that r must be a solution to this equation. It looks like our second guess worked. Therefore,
r r r x2 = t e lth + elt r r will be a solution to the system provided r is a solution to r r ( A - lI ) r =h

Also this solution and the first solution are linearly independent and so they form a fundamental set of solutions and so the general solution in the double eigenvalue case is,
r r r r x = c1 e lth + c2 ( t elth + e lt r )
Lets work an example.
Example 1 Solve the following IVP. r 7 1 r x = x -4 3

Solution First find the eigenvalues for the system.
2 r x ( 0) = -5
det ( A - l I ) =
7-l 1 -4 3 - l
2
= l 2 - 10l + 25 = ( l - 5)
l1,2 = 5
So, we got a double eigenvalue. Of course that shouldnt be too surprising given the section that were in. Lets find the eigenvector for this eigenvalue.
2 1 h1 0 = -4 -2 h2 0
The eigenvector is then,
2h1 + h 2 = 0
h 2 = -2h1
r h h = 1 -2h1 1 r h (1) = -2
The next step is find r . To do this well need to solve,
h1 0 h1 = 1
2 1 r1 1 = -4 -2 r 2 -2
2 r1 + r 2 = 1
r 2 = 1 - 2 r1
Note that this is almost identical to the system that we solve to find the eigenvalue. The only r difference is the right hand side. The most general possible r is
r r1 r = 1 - 2 r1
r 0 r = 1
if r1 = 0
In this case, unlike the eigenvector system we can choose the constant to be anything we want, so we might as well pick it to make our life easier. This usually means picking it to be zero. We can now write down the general solution to the system.
1 1 0 r x ( t ) = c1e5t + c2 e5t t + e5t -2 -2 1

Applying the initial condition to find the constants gives us,
2 r 1 0 -5 = x ( 0 ) = c1 -2 + c2 1 c1 = 2 c1 = 2, c2 = -1 -2c1 + c2 = -5
1 1 0 r x ( t ) = 2e5 t - te5t + e5 t -2 -2 1 2 1 0 = e 5t - e 5t t - e 5t -4 -2 1 2 1 = e 5t - e 5t t -5 -2
Note that we did a little combining here to simplify the solution up a little. So, the next example will be to sketch the phase portrait for this system.
Example 2 Sketch the phase portrait for the system. r 7 1 r x = x -4 3

Solution These will start in the same way that real, distinct eigenvalue phase portraits start. Well first sketch in a trajectory that is parallel to the eigenvector and note that since the eigenvalue is positive the trajectory will be moving away from the origin.
2007 Paul Dawkins
298
Now, it will be easier to explain the remainder of the phase portrait if we actually have one in front of us. So here is the full phase portrait with some more trajectories sketched in.
Trajectories in these cases always emerge from (or move into) the origin in a direction that is parallel to the eigenvector. Likewise they will start in one direction before turning around and moving off into the other direction. The directions in which they move are opposite depending on which side of the trajectory corresponding to the eigenvector we are on. Also, as the trajectories moves away from the origin it should start becoming parallel to the trajectory corresponding to the eigenvector. So, how do we determine the direction? We can do the same thing that we did in the complex case. Well plug in (1,0) into the system and see which direction the trajectories are moving at that point. Since this point is directly to the right of the origin the trajectory at that point must have already turned around and so this will give the direction that it will traveling after turning around. Doing that for this problem to check our phase portrait gives,
7 1 1 7 -4 3 0 = -4
This vector will point down into the fourth quadrant and so the trajectory must be moving into the fourth quadrant as well. This does match up with our phase portrait. In these cases the equilibrium is called a node and is unstable in this case. Note that sometimes you will hear nodes for the repeated eigenvalue case called degenerate nodes or improper nodes. Lets work one more example.
Example 3 Solve the following IVP. r -1 3 r x = 1 2 x - 6 -2

Solution First the eigenvalue for the system.
1 r x ( 2) = 0
3 2
det ( A - l I ) =
-1 - l -1 6
-2 - l 9 4
= l 2 + 3l + 3 = l + 2
Now lets get the eigenvector.
2
l1,2 = -
3 2
1 2 1 - 6
h1 0 = -1 0 2 h 2
3 2
1 3 h1 + h 2 = 0 2 2
h1 = -3h 2
r -3h 2 h = h2 -3 r h (1) = 1
Now find r ,
h2 0 h2 = 1
1 2 1 6
r1 -3 = - r2 1 r -6 - 3r 2 r = r2
3 2 1 2
1 3 r1 + r 2 = -3 2 2 r -6 r = 0
r1 = -6 - 3r 2
if r 2 = 0
The general solution for the system is then,
2007 Paul Dawkins
300
- 3t -3 - 3t -6 r - 3t -3 x ( t ) = c1e 2 + c2 te 2 + e 2 1 1 0
Applying the initial condition gives,
-3 -3 -3 -6 1 r -3 -3 = x ( 2 ) = c1e + c2 2e + e 0 1 1 0
Note that we didnt use t=0 this time! We now need to solve the following system,
-3e-3c1 - 12e -3c2 = 1 e -3 c1 + 2e -3 c2 = 0

c1 =
e3 e3 , c2 = 3 6
r e3 - 32t -3 e3 - 32t -3 - 32t -6 x ( t ) = e - te + e 3 1 6 1 0 =e

t +3 -3 2
1 t +3 0 -3 2 2 1 + te 1 3 - 6
And just to be consistent with all the other problems that weve done lets sketch the phase portrait.
Example 4 Sketch the phase portrait for the system. r -1 3 r x = 1 2 x - 6 -2

Solution Lets first notice that since the eigenvalue is negative in this case the trajectories should all move in towards the origin. Lets check the direction of the trajectories at (1,0)
-1 3 -1 2 1 1 = 1 - 6 -2 0 - 6
So it looks like the trajectories should be pointing into the third quadrant at (1,0). This gives the following phase portrait.
2007 Paul Dawkins
301
2007 Paul Dawkins
302
Nonhomogeneous Systems
We now need to address nonhomogeneous systems briefly. Both of the methods that we looked at back in the second order differential equations chapter can also be used here. As we will see Undetermined Coefficients is almost identical when used on systems while Variation of Parameters will need to have a new formula derived, but will actually be slightly easier when applied to systems. Undetermined Coefficients The method of Undetermined Coefficients for systems is pretty much identical to the second order differential equation case. The only difference is that the coefficients will need to be vectors now. Lets take a quick look at an example.
Example 1 Find the general solution to the following system. r 1 2 r 2 x = x + t -4 3 2

Solution We already have the complimentary solution as we solved that part back in the real eigenvalue section. It is,
-1 2 r xc ( t ) = c1e- t + c2e 4t 1 3
Guessing the form of the particular solution will work in exactly the same way it did back when we first looked at this method. We have a linear polynomial and so our guess will need to be a linear polynomial. The only difference is that the coefficients will need to be vectors instead of constants. The particular solution will have the form,
r r r a b xP = ta + b = t 1 + 1 a2 b2 r r a1 x P =a = a2
So, we need to differentiate the guess
Before plugging into the system lets simplify the notation a little to help with our work. Well write the system as,
r 1 2 r 2 r r x = x + t = Ax + tg 3 2 -4
This will make the following work a little easier. Now, lets plug things into the system.
2007 Paul Dawkins
303
r r r r a = A ta + b + tg r r r r a = tAa + Ab + tg r r r r r 0 = t ( Aa + g ) + Ab - a
Now we need to set the coefficients equal. Doing this gives,
t1 : t0 : r
r r r Aa + g = 0 r r r Ab - a = 0
r r Aa = - g r r Ab = a
Now only a is unknown in the first equation so we can use Gaussian elimination to solve the system. Well leave this work to you to check.
1 2 a1 2 a = - 3 2 2 -4 r
r 3 a = 5 - 2 r - 11 b = 234 8
Now that we know a we can solve the second equation for b .
1 2 b1 3 = 5 3 2 b2 - 2
So, since we were able to solve both equations, the particular solution is then,
3 - 11 r xP = t 5 + 234 - 2 8
-1 2 3 - 11 r x ( t ) = c1e - t + c2e4 t + t 5 + 234 1 3 - 2 8

So, as you can see undetermined coefficients is nearly the same as the first time we saw it. The work in solving for the constants is a little messier however. Variation of Parameters In this case we will need to derive a new formula for variation of parameters for systems. The derivation this time will be much simpler than the when we first saw variation of parameters. First let X(t) be a matrix whose ith column is the ith linearly independent solution to the system,
r r x = Ax
Now it can be shown that X(t) will be a solution to the following differential equation.
X = AX
(1)
This is nothing more than the original system with the matrix in place of the original vector. We are going to try and find a particular solution to
r r r x = Ax + g ( t )
2007 Paul Dawkins
r r xP = X ( t ) v ( t ) r where we will need to determine the vector v ( t ) . To do this we will need to plug this into the
nonhomogeneous system. Dont forget to product rule the particular solution when plugging the guess into the system.
We will assume that we can find a solution of the form,
r r r r X v + X v = A X v + g
Note that we dropped the (t) part of things to simplify the notation a little. Now using (1) we can rewrite this a little.
r r r r X v + X v = X v + g r r X v = g
Because we formed X using linearly independent solutions we know that det(X) must be nonzero and this in turn means that we can find the inverse of X. So, multiply both sides by the inverse of X.
r r v = X -1 g
Now all that we need to do is integrate both sides to get v ( t ) .
r r v ( t ) = X -1 g dt
As with the second order differential equation case we can ignore any constants of integration. The particular solution is then,
r r xP = X X -1 g dt
(2)
Lets work a quick example using this.
Example 2 Find the general solution to the following system. r -5 1 r 2t 6 x = x + e -1 4 -2

Solution We found the complimentary solution to this system in the real eigenvalue section. It is,
1 -1 r xc ( t ) = c1e -t + c2 e-6 t 4 1 e -t X = -t 4e -e -6t e -6 t
Now the matrix X is,
Now, we need to find the inverse of this matrix. We saw how to find inverses of matrices back in the second linear algebra review section and the process is the same here even though we dont have constant entries. Well leave the detail to you to check.
2007 Paul Dawkins
305
1 et X -1 = 54 6t - 5 e
Now do the multiplication in the integral.
t r 1 5e X g = 4 6t - 5 e -1
et 1 6t 5e
1 5
et 6e 2t e3t = 2t 8t 1 6 t -5e 5 e -e
1 5
Now do the integral.
e3t dt 1 e3t e 3t r = 3 dt = X g dt = 8t 5 8t t 8 -5e dt - 8 e -5e

-1
Remember that to integrate a matrix or vector you just integrate the individual entries. We can now get the particular solution.
r r xP = X X -1 g dt e-t = -t 4e
23 2 t 24 e = 17 2t 24 e
e3t -e -6t 1 3 5 8t e e-6t - 8
23 24 = e 2t 17 24
23 1 -1 r 24 x ( t ) = c1e - t + c2e -6t + e2t 17 4 1 24

So, some of the work can be a little messy, but overall not too bad. We looked at two methods of solving nonhomogeneous differential equations here and while the work can be a little messy they arent too bad. Of course we also kept the nonhomogeneous part fairly simple here. More complicated problems will have significant amounts of work involved.
2007 Paul Dawkins
306
Laplace Transforms
Theres not too much to this section. Were just going to work an example to illustrate how Laplace transforms can be used to solve systems of differential equations.
Example 1 Solve the following system. = 3x1 - 3 x2 + 2 x1 = -6 x1 - t x2
x1 ( 0 ) = 1 x2 ( 0 ) = -1
Solution First notice that the system is not given in matrix form. This is because the system wont be solved in matrix form. Also note that the system is nonhomogeneous. We start just as we did when we used Laplace transforms to solve single differential equations. We take the transform of both differential equations.
sX 1 ( s ) - x1 ( 0 ) = 3 X 1 ( s ) - 3 X 2 ( s ) + sX 2 ( s ) - x2 ( 0 ) = -6 X 1 ( s ) 1 s2
2 s
Now plug in the initial condition and simplify things a little.
( s - 3) X 1 ( s ) + 3 X 2 ( s ) =
2 2+s +1 = s s 1 s2 + 1 6 X 1 ( s ) + sX 2 ( s ) = - 2 - 1 = - 2 s s
Now we need to solve this for one of the transforms. Well do this by multiplying the top equation by s and the bottom by -3 and then adding. This gives,
3s 2 + 3 ( s - 3s - 18) X1 ( s ) = 2 + s + s 2
2
Solving for X1 gives,
X1 ( s ) =
Partial fractioning gives,
s 3 + 5s 2 + 3 s 2 ( s + 3)( s - 6 )
X1 ( s ) =
1 133 28 3 18 + - 108 s - 6 s + 3 s s 2
Taking the inverse transform gives us the first solution,
x1 ( t ) =
1 (133e6t - 28e-3t + 3 - 18t ) 108
Now to find the second solution we could go back up and eliminate X1 to find the transform for X2 and sometimes we would need to do that. However, in this case notice that the second
differential equation is,
= -6 x1 - t x2
x2 = -6 x1 - t dt
So, plugging the first solution in and integrating gives,
x2 ( t ) = -
1 133e 6 t - 28e -3t + 3 dt 18 1 =133e 6t + 56e -3t + 18t ) + c ( 108 3 4
Now, reapplying the second initial condition to get the constant of integration gives
-1 = The second solution is then,
1 (133 + 56 ) + c 108
c=
x2 ( t ) = -
1 133e 6t + 56e -3t + 18t - 81) ( 108
So, putting all this together gives the solution to the system as,
1 (133e6t - 28e-3t + 3 - 18t ) 108 1 x2 ( t ) = (133e6t + 56e-3t + 18t - 81) 108 x1 ( t ) =

Compared to the last section the work here wasnt too bad. That wont always be the case of course, but you can see that using Laplace transforms to solve systems isnt too bad in at least some cases.
2007 Paul Dawkins
308
Modeling
In this section were going to go back and revisit the idea of modeling only this time were going to look at it in light of the fact that we now know how to solve systems of differential equations. Were not actually going to be solving any differential equations in this section. Instead well just be setting up a couple of problems that are extensions of some of the work that weve done in earlier modeling sections whether it is the first order modeling or the vibrations work we did in the second order chapter. Almost all of the systems that well be setting up here will be nonhomogeneous systems (which we only briefly looked at), will be nonlinear (which we didnt look at) and/or will involve systems with more than two differential equations (which we didnt look at, although most of what we do know will still be true). Mixing Problems Lets start things by looking at a mixing problem. The last time we saw these was back in the first order chapter. In those problems we had a tank of liquid with some type of contaminate dissolved in it. Liquid, possibly with more contaminate dissolved in it, entered the tank and liquid left the tank. In this situation we want to extend things out to the following situation.
Well now have two tanks that are interconnected with liquid potentially entering both and with an exit for some of the liquid if we need it (as illustrated by the lower connection). For this situation were going to make the following assumptions. 1. The inflow and outflow from each tank are equal, or in other words the volume in each tank is constant. When we worked with a single tank we didnt need to worry about this, but here if we dont well end up with a system with nonconstant coefficients and those can be quite difficult to solve. 2. The concentration of the contaminate in each tank is the same at each point in the tank. In reality we know that this wont be true but without this assumption wed need to deal with partial differential equations. 3. The concentration of contaminate in the outflow from tank 1 (the lower connection in the figure above) is the same as the concentration in tank 1. Likewise, the concentration of contaminate in the outflow from tank 2 (the upper connection) is the same as the concentration in tank 2.
2007 Paul Dawkins
309
4. The outflow from tank 1 is split and only some of the liquid exiting tank 1 actually reaches tank 2. The remainder exits the system completely. Note that if we dont want any liquid to completely exit the system we can think of the exit as having a value that is turned off. Also note that we could just as easily done the same thing for the outflow from tank 2 if wed wanted to. Lets take a look at a quick example.
Example 2 Two 1000 liter tanks are with salt water. Tank 1 contains 800 liters of water
initially containing 20 grams of salt dissolved in it and tank 2 contains 1000 liters of water and initially has 80 grams of salt dissolved in it. Salt water with a concentration of gram/liter of salt enters tank 1 at a rate of 4 liters/hour. Fresh water enters tank 2 at a rate of 7 liters/hour. Through a connecting pipe water flows from tank 2 into tank 1 at a rate of 10 liters/hour. Through a different connecting pipe 14 liters/hour flows out of tank 1 and 11 liters/hour are drained out of the pipe (and hence out of the system completely) and only 3 liters/hour flows back into tank 2. Set up the system that will give the amount of salt in each tank at any given time. Solution Okay, let Q1 ( t ) and Q2 ( t ) be the amount of salt in tank 1 and tank 2 at any time t respectively. Now all we need to do is set up a differential equation for each tank just as we did back when we had a single tank. The only difference is that we now need to deal with the fact that weve got a second inflow to each tank and the concentration of the second inflow will be the concentration of the other tank. Recall that the basic differential equation is the rate of change of salt ( Q ) equals the rate at which salt enters minus the rate at salt leaves. Each entering/leaving rate is found by multiplying the flow rate times the concentration. Here is the differential equation for tank 1.
1 Q Q Q1 = ( 4 ) + (10 ) 2 - (14 ) 1 2 1000 800 Q 7Q = 2+ 2 - 1 100 400
Q1 ( 0 ) = 20
In this differential equation the first pair of numbers is the salt entering from the external inflow. The second set of numbers is the salt that entering into the tank from the water flowing in from tank 2. The third set is the salt leaving tank as water flows out. Heres the second differential equation.
Q1 Q2 = ( 7 )( 0 ) + ( 3) Q2 - (10 ) 800 1000 3Q Q = 1- 2 800 100
Q2 ( 0 ) = 80
Note that because the external inflow into tank 2 is fresh water the concentration of salt in this is zero.
2007 Paul Dawkins
310
In summary here is the system wed need to solve,
Q2 7Q1 100 400 3Q Q = 1- 2 Q2 800 100 Q1 = 2 +
Q1 ( 0 ) = 20 Q2 ( 0 ) = 80
This is a nonhomogeneous system because of the first term in the first differential equation. If we had fresh water flowing into both of these we would in fact have a homogeneous system. Population The next type of problem to look at is the population problem. Back in the first order modeling section we looked at some population problems. In those problems we looked at a single population and often included some form of predation. The problem in that section was we assumed that the amount of predation would be constant. This however clearly wont be the case in most situations. The amount of predation will depend upon the population of the predators and the population of the predators will depend, as least partially, upon the population of the prey. So, in order to more accurately (well at least more accurate than what we originally did) we really need to set up a model that will cover both populations, both the predator and the prey. These types of problems are usually called predator-prey problems. Here are the assumptions that well make when we build up this model. 1. The prey will grow at a rate that is proportional to its current population if there are no predators. 2. The population of predators will decrease at a rate proportional to its current population if there is no prey. 3. The number of encounters between predator and prey will be proportional to the product of the populations. 4. Each encounter between the predator and prey will increase the population of the predator and decrease the population of the prey. So, given these assumptions lets write down the system for this case.
Example 3 Write down the system of differential equations for the population of predators and
prey using the assumptions above. Solution Well start off by letting x represent the population of the predators and y represent the population of the prey. Now, the first assumption tells us that, in the absence of predators, the prey will grow at a rate of ay where a > 0 . Likewise the second assumption tells us that, in the absence of prey, the predators will decrease at a rate of -bx where b > 0 . Next, the third and fourth assumptions tell us how the population is affected by encounters between predators and prey. So, with each encounter the population of the predators will increase
at a rate of a xy and the population of the prey will decrease at a rate of - b xy where a > 0 and b > 0 . Putting all of this together we arrive at the following system.
x = -bx + a xy = x (a y - b ) y = ay - b xy = y ( a - b x )
Note that this is a nonlinear system and weve not (nor will we here) discuss how to solve this kind of system. We simply wanted to give a better model for some population problems and to point out that not all systems will be nice and simple linear systems. Mechanical Vibrations When we first looked at mechanical vibrations we looked at a single mass hanging on a spring with the possibility of both a damper and/or an external force acting on the mass. Here we want to look at the following situation.
In the figure above we are assuming that the system is at rest. In other words all three springs are currently at their natural lengths and are not exerting any forces on either of the two masses and that there are no currently any external forces acting on either mass. We will use the following assumptions about this situation once we start the system in motion. 1.
x1 will measure the displacement of mass m1 from its equilibrium (i.e. resting) position and x2 will measure the displacement of mass m2 from its equilibrium position.
2. As noted in the figure above all displacement will be assumed to be positive if it is to the right of equilibrium position and negative if to the left of the equilibrium position. 3. All forces acting to the right are positive forces and all forces acting to the left are negative forces. 4. The spring constants, k1 , k2 , and k3 , are all positive and may or may not be the same value. 5. The surface that the system is sitting on is frictionless and so the mass of each of the objects will not affect the system in any way.
2007 Paul Dawkins
312
Before writing down the system for this case recall that the force exerted by the spring on the each mass is the spring constant times the amount that the spring has been compressed or stretched and well need to be careful with signs to make sure that the force is acting in the correct direction.
Example 4 Write down the system of differential equations for the spring and mass system
above. Solution To help us out lets first take a quick look at a situation in which both of the masses have been moved. This is shown below.
Before proceeding lets note that this is only a representation of a typical case, but most definitely not all possible cases. In this case were assuming that both x1 and x2 are positive and that x2 - x1 < 0 , or in other words, both masses have been moved to the right of their respective equilibrium points and that m1 has been moved farther than m2 . So, under these assumption on x1 and x2 we know that the spring on the left (with spring constant k1 ) has been stretched past its natural length while the middle spring (spring constant k2 ) and the right spring (spring constant k3 ) are both under compression. Also, weve shown the external forces, F1 ( t ) and F2 ( t ) , as present and acting in the positive direction. They do not, in practice, need to be present in every situation in which case we will assume that F1 ( t ) = 0 and/or F2 ( t ) = 0 . Likewise, if the forces are in fact acting in the negative direction we will then assume that F1 ( t ) < 0 and/or F2 ( t ) < 0 . Before proceeding we need to talk a little bit about how the middle spring will behave as the masses move. Here are all the possibilities that we can have and the affect each will have on x2 - x1 . Note that in each case the amount of compression/stretch in the spring is given by
x2 - x1 although we wont be using the absolute value bars when we set up the differential
equations. 1. If both mass move the same amount in the same direction then the middle spring will not have changed length and well have x2 - x1 = 0 .
2007 Paul Dawkins
313
2. If both masses move in the positive direction then the sign of x2 - x1 will tell us which has moved more. If m1 moves more than m2 then the spring will be in compression and
x2 - x1 < 0 . Likewise, if m2 moves more than m1 then the spring will have been stretched and x2 - x1 > 0 .
3. If both masses move in the negative direction well have pretty much the opposite behavior as #2. If m1 moves more than m2 then the spring will have been stretched and
x2 - x1 > 0 . Likewise, if m2 moves more than m1 then the spring will be in compression and x2 - x1 < 0 .
4. If m1 moves in the positive direction and m2 moves in the negative direction then the spring will be in compression and x2 - x1 < 0 . 5. Finally, if m1 moves in the negative direction and m2 moves in the positive direction then the spring will have been stretched and x2 - x1 > 0 . Now, well use the figure above to help us develop the differential equations (the figure corresponds to case 2 above) and then make sure that they will also hold for the other cases as well. Lets start off by getting the differential equation for the forces acting on m1 . Here is a quick sketch of the forces acting on m1 for the figure above.
In this case x1 > 0 and so the first spring has been stretched and so will exert a negative (i.e. to the left) force on the mass. The force from the first spring is then - k1 x1 and the - is needed because the force is negative but both k1 and x1 are positive. Next, because were assuming that m1 has moved more than m2 and both have moved in the positive direction we also know that x2 - x1 < 0 . Because m1 has moved more than m2 we know that the second spring will be under compression and so the force should be acting in the negative direction on m1 and so the force will be k2 ( x2 - x1 ) . Note that because k2 is positive and x2 - x1 is negative this force will have the correct sign (i.e. negative). The differential equation for m1 is then,
= - k1 x1 + k2 ( x2 - x1 ) + F1 ( t ) m1 x1
Note that this will also hold for all the other cases. If m1 has been moved in the negative
direction the force form the spring on the right that acts on the mass will be positive and - k1 x1 will be a positive quantity in this case. Next, if the middle is has been stretched (i.e. x2 - x1 > 0 ) then the force from this spring on m1 will be in the positive direction and k2 ( x2 - x1 ) will be a positive quantity in this case. Therefore, this differential equation holds for all cases not just the one we illustrated at the start of this problem. Lets now write down the differential equation for all the forces that are acting on m2 . Here is a sketch of the forces acting on this mass for the situation sketched out in the figure above.
In this case x2 is positive and so the spring on the right is under compression and will exert a negative force on m2 and so this force should be - k3 x2 , where the - is required because both
k3 and x2 are positive. Also, the middle spring is still under compression but the force that it exerts on this mass is now a positive force, unlike in the case of m1 , and so is given by
positive. The differential equation for m2 is then,
-k 2 ( x2 - x1 ) . The - on this force is required because x2 - x1 is negative and the force must be
= - k3 x2 - k2 ( x2 - x1 ) + F2 ( t ) m2 x2
Well leave it to you to verify that this differential equation does in fact hold for all the other cases. Putting all of this together and doing a little rewriting will then give the following system of differential equations for this situation.
= - ( k1 + k2 ) x1 + k2 x2 + F1 ( t ) m1 x1 = k2 x1 - ( k2 + k3 ) x2 + F2 ( t ) m2 x2
This is a system to two linear second order differential equations that may or may not be nonhomogeneous depending whether there are any external forces, F1 ( t ) and F2 ( t ) , acting on the masses. We have not talked about how to solve systems of second order differential equations. However, it can be converted to a system of first order differential equations as the next example shows and in many cases we could solve that.
2007 Paul Dawkins
315
Example 5 Convert the system from the previous example to a system of 1st order differential
equations. Solution This isnt too hard to do. Recall that we did this for single higher order differential equations earlier in the chapter when we first started to look at systems. To convert this to a system of first order differential equations we can make the following definitions.
u1 = x1
u2 = x1
u3 = x2
u4 = x 2
We can then convert each of the differential equations as we did earlier in the chapter.
= x1 = u2 u1 = x1 = u2 1 1 - ( k1 + k2 ) x1 + k2 x2 + F1 ( t ) ) = ( - ( k1 + k2 ) u1 + k2u3 + F1 ( t ) ) ( m1 m1 1 1 k2 x1 - ( k2 + k3 ) x2 + F2 ( t ) ) = ( ( k2u1 - ( k2 + k3 ) u3 + F2 ( t ) ) m2 m2
= x2 = u4 u3 = x2 = u4
Eliminating the middle step we get the following system of first order differential equations.
= u2 u1 = u2 1 ( - ( k1 + k2 ) u1 + k2u3 + F1 ( t ) ) m1 1 ( k2u1 - ( k2 + k3 ) u3 + F2 ( t ) ) m2
= u4 u3 = u4
The matrix form of this system would be,
0 - ( k1 + k 2 ) m1 r u = 0 k2 m2
1 0 0 0
0 k2 m1 0 - ( k 2 + k3 ) m2
0 0 F1 ( t ) 0 r m1 u + 1 0 F (t ) 2 0 m 2
u1 r u2 where, u = u3 u4
While we never discussed how to solve systems of more than two linear first order differential equations we know most of what we need to solve this. In an earlier section we discussed briefly solving nonhomogeneous systems and all of that information is still valid here.
For the homogenous system, that wed still need to solve for the general solution to the nonhomogeneous system, we know most of what we need to know in order to solve this. The only issues that we havent dealt with are what to do with repeated complex eigenvalues (which are now a possibility) and what to do with eigenvalues of multiplicity greater than 2 (which are again now a possibility).
2007 Paul Dawkins
317
Series Solutions to Differential Equations

Introduction
In this chapter we will finally be looking at nonconstant coefficient differential equations. While we wont cover all possibilities in this chapter we will be looking at two of the more common methods for dealing with this kind of differential equation. The first method that well be taking a look at, series solutions, will actually find a series representation for the solution instead of the solution itself. You first saw something like this when you looked at Taylor series in your Calculus class. As we will see however, these wont work for every differential equation. The second method that well look at will only work for a special class of differential equations. This special case will cover some of the cases in which series solutions cant be used. Here is a brief listing of the topics in this chapter. Review : Power Series A brief review of some of the basics of power series. Review : Taylor Series A reminder on how to construct the Taylor series for a function. Series Solutions In this section we will construct a series solution for a differential equation about an ordinary point. Euler Equations We will look at solutions to Eulers differential equation in this section.
2007 Paul Dawkins
318
Review : Power Series

Before looking at series solutions to a differential equation we will first need to do a cursory review of power series. A power series is a series in the form,
f ( x ) = an ( x - x0 )
n =0
(1)
where, x0 and an are numbers. We can see from this that a power series is a function of x. The function notation is not always included, but sometimes it is so we put it into the definition above. Before proceeding with our review we should probably first recall just what series really are. Recall that series are really just summations. One way to write our power series is then,
f ( x ) = an ( x - x0 )
n =0
(2)
2 3
= a0 + a1 ( x - x0 ) + a2 ( x - x0 ) + a3 ( x - x0 ) + L
Notice as well that if we needed to for some reason we could always write the power series as,
f ( x ) = an ( x - x0 )
n =0
= a0 + a1 ( x - x0 ) + a2 ( x - x0 ) + a3 ( x - x0 ) + L
2 3
= a0 + an ( x - x0 )
n =1
All that were doing here is noticing that if we ignore the first term (corresponding to n = 0 ) the remainder is just a series that starts at n = 1. When we do this we say that weve stripped out the n = 0 , or first, term. We dont need to stop at the first term either. If we strip out the first three terms well get,
an ( x - x0 ) = a0 + a1 ( x - x0 ) + a2 ( x - x0 ) + an ( x - x0 )
n 2 n= 0 n =3
There are times when well want to do this so make sure that you can do it. Now, since power series are functions of x and we know that not every series will in fact exist, it then makes sense to ask if a power series will exist for all x. This question is answered by looking at the convergence of the power series. We say that a power series converges for x = c if the series,
a (c - x )
n= 0 n 0 N N n =0
converges. Recall that this series will converge if the limit of partial sums,
lim an ( c - x0 )
exists and is finite. In other words, a power series will converge for x=c if
a (c - x )
n= 0 n 0
2007 Paul Dawkins
319
is a finite number. Note that a power series will always converge if x = x0. In this case the power series will become
a (x
n =0 n
- x0 ) = a0
n
With this we now know that power series are guaranteed to exist for at least one value of x. We have the following fact about the convergence of a power series. Fact Given a power series, (1), there will exist a number 0 r so that the power series will converge for x - x0 < r and diverge for x - x0 > r . This number is called the radius of convergence. Determining the radius of convergence for most power series is usually quite simple if we use the ratio test. Ratio Test Given a power series compute,
L = x - x0 lim
n
an +1 an
then,
L <1 L >1 L =1
the series converges the series diverges the series may converge or diverge
Lets take a quick look at how this can be used to determine the radius of convergence.
Example 1 Determine the radius of convergence for the following power series.
n 7 ( x - 5)
n= 0 n +1
( -3 )
Solution So, in this case we have,
an
( -3 ) =
n 7 n +1
( -3) an +1 = ( n + 1) 7 n + 2
n +1
Remember that to compute an+1 all we do is replace all the ns in an with n+1. Using the ratio test then gives,
2007 Paul Dawkins
320
L = x - 5 lim
n
an +1 an
n +1
-3) ( n 7 n+1 x - 5 lim n n ( n + 1) 7 n + 2 ( -3)

n
= x - 5 lim = 3 x -5 7
-3 n ( n + 1) 7 1
Now we know that the series will converge if,
3 x -5 <1 7
and the series will diverge if,
x -5 < x -5 >
7 3 7 3
3 x -5 >1 7
In other words, the radius of the convergence for this series is,
r=
7 3
As this last example has shown, the radius of convergence is found almost immediately upon using the ratio test. So, why are we worried about the convergence of power series? Well in order for a series solution to a differential equation to exist at a particular x it will need to be convergent at that x. If its not convergent at a given x then the series solution wont exist at that x. So, the convergence of power series is fairly important. Next we need to do a quick review of some of the basics of manipulating series. Well start with addition and subtraction. There really isnt a whole lot to addition and subtraction. All that we need to worry about is that the two series start at the same place and both have the same exponent of the x-x0. If they do then we can perform addition and/or subtraction as follows,
n = n0
an ( x - x0 ) bn ( x - x0 ) =
n n n = n0
n = n0
(a
bn )( x - x0 )
In other words all we do is add or subtract the coefficients and we get the new series. One of the rules that were going to have when we get around to finding series solutions to differential equations is that the only x that we want in a series is the x that sits in ( x - x0 ) .
n
This means that we will need to be able to deal with series of the form,
( x - x0 ) an ( x - x0 )n
c n =0
where c is some constant. These are actually quite easy to deal with.
( x - x0 ) an ( x - x0 )
c n =0
= ( x - x0 ) a0 + a1 ( x - x0 ) + a1 ( x - x0 ) + L
c 2
= a0 ( x - x0 ) + a1 ( x - x0 )
c
1+ c
+ a1 ( x - x0 )
2 +c
+L
a (x - x )
n= 0 n 0
n+ c
So, all we need to do is to multiply the term in front into the series and add exponents. Also note that in order to do this both the coefficient in front of the series and the term inside the series must be in the form x-x0. If they are not the same we cant do this, we will eventually see how to deal with terms that arent in this form. Next we need to talk about differentiation of a power series. By looking at (2) it should be fairly easy to see how we will differentiate a power series. Since a series is just a giant summation all we need to do is differentiate the individual terms. The derivative of a power series will be,
f ( x ) = a1 + 2a2 ( x - x0 ) + 3a3 ( x - x0 ) + L
2
= nan ( x - x0 )
n =1
n -1
= nan ( x - x0 )
n =0
n -1
So, all we need to do is just differentiate the term inside the series and were done. Notice as well that there are in fact two forms of the derivative. Since the n=0 term of the derivative is zero it wont change the value of the series and so we can include it or not as we need to. In our work we will usually want the derivative to start at n=1, however there will be the occasion problem were it would be more convenient to start it at n=0. Following how we found the first derivative it should make sense that the second derivative is,
f ( x ) = n ( n - 1) an ( x - x0 )
n= 2
n -2
= n ( n - 1) an ( x - x0 )
n =1
n -2
= n ( n - 1) an ( x - x0 )
n =0
n -2
In this case since the n=0 and n=1 terms are both zero we can start at any of three possible starting points as determined by the problem that were working.
2007 Paul Dawkins
322
Next we need to talk about index shifts. As we will see eventually we are going to want our power series written in terms of ( x - x0 ) and they often wont, initially at least, be in that form.
n
To get them into the form we need we will need to perform an index shift. Index shifts themselves really arent concerned with the exponent on the x term, they instead are concerned with where the series starts as the following example shows.
Example 2 Write the following as a series that starts at n=0 instead of n=3.
n a ( x + 4)
2 n =3 n -1
n+ 2
Solution An index shift is a fairly simple manipulation to perform. First we will notice that if we define i=n-3 then when n=3 we will have i=0. So what well do is rewrite the series in terms of i instead of n. We can do this by noting that n=i+3. So, everywhere we see an n in the actual series term we will replace it with an i+3. Doing this gives,
n a ( x + 4)
2 n =3 n -1
n+ 2
= ( i + 3) ai +3-1 ( x + 4 )
2 i =0
i + 3+ 2
= ( i + 3) ai + 2 ( x + 4 )
2 i =0
i +5
The upper limit wont change in this process since infinity minus three is still infinity. The final step is to realize that the letter we use for the index doesnt matter and so we can just switch back to ns.
n a ( x + 4)
2 n =3 n -1
n+ 2
= ( n + 3) an + 2 ( x + 4 )
2 n =0
n +5
Now, we usually dont go through this process to do an index shift. All we do is notice that we dropped the starting point in the series by 3 and everywhere else we saw an n in the series we increased it by 3. In other words, all the ns in the series move in the opposite direction that we moved the starting point.
Example 3 Write the following as a series that starts at n=5 instead of n=3.
n a ( x + 4)
2 n =3 n -1
n+ 2
Solution To start the series to start at n=5 all we need to do is notice that this means we will increase the starting point by 2 and so all the other ns will need to decrease by 2. Doing this for the series in the previous example would give,
n2 an-1 ( x + 4 )
n =3 n
n+ 2
= ( n - 2 ) an -3 ( x + 4 )
2 n =5
Now, as we noted when we started this discussion about index shift the whole point is to get our series into terms of ( x - x0 ) . We can see in the previous example that we did exactly that with
an index shift. The original exponent on the (x+4) was n+2. To get this down to an n we needed to decrease the exponent by 2. This can be done with an index that increases the starting point by 2. Lets take a look at a couple of more examples of this.
Example 4 Write each of the following as a single series in terms of ( x - x0 ) .

n
(a) ( x + 2 )
nan ( x + 2)
n =3 2
n- 4
- nan ( x + 2 )
n =1
n +1
[Solution]
(b) x Solution (a)
( n - 5)
n =0
bn +1 ( x - 3)
n +3
[Solution]
( x + 2)
nan ( x + 2)
n =3
n- 4
- nan ( x + 2 )
n =1
n +1
First, notice that there are two series here and the instructions clearly ask for only a single series. So, we will need to subtract the two series at some point in time. The vast majority of our work will be to get the two series prepared for the subtraction. This means that the two series cant have any coefficients in front of them (other than one of course), they will need to start at the same value of n and they will need the same exponent on the x-x0. Well almost always want to take care of any coefficients first. So, we have one in front of the first series so lets multiply that into the first series. Doing this gives,
nan ( x + 2 )
n =3
n- 2
- nan ( x + 2 )
n =1
n +1
Now, the instructions specify that the new series must be in terms of ( x - x0 ) , so thats the next
n
thing that weve got to take care of. We will do this by an index shift on each of the series. The exponent on the first series needs to go up by two so well shift the first series down by 2. On the second series will need to shift up by 1 to get the exponent to move down by 1. Performing the index shifts gives us the following,
n n ( n + 2) an+2 ( x + 2 ) - ( n - 1) an-1 ( x + 2 ) n =1 n= 2
Finally, in order to subtract the two series well need to get them to start at the same value of n. Depending on the series in the problem we can do this in a variety of ways. In this case lets notice that since there is an n-1 in the second series we can in fact start the second series at n=1 without changing its value. Also note that in doing so we will get both of the series to start at n=1 and so we can do the subtraction. Our final answer is then,
( n + 2) a ( x + 2 ) - ( n - 1) a ( x + 2 ) = ( n + 2 ) a
n n n =1 n+ 2 n =1 n -1 n =1
n+ 2
- ( n - 1) an-1 ( x + 2)
2007 Paul Dawkins
324
(b) x
( n - 5)
n =0
bn +1 ( x - 3)
n +3
In this part the main issue is the fact that we cant just multiply the coefficient into the series this time since the coefficient doesnt have the same form as the term inside the series. Therefore, the first thing that well need to do is correct the coefficient so that we can bring it into the series. We do this as follows,
x ( n - 5 ) bn +1 ( x - 3)
2 n =0
n +3
= ( x - 3 + 3) ( n - 5) bn +1 ( x - 3)
2 n= 0
n +3
= ( x - 3 ) ( n - 5 ) bn +1 ( x - 3)
2 n =0
n +3
+ 3 ( n - 5) bn +1 ( x - 3)
2 n= 0
n +3
We can now move the coefficient into the series, but in the process of we managed to pick up a second series. This will happen so get used to it. Moving the coefficients of both series in gives,
( n - 5) bn +1 ( x - 3)
2 n= 0
n+4
+ 3 ( n - 5 ) bn +1 ( x - 3)
2 n =0
n+ 3
We now need to get the exponent in both series to be an n. This will mean shifting the first series up by 4 and the second series up by 3. Doing this gives,
( n - 9) bn-3 ( x - 3) + 3 ( n - 8) bn-2 ( x - 3)
2 n 2 n= 4 n =3
In this case we cant just start the first series at n=3 because there is not an n-3 sitting in that series to make the n=3 term zero. So, we wont be able to do this part as we did in the first part of this example. What well need to do in this part is strip out the n=3 from the second series so they will both start at n=4. We will then be able to add the two series together. Stripping out the n=3 term from the second series gives,
2 n 2 3 2 n ( n - 9 ) bn-3 ( x - 3) + 3 ( -5) b1 ( x - 3) + 3 ( n - 8) bn-2 ( x - 3) n= 4 n= 4
We can now add the two series together.

