You are on page 1of 365

The Ordinary Differential Equations

Project

The Ordinary Differential Equations
Project

Thomas W. Judson
Stephen F. Austin State University

DRAFT April 26, 2017 DRAFT

© 2013-–2017 Thomas W. Judson
Permission is granted to copy, distribute and/or modify this document under the terms of
the GNU Free Documentation License, Version 1.3 or any later version published by the Free
Software Foundation; with no Invariant Sections, no Front-Cover Texts, and no Back-Cover
Texts. A copy of the license is included in the section entitled “GNU Free Documentation
License”.

Preface

This text is intended for a one-semester undergraduate course in ordinary differential equa-
tions.
There are additional exercises or computer projects at the ends of many of the chapters.
The computer projects often require a basic knowledge of programming. All of these exer-
cises and projects are more substantial in nature and allow the exploration of new results
and theory.
Sage (sagemath.org) is a free, open source, software system for advanced mathematics,
which is ideal for assisting with a study of ordinary differential equations. Sage can be used
either on your own computer, a local server, or on SageMathCloud (https://cloud.sagemath.com).
The Sage code has been tested for accuracy with the most recent version available at this
time: Sage Version 7.3 (released 2016–08–5).
Thomas W. Judson
Stephen F. Austin State University
Nacogdoches, Texas 2016

v

Acknowledgements

This book was written in MathBook XML.

vi

Contents

Preface v

Acknowledgements vi

1 A First Look at Differential Equations 1
1.1 Modeling with Differential Equations . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Separable Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3 Geometric and Quantitative Analysis . . . . . . . . . . . . . . . . . . . . . . 28
1.4 Analyzing Equations Numerically . . . . . . . . . . . . . . . . . . . . . . . . 38
1.5 First-Order Linear Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 50
1.6 Existence and Uniqueness of Solutions . . . . . . . . . . . . . . . . . . . . . 63
1.7 Bifurcations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

2 Systems of Differential Equations 81
2.1 Modeling with Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
2.2 The Geometry of Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
2.3 Numerical Techniques for Systems . . . . . . . . . . . . . . . . . . . . . . . 103
2.4 Solving Systems Analytically . . . . . . . . . . . . . . . . . . . . . . . . . . 112

3 Linear Systems 119
3.1 Linear Algebra in a Nutshell . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
3.2 Planar Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
3.3 Phase Plane Analysis of Linear Systems . . . . . . . . . . . . . . . . . . . . 133
3.4 Complex Eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
3.5 Repeated Eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
3.6 Changing Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
3.7 The Trace-Determinant Plane . . . . . . . . . . . . . . . . . . . . . . . . . . 163
3.8 Linear Systems in Higher Dimensions . . . . . . . . . . . . . . . . . . . . . . 172
3.9 The Matrix Exponential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183

4 Second-Order Linear Equations 194
4.1 Homogeneous Linear Equations . . . . . . . . . . . . . . . . . . . . . . . . . 194
4.2 Forcing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
4.3 Sinusoidal Forcing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
4.4 Forcing and Resonance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220

vii

. . . . . . . . . . . . . 264 5. . . . . . . . 280 6 The Laplace Transform 295 6. . . . . . 241 5. . . . . .3 Delta Functions and Forcing . . . . . . 300 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Solving Initial Value Problems . . . . . . . . . . . . . . . . . 305 6. . . . . . . . . . . . . . . . 309 A Hints and Solutions to Selected Exercises 313 B GNU Free Documentation License 346 Readings and References 353 Index 355 . . . .3 More Nonlinear Mechanics . . . 295 6.1 The Laplace Transform . . . .1 Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4 The Hopf Bifurcation . . . . . . . . . . . . . . . . . .viii CONTENTS 5 Nonlinear Systems 241 5. . . . . . . . . . . . .2 Hamiltonian Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253 5. . . .4 Convolution . . . . . . . . . . . . . .

Suppose that we wish to study how a population P (t) grows with time t. dt ∆t→0 ∆t the rate of change of the population is proportional to the size of the population. the change in the population ∆P during a small time interval ∆t will be ∆P ≈ kbirth P (t)∆t − kdeath P (t)∆t. For a particular situation that we might wish to investigate. our first task is to write an equation (or equations) that best describes the phenomenon. Equivalently. the theory of the subject has broad and important implications. If we also assume that the population has a constant death rate. 1 A First Look at Differential Equations 1.1) dt The equation dP = kP dt 1 .1 Exponential Growth We begin our study of ordinary differential equations by modeling some real world phenom- ena.1. (1. ∆t where k = kbirth − kdeath . The subject of differential equations is one of the most interesting and useful areas of mathematics. 1. We might make the assumption that a constant fraction of population is having offspring at any particular time. we can write ∆P ≈ kP (t). or dP = kP. We can describe many interesting natural phenomena that involve change using differential equations. where kbirth is the fraction of the population having offspring during the interval and kdeath is the fraction of the population that dies during the interval. In addition. Since the derivative of P is dP ∆P = lim .1 Modeling with Differential Equations Calculus tells us that the derivative of a function measures how the function changes. An equation relating a function to one or more of its derivatives is called a differential equation.

0296 and the solution to our initial value problem is P (t) = 1000e0. we say that the expression x(t) = Cekt is a general solution of the equation x′ = kx.org/indicator/SP. where time is measured in hours. P (0) = P0 is an example of an initial value problem. Since the solution to equation (1. whose initial population is 1000 at t = 0. then 1030 = P (1) = 1000ek . we obtain d P (t) = kCekt = kP (t). Of course.2 CHAPTER 1. The differential equation P ′ (t) = kP (t). A FIRST LOOK AT DIFFERENTIAL EQUATIONS is one of the simplest differential equations that we will consider. if the population at the time t = 0 is P (0) = P0 . The equation tells us that a population grows in proportion to its current size. Then 1000 = P (0) = Ce0 = C. after one hour.1) is P (t) = Cekt . if we differentiate P (t). Consequently.0296t . Japan has expe- rienced negative growth in recent years. If t is small. our colony of bacteria could very well exceed the mass of the earth. For example. it is important to realize that this is only a model. For example. we can also determine the value of C. To summarize. k = ln 1. If the population grows at three percent per hour. Indeed. and x(t) = x0 ekt is a particular solution to the differential equation.GROW. and we say that P (0) = P0 is an initial condition. and we say that the population grows exponentially.worldbank. The growth rate of a population need not be positive. However. As an example. then P0 = P (0) = Cek·0 = C or P (t) = P0 ekt .POP. if we know the value of P (t). and our solution becomes P (t) = 1000ekt . if we let t be very large. which we will call integral curves or solution curves (Figure 1.1 The equation dP /dt = kP can also be used to model such phenomena such as radioactive decay and compound interest—topics which we will explore later. say when t = 0. our model might be reasonably accurate. dt In addition. It is not too difficult to see that P (t) = Cekt is a solution to this equation. . 1 See http://data. The general solution to our equation x(t) = Cekt graphs as an infinite family of curves. where C is an arbitrary constant.03 ≈ 0.1). suppose that P (t) is a population of a colony of bacteria at time t.

• If the population of trout is small and there are abundant resources.1. let us consider the population of fish in a children’s trout pond. In other words. dP ≈ kP. but the growth rate will decline as the population increases and the availability of resources declines. the rate of growth will be approximately exponential. a bacteria culture in a petri dish would grow unbounded and soon be much larger than the size of the laboratory. dt 2 The logistic model was first used by the Belgian mathematician and physician Pierre François Verhulst in 1836 to predict the populations of Belgium and France. We say that N is the carrying capacity for the population. and spawning habitat. let us try to adjust our model of exponential growth to account for the limited resources of the pond.1. otherwise.2 Logistic Growth Not all populations grow exponentially.1: Integral curves 1.2 To see how the logistic model works. We can use the logistic equation to model population growth in a resource limited environment. dP <0 dt for P > N . MODELING WITH DIFFERENTIAL EQUATIONS 3 x(t) 300 200 x(t) = Ce kt 100 -2 -1 1 2 t -100 -200 -300 Figure 1. A small population of fish might grow exponentially if the pond is large and food is abundant. Our assumptions suggest that we might try an equation of the form dP = kf (P )P. then any popu- lation larger than N will decrease. The number of trout will be limited by the available resources. To see what happens if there are limiting factors to population growth. . We will make the following assumptions. food supply.1. dt • If N is the maximum population of trout that the pond can support.

3 If we differentiate P .4 CHAPTER 1. we have not allowed for the possibility that children will be catching fish in our pond.8109t +1 The graph of our solution certainly fits the situation that we are modeling (Figure 1. 1000 and certainly P (0) = 100. 4 Consequently.7. N Thus. ( ) dP d 1000 = dt dt 9e−kt + 1 9e−kt = 1000k −kt (9e + 1)2 9e−kt 1000 = k −kt · −kt (9e + 1) 9e +1 (9e −kt + 1) − 1 1000 =k −kt · −kt (9e + 1) 9e +1 ( ) 1000 1000 =k 1− −kt −kt 1000(9e + 1) 9e +1 ( ) P =k 1− P. For the time being.2.2. 3 We will learn how to solve initial-value problems such as the one described here in Section 1. In addition. the logistic population model is given by the differential equation ( ) dP P =k 1− P. 9e−0. we must solve the initial value problem ( ) dP P =k 1− P dt 1000 P (0) = 100.3). and the initial population is 100 fish. but negative if the population is greater than N . we will obtain the righthand side of the differential equation. The simplest function satisfying these conditions is ( ) P f (P ) = 1 − . we will be satisfied with being able to verify the fact that we have a solution. A FIRST LOOK AT DIFFERENTIAL EQUATIONS where f (P ) is a function of P that is close to 1 if the population is small. Of course.8109. the solution to our intial-value problem is 1000 P (t) = . if we know that the population is 200 fish after one year. It is easy to verify that 1000 P (t) = 9e−kt +1 is the solution to our initial value problem. dt N Example 1. In order to determine the number of fish in the lake at any time t. Suppose we have a pond will support 1000 fish. then 1000 200 = P (1) = −k 9e + 1 and we can determine that ( ) 9 k = ln ≈ 0. . We will examine how fishing affects our model in Section 1.

4: A spring-mass system We can construct a differential equation that models our oscillating mass. F (x). If x < 0. MODELING WITH DIFFERENTIAL EQUATIONS 5 P(t) 1000 800 600 400 200 5 10 15 t Figure 1. we must consider the restorative force on the spring. frictionless surface and that this mass is attached to one end of a spring with the other end of the spring attached to a wall. First. wall mass Position x(t) at time t wall mass Position at rest Figure 1. Suppose that we have a mass lying on a flat.3 A Spring-Mass Model Sometimes it is necessary to consider the second derivative when constructing a mathemat- ical model. If x = 0. 2 2 .3: Logistic growth 1.1. then the spring is in a state of equilibrium (Figure 1. we can expand F to obtain 1 F (x) = F (0) + F ′ (0)x + F ′′ (0)x2 + · · · 2 1 ′′ = −kx + F (0)x + · · · .4).1. If we pull on the mass. We denote the spring displacement by x. Using Taylor’s Theorem from calculus. We will make the assumption that this force depends on the displacement of the spring. then the mass will oscillate back and forth across the table. the spring is compressed. then the spring is stretched.1. If x > 0.

we have a second order differential equation. For example. the force on the mass must be d2 x F = ma = m = mx′′ . x(t) 1. Thus. which describes our oscillating mass. dt2 Setting the two forces equal. then we have the following initial value problem. Think of a dashpot as the small . we might guess that a general solution of our differential equation has the form x(t) = A cos t + B sin t. and it is reasonably accurate if the displacement of the spring. The graph of the position of the mass as a function of time is given in Figure 1. If the displacement is not too large.5 5 10 15 20 25 30 t -0. we might add a dashpot. to our system. A FIRST LOOK AT DIFFERENTIAL EQUATIONS where F ′ (0) = −k and F (0) = 0. x′′ + x = 0 x(0) = 0 x′ (0) = 1. F = −kx. This equation is known as is Hooke’s Law.5 Figure 1.5. x. By Newton’s second law of motion. The spring-mass system is an example of a harmonic oscillator. and using our initial conditions.6 CHAPTER 1. we can consider the restorative force on the spring to be proportional to displacement of the spring from its equilibrium length.5 1 0. Noting that x′ (t) = −A sin t + B cos t. mx′′ = −kx. We can test this law experimentally. and we can ignore higher ordered terms. Since x′′ (t) = −x(t) for both the sine and cosine functions. a mechanical device that resists motion. If the initial velocity of the spring is one unit per second and the initial position is at the equilibrium point.5 -1 -1. is not too large. we can determine that A = 0 and B = 1 or x(t) = sin t. Suppose that we have a spring-mass system where m = 1 and k = 1. then xn will be small for n ≥ 2.5: A undamped spring-mass system Now let us add a damping force to our system.

and k are all positive constants. If the initial velocity of our mass is one unit per second and the initial position is zero. A = 1 and B = −1.1. b. but let us assume that the solution is of the form x(t) = ert for now. then we would have an oscillating mass governed by the equation x′′ + 3x′ + 2x = 0. We will learn how to solve equations of the form ax′′ + bx′ + cx = 0 in Chapter 4. we might guess that x(t) = Ae−t + Be−2t is a general solution to our equation. Thus. Notice that the additional damping negates any oscillation in the system. or mx′′ + bx′ + kx = 0. our initial conditions give us the following system of linear equations. we have will have an additional force. In this case. it must be the case that r = −2 or r = −1 if x(t) = ert is to be a solution to our equation. The simplest assumption would be to take the damping force of the dashpot to be proportional to the velocity of the mass. then we have the initial value problem x′′ + 3x′ + 2x = 0 x(0) = 0 x′ (0) = 1. b = 3. Since ert is never zero. MODELING WITH DIFFERENTIAL EQUATIONS 7 cylinder that keeps your screen door from slamming shut. where b > 0. Then x′′ + 3x′ + 2x = r2 ert + 3rert + 2ert = ert (r2 + 3r + 2) = ert (r + 2)(r + 1) = 0. A+B =0 −A − 2B = 1.1. F = −bx′ acting on our mass. Thus. we say that the system is over-damped (Figure 1. Using the fact that x′ (t) = −Ae−t − 2Be−2t . . Our new equation for the spring-mass system is mx′′ = −bx′ − kx. If we let m = 1. where m. x′ (t). and our spring-mass system is modeled by the function x(t) = e−t − e−2t . and k = 2.6). Thus.