3 2 2 n 75b1 ( x - 3) + ( n - 9 ) bn-3 + 3 ( n - 8 ) bn- 2 ( x - 3) n= 4
This is what were looking for. We wont worry about the extra term sitting in front of the series. When we finally get around to finding series solutions to differential equations we will see how to deal with that term there.
There is one final fact that we need take care of before moving on. Before giving this fact for power series lets notice that the only way for
a + bx + cx 2 = 0
to be zero for all x is to have a=b=c=0.
Weve got a similar fact for power series. Fact If,
a (x - x )
n= 0 n 0
=0
for all x then,
an = 0, n = 0,1, 2, K
This fact will be key to our work with differential equations so dont forget it.
2007 Paul Dawkins
326
Review : Taylor Series

We are not going to be doing a whole lot with Taylor series once we get out of the review, but they are a nice way to get us back into the swing of dealing with power series. By time most students reach this stage in their mathematical career theyve not had to deal with power series for at least a semester or two. Remembering how Taylor series work will be a very convenient way to get comfortable with power series before we start looking at differential equations. Taylor Series If f(x) is an infinitely differential function then the Taylor Series of f(x) about x=x0 is,
f ( x) =
n =0
f(
n)
( x0 )
n!
( x - x0 )
Recall that
f (0) ( x ) = f ( x )
f ( n ) ( x ) = n th derivative of f ( x )
Lets take a look at an example.
Example 1 Determine the Taylor series for f ( x ) = e x about x=0.

Solution This is probably one of the easiest functions to find the Taylor series for. We just need to recall that,
f ( n) ( x ) = e x
n = 0,1, 2,K n = 0,1, 2,K
and so we get,
f ( n) ( 0 ) = 1
The Taylor series for this example is then,
xn e = n =0 n !
x
Of course, its often easier to find the Taylor series about x=0 but we dont always do that.
Example 2 Determine the Taylor series for f ( x ) = e x about x=-4.

Solution This problem is virtually identical to the previous problem. In this case we just need to notice that,
f ( n ) ( -4 ) = e -4 n = 0,1, 2,K e -4 n ( x + 4) n =0 n !
The Taylor series for this example is then,
ex =
Lets now do a Taylor series that requires a little more work.
Example 3 Determine the Taylor series for f ( x ) = cos ( x ) about x=0.

Solution This time there is no formula that will give us the derivative for each n so lets start taking derivatives and plugging in x=0.
f ( 0 ) ( x ) = cos ( x )
f ( 0) ( 0 ) = 1 f (1) ( 0 ) = 0
f (1) ( x ) = - sin ( x ) f ( 2) ( x ) = - cos ( x ) f ( 3) ( x ) = sin ( x ) f ( 4) ( x ) = cos ( x ) M
f ( 2) ( 0 ) = -1 f ( 3) ( 0 ) = 0 f ( 4) ( 0 ) = 1 M
Once we reach this point its fairly clear that there is a pattern emerging here. Just what this pattern is has yet to be determined, but it does seem fairly clear that a pattern does exist. Lets plug what weve got into the formula for the Taylor series and see what we get.
cos ( x ) =
n =0
f (n) ( 0) n x n!
f ( ) ( 0) f ( ) ( 0) 2 f ( ) (0 ) 3 x+ x + x +L 0! 1! 2! 3! 1 x2 x4 x6 x8 = + 0 - + 0 + + 0 - + 0 + +L 0! 2! 4! 6! 8! = f(
0) 1 2 3
( 0) +
So, every other term is zero. We would like to write this in terms of a series, however finding a formula that is zero every other term and gives the correct answer for those that arent zero would be unnecessarily complicated. So, lets rewrite what weve got above and while were at it renumber the terms as follows,
cos ( x ) =
1 x 2 x4 x 6 x8 - + - + +L 0! { 2! { 4! { 6! { 8! {
n= 0 n =1 n =2 n =3 n =4
With this renumbering we can fairly easily get a formula for the Taylor series of the cosine function about x=0.
( -1) x 2n cos ( x ) = ( 2n ) ! n =0
n
For practice you might want to see if you can verify that the Taylor series for the sine function about x=0 is,
-1) x 2 n +1 ( sin ( x ) = n = 0 ( 2 n + 1) !
n
We need to look at one more example of a Taylor series. This example is both tricky and very easy.
Example 4 Determine the Taylor series for f ( x ) = 3 x 2 - 8 x + 2 about x=2.

Solution Theres not much to do here except to take some derivatives and evaluate at the point.
f ( x ) = 3x2 - 8x + 2
f ( 2 ) = -2
f ( x) = 6x - 8 f ( x ) = 6 f ( n) ( x ) = 0, n 3
f ( 2) = 4 f ( 2 ) = 6 f ( n) ( 2 ) = 0, n 3
So, in this case the derivatives will all be zero after a certain order. That happens occasionally and will make our work easier. Setting up the Taylor series then gives,
3x2 - 8x + 2 =
n =0
f ( n) ( 2 ) n ( x - 2) n!
f ( 0) ( 2 ) f (1) ( 2 ) f ( 2) ( 2 ) f ( 3) ( 2 ) 2 3 = + ( x - 2) + ( x - 2) + ( x - 2) + L 0! 1! 2! 3! 6 2 = -2 + 4 ( x - 2 ) + ( x - 2 ) + 0 2 = -2 + 4 ( x - 2 ) + 3 ( x - 2 )
2
In this case the Taylor series terminates and only had three terms. Note that since we are after the Taylor series we do not multiply the 4 through on the second term or square out the third term. All the terms with the exception of the constant should contain an x-2. Note in this last example that if we were to multiply the Taylor series we would get our original polynomial. This should not be too surprising as both are polynomials and they should be equal. We now need a quick definition that will make more sense to give here rather than in the next section were we actually need it since it deals with Taylor series. Definition A function, f(x), is called analytic at x=a if the Taylor series for f(x) about x=a has a positive radius of convergence and converges to f(x). We need to give one final note before proceeding into the next section. We started this section out by saying that we werent going to be doing much with Taylor series after this section. While that is correct it is only correct because we are going to be keeping the problems fairly simple. For more complicated problems we would also be using quite a few Taylor series.
2007 Paul Dawkins
329
Series Solutions to Differential Equations

Before we get into finding series solutions to differential equations we need to determine when we can find series solutions to differential equations. So, lets start with the differential equation,
p ( x ) y + q ( x ) y + r ( x ) y = 0
(1)
This time we really do mean nonconstant coefficients. To this point weve only dealt with constant coefficients. However, with series solutions we can now have nonconstant coefficient differential equations. Also, in order to make the problems a little nicer we will be dealing only with polynomial coefficients. Now, we say that x=x0 is an ordinary point if provided both
q ( x) p ( x)
and
r ( x) p ( x)
are analytic at x=x0. That is to say that these two quantities have Taylor series around x=x0. We are going to be only dealing with coefficients that are polynomials so this will be equivalent to saying that
p ( x0 ) 0
for most of the problems. If a point is not an ordinary point we call it a singular point. The basic idea to finding a series solution to a differential equation is to assume that we can write the solution as a power series in the form,
y ( x ) = an ( x - x0 )
n =0
(2)
and then try to determine what the ans need to be. We will only be able to do this if the point x=x0, is an ordinary point. We will usually say that (2) is a series solution around x=x0. Lets start with a very basic example of this. In fact it will be so basic that we will have constant coefficients. This will allow us to check that we get the correct solution.
Example 1 Determine a series solution for the following differential equation about x0 = 0 . y + y = 0
Solution Notice that in this case p(x)=1 and so every point is an ordinary point. We will be looking for a solution in the form,
y ( x ) = an x n
n =0
We will need to plug this into our differential equation so well need to find a couple of derivatives.
y ( x ) = nan x n-1
n =1
y ( x ) = n ( n - 1) an x n -2
n= 2
2007 Paul Dawkins
330
Recall from the power series review section on power series that we can start these at n=0 if we need to, however its almost always best to start them where we have here. If it turns out that it would have been easier to start them at n=0 we can easily fix that up when the time comes around. So, plug these into our differential equation. Doing this gives,

n ( n - 1) a x
n= 2 n
n -2
+ an x n = 0
n= 0
The next step is to combine everything into a single series. To do this requires that we get both series starting at the same point and that the exponent on the x be the same in both series. We will always start this by getting the exponent on the x to be the same. It is usually best to get the exponent to be an n. The second series already has the proper exponent and the first series will need to be shifted down by 2 in order to get the exponent up to an n. If you dont recall how to do this take a quick look at the first review section where we did several of these types of problems. Shifting the first power series gives us,

( n + 2 )( n + 1) an+2 x n + an xn = 0
n= 0 n =0
Notice that in the process of the shift we also got both series starting at the same place. This wont always happen, but when it does well take it. We can now add up the two series. This gives,
( n + 2 )( n + 1) a
n= 0
n+ 2
n + an x =0
Now recalling the fact from the power series review section we know that if we have a power series that is zero for all x (as this is) then all the coefficients must have been zero to start with. This gives us the following,
( n + 2 )( n + 1) an+ 2 + an = 0,
n = 0,1, 2,K
This is called the recurrence relation and notice that we included the values of n for which it must be true. We will always want to include the values of n for which the recurrence relation is true since they wont always start at n = 0 as it did in this case. Now lets recall what we were after in the first place. We wanted to find a series solution to the differential equation. In order to do this we needed to determine the values of the ans. We are almost to the point where we can do that. The recurrence relation has two different ans in it so we cant just solve this for an and get a formula that will work for all n. We can however, use this to determine what all but two of the ans are. To do this we first solve the recurrence relation for the an that has the largest subscript. Doing this gives,
an + 2 = 2007 Paul Dawkins
an ( n + 2 )( n + 1)
331
n = 0,1, 2,K
Now, at this point we just need to start plugging in some value of n and see what happens, n=0
a2 =
- a0 ( 2 )(1) a2 ( 4 )( 3)
n=1
a3 =
- a1 ( 3)( 2 )
a4 = n=2
a5 = n=3
( 5 )( 4 )
a1
a3
a0 ( 4 )( 3)( 2 )(1) a4 ( 6 )( 5 )
( 5)( 4 )( 3)( 2 ) ( 7 )( 6 )
a5 - a1 ( 7 )( 6 )( 5 )( 4 )( 3)( 2 )
k
a6 = n=4
a7 = n=5
- a0 ( 6 )( 5)( 4 )( 3)( 2 )(1)

k
M ( -1) a0 , k = 1, 2,K a2 k = ( 2k ) !
M ( -1) a1 , k = 1, 2,K a2 k +1 = ( 2k + 1)!
Notice that at each step we always plugged back in the previous answer so that when the subscript was even we could always write the an in terms of a0 and when the coefficient was odd we could always write the an in terms of a1. Also notice that, in this case, we were able to find a general formula for ans with even coefficients and ans with odd coefficients. This wont always be possible to do. Theres one more thing to notice here. The formulas that we developed were only for k=1,2, however, in this case again, the will also work for k=0. Again, this is something that wont always work, but does here. Do not get excited about the fact that we dont know what a0 and a1 are. As you will see, we actually need these to be in the problem to get the correct solution. Now that weve got formulas for the ans lets get a solution. The first thing that well do is write out the solution with a couple of the ans plugged in.
y ( x ) = an x n
n =0
= a0 + a1 x + a2 x 2 + a3 x 3 + L + a2 k x 2k + a2 k +1 x 2 k +1 + L
k
( -1) a0 x 2k + ( -1) a1 x 2 k +1 + L a a = a0 + a1 x - 0 x 2 - 1 x 3 + L + 2! 3! ( 2k ) ! ( 2k + 1)!

k +1
The next step is to collect all the terms with the same coefficient in them and then factor out that coefficient.
2007 Paul Dawkins
332
k k +1 -1) x 2k -1) ( ( x3 x2 y ( x ) = a0 1 - L + x 2 k +1 + L + L + a1 x - + L + 3! ( 2k ) ! ( 2k + 1)! 2!
( -1) x 2k = a0 ( 2k )! k =0
k
( -1) x 2k +1 + a1 k = 0 ( 2 k + 1) !
k
In the last step we also used the fact that we knew what the general formula was to write both portions as a power series. This is also our solution. We are done. Before working another problem lets take a look at the solution to the previous example. First, we started out by saying that we wanted a series solution of the form,
y ( x ) = an x n
n =0
and we didnt get that. We got a solution that contained two different power series. Also, each of the solutions had an unknown constant in them. This is not a problem. In fact, its what we want to have happen. From our work with second order constant coefficient differential equations we know that the solution to the differential equation in the last example is,
y ( x ) = c1 cos ( x ) + c2 sin ( x )
Solutions to second order differential equations consist of two separate functions each with an unknown constant in front of them that are found by applying any initial conditions. So, the form of our solution in the last example is exactly what we want to get. Also recall that the following Taylor series,
-1) x 2n ( cos ( x ) = ( 2n ) ! n =0
n
-1) x 2 n+1 ( sin ( x ) = n = 0 ( 2 n + 1) !

n
Recalling these we very quickly see that what we got from the series solution method was exactly the solution we got from first principles, with the exception that the functions were the Taylor series for the actual functions instead of the actual functions themselves. Now lets work an example with nonconstant coefficients since that is where series solutions are most useful.
Example 2 Find a series solution around x0 = 0 for the following differential equation. y - xy = 0
Solution As with the first example p(x)=1 and so again for this differential equation every point is an ordinary point. Now well start this one out just as we did the first example. Lets write down the form of the solution and get its derivatives.
y ( x ) = an x n
n =0
y ( x ) = nan x n-1
n =1
y ( x ) = n ( n - 1) an x n -2
n= 2
Plugging into the differential equation gives,
2007 Paul Dawkins
333
n ( n - 1) a x
n= 2 n
n -2
- x an x n = 0
n =0
Unlike the first example we first need to get all the coefficients moved into the series.
n ( n - 1) an xn -2 - an x n+1 = 0
n= 2 n =0
Now we will need to shift the first series down by 2 and the second series up by 1 to get both of the series in terms of xn.
( n + 2)( n + 1) an+2 x n - an-1 xn = 0

n= 0 n =1
Next we need to get the two series starting at the same value of n. The only way to do that for this problem is to strip out the n=0 term.
( 2 )(1) a2 x 0 + ( n + 2 )( n + 1) an+ 2 xn - an-1 xn = 0

n =1 n =1 n 2 a2 + ( n + 2 )( n + 1) an + 2 - an -1 x =0 n =1
We now need to set all the coefficients equal to zero. We will need to be careful with this however. The n=0 coefficient is in front of the series and the n=1,2,3 are all in the series. So, setting coefficient equal to zero gives,
n = 0: n = 1, 2,3,K n = 0:
( n + 2 )( n + 1) an + 2 - an -1 = 0
a2 = 0 an + 2 = an-1 ( n + 2 )( n + 1)
2a2 = 0
Solving the first as well as the recurrence relation gives,
n = 1, 2,3,K
Now we need to start plugging in values of n.
a3 = a6 = =
( 3)( 2 )
a3 ( 6 )( 5 ) a0 ( 6 )( 5 )( 3)( 2 ) M
a0
a4 = a7 = =
( 4 )( 3)
( 7 )( 6 ) ( 7 )( 6 )( 4 )( 3)
M a1 a4
a1
a5 =
( 5 )( 4 )
a5
a2
=0
a8 =
( 8)( 7 )
M
=0
2007 Paul Dawkins
334
a0 a1 a3k +1 = ( 2 )( 3)( 5)( 6 )L ( 3k - 1)( 3k ) ( 3)( 4 )( 6 )( 7 )L ( 3k )( 3k + 1) a3k + 2 = 0 k = 0,1, 2,L k = 1, 2,3,L k = 1, 2,3,L a3k =
There are a couple of things to note about these coefficients. First, every third coefficient is zero. Next, the formulas here are somewhat unpleasant and not all that easy to see the first time around. Finally, these formulas will not work for k=0 unlike the first example. Now, get the solution,
y ( x ) = a0 + a1 x + a2 x 2 + a3 x 3 + a4 x 4 + L + a3k x 3k + a3 k +1 x 3k +1 + L a0 3 a1 4 a0 x3 k = a0 + a1 x + x + x L + + 6 12 ( 2 )( 3)( 5 )( 6 )L ( 3k - 1)( 3k ) a1 x 3 k +1 +L ( 3)( 4 )( 6 )( 7 )L ( 3k )( 3k + 1)
Again, collect up the terms that contain the same coefficient, factor the coefficient out and write the results as a new series,
x 3k x3 k +1 y ( x ) = a0 1 + + a x + 1 k =1 ( 3 )( 4 )( 6 )( 7 )L ( 3k )( 3k + 1) k =1 ( 2 )( 3)( 5 )( 6 )L ( 3k - 1)( 3k )
We couldnt start our series at k=0 this time since the general term doesnt hold for k=0. Now, we need to work an example in which we use a point other that x=0. In fact, lets just take the previous example and rework it for a different value of x0. Were also going to need to change up the instructions a little for this example.
Example 3 Find the first four terms in each portion of the series solution around x0 = -2 for
the following differential equation.
y - xy = 0
Solution Unfortunately for us there is nothing from the first example that can be reused here. Changing to x0 = -2 completely changes the problem. In this case our solution will be,
y ( x ) = an ( x + 2 )
n= 0
The derivatives of the solution are,
y ( x ) = nan ( x + 2 )
n =1
n -1
y ( x ) = n ( n - 1) an ( x + 2 )
n =2
n- 2
Plug these into the differential equation.

n ( n - 1) a ( x + 2 )
n= 2 n
n- 2
- x an ( x + 2 ) = 0
n n =0
We now run into our first real difference between this example and the previous example. In this case we cant just multiply the x into the second series since in order to combine with the series it must be x+2. Therefore we will first need to modify the coefficient of the second series before multiplying it into the series.
n ( n - 1) an ( x + 2)
n= 2
n- 2
- ( x + 2 - 2 ) an ( x + 2 ) = 0
n n =0 n n
n ( n - 1) a ( x + 2 )
n= 2 n n= 2 n
n- 2
- ( x + 2 ) an ( x + 2 ) + 2 an ( x + 2 ) = 0
n =0 n =0 n -2
n ( n - 1) a ( x + 2 )
- an ( x + 2 )
n =0
n +1
+ 2an ( x + 2 ) = 0
n n =0
We now have three series to work with. This will often occur in these kinds of problems. Now we will need to shift the first series down by 2 and the second series up by 1 the get common exponents in all the series.
( n + 2)( n + 1) a ( x + 2) - a ( x + 2 ) + 2a ( x + 2)
n n n= 0 n+ 2 n =1 n -1 n =0 n
=0
In order to combine the series we will need to strip out the n=0 terms from both the first and third series.
2a2 + 2a0 + ( n + 2 )( n + 1) an + 2 ( x + 2 ) - an -1 ( x + 2 ) + 2an ( x + 2 ) = 0

n n n n =1 n =1 n =1
2a2 + 2a0 + ( n + 2 )( n + 1) an + 2 - an -1 + 2an ( x + 2) = 0

n n =1
Setting coefficients equal to zero gives,
n=0 n = 1, 2,3,K
( n + 2 )( n + 1) an+ 2 - an-1 + 2an = 0
2a2 + 2a0 = 0
We now need to solve both of these. In the first case there are two options, we can solve for a2 or we can solve for a0. Out of habit Ill solve for a0. In the recurrence relation well solve for the term with the largest subscript as in previous examples.
n=0 n = 1, 2,3,K
a2 = - a0 an + 2 = an -1 - 2an ( n + 2 )( n + 1)
Notice that in this example we wont be having every third term drop out as we did in the previous example. At this point well also acknowledge that the instructions for this problem are different as well. We arent going to get a general formula for the ans this time so well have to be satisfied with just getting the first couple of terms for each portion of the solution. This is often the case for
series solutions. Getting general formulas for the ans is the exception rather than the rule in these kinds of problems. To get the first four terms well just start plugging in terms until weve got the required number of terms. Note that we will already be starting with an a0 and an a1 from the first two terms of the solution so all we will need are three more terms with an a0 in them and three more terms with an a1 in them.
n=0
a2 = - a0
Weve got two a0s and one a1.
n =1
Weve got three a0s and two a1s.
a3 =
a0 - 2a1 a0 a1 = ( 3)( 2 ) 6 3
n=2
a4 =
a1 - 2a2 a1 - 2 ( - a0 ) a0 a1 = = + 6 12 ( 4 )( 3) ( 4 )( 3)
Weve got four a0s and three a1s. Weve got all the a0s that we need, but we still need one more a1. So, well need to do one more term it looks like.
n=3
a5 =
a2 - 2a3 a 31a0 a1 1 a a =- 0 - 0 - 1=+ 20 10 6 3 60 30 ( 5 )( 4 )
Weve got five a0s and four a1s. Weve got all the terms that we need. Now, all that we need to do is plug into our solution.
y ( x ) = an ( x + 2 )
n =0
= a0 + a1 ( x + 2 ) + a2 ( x + 2 ) + a3 ( x + 2 ) + a4 ( x + 2 ) + a5 ( x + 2 ) + L
2 3 4 5 2 3 a a = a0 + a1 ( x + 2 ) - a0 ( x + 2 ) + 0 - 1 ( x + 2 ) + 6 3 4 5 a0 a1 31a0 a1 6 + 12 ( x + 2 ) + - 60 + 30 ( x + 2 ) + L
Finally collect all the terms up with the same coefficient and factor out the coefficient to get,
1 1 31 2 3 4 5 y ( x ) = a0 1 - ( x + 2 ) + ( x + 2 ) + ( x + 2 ) - ( x + 2 ) + L + 6 6 60 1 1 1 3 4 5 a1 ( x + 2 ) - ( x + 2 ) + ( x + 2 ) + ( x + 2 ) + L 3 12 30
Thats the solution for this problem as far as were concerned. Notice that this solution looks nothing like the solution to the previous example. Its the same differential equation, but changing x0 completely changed the solution. Lets work one final problem.
Example 4 Find the first four terms in each portion of the series solution around x0 = 0 for the
following differential equation.
(x
+ 1) y - 4 xy + 6 y = 0
Solution We finally have a differential equation that doesnt have a constant coefficient for the second derivative.
p ( x ) = x2 +1
p (0) = 1 0
So x0 = 0 is an ordinary point for this differential equation. We first need the solution and its derivatives,
y ( x ) = an x n
n =0
y ( x ) = nan x n-1
n =1
y ( x ) = n ( n - 1) an x n -2
n= 2
Plug these into the differential equation.
(x
+ 1) n ( n - 1) an x n- 2 - 4 x nan x n-1 + 6 an x n = 0
n =2 n =1 n =0
Now, break up the first term into two so we can multiply the coefficient into the series and multiply the coefficients of the second and third series in as well.
n ( n - 1) an xn + n ( n - 1) an xn-2 - 4nan x n + 6an x n = 0

n= 2 n= 2 n =1 n= 0
We will only need to shift the second series down by two to get all the exponents the same in all the series.
n ( n - 1) an xn + ( n + 2)( n + 1) an+2 x n - 4nan x n + 6an xn = 0

n= 2 n= 0 n =1 n =0
At this point we could strip out some terms to get all the series starting at n=2, but thats actually more work than is needed. Lets instead note that we could start the third series at n=0 if we wanted to because that term is just zero. Likewise the terms in the first series are zero for both n=1 and n=0 and so we could start that series at n=0. If we do this all the series will now start at n=0 and we can add them up without stripping terms out of any series.
n ( n - 1) a + ( n + 2 )( n + 1) a
n= 0 n
n+2
n - 4nan + 6an x =0
( n
n =0 n =0
n - 5n + 6 ) an + ( n + 2 )( n + 1) an + 2 x =0 n x =0
( n - 2 )( n - 3) a + ( n + 2 )( n + 1) a
n
n+2
Now set coefficients equal to zero.
( n - 2 )( n - 3) an + ( n + 2 )( n + 1) an+ 2 ,
338
n = 0,1, 2,K
2007 Paul Dawkins
Solving this gives,
an + 2 = Now, we plug in values of n.
( n - 2 )( n - 3) an , ( n + 2 )( n + 1)
n = 0: n = 1:
n = 0,1, 2,K
n = 2: n = 3:
a2 = -3a0 1 a3 = - a1 3 0 a4 = - a2 = 0 12 0 a5 = - a3 = 0 20
Now, from this point on all the coefficients are zero. In this case both of the series in the solution will terminate. This wont always happen, and often only one of them will terminate. The solution in this case is,
1 y ( x ) = a0 {1 - 3 x 2 } + a1 x - x 3 3
2007 Paul Dawkins
339
Euler Equations
In this section we want to look for solutions to
ax 2 y + bxy + cy = 0
around x0 = 0 . These type of differential equations are called Euler Equations. Recall from the previous section that a point is an ordinary point if the quotients,
(1)
bx b c = and 2 ax ax ax 2 have Taylor series around x0 = 0 . However, because of the x in the denominator neither of these will have a Taylor series around x0 = 0 and so x0 = 0 is a singular point. So, the method from
the previous section wont work since it required an ordinary point. However, it is possible to get solutions to this differential equation that arent series solutions. Lets start off by assuming that x>0 (the reason for this will be apparent after we work the first example) and that all solutions are of the form,
y ( x ) = xr
(2)
Now plug this into the differential equation to get,
ax 2 ( r )( r - 1) x r -2 + bx ( r ) x r -1 + cx r = 0 ar ( r - 1) x r + b ( r ) x r + cx r = 0
( ar ( r - 1) + b ( r ) + c ) x
Now, we assumed that x>0 and so this will only be zero if,
=0
ar ( r - 1) + b ( r ) + c = 0
(3)
So solutions will be of the form (2) provided r is a solution to (3). This equation is a quadratic in r and so we will have three cases to look at : Real, Distinct Roots, Double Roots, and Quadratic Roots. Real, Distinct Roots There really isnt a whole lot to do in this case. Well get two solutions that will form a fundamental set of solutions (well leave it to you to check this) and so our general solution will be,
y ( x ) = c1 x r1 + c2 x r2
Example 1 Solve the following IVP 2 x 2 y + 3 xy - 15 y = 0,

Solution We first need to find the roots to (3).
y (1) = 0 y (1) = 1
2r ( r - 1) + 3r - 15 = 0
2r 2 + r - 15 = ( 2r - 5)( r + 3) = 0
5 r1 = , r2 = -3 2
y ( x ) = c1 x 2 + c2 x -3
5
To find the constants we differentiate and plug in the initial conditions as we did back in the second order differential equations chapter.
5 3 y ( x ) = c1 x 2 - 3c2 x -4 2 0 = y (1) = c1 + c2 2 2 c1 = , c2 = 5 11 11 1 = y (1) = c1 - 3c2 2

y ( x) =
2 5 2 x 2 - x -3 11 11
With the solution to this example we can now see why we required x>0. The second term would have division by zero if we allowed x=0 and the first term would give us square roots of negative numbers if we allowed x<0. Double Roots This case will lead to the same problem that weve had every other time weve run into double roots (or double eigenvalues). We only get a single solution and will need a second solution. In this case it can be shown that the second solution will be,
y2 ( x ) = x r ln x
and so the general solution in this case is,
y ( x ) = c1 x r + c2 x r ln x = x r ( c1 + c2 ln x )
We can again see a reason for requiring x>0. If we didnt wed have all sorts of problems with that logarithm.
Example 2 Find the general solution to the following differential equation. x 2 y - 7 xy + 16 y = 0

Solution First the roots of (3).
r ( r - 1) - 7r + 16 = 0 r 2 - 8r + 16 = 0
( r - 4)
So the general solution is then,
=0
r=4
y ( x ) = c1 x 4 + c2 x 4 ln x
2007 Paul Dawkins
341
Complex Roots In this case well be assuming that our roots are of the form,
r1,2 = l m i xl + m i
If we take the first root well get the following solution.
This is a problem since we dont want complex solutions, we only want real solutions. We can eliminate this by recalling that,
x r = eln x = e r ln x
Plugging the root into this gives,
x l + m i = e(
l + m i ) ln x
= el ln x e m i ln x = eln x ( cos ( m ln x ) + i sin ( m ln x ) )

l
= x l cos ( m ln x ) + ix l sin ( m ln x )
Note that we had to use Euler formula as well to get to the final step. Now, as weve done every other time weve seen solution like this we can take the real part and the imaginary part and use those for our two solutions. So, in the case of complex roots the general solution will be,
y ( x ) = c1 x l cos ( m ln x ) + c2 x l sin ( m ln x ) = x l ( c1 cos ( m ln x ) + c2 sin ( m ln x ) )
Once again we can see why we needed to require x>0.
Example 3 Find the solution to the following differential equation. x 2 y + 3 xy + 4 y = 0

Solution Get the roots to (3) first as always.
r ( r - 1) + 3r + 4 = 0 r 2 + 2r + 4 = 0 r1,2 = -1 3 i
y ( x ) = c1 x -1 cos
3 ln x + c2 x -1 sin
3 ln x
We should now talk about how to deal with x<0 since that is a possibility on occasion. To deal with this we need to use the variable transformation,
h = -x
In this case since x<0 we will get h>0. Now, define,
u (h ) = y ( x ) = y ( -h )
2007 Paul Dawkins
Then using the chain rule we can see that,
u (h ) = - y ( x )
2
and
u (h ) = y ( x )
With this transformation the differential equation becomes,
a ( -h ) u + b ( -h )( -u ) + cu = 0 ah 2u + bh u + cu = 0
In other words, since h>0 we can use the work above to get solutions to this differential equation. Well also go back to xs by using the variable transformation in reverse.
h = -x
Lets just take the real, distinct case first to see what happens.
u (h ) = c1h r1 + c2h r2
r
y ( x ) = c1 ( - x ) 1 + c2 ( - x ) 2
r
Now, we could do this for the rest of the cases if we wanted to, but before doing that lets notice that if we recall the definition of absolute value,
x if x 0 x = - x if x < 0
we can combine both of our solutions to this case into one and write the solution as,
y ( x ) = c1 x 1 + c2 x 2 ,
r r
x0
Note that we still need to avoid x=0 since we could still get division by zero. However this is now a solution for any interval that doesnt contain x=0. We can do likewise for the other two cases and the following solutions for any interval not containing x=0,.
y ( x ) = c1 x + c2 x ln x
r r
y ( x ) = c1 x cos ( m ln x ) + c2 x sin ( m ln x )
l l
We can make one more generalization before working one more example. A more general form of an Euler Equation is,
a ( x - x0 ) y + b ( x - x0 ) y + cy = 0
2
and we can ask for solutions in any interval not containing x = x0 . The work for generating the solutions in this case is identical to all the above work and so isnt shown here. The solutions in this general case for any interval not containing x=a are,
y ( x ) = c1 x - a 1 + c2 x - a
r
r2
y ( x) = x - a y ( x) = x - a
2007 Paul Dawkins
(c + c
1 1
ln x - a )
2
( c cos ( m ln x - a ) + c sin ( m ln x - a ))
Where the roots are solutions to
ar ( r - 1) + b ( r ) + c = 0
Example 4 Find the solution to the following differential equation on any interval not
containing x=-6.
3 ( x + 6 ) y + 25 ( x + 6 ) y - 16 y = 0
2
Solution So we get the roots from the identical quadratic in this case.
3r ( r - 1) + 25r - 16 = 0 3r 2 + 22r - 16 = 0
( 3r - 2 )( r + 8 ) = 0
2
2 r1 = , r2 = -8 3
-8
y ( x ) = c1 x - a 3 + c2 x - a
2007 Paul Dawkins
344
2007 Paul Dawkins
345
Higher Order Differential Equations

Introduction
In this chapter were going to take a look at higher order differential equations. This chapter will actually contain more than most text books tend to have when they discuss higher order differential equations. We will definitely cover the same material that most text books do here. However, in all the previous chapters all of our examples were 2nd order differential equations or 2 x 2 systems of differential equations. So, in this chapter were also going to do a couple of examples here dealing with 3rd order or higher differential equation with Laplace transforms and series as well as a discussion of some larger systems of differential equations. Here is a brief listing of the topics in this chapter. Basic Concepts for nth Order Linear Equations Well start the chapter off with a quick look at some of the basic ideas behind solving higher order linear differential equations. Linear Homogeneous Differential Equations In this section well take a look at extending the ideas behind solving 2nd order differential equations to higher order. Undetermined Coefficients Here well look at undetermined coefficients for higher order differential equations. Variation of Parameters Well look at variation of parameters for higher order differential equations in this section. Laplace Transforms In this section were just going to work an example of using Laplace transforms to solve a differential equation on a 3rd order differential equation just so say that we looked at one with order higher than 2nd. Systems of Differential Equations Here well take a quick look at extending the ideas we discussed when solving 2 x 2 systems of differential equations to systems of size 3x3. Series Solutions This section serves the same purpose as the Laplace Transform section. It is just here so we can say weve worked an example using series solutions for a differential equations of order higher than 2nd.
2007 Paul Dawkins
346
Basic Concepts for nth Order Linear Equations

Well start this chapter off with the material that most text books will cover in this chapter. We will take the material from the Second Order chapter and expand it out to nth order linear differential equations. As well see almost all of the 2nd order material will very naturally extend out to nth order with only a little bit of new material. So, lets start things off here with some basic concepts for nth order linear differential equations. The most general nth order linear differential equation is,
0 (t ) y = G (t ) Pn ( t ) y ( n) + Pn -1 ( t ) y ( n -1) + L + P 1 (t ) y + P y
( m)
(4)
where youll hopefully recall that,
dmy = m dx
(n ) Many of the theorems and ideas for this material require that y has a coefficient of 1 and so if
we divide out by Pn ( t ) we get,
y ( n ) + pn-1 ( t ) y ( n-1) + L + p1 ( t ) y + p0 ( t ) y = g ( t )
As we might suspect an IVP for an nth order differential equation will require the following n initial conditions.
(5)
y ( t0 ) = y0 ,
y ( t0 ) = y1 ,
L,
y ( n-1) ( t0 ) = yn-1
(6)
The following theorem tells us when we can expect there to be a unique solution to the IVP given by (5) and (6). Theorem 1 Suppose the functions p0 , p1 ,K , pn -1 and g ( t ) are all continuous in some open interval I containing t0 then there is a unique solution to the IVP given by (5) and (6) and the solution will exist for all t in I. This theorem is a very natural extension of a similar theorem we saw in the 1st order material. Next we need to move into a discussion of the nth order linear homogeneous differential equation,
y ( n) + pn -1 ( t ) y ( n -1) + L + p1 ( t ) y + p0 ( t ) y = 0
(7)
Lets suppose that we know y1 ( t ) , y2 ( t ) , K , yn ( t ) are all solutions to (7) then by the an extension of the Principle of Superposition we know that
y ( t ) = c1 y1 ( t ) + c2 y2 ( t ) + L + cn yn ( t )
will also be a solution to (7). The real question here is whether or not this will form a general solution to (7).
2007 Paul Dawkins
347
In order for this to be a general solution then we will have to be able to find constants c1 , c2 ,K , cn for any choice of t0 (in the interval I from Theorem 1) and any choice of
y1 , y2 ,K , yn . Or, in other words we need to be able to find c1 , c2 ,K , cn that will solve, c1 y1 ( t0 ) + c2 y2 ( t0 ) + L + cn yn ( t0 ) = y0 ( t0 ) + c2 y2 ( t0 ) + L + cn yn ( t0 ) = y1 c1 y1 cy

( n -1) 1 1
( t0 ) + c2 y
( n -1) 2
( t0 ) + L + cn y
L L O
( n -1) n
( t0 ) = yn-1
Just as we did for 2nd order differential equations, we can use Cramers Rule to solve this and the denominator of each the answers will be the following determinant of an n x n matrix.
y1 y1 y1( M
n -1)
y2 y2
( y2
yn yn M
n -1)
n -1)
( L yn
As we did back with the 2nd order material well define this to be the Wronskian and denote it by,
W ( y1 , y2 ,K yn )( t ) =
y1( n -1)
y1 y1 M
( n -1) ( n -1) y2 L yn
y2 y2 M
L L O
yn yn M
Now that we have the definition of the Wronskian out of the way we need to get back to the question at hand. Because the Wronskian is the denominator in the solution to each of the ci we can see that well have a solution provided it is not zero for any value of t = t0 that we chose to evaluate the Wronskian at. The following theorem summarizes all this up. Theorem 2 Suppose the functions p0 , p1 ,K , pn -1 are all continuous on the open interval I and further
suppose that y1 ( t ) , y2 ( t ) ,K yn ( t ) are all solutions to (7). If W ( y1 , y2 ,K yn )( t ) 0 for every t in I then y1 ( t ) , y2 ( t ) ,K yn ( t ) form a Fundamental Set of Solutions and the general solution to (7) is,
y ( t ) = c1 y1 ( t ) + c2 y2 ( t ) + L + cn yn ( t )
Recall as well that if a set of solutions form a fundamental set of solutions then they will also be a set of linearly independent functions. Well close this section off with a quick reminder of how we find solutions to the nonhomogeneous differential equation, (5). We first need the nth order version of a theorem we saw back in the 2nd order material.
Theorem 3 Suppose that Y1 ( t ) and Y2 ( t ) are two solutions to (5) and that y1 ( t ) , y2 ( t ) ,K yn ( t ) are a fundamental set of solutions to the homogeneous differential equation (7) then,
Y1 ( t ) - Y2 ( t )
is a solution to (7) and it can be written as
Y1 ( t ) - Y2 ( t ) = c1 y1 ( t ) + c2 y2 ( t ) + L + cn yn ( t )
Now, just as we did with the 2nd order material if we let Y ( t ) be the general solution to (5) and if we let YP ( t ) be any solution to (5) then using the result of this theorem we see that we must have,
where, yc ( t ) = c1 y1 ( t ) + c2 y2 ( t ) + L + cn yn ( t ) is called the complimentary solution and
Y ( t ) = c1 y1 ( t ) + c2 y2 ( t ) + L + cn yn ( t ) + YP ( t ) = yc ( t ) + YP ( t )
YP ( t ) is called a particular solution.
Over the course of the next couple of sections well discuss the differences in finding the complimentary and particular solutions for nth order differential equations in relation to what we know about 2nd order differential equations. Well see that, for the most part, the methods are the same. The amount of work involved however will often be significantly more.
2007 Paul Dawkins
349
Linear Homogeneous Differential Equations