3 0. if we have a very strong spring and only add a small amount of damping to our spring-mass system. For example. say where one species preys on the other.05 2 4 6 8 10 t -0. which owns and operates 4 The predator-prey model was discovered independently by Lotka (1925) and Volterra (1926).4 For example. In other words.1. the mass would continue to oscillate. We have good historical data for the populations of the lynx and snowshoe hare from the Hudson Bay Company.8 CHAPTER 1.1 0.7: An under-damped spring-mass system We will explore harmonic oscillators and second-order differential equations more fully in Chapter 4.1 Figure 1. We might use a system of differential equations to model two interacting species. This large Canadian retail company. we can model how the population of Canadian lynx (lynx canadienis) interacts with a the population of snowshoe hare (lepus americanis) (see https://www.05 -0. 1.1 2 4 6 8 10 t -0.7).1 Figure 1. but the oscillations would become progressively smaller.com/watch?v=ZWucOrSOdCs).2 0.6: An over-damped spring-mass system Of course. A FIRST LOOK AT DIFFERENTIAL EQUATIONS x(t) 0. our spring-mass system might be described by the initial value problem x′′ + 2x′ + 50x = 0 x(0) = 0 x′ (0) = 1. . x(t) 0.youtube.4 A Predator-Prey System Some situations require more than one differential equation to model a particular situation.15 0. our system would be under-damped. It is easy to verify that 1 x(t) = e−t sin 7t 7 is a solution to the initial value problem (Figure 1.

including Galeria Kaufhof. The company noticed that the number of pelts varied from year to year and that the number of lynx pelts reached a peak about every ten years [11]. dt We now have a system of differential equations that describe how the two populations interact. Lord & Taylor. MODELING WITH DIFFERENTIAL EQUATIONS 9 a large number of retail stores in North America and Europe. dH = aH.8 to the system dH = 0. we will give a graphical solution in Figure 1. Consequently. • If no lynx are present. dL = −cL.1. We will denote the population of hares by H(t) and the population of lynx by L(t). That is. The Hudson Bay Company kept accurate records on the number of lynx pelts that were bought from trappers from 1821 to 1940.1. we will assume that the hares reproduce at a rate proportional to their population and are not affected by overcrowding. We will make the following assumptions for our predator-prey model. we can argue that the rate at which the hares are consumed by the lynx is proportional to the rate at which the hares and lynx interact. The primary prey for the Canadian lynx is the snowshoe hare. Home Outfitters. the lynx population will decline at a rate proportional to itself. dt We are thinking of HL as the number of possible interactions between the lynx and the hare populations.4H − 0. where t is the time measured in years. dt dL = −0. The rate at which lynx are born is proportional to the number of hares that are eaten. • If there is no food. was originally founded in 1670 as a fur trading company. the growth rate of the lynx population can be described by dL = −cL + dHL. dt .01HL. Hudson’s Bay. dH = aH − bHL. the hare population will grow exponentially. however.005HL. and this is proportional to the rate at which the hares and lynx interact. dt • Since the lynx prey on the hares. Thus. dt We will learn how to analyze and solve systems of differential equations in subsequent chapters. dt • The lynx receive benefit from the hare population. the equation that predicts the rate of change of the hare population becomes dH = aH − bHL. dt dL = −cL + dHL. and Saks Fifth Avenue.3L + 0. The ten year cycle for lynx can be best understood using a system of differential equations.

dt The first term. . L. L(t) 80 H(t) L(t) 70 60 50 40 30 t 0 20 40 60 80 100 Figure 1. As the prey population declines. where V0 is the initial viral population. The body’s immune system fights the HIV-1 virus with white blood cells. it can be years before an HIV-positive patient exhibits the full symptoms of AIDS. The term −cV is the death rate for the virions. dt and V (t) = V0 e−ct . is especially important since it helps other cells fight the virus.10 CHAPTER 1. Once infected with the HIV-1 virus. dV = −cV.5 H(t). P is some function of t that determines the rate at which new viral particles are created. After an individual is infected with the HIV-1 virus. the amount of the virus in the bloodstream rises dramat- ically and the person will often suffer from flu-like symptoms.5 Modeling the HIV-1 Virus The interaction of the HIV-1 virus with the body’s immune system can be modeled by a system of differential equations similar to a predator-prey system.8: The predator-prey relationship between the lynx and the snowshoe hare 1. the prey population has a chance to recover that the cycle begins again. Once the predator population is smaller. a specific type of white blood cell. dV = P − cV. If someone discovers a drug that blocks the creation of new HIV-1 virions. If a individual has such antibodies. Let V = V (t) be the population of the HIV-1 virus at time t. then P would be zero and the virions would clear the body at the following rate. The CD4-positive T-helper cells. Tests have been developed to determine the presence of HIV-1 antibodies. A FIRST LOOK AT DIFFERENTIAL EQUATIONS Notice that the predator population. the predator population also declines. We can develop a system of differential equations to better understand the dynamics of the HIV-1 virus [20]. 5 An excellent account of the actual lynx and snowshoe hare data and model can be found in [5]. begins to grow and reaches a peak after the prey population. these symptoms will disappear after a period of weeks or months as the body begins to manufacture anti- bodies against the virus. H reaches its peak. the HIV-1 virus can destroy CD4-positive T-helper cells.1. However. then they are said to be HIV-1 positive. However. We will assume that the virus concentration is governed by the following differential equation.

dt One class of drugs that HIV infected patients receive are reverse transcriptase (RT) inhibitors. MODELING WITH DIFFERENTIAL EQUATIONS 11 Now let us consider a model for the concentration of T = T (t) of (uninfected) CD4- positive T-helper cells. Thus. If one could find the perfect RT inhibitor. An RNA virus cannot reproduce on its own and must use the DNA from a host cell. where p is the maximum proliferation rate and Tmax is the T cell population density where proliferation ceases. the HIV-1 virus is an RNA virus. 1 − η. This way the virus can use the T cell’s DNA to replicate itself using a process called reverse transcription. dt Tmax The constant s represents the rate at which T cells are created from sources in the body. where a DNA copy of the virus’s RNA is made.1. We can accomplish this by adding an effectiveness factor. our complete model becomes ( ) dT T = s + pT 1 − − dT T − kT V dt Tmax dT ∗ = kT V − δT ∗ dt dV = N δT ∗ − cV. To do this. Our system now becomes ( ) dT T = s + pT 1 − − dT T − k(1 − η)T V dt Tmax . the virus attaches itself to a CD4-positive T-helper cell and injects its RNA into the cell. so we will need to modify the system to account for the effectiveness of the RT inhibitor. The term kT V tells us the rate at which the HIV-1 virus infects T cells. dt where δ is the rate of loss of the virus producing T cells and N is the number of virions produced per infected T cell during its lifetime. dt Unfortunately. and the T cell eventually bursts releasing the virions into the body. dT is the death rate of the T cells. dT ∗ = kT V − δT ∗ dt dV = N δT ∗ − cV. no one has discovered a perfect RT inhibitor. Finally.1. such as the thymus. then k = 0 and our system becomes ( ) dT T = s + pT 1 − − dT T dt Tmax dT ∗ = −δT ∗ dt dV = N δT ∗ − cV. New CD4-positive T-helper cells can also be created from the proliferation of existing CD4-positive T-helper cells. Like the influenza virus. RT inhibitors block the action of reverse transcription and prevent the virus from duplicating. New virus particles are created. If we let T ∗ be the concentration of infected T cells. ( ) dT T = s + pT 1 − − dT T. and the second term in the equation represents the logistic growth of the T cells. This is the same idea as modeling how predators interact with prey in a predator-prey model. to the kV T term. we can model this process with the following system of equations.

then the RT inhibitor is completely ineffective. legend_color = ' red ' ) p We can use Sage to solve differential equations. t) == k*P # differential equation # solve specifying initial conditions # and independent variable solution solution = desolve (de . For example. You can even access Sage from your smart phone. dt If η = 1. color = ' red ' . You can even change the preloaded commands in the cell if you wish. -2. color = ' blue ' . legend_label = ' $df / dx$ ' . (x . we can plot the function f (x) = x2 cos x as well as its derivative on the same graph. x) p = plot (y . and Matlab each have their advantages and disadvantages.1. Researchers can use data to estimate the parameters and see exactly what types of solutions are possible. For our purposes.␣t ' ) # declare variables k and t P = function ( ' P ' )(t) # declare P to be a function of t de = diff (P . Sage can be run on an individual computer or over the Internet on a server. Software packages such Maple. P . (x .t = var ( ' k . a readily available open source computer algebra system. legend_label = ' $f$ ' . 2) . -2. We can use the following commands. then the RT inhibitor is completely effective. Suppose that we wish to solve the initial value problem dP = kP dt P (0) = 1000.12 CHAPTER 1. evaluate the cell. x) -x ^2* sin (x) + 2* x* cos (x) We can use Sage to plot functions. k . as our choice of software. On the other hand. legend_color = ' blue ' ) p += plot ( yprime . We will use Sage. For example. Simply. Sage cells are embedded into the textbook.6 SageOpen Source Software Technology can prove very useful when studying differential equations. x = var ( ' x ' ) y = x ^2* cos (x) diff (y . 1. We now have a model for how the HIV-1 virus interacts we the immune system. 1000]) solution 1000* e ^( k*t) . A FIRST LOOK AT DIFFERENTIAL EQUATIONS dT ∗ = k(1 − η)T V − δT ∗ dt dV = N δT ∗ − cV. 2) . ics =[0 . Mathematica. if η = 0. x = var ( ' x ' ) y = x ^2* cos (x) yprime = diff (y . let us evaluate the derivative of f (x) = x2 cos x. ivar =t . so there is nothing to install on your computer.

What is the proper way to define a system of differential equations? Does a differential equation or a system of differential equations always have a solution? Are solutions to differential equations unique? If a unique solution to a differential equation exists. can we find it? If it is not possible to find a precise solution algebraically. such as the relationship between a population of predators and a population of prey.sagemath. ◦ Finding—either exactly or approximately—the appropriate solution of the equa- tion or equations. Finally. First. . we need a more rigorous definition of a differential equation. 1. Although we will be using Sage as the technology of choice. ( ) dP P =k 1− P. • The three principle steps in modeling any phenomenon with differential equations are: ◦ Discovering the differential equation or equations that best describe a specified physical situation. can we estimate the solution numerically? If neither is possible. x(t0 ) = x0 . If you make a mistake.8 Important Lessons • A differential equation is an equation relating a function to one or more of its derivatives. • A population that is not affected by overcrowding can be modeled by the differential equation P ′ = kP and is said to grow exponentially. A good place to start is http://www. other questions will come to mind as we continue our study of differential equations. • Some phenomenon. 1. dt N where N is the carrying capacity of the population. however. x). • An initial value problem is a differential equation dx = f (t. The reader will find plenty of resources to learn how to use Sage. ◦ Interpreting the solution in terms of the phenomenon. we would like to emphasize once again that the reader who chooses not to use some sort of technology will be at a disadvantage. are best modeled by systems of differential equations. and we encourage the reader to experiment by altering the Sage commands inside the individual Sage cells. this book can be read independently of Sage. dt where the initial condition.1.1.1. • A population that must compete for limited resources can be modeled by the logistic equation.html or [1]. we have left many questions unanswered.7 Some Questions for Thought In this section we have provided a general notion of what a differential equation is as well as several modeling situations where differential equations are useful. is specified. you can simply reload the webpage and start again.org/help. MODELING WITH DIFFERENTIAL EQUATIONS 13 We will provide abundant examples of how to use Sage to solve and analyze differential equations throughout the book. can we still say anything useful about the solution? Of course.1.

Hint. . There must be a minimum population for the species to continue. . A FIRST LOOK AT DIFFERENTIAL EQUATIONS 1. 3. show that ∫ t −at −at x(t) = Ce +e eas b(s) ds t0 is a solution of this equation. Given the equation x′ +px = q(t). Verify that y(t) = Ce−t is a solution to the differential equation y ′ + y = 0. If the wild population becomes too low. Verify that y(t) = 3e−2t is a solution to the differential equation y ′ = −2y Hint. where c is any constant and t0 ∈ I. once the most numerous of all rhinoceros species and now critically endangered. where p is a constant and q(t) is a continuous function defined on an interval I. where P changes at a rate proportional to the square root of P . Determine C if y(0) = 3.14 CHAPTER 1. . We can modify the logistic growth model to account for a threshold population. Suppose the this minimum or threshold population for the black rhino is 1000 animals and that remaining habitant in Africa will support no more that 20.1. That is. We can determine that C = 3. (c) Verify that y = 0 is a solution to the differential equation in part (a). (a) The population is increasing if dP /dt > 0 and 1000 < P < 20000. Can you find a value for C such that 2 y(t) = = 0? 1 + Ce−2t 4. . 2. competing species. The black rhino. Because of hunting. 6. the species is endanger of extinction.000 around 1900. 5. (a) Verify that y(t) = 2/(1 + Ce−2t ) is a solution to the differential equation y ′ = y(2 − y). habitat changes. if the population of a species drops below a certain level.000 rhinos. Write a differential equation that models a population P . the animals may not be able to find suitible mates and the black rhino will become extinct. (a) For what values of P is the rhino population increasing? What can be said about the value of dP /dt for these values of P ? (b) For what values of P is the rhino population decreasing? What can be said about the value of dP /dt for these values of P ? (c) For what values of P is the rhino population in equilibrium? What can be said about the value of dP /dt for these values of P ? (d) Find a differential equation that models the population of rhinos at time t. was estimated to have a population of about 100. √ Hint. (b) Sketch solution curves for C = 1. Since y ′ + y = −3e−t + 3e−t = 0. Hint.9 Exercises 1. Consider the case of the black rhinoceros. dP /dt = k P 5. y(t) = 3e−t is a solution to the initial value problem. the number of black rhinos today is estimated to be below 3000. . native to eastern and southern Africa. y ′ = −6e−2t = −2(3e−2t ) = −2y 2. and most of all illegal poaching.

Think about the limit of the interaction term as the number of prey becomes very large. (a) Consider at sample of material that contains A(t) atoms of carbon 14 at time t. dt 2 2+y dy xy = y(1 − y) − . During each unit of time a constant fraction of the radioactive atoms will spontaneously decay into another element or a different isotope of the same element. which is incorporated into plants by photosynthesis. Consider the following predator-prey systems of differential equations dx x xy =− + . about 5730 years. Animals acquire carbon 14 by eating plants. t0 we see that ∫ t ′ −at −at x (t) + ax(t) − b(t) = −aCe − ae eas b(s) ds + b(t) t0 ∫ t −at −at + aCe + ae eas b(s) ds − b(t) t0 = 0. what is the approximate age of the sample? 8. Make use of the model of exponential growth to construct a differential equation that models radioactive decay for carbon 14. Thus. The long half-life is what makes carbon 14 dating is very useful in dating objects from antiquity. (Radiocarbon Dating) Carbon 14 is a radioactive isotope of carbon. That is. Carbon 14 has a very long half-life. and the amount of isotope in the organism begins to decay into the more common carbon 12. it ceases to take on carbon 14. the sample behaves like a population with a constant death rate and a zero birth rate. given a sample of carbon 14. When an ani- mal or plant dies. as well as from animal bones and charcoal found on the cave floor.1. The resulting carbon 14 com- bines with atmospheric oxygen to form radioactive carbon dioxide. dt 2+y (a) Which equation models the prey population and which equation models the predator population? (b) How does the prey population grow if there are no predators present? (c) What happens if there are a lot of prey present? Hint. . (c) The Chauvet-Pont-d’Arc Cave in the Ardèche department of southern France contains some of the best preserved cave paintings in the world. Carbon samples from torch marks and from the paintings themselves.1. it will take 5730 years for half of the sample will decay to carbon 12. Carbon 14 is created when cosmic ray bombardment changes nitrogen 14 to carbon 14 in the upper atmosphere. Rewriting the differential equation as x′ + px − q(t) = 0 and using the fact that ∫ t x′ (t) = −aCe−at − ae−at eas b(s) ds + b(t). 7. the most common isotope of carbon being carbon 12. (b) Solve the equation that you proposed in (a) to find an explicit formula for A(t). If a particular sample taken from the Cauvet Cave contains 2% of the expected cabon 14. have been used to estimate the age of the cave paintings. MODELING WITH DIFFERENTIAL EQUATIONS 15 Hint.

1. Example 1. the solution is x(t) = Cekt . solve a differential equation y ′ = f (x. we will consider equations of the form dx = f (t. x′′ . .16 CHAPTER 1. . x(n) ) = 0. That is. We can solve separable equations by integrating the first term with respect to x and the second term with respect to y. .9. where F is a function of n+2 variables. we have 1 ln |y| = x2 + C. In this particular case. .2. b) is a function u = u(t) such that the first n derivatives of u are defined on I. u. 2 . u′ . . and F (t. dt 1. x. A FIRST LOOK AT DIFFERENTIAL EQUATIONS 1. We will concentrate on first-order differential equations in this chapter. We can rewrite this equation in the form 1 dy =x y dx or in the alternate form 1 dy = x dx. however.10 An Introduction to Sage Write an introduction to Sage. y) if we can write the equation in the form dy f (x) + g(y) = 0.1 Separable Differential Equations For some first-order differential equations such as dx = kx dt the solution is given by an explicit formula. 1. dx Such equations are called separable. we cannot generally find such a formula for an arbitrary first-order differ- ential equation. . y Integrating both sides of the equation. In general. . Suppose that we wish to solve the initial value problem dy = xy dx y(0) = 1.2 Separable Differential Equations We will define a differential equation of order n to be an equation that can be put in the form F (t. x). A solution to this equation on an interval I = (a. . We can. u(n) ) = 0. x′ . u′′ .

1]) solution Example 1. Consider the initial-value problem dy t = dt y − t2 y y(0) = 4. x) == x*y # write the differential equation solution = desolve (de . the computation will return an error. Notice that the solution does not make sense for all values of t. we can determine the value of C. we see that 0 = ln |1| = 0 + C. Thus. First. In this example. SEPARABLE DIFFERENTIAL EQUATIONS 17 where C is an arbitrary constant. if we require that our solution be continuous. we separate the variables of the equations and write t y dy = dt. we have 1 2 1 y = − ln |1 − t2 | + C or y 2 = − ln |1 − t2 | + C.2. y . we can actually write down an explicit solution that is defined everywhere. ics =[0 . Let us see what Sage has to say. y . The Sage commands for solving our initial value problem are below. Sage will not solve the initial value problem dy = sin(xy) dx y(0) = 1. In fact. 2 /2 y = ex . y(0) = 1 to find C.10. the solution to our initial value problem can be given implicitly by ln y = x2 /2. √ y 2 = 16 − ln |1 − t2 | or y = 16 − ln |1 − t2 |. ics =[0 . x = var ( ' x ' ) # declare x as a variable y = function ( ' y ' )(x) # declare y as a function of x de = diff (y . Using the initial condition. the solution is only defined on the interval −1 < t < 1. 1 − t2 Integrating both sides of the equation. 2 2 Using the initial condition. x) == sin (x*y) solution = desolve (de .1. . x = var ( ' x ' ) y = function ( ' y ' )(x) de = diff (y . 1]) # find the solution solution e ^(1/2* x ^2) If we ask Sage to solve a differential equation that is impossible to solve analytically. y(0) = 4.

2 is a good example of a separable differential equation. y . but for the moment complex numbers will just muddy the situation.8 Sage does return a solution even if it looks a bit different than the one that we arrived at above.2) P (1000 − P ) P 1000 − P Integrating both sides of (1. we obtain . Using partial fractions. We will examine the role of complex numbers and how useful they are in the study of ordinary differential equations in a later chapter. ( ) dP P =k 1− P dt 1000 P (0) = 100. Example 1. The initial value problem in Example 1.11. A FIRST LOOK AT DIFFERENTIAL EQUATIONS t = var ( ' t ' ) y = function ( ' y ' )(t) de = diff (y . where i2 = −1.2). 4]) show ( solution ) -1/2* y (t) ^2 == -1/4* I* pi + 1/2* t + 1/4* log (2* t . (1. ics =[0 .18 CHAPTER 1. we can rewrite this equation as ( ) 1000 1 1 dP = + dP = k dt. t) == t /( y .1) .t *2* y ) solution = desolve (de . Notice that we have an imaginary term in our solution.

.

.

P .

.

ln |P | − ln |1000 − P | = ln .

.

1000 − P . = kt + C.

then the body will cool down after death. Solving for P yields 1000Cekt 1000 P = = . 1. the temperature of the body will match the temperature of the environment. Cekt + 1 Ce−kt + 1 Using our initial condition P (0) = 100. we know that eC is an arbitrary postive constant. we can determine that C = 9.2 Newton’s Law of Cooling Separable equations arise in a wide range of application problems. 1000 − P Since C is an arbitrary constant. We should not expect the body to cool at a constant rate . Taking the exponential of both sides yields P = ekt+C = ekt eC . If the surrounding temperature is cooler. which we will also call C.2. The time of death of a murder victim is an important question on many popular movies and television programs.6◦ F. How does a forensic scientist or a medical examiner determine the time of death? Human beings have a temperature of 98. Eventually. So we can rewrite this last equation as P = Cekt (1000 − P ) = 1000Cekt − Cekt P.

and we know from experience that a body under these conditions cools off approximately 2◦ F during the first hour after death.2.6. Newton’s law of cooling can be easily stated as a differential equation. which tells us that the rate of change of the temperature of a object is proportional to the difference between the temperature of the object and the temperature of the surrounding medium. we can write T − 70 instead of |T − 70|. tells us that D = 28. dT = k(T − Tm ).12).6ek·1 + 70.6 = T (1) = 28. Letting D = eC . Thus. we can determine the constant k to be ( ) 26. the solution becomes T (t) = Dekt + 70. we must solve the initial value problem dT = k(T − 70) dt T (0) = 98.0725t + 70. and k is the proportionality constant. Thus. T (0) = 98. Think of how a hot cup of coffee or tea cools. The initial condition. . where T (1) = 96. we obtain ln |T − 70| = kt + C.0725.6. T (t) = 28. dt where T is the temperature of the object.6ekt + 70. The answer to our forensic question can be found by using Newton’s law of cooling. The liquid will cool quite quickly during the first few minutes but will remain relatively warm for quite a long period. Since 96. T − 70 dt we see that this equation is separable.6. Integrating both sides of the last equation. we have ln(T − 70) = kt + C or T − 70 = ekt+C = ekt eC .6. SEPARABLE DIFFERENTIAL EQUATIONS 19 either. The graph of the T seems appropriate to our model (Figure 1. 28. In order to determine a formula for the time of death.1.6e−0. Tm is the temperature of the surrounding medium. Since we are assuming that T > 70.6 and T (t) = 28. If we rewrite the equation dT = k(T − 70) dt as 1 dT = k.6 k = ln ≈ −0. Suppose that the temperature of the surrounding environment is 70◦ F.

20 CHAPTER 1. If the tank is also draining at a rate of 10 gallons per minute. Suppose that we have a large tank containing 1000 gallons of water and that water containing 0. 0.t ' ) T = function ( ' T ' )(t) de = diff (T .t = var ( ' k .3 Mixing Problems There is a large class of problems in modeling known as mixing problems. the water level in the tank will remain constant.t) == k *( T . the salt flows into the tank at the rate of 10 · 0. there is x(t)/1000 pounds of salt in one gallon.3) dt Of course.6 * exp ( -0. k . or even how various greenhouse gases mix together across different layers of the atmosphere.0725 * t ) + 70 plot (T . then the rate at which the salt is changing in the tank is the difference between the rate at which salt is flowing into the tank and the rate at which it is leaving the tank. the amount salt in the tank at time t. 24) ) 1. T . the rate at which the salt leaves the tank depends on x(t). salt . how pollutants are mixed together in a pond or a lake. We can model the amount of salt in the tank using differential equations. For example. These problems refer to situations where two are more substances are mixed together in a container or containers. or dx = rate in − rate out. A FIRST LOOK AT DIFFERENTIAL EQUATIONS T(t) 110 100 90 80 70 60 0 10 20 30 40 50 t Figure 1.5 pounds of salt per gallon flows into the tank at a rate of 10 gallons per minute. ivar = t ) solution (c + 70* e ^( .2.5 = 5 pounds of salt per minute.k*t))*e ^( k*t) We can use Sage to plot our solution. Therefore. At time t. how ingredients are mixed together when brewing beer. However. If x(t) is the amount of salt in the tank at time t. t = var ( ' t ' ) T (t ) = 28.70) solution = desolve (de .12: Newton’s law of cooling Let us solve our differential equation using Sage. We will assume that the water in the tank is constantly stirred so that the mixture of salt and water is uniform in the tank. (t . (1. we might wish to model how chemicals are mixed together in a refinery.

where P (t) is the amount in thousands of dollars in the fund at time t.01t .4 A Retirement Model Differential equations have many applications in economics and finance. wisely started saving for his retirement as soon as he entered the work- force. Equation (1.’s investment. 500 − x 100 Integrating both sides of the equation.01t models the amount of salt in the tank at time t. J. plans to retire in 10 years. dP = 0. 100 Consequently. and he now has $500. The solution to our initial value problem is P (t) = 500e0. we can quickly determine that C = 500 and x(t) = 500 − 500e−0. 1. still plans to make contributions to his retirement fund during his next ten years of employment.3) now becomes dx x =5− dt 100 x(0) = 0. Dr. which his employer will generously match.2.05t . we have t − ln |500 − x| = +k 100 or t ln |500 − x| = − − k. where C = e−0. If Dr..05P + 10 dt P (0) = 500 . J. Dr. 500 − x = Ce−0. The initial value problem.000 in a retirement account earning an interest of 5% com- pounded continuously.1.01k . dP = 0.360635350064 or about $824. If we assume that these contributions will spread out evenly over the course of the year.05P dt P (0) = 500 provides a nice model of Dr. Of course. This equation is separable.2.000. he can expect a nest egg of P (10) ≈ 824. For example. dx dt = .360. we can incorporate this information into our original initial value problem. J. a college professor. From our intial condition. His annual contribution will be $5. SEPARABLE DIFFERENTIAL EQUATIONS 21 flows out of the tank at a rate of 10x(t)/1000 = x(t)/100 pounds per minute. J.