As with 2nd order differential equations we cant solve a nonhomogeneous differential equation unless we can first solve the homogeneous differential equation. Well also need to restrict ourselves down to constant coefficient differential equations as solving non-constant coefficient differential equations is quite difficult and so we wont be discussing them here. Likewise, well only be looking at linear differential equations. So, lets start off with the following differential equation,
an y ( n) + an-1 y ( n -1) + L + a1 y + a0 y = 0
Now, assume that solutions to this differential equation will be in the form y ( t ) = e rt and plug this into the differential equation and with a little simplification we get,
e r t ( an r n + an-1r n-1 + L + a1r + a0 ) = 0 an r n + an-1r n-1 + L + a1r + a0 = 0
and so in order for this to be zero well need to require that
This is called the characteristic polynomial/equation and its roots/solutions will give us the solutions to the differential equation. We know that, including repeated roots, an nth degree polynomial (which we have here) will have n roots. So, we need to go through all the possibilities that weve got for roots here. This is where we start to see differences in how we deal with nth order differential equations versus 2nd order differential equations. There are still the three main cases : real distinct roots, repeated roots and complex roots (although these can now also be repeated as well see). In 2nd order differential equations each differential equation could only involve one of these cases. Now, however, that will not necessarily be the case. We could very easily have differential equations that contains each of these cases. For instance suppose that we have an 9th order differential equation. The complete list of roots could have 3 roots which only occur once in the list (i.e. real distinct roots), a root with multiplicity 4 (i.e. occurs 4 times in the list) and a set of complex conjugate roots (recall that because the coefficients are all real complex roots will always occur in conjugate pairs). So, for each nth order differential equation well need to form a set of n linearly independent functions (i.e. a fundamental set of solutions) in order to get a general solution. In the work that follows well discuss the solutions that we get from each case but we will leave it to you to verify that when we put everything together to form a general solution that we do indeed get a fundamental set of solutions. Recall that in order to this we need to verify that the Wronskian is not zero. So, lets get started with the work here. Lets start off by assuming that in the list of roots of the characteristic equation we have r1 , r2 , K , rk and they only occur once in the list. The solution from each of these will then be,
e1 ,
2007 Paul Dawkins
r t
e2 ,
350
r t
L,
rk t
Theres nothing really new here for real distinct roots. Now lets take a look at repeated roots. The result here is a natural extension of the work we saw in the 2nd order case. Lets suppose that r is a root of multiplicity k (i.e. r occurs k times in the list of roots). We will then get the following k solutions to the differential equation,
er t ,
t e rt ,
L,
t k -1e r t
So, for repeated roots we just add in a t for each of the solutions past the first one until we have a total of k solutions. Again, we will leave it to you to compute the Wronskian to verify that these are in fact a set of linearly independent solutions. Finally we need to deal with complex roots. The biggest issue here is that we can now have repeated complex roots for 4th order or higher differential equations. Well start off by assuming that r = l m i occurs only once in the list of roots. In this case well get the standard two solutions,
elt cos ( m t )
el t sin ( m t )
Now lets suppose that r = l m i has a multiplicity of k (i.e. they occur k times in the list of roots). In this case we can use the work from the repeated roots above to get the following set of 2k complex-valued solutions,
e( e(
l +m i) t l - m i )t
, t e( l + m i ) t ,
l - m i )t
, t e(
L, L,
t k -1e(l + m i )t t k -1e(
l - m i )t
The problem here of course is that we really want real-valued solutions. So, recall that in the case where they occurred once all we had to do was use Eulers formula on the first one and then take the real and imaginary part to get two real valued solutions. Well do the same thing here and use Eulers formula on the first set of complex-valued solutions above, split each one into its real and imaginary parts to arrive at the following set of 2k real-valued solutions.
elt cos ( m t ) ,
elt sin ( m t ) ,
telt cos ( m t ) , telt sin ( m t ) ,
t k -1e lt cos ( m t ) ,
L,
t k -1el t sin ( m t )
Once again well leave it to you to verify that these do in fact form a fundamental set of solutions. Before we work a couple of quick examples here we should point out that the characteristic polynomial is now going to be at least a 3rd degree polynomial and finding the roots of these by hand is often a very difficult and time consuming process and in many cases if the roots are not rational (i.e. in the form
p q
) it can be almost impossible to find them all by hand. To see a
process for determining all the rational roots of a polynomial check out the Finding Zeroes of Polynomials page in my Algebra notes. In practice however, we usually use some form of computation aid such as Maple or Mathematica to find all the roots. So, lets work a couple of example here to illustrate at least some of the ideas discussed here.
2007 Paul Dawkins
351
Example 1 Solve the following IVP. y ( 3) - 5 y - 22 y + 56 y = 0

Solution The characteristic equation is,
y ( 0) = 1
y ( 0) = -2
y ( 0 ) = -4
r 3 - 5r 2 - 22r + 56 = ( r + 4 )( r - 2 )( r - 7 ) = 0
r1 = -4, r2 = 2, r3 = 7
So we have three real distinct roots here and so the general solution is,
y ( t ) = c1e-4 t + c2e 2 t + c3e7 t
Differentiating a couple of times and applying the initial conditions gives the following system of equations that well need to solve in order to find the coefficients.
1= -2 = -4 =
y ( 0 ) = c1 + c2 + c3 y ( 0 ) = -4c1 + 2c2 + 7c3 y ( 0 ) = 16c1 + 4c2 + 49c3
c1 = 14 33 13 c2 = 15 c3 = - 16 55
-4 t 2t 16 7 t y ( t ) = 14 + 13 33 e 15 e - 55 e
So, outside of needing to solve a cubic polynomial (which we left the details to you to verify) and needing to solve a system of 3 equations to find the coefficients (which we also left to you to fill in the details) the work here is pretty much identical to the work we did in solving a 2nd order IVP. Because the initial condition work is identical to work that we should be very familiar with to this point with the exception that it involved solving larger systems were going to not bother with solving IVPs for the rest of the examples. The main point of this section is the new ideas involved in finding the general solution to the differential equation anyway and so well concentrate on that for the remaining examples. Also note that well not be showing very much work in solving the characteristic polynomial. We are using computational aids here and would encourage you to do the same here. Solving these higher degree polynomials is just too much work and would obscure the point of these examples. So, lets move into a couple of examples where we have more than one case involved in the solution.
Example 2 Solve the following differential equation. 2 y ( 4) + 11 y ( 3) + 18 y + 4 y - 8 y = 0

2r 4 + 11r 3 + 18r 2 + 4r - 8 = ( 2r - 1)( r + 2 ) = 0

3
So, we have two roots here, r1 =

2007 Paul Dawkins
1 2
and r2 = -2 which is multiplicity of 3. Remember that well

get three solutions for the second root and after the first we add ts only the solution until we reach three solutions. The general solution is then,
y ( t ) = c1e 2 + c2 e2t + c3te 2t + c4t 2 e2t

1t
Example 3 Solve the following differential equation. y ( 5) + 12 y ( 4) + 104 y ( 3) + 408 y + 1156 y = 0

r 5 + 12r 4 + 104r 3 + 408r 2 + 1156r = r r 2 + 6r + 34
=0
So, we have one real root r = 0 and a pair of complex roots r = -3 5 i each with multiplicity 2. So, the solution for the real root is easy and for the complex roots well get a total of 4 solutions, 2 will be the normal solutions and two will be the normal solution each multiplied by t. The general solution is,
y ( t ) = c1 + c2e -3t cos ( 5t ) + c3e-3t sin ( 5t ) + c4 t e -3t cos ( 5t ) + c5 t e -3t sin ( 5t )
Lets now work an example that contains all three of the basic cases just to say that we that weve got one work here.
Example 4 Solve the following differential equation. y ( 5) - 15 y ( 4 ) + 84 y ( 3) - 220 y + 275 y - 125 y = 0

r 5 - 15r 4 + 84r 3 - 220r 2 + 275r - 125 = ( r - 1)( r - 5) ( r 2 - 4r + 5) = 0

2
In this case weve got one real distinct root, r = 1 , and double root, r = 5 , and a pair of complex roots, r = 2 i that only occur once. The general solution is then,
y ( t ) = c1et + c2e5t + c3 t e5t + c4 e2t cos ( t ) + c5e 2t sin ( t )
Weve got one final example to work here that on the surface at least seems almost too easy. The problem here will be finding the roots as well see.
Example 5 Solve the following differential equation. y ( 4) + 16 y = 0

r 4 + 16 = 0
So, a really simple characteristic equation. However, in order to find the roots we need to
compute the fourth root of -16 and that is something that most people havent done at this point in their mathematical career. Well just give the formula here for finding them, but if youre interested in seeing a little more about this you might want to check out the Powers and Roots section of my Complex Numbers Primer. The 4 (and yes there are 4!) 4th roots of -16 can be found by evaluating the following,
4
-16 = ( -16 ) 4 = 4 16e

1
( p4 + p2k )i
pk p pk = 2 ( cos ( p 4 + 2 ) + i cos ( 4 + 2 ) )
k = 0,1, 2,3
Note that each value of k will give a distinct 4th root of -16. Also, note that for the 4th root (and ONLY the 4th root) of any negative number all we need to do is replace the 16 in the above formula with the absolute value of the number in question and this formula will work for those as well. Here are the 4th roots of -16.
p k = 0 : 2 ( cos ( p 4 ) + i sin ( 4 ) ) = 2
1 2
p k = 1: 2 ( cos ( 34 ) + i sin ( 34p + p2k ) ) = 2 p k = 2 : 2 ( cos ( 54 ) + i sin ( 54p ) ) = 2 -
k = 3: 2 ( cos ( 74p ) + i sin ( 74p
( )) = 2 (
1 2
i = 2 +i 2
1 2 1 2
1 2
1 2
i = - 2 -i 2
1 2
i = 2 -i 2
1 2
i = - 2 +i 2
So, we have two sets of complex roots : r = 2 i 2 and r = - 2 i 2 . The general solution is,
y ( t ) = c1e
2t
cos
2 t + c2e
2t
sin
2 t + c3e-
2t
cos
2 t + c4e -
2t
sin
2t
So, weve worked a handful of examples here of higher order differential equations that should give you a feel for how these work in most cases. There are of course a great many different kinds of combinations of the basic cases than what we did here and of course we didnt work any case involving 6th order or higher, but once youve got an idea on how these work its pretty easy to see that they all work pretty in pretty much the same manner. The biggest problem with the higher order differential equations is that the work in solving the characteristic polynomial and the system for the coefficients on the solution can be quite involved.
2007 Paul Dawkins
354
Undetermined Coefficients
We now need to start looking into determining a particular solution for nth order differential equations. The two methods that well be looking at are the same as those that we looked at in the 2nd order chapter. In this section well look at the method of Undetermined Coefficients and this will be a fairly short section. With one small extension, which well see in the lone example in this section, the method is identical to what we saw back when we were looking at undetermined coefficients in the 2nd order differential equations chapter. Given the differential equation,
if g ( t ) is an exponential function, polynomial, sine, cosine, sum/difference of one of these and/or a product of one of these then we guess the form of a particular solution using the same guidelines that we used in the 2nd order material. We then plug the guess into the differential equation, simplify and set the coefficients equal to solve for the constants. The one thing that we need to recall is that we first need the complimentary solution prior to making our guess for a particular solution. If any term in our guess is in the complimentary solution then we need to multiply the portion of our guess that contains that term by a t. This is where the one extension to the method comes into play. With a 2nd order differential equation the most wed ever need to multiply by is t 2 . With higher order differential equations this may need to be more than t 2 . The work involved here is almost identical to the work weve already done and in fact it isnt even that much more difficult unless the guess is particularly messy and that makes for more mess when we take the derivatives and solve for the coefficients. Because there isnt much difference in the work here were only going to do a single example in this section illustrating the extension. So, lets take a look at the lone example were going to do here.
Example 1 Solve the following differential equation. 3 y ( ) - 12 y + 48 y - 64 y = 12 - 32e-8t + 2e4t

Solution We first need the complimentary solution so the characteristic equation is,
r 3 - 12r 2 + 48r - 64 = ( r - 4) = 0
3
r = 4 ( multiplicity 3)
Weve got a single root of multiplicity 3 so the complimentary solution is,
yc ( t ) = c1e 4t + c2te 4 t + c3t 2 e4 t YP = A + Be-8t + Ce 4t
Now, our first guess for a particular solution is,
Notice that the last term in our guess is in the complimentary solution so well need to add one at least one t to the third term in our guess. Also notice that multiplying the third term by either t or
t 2 will result in a new term that is still in the complimentary solution and so well need to multiply the third term by t 3 in order to get a term that is not contained in the complimentary
solution. Our final guess is then,
YP = A + Be -8t + Ct 3e 4 t
Now all we need to do is take three derivatives of this, plug this into the differential equation and simplify to get (well leave it to you to verify the work here),
-64 A - 1728Be-8t + 6Ce4 t = 12 - 32e-8t + 2e 4t

Setting coefficients equal and solving gives,
t0 : -64 A = 12 -8 t e : -1728B = -32 e4t : 6C = 2

A particular solution is then,
3 A = - 16 1 B = 54 C=1 3
3 4t 3 1 -8 t YP = - 16 + 54 e +1 3t e
The general solution to this differential equation is then,
3 4t 3 1 -8 t + 54 y ( t ) = c1e4 t + c2 te4t + c3t 2e 4t - 16 e +1 3t e
Okay, weve only worked one example here, but remember that we mentioned earlier that with the exception of the extension to the method that we used in this example the work here is identical to work we did the 2nd order material.
2007 Paul Dawkins
356
Variation of Parameters
We now need to take a look at the second method of determining a particular solution to a differential equation. As we did when we first saw Variation of Parameters well go through the whole process and derive up a set of formulas that can be used to generate a particular solution. However, as we saw previously when looking at 2nd order differential equations this method can lead to integrals that are not easy to evaluate. So, while this method can always be used, unlike Undetermined Coefficients, to at least write down a formula for a particular solution it is not always going to be possible to actually get a solution. So lets get started on the process. Well start with the differential equation, and assume that weve found a fundamental set of solutions, y1 ( t ) , y2 ( t ) , K , yn ( t ) , for the associated homogeneous differential equation. Because we have a fundamental set of solutions to the homogeneous differential equation we now know that the complimentary solution is,
(1)
y ( t ) = c1 y1 ( t ) + c2 y2 ( t ) + L + cn yn ( t )
The method of variation of parameters involves trying to find a set of new functions, u1 ( t ) , u2 ( t ) ,K , un ( t ) so that,
Y ( t ) = u1 ( t ) y1 ( t ) + u2 ( t ) y2 ( t ) + L + un ( t ) yn ( t )
(2)
will be a solution to the nonhomogeneous differential equation. In order to determine if this is possible, and to find the ui ( t ) if it is possible, well need a total of n equations involving the unknown functions that we can (hopefully) solve. One of the equations is easy. The guess, (2), will need to satisfy the original differential equation, (1). So, lets start taking some derivatives and as we did when we first looked at variation of parameters well make some assumptions along the way that will simplify our work and in the process generate the remaining equations well need. The first derivative of (2) is,
+ u2 y2 + L + un yn + u1 y1 + u2 y 2 + L + un yn Y ( t ) = u1 y1
Note that we rearranged the results of the differentiation process a little here and we dropped the ( t ) part on the u and y to make this a little easier to read. Now, if we keep differentiating this it will quickly become unwieldy and so lets make as assumption to simplify things here. Because we are after the ui ( t ) we should probably try to avoid letting the derivatives on these become too large. So, lets make the assumption that,
y1 + u2 y2 + L + un yn = 0 u1
The natural question at this point is does this even make sense to do? The answer is, if we end up with a system of n equation that we can solve for the ui ( t ) then yes it does make sense to do. Of
course, the other answer is, we wouldnt be making this assumption if we didnt know that it was going to work. However to accept this answer requires that you trust us to make the correct assumptions so maybe the first answer is the best at this point. At this point the first derivative of (2) is,
+ u2 y2 + L + un yn Y ( t ) = u1 y1
and so we can now take the second derivative to get,
+ u2 y Y ( t ) = u1 y1 2 + L + u n yn + u1 y1 + u 2 y2 + L + un yn
This looks an awful lot alike the original first derivative prior to us simplifying it so lets again make a simplification. Well again want to keep the derivatives on the ui ( t ) to a minimum so this time lets assume that,
and with this assumption the second derivative becomes,
y1 + u2 y2 + L + un yn =0 u1
+ u2 y2 + L + un yn Y ( t ) = u1 y1
Hopefully youre starting to see a pattern develop here. If we continue this process for the first n - 1 derivatives we will arrive that the following formula for these derivatives.
(k ) (k ) Y ( k ) ( t ) = u1 y1( k ) + u2 y2 + L + un yn
k = 1, 2,K , n - 1 k = 0,1,K n - 2
(3)
To get to each of these formulas we also had to assume that, and recall that the 0th derivative of a function is just the function itself. So, for example,
(0) y2 ( t ) = y2 ( t ) . (k) (k ) y1( k ) + u2 y2 yn u1 + L + un =0
(4)
Notice as well that the set of assumptions in (4) actually give us n - 1 equations in terms of the derivatives of the unknown functions : u1 ( t ) , u2 ( t ) , K , un ( t ) . All we need to do then is finish generating the first equation we started this process to find (i.e. plugging (2) into (1)). To do this well need one more derivative of the guess. Differentiating the
( n - 1)
st
derivative, which we can get from (3), to get the nth derivative gives,
( n) (n ) ( n -1) ( n -1) y1( n-1) + u2 y2 yn Y ( n ) ( t ) = u1 y1( n ) + u2 y2 + L + un yn + u1 + L + un
This time well also not be making any assumptions to simplify this but instead just plug this along with the derivatives given in (3) into the differential equation, (1)
2007 Paul Dawkins
358
( n) (n) ( n -1) ( n -1) y1( n -1) + u2 y2 u1 y1( n ) + u2 y2 + L + un yn + u1 + L + u + n yn ( n -1) ( n -1) ( n -1) pn-1 ( t ) u1 y1 + u2 y2 + L + un yn +
p0 ( t ) [u1 y1 + u2 y2 + L + un yn ] = g ( t )
Next, rearrange this a little to get,
( n) ( n -1) u1 y1 + pn -1 ( t ) y1 + L + p1 ( t ) y1 + p0 ( t ) y1 + ( n) ( n -1) y2 + p0 ( t ) y2 + u2 + pn -1 ( t ) y2 + L + p1 ( t ) y2
+ u2 y2 + L + un yn ]+ p1 ( t ) [u1 y1
un y
( n) n
+ pn -1 ( t ) y
( n -1) n
+ p0 ( t ) yn + + L + p1 ( t ) yn
( n -1) ( n -1) y1( n -1) + u2 y2 yn u1 + L + un = g (t )
Recall that y1 ( t ) , y2 ( t ) , K , yn ( t ) are all solutions to the homogeneous differential equation and so all the quantities in the [ ] are zero and this reduces down to,
( n -1) ( n -1) y1( n -1) + u2 y2 yn u1 + L + un = g (t )
So this equation, along with those given in (4), give us the n equations that we needed. Lets list them all out here for the sake of completeness.
y1 + u2 y2 + L + un yn = 0 u1 y1 + u2 y2 + L + un yn =0 u1 y1 + u2 y2 + L + un yn = 0 u1 M
y1( u1
n -2 )
( n -1) ( n -1) y1( n -1) + u2 y2 yn u1 + L + un = g (t )
( n - 2) ( n -2 ) y2 yn + u2 + L + un =0
So, weve got n equations, but notice that just like we got when we did this for 2nd order differential equations the unknowns in the system are not u1 ( t ) , u2 ( t ) , K , un ( t ) but instead they
( t ) , u2 ( t ) , K , un ( t ) . This isnt a major problem however. Provided we are the derivatives, u1

Also, recall that the y1 ( t ) , y2 ( t ) , K , yn ( t ) are assumed to be known functions and so they
can solve this system we can then just integrate the solutions to get the functions that were after.
along with their derivatives (which appear in the system) are all known quantities in the system. Now, we need to think about how to solve this system. If there arent too many equations we can just solve it directly if we want to. However, for large n (and it wont take much to get large
here) that could be quite tedious and prone to error and it wont work at all for general n as we have here. The best solution method to use at this point is then Cramers Rule. Weve used Cramers Rule several times in this course, but the best reference for our purposes here is when we used it when we first defined Fundamental Sets of Solutions back in the 2nd order material. Upon using Cramers Rule to solve the system the resulting solution for each ui will be a quotient of two determinants of n x n matrices. The denominator of each solution will be the determinant of the matrix of the known coefficients,
y1( n-1)
y1 y1 M
( n -1) ( n -1) y2 L yn
y2 y2 M
L L O
yn yn M
= W ( y1 , y2 , K yn )( t )
This however, is just the Wronskian of y1 ( t ) , y2 ( t ) , K , yn ( t ) as noted above and because we have assumed that these form a fundamental set of solutions we also know that the Wronskian will not be zero. This in turn tells us that the system above is in fact solvable and all of the assumptions we apparently made out of the blue above did in fact work. The numerators of the solution for ui will be the determinant of the matrix of coefficients with
the ith column replaced with the column 0, 0, 0, K , 0, g ( t ) . For example, the numerator for the
is, first one, u1
g (t )
0 0 M
( n -1) ( n -1) y2 L yn
y2 y 2 M
L L O
yn yn M
Now, by a nice property of determinants if we factor something out of one of the columns of a matrix then the determinant of the resulting matrix will be the factor times the determinant of new matrix. In other words, if we factor g ( t ) out of this matrix we arrive at,
g (t )
0 0 M
y2 y2 y
( n -1) 2
L L O
yn yn
( n -1) n
L y
0 0 = g (t ) M 1
( n -1) ( n -1) y2 L yn
y2 y2 M
L L O
yn yn M
We did this only for the first one, but we could just as easily done this with any of the n solutions. So, let Wi represent the determinant we get by replacing the ith column of the Wronskian with the column (0,0,0,,0,1) and the solution to the system can then be written as,
2007 Paul Dawkins
360
= u1
g ( t )W1 ( t ) , W (t )
= u2
g ( t )W2 ( t ) g ( t )Wn ( t ) = , L , un W (t ) W (t )
Wow! That was a lot of effort to generate and solve the system but were almost there. With the solution to the system in hand we can now integrate each of these terms to determine just what the unknown functions, u1 ( t ) , u2 ( t ) , K , un ( t ) weve after all along are.
g ( t )W1 ( t ) u1 = dt , W (t )
g ( t ) W2 ( t ) g ( t )Wn ( t ) u2 = dt , L , un = dt W (t ) W (t )
Finally, a particular solution to (1) is then given by,
g ( t )W1 ( t ) g ( t )W2 ( t ) g ( t )Wn ( t ) Y ( t ) = y1 ( t ) dt + y2 ( t ) dt + L + yn ( t ) dt W (t ) W (t ) W (t )
We should also note that in the derivation process here we assumed that the coefficient of the
y ( n ) term was a one and that has been factored into the formula above. If the coefficient of this
term is not one then well need to make sure and divide it out before trying to use this formula. Before we work an example here we really should note that while we can write this formula down actually computing these integrals may be all but impossible to do. Okay lets take a look at a quick example.
Example 1 Solve the following differential equation. y ( 3) - 2 y - 21y - 18 y = 3 + 4e -t

r 3 - 2r 2 - 21r - 18 = ( r + 3)( r + 1)( r - 6 ) = 0
r1 = -3, r2 = -1, r3 = 6
So we have three real distinct roots here and so the general solution is,
yc ( t ) = c1e -3t + c2e -t + c3e6t
Okay, weve now got several determinants to compute. Well leave it to you to verify the following determinant computations.
e -3 t W = -3e -3t 9e -3t e -3t W2 = -3e -3t 9e -3t
e -t -e -t e -t
e 6t 6e6 t = 126e 2 t 36e6t
0 e -t W1 = 0 -e -t 1 e -t e -3t W3 = -3e -3t 9e -3t
e6 t 6e6 t = 7e5t 36e 6t e- t -e -t e- t 0 0 = 2e -4t 1
0 e6 t 0 6e 6t = -9e3t 1 36e 6t
2007 Paul Dawkins
361
Now, given that g ( t ) = 3 + 4e- t we can compute each of the ui . Here are those integrals.
( 3 + 4e )( 7e ) 1 1 u1 = dt = 3e3t + 4e2 t dt = ( e3t + 2e 2 t ) 2t 126e 18 18 -t 3t ( 3 + 4e )( -9e ) 1 1 u2 = dt = - 3et + 4 dt = - ( 3et + 4t ) 2t 126e 14 14 -t -4 t ( 3 + 4e )( 2e ) 1 1 1 4 u3 = dt = 3e -6t + 4e -7t dt = - e -6 t - e -7 t 2t 126e 63 63 2 7
-t 5t
Note that we didnt include the constants of integration in each of these because including them would just have introduced a term that would get absorbed into the complimentary solution just as we saw when we were dealing with 2nd order differential equations. Finally, a particular solution for this differential equation is then,
YP = u1 y1 + u2 y2 + u3 y3
1 1 1 = 18 (e3t + 2e2t ) e-3t - 14 (3et + 4t ) e-t + 63 ( - 12 e-6t - 74 e-7t ) e6t -t 5 -t =-1 -2 6 + 49 e 7 te
-t 5 -t y ( t ) = c1e-3t + c2e -t + c3e 6t - 1 -2 6 + 49 e 7 te
Were only going to do a single example in this section to illustrate the process more than anything so with that well close out this section.
2007 Paul Dawkins
362
Laplace Transforms
There really isnt all that much to this section. All were going to do here is work a quick example using Laplace transforms for a 3rd order differential equation so we can say that we worked at least one problem for a differential equation whose order was larger than 2. Everything that we know from the Laplace Transforms chapter is still valid. The only new bit that well need here is the Laplace transform of the third derivative. We can get this from the general formula that we gave when we first started looking at solving IVPs with Laplace transforms. Here is that formula,
L { y } = s 3Y ( s ) - s 2 y ( 0 ) - sy ( 0 ) - y ( 0 )
Heres the example for this section.
Example 1 Solve the following IVP. y - 4 y = 4t + 3u6 ( t ) e30-5t ,
y ( 0 ) = -3 y ( 0 ) = 1 y ( 0 ) = 4
Solution As always we first need to make sure the function multiplied by the Heaviside function has been properly shifted.
y - 4 y = 4t + 3u6 ( t ) e -5( t -6 )
It has been properly shifted and we can see that were shifting e -5t . All we need to do now is take the Laplace transform of everything, plug in the initial conditions and solve for Y ( s ) . Doing all of this gives,
s 3Y ( s ) - s 2 y ( 0 ) - sy ( 0 ) - y ( 0 ) - 4 ( s 2Y ( s ) - sy ( 0 ) - y ( 0 ) ) =
4 3e -6s + s2 s + 5
4 3e -6 s + s2 s + 5 4 3e -6 s 3 2 2 s 4 s Y s = 3 s + 13 s + ( ) ( ) s2 s+5 4 3 e -6 s ( s3 - 4s 2 )Y ( s ) = 4 - 3ss 2+ 13s + 3 s+5 4 3 4 - 3s + 13s 3e -6 s Y (s) = + s4 ( s - 4) s 2 ( s - 4 )( s + 5 )
(s
- 4s 2 ) Y ( s ) + 3s 2 - 13s =
Y ( s ) = F ( s ) + 3e -6 s G ( s )
Now we need to partial fraction and inverse transform F(s) and G(s). Well leave it to you to verify the details.
F (s) =
1 2! 3! 17 209 1 1( 3! ) 4 - 3s 4 + 13s 3 4 ( 2! ) 64 64 16 = 4 2 3 4 s ( s - 4) s-4 s s s s
4t 209 1 1 2 1 3 f ( t ) = 17 64 e - 64 - 16 t - 8 t - 6 t
2007 Paul Dawkins
363
G (s) =
1 1 1 1 1 144 225 400 20 = s 2 ( s - 4 )( s + 5 ) s - 4 s + 5 s s 2
1 1 1 1 e 4t - 225 e-5t - 400 - 20 g ( t ) = 144 t
Y ( s ) = F ( s ) + 3e -6 s G ( s )
y ( t ) = f ( t ) + 3u6 ( t ) g ( t - 6 )
where f(t) and g(t) are the functions shown above. Okay, there is the one Laplace transform example with a differential equation with order greater than 2. As you can see the work in identical except for the fact that the partial fraction work (which we didnt show here) is liable to be messier and more complicated.
2007 Paul Dawkins
364

In this section we want to take a brief look at systems of differential equations that are larger than 2 x 2. The problem here is that unlike the first few sections where we looked at nth order differential equations we cant really come up with a set of formulas that will always work for every system. So, with that in mind were going to look at all possible cases for a 3 x 3 system (leaving some details for you to verify at times) and then a couple of quick comments about 4 x 4 systems to illustrate how to extend things out to even larger systems and then well leave it to you to actually extend things out if youd like to. We will also not be doing any actual examples in this section. The point of this section is just to show how to extend out what we know about 2 x 2 systems to larger systems. Initially the process is identical regardless of the size of the system. So, for a system of 3 differential equations with 3 unknown functions we first put the system into matrix form,
r r x = A x
where the coefficient matrix, A, is a 3 x 3 matrix. We next need to determine the eigenvalues and eigenvectors for A and because A is a 3 x 3 matrix we know that there will be 3 eigenvalues (including repeated eigenvalues if there are any). This is where the process from the 2 x 2 systems starts to vary. We will need a total of 3 linearly independent solutions to form the general solution. Some of what we know from the 2 x 2 systems can be brought forward to this point. For instance, we know that solutions corresponding to simple eigenvalues (i.e. they only occur once in the list of eigenvalues) will be linearly independent. We know that solutions from a set of complex conjugate eigenvalues will be linearly independent. We also know how to get a set of linearly independent solutions from a double eigenvalue with a single eigenvector. There are also a couple of facts about eigenvalues/eigenvectors that we need to review here as well. First, provided A has only real entries (which it always will here) all complex eigenvalues will occur in conjugate pairs (i.e. l = a b i ) and their associated eigenvectors will also be complex conjugates of each other. Next, if an eigenvalue has multiplicity k 2 (i.e. occurs at least twice in the list of eigenvalues) then there will be anywhere from 1 to k linearly independent eigenvectors for the eigenvalue. With all these idea in mind lets start going through all the possible combinations of eigenvalues that we can possibly have for a 3 x 3 case. Lets also note that for a 3 x 3 system it is impossible to have only 2 real distinct eigenvalues. The only possibilities are to have 1 or 3 real distinct eigenvalues. Here are all the possible cases. 3 Real Distinct Eigenvalues In this case well have the real, distinct eigenvalues l1 l 2 l 3 and their associated eigenvectors, h1 , h 2 and h3 are guaranteed to be linearly independent and so the three linearly independent solutions we get from this case are,
ltr e 1 h1 l tr e 2 h2 l tr e 3 h3
2007 Paul Dawkins
365
1 Real and 2 Complex Eigenvalues r From the real eigenvalue/eigenvector pair, l1 and h1 , we get one solution,
ltr e 1 h1
We get the other two solutions in the same manner that we did with the 2 x 2 case. If the eigenvalues are l 2,3 = a b i with eigenvectors h 2 and h3 = (h 2 ) we can get two real-valued solution by using Eulers formula to expand,
r l tr a + b i )t r e 2 h2 = e( h 2 = ea t ( cos ( b t ) + i sin ( b t ) )h 2 r r into its real and imaginary parts, u + i v . The final two real valued solutions we need are then, r r u v
1 Real Distinct and 1 Double Eigenvalue with 1 Eigenvector r From the real eigenvalue/eigenvector pair, l1 and h1 , we get one solution,
ltr e 1 h1
From our work in the 2 x 2 systems we know that from the double eigenvalue l 2 with single eigenvector, h 2 , we get the following two solutions,
where x and r must satisfy the following equations,
l tr e 2 h2
l t r l t r te 2 x +e 2 r
( A - l I )x = 0
2
r =x (A-l I)r
2
Note that the first equation simply tells us that x must be the single eigenvector for this eigenvalue, h 2 , and we usually just say that the second solution we get from the double root case is,
l tr l t r t e 2 h2 + e 2 r
r r r where r satisfies A - l 2 I r = h2
1 Real Distinct and 1 Double Eigenvalue with 2 Linearly Independent Eigenvectors We didnt look at this case back when we were examining the 2 x 2 systems but it is easy enough r to deal with. In this case well have a single real distinct eigenvalue/eigenvector pair, l1 and h1 , as well as a double eigenvalue l 2 and the double eigenvalue has two linearly independent eigenvectors, h 2 and h3 . In this case all three eigenvectors are linearly independent and so we get the following three linearly independent solutions,
ltr e 1 h1 l tr e 2 h2 l tr e 2 h3
We are now out of the cases that compare to those that we did with 2 x 2 systems and we now need to move into the brand new case that we pick up for 3 x 3 systems. This new case involves eigenvalues with multiplicity of 3. As we noted above we can have 1, 2, or 3 linearly
independent eigenvectors and so we actually have 3 sub cases to deal with here. So lets go through these final 3 cases for a 3 x 3 system. 1 Triple Eigenvalue with 1 Eigenvector r The eigenvalue/eigenvector pair in this case are l and h . Because the eigenvalue is real we know that the first solution we need is,
r elth
We can use the work from the double eigenvalue with one eigenvector to get that a second solution is,
r r t elth + elt r
r r r where r satisfies ( A - l I ) r = h
For a third solution we can take a clue from how we dealt with nth order differential equations with roots multiplicity 3. In those cases we multiplied the original solution by a t 2 . However, just as with the double eigenvalue case that wont be enough to get us a solution. In this case the third solution will be, where x , r , and m must satisfy,
r r
1 2 lt 2
r r r t e x + te l t r + e l t m
( A - l I )x
r =0
( A - lI ) r = x
( A - lI ) m = r
You can verify that this is a solution and the conditions by taking a derivative and plugging into the system. Now, the first condition simply tells us that x = h because we only have a single eigenvector here and so we can reduce this third solution to, where r , and m must satisfy,
1 2 lt 2
r r r t e h + t elt r + e lt m r
( A - lI ) r =h
( A - lI ) m = r
and finally notice that we would have solved the new first condition for in determining the second solution above and so all we really need to do here is solve the final condition. As a final note in this case, the 1 2 is in the solution solely to keep any extra constants from appearing in the conditions which in turn allows us to reuse previous results. 1 Triple Eigenvalue with 2 Linearly Independent Eigenvectors r In this case well have the eigenvalue l with the two linearly independent eigenvectors h1 and
r h 2 so we get the following two linearly independent solutions, r r elth1 e lth 2

l t r l t r te 2 x +e 2 r
We now need a third solution. The third solution will be in the form, where x and r must satisfy the following equations,
( A - l I )x = 0
2
r =x (A-l I)r
2
Weve already verified that this will be a solution with these conditions in the double eigenvalue case (that work only required a repeated eigenvalue, not necessarily a double one). However, unlike the previous times weve seen this we cant just say that x is an eigenvalue. In all the previous cases in which weve seen this condition we had a single eigenvalue and this time we have two linearly independent eigenvalues. This means that the most general possible solution to the first condition is,
r r r x = c1h1 + c2h 2
This creates problems in solving the second condition. The second condition will not have solutions for every choice of c1 and c2 and the choice that we use will be dependent upon the eigenvectors. So upon solving the first condition we would need to plug the general solution into the second condition and then proceed to determine conditions on c1 and c2 that would allow us to solve the second condition. 1 Triple Eigenvalue with 3 Linearly Independent Eigenvectors r r In this case well have the eigenvalue l with the three linearly independent eigenvectors h1 , h 2 , and h3 so we get the following three linearly independent solutions,
r el th1
r el th 2
r e lth3
4 x 4 Systems Well close this section out with a couple of comments about 4 x 4 systems. In these cases we will have 4 eigenvalues and will need 4 linearly independent solutions in order to get a general solution. The vast majority of the cases here are natural extensions of what 3 x 3 systems cases and in fact will use a vast majority of that work. Here are a couple of new cases that we should comment briefly on however. With 4 x 4 systems it will now be possible to have two different sets of double eigenvalues and two different sets of complex conjugate eigenvalues. In either of these cases we can treat each one as a separate case and use our previous knowledge about double eigenvalues and complex eigenvalues to get the solutions we need. It is also now possible to have a double complex eigenvalue. In other words we can have l = a b i each occur twice in the list of eigenvalues. The solutions for this case arent too bad. We get two solutions in the normal way of dealing with complex eigenvalues. The remaining two solutions will come from the work we did for a double eigenvalue. The work we did in that case did not require that the eigenvalue/eigenvector pair to be real. Therefore if the r eigenvector associated with l = a + b i is h then the second solution will be,
r r r r a + b i )t r h + e( r where r satisfies ( A - l I ) r = h r and once weve determined r we can again split this up into its real and imaginary parts using t e(
a + b i )t
Eulers formula to get two new real valued solutions.

Finally with 4 x 4 systems we can now have eigenvalues with multiplicity of 4. In these cases we can have 1, 2, 3, or 4 linearly independent eigenvectors and we can use our work with 3 x 3 systems to see how to generate solutions for these cases. The one issue that youll need to pay attention to is the conditions for the 2 and 3 eigenvector cases will have the same complications that the 2 eigenvector case has in the 3 x 3 systems. So, weve discussed some of the issues involved in systems larger than 2 x 2 and it is hopefully clear that when we move into larger systems the work can be become vastly more complicated.
2007 Paul Dawkins
369
Series Solutions
The purpose of this section is not to do anything new with a series solution problem. Instead it is here to illustrate that moving into a higher order differential equation does not really change the process outside of making it a little longer. Back in the Series Solution chapter we only looked at 2nd order differential equations so were going to do a quick example here involving a 3rd order differential equation so we can make sure and say that weve done at least one example with an order larger than 2.
Example 1 Find the series solution around x0 = 0 for the following differential equation.
y + x 2 y + xy = 0
Solution Recall that we can only find a series solution about x0 = 0 if this point is an ordinary point, or in other words, if the coefficient of the highest derivative term is not zero at x0 = 0 . That is clearly the case here so lets start with the form of the solutions as well as the derivatives that well need for this solution.
y ( x ) = an x n
n =0
y ( x ) = nan x n -1
n =1
y ( x ) = n ( n - 1)( n - 2 ) an x n -3
n =3
Plugging into the differential equation gives,
n ( n - 1)( n - 2 ) a x
n =3 n
n -3
+ x 2 nan x n-1 + x an x n = 0
n =1 n =0
Now, move all the coefficients into the series and do appropriate shifts so that all the series are in terms of xn .
n ( n - 1)( n - 2 ) a x
n =3 n n =0
n -3
+ nan x n+1 + an x n+1 = 0

n =1 n =0
( n + 3)( n + 2 )( n + 1) an +3 x n + ( n - 1) an-1 x n + an -1 xn = 0
n= 2 n =1
Next, lets notice that we can start the second series at n = 1 since that term will be zero. So lets do that and then we can combine the second and third terms to get,
n ( n + 3)( n + 2 )( n + 1) an+3 xn + ( n - 1) + 1 an-1 x = 0 n= 0 n =1
( n + 3)( n + 2 )( n + 1) an+3 xn + nan-1 xn = 0

n =0 n =1
So, we got a nice simplification in the new series that will help with some further simplification. The new second series can now be started at n = 0 and then combined with the first series to get,
( n + 3)( n + 2 )( n + 1) a
n= 0
n +3
n + nan -1 x =0
2007 Paul Dawkins
370
We can now set the coefficients equal to get a fairly simply recurrence relation.
( n + 3)( n + 2 )( n + 1) an+3 + nan-1 = 0

an +3 = - nan-1 ( n + 1)( n + 2 )( n + 3 )
n = 0,1, 2,K
Solving the recurrence relation gives,
n = 0,1, 2,K
Now we need to start plugging in values of n and this will be one of the main areas where we can see a somewhat significant increase in the amount of work required when moving into a higher order differential equation.
n = 0 : a3 = 0 n = 1: a4 = n = 2 : a5 = n = 3 : a6 = - a0 ( 2 )( 3)( 4 ) -2a1 ( 3)( 4 )( 5 )
-3a2 ( 4 )( 5 )( 6 ) -4a3 n = 4 : a7 = =0 ( 5 )( 6 )( 7 ) n = 5 : a8 = n = 6 : a9 = n = 7 : a10 = 5a0 -5a4 = ( 6 )( 7 )( 8) ( 2 )( 3)( 4 )( 6 )( 7 )(8 ) -6a5 ( 2 )( 6 ) a1 = ( 7 )( 8)( 9 ) ( 3)( 4 )( 5 )( 7 )(8 )( 9 )
-7 a6 ( 3)( 7 ) a2 = (8 )( 9 )(10 ) ( 4 )( 5)( 6 )( 8 )( 9 )(10 ) -8a7 n = 8 : a11 = =0 ( 9 )(10 )(11)
n = 9 : a12 = n = 10 : a13 = n = 11: a14 =
- ( 2 )( 6 )(10 ) a1 -10a9 = (11)(12 )(13) ( 3)( 4 )( 5 )( 7 )(8 )( 9 )(11)(12 )(13)
- ( 5 )( 9 ) a0 -9a8 = (10 )(11)(12 ) ( 2 )( 3)( 4 )( 6 )( 7 )(8 )(10 )(11)(12 )
- ( 3)( 7 )(11) a2 -11a10 = (12 )(13)(14 ) ( 4 )( 5 )( 6 )(8 )( 9 )(10 )(12 )(13)(14 )
Okay, we can now break the coefficients down into 4 sub cases given by a4 k , a4 k +1 , a4 k + 2 and you with only couple of comments as they are nearly identical derivations.
a4 k +3 for k = 0,1, 2, 3, K Well give a semi-detailed derivation for a4 k and then leave the rest to
First notice that all the a4 k terms have a0 in them and they will alternate in sign. Next notice that we can turn the denominator into a factorial, ( 4k )! to be exact, if we multiply top and bottom by the numbers that are already in the numerator and so this will turn these numbers into squares. Next notice that the product in the top will start at 1 and increase by 4 until we reach 4k - 3 . So, taking all of this into account we get,
( -1) (1) ( 5 ) L ( 4k - 3) a4 k = ( 4k )!
k 2 2
a0
k = 1, 2,3,K
and notice that this will only work starting with k = 1 as we wont get a0 out of this formula as we should by plugging in k = 0 . Now, for a4 k +1 the derivation is almost identical and so the formula is,
( -1) ( 2 ) ( 6 ) L ( 4k - 2 ) a4 k +1 = ( 4k + 1)!
k 2 2
a1
k = 1, 2,3,K
and again notice that this wont work for k = 0 The formula for a4 k + 2 is again nearly identical except for this one note that we also need to multiply top and bottom by a 2 in order to get the factorial to appear in the denominator and so the formula here is,
2 ( -1) ( 3) ( 7 ) L ( 4k - 1) a2 a4 k + 2 = ( 4k + 2 ) ! noticing yet one more time that this wont work for k = 0 .
k 2 2 2
k = 1, 2,3,K
Finally, we have a4 k +3 = 0 for k = 0,1, 2, 3, K Now that we have all the coefficients lets get the solution,
k 2 2
y ( x ) = a0 + a1 x + a2 x 2 + a3 x 3 + L + a4 k x 4 k + a4 k +1 x 4 k +1 + a4 k +3 x 4 k +3 + a4 k +3 x 4 k +3 + L = a0 + a1 x + a2 x
2
( -1) (1) ( 5 ) L ( 4k - 3) a0 x4 k + L+ ( 4k ) ! k 2 2 2 ( -1) ( 2 ) ( 6 ) L ( 4k - 2 ) a1 x 4k +1 + ( 4k + 1)! k 2 2 2 2 ( -1) ( 3) ( 7 ) L ( 4k - 1) a2 4 k + 2 x +L ( 4k + 2 ) !

2
Collecting up the terms that contain the same coefficient (except for the first one in each case since the formula wont work for those) and writing everything as a set of series gives us our solution,
2007 Paul Dawkins
372
k 2 2 2 -1) (1) ( 5 ) L ( 4k - 3) x 4 k ( y ( x ) = a0 1 + ( 4k ) ! k =1
k 2 2 2 ( -1) ( 2 ) ( 6 ) L ( 4k - 2 ) x4 k +1 a1 x + + k 4 1 ! + ( ) = k 1 k 2 2 2 4k +2 2 2 ( -1) ( 3) ( 7 ) L ( 4k - 1) x a2 x + ( 4k + 2 ) ! k =1
So, there we have it. As we can see the work in getting formulas for the coefficients was a little messy because we had three formulas to get, but individually they were not as bad as even some of them could be when dealing with 2nd order differential equations. Also note that while we got lucky with this problem and we were able to get general formulas for the terms the higher the order the less likely this will become.
2007 Paul Dawkins
373
Boundary Value Problems & Fourier Series

Introduction
In this chapter well be taking a quick and very brief look at a couple of topics. The two main topics in this chapter are Boundary Value Problems and Fourier Series. Well also take a look at a couple of other topics in this chapter. The main point of this chapter is to get some of the basics out of the way that well need in the next chapter where well take a look at one of the more common solution methods for partial differential equations. It should be pointed out that both of these topics are far more in depth than what well be covering here. In fact you can do whole courses on each of these topics. What well be covering here are simply the basics of these topics that well need in order to do the work in the next chapter. There are whole areas of both of these topics that well not be even touching on. Here is a brief listing of the topics in this chapter. Boundary Value Problems In this section well define the boundary value problems as well as work some basic examples. Eigenvalues and Eigenfunctions Here well take a look at the eigenvalues and eigenfunctions for boundary value problems. Periodic Functions and Orthogonal Functions Well take a look at periodic functions and orthogonal functions in section. Fourier Sine Series In this section well start looking at Fourier Series by looking at a special case : Fourier Sine Series. Fourier Cosine Series Well continue looking at Fourier Series by taking a look at another special case : Fourier Cosine Series. Fourier Series Here we will look at the full Fourier series. Convergence of Fourier Series Here well take a look at some ideas involved in the just what functions the Fourier series converge to as well as differentiation and integration of a Fourier series.
2007 Paul Dawkins
374
Boundary Value Problems

Before we start off this section we need to make it very clear that we are only going to scratch the surface of the topic of boundary value problems. There is enough material in the topic of boundary value problems that we could devote a whole class to it. The intent of this section is to give a brief (and we mean very brief) look at the idea of boundary value problems and to give enough information to allow us to do some basic partial differential equations in the next chapter. Now, with that out of the way, the first thing that we need to do is to define just what we mean by a boundary value problem (BVP for short). With initial value problems we had a differential equation and we specified the value of the solution and an appropriate number of derivatives at the same point (collectively called initial conditions). For instance for a second order differential equation the initial conditions are,
y ( t0 ) = y0
y ( t0 ) = y0
With boundary value problems we will have a differential equation and we will specify the function and/or derivatives at different points, which well call boundary values. For second order differential equations, which will be looking at pretty much exclusively here, any of the following can, and will, be used for boundary conditions.
y ( x0 ) = y0 y ( x0 ) = y0 y ( x0 ) = y0 y ( x0 ) = y0
y ( x1 ) = y1 y ( x1 ) = y1 y ( x1 ) = y1 y ( x1 ) = y1
(1) (2) (3) (4)
As mentioned above well be looking pretty much exclusively at second order differential equations. We will also be restricting ourselves down to linear differential equations. So, for the purposes of our discussion here well be looking almost exclusively at differential equations in the form,
y + p ( x ) y + q ( x ) y = g ( x )
(5)
along with one of the sets of boundary conditions given in (1) (4). We will, on occasion, look at some different boundary conditions but the differential equation will always be on that can be written in this form. As well soon see much of what we know about initial value problems will not hold here. We can, of course, solve (5) provided the coefficients are constant and for a few cases in which they arent. None of that will change. The changes (and perhaps the problems) arise when we move from initial conditions to boundary conditions. One of the first changes is a definition that we saw all the time in the earlier chapters. In the earlier chapters we said that a differential equation was homogeneous if g ( x ) = 0 for all x. Here we will say that a boundary value problem is homogeneous if in addition to g ( x ) = 0 we also have y0 = 0 and y1 = 0 (regardless of the boundary conditions we use). If any of these are not zero we will call the BVP nonhomogeneous.
It is important to now remember that when we say homogeneous (or nonhomogeneous) we are saying something not only about the differential equation itself but also about the boundary conditions as well. The biggest change that were going to see here comes when we go to solve the boundary value problem. When solving linear initial value problems a unique solution will be guaranteed under very mild conditions. We only looked at this idea for first order IVPs but the idea does extend to higher order IVPs. In that section we saw that all we needed to guarantee a unique solution was some basic continuity conditions. With boundary value problems we will often have no solution or infinitely many solutions even for very nice differential equations that would yield a unique solution if we had initial conditions instead of boundary conditions. Before we get into solving some of these lets next address the question of why were even talking about these in the first place. As well see in the next chapter in the process of solving some partial differential equations we will run into boundary value problems that will need to be solved as well. In fact, a large part of the solution process there will be in dealing with the solution to the BVP. In these cases the boundary conditions will represent things like the temperature at either end of a bar, or the heat flow into/out of either end of a bar. Or maybe they will represent the location of ends of a vibrating string. So, the boundary conditions there will really be conditions on the boundary of some process. So, with some of basic stuff out of the way lets find some solutions to a few boundary value problems. Note as well that there really isnt anything new here yet. We know how to solve the differential equation and we know how to find the constants by applying the conditions. The only difference is that here well be applying boundary conditions instead of initial conditions.
Example 1 Solve the following BVP. y + 4 y = 0 y ( 0 ) = -2 p y = 10 4
Solution Okay, this is a simple differential equation so solve and so well leave it to you to verify that the general solution to this is,
y ( x ) = c1 cos ( 2 x ) + c2 sin ( 2 x ) -2 = y ( 0 ) = c1
Now all that we need to do is apply the boundary conditions.
10 = y ( p 4 ) = c2
y ( x ) = -2 cos ( 2 x ) + 10sin ( 2 x )
We mentioned above that some boundary value problems can have no solutions or infinite solutions we had better do a couple of examples of those as well here. This next set of examples will also show just how small of a change to the BVP it takes to move into these other possibilities.
2007 Paul Dawkins
376
Example 2 Solve the following BVP. y + 4 y = 0
y ( 0 ) = -2
y ( 2p ) = -2
Solution Were working with the same differential equation as the first example so we still have,
y ( x ) = c1 cos ( 2 x ) + c2 sin ( 2 x ) -2 = y ( 0 ) = c1 -2 = y ( 2p ) = c1
Upon applying the boundary conditions we get,
So in this case, unlike previous example, both boundary conditions tell us that we have to have c1 = -2 and neither one of them tell us anything about c2 . Remember however that all were asking for is a solution to the differential equation that satisfies the two given boundary conditions and the following function will do that,
y ( x ) = -2cos ( 2 x ) + c2 sin ( 2 x )
In other words, regardless of the value of c2 we get a solution and so, in this case we get infinitely many solutions to the boundary value problem.

Solution Again, we have the following general solution,
y ( 0 ) = -2
y ( 2p ) = 3
y ( x ) = c1 cos ( 2 x ) + c2 sin ( 2 x ) -2 = y ( 0 ) = c1 3 = y ( 2p ) = c1
This time the boundary conditions give us,
In this case we have a set of boundary conditions each of which requires a different value of c1 in order to be satisfied. This, however, is not possible and so in this case have no solution. So, with Examples 2 and 3 we can see that only a small change to the boundary conditions, in relation to each other and to Example 1, can completely change the nature of the solution. All three of these examples used the same differential equation and yet a different set of initial conditions yielded, no solutions, one solution, or infinitely many solutions. Note that this kind of behavior is not always unpredictable however. If we use the conditions y ( 0 ) and y ( 2p ) the only way well ever get a solution to the boundary value problem is if we have,
y (0) = a
377
y ( 2p ) = a
2007 Paul Dawkins
for any value of a. Also, note that if we do have these boundary conditions well in fact get infinitely many solutions. All the examples weve worked to this point involved the same differential equation and the same type of boundary conditions so lets work a couple more just to make sure that weve got some more examples here. Also, note that with each of these we could tweak the boundary conditions a little to get any of the possible solution behaviors to show up (i.e. zero, one or infinitely many solutions).
Example 4 Solve the following BVP. y + 3 y = 0 y ( x ) = c1 cos

Applying the boundary conditions gives,
y (0) = 7
y ( 2p ) = 0
Solution The general solution for this differential equation is,
3 x + c2 sin
3x
)
c2 = -7 cot 2 3 p
7 = y ( 0 ) = c1
0 = y ( 2p ) = c1 cos 2 3 p + c2 sin 2 3 p
In this case we get a single solution,
y ( x ) = 7 cos
3 x - 7 cot 2 3 p sin
) (
3x

Solution Here the general solution is,
y ( 0 ) = 6
y ( p ) = -9
y ( x ) = c1 cos ( 5 x ) + c2 sin ( 5 x ) y ( x ) = -5c1 sin ( 5 x ) + 5c2 cos ( 5 x ) c2 = c2 =
and well need the derivative to apply the boundary conditions,
6 = y ( 0 ) = 5c2
6 5 9 5
-9 = y (p ) = -5c2
This is not possible and so in this case have no solution. All of the examples worked to this point have been nonhomogeneous because at least one of the boundary conditions have been non-zero. Lets work one nonhomogeneous example where the differential equation is also nonhomogeneous before we work a couple of homogeneous examples.
2007 Paul Dawkins
378
Example 6 Solve the following BVP. y + 9 y = cos x y ( 0 ) = 5 5 p y = 3 2
Solution The complimentary solution for this differential equation is,
yc ( x ) = c1 cos ( 3 x ) + c2 sin ( 3 x )
Using Undetermined Coefficients or Variation of Parameters it is easy to show (well leave the details to you to verify) that a particular solution is,
YP ( x ) = 1 8 cos x
The general solution and its derivative (since well need that for the boundary conditions) are,
y ( x ) = c1 cos ( 3 x ) + c2 sin ( 3 x ) + 1 8 cos x
y ( x ) = -3c1 sin ( 3 x ) + 3c2 cos ( 3 x ) - 1 8 sin x

5 = y ( 0 ) = 3c2

5 3
c2 = 5 3 c2 =
5 3
p -5 3 = y ( 2 ) = -c2
The boundary conditions then tell us that we must have c2 =
and they dont tell us anything
about c1 and so it is can be arbitrarily chosen. The solution is then, and there will be infinitely many solutions to the BVP.
1 y ( x ) = c1 cos ( 3 x ) + 5 3 sin ( 3 x ) + 8 cos x
Lets now work a couple of homogeneous examples that will also be helpful to have worked once we get to the next section.

Solution Here the general solution is,
y (0) = 0
y ( 2p ) = 0
y ( x ) = c1 cos ( 2 x ) + c2 sin ( 2 x ) 0 = y ( 0 ) = c1 0 = y ( 2p ) = c1
So c2 is arbitrary and the solution is,
y ( x ) = c2 sin ( 2 x )
and in this case well get infinitely many solutions.


Solution The general solution in this case is,
y (0) = 0
y ( 2p ) = 0
y ( x ) = c1 cos
3 x + c2 sin
3x
)
c2 = 0
0 = y ( 0 ) = c1
0 = y ( 2p ) = c2 sin 2 3 p
In this case we found both constants to be zero and so the solution is,
y ( x) = 0
In the previous example the solution was y ( x ) = 0 . Notice however, that this will always be a solution to any homogenous system given by (5) and any of the (homogeneous) boundary conditions given by (1) (4). Because of this we usually call this solution the trivial solution. Sometimes, as in the case of the last example the trivial solution is the only solution however we generally prefer solutions to be non-trivial. This will be a major idea in the next section. Before we leave this section an important point needs to be made. In each of the examples, with one exception, the differential equation that we solved was in the form,
y + l y = 0
The one exception to this still solved this differential equation except it was not a homogeneous differential equation and so we were still solving this basic differential equation in some manner. So, there are probably several natural questions that can arise at this point. Do all BVPs involve this differential equation and if not why did we spend so much time solving this one to the exclusion of all the other possible differential equations? The answers to these questions are fairly simple. First, this differential equation is most definitely not the only one used in boundary value problems. It does however exhibit all of the behavior that we wanted to talk about here and has the added bonus of being very easy to solve. So, by using this differential equation almost exclusively we can see and discuss the important behavior that we need to discuss and frees us up from lots of potentially messy solution details and or messy solutions. We will, on occasion, look at other differential equations in the rest of this chapter, but we will still be working almost exclusively with this one. There is another important reason for looking at this differential equation. When we get to the next chapter and take a brief look at solving partial differential equations we will see that almost every one of the examples that well work there come down to exactly this differential equation. Also, in those problems we will be working some real problems that are actually solved in places and so are not just made up problems for the purposes of examples. Admittedly they will have some simplifications in them, but they do come close to realistic problem in some cases.
2007 Paul Dawkins
380
Eigenvalues and Eigenfunctions

As we did in the previous section we need to again note that we are only going to give a brief look at the topic of eigenvalues and eigenfunctions for boundary value problems. There are quite a few ideas that well not be looking at here. The intent of this section is simply to give you an idea of the subject and to do enough work to allow us to solve some basic partial differential equations in the next chapter. Now, before we start talking about the actual subject of this section lets recall a topic from Linear Algebra that we briefly discussed previously in these notes. For a given square matrix, A, if we could find values of l for which we could find nonzero solutions, i.e. x 0 , to, then we called l an eigenvalue of A and x was its corresponding eigenvector. Its important to recall here that in order for l to be an eigenvalue then we had to be able to find nonzero solutions to the equation. So, just what does this have to do with boundary value problems? Well go back to the previous section and take a look at Example 7 and Example 8. In those two examples we solved homogeneous (and thats important!) BVPs in the form,
r r Ax = l x
y + l y = 0
y (0) = 0
y ( 2p ) = 0
(1)
In Example 7 we had l = 4 and we found nontrivial (i.e. nonzero) solutions to the BVP. In Example 8 we used l = 3 and the only solution was the trivial solution (i.e. y ( t ) = 0 ). So, this homogeneous BVP (recall this also means the boundary conditions are zero) seems to exhibit similar behavior to the behavior in the matrix equation above. There are values of l that will give nontrivial solutions to this BVP and values of l that will only admit the trivial solution. So, for those values of l that give nontrivial solutions well call l an eigenvalue for the BVP and the nontrivial solutions will be called eigenfunctions for the BVP corresponding to the given eigenvalue. We now know that for the homogeneous BVP given in (1) l = 4 is an eigenvalue (with eigenfunctions y ( x ) = c2 sin ( 2 x ) ) and that l = 3 is not an eigenvalue. Eventually well try to determine if there are any other eigenvalues for (1), however before we do that lets comment briefly on why it is so important for the BVP to homogeneous in this discussion. In Example 2 and Example 3 of the previous section we solved the homogeneous differential equation
y + 4 y = 0
with two different nonhomogeneous boundary conditions in the form,
y (0) = a
y ( 2p ) = b
In these two examples we saw that by simply changing the value of a and/or b we were able to get either nontrivial solutions or to force no solution at all. In the discussion of eigenvalues/eigenfunctions we need solutions to exist and the only way to assure this behavior is
to require that the boundary conditions also be homogeneous. In other words, we need for the BVP to be homogeneous. There is one final topic that we need to discuss before we move into the topic of eigenvalues and eigenfunctions and this is more of a notational issue that will help us with some of the work that well need to do. Lets suppose that we have a second order differential equations and its characteristic polynomial has two real, distinct roots and that they are in the form
r1 = a
r x
r2 = - a
Then we know that the solution is,
y ( x ) = c1e 1 + c2e 2 = c1ea x + c2e - a x

r x
While there is nothing wrong with this solution lets do a little rewriting of this. Well start by splitting up the terms as follows,
y ( x ) = c1ea x + c2 e - a x =
c1 a x c1 a x c2 - a x c2 - a x e + e + e + e 2 2 2 2
Now well add/subtract the following terms (note were mixing the ci and a up in the new terms) to get,
y ( x) =
c1 a x c1 a x c2 - a x c2 - a x c1 - a x c1 - a x c2 a x c2 a x e + e + e + e + e - e + e - e 2 2 2 2 2 2 2 2 1 1 c1ea x + c1e - a x + c2 ea x + c2e - a x ) + ( c1ea x - c1e - a x - c2 ea x + c2 e- a x ) ( 2 2
Next, rearrange terms around a little,
y ( x) =
Finally, the quantities in parenthesis factor and well move the location of the fraction as well. Doing this, as well as renaming the new constants we get,
ea x + e - a x ea x - e - a x y ( x ) = ( c1 + c2 ) + ( c1 - c2 ) 2 2 -a x -a x ax ax e +e e -e = c1 + c2 2 2
All this work probably seems very mysterious and unnecessary. However there really was a reason for it. In fact you may have already seen the reason, at least in part. The two new functions that we have in our solution are in fact two of the hyperbolic functions. In particular,
cosh ( x ) =
e x + e- x 2
sinh ( x ) =
e x - e- x 2
So, another way to write the solution to a second order differential equation whose characteristic polynomial has two real, distinct roots in the form r1 = a , r2 = - a is,
y ( x ) = c1 cosh (a x ) + c2 sinh (a x )
Having the solution in this form for some (actually most) of the problems well be looking will make our life a lot easier. The hyperbolic functions have some very nice properties that we can (and will) take advantage of. First, since well be needing them later on, the derivatives are,
d (cosh ( x ) ) = sinh ( x ) dx
d (sinh ( x ) ) = cosh ( x ) dx
Next lets take a quick look at the graphs of these functions.
Note that cosh ( 0 ) = 1 and sinh ( 0 ) = 0 . Because well often be working with boundary conditions at x = 0 these will be useful evaluations. Next, and possibly more importantly, lets notice that cosh ( x ) > 0 for all x and so the hyperbolic cosine will never be zero. Likewise, we can see that sinh ( x ) = 0 only if x = 0 . We will be using both of these facts in some of our work so we shouldnt forget them. Okay, now that weve got all that out of the way lets work an example to see how we go about finding eigenvalues/eigenfunctions for a BVP.
Example 1 Find all the eigenvalues and eigenfunctions for the following BVP. y + l y = 0 y (0) = 0 y ( 2p ) = 0
Solution We started off this section looking at this BVP and we already know one eigenvalue ( l = 4 ) and we know one value of l that is not an eigenvalue ( l = 3 ). As we go through the work here we need to remember that we will get an eigenvalue for a particular value of l if we get non-trivial solutions of the BVP for that particular value of l . In order to know that weve found all the eigenvalues we cant just start randomly trying values of l to see if we get non-trivial solutions or not. Luckily there is a way to do this thats not too bad and will give us all the eigenvalues/eigenfunctions. We are going to have to do some cases
however. The three cases that we will need to look at are : l > 0 , l = 0 , and l < 0 . Each of these cases gives a specific form of the solution to the BVP which we can then apply the boundary conditions to see if well get non-trivial solutions or not. So, lets get started on the cases.
l >0
In this case the characteristic polynomial we get from the differential equation is,
r2 + l = 0
r1,2 = -l
In this case since we know that l > 0 these roots are complex and we can write them instead as,
r1,2 = l i
The general solution to the differential equation is then,
y ( x ) = c1 cos
l x + c2 sin
lx
Applying the first boundary condition gives us,
0 = y ( 0 ) = c1
So, taking this into account and applying the second boundary condition we get,
0 = y ( 2p ) = c2 sin 2p l
This means that we have to have one of the following,
c2 = 0
or
sin 2p l = 0
However, recall that we want non-trivial solutions and if we have the first possibility we will get the trivial solution for all values of l > 0 . Therefore lets assume that c2 0 . This means that we have,
sin 2p l = 0
2p l = np
n = 1, 2,3,K
In other words, taking advantage of the fact that we know where sine is zero we can arrive at the second equation. Also note that because we are assuming that l > 0 we know that
2p l > 0 and so n can only be a positive integer for this case.

Now all we have to do is solve this for l and well have all the positive eigenvalues for this BVP. The positive eigenvalues are then,
2 n n ln = = 4 2 2
n = 1, 2,3,K
and the eigenfunctions that correspond to these eigenvalues are,
2007 Paul Dawkins
384
nx yn ( x ) = sin 2
n = 1, 2,3,K
Note that we subscripted an n on the eigenvalues and eigenfunctions to denote the fact that there is one for each of the given values of n. Also note that we dropped the c2 on the eigenfunctions. For eigenfunctions we are only interested in the function itself and not the constant in front of it and so we generally drop that. Lets now move into the second case.
l =0
In this case the BVP becomes,
y = 0
y (0) = 0 y ( x ) = c1 + c2 x 0 = y ( 0 ) = c1
y ( 2p ) = 0
and integrating the differential equation a couple of times gives us the general solution,
Applying the first boundary condition gives,
Applying the second boundary condition as well as the results of the first boundary condition gives,
0 = y ( 2p ) = 2c2p
Here, unlike the first case, we dont have a choice on how to make this zero. This will only be zero if c2 = 0 . Therefore, for this BVP (and thats important), if we have l = 0 the only solution is the trivial solution and so l = 0 cannot be an eigenvalue for this BVP. Now lets look at the final case.
l<0
In this case the characteristic equation and its roots are the same as in the first case. So, we know that,
r1,2 = -l
However, because we are assuming l < 0 here these are now two real distinct roots and so using our work above for these kinds of real, distinct roots we know that the general solution will be,
y ( x ) = c1 cosh
-l x + c2 sinh
-l x
Note that we could have used the exponential form of the solution here, but our work will be significantly easier if we use the hyperbolic form of the solution here. Now, applying the first boundary condition gives,
0 = y ( 0 ) = c1 cosh ( 0 ) + c2 sinh ( 0 ) = c1 (1) + c2 ( 0 ) = c1

Applying the second boundary condition gives,
c1 = 0
0 = y ( 2p ) = c2 sinh 2p -l
Because we are assuming l < 0 we know that 2p -l 0 and so we also know that
sinh 2p -l 0 . Therefore, much like the second case, we must have c2 = 0 .

So, for this BVP (again thats important), if we have l < 0 we only get the trivial solution and so there are no negative eigenvalues. In summary then we will have the following eigenvalues/eigenfunctions for this BVP.
ln =
n2 4
nx yn ( x ) = sin 2
n = 1, 2,3,K
Lets take a look at another example with slightly different boundary conditions.
Example 2 Find all the eigenvalues and eigenfunctions for the following BVP. y + l y = 0 y ( 0 ) = 0 y ( 2p ) = 0
Solution Here we are going to work with derivative boundary conditions. The work is pretty much identical to the previous example however so we wont put in quite as much detail here. Well need to go through all three cases just as the previous example so lets get started on that.
l >0
The general solution to the differential equation is identical to the previous example and so we have,
y ( x ) = c1 cos 0 = y ( 0 ) = l c2
l x + c2 sin
lx
c2 = 0
Recall that we are assuming that l > 0 here and so this will only be zero if c2 = 0 . Now, the second boundary condition gives us,
0 = y ( 2p ) = - l c1 sin 2p l
Recall that we dont want trivial solutions and that l > 0 so we will only get non-trivial solution if we require that,
sin 2p l = 0
2p l = np
n = 1, 2,3,K
2007 Paul Dawkins
386
Solving for l and we see that we get exactly the same positive eigenvalues for this BVP that we got in the previous example.
2 n n ln = = 4 2 2
n = 1, 2,3,K
The eigenfunctions that correspond to these eigenvalues however are,
nx yn ( x ) = cos 2
n = 1, 2,3,K
So, for this BVP we get cosines for eigenfunctions corresponding to positive eigenvalues. Now the second case.
l =0
y ( x ) = c1 + c2 x 0 = y ( 0 ) = c2 y ( x ) = c1 0 = y ( 2p ) = 0
Using this the general solution is then,
and note that this will trivially satisfy the second boundary condition,
Therefore, unlike the first example, l = 0 is an eigenvalue for this BVP and the eigenfunctions corresponding to this eigenvalue is,
y ( x) = 1
Again, note that we dropped the arbitrary constant for the eigenfunctions. Finally lets take care of the third case.
l<0
The general solution here is,
y ( x ) = c1 cosh
-l x + c2 sinh
-l x
)
c2 = 0
0 = y ( 0 ) = -l c1 sinh ( 0 ) + -l c2 cosh ( 0 ) = -l c2 0 = y ( 2p ) = -l c1 sinh 2p -l
2007 Paul Dawkins
387
As with the previous example we again know that 2p -l 0 and so sinh 2p -l 0 . Therefore we must have c1 = 0 . So, for this BVP we again have no negative eigenvalues. In summary then we will have the following eigenvalues/eigenfunctions for this BVP.
ln =
n2 4
l0 = 0
nx yn ( x ) = cos 2 y0 ( x ) = 1
n = 1, 2,3,K
Notice as well that we can actually combine these if we allow the list of ns for the first one to start at zero instead of one. This will often not happen, but when it does well take advantage of it. So the official list of eigenvalues/eigenfunctions for this BVP is,
ln =
n2 4
nx yn ( x ) = cos 2
n = 0,1, 2, 3, K
So, in the previous two examples we saw that we generally need to consider cases for l as different values will often lead to different general solutions. Do not get too locked into the cases we did here. We will mostly be solving this particular differential equation and so it will be tempting to assume that these are always the cases that well be looking at, but there are BVPs that will require other/different cases. Also, as we saw in the two examples sometimes one or more of the cases will not yield any eigenvalues. This will often happen, but again we shouldnt read anything into the fact that we didnt have negative eigenvalues for either of these two BVPs. There are BVPs that will have negative eigenvalues. Lets take a look at another example with a very different set of boundary conditions. These are not the traditional boundary conditions that weve been looking at to this point, but well see in the next chapter how these can arise from certain physical problems.
Example 3 Find all the eigenvalues and eigenfunctions for the following BVP. y + l y = 0 y ( -p ) = y (p ) y ( -p ) = y (p )
Solution So, in this example we arent actually going to specify the solution or its derivative at the boundaries. Instead well simply specify that the solution must be the same at the two boundaries and the derivative of the solution must also be the same at the two boundaries. Also, this type of boundary condition will typically be on an interval of the form [-L,L] instead of [0,L] as weve been working on to this point. As mentioned above this kind of boundary conditions arise very naturally in certain physical problems and well see that in the next chapter. As with the previous two examples we still have the standard three cases to look at.
l >0
The general solution for this case is,
y ( x ) = c1 cos
l x + c2 sin
lx
Applying the first boundary condition and using the fact that cosine is an even function (i.e. cos ( - x ) = cos ( x ) ) and that sine is an odd function (i.e. sin ( - x ) = - sin ( x ) ). gives us,
c1 cos -p l + c2 sin -p l = c1 cos p l + c2 sin p l

1 2 1 2
) ( ) ( ) c cos (p l ) - c sin (p l ) = c cos (p l ) + c sin (p l ) -c sin (p l ) = c sin (p l ) 0 = 2c sin (p l )

2 2 2
This time, unlike the previous two examples this doesnt really tell us anything. We could have
sin p l = 0 but it is also completely possible, at this point in the problem anyway, for us to
have c2 = 0 as well. So, lets go ahead and apply the second boundary condition and see if we get anything out of that.
- l c1 sin -p l + l c2 cos -p l = - l c1 sin p l + l c2 cos p l

1 2 1 2
) l c sin (p l ) + l c cos (p l ) = l c sin (p l ) = 2 l c sin (p l ) = 0

1 1
( ) l c sin (p l ) + l c sin (p l )
1
( ) l c cos (p l )
So, we get something very similar to what we got after applying the first boundary condition. Since we are assuming that l > 0 this tells us that either sin p l = 0 or c1 = 0 . Note however that if sin p l 0 then we will have to have c1 = c2 = 0 and well get the trivial solution. We therefore need to require that sin p l = 0 and so just as weve done for the previous two examples we can now get the eigenvalues,
p l = np
l = n 2 n = 1, 2,3,K
Recalling that l > 0 and we can see that we do need to start the list of possible ns at one instead of zero. So, we now know the eigenvalues for this case, but what about the eigenfunctions. The solution for a given eigenvalue is,
y ( x ) = c1 cos ( n x ) + c2 sin ( n x )
and weve got no reason to believe that either of the two constants are zero or non-zero for that matter. In cases like these we get two sets of eigenfunctions, one corresponding to each constant. The two sets of eigenfunctions for this case are,
yn ( x ) = cos ( n x )
yn ( x ) = sin ( n x )
n = 1, 2,3,K
Now the second case.
l =0
y ( x ) = c1 + c2 x
c1 + c2 ( -p ) = c1 + c2 (p ) 2p c2 = 0 y ( x ) = c1 c2 = 0
and note that this will trivially satisfy the second boundary condition just as we saw in the second example above. Therefore we again have l = 0 as an eigenvalue for this BVP and the eigenfunctions corresponding to this eigenvalue is,
y ( x) = 1
Finally lets take care of the third case.
l<0
y ( x ) = c1 cosh
-l x + c2 sinh
-l x
) ( )
Applying the first boundary condition and using the fact that hyperbolic cosine is even and hyperbolic sine is odd gives,
c1 cosh -p - l + c2 sinh -p -l = c1 cosh p -l + c2 sinh p - l

2 2
( -c sinh ( -p 2c sinh (p
2
) ( -l ) = c sinh (p -l ) = 0
) -l )
Now, in this case we are assuming that l < 0 and so we know that p -l 0 which in turn tells us that sinh p -l 0 . We therefore must have c2 = 0 . Lets now apply the second boundary condition to get,
- l c1 sinh -p -l = -l c1 sinh p -l 2
1
) -l c sinh (p -l ) = 0
)
c1 = 0
By our assumption on l we again have no choice here but to have c1 = 0 . Therefore, in this case the only solution is the trivial solution and so, for this BVP we again have no negative eigenvalues. In summary then we will have the following eigenvalues/eigenfunctions for this BVP.
l n = n2 l n = n2 l0 = 0
yn ( x ) = sin ( n x ) yn ( x ) = cos ( n x ) y0 ( x ) = 1
n = 1, 2,3,K n = 1, 2,3,K
Note that weve acknowledged that for l > 0 we had two sets of eigenfunctions by listing them each separately. Also, we can again combine the last two into one set of eigenvalues and eigenfunctions. Doing so gives the following set of eigenvalues and eigenfunctions.
l n = n2 l n = n2
yn ( x ) = sin ( n x )
yn ( x ) = cos ( n x )
n = 1, 2,3,K
n = 0,1, 2, 3, K
Once again weve got an example with no negative eigenvalues. We cant stress enough that this is more a function of the differential equation were working with than anything and there will be examples in which we may get negative eigenvalues. Now, to this point weve only worked with one differential equation so lets work an example with a different differential equation just to make sure that we dont get too locked into this one differential equation. Before working this example lets note that we will still be working the vast majority of our examples with the one differential equation weve been using to this point. Were working with this other differential equation just to make sure that we dont get too locked into using one single differential equation.
Example 4 Find all the eigenvalues and eigenfunctions for the following BVP. x 2 y + 3 xy + l y = 0 y (1) = 0 y (2) = 0
Solution This is an Euler differential equation and so we know that well need to find the roots of the following quadratic.
r ( r - 1) + 3r + l = r 2 + 2r + l = 0
The roots to this quadratic are,
2007 Paul Dawkins
391
r1,2 =
-2 4 - 4l = -1 1 - l 2
Now, we are going to again have some cases to work with here, however they wont be the same as the previous examples. The solution will depend on whether or not the roots are real distinct, double or complex and these cases will depend upon the sign/value of 1 - l . So, lets go through the cases.
1 - l < 0, l > 1
In this case the roots will be complex and well need to write them as follows in order to write down the solution.
r1,2 = -1 1 - l = -1 - ( l - 1) = -1 i l - 1
By writing the roots in this fashion we know that l - 1 > 0 and so which we need in order to write the following solution,
l - 1 is now a real number,
y ( x ) = c1 x -1 cos ln ( x ) l - 1 + c2 x -1 sin ln ( x ) l - 1
0 = y (1) = c1 cos ( 0 ) + c2 sin ( 0 ) = c1
c1 = 0
The second boundary condition gives us,
0 = y ( 2) = 1 2 c2 sin ln ( 2 ) l - 1
In order to avoid the trivial solution for this case well require,
)
n = 1, 2,3,K
sin ln ( 2 ) l - 1 = 0
ln ( 2 ) l - 1 = np
This is much more complicated of a condition than weve seen to this point, but other than that we do the same thing. So, solving for l gives us the following set of eigenvalues for this case.
np ln = 1+ ln 2
n = 1, 2,3,K
Note that we need to start the list of ns off at one and not zero to make sure that we have l > 1 as were assuming for this case. The eigenfunctions that correspond to these eigenvalues are,
np yn ( x ) = sin ln ( x ) ln 2
Now the second case.
n = 1, 2,3,K
2007 Paul Dawkins
392
1 - l = 0, l = 1
In this case we get a double root of r1,2 = -1 and so the solution is,
y ( x ) = c1 x -1 + c2 x -1 ln ( x )
0 = y (1) = c1 c2 = 0
The second boundary condition gives,
0 = y ( 2) = 1 2 c2 ln ( 2 )
We therefore have only the trivial solution for this case and so l = 1 is not an eigenvalue. Lets now take care of the third (and final) case.
1 - l > 0, l < 1
This case will have two real distinct roots and the solution is,
y ( x ) = c1 x -1+
1- l
+ c2 x -1
1- l
0 = y (1) = c1 + c2
c2 = -c1
Using this our solution becomes,
y ( x ) = c1 x -1+
1- l
- c1 x -1-
1- l
0 = y ( 2 ) = c1 2-1+
1- l
- c1 2-1-
1- l
= c1 2-1+
1- l
- 2-1-
1- l
Now, because we know that l 1 for this case the exponents on the two terms in the parenthesis are not the same and so the term in the parenthesis is not the zero. This means that we can only have,
c1 = c2 = 0
and so in this case we only have the trivial solution and there are no eigenvalues for which l < 1 . The only eigenvalues for this BVP then come from the first case. So, weve now worked an example using a differential equation other than the standard one weve been using to this point. As we saw in the work however, the basic process was pretty much the same. We determined that there were a number of cases (three here, but it wont always be three) that gave different solutions. We examined each case to determine if non-trivial solutions were possible and if so found the eigenvalues and eigenfunctions corresponding to that case.
2007 Paul Dawkins
393
We need to work one last example in this section before we leave this section for some new topics. The four examples that weve worked to this point were all fairly simple (with simple being relative of course), however we dont want to leave without acknowledging that many eigenvalue/eigenfunctions problems are so easy. In many examples it is not even possible to get a complete list of all possible eigenvalues for a BVP. Often the equations that we need to solve to get the eigenvalues are difficult if not impossible to solve exactly. So, lets take a look at one example like this to see what kinds of things can be done to at least get an idea of what the eigenvalues look like in these kinds of cases.
Example 5 Find all the eigenvalues and eigenfunctions for the following BVP. y + l y = 0 y (0) = 0 y (1) + y (1) = 0
Solution The boundary conditions for this BVP are fairly different from those that weve worked with to this point. However, the basic process is the same. So lets start off with the first case.
l >0
The general solution to the differential equation is identical to the first few examples and so we have,
y ( x ) = c1 cos 0 = y ( 0 ) = c1
l x + c2 sin
lx
c1 = 0
The second boundary condition gives us,
0 = y (1) + y (1) = c2 sin