05P + 10 > 0.05P + 10. He estimates that he will need to withdraw $60. then dP = 0.’s retirement fund will be disappear.05P + 10) = t + k. the solution that we seek is P (t) = 700e0.000 a year for living expenses if he wishes to travel and enjoy his golden years. t ≤ 10 = dt 0. so we have ∫ ∫ dP = dt. if Dr. we obtain P = 20Ce0.05P − 60 Intergrating both sides of this equation yields 20 ln |0.05P + 10| = t + k. { dP 0.05k . J. Using our initial condition. plans to retire early and live a long life. J. Again. This may pose a problem.104889490090 or about $954.05t − 200. Dr. J. the rate of withdrawal exceed the rate at which Dr. the differential equation dP /dt = 0.’s account is earning interest. and we have ∫ ∫ dP = dt.3 < 0. whatever remains in his account at any given time will still collect interest. we have C = 35.’s nest egg is now P (10) ≈ 954. we have 20 ln(0. where k is an arbitrary constant. 500 = P (0) = 20Ce0.05P + 10 = e0. We describe J.05·0 − 200 = 20C − 200. Once Dr. Thus. 0. he will need to begin withdrawing money from his account. t > 10 P (0) = 5. retires. Since 0.05P − 60. Dr. dt Hence. Of course.22 CHAPTER 1.05P − 60| = t + k.05P − 60 is separable. A FIRST LOOK AT DIFFERENTIAL EQUATIONS This differential equation is separable.05t − 200.05k and solve for P . This last equation is equivalent to 0. If we let C = e0.05P − 60 ≈ −12. Eventually.105. If P = 954. . J.05t e0. we have 20 ln |0.05(t+k) = e0. 0.’s retirement situation with the initial value problem.05P + 10 Integrating both sides of this equation. J.

retires in 10 years at the age of 65. In this case.05t describes how much money Dr.21 This means that if Dr.05t .4) dx We now will prove that such an equation can solved by integrating the first term with respect to x and the second term with respect to y.2. then we can rewrite equation (1. 149. If ∫ h1 (s) = f (s) ds ∫ h2 (s) = g(s) ds.21e0. (1.7.05t = 0. y) is called separable if it can be written in the form dy f (x) + g(y) = 0.05P − 60 < 0. ( ) 1200 t = 20 ln ≈ 41. A differential equation y ′ = F (x. dx .05P − 60| = 60 − 0. J.4) as dy h′1 (x) + h′2 (y) = 0.05k .05P = e0. we can apply our initial condition P (10) = 954 to determine that C ≈ 149. and 20 ln(60 − 0.21. we obtain dy d h′2 (y) = [h2 (y)].05t . P = 1200 − 149.05P ) = t + k.4) now becomes d (h1 (x) + h2 (y)) = 0. he must solve the equation 1200 − 149. Therefore. he can expect his retirement to last into his mid 90s.05(t+k)/20 = Ce0. 60 − 0. equation (1. If Dr. |0.05P.21e0. Consequently.5 Some Theory We now give a theoretical basis for solving first-order separable differential equations. J. after he retires (t ≥ 10). Now. J.1. or P = 1200 − Ce0. dx dx Hence.2. 1. dx Applying the chain rule to the second term. where C = e0. SEPARABLE DIFFERENTIAL EQUATIONS 23 Since 0. want to know how long his retirement fund will last.

we have arctan y + ln |y| = x + C. Another difficulty arises if we consider the equation y ′ = te−y . 0 2 . we simply need to integrate each term to solve the differential equation.13.2. we have no method of solving this last equation explicitly for y. dx Then h1 (x0 ) + h2 (y0 ) = C. Although the Fundamental Theorem of Calculus guarantees that every continuous function 2 has an antiderivative. Consider the equation dy y(1 + y 2 ) = 2 . we cannot find an antiderivative for the function ey in terms of elementary functions. dx y +y+1 This equation can be rewritten in the form ( ) 1 1 + dy = dx 1 + y2 y Integrating both sides of the equation. x0 ∫ y h2 (y) − h2 (y0 ) = g(s) ds. 1.24 CHAPTER 1. A FIRST LOOK AT DIFFERENTIAL EQUATIONS Integrating.6 What Can Go Wrong Example 1. we are forced to write our solution as ∫ y 2 1 es ds = t2 + C. It is not always possible to explicitly solve a separable differential equation. y0 Consequently. we obtain h1 (x) + h2 (y) = C. where C is any arbitrary constant. Now suppose that y(x0 ) = y0 is an initial condition for dy f (x) + g(y) = 0. we can replace equation (1.14. 2 This equation is separable since we can rewrite it in the form 2 ey dy = t dt. Example 1. However. Thus.4) with the integral equation ∫ x ∫ y f (s) ds + g(s) ds = 0. By the Fundamental Theorem of Calculus ∫ x h1 (x) − h1 (x0 ) = f (s) ds. x0 y0 In other words.

3 If C = 0.16. Separating the variables. where F is a function of n + 2 variables. Separating the variables. the solution is 1 y=− . However. y2 we see that 1 y=− t+C or 1 y= .5 we will learn sufficient conditions for a first-order initial value problem to have a unique solution. . Even if we have a separable differential equation. 1. and F (t. 3 y ≡ 0. b) is a function u = u(t) such that the first n derivatives of u are defined on I. 1) if y(0) = −1. we can find at least two additional solutions for t ≥ 0: ( )3/2 2 y=− t .15. Consider the initial value problem y ′ = y 1/3 with y(0) = 0 and t ≥ 0. u′′ . the initial condition is satisfied and ( ) 2 3/2 y= t 3 is a solution for t ≥ 0. . In Section 1. Example 1. Thus. .7 Important Lessons • A differential equation of order n is an equation that can be put in the form F (t. x. 1−t Therefore. In the case that y(0) = −1. t+1 and a continuous solution exists on (−1. x(n) ) = 0. we are not guaranteed a unique solution. a continuous solution also exists on (−∞. . . 1 dy = dt. SEPARABLE DIFFERENTIAL EQUATIONS 25 Example 1.2. . . . Suppose that y ′ = y 2 with y(0) = 1. u(n) ) = 0. . y −1/3 dy = dt. 3 2/3 y =t+C 2 or ( )3/2 2 y= (t + C) . u. x′ . A solution to the equation on an interval I = (a.2. ∞). x′′ . u′ .1.

2* x desolve (de . A FIRST LOOK AT DIFFERENTIAL EQUATIONS • A first-order differential equation is an equation that can be written in the form dx = f (t. y) = f (tx. 1. 2. What is the domain of your solution? 4. . Solve the initial value problem dy = t2 y 2 + t2 .2. y(0) = −1 y+1 Hint. (a) x = var ( ' x ' ) y = function ( ' y ' )(x) de = diff (y . dx In this case we can rewrite the equation in the form dy f (x) + g(y) =0 dx or g(y) dy = f (x) dx and solve by integrating both sides. Homogeneous Equations. dy (a) = 8 − 2x dx dy x+1 (b) = dx y dx (c) = x2 t2 . y). dt y(0) = 1.26 CHAPTER 1. x) == 8 . A first-order differential equation. Hint.2. (c) x = 3/(3 − t3 ). Solve each of the following equations in Exercise 1. ty). x(0) = 1 dt t2 y − y (d) y ′ = . (a) y = −x2 + 8x + C.1 using Sage. y) -x ^2 + _C + 8* x 3. y ′ = f (x.8. x).8 Exercises 1. is homogeneous if f (x. Solve each of the following differential equations. dt • A differential equation is separable if it can be written in the form dy = M (x)N (y).

can be transformed into a separable differential equation using the substitution y = vx. (c) Show that any homogeneous equation y ′ = f (x. (e) Solve the differential equation x2 y ′ = 2y 2 − x2 . Show that differential equation P (x. Although some differential equations have an exact solution and can be solved using analytic techniques with calculus. dx 2vx2 Use the fact that this new equation is separable to solve for y. 2. . the famous detective Hercule Poirot noted that Ratchett had a body temperature of 28 degress. y) can be transformed into a separable differential equation by making the substitution y = vx. y)dx + Q(x. .9 Sage—Quick Start Guide to Solving Ordinary Differential Equa- tions Sage has powerful algorithms for finding exact and numerical solutions of differential equa- tions. Mr. Sage needs three things to solve a differential equation: • An abstract function • A differential equation • A Sage command to solve the equation.2. many differential equations can only be solved nusing numerical techniques. The body had cooled to a temperture of 27 degrees one hour later. .M. SEPARABLE DIFFERENTIAL EQUATIONS 27 (a) Show that the equation dy x2 + y 2 = dx 2xy is homogeneous.. ty) = tn f (x. Ratchett. estimate the time of Ratchett’s death using Newton’s Law of Cooling. (d) A function f is said to homogeneous of degree n if f (tx. an elderly American. y)dy = 0. We can even encounter difficulties when using numerical algorithms such as the Newton-Raphson algorithm. Unlike the situation for quadratic equations. 5. This should not be too surprising if we consider how we solve polynomials. where P and Q are both homogeneous of degree n. When his body was discovered. In general. 1. . In addition. (b) Let y = xv and show that the equation in part (a) can be written as dv x2 + v 2 x2 v+x = . y) for n = 1. we can plot solutions and direction fields. but solving an equation such as 5x7 − 6x4 + 3x3 − 23x2 + 3x − 17 = 0 is a much more difficult problem. It is quite easy to find the roots of any equation of the form ax2 + bx + c = 0 by either factoring or using the quadratic equation. there does not exist a general formula for solving seventh degree equations.2. If the normal temperature of a human being is 37 degrees and the air temperature in the train is 22 degrees. was found murdered in his train compartment on the Orient Express at 7 A.1.

then a solution curve will have a slope of 2 at the point (1._C )*e^x The first command defines the abstract function. Try replacing the h command with show(h) or show(expand(h)). 1). The second describes the actual differential equation. 1. y) as a formula for the slope of a tangent line to a solution curve.x) .3 Geometric and Quantitative Analysis If we view the differential equation y ′ = f (t. and a resistor (Figure 1.3. A capacitor stores an electrical charge and can be made by separating two metal plates with an insulating material. A current flowing in the opposite direction will be given negative values.1]) h h There are many other commands and packages to solve ordinary differential equations using Sage. ics =[0 .28 CHAPTER 1.sagemath.17). if we consider the equation y ′ = t + y.x) == x + y h = desolve (de . For more information. Finally. An empty Sage cell is below for practice and exploration. see http://www. I(t). We can also specify an initial condition for our differential equation. 1. y) h -(( x + 1) *e ^( . The glowing red heating element in a toaster or an electric stove is an example of something that provides resistance in a circuit. We can use this information to obtain a geometric description of the solutions to the equation. An RC circuit is a very simple circuit might contain a voltage source.org/doc/reference/calculus/ sage/calculus/desolvers. Current.1 RC Circuits Suppose that we wish to analyze how an electric current flows through a circuit. A FIRST LOOK AT DIFFERENTIAL EQUATIONS Suppose we wish to solve the equation dy = x + y. a capacitor. is the rate at which a charge flows through this circuit and is measured in amperes or amps (A). We assign a direction to the current. dx We can use the following sequence of Sage commands. say y(0) = 1. . y . Capacitors are used to power the electronic flashes for cameras. we can approximate the graph of a solution curve.html. A battery or generator is an example of a voltage source.x) == x + y h = desolve (de . For example. y = function ( ' y ' )(x) de = diff (y . y = function ( ' y ' )(x) de = diff (y . we use the Sage command desolve to find the actual solution.