2
( l )+ = c ( sin ( l ) +
l c2 cos
( l) l cos ( l ) )
So, if we let c2 = 0 well get the trivial solution and so in order to satisfy this boundary condition well need to require instead that,
0 = sin
( l)=tan ( l ) = sin
( l ) + l cos ( l ) l cos ( l )
l
Now, this equation has solutions but well need to use some numerical techniques in order to get them. In order to see whats going on here lets graph tan
( l ) and -
l on the same graph.
Here is that graph and note that the horizontal axis really is values of l as that will make things a little easier to see and relate to values that were familiar with.
2007 Paul Dawkins
394
So, eigenvalues for this case will occur where the two curves intersect. Weve shown the first five on the graph and again what is showing on the graph is really the square root of the actual eigenvalue as weve noted. The interesting thing to note here is that the farther out on the graph the closer the eigenvalues come to the asymptotes of tangent and so well take advantage of that and say that for large enough n we can approximate the eigenvalues with the (very well known) locations of the asymptotes of tangent. How large the value of n is before we start using the approximation will depend on how much accuracy we want, but since we know the location of the asymptotes and as n increases the accuracy of the approximation will increase so it will be easy enough to check for a given accuracy. For the purposes of this example we found the first five numerically and then well use the approximation of the remaining eigenvalues. Here are those values/approximations.
l1 = 2.0288 l 2 = 4.9132 l 3 = 7.9787 l 4 = 11.0855 l 5 = 14.2074 ln

2 n -1 2
l1 = 4.1160 l 2 = 24.1395 l 3 = 63.6597 l 4 = 122.8883 l 5 = 201.8502 ln (

2 n -1) 4
2
( 2.4674 ) ( 22.2066 ) ( 61.6850 ) (120.9027 ) (199.8595 )

n = 6, 7,8,K
p2
The number in parenthesis after the first five is the approximate value of the asymptote. As we can see they are a little off, but by the time we get to n = 5 the error in the approximation is 0.9862%. So less than 1% error by the time we get to n = 5 and it will only get better for larger value of n.
The eigenfunctions for this case are,
yn ( x ) = sin
ln x
n = 1, 2,3,K
where the values of l n are given above. So, now that all that work is out of the way lets take a look at the second case.
l =0
y ( x ) = c1 + c2 x 0 = y ( 0 ) = c1 y ( x ) = c2 x c2 = 0
Applying the second boundary condition to this gives,
0 = y (1) + y (1) = c2 + c2 = 2c2
Therefore for this case we get only the trivial solution and so l = 0 is not an eigenvalue. Note however that had the second boundary condition been y (1) - y (1) = 0 then l = 0 would have been an eigenvalue (with eigenfunctions y ( x ) = x ) and so again we need to be careful about reading too much into our work here. Finally lets take care of the third case.
l<0
y ( x ) = c1 cosh
-l x + c2 sinh
-l x
)
c1 = 0
0 = y ( 0 ) = c1 cosh ( 0 ) + c2 sinh ( 0 ) = c1 y ( x ) = c2 sinh
Using this the general solution becomes,
( (
-l x
) (
-l -l
Applying the second boundary condition to this gives,
0 = y (1) + y (1) = - l c2 cosh = c2

2007 Paul Dawkins
-l + c2 sinh
-l cosh
-l + sinh
))
396
Now, by assumption we know that l < 0 and so
sinh
-l > 0 and we know that cosh ( x ) > 0 for all x. Therefore, - l cosh
-l > 0 . This in turn tells us that
-l + sinh
-l 0
and so we must have c2 = 0 and once again in this third case we get the trivial solution and so this BVP will have no negative eigenvalues. In summary the only eigenvalues for this BVP come from assuming that l > 0 and they are given above. So, weve worked several eigenvalue/eigenfunctions examples in this section. Before leaving this section we do need to note once again that there are a vast variety of different problems that we can work here and weve really only shown a bare handful of examples and so please do now walk away from this section believing that weve shown you everything. The whole purpose of this section is to prepare us for the types of problems that well be seeing in the next chapter. Also, in the next chapter we will again be restricting ourselves down to some pretty basic and simple problems in order to illustrate one of the more common methods for solving partial differential equations.
2007 Paul Dawkins
397
Periodic Functions, Even/Odd Functions and Orthogonal Functions

This is going to be a short section. We just need to have a brief discussion about a couple of ideas that well be dealing with on occasion as we move into the next topic of this chapter. Periodic Function The first topic we need to discuss is that of a periodic function. A function is said to be periodic with period T if the following is true,
f ( x + T ) = f ( x)
for all x
The following is a nice little fact about periodic functions. Fact 1 If f and g are both periodic functions with period T then so is f + g and fg. This is easy enough to prove so lets do that.
( f + g )( x + T ) = f ( x + T ) + g ( x + T ) = f ( x ) + g ( x ) = ( f + g )( x ) ( fg )( x + T ) = f ( x + T ) g ( x + T ) = f ( x ) g ( x ) = ( fg )( x )
The two periodic functions that most of us are familiar are sine and cosine and in fact well be using these two functions regularly in the remaining sections of this chapter. So, having said that lets close off this discussion of periodic functions with the following fact, Fact 2
sin (w x ) and cos (w x ) are periodic functions with period T =
2p . w
Even and Odd Functions The next quick idea that we need to discuss is that of even and odd functions. Recall that a function is said to be even if, and a function is said to be odd if,
f (-x) = f ( x)
f (-x) = - f ( x)
The standard examples of even functions are f ( x ) = x 2 and g ( x ) = cos ( x ) while the standard examples of odd functions are f ( x ) = x 3 and g ( x ) = sin ( x ) . The following fact about certain integrals of even/odd functions will be useful in some of our work.
2007 Paul Dawkins
398
Fact 3 1. If f ( x ) is an even function then,
2. If f ( x ) is an odd function then,
-L
f ( x ) dx = 2 f ( x ) dx
0
-L
f ( x ) dx = 0
Note that this fact is only valid on a symmetric interval, i.e. an interval in the form [ - L, L ] . If we arent integrating on a symmetric interval then the fact may or may not be true. Orthogonal Functions The final topic that we need to discuss here is that of orthogonal functions. This idea will be integral to what well be doing in the remainder of this chapter and in the next chapter as we discuss one of the basic solution methods for partial differential equations. Lets first get the definition of orthogonal functions out of the way. Definition 1. Two non-zero functions, f ( x ) and g ( x ) , are said to be orthogonal on a x b if,
a f ( x ) g ( x ) dx = 0
2. A set of non-zero functions, In other words,
{ f ( x )} , is said to be mutually orthogonal on a x b (or just

i
an orthogonal set if were being lazy) if fi ( x ) and f j ( x ) are orthogonal for every i j .
i j 0 f i ( x ) f j ( x ) dx = c > 0 i = j
Note that in the case of i = j for the second definition we know that well get a positive value from the integral because,
f ( x ) a f ( x ) f ( x ) dx = a
i i i
dx > 0
Recall that when we integrate a positive function we know the result will be positive as well. Also note that the non-zero requirement is important because otherwise the integral would be trivially zero regardless of the other function we were using. Before we work some examples there are a nice set of trig formulas that well need to help us with some of the integrals.
2007 Paul Dawkins
399
1 sin (a - b ) + sin (a + b ) 2 1 sin a sin b = cos (a - b ) - cos (a + b ) 2 1 cos a cos b = cos (a - b ) + cos (a + b ) 2 sin a cos b =
Now lets work some examples that well need over the course of the next couple of sections.
np x Example 1 Show that cos is mutually orthogonal on - L x L . L n = 0

Solution This is not too difficult to do. All we really need to do is evaluate the following integral.
np x mp x cos cos dx - L L L
Before we start evaluating this integral lets notice that the integrand is the product of two even functions and so must also be even. This means that we can use Fact 3 above to write the integral as,
np x mp x np x mp x cos cos dx = 2 cos cos dx L L L L - L 0

There are two reasons for doing this. First having a limit of zero will often make the evaluation step a little easier and that will be the case here. Well discuss the second reason after were done with the example. Now, in order to do this integral well actually need to consider three cases.
n=m=0
In this case the integral is very easy and is,
n=m0
-L
dx = 2 dx = 2 L
0
This integral is a little harder than the first case, but not by much (provided we recall a simple trig formula). The integral for this case is,
2np x 2 np x 2 np x cos dx = 2 cos dx = 1 + cos dx - L L 0 L 0 L L L 2np x =x+ sin sin ( 2np ) = L + 2np 2np L 0
L
Now, at the point we need to recall that n is an integer and so sin ( 2np ) = 0 and our final value for the is,
2 np x 2 np x cos dx = 2 cos dx = L - L L 0 L
The first two cases are really just showing that if n = m the integral is not zero (as it shouldnt be) and depending upon the value of n (and hence m) we get different values of the integral. Now we need to do the third case and this, in some ways, is the important case since we must get zero out of this integral in order to know that the set is an orthogonal set. So, lets take care of the final case.
nm
This integral is the messiest of the three that weve had to do here. Lets just start off by writing the integral down.
L L
np x mp x np x mp x cos cos dx = 2 cos cos dx L L L L - L 0

In this case we cant combine/simplify as we did in the previous two cases. We can however, acknowledge that weve got a product of two cosines with different arguments and so we can use one of the trig formulas above to break up the product as follows,
np x mp x np x mp x cos dx = 2 cos cos cos dx - L L L 0 L L (n - m)p x ( n + m )p x = cos + cos dx L L 0

L
( n - m)p x ( n + m )p x L L = sin sin + L L ( n - m )p 0 ( n + m)p L L = sin ( ( n - m ) p ) + sin ( ( n + m ) p ) ( n - m)p (n + m)p
Now, we know that n and m are both integers and so n - m and n + m are also integers and so both of the sines above must be zero and all together we get,
np x mp x np x mp x cos cos dx = 2 cos cos dx = 0 - L L L 0 L L

So, weve shown that if n m the integral is zero and if n = m the value of the integral is a positive constant and so the set is mutually orthogonal. In all of the work above we kept both forms of the integral at every step. Lets discuss why we did this a little bit. By keeping both forms of the integral around we were able to show that not
2007 Paul Dawkins
401
np x only is cos mutually orthogonal on - L x L but it is also mutually orthogonal L n = 0 on 0 x L . The only difference is the value of the integral when n = m and we can get those
values from the work above. Lets take a look at another example.
np x is mutually orthogonal on - L x L and on Example 2 Show that sin L n = 1 0 x L.

Solution First well acknowledge from the start this time that well be showing orthogonality on both of the intervals. Second, we need to start this set at n = 1 because if we used n = 0 the first function would be zero and we dont want the zero function to show up on our list. As with the first example all we really need to do is evaluate the following integral.
np x mp x sin sin dx - L L L
In this case integrand is the product of two odd functions and so must be even. This means that we can again use Fact 3 above to write the integral as,
np x mp x np x mp x sin sin dx = 2 sin sin dx - L L L 0 L L

We only have two cases to do for the integral here.
n=m
Not much to this integral. Its pretty similar to the previous examples second case.
2np x 2 np x 2 np x sin dx = 2 sin dx = 1 - cos dx - L 0 0 L L L L L 2np x =xsin sin ( 2np ) = L = L 2np 2np L 0
Summarizing up we get,
L
2 np x 2 np x sin dx = 2 sin dx = L - L L 0 L nm
As with the previous example this can be a little messier but it is also nearly identical to the third case from the previous example so well not show a lot of the work.
np x mp x np x mp x sin sin dx = 2 sin sin dx - L L L 0 L L ( n - m )p x ( n + m )p x = cos - cos dx L L 0

L
(n - m)p x ( n + m )p x L L = sin sin L L ( n + m )p ( n - m)p 0 L L = sin ( ( n - m ) p ) sin ( ( n + m ) p ) ( n - m)p (n + m)p

As with the previous example we know that n and m are both integers a and so both of the sines above must be zero and all together we get,
np x mp x np x mp x sin sin dx = 2 sin sin dx = 0 L L - L L L 0

So, weve shown that if n m the integral is zero and if n = m the value of the integral is a positive constant and so the set is mutually orthogonal.
np x Weve now shown that sin mutually orthogonal on - L x L and on L n = 1 0 x L.

We need to work one more example in this section.
np x np x Example 3 Show that sin and cos are mutually orthogonal on L n = 1 L n = 0 -L x L .

Solution This example is a little different from the previous two examples. Here we want to show that together both sets are mutually orthogonal on - L x L . To show this we need to show three things. First (and second actually) we need to show that individually each set is mutually orthogonal and weve already done that in the previous two examples. The third (and only) thing we need to show here is that if we take one function from one set and another function from the other set and we integrate them well get zero. Also, note that this time we really do only want to do the one interval as the two sets, taken together, are not mutually orthogonal on 0 x L . You might want to do the integral on this interval to verify that it wont always be zero. So, lets take care of the one integral that we need to do here and there isnt a lot to do. Here is
the integral.
np x mp x sin cos dx - L L L
The integrand in this case is the product of an odd function (the sine) and an even function (the cosine) and so the integrand is an odd function. Therefore, since the integral is on a symmetric interval, i.e. - L x L , and so by Fact 3 above we know the integral must be zero or,
np x mp x sin cos dx = 0 - L L L
So, in previous examples weve shown that on the interval - L x L the two sets are mutually orthogonal individually and here weve shown that integrating a product of a sine and a cosine gives zero. Therefore, as a combined set they are also mutually orthogonal. Weve now worked three examples here dealing with orthogonality and we should note that these were not just pulled out of the air as random examples to work. In the following sections (and following chapter) well need the results from these examples. So, lets summarize those results up here.
np x np x 1. cos and sin are mutually orthogonal on - L x L as L n = 0 L n = 1

individual sets and as a combined set.
np x 2. cos is mutually orthogonal on 0 x L . L n = 0
np x 3. sin is mutually orthogonal on 0 x L . L n = 1

We will also be needing the results of the integrals themselves, both on - L x L and on 0 x L so lets also summarize those up here as well so we can refer to them when we need to.
2007 Paul Dawkins
404
2 L if n = m = 0 L np x mp x 1. cos cos dx = L if n = m 0 - L L L 0 if n m L L np x mp x 2. cos cos dx = 0 L L 2 0

L L
if n = m = 0 if n = m 0 if n m
3. sin
- L
L if n = m np x mp x sin dx = 0 if n m L L
L L np x mp x 4. sin sin dx = 2 0 L L 0 np x mp x 5. sin cos dx = 0 L - L L

L
if n = m if n m
With this summary well leave this section and move off into the second major topic of this chapter : Fourier Series.
2007 Paul Dawkins
405
Fourier Sine Series

In this section we are going to start taking a look at Fourier series. We should point out that this is a subject that can span a whole class and what well be doing in this section (as well as the next couple of sections) is intended to be nothing more than a very brief look at the subject. The point here is to do just enough to allow us to do some basic solutions to partial differential equations in the next chapter. There are many topics in the study of Fourier series that well not even touch upon here. So, with that out of the way lets get started, although were not going to start off with Fourier series. Lets instead think back to our Calculus class where we looked at Taylor Series. With Taylor Series we wrote a series representation of a function, f ( x ) , as a series whose terms were powers of x - a for some x = a . With some conditions we were able to show that,
f ( n) ( a ) n f ( x) = ( x - a) n! n =0 and that the series will converge to f ( x ) on x - a < R for some R that will be dependent upon
the function itself. There is nothing wrong with this, but it does require that derivatives of all orders exist at x = a .
n Or in other words f ( ) ( a ) exists for n = 0,1, 2,3, K Also for some functions the value of R
may end up being quite small. These two issues (along with a couple of others) mean that this is not always the best way or writing a series representation for a function. In many cases it works fine and there will be no reason to need a different kind of series. There are times however where another type of series is either preferable or required. Were going to build up an alternative series representation for a function over the course of the next couple of sections. The ultimate goal for the rest of this chapter will be to write down a series representation for a function in terms of sines and cosines. Well start things off by assuming that the function, f ( x ) , we want to write a series representation for is an odd function (i.e. f ( - x ) = - f ( x ) ). Because f ( x ) is odd it makes some sense that should be able to write a series representation for this in terms of sines only (since they are also odd functions). What well try to do here is write f ( x ) as the following series representation, called a Fourier sine series, on - L x L .
B
n =1
np x sin L
There are a couple of issues to note here. First, at this point, we are going to assume that the series representation will converge to f ( x ) on - L x L . We will be looking into whether or not it will actually converge in a later section. However, assuming that the series does converge
to f ( x ) it is interesting to note that, unlike Taylor Series, this representation will always converge on the same interval and that the interval does not depend upon the function. Second, the series representation will not involve powers of sine (again contrasting this with Taylor Series) but instead will involve sines with different arguments.
x , may seem like an odd choice that was arbitrarily chosen Finally, the argument of the sines, np L
and in some ways it was. For Fourier sine series the argument doesnt have to necessarily be this but there are several reasons for the choice here. First, this is the argument that will naturally arise in the next chapter when we use Fourier series (in general and not necessarily Fourier sine series) to help us solve some basic partial differential equations. The next reason for using this argument is that fact that the set of functions that we chose to work
x with, sin ( np L )
}n
=1
in this case, need to be orthogonal on the given interval, - L x L in this
case, and note that in the last section we showed that in fact they are. In other words, the choice of functions were going to be working with and the interval were working on will be tied together in some way. We can use a different argument, but will need to also choose an interval on which we can prove that the sines (with the different argument) are orthogonal. So, lets start off by assuming that given an odd function, f ( x ) , we can in fact find a Fourier sine series, of the form given above, to represent the function on - L x L . This means we will have,
np x f ( x ) = Bn sin L n =1
As noted above well discuss whether or not this even can be done and if the series representation does in fact converge to the function in later section. At this point were simply going to assume that it can be done. The question now is how to determine the coefficients, Bn , in the series.
x Lets start with the series above and multiply both sides by sin ( mp where m is a fixed integer L )
in the range {1, 2,3, K} . In other words we multiply both sides by any of the sines in the set of sines that were working with here. Doing this gives,
mp x np x mp x f ( x ) sin = Bn sin sin L n=1 L L

Now, lets integrate both sides of this from x = - L to x = L .
L L
mp x np x mp x f x sin dx = ( ) Bn sin L sin dx -L - L n =1 L L

At this point weve got a small issue to deal with. We know from Calculus that an integral of a finite series (more commonly called a finite sum.) is nothing more than the (finite) sum of the integrals of the pieces. In other words for finite series we can interchange an integral and a series.
For infinite series however, we cannot always do this. For some integrals of infinite series we cannot interchange an integral and a series. Luckily enough for us we actually can interchange the integral and the series in this case. Doing this gives and factoring the constant, Bn , out of the integral gives,
mp x np x mp x f x dx = sin ( ) Bn sin L sin dx -L L L n =1 - L L L
np x mp x = Bn sin sin dx -L L L n =1
x Now, recall from the last section we proved that sin ( np L )
}n
=1
is orthogonal on - L x L and
that,
L L if n = m np x mp x sin sin dx = 0 if n m - L L L
So, what does this mean for us. As we work through the various values of n in the series and compute the value of the integrals all but one of the integrals will be zero. The only non-zero integral will come when we have n = m , in which case the integral has the value of L. Therefore, the only non-zero term in the series will come when we have n = m and our equation becomes,
mp x f ( x ) sin dx = B m L L -L
Finally all we need to do is divide by L and we know have an equation for each of the coefficients.
1 mp x B m = f ( x ) sin dx L -L L
m = 1, 2, 3,K
Next, note that because were integrating two odd functions the integrand of this integral is even and so we also know that,
2 mp x B m = f ( x ) sin dx L 0 L
m = 1, 2,3,K
Summarizing all this work up the Fourier sine series of an odd function f ( x ) on - L x L is given by,
1 np x B n = f ( x ) sin dx L -L L 2 np x = f ( x ) sin dx L 0 L
L
n = 1, 2,3,K n = 1, 2,3,K
2007 Paul Dawkins
408
Lets take a quick look at an example.
Example 1 Find the Fourier sine series for f ( x ) = x on - L x L .

Solution First note that the function were working with is in fact an odd function and so this is something we can do. There really isnt much to do here other than to compute the coefficients for f ( x) = x . Here is that work and note that were going to leave the integration by parts details to you to verify. Dont forget that n, L, and p are constants!
2 2 L np x np x np x B n = x sin dx = 2 2 L sin - np x cos L 0 L n p L L L 0 = 2 ( L sin ( n p ) - np L cos ( n p ) ) n p2

2
These integrals can, on occasion, be somewhat messy especially when we use a general L for the endpoints of the interval instead of a specific number. Now, taking advantage of the fact that n is an integer we know that sin ( np ) = 0 and that
cos ( np ) = ( -1) . We therefore have,

n
( -1) 2 L 2 n B n = 2 2 - np L ( -1) = np np
n +1
n = 1, 2, 3K
The Fourier sine series is then,
x=
n =1
( -1)
n +1
2L
np
n p x 2 L ( -1) sin = L p n =1 n
n +1
np x sin L
At this point we should probably point out that well be doing most, if not all, of our work here on a general interval ( - L x L or 0 x L ) instead of intervals with specific numbers for the endpoints. There are a couple of reasons for this. First, it gives a much more general formula that will work for any interval of that form which is always nice. Secondly, when we run into this kind of work in the next chapter it will also be on general intervals so we may as well get used to them now. Now, finding the Fourier sine series of an odd function is fine and good but what if, for some reason, we wanted to find the Fourier sine series for a function that is not odd? To see how to do this were going to have to make a change. The above work was done on the interval - L x L . In the case of a function that is not odd well be working on the interval 0 x L . The reason for this will be made apparent in a bit.
So, we are now going to do is to try to find a series representation for f ( x ) on the interval
0 x L that is in the form,
B
n =1
np x sin L
or in other words,
As we did with the Fourier sine series on - L x L we are going to assume that the series will in fact converge to f ( x ) and well hold off discussing the convergence of the series for a later section. There are two methods of generating formulas for the coefficients, Bn , although well see in a bit that they really the same way, just looked at from different perspectives. The first method is to just ignore the fact that f ( x ) is odd and proceed in the same manner that we did above only this time well take advantage of the fact that we proved in the previous
x section that sin ( np L )
}n
=1
also forms an orthogonal set on 0 x L and that,
L L np x mp x sin sin dx = 2 0 L L 0
if n = m if n m
x So, if we do this then all we need to do is multiply both sides of our series by sin ( mp , integral L )
from 0 to L and interchange the integral and series to get,

mp x np x mp x f x dx = Bn sin sin ( ) sin dx L L L 0 0 n =1 L L
Now, plugging in for the integral we arrive at,
mp x L f ( x ) sin dx = Bm 0 L 2
Upon solving for the coefficient we arrive at,
L
2 mp x Bm = f ( x ) sin dx L 0 L
m = 1, 2,3,K
Note that this is identical to the second form of the coefficients that we arrived at above by assuming f ( x ) was odd and working on the interval - L x L . The fact that we arrived at essentially the same coefficients is not actually all the surprising as well see once weve looked the second method of generating the coefficients.
Before we look at the second method of generating the coefficients we need to take a brief look at another concept. Given a function, f ( x ) , we define the odd extension of f ( x ) to be the new function,
f ( x) g ( x) = - f ( - x )
if 0 x L if - L x 0
Its pretty easy to see that this is an odd function.
g ( - x ) = - f ( - ( - x )) = - f ( x ) = - g ( x )
for 0 < x < L
and we can also know that on 0 x L we have that g ( x ) = f ( x ) . Also note that if f ( x ) is already an odd function then we in fact get g ( x ) = f ( x ) on - L x L . Lets take a quick look at a couple of odd extensions before we proceed any further.
Example 2 Sketch the odd extension of each of the given functions. (a) f ( x ) = L - x on 0 x L [Solution]
(b) f ( x ) = 1 + x 2 on 0 x L [Solution] (c) f ( x ) =
L 2 L x -2
if 0 x L 2 [Solution] if L x L 2
Solution Not much to do with these other than to define the odd extension and then sketch it. (a) f ( x ) = L - x on 0 x L Here is the odd extension of this function.
if 0 x L f ( x) g ( x) = if - L x 0 - f ( - x ) if 0 x L L - x = if - L x 0 - L - x
Below is the graph of both the function and its odd extension. Note that weve put the extension in with a dashed line to make it clear the portion of the function that is being added to allow us to get the odd extension.
2007 Paul Dawkins
411
(b) f ( x ) = 1 + x 2 on 0 x L First note that this is clearly an even function. That does not however mean that we cant define the odd extension for it. The odd extension for this function is,
f ( x) g ( x) = - f ( - x ) 1 + x 2 = 2 -1 - x
if 0 x L if - L x 0 if 0 x L if - L x 0
The sketch of the original function and its odd extension are ,
2007 Paul Dawkins
412
(c) f ( x ) =
L 2 L x -2
if 0 x L 2 if L x L 2 x - L 2 L if 0 x L = 2 L if - L x 0 - 2 L x + 2 if L 2 x L if 0 x L 2 L if - 2 x 0 if - L x - L 2
Lets first write down the odd extension for this function.
f ( x) g ( x) = - f ( - x )
The sketch of the original function and its odd extension are,
With the definition of the odd extension (and a couple of examples) out of the way we can now take a look at the second method for getting formulas for the coefficients of the Fourier sine series for a function f ( x ) on 0 x L . First, given such a function define its odd extension as above. At this point, because g ( x ) is an odd function, we know that on - L x L the Fourier sine series for g ( x ) (and NOT f ( x ) yet) is,
np x g ( x ) = Bn sin L n =1
2 np x B n = g ( x ) sin dx L 0 L
n = 1, 2,3, K
However, because we know that g ( x ) = f ( x ) on 0 x L we can also see that as long as we are on 0 x L we have,
2 np x B n = f ( x ) sin dx L 0 L
n = 1, 2, 3,K
2007 Paul Dawkins
413
So, exactly the same formula for the coefficients regardless of how we arrived at the formula and the second method justifies why they are the same here as they were when we derived them for the Fourier sine series for an odd function. Now, lets find the Fourier sine series for each of the functions that we looked at in Example 2. Note that again we are working on general intervals here instead of specific numbers for the right endpoint to get a more general formula for any interval of this form and because again this is the kind of work well be doing in the next chapter. Also, well again be leaving the actually integration details up to you to verify. In most cases it will involve some fairly simple integration by parts complicated by all the constants (n, L, p , etc.) that show up in the integral.
Example 3 Find the Fourier sine series for f ( x ) = L - x on 0 x L .

Solution There really isnt much to do here other than computing the coefficients so here they are,
Bn =
2 2 np x np x f ( x ) sin dx = ( L - x ) sin dx L 0 L 0 L L
L
L 2 np x n p x = - 2 2 L sin - np ( x - L ) cos L n p L L 0 2L 2 L2 = 2 2 ( np - sin ( np ) ) = L n p np

In the simplification process dont forget that n is an integer.
So, with the coefficients we get the following Fourier sine series for this function.
f ( x) =
2L np x sin L n =1 np
In the next example it is interesting to note that while we started out this section looking only at odd functions were now going to be finding the Fourier sine series of an even function on 0 x L . Recall however that were really finding the Fourier sine series of the odd extension of this function and so were okay.
2007 Paul Dawkins
414
Example 4 Find the Fourier sine series for f ( x ) = 1 + x 2 on 0 x L .

Solution In this case the coefficients are liable to be somewhat messy given the fact that the integrals will involve integration by parts twice. Here is the work for the coefficients.
2 2 np x np x 2 B n = f ( x ) sin dx = (1 + x ) sin dx L 0 L 0 L L 2 L np x n p x = 3 3 2 L2 - n2p 2 (1 + x 2 ) cos + 2 Lnp x sin L n p L L 0
( (
2 L 2 2 2 2 2 2 2 2 3p 3 2 L - n p (1 + L ) cos ( np ) + 2 L np sin ( np ) - ( 2 L - n p ) Ln 2 n = 3 3 2 L2 - n 2p 2 (1 + L2 ) ( -1) - 2 L2 + n 2p 2 np =
As noted above the coefficients are not the most pleasant ones, but there they are. The Fourier sine series for this function is then,
f ( x) =
2 2 n np x 2 L - n 2p 2 (1 + L2 ) ( -1) - 2 L2 + n 2p 2 sin 3 L n =1 n p
3
In the last example of this section well be finding the Fourier sine series of a piecewise function and can definitely complicate the integrals a little but they do show up on occasion and so we need to be able to deal with them.
L Example 5 Find the Fourier sine series for f ( x ) = 2 L x - 2

Solution Here is the integral for the coefficients.
if 0 x L 2 on 0 x L . L if 2 x L
L L 2 2 2 np x np x np x B n = f ( x ) sin dx = f ( x ) sin dx + f ( x ) sin dx L 0 L L L L L 0 2

L
2 2 L np x = dx + sin L L L 0 2 2
L
L np x x - 2 sin L dx
Note that we need to split the integral up because of the piecewise nature of the original function. Lets do the two integrals separately
2 L2 2L np x L L np x np sin dx = - cos 1 - cos = 0 2 L 2 np L 0 2np 2 L L
2007 Paul Dawkins
415
L 2
L np x L x - sin dx = 2 2 np 2 L = L n p2
2
L np x n p x L sin L - np x - 2 cos L L 2
np L np L sin ( np ) - 2 cos ( np ) - L sin 2 n L2 np ( -1) n p =- 2 2 + sin np 2 2

Putting all of this together gives,
2 2 L2 n +1 np x np B n = f ( x ) sin dx = 1 + ( -1) - cos L 0 L 2np L 2 = L np n +1 np 1 + ( -1) - cos 2
2 n p sin + np 2
2 n p + np sin 2 n p x sin L
So, the Fourier sine series for this function is,
f ( x) =
L n =1 np
n +1 np 1 + ( -1) - cos 2
2 np sin + np 2
As the previous two examples has shown the coefficients for these can be quite messy but that will often be the case and so we shouldnt let that get us too excited.
2007 Paul Dawkins
416
Fourier Cosine Series

In this section were going to take a look at Fourier cosine series. Well start off much as we did in the previous section where we looked at Fourier sine series. Lets start by assuming that the function, f ( x ) , well be working with initially is an even function (i.e. f ( - x ) = f ( x ) ) and that we want to write a series representation for this function on - L x L in terms of cosines (which are also even). In other words we are going to look for the following,
np x f ( x ) = An cos L n =0
This series is called a Fourier cosine series and note that in this case (unlike with Fourier sine series) were able to start the series representation at n = 0 since that term will not be zero as it x was with sines. Also, as with Fourier Sine series, the argument of np in the cosines is being L used only because it is the argument that well be running into in the next chapter. The only real requirement here is that the given set of functions were using be orthogonal on the interval were working on. Note as well that were assuming that the series will in fact converge to f ( x ) on - L x L at this point. In a later section well be looking into the convergence of this series in more detail.
x So, to determine a formula for the coefficients, An , well use the fact that cos ( np L )
}n
=0
do
form an orthogonal set on the interval - L x L as we showed in a previous section. In that section we also derived the following formula that well need in a bit.
2 L if n = m = 0 L np x mp x cos cos dx = L if n = m 0 L L - L 0 if n m
Well get a formula for the coefficients in almost exactly the same fashion that we did in the x previous section. Well start with the representation above and multiply both sides by cos ( mp L )
where m is a fixed integer in the range {0,1, 2, 3, K} . Doing this gives,
mp x np x mp x f ( x ) cos = An cos cos L n= 0 L L

Next, we integrate both sides from x = - L to x = L and as we were able to do with the Fourier Sine series we can again interchange the integral and the series.
mp x np x mp x f x dx = cos cos ( ) An cos dx -L L - L n =0 L L np x mp x = An cos cos dx -L L L n =0

L
2007 Paul Dawkins
417
We now know that the all of the integrals on the right side will be zero except when n = m because the set of cosines form an orthogonal set on the interval - L x L . However, we need to be careful about the value of m (or n depending on the letter you want to use). So, after evaluating all of the integrals we arrive at the following set of formulas for the coefficients.
m = 0:
m0:
L
L -L
f ( x ) dx = A0 ( 2 L )
A0 =
1 L f ( x ) dx 2L -L 1 mp x f ( x ) cos dx L -L L
L
mp x f ( x ) cos dx = Am ( L ) -L L
Am =
Summarizing everything up then, the Fourier cosine series of an even function, f ( x ) on
- L x L is given by, np x f ( x ) = An cos L n =0
1 L 2 L - L f ( x ) dx An = L np x 1 f ( x ) cos dx L L L
n=0 n0
Finally, before we work an example, lets notice that because both f ( x ) and the cosines are even the integrand in both of the integrals above is even and so we can write the formulas for the An s as follows,
1 L L 0 f ( x ) dx An = L np x 2 f ( x ) cos dx L L 0
Now lets take a look at an example.
n=0 n0
Example 1 Find the Fourier cosine series for f ( x ) = x 2 on - L x L .

Solution We clearly have an even function here and so all we really need to do is compute the coefficients and they are liable to be a little messy because well need to do integration by parts twice. Well leave most of the actual integration details to you to verify.
A0 =
1 L 1 L 2 1 L3 L2 f x dx = x dx = ( ) = L 0 L 0 L 3 3
2007 Paul Dawkins
An =
2 2 2 np x np x f ( x ) cos dx = x cos dx L 0 L 0 L L
L
2 L np x np x = 3 3 2 Lnp x cos + ( n 2p 2 x 2 - 2 L2 ) sin L n p L L 0 = 2 2 L2 np cos ( n p ) + ( n 2p 2 L2 - 2 L2 ) sin ( n p ) n p3

3 n
4 L2 ( -1) = n 2p 2
The coefficients are then,
n = 1, 2,3,K
L2 A0 = 3
The Fourier cosine series is then,
4 L2 ( -1) , n = 1, 2, 3, K An = n 2p 2
n 2 n
2 4 L ( -1) np x np x L np x cos cos x = An cos = A + A = + 0 n 2 2 L L 3 n=1 n p L n= 0 n =1 2
Note that well often strip out the n = 0 from the series as weve done here because it will almost always be different from the other coefficients and it allows us to actually plug the coefficients into the series. Now, just as we did in the previous section lets ask what we need to do in order to find the Fourier cosine series of a function that is not even. As with Fourier sine series when we make this change well need to move onto the interval 0 x L now instead of - L x L and again well assume that the series will converge to f ( x ) at this point and leave the discussion of the convergence of this series to a later section. We could go through the work to find the coefficients here twice as we did with Fourier sine series, however theres no real reason to. So, while we could redo all the work above to get formulas for the coefficients lets instead go straight to the second method of finding the coefficients. In this case, before we actually proceed with this well need to define the even extension of a function, f ( x ) on - L x L . So, given a function f ( x ) well define the even extension of the function as,
f ( x) g ( x) = f (-x)
if 0 x L if - L x 0
Showing that this is an even function is simple enough.
g ( - x ) = f ( - ( - x )) = f ( x ) = g ( x )
419
for 0 < x < L

2007 Paul Dawkins
and we can see that g ( x ) = f ( x ) on 0 x L and if f ( x ) is already an even function we get
g ( x ) = f ( x ) on - L x L .
Lets take a look at some functions and sketch the even extensions for the functions.
Example 2 Sketch the odd extension of each of the given functions. (a) f ( x ) = L - x on 0 x L [Solution]
(b) f ( x ) = x 3 on 0 x L [Solution] (c) f ( x ) = 2
L L x - 2
if 0 x L 2 [Solution] L if 2 x L
Solution (a) f ( x ) = L - x on 0 x L Here is the even extension of this function.
if 0 x L f ( x) g ( x) = if - L x 0 f (-x) if 0 x L L - x = if - L x 0 L + x
Here is the graph of both the original function and its even extension. Note that weve put the extension in with a dashed line to make it clear the portion of the function that is being added to allow us to get the even extension
2007 Paul Dawkins
420
(b) f ( x ) = x 3 on 0 x L The even extension of this function is,
if 0 x L f ( x) g ( x) = if - L x 0 f (-x) x3 if 0 x L = 3 if - L x 0 - x
The sketch of the function and the even extension is,
(c) f ( x ) =
L 2 L x - 2
if 0 x L 2 L if 2 x L f ( x) g ( x) = f (-x) x - L 2 L 2 = L 2 L - x - 2 if 0 x L if - L x 0 if L 2 x L if 0 x L 2 if - L x 0 2 if - L x - L 2
Here is the even extension of this function,
The sketch of the function and the even extension is,
2007 Paul Dawkins
421
Okay, lets now think about how we can use the even extension of a function to find the Fourier cosine series of any function f ( x ) on 0 x L . So, given a function f ( x ) well let g ( x ) be the even extension as defined above. Now, g ( x ) is an even function on - L x L and so we can write down its Fourier cosine series. This is,
np x g ( x ) = An cos L n =0
1 L L 0 f ( x ) dx An = L 2 np x f ( x ) cos dx L L 0
n=0 n0
and note that well use the second form of the integrals to compute the constants. Now, because we know that on 0 x L we have f ( x ) = g ( x ) and so the Fourier cosine series of f ( x ) on 0 x L is also given by,
np x f ( x ) = An cos L n =0
n=0 n0
Lets take a look at a couple of examples.
2007 Paul Dawkins
422
Example 3 Find the Fourier cosine series for f ( x ) = L - x on 0 x L .

Solution All we need to do is compute the coefficients so here is the work for that,
1 L 1 L L f ( x ) dx = L - x dx = L 0 L 0 2 L L 2 2 np x np x An = f ( x ) cos dx = ( L - x ) cos dx L 0 L 0 L L A0 = 2 L np x np x = 2 2 np ( L - x ) sin - L cos L n p L L 0 = 2 L 2L n +1 - L cos ( n p ) + L ) = 2 2 1 + ( -1) 2 2 ( Ln p np

L
n = 1, 2, 3, K
The Fourier cosine series is then,
L 2L n +1 np x f ( x ) = + 2 2 1 + ( -1) cos 2 n =1 n p L
Note that as we did with the first example in this section we stripped out the A0 term before we plugged in the coefficients. Next, lets find the Fourier cosine series of an odd function. Note that this is doable because we are really finding the Fourier cosine series of the even extension of the function.
Example 4 Find the Fourier cosine series for f ( x ) = x 3 on 0 x L .

Solution The integral for A0 is simple enough but the integral for the rest will be fairly messy as it will require three integration by parts. Well leave most of the details of the actual integration to you to verify. Heres the work,
1 L 1 L 3 L3 f x dx = x dx = ( ) L 0 L 0 4 L L 2 2 3 np x np x An = f ( x ) cos dx = x cos dx L 0 L 0 L L A0 = 2 L np x np x 2 2 2 3 = 4 4 np x ( n 2p 2 x 2 - 6 L2 ) sin + ( 3Ln p x - 6 L ) cos L n p L L 0 = = 2 L 2 2 2 2 3 2 2 3 3 4 4 np L ( n p L - 6 L ) sin ( n p ) + ( 3L n p - 6 L ) cos ( n p ) + 6 L Ln p 2 3L4 6 L3 n n 2 2 2 + n p 2 1 = 2 + ( n2p 2 - 2 ) ( -1) ( ) ( ) 4 4 4 4 Ln p np
423
L
n = 1, 2,3, K
2007 Paul Dawkins
The Fourier cosine series for this function is then,
f ( x) =
L3 6 L3 n np x + 4 4 2 + ( n 2p 2 - 2 ) ( -1) cos 4 n =1 n p L
Finally, lets take a quick look at a piecewise function.
L Example 5 Find the Fourier cosine series for f ( x ) = 2 L x - 2
if 0 x L 2 on 0 x L . x L if L 2
Solution Well need to split up the integrals for each of the coefficients here. Here are the coefficients.
A0 =
L 1 L 1 L 2 f x dx = f x dx + f ( x ) dx ( ) ( ) L 0 0 2 L L L L 1 2L L 1 L2 L2 1 3L2 3L = dx + x - dx = + = = L 2 L 4 8 L 8 8 L 0 2 2
For the rest of the coefficients here is the integral well need to do.
L L 2 2 2 np x np x np x An = f ( x ) cos dx = f x cos dx + ( ) f ( x ) cos dx L 0 L L L L L 0 2 L L L 2 2 L np x np x = cos dx + x cos L dx L L 2 2 L 0 2
L
To make life a little easier lets do each of these separately.
L L np x 2 L L 2L np x cos dx = L sin = 2 0 2 np L 0 2 np L 2
L
np sin 2
2 L np = 2np sin 2 L
L L L L np x np x np x cos x - cos dx = + x - sin 2 2 L np np L L L 2 =
L L L L np cos ( n p ) + sin ( n p ) cos 2 np np np 2 L2 n np = 2 2 ( -1) - cos np 2

Putting these together gives,
2007 Paul Dawkins
424
An = =
2 L2 np sin L 2np 2
L2 n np + -1) - cos 2 2 ( np 2 np np sin + 2 2
2L np ( -1)n - cos np 2
So, after all that work the Fourier cosine series is then,
f ( x) =
3L 2 L np n + ( -1) - cos 8 n =1 np 2
np np sin + 2 2
np x cos L
Note that much as we saw with the Fourier sine series many of the coefficients will quite messy to deal with.
2007 Paul Dawkins
425
Fourier Series
Okay, in the previous two sections weve looked at Fourier sine and Fourier cosine series. It is now time to look at a Fourier series. With a Fourier series we are going to try to write a series representation for f ( x ) on - L x L in the form,
np x np x f ( x ) = An cos + Bn sin L n=1 L n =0
So, a Fourier series is, in some way a combination of the Fourier sine and Fourier cosine series. Also, like the Fourier sine/cosine series well not worry about whether or not the series will actually converge to f ( x ) or not at this point. For now well just assume that it will converge and well discuss the convergence of the Fourier series in a later section. Determining formulas for the coefficients, An and Bn , will be done in exactly the same manner as we did in the previous two sections. We will take advantage of the fact that and sin (
{cos ( npL x )}n
=0
np x
L
)}n = 1 are mutually orthogonal on - L x L as we proved earlier.
Well also
need the following formulas that we derived when we proved the two sets were mutually orthogonal.
2 L if n = m = 0 L np x mp x cos cos dx = L if n = m 0 - L L L 0 if n m
L L if n = m np x mp x sin sin dx = 0 if n m - L L L
np x mp x sin cos dx = 0 L - L L
x So, lets start off by multiplying both sides of the series above by cos ( mp and integrating from L )
L to L. Doing this gives,
L -L
f ( x ) cos
( )
mp x L
dx = An cos - L n = 0
( ) cos ( )
np x L mp x L
dx + Bn sin - L n=1
x mp x ( np L ) cos ( L ) dx
Now, just as weve been able to do in the last two sections we can interchange the integral and the summation. Doing this gives,
L -L
f ( x ) cos
( p ) dx = A
m x L
n =0 n
-L
cos
( p ) cos ( p ) dx + B
n x L m x L
n =1 n
-L
sin
( p ) cos ( p ) dx
n x L m x L
2007 Paul Dawkins
426
We can now take advantage of the fact that the sines and cosines are mutually orthogonal. The integral in the second series will always be zero and in the first series the integral will be zero if n m and so this reduces to,
Solving for Am gives,
L -L
f ( x ) cos
( p ) dx = A
m x L
Am ( 2 L ) if n = m = 0 m ( L ) if n = m 0
1 L f ( x ) dx 2L -L 1 L Am = f ( x ) cos L -L A0 =
( mLp x ) dx
m = 1, 2, 3,K
x , integrate both sides Now, do it all over again only this time multiply both sides by sin ( mp L )
from L to L and interchange the integral and summation to get,
L -L
f ( x ) sin
( )
mp x L
dx = An cos
L n =0 -L
( ) ( )
np x L
sin
mp x L
dx + Bn sin
L n =1 -L
( p ) sin ( p ) dx
n x L m x L
In this case the integral in the first series will always be zero and the second will be zero if n m and so we get,
Finally, solving for Bm gives,
L -L
f ( x ) sin
( p ) dx = B
m x L
( L)
Bm =
1 L f ( x ) sin L -L
( mLp x ) dx
m = 1, 2, 3,K
In the previous two sections we also took advantage of the fact that the integrand was even to give a second form of the coefficients in terms of an integral from 0 to L. However, in this case we dont know anything about whether f ( x ) will be even, odd, or more likely neither even nor odd. Therefore, this is the only form of the coefficients for the Fourier series. Before we start examples lets remind ourselves of a couple of formulas that well make heavy use of here in this section, as weve done in the previous two sections as well. Provided n in an integer then,
cos ( np ) = ( -1)
sin ( np ) = 0
In all of the work that well be doing here n will be an integer and so well use these without comment in the problems so be prepared for them. Also dont forget that sine is an odd function, i.e. sin ( - x ) = - sin ( x ) and that cosine is an even function, i.e. cos ( - x ) = cos ( x ) . Well also be making heave use of these ideas without comment in many of the integral evaluations so be ready for these as well.
Now lets take a look at an example.
Example 1 Find the Fourier series for f ( x ) = L - x on - L x L .

Solution So, lets go ahead and just run through formulas for the coefficients.
1 An = L -L
1 L 1 L f x dx = L - x dx = L ( ) 2L -L 2 L -L L 1 np x np x f ( x ) cos dx = ( L - x ) cos dx L -L L L A0 =
L
1 L np x np x = 2 2 np ( L - x ) sin - L cos L n p L L -L = 1 L ( -2np L sin ( -n p ) ) = 0 L n 2p 2 1 1 np x np x f ( x ) sin dx = ( L - x ) sin dx L -L L -L L L

L L L
n = 1, 2,3,K
Bn =
L 1 np x n p x = - 2 2 L sin - np ( x - L ) cos L n p L L -L 2 L ( -1) 1 L2 = 2 2 ( 2np cos ( np ) - 2sin ( np ) ) = L n p np

n
n = 1, 2,3,K
Note that in this case we had A0 0 and An = 0, n = 1, 2,3, K This will happen on occasion so dont get excited about this kind of thing when it happens. The Fourier series is then,
np x np x f ( x ) = An cos + Bn sin L n =1 L n =0 2 L ( -1) np x np x np x = A0 + An cos + B sin = L + sin n np L n =1 L L n =1 n =1 n
As we saw in the previous example sometimes well get A0 0 and An = 0, n = 1, 2,3, K but when it does dont get excited about it. Lets take a look at another problem.
Whether or not this will happen will depend upon the function f ( x ) and often wont happen,
2007 Paul Dawkins
428
L Example 2 Find the Fourier series for f ( x ) = 2 x
if - L x 0 on - L x L . if 0 x L
Solution Because of the piece-wise nature of the function the work for the coefficients is going to be a little unpleasant but lets get on with it.
A0 =
L 1 L 1 0 f x dx = f x dx + ( ) ( ) 0 f ( x ) dx -L 2L -L 2L L 1 0 1 2 L dx + 2 x dx = L + L2 = =L L 0 2L 2L
L 0 L 1 1 np x np x np x An = f ( x ) cos dx f x cos dx f x cos = + ( ) ( ) dx L 0 L L L L 0 - L
0 L 1 np x np x L cos dx + 2 x cos dx L L L L 0
At this point it will probably be easier to do each of these individually.
L2 L2 np x np x L cos dx = sin = sin ( np ) = 0 L L - L np -L np np x 2L np x np x 2 x cos dx = 2 2 L cos + np x sin L n p L L 0 0

L L
2L = 2 2 ( L cos ( np ) + np L sin ( np ) - L cos ( 0 ) ) n p 2 L2 n = 2 2 ( -1) - 1 n p
So, if we put all of this together we have,

L 1 1 2 L2 n np x An = f ( x ) cos dx = 0 + 2 2 ( -1) - 1 L 0 L n p L 2L n = 2 2 ( -1) - 1 , n = 1, 2,3,K np
So, weve gotten the coefficients for the cosines taken care of and now we need to take care of the coefficients for the sines.
2007 Paul Dawkins
429
Bn =
0 L L 1 1 np x np x np x f x dx = f x dx + sin sin ( ) ( ) f ( x ) sin dx L 0 L L L L 0 L 0 L 1 np x np x = L sin dx + x 2 sin dx L L 0 L - L
As with the coefficients for the cosines will probably be easier to do each of these individually.
L2 L2 L2 np x np x L sin dx = - np cos L = np ( -1 + cos ( np ) ) = np -L L -L

L
(( -1) - 1)
n L
np x 2L np x np x 2 x sin dx = 2 2 L sin - np x cos 0 L n p L L 0 2L = 2 2 ( L sin ( np ) - np L cos ( np ) ) n p 2 L2 2 L2 n n = 2 2 - np ( -1) = ( -1) np n p
So, if we put all of this together we have,
1 1 L2 2 L2 n n np x B n = f ( x ) sin ( -1) - 1 - ( -1) dx = L 0 L np np L L L n n = -1 - ( -1) = 1 + ( -1) np np
n = 1, 2,3,K
So, after all that work the Fourier series is,

np x np x f ( x ) = An cos + Bn sin L n=1 L n =0 np x np x = A0 + An cos + Bn sin L n=1 L n =1 2L n n np x L np x 1 + ( -1) sin = L + 2 2 ( -1) - 1 cos - L n =1 np L n =1 n p
As we saw in the previous example there is often quite a bit of work involved in computing the integrals involved here. The next couple of examples are here so we can make a nice observation about some Fourier series and their relation to Fourier sine/cosine series
2007 Paul Dawkins
430
Example 3 Find the Fourier series for f ( x ) = x on - L x L .

Solution Lets start with the integrals for An .
A0 =
L
1 L 1 L f x dx = x dx = 0 ( ) 2L -L 2L -L
L
An =
1 1 np x np x f ( x ) cos dx = x cos dx = 0 L -L L -L L L
In both cases note that we are integrating an odd function (x is odd and cosine is even so the product is odd) over the interval [ - L, L ] and so we know that both of these integrals will be zero. Next here is the integral for Bn
1 1 2 np x np x np x B n = f ( x ) sin dx = x sin dx = x sin dx L -L L -L L 0 L L L

In this case were integrating an even function (x and sine are both odd so the product is even) on the interval [ - L, L ] and so we can simplify the integral as shown above. The reason for doing this here is not actually to simplify the integral however. It is instead done so that we can note that we did this integral back in the Fourier sine series section and so dont need to redo it in this section. Using the previous result we get,
Bn
( -1) =
n +1
2L
np
n = 1, 2,3,K
In this case the Fourier series is,
np x n p x ( -1) 2 L np x sin sin f ( x ) = An cos + B n = np L n =1 L n=1 L n =0

n +1
If you go back and take a look at Example 1 in the Fourier sine series section, the same example we used to get the integral out of, you will see that in that example we were finding the Fourier sine series for f ( x ) = x on - L x L . The important thing to note here is that the answer that we got in that example is identical to the answer we got here. If you think about it however, this should not be too surprising. In both cases we were using an odd function on - L x L and because we know that we had an odd function the coefficients of the cosines in the Fourier series, An , will involve integrating and odd function over a symmetric interval, - L x L , and so will be zero. So, in these cases the Fourier sine series of an odd function on - L x L is really just a special case of a Fourier series.
Note however that when we moved over to doing the Fourier sine series of any function on 0 x L we should no longer expect to get the same results. You can see this by comparing Example 1 above with Example 3 in the Fourier sine series section. In both examples we are finding the series for f ( x ) = x - L and yet got very different answers. So, why did we get different answers in this case? Recall that when we find the Fourier sine series of a function on 0 x L we are really finding the Fourier sine series of the odd extension of the function on - L x L and then just restricting the result down to 0 x L . For a Fourier series we are actually using the whole function on - L x L instead of its odd extension. We should therefore not expect to get the same results since we are really using different functions (at least on part of the interval) in each case. So, if the Fourier sine series of an odd function is just a special case of a Fourier series it makes some sense that the Fourier cosine series of an even function should also be a special case of a Fourier series. Lets do a quick example to verify this.
Example 4 Find the Fourier series for f ( x ) = x 2 on - L x L .

Solution Here are the integrals for the An and in this case because both the function and cosine are even well be integrating an even function and so can simplify the integral.
A0 =
L
1 L 1 L 2 1 L 2 f x dx = x dx = x dx ( ) 2L -L 2L -L L 0
L L
An =
1 1 2 np x np x np x dx = x 2 cos dx = x 2 cos f ( x ) cos dx L -L L -L L 0 L L L
As with the previous example both of these integrals were done in Example 1 in the Fourier cosine series section and so well not bother redoing them here. The coefficients are,
L2 A0 = 3
Next here is the integral for the Bn
An =
4 L2 ( -1) n 2p 2
, n = 1, 2,3, K
1 1 np x np x B n = f ( x ) sin dx = x 2 sin dx = 0 L -L L -L L L
In this case the function is even and sine is odd so the product is odd and were integrating over - L x L and so the integral is zero. The Fourier series is then,
2 4 L ( -1) np x np x L np x sin cos f ( x ) = An cos + B = + n 2 2 L n=1 L 3 n =1 n p L n =0 2 n
2007 Paul Dawkins
432
As suggested before we started this example the result here is identical to the result from Example 1 in the Fourier cosine series section and so we can see that the Fourier cosine series of an even function is just a special case a Fourier series.
2007 Paul Dawkins
433
Convergence of Fourier Series

Over the last few sections weve spent a fair amount of time to computing Fourier series, but weve avoided discussing the topic of convergence of the series. In other words, will the Fourier series converge to the function on the given interval? In this section were going to address this issue as well as a couple of other issues about Fourier series. Well be giving a fair number of theorems in this section but are not going to be proving any of them. Well also not be doing a whole lot of in the way of examples in this section. Before we get into the topic of convergence we need to first define a couple of terms that well run into in the rest of the section. First, we say that f ( x ) has a jump discontinuity at x = a if the limit of the function from the left, denoted f a - , and the limit of the function from the
+ right, denoted f a , both exist and
( )
( ) f ( a - ) f ( a+ ) .
Next, we say that f ( x ) is piecewise smooth if the function can be broken into distinct pieces and on each piece both the function and its derivative, f ( x ) , are continuous. A piecewise smooth function may not be continuous everywhere however the only discontinuities that are allowed are a finite number of jump discontinuities. Lets consider the function,
L f ( x) = 2 x
if - L x 0 if 0 x L
We found the Fourier series for this function in Example 2 of the previous section. Here is a sketch of this function on the interval on which it is defined, i.e. - L x L .
This function has a jump discontinuity at x = 0 because f 0- = L 0 = f 0+ and note that on the intervals - L x 0 and 0 x L both the function and its derivative are continuous. This is therefore an example of a piecewise smooth function. Note that the function itself is not
( )
( )
continuous at x = 0 but because this point of discontinuity is a jump discontinuity the function is still piecewise smooth. The last term we need to define is that of periodic extension. Given a function, f ( x ) , defined on some interval, well be using - L x L exclusively here, the periodic extension of this function is the new function we get by taking the graph of the function on the given interval and then repeating that graph to the right and left of the graph of the original function on the given interval. It is probably best to see an example of a periodic extension at this point to help make the words above a little clearer. Here is a sketch of the period extension of the function we looked at above,
The original function is the solid line in the range - L x L . We then got the periodic extension of this by picking this piece up and copying it every interval of length 2L to the right and left of the original graph. This is shown with the two sets of dashed lines to either side of the original graph. Note that the resulting function that we get from defining the periodic extension is in fact a new periodic function that is equal to the original function on - L x L . With these definitions out of the way we can now proceed to talk a little bit about the convergence of Fourier series. We will start off with the convergence of a Fourier series and once we have that taken care of the convergence of Fourier Sine/Cosine series will follow as a direct consequence. Here then is the theorem giving the convergence of a Fourier series. Convergence of Fourier series Suppose f ( x ) is a piecewise smooth on the interval - L x L . The Fourier series of f ( x ) will then converge to, 1. the periodic extension of f ( x ) if the periodic extension is continuous. 2. the average of the two one-sided limits, has a jump discontinuity at x = a .
1 2
f ( a - ) + f ( a + ) , if the periodic extension

The first thing to note about this is that on the interval - L x L both the function and the periodic extension are equal and so where the function is continuous on - L x L the periodic extension will also be continuous and hence at these points the Fourier series will in fact converge to the function. The only points in the interval - L x L where the Fourier series will not converge to the function is where the function has a jump discontinuity. Lets again consider Example 2 of the previous section. In that section we found that the Fourier series of,
on - L x L to be,
L f ( x) = 2 x 2L 2 2 n =1 n p
if - L x 0 if 0 x L np x L n np x 1 + ( -1) sin - L n =1 np L
f ( x) = L +
(( -1)n - 1) cos
We now know that in the intervals - L < x < 0 and 0 < x < L the function and hence the periodic extension are both continuous and so on these two intervals the Fourier series will converge to the periodic extension and hence will converge to the function itself. At the point x = 0 the function has a jump discontinuity and so the periodic extension will also have a jump discontinuity at this point. That means that at x = 0 the Fourier series will converge to,
1 1 L f ( 0- ) + f ( 0+ ) = [ L + 0] = 2 2 2
At the two endpoints of the interval, x = - L and x = L , we can see from the sketch of the periodic extension above that the periodic extension has a jump discontinuity here and so the Fourier series will not converge to the function there but instead the averages of the limits. So, at x = - L the Fourier series will converge to,
and at x = L the Fourier series will converge to,
1 1 3L + f L + f L = 2 L + L = [ ] ( ) ( ) 2 2 2 1 = 1 [ 2 L + L ] = 3L f ( L- ) + f ( L+ ) 2 2 2
Now that we have addressed the convergence of a Fourier series we can briefly turn our attention to the convergence of Fourier sine/cosine series. First, as noted in the previous section the Fourier sine series of an odd function on - L x L and the Fourier cosine series of an even function on - L x L are both just special cases of a Fourier series we now know that both of these will have the same convergence as a Fourier series. Next, if we look at the Fourier sine series of any function, g ( x ) , on 0 x L then we know that this is just the Fourier series of the odd extension of g ( x ) restricted down to the interval
0 x L . Therefore we know that the Fourier series will converge to the odd extension on - L x L where it is continuous and the average of the limits where the odd extension has a
jump discontinuity. However, on 0 x L we know that g ( x ) and the odd extension are equal and so we can again see that the Fourier sine series will have the same convergence as the Fourier series. Likewise, we can go through a similar argument for the Fourier cosine series using even extensions to see that Fourier cosine series for a function on 0 x L will also have the same convergence as a Fourier series. The next topic that we want to briefly discuss here is when will a Fourier series be continuous. From the theorem on the convergence of Fourier series we know that where the function is continuous the Fourier series will converge to the function and hence be continuous at these points. The only places where the Fourier series may not be continuous is if there is a jump discontinuity on the interval - L x L and potentially at the endpoints as we saw that the periodic extension may introduce a jump discontinuity there. So, if were going to want the Fourier series to be continuous everywhere well need to make sure that the function does not have any discontinuities in - L x L . Also, in order to avoid having the periodic extension introduce a jump discontinuity well need to require that f ( - L ) = f ( L ) . By doing this the two ends of the graph will match up when we form the periodic extension and hence we will avoid a jump discontinuity at the end points. Here is a summary of these ideas for a Fourier series. Suppose f ( x ) is a piecewise smooth on the interval - L x L . The Fourier series of f ( x ) will be continuous and will converge to f ( x ) on - L x L provided f ( x ) is continuous on
- L x L and f ( - L ) = f ( L ) .
Now, how can we use this to get similar statements about Fourier sine/cosine series on 0 x L ? Lets start with a Fourier cosine series. The first thing that we do is form the even extension of f ( x ) on - L x L . For the purposes of this discussion lets call the even extension g ( x ) As we saw when we sketched several even extensions in the Fourier cosine series section that in order for the sketch to be the even extension of the function we must have both,
g ( 0 - ) = g ( 0+ )
g (-L) = g ( L)
If one or both of these arent true then g ( x ) will not be an even extension of f ( x ) . So, in forming the even extension we do not introduce any jump discontinuities at x = 0 and we get for free that g ( - L ) = g ( L ) . If we now apply the above theorem to the even extension we see that the Fourier series of the even extension is continuous on - L x L . However, because the even extension and the function itself are the same on 0 x L then the Fourier cosine series of f ( x ) must also be continuous on 0 x L .
Here is a summary of this discussion for the Fourier cosine series. Suppose f ( x ) is a piecewise smooth on the interval 0 x L . The Fourier cosine series of
f ( x ) will be continuous and will converge to f ( x ) on 0 x L provided f ( x ) is continuous on 0 x L .

Note that we dont need any requirements on the end points here because they are trivially satisfied when we convert over to the even extension. For a Fourier sine series we need to be a little more careful. Again, the first thing that we need to do is form the odd extension on - L x L and lets call it g ( x ) . We know that in order for it to be the odd extension then we know that at all points in - L x L it must satisfy g ( - x ) = - g ( x ) and that is what can lead to problems. As we saw in the Fourier sine series section it is very easy to introduce a jump discontinuity at x = 0 when we form the odd extension. In fact, the only way to avoid forming a jump discontinuity at this point is to require that f ( 0 ) = 0 . Next, the requirement that at the endpoints we must have g ( - L ) = - g ( L ) will practically guarantee that well introduce a jump discontinuity here as well when we form the odd extension. Again, the only way to avoid doing this is to require f ( L ) = 0 . So, with these two requirements we will get an odd extension that is continuous and so we know that the Fourier series of the odd extension on - L x L will be continuous and hence the Fourier sine series will be continuous on 0 x L . Here is a summary of all this for the Fourier sine series. Suppose f ( x ) is a piecewise smooth on the interval 0 x L . The Fourier sine series of
f ( x ) will be continuous and will converge to f ( x ) on 0 x L provided f ( x ) is

continuous on 0 x L , f ( 0 ) = 0 and f ( L ) = 0 . The next topic of discussion here is differentiation and integration of Fourier series. In particular we want to know if we can differentiate a Fourier series term by term and have the result be the Fourier series of the derivative of the function. Likewise we want to know if we can integrate a Fourier series term by term and arrive at the Fourier series of the integral of the function. Note that well not be doing much discussion of the details here. All were really going to be doing is giving the theorems that govern the ideas here so that we can say weve given them. Lets start off with the theorem for term by term differentiation of a Fourier series.
2007 Paul Dawkins
438
Given a function f ( x ) if the derivative, f ( x ) , is piecewise smooth and the Fourier series of
f ( x ) is continuous then the Fourier series can be differentiated term by term. The result of the
differentiation is the Fourier series of the derivative, f ( x ) . One of the main condition of this theorem is that the Fourier series be continuous and from above we also know the conditions on the function that will give this. So, if we add this into the theorem to get this form of the theorem, Suppose f ( x ) is a continuous function, its derivative f ( x ) is piecewise smooth and
f ( - L ) = f ( L ) then the Fourier series of the function can be differentiated term by term and the
result is the Fourier series of the derivative. For Fourier cosine/sine series the basic theorem is the same as for Fourier series. All thats required is that the Fourier cosine/sine series be continuous and then you can differentiate term by term. The theorems that well give here will merge the conditions for the Fourier cosine/sine series to be continuous into the theorem. Lets start with the Fourier cosine series. Suppose f ( x ) is a continuous function and its derivative f ( x ) is piecewise smooth then the Fourier cosine series of the function can be differentiated term by term and the result is the Fourier sine series of the derivative. Next the theorem for Fourier sine series. Suppose f ( x ) is a continuous function, its derivative f ( x ) is piecewise smooth, f ( 0 ) = 0 and
f ( L ) = 0 then the Fourier sine series of the function can be differentiated term by term and the
result is the Fourier cosine series of the derivative. The theorem for integration of Fourier series term by term is simple so there it is. Suppose f ( x ) is piecewise smooth then the Fourier sine series of the function can be integrated term by term and the result is a convergent infinite series that will converge to the integral of f ( x) . Note however that the new series that results from term by term integration may not be the Fourier series for the integral of the function.
2007 Paul Dawkins
439
Partial Differential Equations

Introduction
In this chapter we are going to take a very brief look at one of the more common methods for solving simple partial differential equations. The method well be taking a look at is that of Separation of Variables. We need to make it very clear before we even start this chapter that we are going to be doing nothing more than barely scratching the surface of not only partial differential equations but also of the method of separation of variables. It would take several classes to cover most of the basic techniques for solving partial differential equations. The intent of this chapter is to do nothing more than to give you a feel for the subject and if youd like to know more taking a class on partial differential equations should probably be your next step. Also note that in several sections we are going to be making heavy use of some of the results from the previous chapter. That in fact was the point of doing some of the examples that we did there. Having done them will, in some cases, significantly reduce the amount of work required in some of the examples well be working in this chapter. When we do make use of a previous result we will make it very clear where the result is coming from. Here is a brief listing of the topics covered in this chapter. The Heat Equation We do a partial derivation of the heat equation in this section as well as a discussion of possible boundary values. The Wave Equation Here we do a partial derivation of the wave equation. Terminology In this section we take a quick look at some of the terminology used in the method of separation of variables. Separation of Variables We take a look at the first step in the method of separation of variables in this section. This first step is really the step motivates the whole process. Solving the Heat Equation In this section we go through the complete separation of variables process and along the way solve the heat equation with three different sets of boundary conditions. Heat Equation with Non-Zero Temperature Boundaries Here we take a quick look at solving the heat equation in which the boundary conditions are fixed, non-zero temperature conditions. Laplaces Equation We discuss solving Laplaces equation on both a rectangle and a disk in this section.
2007 Paul Dawkins
440
Vibrating String Here we solve the wave equation for a vibrating string. Summary of Separation of Variables In this final section we give a quick summary of the method of separation of variables.
2007 Paul Dawkins
441
The Heat Equation

Before we get into actually solving partial differential equations and before we even start discussing the method of separation of variables we want to spend a little bit of time talking about the two main partial differential equations that well be solving later on in the chapter. Well look at the first one in this section and the second one in the next section. The first partial differential equation that well be looking at once we get started with solving will be the heat equation, which governs the temperature distribution in an object. We are going to give several forms of the heat equation for reference purposes, but we will only be really solving one of them. We will start out by considering the temperature in a 1-D bar of length L. What this means is that we are going to assume that the bar starts off at x = 0 and ends when we reach x = L . We are also going to so assume that at any location, x the temperature will be constant an every point in the cross section at that x. In other words, temperature will only vary in x and we can hence consider the bar to be a 1-D bar. Note that with this assumption the actual shape of the cross section (i.e. circular, rectangular, etc.) doesnt matter. Note that the 1-D assumption is actually not all that bad of an assumption as it might seem at first glance. If we assume that the lateral surface of the bar is perfectly insulated (i.e. no heat can flow through the lateral surface) then the only way heat can enter or leave the bar as at either end. This means that heat can only flow from left to right or right to left and thus creating a 1-D temperature distribution. The assumption of the lateral surfaces being perfectly insulated is of course impossible, but it is possible to put enough insulation on the lateral surfaces that there will be very little heat flow through them and so, at least for a time, we can consider the lateral surfaces to be perfectly insulated. Okay, lets now get some definitions out of the way before we write down the first form of the heat equation.
u ( x, t ) = Temperature at any point x and any time t c ( x ) = Specific Heat
r ( x ) = Mass Density j ( x, t ) = Heat Flux

Q ( x, t ) = Heat energy generated per unit volume per unit time
We should probably make a couple of comments about some of these quantities before proceeding. The specific heat, c ( x ) > 0 , of a material is the amount of heat energy that it takes to raise one unit of mass of the material by one unit of temperature. As indicated we are going to assume, at least initially, that the specific heat may not be uniform throughout the bar. Note as well that in practice the specific heat depends upon the temperature. However, this will generally only be an
issue for large temperature differences (which in turn depends on the material the bar is made out of) and so were going to assume for the purposes of this discussion that the temperature differences are not large enough to affect our solution. The mass density, r ( x ) , is the mass per unit volume of the material. As with the specific heat were going to initially assume that the mass density may not be uniform throughout the bar. The heat flux, j ( x, t ) , is the amount of thermal energy that flows to the right per unit surface area per unit time. The flows to the right bit simply tells us that if j ( x, t ) > 0 for some x and t then the heat is flowing to the right at that point and time. Likewise if j ( x, t ) < 0 then the heat will be flowing to the left at that point and time. The final quantity we defined above is Q ( x, t ) and this is used to represent any external sources or sinks (i.e. heat energy taken out of the system) of heat energy. If Q ( x, t ) > 0 then heat energy is being added to the system at that location and time and if Q ( x, t ) < 0 then heat energy is being removed from the system at that location and time. With these quantities the heat equation is,
c ( x) r ( x)
u j =+ Q ( x, t ) t x
(2)
While this is a nice form of the heat equation it is not actually something we can solve. In this form there are two unknown functions, u and j , and so we need to get rid of one of them. With Fouriers law we can easily remove the heat flux from this equation. Fouriers law states that,
j ( x, t ) = - K 0 ( x )
where K 0 ( x ) > 0 is the thermal conductivity of the material and measures the ability of a given material to conduct heat. The better a material can conduct heat the larger K 0 ( x ) will be. As noted the thermal conductivity can vary with the location in the bar. Also, much like the specific heat the thermal conductivity can vary with temperature, but we will assume that the total temperature change is not so great that this will be an issue and so we will assume for the purposes here that the thermal conductivity will not vary with temperature. Fouriers law does a very good job of modeling what we know to be true about heat flow. First, we know that if the temperature in a region is constant, i.e.
u x
u = 0 , then there is no heat flow. x
Next, we know that if there is a temperature difference in a region we know the heat will flow from the hot portion to the cold portion of the region. For example, if it is hotter to the right then we know that the heat should flow to the left. When it is hotter to the right then we also know
u > 0 (i.e. the temperature increases as we move to the right) and so well have j < 0 and x u < 0 (i.e. it is hotter to the left) then so the heat will flow to the left as it should. Likewise, if x well have j > 0 and heat will flow to the right as it should.
that Finally, the greater the temperature difference in a region (i.e. the larger the heat flow. So, if we plug Fouriers law into (2), we get the following form of the heat equation,
u is) then the greater x
c ( x) r ( x)
u u = K 0 ( x ) + Q ( x, t ) t x x
(3)
Note that we factored the minus sign out of the derivative to cancel against the minus sign that was already there. We cannot however, factor the thermal conductivity out of the derivative since it is a function of x and the derivative is with respect to x. Solving (3) is quite difficult due to the non uniform nature of the thermal properties and the mass density. So, lets now assume that these properties are all constant, i.e.,
c ( x) = c
r ( x) = r
K 0 ( x ) = K0
where c, r and K 0 are now all fixed quantities. In this case we generally say that the martial in the bar is uniform. Under these assumptions the heat equation becomes,
cr
u 2u = K 0 2 + Q ( x, t ) t x
(4)
For a final simplification to the heat equation lets divide both sides by c r and define the thermal diffusivity to be,
k=
The heat equation is then,
K0 cr
u 2u Q ( x, t ) =k 2 + t x cr
(5)
To most people this is what they mean when they talk about the heat equation and in fact it will be the equation that well be solving. Well, actually well be solving (5) with no external sources, i.e. Q ( x, t ) = 0 , but well be considering this form when we start discussing separation of variables in a couple of sections. Well only drop the sources term when we actually start solving the heat equation. Now that weve got the 1-D heat equation taken care of we need to move into the initial and boundary conditions well also need in order to solve the problem. If you go back to any of our solutions of ordinary differential equations that weve done in previous sections you can see that
the number of conditions required always matched the highest order of the derivative in the equation. In partial differential equations the same idea holds except now we have to pay attention to the variable were differentiating with respect to as well. So, for the heat equation weve got a first order time derivative and so well need one initial condition and a second order spatial derivative and so well need two boundary conditions. The initial condition that well use here is,
u ( x, 0 ) = f ( x )
and we dont really need to say much about it here other than to note that this just tells us what the initial temperature distribution in the bar is. The boundary conditions will tell us something about what the temperature and/or heat flow is doing at the boundaries of the bar. There are four of them that are fairly common boundary conditions. The first type of boundary conditions that we can have would be the prescribed temperature boundary conditions, also called Dirichlet conditions. The prescribed temperature boundary conditions are,
u ( 0, t ) = g1 ( t )
u ( L, t ) = g 2 ( t )
The next type of boundary conditions are prescribed heat flux, also called Neumann conditions. Using Fouriers law these can be written as,
-K0 ( 0)
u ( 0, t ) = j1 ( t ) x
-K0 ( L)
u ( L, t ) = j 2 ( t ) x
If either of the boundaries are perfectly insulated, i.e. there is no heat flow out of them then these boundary conditions reduce to,
u ( 0, t ) = 0 x
u ( L, t ) = 0 x
and note that we will often just call these particular boundary conditions insulated boundaries and drop the perfectly part. The third type of boundary conditions use Newtons law of cooling and are sometimes called Robins conditions. These are usually used when the bar is in a moving fluid and note we can consider air to be a fluid for this purpose.
u ( L, t ) = H u ( L, t ) - g 2 ( t ) x where H is a positive quantity that is experimentally determined and g1 ( t ) and g 2 ( t ) give the -K0 ( 0) -K0 ( L)
temperature of the surrounding fluid at the respective boundaries. Note that the two conditions do vary slightly depending on which boundary we are at. At x = 0 we have a minus sign on the right side while we dont at x = L . To see why this is lets first
Here are the equations for this kind of boundary condition.
u ( 0, t ) = - H u ( 0, t ) - g1 ( t ) x
assume that at x = 0 we have u ( 0, t ) > g1 ( t ) . In other words the bar is hotter than the surrounding fluid and so at x = 0 the heat flow (as given by the left side of the equation) must be to the left, or negative since the heat will flow from the hotter bar into the cooler surrounding liquid. If the heat flow is negative then we need to have a minus sign on the right side of the equation to make sure that it has the proper sign. If the bar is cooler than the surrounding fluid at x = 0 , i.e. u ( 0, t ) < g1 ( t ) we can make a similar argument to justify the minus sign. Well leave it to you to verify this. If we now look at the other end, x = L , and again assume that the bar is hotter than the surrounding fluid or, u ( L, t ) > g 2 ( t ) . In this case the heat flow must be to the right, or be positive, and so in this case we cant have a minus sign. Finally, well again leave it to you to verify that we cant have the minus sign at x = L is the bar is cooler than the surrounding fluid as well. Note that we are not actually going to be looking at any of these kinds of boundary conditions here. These types of boundary conditions tend to lead to boundary value problems such as Example 5 in the Eigenvalues and Eigenfunctions section of the previous chapter. As we saw in that example it is often very difficult to get our hands on the eigenvalues and as well eventually see we will need them. It is important to note at this point that we can also mix and match these boundary conditions so to speak. There is nothing wrong with having a prescribed temperature at one boundary a prescribed flux at the other boundary for example so dont always expect the same boundary condition to show up at both ends. This warning is more important that it might seem at this point because once we get into solving the heat equation we are going to have the same kind of condition on each end to simplify the problem somewhat. The final type of boundary conditions that well need here are periodic boundary conditions. Periodic boundary conditions are,
u ( - L, t ) = u ( L, t )
u u ( - L, t ) = ( L, t ) x x
Note that for these kinds of boundary conditions the left boundary tends to be x = - L instead of x = 0 as we were using in the previous types of boundary conditions. The periodic boundary conditions will arise very naturally from a couple of particular geometries that well be looking at down the road. We will now close out this section with a quick look at the 2-D and 3-D version of the heat equation. However, before we jump into that we need to introduce a little bit of notation first. The del operator is defined to be,
r r i+ j x y
r r r i+ j+ k x y k
depending on whether we are in 2 or 3 dimensions. Think of the del operator as a function that takes functions as arguments (instead of numbers as were used to). Whatever function we plug into the operator gets put into the partial derivatives.
So, for example in 3-D we would have,
f r f r f r i+ j+ k x y k This of course is also the gradient of the function f ( x, y , z ) . f =

The del operator also allows us to quickly write down the divergence of a function. So, again using 3-D as an example the divergence of f ( x, y , z ) can be written as the dot product of the del operator and the function. Or,
g f =
f f f + + x y k
Finally, we will also see the following show up in the our work,
g( f ) =
This is usually denoted as,
f f f 2 f 2 f 2 f + + + + = x x y y z k x 2 y 2 z 2 2 f 2 f 2 f + + x 2 y 2 z 2
2 f =
and is called the Laplacian. The 2-D version of course simply doesnt have the third term. Okay, we can now into the 2-D and 3-D version of the heat equation and where ever the del operator and or Laplacian appears assume that it is the appropriate dimensional version. The higher dimensional version of (2) is,
u = -g j + Q (6) t and note that the specific heat, c, and mass density, r , are may not be uniform and so may be cr
functions of the spatial variables. Likewise, the external sources term, Q, may also be a function of both the spatial variables and time. Next, the higher dimensional version of Fouriers law is,
j = - K 0 u
where the thermal conductivity, K 0 , is again assumed to be a function of the spatial variables. If we plug this into (6) we get the heat equation for a non uniform bar (i.e. the thermal properties may be functions of the spatial variables) with external sources/sinks,
cr
u = g ( K 0 u ) + Q t
(7)
If we now assume that the specific heat, mass density and thermal conductivity are constant (i.e. the bar is uniform) the heat equation becomes,
u Q = k 2u + t cp
447
(8)
2007 Paul Dawkins
where we divided both sides by c r to get the thermal diffusivity, k in front of the Laplacian. The initial condition for the 2-D or 3-D heat equation is,
u ( x, y , t ) = f ( x, y )
or
u ( x, y, z , t ) = f ( x, y, z )
depending upon the dimension were in. The prescribed temperature boundary condition becomes,
u ( x, y , t ) = T ( x, y , t )
or
u ( x, y , z , t ) = T ( x, y, z , t )
where ( x, y ) or ( x, y , z ) , depending upon the dimension were in, will range over the portion of the boundary in which we are prescribing the temperature. The prescribed heat flux condition becomes,
r - K 0 u g n = j ( t )
where the left side is only being evaluated at points along the boundary and n is the outward unit normal on the surface. Newtons law of cooling will become,
r - K 0 u g n = H ( u - u B )
where H is a positive quantity that is experimentally determine, uB is the temperature of the fluid at the boundary and again it is assumed that this is only being evaluated at points along the boundary. We dont have periodic boundary conditions here as they will only arise from specific 1-D geometries. We should probably also acknowledge at this point that well not actually be solving (8) at any point, but we will be solving a special case of it in the Laplaces Equation section.
2007 Paul Dawkins
448
The Wave Equation