7) dt We will now investigate how our circuit reacts under different voltage sources. Coulomb’s Law tells us how current flows across a capacitor. b] × [c. Suppose that our differential equation is defined on the rectangle R = [a. R. Kirchhoff’s Second Law tells us that the impressed voltage in a closed circuit is equal to the sum of the voltage drops in the rest of the circuit. (1.7) for different voltage sources E(t). and the voltage drop across the capacitor. C. Resistance. EC . For ex- ample. ER . we can analyze our circuit from a geometric point of view and can actually say a great deal about circuits without solving a differential equation. y0 ). is measured in volts (V). and let y(t) be a solution curve for y ′ = f (t. capacitance. E(t).1. We would like to be able to understand the solutions to the equation (1. to the current is measured in ohms (Ω). dEC I=C .5) and (1.2 Direction Fields Any differential equation dy = f (t. Geometrically. this is ER + EC = E. y) dt can be viewed as a formula for the slope of a function y = y(t). the equation tells us that. We could also have a constant nonzero source of voltage such as a battery or a fluctuating source of voltage such as a generator. is measured in farads (F). our equation ER + EC = E becomes dEC RC + EC = E(t).6).3. GEOMETRIC AND QUANTITATIVE ANALYSIS 29 R + E(t) I(t) C − Figure 1. at any point (t0 . if we combine the equations (1. d].3. we need only compute the voltage drop across the resistor. Thus. (1. (1. the slope of a solution curve is given by f (t0 .17: An RC circuit The source voltage. Ohm’s Law tells us that the voltage drop across a resistor is given by ER = IR. 1.6) dt Thus.7) as an expression for computing how fast current is flowing across the capacitor. We might even have a series of pulses of voltage where the current is periodically turned on and off. y0 ). If we view the differ- ential equation (1. y0 ). we might have a zero voltage source (the capacitor could still hold a charge). y) that passes through the point (t0 .5) Finally. Then the differential . According to Kirchhoff’s Law.

y ) = y ^2/2 . y0 ).3. (y . Most computer algebra systems.19.y = var ( ' t . we can obtain a direction field or slope field for the differential equation. we will get the direction field of given in Figure 1. headaxislength =3 .18 along with several solution curves. For example. . For example. we can easily generate direction fields for any differential equation with the use of computer software. -1 . dt RC with R = 1 and C = 1.3 RC Circuits Revisited Now let us return to our RC circuit and consider different functions E(t) for the differential equation dEC E(t) − EC = .5) .t plot_slope_field (f . y0 ). (t . have facilities for generating and graphing direction fields. t . y) with slope f (t0 . where v0 = EC (0). axes_labels =[ ' $t$ ' . y)-plane by drawing a short line segment at the point (t. We can indicate this on the (t. Thus.18: The direction field for y ′ = y 2 /2 − t Although direction fields can be tedious to compute using pencil and paper. including Sage.10) .30 CHAPTER 1. First suppose that there is no voltage source in the circuit. 10 8 6 4 y(t) 2 0 -2 -4 -1 0 1 2 3 4 5 t Figure 1. The direction field agrees with our analytic solution EC (t) = v0 e−t . We will learn how to add solution curves at the end of this section. A FIRST LOOK AT DIFFERENTIAL EQUATIONS equation tells us the slope of this solution curve at (t0 . the following Sage cell will generate the direction field for y ′ = y 2 /2 − t using the command plot_slope_field. consider the differential equation y ′ = y 2 /2 − t. headlength =3 . A solution curve must be tangent to its direction field at every point.y ' ) f (t . If we let E(t) = 0 for all t ≥ 0. -5 . ' $y ( t ) $ ' ]) 1. The direction field for this equation is given in Figure 1.

3. E(t) = 0 t > 4. .20: The direction field for a constant current If we attach a battery to our circuit at time t = 0 and then disconnect the battery at t = 4.19: The direction field for no current If we assume that we have a nonzero constant source of voltage.21). 10 5 E(t) 0 -5 -10 0 1 2 3 4 t Figure 1. E(t) = K. in our circuit such as a battery. For example.20. if { 10 0 ≤ t ≤ 4. we will obtain a different direction field (Figure 1. then we obtain a different solution.1. dt The direction field for this differential equation for K = 10 is given in Figure 1. GEOMETRIC AND QUANTITATIVE ANALYSIS 31 10 5 E(t) 0 -5 -10 0 2 4 6 8 10 t Figure 1. then we obtain the separable differential equation dEC = K − EC .

A FIRST LOOK AT DIFFERENTIAL EQUATIONS 20 15 E(t) 10 5 0 0 1 2 3 4 5 6 7 8 t Figure 1. .22: The direction field for a series of pulses Finally.22. the voltage source might be given by a function such as E(t) = sin(πt/2). E(t) =   10 8 ≤ t < 12.32 CHAPTER 1. then the direction field for our differential equation is given in Figure 1. 20 15 E(t) 10 5 0 0 5 10 15 t Figure 1.     . .23. 10   0 4 ≤ t < 8.21: The direction field for a single pulse If our voltage source emits a series of pulses. if we use a generator for a voltage source. say    0 ≤ t < 4. The direction field for this circuit is given in Figure 1..

size =60 . x_start ]. You can then change each field in the cell output to solve a particular initial value problem. color = " red " ) p.4 Plotting Direction Fields with Sage It is very tedious to plot direction fields by hand. arrowsize = 8) p += point (( t_start .1.3. step_size = (0.5 Autonomous Differential Equations An autonomous differential equation is one of the form dx = f (x). thickness =3) p = line (P . x_max ) . ics =[ t_start . show () pretty_print ( html (r"$\ displaystyle \ frac { dx }{ dt } ␣ = ␣ % s$ " % latex (f))) Of course.001 . we have technology to help.1) . 1. Fortunately. t_start = input_box ( default = 1) . headaxislength =3 . dt . headlength =3) p += arrow (P[ int ( len (P) /2) ] . (t .1 . x_start = input_box ( default = -0. x . t_max ) . len ( P ) -1) ) [0] p += plot_slope_field (f . 1. step = step_size .t ) .. The Sage interact below will help you plot direction fields for a given initial value problem.3. it is quite easy to give the interact a problem that yield unitelligible results. P [ int ( len ( P ) /2) + 1] . Simply evaluate the Sage cell.23: The direction field for an oscillating voltage 1.. thickness = 2 .t ' ) @interact def _(f = input_box ( default = x ^2/2 .3. t = var ( ' x . x_start ) . GEOMETRIC AND QUANTITATIVE ANALYSIS 33 3 2 1 E(t) 0 -1 -2 -3 0 2 4 6 8 10 t Figure 1.0) .) x_min = min ([ P[i ][1]] for i in (0.x . end_points =[ t_min . t_max = input_box ( default = 6) . (x . x_min . A Sage interact is a menu driven Sage applet. ivar =t . (0. width = 1 . t_max ] . axes_labels = [ ' $t$ ' . t_min = input_box ( default = -1. t_min .0) ) ) : P = desolve_rk4 (f . len ( P ) -1) ) [0] x_max = max ([ P[i ][1]] for i in (0. ' $x$ ' ] .

we can draw a single line containing the same information. Thus. and we will indicate this by a downward pointing arrow. Conse- quently. Let us consider a trout pond that has a carrying capacity of 200 fish. If so. our equation becomes ( ) dP P =P 1− − 32. We can do this with a phase line. In this case any solution curve will be decreasing. we know that dP /dt > 0. Thus. . If we allow the fish to be harvested at a constant rate of 32 per year. Instead of drawing the entire direction field. we know that dP /dt < 0 when P < 40 or P > 160.34 CHAPTER 1. A FIRST LOOK AT DIFFERENTIAL EQUATIONS In other words. We say that such a solution is an equilibrium solution. under what conditions for sustainablility? Since ( ) dP P 1 =P 1− − 32 = − (P − 40)(P − 160) dt 200 200 is an autonomous differential equation. Suppose that the trout population can be modeled according to the logistic equation ( ) dP P =P 1− .25: Logistic growth with harvesting One of the basic questions that we can ask of our model is whether or not we have a sustainable population in our trout pond given this harvest rate.26).24 (A Logistic Model with Harvesting). dt 200 where t is the time in years. On the other hand. any solution curve must be increasing. the two constant solutions P (t) = 40 and P (t) = 160 are the same for all values of the independent variable t. We denote this property on the phase line by drawing an upward pointing arrow. a differential equation is autonomous if it does not depend on t. Notice that dP /dt = 0 when P = 40 or P = 160. Example 1. dt 200 The direction field for this equation is given in Figure 1. Equilibrium solutions graph as horizontal lines. the direction field does not depend on t. We can identify equilibrium solutions by setting the derivative of the function equal to zero.25. On our phase line we will represent these solutions as equilibrium points (Figure 1. 250 200 150 P(t) 100 50 0 0 2 4 6 8 10 t Figure 1. Since an autonomous differential equation is time independent. For values of P between 40 and 160. we need only keep track of what happens on the vertical axis. The slope only depends on x and is the same for all values of t. its direction field is particularly easy to draw.

Therefore.27: Sinks. we have a node.26).26: Phase line diagram An equilibrium solution is stable if a small change in the initial conditions gives a solu- tion which tends toward the equilibrium as the independent variable tends towards positive infinity.28). GEOMETRIC AND QUANTITATIVE ANALYSIS 35 P = 160 is a sink P = 40 is a source Figure 1. We say this solution is a sink if for any solution y(t) with initial condition sufficiently close to y0 . we have a source.8) dt The graph of f (y) = y 4 − 4y 2 is given in Figure 1. dy/dt 6 4 2 y -3 -2 -1 1 2 3 -2 -4 Figure 1. Let y ′ = f (y) and suppose that y = y0 is an equilibrium solution. An equilibrium solution is unstable if a small change in the initial conditions gives a solution which veers away from the equilibrium as the independent variable tends towards positive infinity. An equilibrium solution that is neither a sink or a source is called a node (Figure 1. then y is increasing. t→∞ We say that an equilibrium point is a source if all solutions that start sufficiently close to y0 tend toward y0 as t → −∞. Consider the differential equation dy = y 4 − 4y 2 = y 2 (y + 2)(y − 2).27). Finally. we have lim y(t) = y0 .8) from the graph of f (y) = y 2 (y + 2)(y − 2) (Figure 1. we have a sink.27. If y = −2. If the graph is above the y-axis. If y = 2.3. then y is decreasing. (1. .1. the phase line is easy to sketch (Figure 1. if y = 0. sources. and nodes It is easy to generate a phase line diagram for equation (1. If the graph is below the y-axis.

In this case. source. an equilibrium solution is either a sink.36 CHAPTER 1. We can use a phase line to analyze autonomous differential equations. f ′′ (x0 ) f (n) (x0 ) f (x) ≈ f (x0 ) + f ′ (x0 )(x − x0 ) + (x − x0 )2 + · · · + (x − x0 )n 2! n! near x0 . this strategy might not work very well if f (x0 ) = cos(x20 + π) = 0 or f ′ (x0 ) = 2x0 sin(x20 + π) = 0. which tells us that any function f (x) can be approximated near a point x0 by an nth degree polynomial. 1. In particular. .3. • An autonomous equation is a differential equation of the form y ′ = f (y). We can classify equilibrium solutions according to whether they are stable or unstable.6 Important Lessons • Direction fields and phase lines are a useful way of analyzing a differential equation from a geometric point of view. • Equilibrium solutions to a differential equation y ′ = f (y) are those solutions given by f (y) = 0 for all y. any solution must be constant.28: Phase line diagram for y ′ = y 2 (y + 2)(y − 2) One of the reasons why autonomous equations are so important is Taylor’s theorem. if dx = f (x) = cos(x2 + π). especially since not all differential equations can be solved analytically. dt with x(0) = x0 then we may approximate this initial value problem near x0 with dx = f (x0 ) + f ′ (x0 )(x − x0 ) = cos(x20 + π) + 2x0 sin(x20 + π)(x − x0 ) dt x(0) = x0 . For example. A FIRST LOOK AT DIFFERENTIAL EQUATIONS y = 2 is a source y = 0 is a node y = −2 is a sink Figure 1. Of course. or node.