In this section we want to consider a vertical string of length L that has been tightly stretched between two points at x = 0 and x = L . Because the string has been tightly stretched we can assume that the slope of the displaced string at any point is small. So just what does this do for us? Lets consider a point x on the string in its equilibrium position, i.e. the location of the point at t = 0 . As the string vibrates this point will be displaced both vertically and horizontally, however, if we assume that at any point the slope of the string is small then the horizontal displacement will be very small in relation to the vertical displacement. This means that we can now assume that at any point x on the string the displacement will be purely vertical. So, lets call this displacement u ( x, t ) . We are going to assume, at least initially, that the string is not uniform and so the mass density of the string, r ( x ) may be a function of x. Next we are going to assume that the string is perfectly flexible. This means that the string will have no resistance to bending. This in turn tells us that the force exerted by the string at any point x on the endpoints will be tangential to the string itself. This force is called the tension in the string and its magnitude will be given by T ( x, t ) . Finally, we will let Q ( x, t ) represent the vertical component per unit mass of any force acting on the string. Provided we again assume that the slope of the string is small the vertical displacement of the string at any point is then given by,
r ( x)
2u u = T ( x, t ) + r ( x ) Q ( x, t ) 2 t x x
(9)
This is a very difficult partial differential equation to solve so we need to make some further simplifications. First, were now going to assume that the string is perfectly elastic. This means that the magnitude of the tension, T ( x, t ) , will only depend upon how much the string stretches near x. Again, recalling that were assuming that the slope of the string at any point is small this means that the tension in the string will then very nearly be the same as the tension in the string in its equilibrium position. We can then assume that the tension is a constant value, T ( x, t ) = T0 . Further, in most cases the only external force that will act upon the string is gravity and if the string light enough the effects of gravity on the vertical displacement will be small and so will also assume that Q ( x, t ) = 0 . This leads to
2u 2u r 2 = T0 2 t x
If we know divide by the mass density and define,
c2 =
we arrive at the 1-D wave equation,
T0 r (10)
2 2u 2 u = c t 2 x 2
In the previous section when we looked at the heat equation he had a number of boundary conditions however in this case we are only going to consider one type of boundary conditions. For the wave equation we are only going to consider the boundary condition will that of prescribed location of the boundaries or,
u ( 0, t ) = h1 ( t )
u ( L, t ) = h 2 ( t )
The initial conditions (and yes we meant more than one) will also be a little different here from what we saw with the heat equation. Here we have a 2nd order time derivative and so well also need two initial conditions. At any point we will specify both the initial displacement of the string as well as the initial slope of the string. The initial conditions are then,
u ( x, 0 ) = f ( x )
u ( x, 0 ) = g ( x ) t
For the sake of completeness well close out this section with the 2-D and 3-D version of the wave equation. Well not actually be solving this at any point, but since we gave the higher dimensional version of the heat equation (in which we will solve a special case) well give this as well. The 2-D and 3-D version of the wave equation is,
2u = c 2 2u t 2
where 2 is the Laplacian.
2007 Paul Dawkins
450
Terminology
Weve got one more section that we need to take care of before we actually start solving partial differential equations. This will be a fairly short section that will cover some of the basic terminology that well need in the next section as we introduce the method of separation of variables. Lets start off with the idea of an operator. An operator is really just a function that takes a function as an argument instead of numbers as were used to dealing with in functions. You already know of a couple of operators even if you didnt know that they were operators. Here are some examples of operators.
L=
d dx
L = dx
L = dx
a
L=
Or, if we plug in a function, say u ( x ) , into each of these we get,
L (u ) =
du dx
L ( u ) = u ( x ) dx
L ( u ) = u ( x ) dx
a
L (u ) =
u t
These are all fairly simple examples of operators but the derivative and integral are operators. A more complicated operator would be the heat operator. We get the heat operator from a slight rewrite of the heat equation without sources. The heat operator is,
L=
2 -k 2 t x
Now, what we really want to define here is not an operator but instead a linear operator. A linear operator is any operator that satisfies,
L ( c1u1 + c2 u2 ) = c1 L ( u1 ) + c2 L ( u2 )
The heat operator is an example of a linear operator and this is easy enough to show using the basic properties of the partial derivative so lets do that.
L ( c1u1 + c2 u2 ) =
2 ( c1u1 + c2u2 ) - k 2 ( c1u1 + c2u2 ) t x 2 2 = ( c1u1 ) + ( c2u2 ) - k 2 ( c1u1 ) + 2 ( c2 u2 ) t t x x
2u u1 u 2u + c2 2 - k c1 21 + c2 22 t t x x u 2u u 2u = c1 1 - kc1 21 + c2 2 - kc2 22 t x t x 2 u1 u2 u1 2u 2 = c1 - k 2 + c2 -k 2 x x t t = c1 = c1 L ( u1 ) + c2 L ( u2 )
You might want to verify for yourself that the derivative and integral operators we gave above are also linear operators. In fact, in the process of showing that the heat operator is a linear operator we actually showed as well that the first order and second order partial derivative operators are also linear. The next term we need to define is a linear equation. A linear equation is an equation in the form,
L (u ) = f
(11)
where L is a linear operator and f is a known function. Here are some examples of linear partial differential equations.
u 2u Q ( x , t ) =k 2 + t x cr
2 2u 2 u = c t 2 x 2 2u 2u 2u + + = 2u = 0 x 2 y 2 z 2
u 2u 3u - 4 2 = 3 + 8u - g ( x, t ) t t x
The first two from this list are of course the heat equation and the wave equation. The third uses the Laplacian and is usually called Laplaces Equation. Well actually be solving the 2-D version of Laplaces Equation in a few sections. The fourth equation was just made up to give a more complicated example. Notice as well with the heat equation and the fourth example above that the presence of the Q ( x, t ) and g ( x, t ) do not prevent these from being linear equations. The main issue that allows these to be linear equations is the fact that the operator in each is linear. Now just to be complete here are a couple of examples of nonlinear partial differential equations.
u 2u = k 2 + u2 t x 2 u u u = u + f ( x, t ) t 2 x t
Well leave it to you to verify that the operators in each of these are not linear however the problem term in the first is the u 2 while in the second the product of the two derivatives is the problem term. Now, if we go back to (11) and suppose that f = 0 then we arrive at,
L (u ) = 0
We call this a linear homogeneous equation (recall that L is a linear operator).
(12)
Notice that u = 0 will always be a solution to a linear homogeneous equation (go back to what it means to be linear and use c1 = c2 = 0 with any two solutions and this is easy to verify). We call
u = 0 the trivial solution. In fact this is also a really nice way of determining if an equation is homogeneous. If L is a linear operator and we plug in u = 0 into the equation and we get L ( u ) = 0 then we will know that the operator is homogeneous.
We can also extend the ideas of linearity and homogeneous to boundary conditions. If we go back to the various boundary conditions we discussed for the heat equation for example we can also classify them as linear and/or homogeneous. The prescribed temperature boundary conditions,
u ( 0, t ) = g1 ( t )
u ( L, t ) = g 2 ( t )
are linear and will only be homogenous if g1 ( t ) = 0 and g 2 ( t ) = 0 . The prescribed heat flux boundary conditions,
u ( L, t ) = j 2 ( t ) x are linear and will again only be homogeneous if j1 ( t ) = 0 and j2 ( t ) = 0 . -K0 ( 0) -K0 ( L) u ( 0, t ) = - H u ( 0, t ) - g1 ( t ) x
u ( 0, t ) = j1 ( t ) x
Next, the boundary conditions from Newtons law of cooling,
u ( L, t ) = H u ( L, t ) - g 2 ( t ) x are again linear and will only be homogenous if g1 ( t ) = 0 and g 2 ( t ) = 0 . -K0 ( 0) -K0 ( L)
The final set of boundary conditions that we looked at were the periodic boundary conditions,
u ( - L, t ) = u ( L, t )
and these are both linear and homogeneous.
u u ( - L, t ) = ( L, t ) x x
The final topic in this section is not really terminology but is a restatement of a fact that weve seen several times in these notes already. Principle of Superposition If u1 and u2 are solutions to a linear homogeneous equation then so is c1u1 + c2 u2 for any values of c1 and c2 . Now, as stated earlier weve seen this several times this semester but we didnt really do much with it. However this is going to be a key idea when we actually get around to solving partial differential equations. Without this fact we would not be able to solve all but the most basic of partial differential equations.
2007 Paul Dawkins
453
Separation of Variables
Okay, it is finally time to at least start discussing one of the more common methods for solving basic partial differential equations. The method of Separation of Variables cannot always be used and even when it can be used it will not always be possible to get much past the first step in the method. However, it can be used to easily solve the 1-D heat equation with no sources, the 1D wave equation, and the 2-D version of Laplaces Equation, 2u = 0 . In order to use the method of separation of variables we must be working with a linear homogenous partial differential equations with linear homogeneous boundary conditions. At this point were not going to worry about the initial condition(s) because the solution that we initially get will rarely satisfy the initial condition(s). As well see however there are ways to generate a solution that will satisfy initial condition(s) provided they meets some fairly simple requirements. The method of separation of variables relies upon the assumption that a function of the form,
u ( x, t ) = j ( x ) G ( t )
(13)
will be a solution to a linear homogeneous partial differential equation in x and t. This is called a product solution and provided the boundary conditions are also linear and homogeneous this will also satisfy the boundary conditions. However, as noted above this will only rarely satisfy the initial condition, but that is something for us to worry about in the next section. Now, before we get started on some examples there is probably a question that we should ask at this point and that is : Why? Why did we choose this solution and how do we know that it will work? This seems like a very strange assumption to make. After all there really isnt any reason to believe that a solution to a partial differential equation will in fact be a product of a function of only xs and a function of only ts. This seems more like a hope than a good assumption/guess. Unfortunately the best answer is that we chose it because it will work. As well see it works because it will reduce our partial differential equation down to two ordinary differential equations and provided we can solve those then were in business and the method will allow us to get a solution to the partial differential equations. So, lets do a couple of examples to see how this method will reduce a partial differential equation down to two ordinary differential equations.
Example 1 Use Separation of Variables on the following partial differential equation. u 2u =k 2 t x u ( x, 0 ) = f ( x ) u ( 0, t ) = 0 u ( L, t ) = 0

Solution So, we have the heat equation with no sources, fixed temperature boundary conditions (that are also homogeneous) and an initial condition. The initial condition is only here because it belongs here, but we will be ignoring it until we get to the next section. The method of separation of variables tells us to assume that the solution will take the form of the product,
u ( x, t ) = j ( x ) G ( t )
454
2007 Paul Dawkins
so all we really need to do here is plug this into the differential equation and see what we get.
2 j ( x ) G ( t ) ) = k 2 (j ( x ) G ( t ) ) ( t x dG d 2j j ( x) = k G (t ) 2 dt dx
As shown above we can factor the j ( x ) out of the time derivative and we can factor the G ( t ) out of the spatial derivative. Also notice that after weve factored these out we no longer have a partial derivative left in the problem. In the time derivative we are now differentiating only G ( t ) with respect to t and this is now an ordinary derivative. Likewise, in the spatial derivative we are now only differentiating j ( x ) with respect to x and so we again have an ordinary derivative. At this point it probably doesnt seem like weve done much to simplify the problem. However, just the fact that weve gotten the partial derivatives down to ordinary derivatives is liable to be good thing even if it still looks like weve got a mess to deal with. Speaking of that apparent (and yes I said apparent) mess, is it really the mess that it looks like? The idea is to eventually get all the ts on one side of the equation and all the xs on the other side. In other words we want to separate the variables and hence the name of the method. In this case lets notice that if we divide both sides by j ( x ) G ( t ) we get want we want and we should point out that it wont always be as easy as just dividing by the product solution. So, dividing out gives us,
1 dG 1 d 2j =k G dt j dx 2
1 dG 1 d 2j = kG dt j dx 2
Notice that we also divided both sides by k. This was done only for convenience down the road. It doesnt have to be done and nicely enough if it turns out to be a bad idea we can always come back to this step and put it back on the right side. Likewise, if we dont do it and it turns out to maybe not be such a bad thing we can always come back and divide it out. For the time being however, please accept our word that this was a good thing to do for this problem. We will discuss the reasoning for this after were done with this example. Now, while we said that this is what we wanted it still seems like weve got a mess. Notice however that the left side is a function of only t and the right side is a function only of x as we wanted. Also notice these two functions must be equal. Lets think about this for a minute. How is it possible that a function of only ts can be equal to a function of only xs regardless of the choice of t and/or x that we have? This may seem like an impossibility until you realize that there is one way that this can be true. If both functions (i.e. both sides of the equation) were in fact constant and not only a constant, but the same constant then they can in fact be equal. So, we must have,
2007 Paul Dawkins
455
1 dG 1 d 2j = = -l kG dt j dx 2
where the -l is called the separation constant and is arbitrary. The next question that we should now address is why the minus sign? Again, much like the dividing out the k above, the answer is because it will be convenient down the road to have chosen this. The minus sign doesnt have to be there and in fact there are times when we dont want it there. So how do we know it should be there or not? The answer to that is to proceed to the next step in the process (which well see in the next section) and at that point well know if would be convenient to have it or not and we can come back to this step and add it in or take it our depending what we chose to do here. Okay, lets proceed with the process. The next step is to acknowledge that we can take the equation above and split it into the following two ordinary differential equations.
dG = -k lG dt
d 2j = -lj dx 2
Both of these are very simple differential equations, however because we dont know what l is we actually cant solve the spatial one yet. The time equation however could be solved at this point if we wanted to, although that wont always be the case. At this point we dont want to actually think about solving either of these yet however. The last step in the process that well be doing in this section is to also make sure that our product solution, u ( x, t ) = j ( x ) G ( t ) , satisfies the boundary conditions so lets plug it into both of those.
u ( 0, t ) = j ( 0 ) G ( t ) = 0
u ( L, t ) = j ( L ) G ( t ) = 0
Lets consider the first one for a second. We have two options here. Either j ( 0 ) = 0 or
G ( t ) = 0 for every t. However, if we have G ( t ) = 0 for every t then well also have u ( x, t ) = 0 , i.e. the trivial solution, and as we discussed in the previous section this is definitely
a solution to any linear homogeneous equation we would really like a non-trivial solution. Therefore we will assume that in fact we must have j ( 0 ) = 0 . Likewise, from the second boundary condition we will get j ( L ) = 0 to avoid the trivial solution. Note as well that we were only able to reduce the boundary conditions down like this because they were homogeneous. Had they not been homogeneous we could not have done this. So, after applying separation of variables to the given partial differential equation we arrive at a 1st order differential equation that well need to solve for G ( t ) and a 2nd order boundary value problem that well need to solve for j ( x ) . The point of this section however is just to get to this
point and well hold off solving these until the next section. Lets summarize everything up that weve determined here.
dG = -k lG dt
d 2j + lj = 0 dx 2 j ( 0) = 0 j ( L) = 0
and note that we dont have a condition for the time differential equation and is not a problem. Also note that we rewrote the second one a little. Okay, so just what have we learned here? By using separation of variables we were able to reduce our linear homogeneous partial differential equation with linear homogeneous boundary conditions down to an ordinary differential equation for one of the functions in our product solution (13), G ( t ) in this case, and a boundary value problem that we can solve for the other function, j ( x ) in this case. Note as well that the boundary value problem is in fact an eigenvalue/eigenfunction problem. When we solve the boundary value problem we will be identifying the eigenvalues, l , that will generate non-trivial solutions the their corresponding eigenfunctions. Again, well look into this more in the next section. At this point all we want to do is identify the two ordinary differential equations that we need to solve to get a solution. Before we do a couple of other examples we should take a second to address the fact that we made two very arbitrary seeming decisions in the above work. We divided both sides of the equation by k at one point and chose to use -l instead of l as the separation constant. Both of these decisions were made to simplify the solution to the boundary value problem we got from our work. The addition of the k in the boundary value problem would just have complicated the solution process with another letter wed have to keep track of so we moved it into the time problem were it wont cause as many problems in the solution process. Likewise, we chose -l because weve already solved that particular boundary value problem (albeit with a specific L, but the work will be nearly identical) when we first looked at finding eigenvalues and eigenfunctions. This by the way was the reason we rewrote the boundary value problem to make it a little clearer that we have in fact solved this one already. We can now at least partially answer the question of how do we know to make these decisions. We wait until we get the ordinary differential equations and then look at them and decide of moving things like the k or which separation constant to use based on how it will affect the solution of the ordinary differential equations. There is also, of course, a fair amount of experience that comes into play at this stage. The more experience you have in solving these the easier it often is to make these decisions. Again, we need to make clear here that were not going to go any farther in this section than getting things down to the two ordinary differential equations. Of course we will need to solve them in order to get a solution to the partial differential equation but that is the topic of the remaining sections in this chapter. All well say about it here is that we will need to first solve the boundary value problem, which will tell us what l must be and then we can solve the other differential equation. Once that is done we can then turn our attention to the initial condition.
Okay, we need to work a couple of other examples and these will go a lot quicker because we wont need to put in all the explanations. After the first example this process always seems like a very long process but it really isnt. It just looked that way because of all the explanation that we had to put into it. So, lets start off with a couple of more examples with the heat equation using different boundary conditions.
Example 2 Use Separation of Variables on the following partial differential equation. u 2u =k 2 t x u u u ( x, 0 ) = f ( x ) ( 0, t ) = 0 ( L, t ) = 0 x x

Solution In this case were looking at the heat equation with no sources and perfectly insulated boundaries. So, well start off by again assuming that our product solution will have the form,
u ( x, t ) = j ( x ) G ( t )
and because the differential equation itself hasnt changed here we will get the same result from plugging this in as we did in the previous example so the two ordinary differential equations that well need to solve are,
dG = -k lG dt
d 2j = -lj dx 2
Now, the point of this example was really to deal with the boundary conditions so lets plug the product solution into them to get,
(G (t )j ( x )) x G (t )
( 0, t ) = 0
(G (t )j ( x )) x G (t )
( L, t ) = 0
dj (0) = 0 dx
dj (L) = 0 dx
Now, just as with the first example if we want to avoid the trivial solution and so we cant have G ( t ) = 0 for every t and so we must have,
dj (0) = 0 dx dG = -k lG dt
dj ( L) = 0 dx d 2j + lj = 0 dx 2 dj dj (0) = 0 ( L) = 0 dx dx
Here is a summary of what we get by applying separation of variables to this problem.
Next, lets see what we get if use periodic boundary conditions with the heat equation.
Example 3 Use Separation of Variables on the following partial differential equation. u 2u =k 2 t x u u u ( x, 0 ) = f ( x ) u ( - L, t ) = u ( L, t ) ( - L, t ) = ( L, t ) x x

Solution First note that these boundary conditions really are homogeneous boundary conditions. If we rewrite them as,
u ( - L, t ) - u ( L, t ) = 0
its a little easier to see.
u u ( - L, t ) - ( L, t ) = 0 x x
Now, again weve done this partial differential equation so well start off with,
u ( x, t ) = j ( x ) G ( t )
and the two ordinary differential equations that well need to solve are,
dG = -k lG dt
d 2j = -lj dx 2
Plugging the product solution into the rewritten boundary conditions gives,
G (t )j ( -L ) - G (t )j ( L ) = G (t ) j ( - L ) - j ( L ) =0 G (t )
dj dj dj dj =0 ( - L ) - G (t ) ( L ) = G (t ) ( - L ) - ( L ) dx dx dx dx dj dj ( -L ) - ( L) = 0 dx dx dj dj (-L) = ( L) dx dx
and we can see that well only get non-trivial solution if,
j (-L) - j ( L ) = 0 j (-L) = j ( L )
So, here is what we get by applying separation of variables to this problem.
dG = - k lG dt
d 2j + lj = 0 dx 2
j ( -L) = j ( L )
dj dj ( -L ) = ( L) dx dx
Lets now take a look at what we get by applying separation of variables to the wave equation with fixed boundaries.
2007 Paul Dawkins
459
Example 4 Use Separation of Variables on the following partial differential equation. 2 2u 2 u c = t 2 x 2 u u ( x, 0 ) = f ( x ) ( x, 0 ) = g ( x ) t u ( 0, t ) = 0 u ( L, t ) = 0

Solution Now, as with the heat equation the two initial conditions are here only because they need to be here for the problem. We will not actually be doing anything with them here and as mentioned previously the product solution will rarely satisfy them. We will be dealing with those in a later section when we actually go past this first step. Again, the point of this example is only to get down to the two ordinary differential equations that separation of variables gives. So, lets get going on that and plug the product solution, u ( x, t ) = j ( x ) h ( t ) (we switched the G to an h here to avoid confusion with the g in the second initial condition) into the wave equation to get,
2 2 2 j x h t = c ( ) ( ) ( ) (j ( x ) h ( t ) ) t 2 x 2 d 2h d 2j j ( x ) 2 = c2 h (t ) 2 dt dx 2 2 1 d h 1d j = 2 c h dt 2 j dx 2
Note that we moved the c 2 to the right side for the same reason we moved the k in the heat equation. It will make solving the boundary value problem a little easier. Now that weve gotten the equation separated into a function of only t on the left and a function of only x on the right we can introduce a separation constant and again well use -l so we can arrive at a boundary value problem that we are familiar with. So, after introducing the separation constant we get,
1 d 2 h 1 d 2j = = -l c 2 h dt 2 j dx 2
The two ordinary differential equations we get are then,
d 2h = -l c 2 h 2 dt
d 2j = -lj dx 2
The boundary conditions in this example are identical to those from the first example and so plugging the product solution into the boundary conditions gives,
j ( 0) = 0
j ( L) = 0
Applying separation of variables to this problem gives,

d 2h = -l c 2 h 2 dt
d 2j = -lj dx 2 j ( 0) = 0
j ( L) = 0
Next, lets take a look at the 2-D Laplaces Equation.
Example 5 Use Separation of Variables on the following partial differential equation. 2 u 2u 0 x L 0 y H + =0 x 2 y 2 u ( 0, y ) = g ( y ) u ( x, 0 ) = 0 u ( L, y ) = 0 u ( x, H ) = 0
Solution This problem is a little (well actually quite a bit in some ways) different from the heat and wave equations. First, we no longer really have a time variable in the equation but instead we usually consider both variables to be spatial variables and well be assuming that the two variables are in the ranges shown above in the problems statement. Note that this also means that we no longer have initial conditions, but instead we now have two sets of boundary conditions, one for x and one for y. Also, we should point out that we have three of the boundary conditions homogeneous and one nonhomogeneous for a reason. When we get around to actually solving this Laplaces Equation well see that this is in fact required in order for us to find a solution. For this problem well use the product solution,
u ( x, y ) = h ( x ) j ( y )
It will often be convenient to have the boundary conditions in hand that this product solution gives before we take care of the differential equation. In this case we have three homogeneous boundary conditions and so well need to convert all of them. Because weve already converted these kind of boundary conditions well leave it to you to verify that these will become,
h(L) = 0
j ( 0) = 0
j (H ) = 0
Plugging this into the differential equation and separating gives,
2 2 h x j y + ( ) ( ) ( ) y 2 ( h ( x ) j ( y ) ) = 0 x 2
j ( y)
d 2h d 2j + h x =0 ( ) dx 2 dy 2 1 d 2h 1 d 2j = h dx 2 j dy 2
Okay, now we need to decide upon a separation constant. Note that every time weve chosen the separation constant we did so to make sure that the differential equation
d 2j + lj = 0 dx 2
would show up. Of course, the letters might need to be different depending on how we defined our product solution (as theyll need to be here). We know how to solve this eigenvalue/eigenfunction problem as we pointed out in the discussion after the first example. However, in order to solve it we need two boundary conditions. So, for our problem here we can see that weve got two boundary conditions for j ( y ) but only one for h ( x ) and so we can see that the boundary value problem that well have to solve will involve j ( y ) and so we need to pick a separation constant that will give use the boundary value problem weve already solved. In this case that means that we need to choose l for the separation constant. If youre not sure you believe that yet hold on for a second and youll soon see that it was in fact the correct choice here. Putting the separation constant gives,
1 d 2h 1 d 2j = =l j dy 2 h dx 2
The two ordinary differential equations we get from Laplaces Equation are then,
d 2h = lh dx 2
and notice that if we rewrite these a little we get,
d 2j = lj dy 2
d 2h - lh = 0 dx 2
d 2j + lj = 0 dy 2
We can now see that the second one does now look like one weve already solved (with a small change in letters of course, but that really doesnt change things). So, lets summarize up here.
d 2h - lh = 0 dx 2 h(L) = 0
d 2j + lj = 0 dy 2
j ( 0) = 0
j (H ) = 0
So, weve finally seen an example where the constant of separation didnt have a minus sign and again note that we chose it so that the boundary value problem we need to solve will match one weve already seen how to solve so there wont be much work to there. All the examples worked in this section to this point are all problems that well continue in later sections to get full solutions for. Lets work one more however to illustrate a couple of other ideas. We will not however be doing any work with this in later sections however, it is only hear to illustrate a couple of points.
2007 Paul Dawkins
462
Example 6 Use Separation of Variables on the following partial differential equation. u 2u = k 2 -u t x u u ( x, 0 ) = f ( x ) u ( 0, t ) = 0 - ( L, t ) = u ( L, t ) x

Solution Note that this is a heat equation with the source term of Q ( x, t ) = -c r u and is both linear and homogenous. Also note that for the first time weve mixed boundary condition types. At x = 0 weve got a prescribed temperature and at x = L weve got a Newtons law of cooling type boundary condition. We should not come away from the first few examples with the idea that the boundary conditions at both boundaries always the same type. Having them the same type just makes the boundary value problem a little easier to solve in many cases. So well start off with,
u ( x, t ) = j ( x ) G ( t ) 2 j ( x ) G ( t ) ) = k 2 (j ( x ) G ( t ) ) - j ( x ) G ( t ) ( t x dG d 2j j ( x) = k G (t ) 2 - j ( x ) G ( t ) dt dx
and plugging this into the partial differential equation gives,
Now, the next step is to divide by j ( x ) G ( t ) and notice that upon doing that the second term on the right will become a one and so can go on either side. Theoretically there is no reason that the one cant be on either side, however from a practical standpoint we again want to keep things a simple as possible so well move it to the t side as this will guarantee that well get a differential equation for the boundary value problem that weve seen before. So, separating and introducing a separation constant gives,
1 1 dG 1 d 2j + 1 = = -l 2 k G dt j dx
The two ordinary differential equations that we get are then (with some rewriting),
dG = - ( l k + 1) G dt
Now lets deal with the boundary conditions.
d 2j = -lj dx 2
G (t )j ( 0) = 0 G (t )
dj dj =0 ( L ) + G (t )j ( L ) = G (t ) ( L ) + j ( L ) dx dx dj ( L) + j ( L) = 0 dx
and we can see that well only get non-trivial solution if,
j ( 0) = 0
So, here is what we get by applying separation of variables to this problem.
dG = - ( l k + 1) G dt
d 2j + lj = 0 dx 2
j ( 0) = 0
dj ( L) + j ( L) = 0 dx
On a quick side note we solved the boundary value problem in this example in Example 5 of the Eigenvalues and Eigenfunctions section and that example illustrates why separation of variables is not always so easy to use. As well see in the next section to get a solution that will satisfy any sufficiently nice initial condition we really need to get our hands on all the eigenvalues for the boundary value problem. However, as the solution to this boundary value problem shows this is not always possible to do. There are ways (which we wont be going into here) to use the information here to at least get approximations to the solution but we wont ever be able to get a complete solution to this problem. Okay, thats it for this section. It is important to remember at this point that what weve done here is really only the first step in the separation of variables method for solving partial differential equations. In the upcoming sections well be looking at what we need to do to finish out the solution process and in those sections well finish the solution to the partial differential equations we started in Example 1 Example 5 above. Also, in the Laplaces Equation section the last two examples show pretty much the whole separation of variable process from defining the product solution to getting an actual solution. The only step thats missing from those two examples is the solving of a boundary value problem that will have been already solved at that point and so was not put into the solution given that they tend to be fairly lengthy to solve. Well also see a worked example (without the boundary value problem work again) in the Vibrating String section.
2007 Paul Dawkins
464
Solving the Heat Equation

Okay, it is finally time to completely solve a partial differential equation. In the previous section we applied separation of variables to several partial differential equations and reduced the problem down to needing to solve two ordinary differential equations. In this section we will now solve those ordinary differential equations and use the results to get a solution to the partial differential equation. We will be concentrating on the heat equation in this section and will do the wave equation and Laplaces equation in later sections. The first problem that were going to look at will be the temperature distribution in a bar with zero temperature boundaries. We are going to do the work in a couple of steps so we can take our time and see how everything works. The first thing that we need to do is find a solution that will satisfy the partial differential equation and the boundary conditions. At this point we will not worry about the initial condition. The solution well get first will not satisfy the vast majority of initial conditions but as well see it can be used to find a solution that will satisfy a sufficiently nice initial condition.
Example 1 Find a solution to the following partial differential equation that will also satisfy the
boundary conditions.
u 2u =k 2 t x u ( x, 0 ) = f ( x )
u ( 0, t ) = 0
u ( L, t ) = 0
Solution Okay the first thing we technically need to do here is apply separation of variables. Even though we did that in the previous section lets recap here what we did. First, we assume that the solution will take the form,
u ( x, t ) = j ( x ) G ( t )
and we plug this into the partial differential equation and boundary conditions. We separate the equation to get a function of only t on one side and a function of only x on the other side and then introduce a separation constant. This leaves us with two ordinary differential equations. We did all of this in Example 1 of the previous section and the two ordinary differential equations are,
dG = -k lG dt
d 2j + lj = 0 dx 2 j ( 0) = 0 j ( L) = 0
The time dependent equation can really be solved at any time, but since we dont know what l is yet lets hold off on that one. Also note that in many problems only the boundary value problem can be solved at this point so dont always expect to be able to solve either one at this point. The spatial equation is a boundary value problem and we know from our work in the previous chapter that it will only have non-trivial solutions (which we want) for certain values of l , which well recall are called eigenvalues. Once we have those we can determine the non-trivial solutions for each l , i.e. eigenfunctions.
Now, we actually solved the spatial problem,
d 2j + lj = 0 dx 2 j ( 0) = 0 j ( L) = 0
in Example 1 of the Eigenvalues and Eigenfunctions section of the previous chapter for L = 2p . So, because weve solved this once for a specific L and the work is not all that much different for a general L were not going to be putting in a lot of explanation here and if you need a reminder on how something works or why we did something go back to Example 1 from the Eigenvalues and Eigenfunctions section for a reminder. Weve got three cases to deal with so lets get going.
l >0
In this case we know the solution to the differential equation is,
j ( x ) = c1 cos
l x + c2 sin
lx
0 = j ( 0 ) = c1
Now applying the second boundary condition, and using the above result of course, gives,
0 = j ( L ) = c2 sin L l
Now, we are after non-trivial solutions and so this means we must have,
sin L l = 0
L l = np
n = 1, 2,3,K
The positive eigenvalues and their corresponding eigenfunctions of this boundary value problem are then,
np ln = L
np x jn ( x ) = sin L
n = 1, 2,3,K
Note that we dont need the c2 in the eigenfunction as it will just get absorbed into another constant that well be picking up later on.
l =0
The solution to the differential equation in this case is,
j ( x ) = c1 + c2 x
0 = j ( L ) = c2 L c2 = 0 So, in this case the only solution is the trivial solution and so l = 0 is not an eigenvalue for this
boundary value problem.
0 = j ( 0 ) = c1
l<0
Here the solution to the differential equation is,
j ( x ) = c1 cosh
-l x + c2 sinh
-l x
Applying the first boundary condition gives, and applying the second gives,
0 = j ( 0 ) = c1
0 = j ( L ) = c2 sinh L -l
) ( )
So, we are assuming l < 0 and so L -l 0 and this means sinh L - l 0 . We therefore we must have c2 = 0 and so we can only get the trivial solution in this case. Therefore, there will be no negative eigenvalues for this boundary value problem. The complete list of eigenvalues and eigenfunctions for this problem are then,
np ln = L
np x jn ( x ) = sin L
n = 1, 2,3,K
Now lets solve the time differential equation,
dG = - k lnG dt and note that even though we now know l were not going to plug it in quite yet to keep the mess to a minimum. We will however now use ln to remind us that we actually have an infinite
number of possible values here. This is a simple linear (and separable for that matter) 1st order differential equation and so well let you verify that the solution is,
G ( t ) = ce
- kl nt
= ce
np -k t L
Okay, now that weve gotten both of the ordinary differential equations solved we can finally write down a solution. Note however that we have in fact found infinitely many solutions since there are infinitely many solutions (i.e. eigenfunctions) to the spatial problem. Our product solution are then,
t np x - k L un ( x, t ) = Bn sin e n = 1, 2,3,K L Weve denoted the product solution un to acknowledge that each value of n will yield a different solution. Also note that weve changed the c in the solution to the time problem to Bn to denote np
2
the fact that it will probably be different for each value of n as well and because had we kept the c2 with the eigenfunction wed have absorbed it into the c to get a single constant in our solution.
So, there we have it. The function above will satisfy the heat equation and the boundary condition of zero temperature on the ends of the bar. The problem with this solution is that it simply will not satisfy almost every possible initial condition we could possibly want to use. That does not mean however, that there arent at least a few that it will satisfy as the next example illustrates.
Example 2 Solve the following heat problem for the given initial conditions. u 2u =k 2 t x u ( x, 0 ) = f ( x ) u ( 0, t ) = 0 u ( L, t ) = 0
(a) f ( x ) = 6sin
px L 9p x 4p x (b) f ( x ) = 12sin - 7 sin L L
Solution (a) This is actually easier than it looks like. All we need to do is choose n = 1 and B1 = 6 in the product solution above to get,
p x -k L u ( x, t ) = 6sin e L p
2
and weve got the solution we need. This is a product solution for the first example and so satisfies the partial differential equation and boundary conditions and will satisfy the initial condition since plugging in t = 0 will drop out the exponential. (b) This is almost as simple as the first part. Recall from the Principle of Superposition that if we have two solutions to a linear homogeneous differential equation (which weve got here) then their sum is also a solution. So, all we need to do is choose n and Bn as we did in the first part to get a solution that satisfies each part of the initial condition and then add them up. Doing this gives,
9p x - k u ( x, t ) = 12sin e L
9p t L
4p x - k L - 7sin e L
4p
Well leave it to you to verify that this does in fact satisfy the initial condition and the boundary conditions. So, weve seen that our solution from the first example will satisfy at least a small number of highly specific initial conditions. Now, lets extend the idea out that we used in the second part of the previous example a little to see how we can get a solution that will satisfy any sufficiently nice initial condition. The Principle of Superposition is, of course, not restricted to only two solutions. For instance the following is also a solution to the partial differential equation.
np x - k L u ( x, t ) = Bn sin e L n =1 M
np
and notice that this solution will not only satisfy the boundary conditions but it will also satisfy the initial condition,
M np x u ( x, 0 ) = Bn sin L n =1
Lets extend this out even further and take the limit as M . Doing this our solution now becomes,
np x - k L u ( x, t ) = Bn sin e L n =1 np
2
This solution will satisfy any initial condition that can be written in the form,
np x u ( x, 0 ) = f ( x ) = Bn sin L n =1
This may still seem to be very restrictive, but the series on the right should look awful familiar to you after the previous chapter. The series on the left is exactly the Fourier sine series we looked at in that chapter. Also recall that when we can write down the Fourier sine series for any piecewise smooth function on 0 x L . So, provided our initial condition is piecewise smooth after applying the initial condition to our solution we can determine the Bn as if we were finding the Fourier sine series of initial condition. So we can either proceed as we did in that section and use the orthogonality of the sines to derive them or we can acknowledge that weve already done that work and know that coefficients are given by,
2 np x Bn = f ( x ) sin dx L 0 L
n = 1, 2,3,K
So, we finally can completely solve a partial differential equation.
Example 3 Solve the following BVP. u 2u =k 2 t x u ( x, 0 ) = 20
u ( 0, t ) = 0
u ( L, t ) = 0
Solution There isnt really all that much to do here as weve done most of it in the examples and discussion above. First, the solution is,
np x - k L u ( x, t ) = Bn sin e L n =1 np
2
2007 Paul Dawkins
469
The coefficients are given by,
L 2 2 20 L (1 - cos ( np ) ) 40 1 - ( -1) np x Bn = 20sin dx = = L 0 L n np p L
If we plug these in we get the solution,
u ( x, t ) =
n =1
40 1- -1 n ( )
np
np x - k L sin e L
np
That almost seems anti-climactic. This was a very short problem. Of course some of that came about because we had a really simple constant initial condition and so the integral was very simple. However, dont forget all the work that we had to put into discussing Fourier sine series, solving boundary value problems, applying separation of variables and then putting all of that together to reach this point. While the example itself was very simple, it was only simple because of all the work that we had to put into developing the ideas that even allowed us to do this. Because of how simple it will often be to actually get these solutions were not actually going to do anymore with specific initial conditions. We will instead concentrate on simply developing the formulas that wed be required to evaluate in order to get an actual solution. So, having said that lets move onto the next example. In this case were going to again look at the temperature distribution in a bar with perfectly insulated boundaries. We are also no longer going to go in steps. We will do the full solution as a single example and end up with a solution that will satisfy any piecewise smooth initial condition.
Example 4 Find a solution to the following partial differential equation. u 2u =k 2 t x u u u ( x, 0 ) = f ( x ) ( 0, t ) = 0 ( L, t ) = 0 x x