This is a fourth-order Runge-Kutta method. (y .8 Sage—Plotting Direction Fields and Solutions Plotting direction fields If we wish to plot a direction field for a differential equation. or node.10) .1. and returns a numerical solution (a table of values). y = var ( ' t . . (a) y ′ = 2y(1 − y) dP (b) = 0. y) = y ^2/2 . ' $y ( t ) $ ' ]) v For more examples and options. (t . Identify any equilibrium solutions and classify each equilibrium solution as a sink. see http://doc. (t .3.5) . axes_labels =[ ' $t$ ' . Hint. 1. Use Sage to graph the direction field of each equation and then superimpose a plot of the equilibrium solution(s) on the direction field. ' $y ( t ) $ ' ]) v There are a few extra commands to specify the size of the arrows in the plot and to label the axes.␣y ' ) f (t .html Plotting solutions Now let us find a numerical solution to the equation using the command desolve_rk4. Find all of the equilibrium solutions for each of the following differential equations. -1 . -5 . Let us plot the direction field for the equation y ′ = y 2 /2 − t.3. 3.t v = plot_slope_field (f . Here. we must supply the dependent variable and initial conditions.org/html/en/reference/plotting/ sage/plot/plot_field.3. y) = y ^2/2 .24. we can use the command plot_slope_field.? What should be our strategy to maintain a viable population in the trout pond? Hint.1P (P − 100) dt dx (c) = (2 − x) cos x dt dy (d) = (2 − y) cos x dx 2. What happens if we increase the harvest rate to 100 in Example 1.␣y ' ) f (t .5) . -1 .sagemath. a source. where the graph of f (y) is given below. t . headaxislength =3 . Try changing or omitting these commands and see what happens. source. y = var ( ' t .10) .7 Exercises 1. or a node. headlength =3 . headlength =3 . GEOMETRIC AND QUANTITATIVE ANALYSIS 37 1. headaxislength =3 . t . axes_labels =[ ' $t$ ' . Consider the differential equation y ′ = f (y). -5 .t v = plot_slope_field (f . (y . Classify each equilibrium solution as a sink.

we will change our original model to the folllowing:  ( p ) dp  0. headlength =3 . y) = y ^2/2 . t . For more information. ics =[ -1 .5) . (c) Find formulas for the solutions of this equation for initial conditions p(0) = 30. 6 See any calculus text for a description of the Newton-Raphson Algorithm.38 CHAPTER 1. (d) Give a qualitative description of how the disease affects the population.4 Analyzing Equations Numerically Just as numerical algorithms are useful when finding the roots of polynomials. -5 .4p 1 − − 0. numerical methods will prove very useful in our study of ordinary differential equations. Suppose that the population of a trout pond can be accurately modeled by the logistic equation dp ( p ) = 0. -1 . = ( 500 p ) dt 0. 500 (a) Plot the direction field for this equation using Sage. To see how the disease affects the fish population.4p 1 − for 0 ≤ t < 30.1p for t > 30. it may be easier to generate a numerical solution for a differential equations if our goal is simply to plot a solution. 1. output = ' plot ' . ivar =t . Similarly. thickness =2) p Of course.t p = plot_slope_field (f .␣y ' ) f (t . y .10) . (y . y = var ( ' t .0] . end_points =[ -1 . Sage Exercises 1. Consider the polynomial f (x) = x2 − 2. headaxislength =3 . A FIRST LOOK AT DIFFERENTIAL EQUATIONS t . y) = y ^2/2 . dt 500 At time t = 30. y = var ( ' t . output = ' plot ' . axes_labels =[ ' $t$ ' .␣y ' ) f (t . where you can practice. we can combine the two plots.0] .org/doc/reference/calculus/ sage/calculus/desolvers. We do √ not need a √ numerical algorithm to see that the roots roots of this polynomial are x = 2 and x = − 2. However. (t . ics =[ -1 . a disease is introduced into the population that kills 10% of the population per year. see http://www. Below is an empty Sage cell.html.4p 1 − .sagemath. y .5] . (b) Plot the graphs of two or three representative solutions to this equation on the direction field. . and there will be differential equations for which we cannot find an analytic solution of course. ' $y ( t ) $ ' ]. fontsize =12) p += desolve_rk4 (f . a numerical method such as the 6 √ Newton-Raphson Algorithm could be useful if we need a decimal approximation for 2. thickness =2) p There are many other commands and packages to solve ordinary differential equations using Sage. end_points =[ -1 .t p = desolve_rk4 (f .5] . ivar =t .

9) y(0) = 1.9)–(1.10) on the interval [0.1. t0 ). however. Even with the use of a computer. 2. we can try to find an approximation for the solution.1 Euler’s Method Suppose that we wish to solve the initial value problem y ′ = f (t. we might be able to find approximations at a = t0 .10) The equation y ′ = y + t is not separable. . Our general algorithm is Yk+1 = Yk + hf (tk . the approximation at t2 = t0 + 2h will be Y2 = Y1 + hf (t1 . Y0 = y(0) = 1. we cannot approximate the solution at every single point on an interval. . y) = y + t. t2 . . Yk ). . Since y(0) = 1. y) y(t0 ) = y0 . If we choose t1 . . tN = b in [a. To generate the second approximation. tN to be equally spaced on [a. For the initial value problem y ′ = f (t. We will try to find an approximate solution on the interval [0. . . . (1. t2 .4. y0 ). Y0 ). . N . Let us estimate a solution to (1. The algorithm that we have described is known as Euler’s method. t1 . If we use slope of the solution curve at t0 . The idea is to compute tangent lines at each step and use this information to get our next approximation. then y ′ (t0 ) = f (y0 . 1]. . . which currently is the only analytic technique at our disposal. We can make our first approximation exact. We say that h is the step size for our approximation. b] at best. Y1 ). ANALYZING EQUATIONS NUMERICALLY 39 1.1. the question is how to find an approximate solution Yk+1 at tk+1 . Given an approximation Yk for the solution yk = y(tk ). we can write tk = t0 + kh. (1. the estimate for our solution at t1 = t0 + h is Y1 = Y0 + hf (t0 . b].4. Similarly. . y(t0 )) = y ′ (t0 . y0 ) h or equivalently y(t0 + h) = y(t1 ) ≈ y(t0 ) + hy ′ (t0 . where h = 1/N and k = 1. we will construct a tangent line to the solution at y(t0 ) = y0 . . Making use of the fact that y(t0 + h) − y(t0 ) ≈ y ′ (t0 . 1] with step size h = 0.

the approximation at t = 0.0000 1.2200 1. 0) = y(0) + 0 = 1 + 0 = 1.2 will be Y2 = Y1 + hf (t1 . k tk Yk yk |yk − Yk | Percent Error 0 0. a smaller step size gives us a more accurate approximation (Table 1.31.4366 0. in fact. and make use of the fact that y(h) − y(0) ≈ y ′ (0.1)[1 + 0] = 1. We can compare our approximation to the exact solution in Table 1.5836 0.6 1.2 1. Our general algorithm is Yk+1 = Yk + hf (tk .7210 1.01)k.1103 0. Notice that the error grows as we get further away from our initial value.0442 0.2200. Y0 ) = Y0 + h[Y0 + t0 ] = 1 + (0.3620 1.0764 4.1011 4. Y1 ) = Y1 + h[Y1 + t1 ] = 1.25% Table 1.1000 + 0.1 is Y1 = Y0 + hf (t0 . Yk ) = Yk + h[Yk + tk ] = (1.69% 4 0. This equation is.0554 3. y(0)) or y(h) ≈ y(0) + y ′ (0.2428 0.29.1] = 1. A FIRST LOOK AT DIFFERENTIAL EQUATIONS To generate the second approximation.1)[1.50% 5 0. we will construct a tangent line to the solution at y(0) = 1.1875 3. In fact.9431 2. y(0)).7974 0.0103 0.84% 3 0. If we use slope of the solution curve at t0 = 0.5282 1.0000 0.95% 10 1.1 1. h the estimate for our solution at t = 0.1000 1.1000 + (0.2491 7.0 1.3997 0.1)Yk + (0. solvable analytically with solution y(t) = 2et − t − 1. In addition.00% 1 0.30).4 1. We can also see graphs of the approximate and exact solutions in Figure 1.0 3.40 CHAPTER 1.0228 1. the graph of the approximation for h = 0. y ′ (0) = f (y(0).0377 2.5 1.93% 2 0.29: Euler’s approximation for y ′ = y + t .0000 0.001 is obscured by the graph of the exact solution. Similarly.1000.25% 6 0.3 1.

6 1. . choose t1 .3993 1.5282 1.1102 1.2 1.7966 1.6511 0.1 h = 0.0 3.2426 1.4366 Table 1.3832 3.5719 1.2 0.9431 2. . 02 h = 0.6493 2.3 1.32.2 Finding an Error Bound To fully understand Euler’s method. tN to be equally spaced on [t0 .1000 1. then f ′′ (x0 ) f (n) (ξ) f (x) = f (x0 ) + f ′ (x0 )(x − x0 ) + (x − x0 )2 + · · · + (x − x0 )n .1 1.3997 0. t2 .9757 3.3261 2.31: Euler’s approximation for y ′ = y + t 1.8 1 Figure 1.2200 1.1875 3. 1 h = 0.30: Step sizes for Euler’s approximation 3.8 2.3917 1.9 2. .0442 0.7974 0.4. Given the initial value problem y ′ = f (t. x).001 Exact Solution 0.1.6161 2. we will need to recall Taylor’s theorem from calculus.7 2. we can write tk = t0 + kh.4 1.5 2 1.1082 1.4238 3. y). 2! n! where ξ ∈ (x0 .2998 2.2380 1.5 1 0 0. .5 h = 0.4 0.02 h = 0.4.3620 1. Theorem 1.8159 2. 001 yexact 3 2. y0 = y(t0 ).1974 2.0431 2.1092 1.7210 1.4872 2. If x > x0 .7812 1.1103 0.5 1. .3275 0. a].0170 3. ANALYZING EQUATIONS NUMERICALLY 41 tk h = 0.5836 0.5831 1.2428 0.6 0.0227 2.

b] × R is called a Lipschitz condition. A FIRST LOOK AT DIFFERENTIAL EQUATIONS where h = (a − t0 )/N and k = 1. . . If the condition is satisfied. . y1 ) and (t. . The simplest approximation can be obtained by taking the first two terms of the Taylor series. whenever (t. 2L The condition that there exists a constant L > 0 such that |f (t. b] × R. YN are the approximations generated by Euler’s method for some positive integer N . we can usually say a great deal about the function. If Y0 . . y2 ) are in D = [a.. The terms that we are omitting. b]. b]. will be very small and these terms will not matter much. y). . Yk+1 = Yk + hf (tk . y2 )| ≤ L|y1 − y2 |. . . Many of the functions that we will consider satisfy such a condition. all contain powers of h of at least degree two. where t ∈ [a.42 CHAPTER 1. Theorem 1. then we can determine the value of y at tk+1 . y2 )| ≤ L|y1 − y2 |. N . Suppose that f is continuous and there exists a constant L > 0 such that |f (t. y1 ) and (t. y(a) = α. y2 ) are in D = [a. . Y1 ) . we will use a linear approximation. Let y be the unique solution to the initial value problem y ′ = f (t. . Taylor’s Theorem tells us that y ′′ (tk ) 2 y(tk+1 ) = y(tk + h) = y(tk ) + y ′ (tk )h + h + ··· . Also assume that there exists an M such that |y ′′ (t)| ≤ M for all t ∈ [a. whenever (t. Yk ). 2! If we know the values of y and its derivatives at tk . . then hM L(ti −a) |y(ti ) − Yi | ≤ [e − 1]. We can actually estimate the error incurred by Euler’s method if we make use of Taylor’s Theorem. If h is small. then hn for n = 2. . yk )h. y1 ) − f (t. y1 ) − f (t. Y0 ) Y2 = Y1 + hf (t1 . That is. .33. 3. 2. yk+1 = y(tk+1 ) ≈ y(tk ) + y ′ (tk )h = y(tk ) + f (tk . Y0 = y(t0 ) Y1 = Y0 + hf (t0 . This gives us Euler’s method.

1337 5 0.9 2.0 3.2033 0. we could add more terms of Taylor series.8159 3.34: Error bound and actual error We can now compare the estimated error from our theorem to the actual error of our example.1301 0.3997 0.5 1. ANALYZING EQUATIONS NUMERICALLY 43 k tk Yk yk = y(tk ) |yk − Yk | Estimated Error 0 0. These methods. indepen- dently invented several numerical algorithms to solve differential equations.0764 0. known as Runge-Kutta methods.7 2.4671 Table 1. .6 1.1 1. our approximation becomes 1 y(t1 ) ≈ y0 + f (t0 . although this is less of a problem with the availability of computer algebra systems such as Sage. y1 ) − f (t. we must find ∂f /∂t and ∂f /∂y. we can bound the error by hM L(ti −a) |y(ti ) − Yi | ≤ [e − 1] = 0. y) to obtain ∂f dt ∂f dy y ′′ = + = ft + f fy .1092 0.4.9431 2.0 1. Thus.3 1.3968 10 1. We first need to determine M and L.0000 0. 2 However. Using the chain rule.1875 3. y0 )fy (t0 .0442 0.1e(eti − 1) 2L for h = 0.4.2235 7 0. y2 )| = |(y1 + t) − (y2 + t)| = |y1 − y2 |.0286 2 0. before we can write a program to compute our solution. Since y ′′ = 2et .1103 0.4872 2.2491 0.1. Our results are in Table 1. estimate the higher-order terms of the Taylor series to find an approximation that does not depend on computing derivatives of f (t.1763 6 0.4 1. we can bound y ′′ on the interval [0.0602 3 0.1011 0.0951 4 0.5282 1. 1] by M = 2e.1000 1. Since |f (t.3620 1.7210 1.34. two German mathematicians.1639 0. y0 )) h2 . we can take L to be one. 1.2 1. Carle Runge and Martin Kutta. y).0000 1.0103 0.0000 0. we need to know y ′′ (t0 ) in order to use this approximation. 2 The problem is that some preliminary analytic work must be done. we can obtain a more accurate approximation by using a quadratic Taylor polynomial.3275 0.2200 1. y0 ) + f (t0 .7974 0.3331 9 0. y0 )h + (ft (t0 . y0 )h + h .1974 2. y ′′ (t0 ) 2 y(t1 ) ≈ y0 + f (t0 .0377 0.0228 0.8 2.1. we can differentiate both sides of y ′ = f (t.0554 0.3 Improving Euler’s Method If we wish to improve upon Euler’s method.6511 0.0000 1 0. For example.5836 0. ∂t dt ∂y dt Thus.2428 0. Around 1900.2756 8 0. That is.4366 0.