Solution We applied separation of variables to this problem in Example 2 of the previous section. So, after assuming that our solution is in the form,
u ( x, t ) = j ( x ) G ( t )
and applying separation of variables we get the following two ordinary differential equations that we need to solve.
dG = -k lG dt
d 2j + lj = 0 dx 2 dj dj (0) = 0 ( L) = 0 dx dx
We solved the boundary value problem in Example 2 of the Eigenvalues and Eigenfunctions section of the previous chapter for L = 2p so as with the first example in this section were not
going to put a lot of explanation into the work here. If you need a reminder on how this works go back to the previous chapter and review the example we worked there. Lets get going on the three cases weve got to work for this problem.
l >0
The solution to the differential equation is,
j ( x ) = c1 cos
l x + c2 sin
lx
)
c2 = 0
0=
dj ( 0 ) = l c2 dx
0=
dj ( L ) = - l c1 sin L l dx
)
n = 1, 2,3,K
Recall that l > 0 and so we will only get non-trivial solutions if we require that,
sin L l = 0
L l = np
The positive eigenvalues and their corresponding eigenfunctions of this boundary value problem are then,
np ln = L
np x jn ( x ) = cos L
n = 1, 2,3,K
l =0
j ( x ) = c1 + c2 x
dj ( 0 ) = c2 dx
0=
j ( x ) = c1 j ( x) = 1
and note that this will trivially satisfy the second boundary condition. Therefore l = 0 is an eigenvalue for this BVP and the eigenfunctions corresponding to this eigenvalue is,
l<0
j ( x ) = c1 cosh
2007 Paul Dawkins
-l x + c2 sinh
-l x
471
0=
dj ( 0 ) = -l c2 dx
c2 = 0
0=
dj ( L ) = -l c1 sinh L -l dx
We know that L -l 0 and so sinh L - l 0 . Therefore we must have c1 = 0 and so, this boundary value problem will have no negative eigenvalues. So, the complete list of eigenvalues and eigenfunctions for this problem is then,
2
np ln = L l0 = 0
np x jn ( x ) = cos L j0 ( x ) = 1
n = 1, 2,3,K
and notice that we get the l 0 = 0 eigenvalue and its eigenfunction if we allow n = 0 in the first set and so well use the following as our set of eigenvalues and eigenfunctions.
np ln = L
np -k t L
2
n = 0,1, 2, 3, K
The time problem here is identical to the first problem we looked at so,
G ( t ) = ce
Our product solutions will then be,
np x - k L un ( x, t ) = An cos e L np
2
n = 0,1, 2, 3, K
np
2
and the solution to this partial differential equation is,

np x - k L u ( x, t ) = An cos e L n =0
If we apply the initial condition to this we get,

np x u ( x, 0 ) = f ( x ) = An cos L n =0
and we can see that this is nothing more than the Fourier cosine series for f ( x ) on 0 x L and so again we could use the orthogonality of the cosines to derive the coefficients or we could recall that weve already done that in the previous chapter and know that the coefficients are given by,
2007 Paul Dawkins
472
n=0 n0
The last example that were going to work in this section is a little different from the first two. We are going to consider the temperature distribution in a thin circular ring. We will consider the lateral surfaces to be perfectly insulated and we are also going to assume that the ring is thin enough so that the temperature does not vary with distance from the center of the ring. So, what does that leave us with? Lets set x = 0 as shown below and then let x be the arc length of the ring as measured from this point.
We will measure x as positive if we move to the right and negative if we move to the left of x = 0 . This means that at the top of the ring well meet where x = L (if we move to the right) and x = - L (if we move to the left). By doing this we can consider this ring to be a bar of length 2L and the heat equation that we developed earlier in this chapter will still hold. At the point of the ring we consider the two ends to be in perfect thermal contact. This means that at the two ends both the temperature and the heat flux must be equal. In other words we must have,
u ( - L, t ) = u ( L, t )
u u ( - L, t ) = ( L, t ) x x
If you recall from the section in which we derived the heat equation we called these periodic boundary conditions. So, the problem we need to solve to get the temperature distribution in this case is,
Example 5 Find a solution to the following partial differential equation. u 2u =k 2 t x u u u ( x, 0 ) = f ( x ) u ( - L, t ) = u ( L, t ) ( - L, t ) = ( L, t ) x x

Solution We applied separation of variables to this problem in Example 3 of the previous section. So, if
we assume the solution is in the form,
u ( x, t ) = j ( x ) G ( t ) d 2j + lj = 0 dx 2
we get the following two ordinary differential equations that we need to solve.
dG = - k lG dt
j ( -L) = j ( L )
dj dj ( -L ) = ( L) dx dx
As weve seen with the previous two problems weve already solved a boundary value problem like this one back in the Eigenvalues and Eigenfunctions section of the previous chapter, Example 3 to be exact with L = p . So, if you need a little more explanation of whats going on here go back to this example and you can see a little more explanation. We again have three cases to deal with here.
l >0
The general solution to the differential equation is,
j ( x ) = c1 cos
l x + c2 sin
lx
) ( )
Applying the first boundary condition and recalling that cosine is an even function and sine is an odd function gives us,
c1 cos - L l + c2 sin - L l = c1 cos L l + c2 sin L l

2 2
) ( ) -c sin ( L l ) = c sin ( L l ) 0 = 2c sin ( L l )

2
At this stage we cant really say anything as either c2 or sine could be zero. So, lets apply the second boundary condition and see what we get.
- l c1 sin - L l + l c2 cos - L l = - l c1 sin L l + l c2 cos L l

1 1
) l c sin ( L l ) = l c sin ( L l ) = 0
1
( ) l c sin ( L l ) )
We get something similar. However notice that if sin L l 0 then we would be forced to have c1 = c2 = 0 and this would give us the trivial solution which we dont want. This means therefore that we must have sin L l = 0 which in turn means (from work in our previous examples) that the positive eigenvalues for this problem are,
2007 Paul Dawkins
474
np ln = L
n = 1, 2,3,K
Now, there is no reason to believe that c1 = 0 or c2 = 0 . All we know is that they both cant be zero and so that means that we in fact have two sets of eigenfunctions for this problem corresponding to positive eigenvalues. They are,
np x j n ( x ) = sin L
n = 1, 2,3,K
l =0
The general solution in this case is,
j ( x ) = c1 + c2 x
c1 + c2 ( - L ) = c1 + c2 ( L ) 2 Lc2 = 0 c2 = 0
j ( x ) = c1 j ( x) = 1
and this will trivially satisfy the second boundary condition. Therefore l = 0 is an eigenvalue for this BVP and the eigenfunctions corresponding to this eigenvalue is,
l<0
For this final case the general solution here is,
j ( x ) = c1 cosh
-l x + c2 sinh
-l x
) ( )
Applying the first boundary condition and using the fact that hyperbolic cosine is even and hyperbolic sine is odd gives,
c1 cosh - L -l + c2 sinh - L -l = c1 cosh L - l + c2 sinh L -l

2 2
( -c sinh ( - L 2c sinh ( L
2
) ( -l ) = c sinh ( L -l ) = 0
) -l )
Now, in this case we are assuming that l < 0 and so L -l 0 . This turn tells us that
sinh L - l 0 . We therefore must have c2 = 0 .

Lets now apply the second boundary condition to get,
2007 Paul Dawkins
475
- l c1 sinh - L - l = -l c1 sinh L -l 2
1
) -l c sinh ( L - l ) = 0
)
c1 = 0
By our assumption on l we again have no choice here but to have c1 = 0 and so for this boundary value problem there are negative eigenvalues. Summarizing up then we have the following sets of eigenvalues and eigenfunctions and note that weve merged the l = 0 case into the cosine case since it can be here to simplify things up a little.
np ln = L np ln = L
np x jn ( x ) = cos L np x jn ( x ) = sin L
np -k t L
2
n = 0,1, 2, 3, K n = 1, 2,3,K
The time problem is again identical to the two weve already worked here and so we have,
G ( t ) = ce
Now, this example is a little different from the previous two heat problems that weve looked at. In this case we actually have two different possible product solutions that will satisfy the partial differential equation and the boundary conditions. They are,
np x - k un ( x, t ) = An cos e L L np x - k un ( x, t ) = Bn sin e L L np np
2
n = 0,1, 2, 3, K n = 1, 2,3,K
The Principle of Superposition is still valid however and so a sum of any of these will also be a solution and so the solution to this partial differential equation is,
t np x - k np x - k L L u ( x, t ) = An cos e + Bn sin e L L n =0 n =1 np
2
np
If we apply the initial condition to this we get,

np x np x u ( x, 0 ) = f ( x ) = An cos + Bn sin L n =1 L n =0
and just as we saw in the previous two examples we get a Fourier series. The difference this time is that we get the full Fourier series for a piecewise smooth initial condition on - L x L . As noted for the previous two examples we could either rederive formulas for the coefficients using the orthogonality of the sines and cosines or we can recall the work weve already done. Theres really no reason at this point to redo work already done so the coefficients are given by,
1 L f ( x ) dx 2L -L 1 L An = f ( x ) cos L -L 1 L Bn = f ( x ) sin L -L A0 =
( npL x ) dx
n = 1, 2,3,K n = 1, 2,3,K
( npL x ) dx
Note that this is the reason for setting up x as we did at the start of this problem. A full Fourier series needs an interval of - L x L whereas the Fourier sine and cosines series we saw in the first two problems need 0 x L . Okay, weve now seen three heat equation problems solved and so well leave this section. You might want to go through and do the two cases where we have a zero temperature on one boundary and a perfectly insulated boundary on the other to see if youve got this process down.
2007 Paul Dawkins
477
Heat Equation with Non-Zero Temperature Boundaries

In this section we want to expand one of the cases from the previous section a little bit. In the previous section we look at the following heat problem.
u 2u =k 2 t x u ( x, 0 ) = f ( x )
u ( 0, t ) = 0
u ( L, t ) = 0
Now, there is nothing inherently wrong with this problem, but the fact that were fixing the temperature on both ends at zero is a little unrealistic. The other two problems we looked at, insulated boundaries and the thin ring, are a little more realistic problems, but this one just isnt all that realistic so wed like to extend it a little. What wed like to do in this section is instead look at the following problem.
u 2u =k 2 t x u ( x, 0 ) = f ( x )
u ( 0, t ) = T1
u ( L, t ) = T2
(14)
In this case well allow the boundaries to be any fixed temperature, T1 or T2 . The problem here is that separation of variables will no longer work on this problem because the boundary conditions are no longer homogeneous. Recall that separation of variables will only work if both the partial differential equation and the boundary conditions are linear and homogeneous. So, were going to need to deal with the boundary conditions in some way before we actually try and solve this. Luckily for us there is an easy way to deal with them. Lets consider this problem a little bit. There are no sources to add/subtract heat energy anywhere in the bar. Also our boundary conditions are fixed temperatures and so cant change with time and we arent prescribing a heat flux on the boundaries to continually add/subtract heat energy. So, what this all means is that there will not ever be any forcing of heat energy into or out of the bar and so while some heat energy may well naturally flow into our out of the bar at the end points as the temperature changes eventually the temperature distribution in the bar should stabilize out and no longer depend on time. Or, in other words it makes some sense that we should expect that as t our temperature distribution, u ( x, t ) should behave as,
lim u ( x, t ) = uE ( x )
where u E ( x ) is called the equilibrium temperature. Note as well that is should still satisfy the heat equation and boundary conditions. It wont satisfy the initial condition however because it is the temperature distribution as t whereas the initial condition is at t = 0 . So, the equilibrium temperature distribution should satisfy,
t
d 2 uE 0= dx 2
2007 Paul Dawkins
uE ( 0 ) = T1
uE ( L ) = T2
(15)
478
This is a really easy 2nd order ordinary differential equation to solve. If we integrate twice we get,
uE ( x ) = c1 x + c2
and applying the boundary conditions (well leave this to you to verify) gives us,
uE ( x ) = T1 +
T2 - T1 x L
Okay, just what does this have to do with solving the problem given by (14) above? Well lets define the function,
v ( x, t ) = u ( x, t ) - u E ( x )
(16)
where u ( x, t ) is the solution to (14) and u E ( x ) is the equilibrium temperature for (14). Now lets rewrite this as, and lets take some derivatives.
u ( x, t ) = v ( x, t ) + u E ( x ) 2 u 2 v 2 uE 2 v = + = 2 x 2 x 2 x 2 x
u v uE v = + = t t t t
In both of these derivatives we used the fact that u E ( x ) is the equilibrium temperature and so is independent of time t and must satisfy the differential equation in (15). What this tells us is that both u ( x, t ) and v ( x, t ) must satisfy the same partial differential equation. Lets see what the initial conditions and boundary conditions would need to be for v ( x, t ) .
v ( x, 0 ) = u ( x, 0 ) - u E ( x ) = f ( x ) - uE ( x ) v ( 0, t ) = u ( 0, t ) - uE ( 0 ) = T1 - T1 = 0 v ( L, t ) = u ( L, t ) - u E ( L ) = T2 - T2 = 0
So, the initial condition just gets potentially messier, but the boundary conditions are now homogeneous! The partial differential equation that v ( x, t ) must satisfy is,
v 2v =k 2 t x v ( x, 0 ) = f ( x ) - u E ( x )
v ( 0, t ) = 0
2
v ( L, t ) = 0
We saw how to solve this in the previous section and so we the solution is,
np x - k L v ( x, t ) = Bn sin e L n =1 np
where the coefficients are given by,
2 np x Bn = ( f ( x ) - uE ( x ) ) sin dx L 0 L
n = 1, 2,3,K
The solution to (14) is then,
u ( x, t ) = uE ( x ) + v ( x, t )
T -T np x - k = T1 + 2 1 x + Bn sin e L L L n =1 np
2
and the coefficients are given above.
2007 Paul Dawkins
480
Laplaces Equation
The next partial differential equation that were going to solve is the 2-D Laplaces equation,
2u =
2u 2u + =0 x 2 y 2
A natural question to ask before we start learning how to solve this is does this equation come up naturally anywhere? The answer is a very resounding yes! If we consider the 2-D heat equation,
u Q = k 2u + t cp
We can see that Laplaces equation would correspond to finding the equilibrium solution (i.e. time independent solution) if there were not sources. So, this is an equation that can arise from physical situations. How we solve Laplaces equation will depend upon the geometry of the 2-D object were solving it on. Lets start out by solving it on the rectangle given by 0 x L , 0 y H . For this geometry Laplaces equation along with the four boundary conditions will be,
2u =
2u 2u + =0 x 2 y 2 u ( L, y ) = g 2 ( y ) u ( x, H ) = f 2 ( x ) (17)
u ( 0, y ) = g1 ( y ) u ( x, 0 ) = f1 ( x )
One of the important things to note here is that unlike the heat equation we will not have any initial conditions here. Both variables are spatial variables and each variable occurs in a 2nd order derivative and so well need two boundary conditions for each variable. Next, lets notice that while the partial differential equation is both linear and homogeneous the boundary conditions are only linear and are not homogeneous. This creates a problem because separation of variables requires homogeneous boundary conditions. To completely solve Laplaces equation were in fact going to have to solve it four times. Each time we solve it only one of the four boundary conditions can be nonhomogeneous while the remaining three will be homogeneous. The four problems are probably best shown with a quick sketch so lets consider the following sketch.
2007 Paul Dawkins
481
Now, once we solve all four of these problems the solution to our original system, (17), will be,
u ( x, y ) = u1 ( x, y ) + u2 ( x, y ) + u3 ( x, y ) + u4 ( x, y )
Because we know that Laplaces equation is linear and homogeneous and each of the pieces is a solution to Laplaces equation then the sum will also be a solution. Also, this will satisfy each of the four original boundary conditions. Well verify the first one and leave the rest to you to verify.
u ( x, 0 ) = u1 ( x, 0 ) + u2 ( x, 0 ) + u3 ( x, 0 ) + u4 ( x,0 ) = f1 ( x ) + 0 + 0 + 0 = f1 ( x )
In each of these cases the lone nonhomogeneous boundary condition will take the place of the initial condition in the heat equation problems that we solved a couple of sections ago. We will apply separation of variables to the each problem and find a product solution that will satisfy the differential equation and the three homogeneous boundary conditions. Using the Principle of Superposition well find a solution to the problem and then apply the final boundary condition to
determine the value of the constant(s) that are left in the problem. The process is nearly identical in many ways to what we did when we were solving the heat equation. Were going to do two of the cases here and well leave the remaining two for you to do.
Example 1 Find a solution to the following partial differential equation. 2u 2 u 2u4 = 24 + 24 = 0 x y u4 ( 0, y ) = g1 ( y ) u4 ( x, 0 ) = 0 u 4 ( L, y ) = 0 u4 ( x, H ) = 0
Solution Well start by assuming that our solution will be in the form,
u4 ( x, y ) = h ( x ) j ( y )
and then recall that we performed separation of variables on this problem (with a small change in notation) back in Example 5 of the Separation of Variables section. So from that problem we know that separation of variables yields the following two ordinary differential equations that well need to solve.
d 2h - lh = 0 dx 2 h(L) = 0
d 2j + lj = 0 dy 2
j ( 0) = 0
j (H ) = 0
Note that in this case, unlike the heat equation we must solve the boundary value problem first. Without knowing what l is there is no way that we can solve the first differential equation here with only one boundary condition since the sign of l will affect the solution. Lets also notice that we solved the boundary value problem in Example 1 of Solving the Heat Equation and so there is no reason to resolve it here. Taking a change of letters into account the eigenvalues and eigenfunctions for the boundary value problem here are,
np ln = H
np y jn ( y ) = sin H
n = 1, 2,3,K
Now that we know what the eigenvalues are lets write down the first differential equation with l plugged in.
d 2 h np - h=0 dx 2 H h(L) = 0
Because the coefficient of h ( x ) in the differential equation above is positive we know that a solution to this is,
np x np x h ( x ) = c1 cosh + c2 sinh H H
However, this is not really suited for dealing with the h ( L ) = 0 boundary condition. So, lets
also notice that the following is also a solution.
np ( x - L ) np ( x - L ) h ( x ) = c1 cosh + c2 sinh H H
You should verify this by plugging this into the differential equation and checking that it is in fact a solution. Applying the lone boundary condition to this shifted solution gives,
0 = h ( L ) = c1
The solution to the first differential equation is now,
np ( x - L ) h ( x ) = c2 sinh H
and this is all the farther we can go with this because we only had a single boundary condition. That is not really a problem however because we now have enough information to form the product solution for this partial differential equation. A product solution for this partial differential equation is,
np ( x - L ) n p y un ( x, y ) = Bn sinh sin H H
n = 1, 2,3,K
The Principle of Superposition then tells us that a solution to the partial differential equation is,
np ( x - L ) n p y u4 ( x, y ) = Bn sinh sin H n =1 H
and this solution will satisfy the three homogeneous boundary conditions. To determine the constants all we need to do is apply the final boundary condition.
np ( - L ) n p y u4 ( 0, y ) = g1 ( y ) = Bn sinh sin n =1 H H
Now, in the previous problems weve done this has clearly been a Fourier series of some kind and in fact it still is. The difference here is that the coefficients of the Fourier sine series are now,
np ( - L ) Bn sinh H
instead of just Bn . We might be a little more tempted to use the orthogonality of the sines to derive formulas for the Bn , however we can still reuse the work that weve done previously to get formulas for the coefficients here. Remember that a Fourier sine series is just a series of coefficients (depending on n) times a sine. We still have that here, except the coefficients are a little messier this time that what we saw when we first dealt with Fourier series. So, the coefficients can be found using exactly the same formula from the Fourier sine series section of a function on 0 y H we just need to be careful with the coefficients.
np ( - L ) 2 np y B n sinh = g1 ( y ) sin dy H H H 0
H
n = 1, 2, 3,K n = 1, 2,3,K
Bn =
2 H sinh
np ( - L ) H
np y g1 ( y ) sin dy 0 H
The formulas for the Bn are a little messy this time in comparison to the other problems weve done but they arent really all that messy. Okay, lets do one of the other problems here so we can make a couple of points.
Example 2 Find a solution to the following partial differential equation. 2 u3 2u3 2 u3 = 2 + 2 = 0 x y u3 ( 0, y ) = 0 u3 ( x, 0 ) = 0 u3 ( L, y ) = 0 u3 ( x , H ) = f 2 ( x )
Solution Okay, for the first time weve hit a problem where we havent previous done the separation of variables so lets go through that. Well assume the solution is in the form,
u3 ( x, y ) = h ( x ) j ( y )
Well apply this to the homogeneous boundary conditions first since well need those once we get reach the point of choosing the separation constant. Well let you verify that the boundary conditions become,
h (0) = 0
h(L) = 0
j ( 0) = 0
Next, well plug the product solution into the differential equation.
2 2 h x j y + ( ) ( ) ( ) ( h ( x )j ( y )) = 0 x 2 y 2
j ( y)
d 2h d 2j + h x =0 ( ) dx 2 dy 2 1 d 2h 1 d 2j = h dx 2 j dy 2
Now, at this point we need to choose a separation constant. Weve got two homogeneous boundary conditions on h so lets choose the constant so that the differential equation for h yields a familiar boundary value problem so we dont need to redo any of that work. In this case, unlike the u4 case, well need -l . This is a good problem in that is clearly illustrates that sometimes you need l as a separation constant and at other times you need -l . Not only that but sometimes all it takes is a small change in the boundary conditions it force the change.
So, after adding in the separation constant we get,
1 d 2h 1 d 2j = = -l h dx 2 j dy 2
and two ordinary differential equations that we get from this case (along with their boundary conditions) are,
d 2h + lh = 0 dx 2 h (0) = 0 h ( L) = 0
d 2j - lj = 0 dy 2
j (0) = 0
Now, as we noted above when we were deciding which separation constant to work with weve already solved the first boundary value problem. So, the eigenvalues and eigenfunctions for the first boundary value problem are,
np ln = L
np x hn ( x ) = sin L d 2j np - j =0 dx 2 L j ( 0) = 0
2
n = 1, 2,3,K
The second differential equation is then,
Because the coefficient of the j is positive we know that a solution to this is,
np y np y j ( y ) = c1 cosh + c2 sinh L L
In this case, unlike the previous example, we wont need to use a shifted version of the solution because this will work just fine with the boundary condition weve got for this. So, applying the boundary condition to this gives,
0 = j ( 0 ) = c1
and this solution becomes,
np y j ( y ) = c2 sinh L
The product solution for this case is then,
np y n p x un ( x, y ) = Bn sinh sin L L
The solution to this partial differential equation is then,
n = 1, 2,3,K
np y n p x u3 ( x, y ) = Bn sinh sin L L n =1
Finally, lets apply the nonhomogeneous boundary condition to get the coefficients for this
solution.
np H n p x u3 ( x, H ) = f 2 ( x ) = Bn sinh sin L L n =1
As weve come to expect this is again a Fourier sine (although it wont always be a sine) series and so using previously done work instead of using the orthogonality of the sines to we see that,
np H 2 np x B n sinh = f 2 ( x ) sin dx L L 0 L
L
n = 1, 2,3,K n = 1, 2,3,K
Bn =
2 np x f 2 ( x ) sin dx np H L sinh L 0 L
Okay, weve worked two of the four cases that would need to be solved in order to completely solve (17). As weve seen each case was very similar and yet also had some differences. We saw the use of both separation constants and that sometimes we need to use a shifted solution in order to deal with one of the boundary conditions. Before moving on lets note that we used prescribed temperature boundary conditions here, but we could just have easily used prescribed flux boundary conditions or a mix of the two. No matter what kind of boundary conditions we have they will work the same. As a final example in this section lets take a look at solving Laplaces equation on a disk of radius a and a prescribed temperature on the boundary. Because we are now on a disk it makes sense that we should probably do this problem in polar coordinates and so the first thing we need to so do is write down Laplaces equation in terms of polar coordinates. Laplaces equation in terms of polar coordinates is,
2u =
1 u 1 2 u r + r r r r 2 q 2
Okay, this is a lot more complicated than the Cartesian form of Laplaces equation and it will add in a few complexities to the solution process, but it isnt as bad as it looks. The main problem that weve got here really is that fact that weve got a single boundary condition. Namely,
u ( a, q ) = f (q )
This specifies the temperature on the boundary of the disk. We are clearly going to need three more conditions however since weve got a 2nd derivative in both r and q . When we solved Laplaces equation on a rectangle we used conditions at the end points of the range of each variable and so it makes some sense here that we should probably need the same kind of conditions here as well. The range on our variables here are,
0r a
-p q p
Note that the limits on q are somewhat arbitrary here and are chosen for convenience here. Any set of limits that covers the complete disk will work, however as well see with these limits we will get another familiar boundary value problem arising. The best choice here is often not
known until the separation of variables is done. At that point you can go back and make your choices. Okay, we now need conditions for r = 0 and q = p . First, note that Laplaces equation in terms of polar coordinates is singular at r = 0 (i.e. we get division by zero). However, we know from physical considerations that the temperature must remain finite everywhere in the disk and so lets impose the condition that,
u ( 0, q ) <
This may seem like an odd condition and it definitely doesnt conform to the other boundary conditions that weve seen to this point, but it will work out for us as well see. Now, for boundary conditions for q well do something similar to what we did for the 1-D head equation on a thin ring. The two limits on q are really just different sides of a line in the disk and so lets use the periodic conditions there. In other words,
u ( -p , t ) = u (p , t )
u u ( -p , t ) = (p , t ) r r
With all of this out of the way lets solve Laplaces equation on a disk of radius a.
Example 3 Find a solution to the following partial differential equation. 1 u 1 2 u 2u = =0 r + r r r r 2 q 2 u ( 0, q ) < u ( -p , t ) = u (p , t ) u ( a, q ) = f (q ) u u ( -p , t ) = (p , t ) r r
Solution In this case well assume that the solution will be in the form,
u (q , r ) = j (q ) G ( r )
Plugging this into the periodic boundary conditions gives,
j ( -p ) = j (p )
G ( 0) <
dj dj ( -p ) = (p ) dq dq
Now lets plug the product solution into the partial differential equation.
2 1 1 r j q G r + ( ( ) ( ) ) r 2 q 2 (j (q ) G ( r ) ) = 0 r r r
j (q )
1 d dG 1 d 2j + r G r ( ) 2 2 =0 r dr dr r dq
This is definitely more of a mess that weve seen to this point when it comes to separating variables. In this case simply dividing by the product solution, while still necessary, will not be
sufficient to separate the variables. We are also going to have to multiply by r 2 to completely separate variables. So, doing all that, moving each term to one side of the equal sign and introduction a separation constant gives,
r d dG 1 d 2j r = =l G dr dr j dq 2
We used l as the separation constant this time to get the differential equation for j to match up with one weve already done. The ordinary differential equations we get are then,
d dG r - lG = 0 dr dr
d 2j + lj = 0 dq 2
G (0) <
j ( -p ) = j (p )
dj dj ( -p ) = (p ) dq dq
Now, we solved the boundary value problem above in Example 3 of the Eigenvalues and Eigenfunctions section of the previous chapter and so there is no reason to redo it here. The eigenvalues and eigenfunctions for this problem are,
l n = n2 l n = n2
jn (q ) = sin ( n q )
jn (q ) = cos ( n q )
n = 1, 2,3,K
n = 0,1, 2, 3, K
Plugging this into the first ordinary differential equation and using the product rule on the derivative we get,
d dG 2 r -n G =0 dr dr
d 2 G dG 2 rr 2 + -n G =0 dr dr d 2G dG r2 2 + r - n 2G = 0 dr dr
This is an Euler differential equation and so we know that solutions will be in the form G ( r ) = r p provided p is a root of,
p ( p - 1) + p - n 2 = 0 p2 - n2 = 0 p = n n = 0,1, 2,3,K
So, because the n = 0 case will yield a double root, versus two real distinct roots if n 0 we have two cases here. They are,
G ( r ) = c1 + c2 ln r G ( r ) = c1r + c2 r
n
n=0 n = 1, 2,3,K
-n
2007 Paul Dawkins
489
Now we need to recall the condition that G ( 0 ) < . Each of the solutions above will have
G ( r ) as r 0 Therefore in order to meet this boundary condition we must have c2 = c2 = 0 .

Therefore, the solution reduces to,
G ( r ) = c1r n
n = 0,1, 2, 3, K
and notice that with the second term gone we can combine the two solutions into a single solution. So, we have two product solutions for this problem. They are,
un (q , r ) = An r n cos ( n q ) un (q , r ) = Bn r n sin ( n q )
n = 0,1, 2, 3, K n = 1, 2,3,K
Our solution is then the sum of all these solution or,
u (q , r ) = An r n cos ( n q ) + An r n sin ( n q )
n= 0 n =1
Applying our final boundary condition to this gives,
u ( a, q ) = f (q ) = An a n cos ( n q ) + An a n sin ( n q )
n =0 n =1
This is a full Fourier series for f (q ) on the interval -p q p , i.e. L = p . Also note that once again the coefficients of the Fourier series are a little messier than normal, but not quite as messy as when we were working on a rectangle above. We could once again use the orthogonality of the sines and cosines to derive formulas for the An and Bn or we could just use the formulas from the Fourier series section with L = p to get,
1 p f (q ) dq 2p -p 1 p An a n = f (q ) cos ( nq ) dq p -p 1 p Bn a n = f (q ) sin ( nq ) dq p -p A0 =
Upon solving for the coefficients we get,
n = 1, 2,3,K n = 1, 2,3,K
1 p f (q ) dq 2p -p 1 p An = n f (q ) cos ( nq ) dq p a -p 1 p Bn = n f (q ) sin ( nq ) dq p a -p A0 =
n = 1, 2,3,K n = 1, 2,3,K
Prior to this example most of the separation of variable problems tended to look very similar and it is easy to fall in to the trap of expecting everything to look like what wed seen earlier. With this example we can see that the problems can definitely be different on occasion so dont get too locked into expecting them to always work in exactly the same way. Before we leave this section lets briefly talk about what youd need to do on a partial disk. The periodic boundary conditions above were only there because we had a whole disk. What if we only had a disk between say a q b . When weve got a partial disk we now have two new boundaries that we not present in the whole disk and the periodic boundary conditions will no longer make sense. The periodic boundary conditions are only used when we have the two boundaries in contact with each other and that clearly wont be the case with a partial disk. So, if we stick with prescribed temperature boundary conditions we would then have the following conditions
u ( 0,q ) <
u ( a , q ) = f (q ) u ( r , a ) = g1 ( r ) u ( r, b ) = g2 ( r )
a q b
0r a 0r a
Also note that in order to use separation of variables on these conditions wed need to have g1 ( r ) = g 2 ( r ) = 0 to make sure they are homogeneous. As a final note we could just have easily used flux boundary conditions for the last two if wed wanted to. The boundary value problem would be different, but outside of that the problem would work in the same manner. We could also use a flux condition on the r = a boundary but we havent really talked yet about how to apply that kind of condition to our solution. Recall that this is the condition that we apply to our solution to determine the coefficients. Its not difficult to use we just havent talked about this kind of condition yet. Well be doing that in the next section.
2007 Paul Dawkins
491
Vibrating String
This will be the final partial differential equation that well be solving in this chapter. In this section well be solving the 1-D wave equation to determine the displacement of a vibrating string. There really isnt much in the way of introduction to do here so lets just jump straight into the example.
Example 1 Find a solution to the following partial differential equation. 2 2u 2 u c = t 2 x 2 u u ( x, 0 ) = f ( x ) ( x, 0 ) = g ( x ) t u ( 0, t ) = 0 u ( L, t ) = 0

Solution One of the main differences here that were going to have to deal with is the fact that weve now got two initial conditions. That is not something weve seen to this point, but will not be all that difficult to deal with when the time rolls around. Weve already done the separation of variables for this problem, but lets go ahead and redo it here so we can say weve got another problem almost completely worked out. So, lets start off with the product solution.
u ( x, t ) = j ( x ) h ( t )
Plugging this into the two boundary conditions gives,
j ( 0) = 0
j ( L) = 0
Plugging the product solution into the differential equation, separating and introducing a separation constant gives,
2 2 2 = j x h t c ( ) ( ) ( ) (j ( x ) h ( t ) ) t 2 x 2 d 2h d 2j j ( x ) 2 = c 2h ( t ) 2 dt dx 2 2 1 d h 1d j = = -l c 2 h dt 2 j dx 2
We moved the c 2 to the left side for convenience and chose -l for the separation constant so the differential equation for j would match a known (and solved) case. The two ordinary differential equations we get from separation of variables are then,
d 2h 2 + c lh = 0 dt 2
d 2j + lj dx 2 j ( 0) = 0
j ( L) = 0
We solved the boundary value problem above in Example 1 of the Solving the Heat Equation section of this chapter and so the eigenvalues and eigenfunctions for this problem are,
np ln = L
np x jn ( x ) = sin L d 2 h np c + h=0 dt 2 L
2
n = 1, 2,3,K
The first ordinary differential equation is now,
and because the coefficient of the h is clearly positive the solution to this is,
np c t np c t h ( t ) = c1 cos + c2 sin L L
Because there is no reason to think that either of the coefficients above are zero we then get two product solutions,
np c t n p x un ( x, t ) = An cos sin L L np c t n p x un ( x, t ) = Bn sin sin L L

n = 1, 2,3,K
np c t n p x np c t n p x u ( x, t ) = An cos sin + Bn sin sin L L L L n =1
Now, in order to apply the second initial condition well need to differentiate this with respect to t so,
u np c np c t n p x np c np c t n p x = An sin Bn cos sin + sin t n=1 L L L L L L

If we now apply the initial conditions we get,
np x n p x np x u ( x, 0 ) = f ( x ) = An cos ( 0 ) sin + B sin 0 sin ( ) n = An sin L L n =1 L n =1 u np c np x Bn sin ( x, 0 ) = g ( x ) = t L n =1 L
Both of these are Fourier sine series. The first is for f ( x ) on 0 x L while the second is for
g ( x ) on 0 x L with a slightly messy coefficient. As in the last few sections were faced
with the choice of either using the orthogonality of the sines to derive formulas for An and Bn or we could reuse formula from previous work. Its easier to reuse formulas so using the formulas form the Fourier sine series section we get,
2 np x An = f ( x ) sin dx L 0 L np c 2 np x B n = g ( x ) sin dx L L 0 L
Upon solving the second one we get,
L
n = 1, 2,3,K n = 1, 2,3,K
2 np x An = f ( x ) sin dx L 0 L 2 np x Bn = g ( x ) sin dx np c 0 L
L
n = 1, 2,3,K n = 1, 2,3, K
So, there is the solution to the 1-D wave equation and with that weve solved the final partial differential equation in this chapter.
2007 Paul Dawkins
494
Summary of Separation of Variables

Throughout this chapter weve been talking about and solving partial differential equations using the method of separation of variables. However, the one thing that weve not really done is completely work an example from start to finish showing each and every step. Each partial differential equation that we solved made use somewhere of the fact that wed done at least part of the problem in another section and so it makes some sense to have a quick summary of the method here. Also note that each of the partial differential equations only involved two variables. The method can often be extended out to more than two variables, but the work in those problems can be quite involved and so we didnt cover any of that here. So with all of that out of the way here is a quick summary of the method of separation of variables for partial differential equations in two variables. 1. Verify that the partial differential equation is linear and homogeneous. 2. Verify that the boundary conditions are in proper form. Note that this will often depend on what is in the problem. So, a. If you have initial conditions verify that all the boundary conditions are linear and homogeneous. b. If there are no initial conditions (such as Laplaces equation) the verify that all but one of the boundary conditions are linear and homogeneous. c. In some cases (such as we saw with Laplaces equation on a disk) a boundary condition will take the form of requiring that the solution stay finite and in these cases we just need to make sure the boundary condition is met. 3. Assume that solutions will be a product of two functions each a function in only one of the variables in the problem. This is called a product solution. 4. Plug the product solution into the partial differential equation, separate variables and introduce a separation constant. This will produce two ordinary differential equations. 5. Plug the product solution into the homogeneous boundary conditions. Note that often it will be better to do this prior to doing the differential equation so we can use these to help us chose the separation constant. 6. One of the ordinary differential equations will be a boundary value problem. Solve this to determine the eigenvalues and eigenfunctions for the problem. Note that this is often very difficult to do and in some cases it will be impossible to completely find all eigenvalues and eigenfunctions for the problem. These cases can be dealt with to get at least an approximation of the solution, but that is beyond the scope of this quick introduction. 7. Solve the second ordinary differential equation using any remaining homogeneous boundary conditions to simplify the solution if possible.
8. Use the Principle of Superposition and the product solutions to write down a solution to the partial differential equation that will satisfy the partial differential equation and homogeneous boundary conditions. 9. Apply the remaining conditions (these may be initial condition(s) or a single nonhomogeneous boundary condition) and use the orthogonality of the eigenfunctions to find the coefficients. Note that in all of our examples the eigenfunctions were sines and/or cosines however they wont always be sines and cosines. If the boundary value problem is sufficiently nice (and thats beyond the scope of this quick introduction to the method) we can always guarantee that the eigenfunctions will be orthogonal regardless of what they are.
2007 Paul Dawkins
496

Differential Equations

Uploaded by

Copyright:

Available Formats

Differential Equations

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Differential Equations

Uploaded by

Copyright:

Available Formats

DIFFERENTIAL EQUATIONS

First Order Differential Equations .........................................................................................20

Second Order Differential Equations ................................................................................... 102

Laplace Transforms .............................................................................................................. 181

Systems of Differential Equations ......................................................................................... 241

Series Solutions to Differential Equations ............................................................................ 318

Higher Order Differential Equations .................................................................................... 346

Boundary Value Problems & Fourier Series ........................................................................ 374

Partial Differential Equations ............................................................................................... 440

2007 Paul Dawkins

2007 Paul Dawkins

2007 Paul Dawkins

2007 Paul Dawkins

2007 Paul Dawkins

2007 Paul Dawkins

(5) (6) (7) (8) (9) (10)

2007 Paul Dawkins

Example 1 Show that y ( x ) = x

is a solution to 4 x 2 y + 12 xy + 3 y = 0 for x > 0 .

Solution Well need the first and second derivative to do this.

Plug these as well as the function into the differential equation.

2007 Paul Dawkins

and so this solution also meets the initial conditions of y ( 4 ) =

Example 3 The following is an IVP. 4 x 2 y + 12 xy + 3 y = 0

Example 4 Heres another IVP. 2t y + 4 y = 3

3 c + is the general solution to 4 t2 2t y + 4 y = 3

Example 6 What is the actual solution to the following IVP? 2t y + 4 y = 3 y (1) = -4

Example 7 y 2 = t 2 - 3 is the actual implicit solution to y =

Example 8 Find an actual explicit solution to y =

2007 Paul Dawkins

To simplify the differential equation lets divide out the mass, m.

2007 Paul Dawkins

2007 Paul Dawkins

2007 Paul Dawkins

2007 Paul Dawkins

2007 Paul Dawkins

2007 Paul Dawkins

y (0) < 1 1 y ( 0) < 2 y (0) = 2 y (0) > 2

started. For our case the family of isoclines is.

2007 Paul Dawkins

2007 Paul Dawkins

2007 Paul Dawkins

First Order Differential Equations

2007 Paul Dawkins

Linear Differential Equations

m ( t ) is, it will satisfy the following. m (t ) p (t ) = m (t )

2007 Paul Dawkins

2007 Paul Dawkins

Divide both sides by m ( t ) ,

2007 Paul Dawkins

2007 Paul Dawkins

dv + 0.196e 0.196t v = 9.8e 0.196t dt e 0.196t v = 9.8e 0.196t

( e0.196t v ) dt = 9.8e 0.196t dt e0.196t v + k = 50e 0.196t + c

e0.196t v = 50e 0.196t + c - k

2007 Paul Dawkins

Example 2 Solve the following IVP. dv = 9.8 - 0.196v dt

sin ( x ) 1 y = 2cos 2 ( x ) sin ( x ) cos ( x ) cos ( x )

y + tan ( x ) y = 2cos 2 ( x ) sin ( x ) - sec ( x )

= eln sec x = sec x

sec ( x ) y + sec ( x ) tan ( x ) y = 2sec ( x ) cos 2 ( x ) sin ( x ) - sec 2 ( x )