y) + k f (x. y0 )h)) . we will need to use the Taylor approximation for a function of two variables. y) + h f (x. y) 3! ∂ x ∂ x∂y ) ∂3 3 ∂ 3 + hk 2 f (x. y(s)) ds t0 or ∫ t1 y1 − y0 = y(t1 ) − y(t0 ) = f (s. Then ∂ ∂ f (x + h. In terms of the definite integral. y) 2! ∂ x ∂x∂y ∂ y ( 3 3 1 ∂ ∂ + h3 3 f (x. A FIRST LOOK AT DIFFERENTIAL EQUATIONS If we consider the initial value problem y ′ = f (t. 2 Thus. y) + 3h2 k 2 f (x. ∫ t1 y(t1 ) − y(t0 ) = f (s.11) by y1 − y0 = f (t0 . y) + k 2 f (x. y(s)) ds t0 h ≈ (f (t0 . then ∫ t1 y(t1 ) = y(t0 ) + f (s. y) n! ∂x ∂y n=0 . y + k) = f (x. y). our algorithm becomes h y1 = y0 + (f (t0 . y0 ) + f (t1 . we approximate the right- hand side of (1. y) + hk f (x. Let us assume that f (x. y(t0 ) = y0 . ∂x∂ 2 y ∂3y As in the case of the single variable Taylor series. y1 )) . y1 )) . y0 ) + f (t1 . we can write a Taylor polynomial if the Taylor series is truncated. ∑ N ( ) 1 ∂ ∂ n f (x + h. y) + k f (x. y0 ) + f (t1 . 2 To understand that the second-order Runge-Kutta method is actually an improvement over the traditional Euler’s method. we have a problem since y1 appears in the right-hand side of our approximation. To get around this difficulty. this is simply a left-hand sum.12) with the Euler approximation for y1 .12) 2 However.11) t0 by the Fundamental Theorem of Calculus.11) using the trapezoid rule from calculus. y0 + f (t0 . In the improved Euler’s method or the second-order Runge-Kutta method we will estimate the right-hand side of (1. y) is defined on some rectangle and that all of the derivatives of f are continuously differentiable. h y1 = y0 + (f (t0 . Thus. y + k) = h +k f (x. y) ∂x ∂y ( 2 ) 1 2 ∂ ∂2 2 ∂ 2 + h 2 f (x. y(s)) ds (1. we will replace y1 in the right-hand side of (1. In Euler’s method.44 CHAPTER 1. y0 )h. y) + ··· . (1.

y) F2 = hf (t + αh. ANALYZING EQUATIONS NUMERICALLY 45 ( ) 1 ∂ ∂ N +1 + h +k f (x. If we use the two variable Taylor series to expand f (t + αh. where O(h2 ) means that of the subsequent terms have a factor of hn with n ≥ 2. 2 2 . y)). and β as accurately as possible in order to duplicate as many terms as possible in the Taylor series h2 ′′ h3 y(t + h) = y(t) + hy ′ (t) + y (t) + y ′′ (t) + · · · . (N + 1)! ∂x ∂y where the second term is the remainder term and (x.14) as h2 y(t + h) = y(t) + hf + (ft + f fy ) + O(h3 ).15) and ((1. y + hf (t. y(t + h) = y(t) + (w1 + w2 )hf + αw2 h2 ft + βw2 h2 f fy + O(h3 ). these equations are satisfied and we obtain the improved Euler’s method h h y(t + h) = y(t) + f (t.13). (1. we obtain a new form for (1. y(t + h) = y(t) + w1 hf (t. y) lies on the line segment joining (x. (1. y) + w2 hf (t + αh. w2 . y) and (x + h. y + βF1 ).16) 2! We can make equations (1.14) agree if we choose w1 = 1 and w2 = 0. y + k). 2 If we choose α = β = 1 and w1 = w2 = 1/2. we obtain Euler’s method. we have f (t + αh. where F1 = hf (t.1. That is. 2 1 βw2 = . y + βhf ) = f + αhft + βhf fy + O(h2 ). we adopt a formula y(t + h) = y(t) + w1 F1 + w2 F2 . If we are more careful about choosing our parameters. (1. y + βhf ). we can rewrite (1.13)) and (1. (1. y)). 1 αw2 = . we can obtain agreement up through the h2 term. h) + f (t + h. Since y ′ = f . Using this expression. y).14) 2! 3! We can make equations ((1.13) The idea is to choose the constants w1 . y + βhf (t.16)) agree up through the quadratic terms if we require that w1 + w2 = 1. In the Improved Euler’s Method.15) Since y ′′ = ft + fy f by the chain rule. α.4.

2! If we use Simpson’s rule to estimate the integral in ∫ t1 y(t1 ) − y(t0 ) = f (s. Since y(t + h) = y(t) + hy ′ (t) + O(h2 ). our approximations should be Y0 = α. y + F1 2 2 ( ) 1 1 F3 = hf t + h. y + F3 ). y). since h2 ′′ y(t + h) = y(t) + hy ′ (t) + y (t) + O(h3 ). y(s)) ds t0 we can improve our accuracy up to h4 . This method is known as the Runge-Kutta method of order 4 and is given by 1 y(t + h) = y(t) + (F1 + 2F2 + 2F3 + F4 ). . Y2 = Y1 + hf (t1 . we know that the error depends on h. . tN .. Yk ). On the other hand. Euler’s method is based on truncating the Taylor series after the linear term. YN = YN −1 + hf (tN −1 . . but the algebra become much more tedious. ) . YN −1 ). Y1 = Y0 + hf (α.4. b]. A FIRST LOOK AT DIFFERENTIAL EQUATIONS The improved Euler’s method or the second-order Runge-Kutta method is a more so- phisticated algorithm that is less prone to error due to the step size h. The idea is exactly the same. where h = (b − a)/N and ti = a + ih. . Yk+1 = Yk + hf (tk . y) ( ) 1 1 F2 = hf t + h. the error for the improved Euler’s method depends on h2 . . y + F2 2 2 F4 = hf (t + h.4 Important Lessons • We can use Euler’s method to find an approximate solution to the initial value problem y ′ = f (t. Y1 .46 CHAPTER 1. 1. . 6 where F1 = hf (t. y(a) = α on an interval [a. Y0 ). If we wish to find approximations at N equally spaced points t1 .

h) + f (t + h. Our approximation is more accurate for smaller values of h. x). y + F2 2 2 F4 = hf (t + h. y2 ) are in D = [a. y1 ) − f (t. • Error analysis rate of convergence is very important for any numerical algorithm. then f ′′ (x0 ) f (n) (ξ) f (x) = f (x0 ) + f ′ (x0 )(x − x0 ) + (x − x0 )2 + · · · + (x − x0 )n . 6 where F1 = hf (t. whenever (t. y(a) = α. y2 )| ≤ L|y1 − y2 |. y + F3 ) with the error bound depending on h4 . y)) 2 2 with the error bound depending on h2 . • The condition that there exists a constant L > 0 such that |f (t. The Runge-Kutta methods are an important class of these algorithms. • The improved Euler’s method is given by h h y(t + h) = y(t) + f (t. y + hf (t.1. y + F1 2 2 ( ) 1 1 F3 = hf t + h. Under reasonable conditions we can also bound the error by hM L(ti −a) |y(ti ) − Yi | ≤ [e − 1]. • The Runge-Kutta method of order 4 is given by 1 y(t + h) = y(t) + (F1 + 2F2 + 2F3 + F4 ). ANALYZING EQUATIONS NUMERICALLY 47 • Taylor’s Theorem theorem is a very useful tool for studying differential equations.4. we can develop better numerical algorithms to compute solutions of differential equations. b] × R is called a Lipschitz condition. y). . 2! n! where ξ ∈ (x0 . If x > x0 . 2L where y is the unique solution to the initial value problem y ′ = f (t. y1 ) and (t. • Using Taylor series. y) ( ) 1 1 F2 = hf t + h.

s = hL. and t = M h2 /2 and apply part (2) to show ( ) M h2 M h2 |yi+1 − Yi+1 | ≤ e (1+i)hL |y0 − Y0 | + − . Use Euler’s method and Taylor’s Theorem to show that h2 ′′ |yi+1 − Yi+1 | ≤ |yi − Yi | + h|f (ti .48 CHAPTER 1. .5 to approximate the solution to the initial value problem on the interval [0. . k. 1.33). yi ) − f (ti . Consider the differential equation dy = 3y − 1 dt with initial value y(0) = 2. (a) Use Taylor’s Theorem to show that for all x ≥ −1 and any positive m.33) to estimate the error at each approxi- mation. Make sure that your approximations are accurate to four decimal places. . . (b) Use part (1) and geometric series to prove the following statement: If s and t are positive real numbers. . . 2 (e) Let aj = |yj − Yj | for j = 0. . the error estimates.5 Exercises 1. . . (c) Sketch the graph of the approximate and exact solutions. (d) Use the error bound theorem (Theorem 1. Make sure that your approximations are accurate to four decimal places. . s s (c) When i = 0. 0 ≤ (1 + x)m ≤ emx . then ( ) t t ai+1 ≤ e (i+1)s + a0 − .4. Your solution should include a table of approximate values of the dependent variable the exact values of the dependent variable. (a) Find the exact solution of the initial value problem. y(t0 ) = Y0 = α and the theorem is true. 2hL 2hL and derive that M h ( (ti+1 −a)L ) |yi+1 − Yi+1 | ≤ e −1 2L for each i = 0. Yi )| + |y (ξi )| 2 where ξi ∈ (ti . 2. 1. In this series of exercises. (b) Use Euler’s method with step size h = 0. 2] Your solution should include a table of approximate values of the dependent variable as well as the exact values of the dependent variable. . N . 1. 2. . (d) Show that M h2 |yi+1 − Yi+1 | ≤ |yi − Yi |(1 + hL) + . and {ai }ki=0 is a sequence satisfying t a0 ≥ − s ai+1 ≤ (1 + s)ai + t. and the actual error. we will prove the error bound theorem for Euler’s method (Theorem 1. Hint. N − 1. ti+1 ) and yk = y(tk ). A FIRST LOOK AT DIFFERENTIAL EQUATIONS 1. . for i = 0.

s s 1. y = function ( ' y ' )(x) de = diff (y . y(0) = 1. We can use Sage to find an algebraic solution. y) f (x0 ) = y0 with step size h on the interval [x0 . 5) We can also use Euler’s method to find a solution for our initial value problem. y .y = PolynomialRing (RR . Hints for part (2): • For fixed i show that ai+1 ≤ (1 + s)ai + t ≤ (1 + s)[(1 + s)ai−1 + t] + t ≤ (1 + s){(1 + s)[(1 + s)ai−2 + t] + t} + t . Suppose we wish to solve the initial value problem dy/dx = x + y. y as generators of a polynomial ring eulers_method (x + y .5 . " xy " ) .x) == x + y h = desolve (de . x1 ] is eulers_method(f. Sage has some very good numerical solvers. 2.4. ≤ (1 + s)i+1 a0 + [1 + (1 + s) + (1 + s)2 + · · · + (1 + s)i ]t. y). If we encounter an equation that we cannot solve or use Sage to solve. x . x0. . 0 .6 Project—Numerical Routines for solving ODEs Not all differential equations can be solved using algebra and calculus even if we are very clever.1. However. • Now use a geometric sum to show that ( ) t t t ai+1 ≤ (1 + s) i+1 a0 + [(1 + s)i+1 − 1] = (1 + s)i+1 + a0 − . 0.1 . including an initial condition • A Sage command to solve the equation. which are best implemented using a computer. we must resort to numerical algorithms like Euler’s method or one of the Runge-Kutta methods. ANALYZING EQUATIONS NUMERICALLY 49 Hint. x1) Notice that we obtained a table of values.. s s s • Apply part (1) to derive ( ) t t ai+1 ≤ e (i+1)s + a0 − . y0. we can use the line command from Sage to plot the values (x. . 0. Fortunately.1]) show ( h) plot (h . h. gens () # declare x . Sage will need to know the following to solve a differential equation: • An abstract function • A differential equation. x . ics = [0 . 5) The syntax of eulers_method for the inital value problem y = f (x.4.

17).5. ore might be pulverized and then processed using chemicals to remove a certain mineral.5) Again. Soft rock mining operations such as coal mining or extracting oil from tar sands might use solvents or water to extract any commodity of value. which implements the fourth order Runge-Kutta method. coal. or oil is extracted can often present huge environmental challenges. algorithm = " none " ) p = list_plot ( pts . x . end_points = 5 . and the method can generate errors if we are not careful. ics =[0 . Not only is desolve_rk4 more accurate. the fact that can obtain such a solution analytically will prove very useful when we investigate more complicated equations and systems of equations. A FIRST LOOK AT DIFFERENTIAL EQUATIONS x . We have to use a very small step size to get good accuracy. output = ' slope_field ' . For example. 1. we shall find that we can always find an algebraic solution to a first-order linear differential equation. eulers_method is not very sophisticated. step = 0. especially if water is used in the mining operation. however. Moreover.1 Mine Tailings In any mining operation. Fortunately. However.y = var ( ' x␣y ' ) p = desolve_rk4 (x + y . 1. 4.5 First-Order Linear Equations First-order differential equations occur naturally in many modeling problems. 1/2 . we just get a list of points. in a hard rock mining operation. " xy " ) . We could. end_points = [0 . x .50 CHAPTER 1. gens () pts = eulers_method (x + y . of course. (1. ivar = x .5 . .y = PolynomialRing (RR . y .1] . desolve_rk4 has some nice built-in graphing utilities.17) dt This equation will not be separable if p(t) in not a constant. 0 . ics = [0 . These are equations of the form dy + p(t)y = g(t).html. We shall have to find a new approach to solving such an equation. thickness = 2 . it is much easier to use. 1] .sagemath.y = var ( ' x␣y ' ) desolve_rk4 (x + y .org/doc/reference/calculus/sage/calculus/desolvers. For more information. 5] . Sage has much better algorithms for solving initial value problems. color = ' red ' ) p Write the Sage commands to compare the graphs obtained using eulers_method and desolve with the exact solution. tailings are what is left after everything of value has been extracted. The material that is left over after the minerals. y . 2. There are different ways of processing mine tailings. see http://www. 1 . color =" red " ) p += line ( pts ) p As we pointed out. use a numerical algorithm to solve (1. One such algorithm is desolve_rk4. but one way is to store them in a pond.

dx = rate in − rate out. Suppose that 1500 cubic meters per day flow into the pond from stream and 1500 cubic meters are pumped from the pond each day to be processed and recycled.1. and our tailings pond will become full after 400 days of operation. the water must be treated before it is released into the watershed. when 0 ≤ t < 400. it is not even separable. the differential equation that models the amount of chemical in the tailings pond at time t is dx 20x =5− . The pond has a stream flowing into it. the rate at which the chemicals leave the pond will depend on the amount of chemicals in the pond at time t. Chemicals are used in processing gold ore. Notice that equation (1. Thus. dt 400 − t If we multiply both sides of this equation by (400 − t)−20 . Thus. We will assume that water in our tailings pond is well mixed so that the concentration of chemicals through out the pond is fairly uniform. (1. 20000 − 50t 400 − t Hence. These chemicals such as sodium cyanide are often highly poisonous and dangerous to the environment. that is. We will have to use a different approach to find a solution. In addition. Let x(t) be the amount of chemicals in the pond at time t. On the other hand. The water can then be treated and recycled. (1. and water is also pumped out of the pond. the water from stream becomes contaminated with chemicals from the mining operation. the concentration of chemicals in the pond at time t is x/(20000 − 50t). and at time t it is V (t) = 20000 − 50t. dt Since water flows into the pond from the stream at a rate of 1000 cubic meters per day. When we begin our operation. We wish to find a differential equation that will model the amount of chemicals in the tailings pond at any particular time. The volume of the pond is decreasing due to sediment. the volume of our tailings pond is reduced by 50 cubic meters each day. Then dx/dt is the difference between the rate at which the chemicals are entering the pond and the rate at which the chemicals leave the pond. the rate as which the chemicals enter the pond is 5 kilograms per day. FIRST-ORDER LINEAR EQUATIONS 51 This method allows any particles that are suspended in the water to settle to the bottom of the pond. Suppose that we have a gold mining operation and we are storing our tailings in a pond that has an initial volume of 20. we obtain dx (400 − t)−20 + 20(400 − t)−21 x = 5(400 − t)−20 .18) is not autonomous.000 cubic meters.19) dt . First.18) dt 400 − t Of course. Thus.5. say at a rate of 5 kilograms of chemicals per 1000 cubic meters. any particulate matter pumped into the pond from the stream settles to the bottom of the pond at a rate of 50 cubic meters per day. and the rate at which the chemicals are flowing out of the pond to be recycled is ( ) x 20x 1000 = . since there will only be water in the pond if V (t) = 20000 − 50t ≥ 0. the tailings pond is filled with clean water. We shall assume that the particulate matter and the chemicals are included in the 1000 cubic meters that flow into the pond from the stream each day. In fact. At time t = 0. we will have to cease mining operations once the pond is full. we will rewrite the equation in the form dx 20 + x = 5. the water level of the pond remains constant.

there are no chemicals at t = 400. there are no chemicals in the pond. equation (1. x(t) 80 60 40 20 100 200 300 400 t Figure 1. dt dt Thus. 19 where C is an arbitrary constant.52 CHAPTER 1.35) fits the situation. we obtain 5 x(t) = (400 − t) + C(400 − t)20 . 19 400 The graph of the solution to our differential equation (Figure 1. Eventually. 19 Since x(0) = 0.19) becomes d ( ) (400 − t)−20 x = 5(400 − t)−20 . A FIRST LOOK AT DIFFERENTIAL EQUATIONS We now make the crucial observation that the product rule applies to the lefthand side of our equation. we have ∫ −20 5(400 − t)−19 (400 − t) x=5 (400 − t)−20 dt = + C. the amount of chemicals decreases as the pond begins to fill with sediment. we can quickly determine that C = −(5/19)400−19 and that the solution to our initial value problem is [ ( ) ] 5 400 − t 19 x(t) = (400 − t) 1 − . However. Ini- tially. Solving for x.35: Chemical waste in a tailings pond . dx d ( ) (400 − t)−20 + 20(400 − t)−21 x = (400 − t)−20 x . dt Integrating both sides of this equation. but x(t) quickly increases.

dt dt If we integrate both sides of d ( ) x(t)ept = qept . we obtain d ( −2t ) dx e x = e−2t − 2e−2t x = 3e−2t . we can conclude that C = 5/2.21) Multiplying both sides of equation (1.36.2 First-Order Linear Equations The differential equation dx 20 + x=5 dt 400 − t is an example of a first-order linear differential equation. That is. (1. dt then q x(t)ept = ept + C. If dx + px = q. p If we apply the initial condition.5. (1. Suppose we wish to solve the initial value problem dx − 2x = 3. FIRST-ORDER LINEAR EQUATIONS 53 1. gives us the following ∫ 3 e x = 3 e−2t dt + C = − e−2t + C. dx d ( ) ept + ept px = x(t)ept . dt The left-hand side of the equation is exact.5.20) by e−2t . 2 2 . a first-order linear differential equation is an equation that can be written in the form dx + p(t)x = g(t). More specifically. −2t 2 Applying the initial condition x(0) = 1. then we first multiply both sides of the equation by ept to obtain dx ept+ ept px = qept .20) dt x(0) = 1. p p Example 1. we can determine C. and 5 3 x(t) = e2t − . p Thus. dt Let us first show how to solve first-order linear equations when the coefficient functions are constant.1. the solution that we seek is ( ) q q x(t) = + x0 − ep(t0 −t) . dt dt Integrating both sides of this last equation. ( ) q C = x0 − ept0 . dt x(t0 ) = x0 .

37 will always work.x ^2) Surprisingly. 2 2 2 2 dx Integrating both sides. 2 Using the initial condition to solve for C.22) If we choose P (x) such that P ′ (x) = p(x) and multiply both sides of the equation by eP (x) .38.54 CHAPTER 1. 2 2 where y(0) = 1. Thus. x = var ( ' x ' ) y = function ( ' y ' )(x) de = diff (y . the strategy in Example 1. We will multiply both sides of the equation by ex and use the product rule to obtain d ( x2 ) e y = ex y ′ + 2xex y = ex e−x = 1. we find that C = 1 and y = (x + 1)e−x . We say that ∫ (∫ ) p(x) dx µ(x) = e = exp p(x) dx is an integrating factor for the differential equation (1. Now let us now solve a first-order linear differential equation where the coefficients are not constant. Example 1.37.22). Integrating both sides. we get ∫ 2 ex y = dx + C. then ( )′ eP (x) y = eP (x) (y ′ (x) + p(x)y(x) = eP (x) g(x). x) + 2* x*y == exp ( . y . Suppose that y ′ + 2xy = e−x . the general solution is y = (x + C)e−x . Suppose that y ′ + p(x)y = g(x).x ^2) solution = desolve (de . ics = [0 . . A FIRST LOOK AT DIFFERENTIAL EQUATIONS Example 1. 1]) solution (x + 1) *e ^( . (1. Consider the initial value problem dy 2 − y = 2x2 dx x y(−2) = 4. 2 We can use Sage to check our solution. ∫ P (x) e y= eP (x) g(x) dx + C or (∫ ) 1 y= eP (x) g(x) dx + C . eP (x) ∫ The Fundamental Theorem of Calculus tells us that P (x) = p(x) dx.

(1. To construct our model.5. dx dx We can now integrate this equation to get x−2 y = 2x + C. Suppose that a 100-gallon tank initially contains 50 gallons of salt water containing five pounds of salt.3 Mixing Models Many applications involve the mixing of two or more substances together. x(t). how various ingredients are mixed together in a brewery.5.23) dt 50 + t . the solution to our initial value problem is y = 2x3 + 5x2 . We have the initial condition x(0) = 5. or how greenhouse gases mix move across various layers of the earth’s atmosphere. The initial condition y(−2) = 4 allows us to find C = 5. V (t) in flow | {z } out flow where V (t) is the volume at time t. We can model how petroleum products are mixed together in a refinery. we obtain ( ) −2 dy 2 x − y =2 dx x or d ( −2 ) dy x y = x−2 − 2x−3 y = 2. we will let t be the time (measured in minutes) and set up a differential equation that will measure how fast the amount of salt at time t.1. A well mixed solution leaves the tank at rate of 4 gallons per minute. We wish to know how much salt is in the tank. since the tank starts with 20 gallons and four gallons are pumped into the tank per minute while three gallons leaves the tank during the same time interval. x Multiplying both of our differential equation by µ(x) = x−2 . is changing. when the tank is full. The expression x/V (t) is the amount of salt in one gallon at time t. 1. and dx = amount of salt flowing in − amount of salt flowing out dt x 5 −4 = |{z} . our differential equation becomes dx 4 =5− x. We have V (t) = 50 + t. Therefore. FIRST-ORDER LINEAR EQUATIONS 55 Our integrating factor is ( ∫ ) 2 −2 µ(x) = exp − dx = e−2 ln x = eln x = x−2 . dt 50 + t Our equation is linear since we can rewrite it as dx 4 + x = 5. Thus. A brine mixture containing one pound of salt per gallon flows into the top of the tank at a rate of 5 gallons per minute.

dt Integrating both sides and simplifying gives us C x = t + 50 + . P (t + ∆t) − P (t) P ′ (t) = lim = rP − W. A FIRST LOOK AT DIFFERENTIAL EQUATIONS An integrating factor for this differential equation is (∫ ) 4 µ(t) = exp dt = e4 ln(50+t) = (50 + t)4 . t = 50 and the amount of salt in the tank is x(50) = 97. when the tank is full. 50 + t Therefore. (t + 50)4 Our initial condition. 250. 000 and 31250000 x(t) = t + 50 − .23) by µ(t).4 Finance Models There are a number of problems in finance that can be modeled using differential equations.188 pounds.188 1. dt We can now apply the product rule to obtain d [(50 + t)4 x] = 5(50 + t)4 . The net increase in the balance between times t and t + ∆t can now be described as P (t + ∆t) − P (t) ≈ rP ∆t − W ∆t Thus. Suppose that we also allow withdrawals of W dollars per year. we get dx (50 + t)4 + 4(t + 50)3 x = 5(50 + t)4 . x(0) = 5 tells us that C = 31. 5]) solution (50) . x .5. ics =[0 .56 CHAPTER 1. t = var ( ' t ' ) x = function ( ' x ' )(t) de = diff (x . We can use Sage to easily check the solution of our initial value problem. if we multiply both sides of equation (1. t) + 4* x /(50 + t ) == 5 solution (t) = desolve (de . . ∆t→0 ∆t We can solve the equation P ′ = rP − W by multiplying both sides of the equation by the integrating factor ( ∫ ) µ(t) = exp − r dt = e−rt . Let P (t) be the balance of an account at time t and suppose that the account pays interest at a rate of r percent per year compounded continuously. (t + 50)4 Thus.n( digits = 5) 97.

r If we know the initial balance in the account. we have W −rt e−rt P = e +C r or W P = + Cert . then ( ) d dy (µ(t)y) = µ(t) + p(t)y = µ(t)g(t). dt dt Integrating both sides of this equation and solving for y.39. dt Integrating both sides of this equation.5. say P (0) = P0 . If y ′ + p(t)y = g(t) is a differential equation such that y(t0 ) = y0 . FIRST-ORDER LINEAR EQUATIONS 57 to obtain d −rt [e P ] = −W e−rt .5 Existence and Uniqueness of Solutions Several questions about the existence and uniqueness of solutions to first order linear dif- ferential equations now arise. the solution to our initial value problem is ( ) W W P = + P0 − ert . then there exists a unique function y = ϕ(t) satisfying the differential equation and the initial condition on I. r r 1. Proof. If (∫ ) µ(t) = exp p(t) dt . we have (∫ ) 1 y= µ(t)g(t) dt + C . • Does an initial-value problem always have a solution? • Is the solution unique? • Is the solution globally defined or does it only hold for a small interval? We can use the following theorem to answer these questions.5. the constant C is uniquely determined. Notice that we have used continuity to guarantee that the integrals exist. . Theorem 1. and p(t) and g(t) are continuous on the open interval I = (α. we can determine C. µ(t) Since y(t0 ) = y0 . W P0 = +C r or W C = P0 − . That is.1. β). r Thus.

7 Exercises 1.5.5. y = 6(x + 1)2 3. ∫ P (x) e y= eP (x) g(x) dx or ∫ 1 y= eP (x) g(x) dx. then there exists a unique function y = ϕ(t) satisfying the differential equation and the initial condition on I. β). dx x x 2. 3x4 + 8x3 + 6x2 + 12 Hint.6 Important Lessons • Any first order linear differential equation y ′ + p(x)y = g(x) can be solved by multiplying each side of the equation by an integrating factor ∫ p(x) dx µ(s) = e .58 CHAPTER 1. 1. Hint. Solve the initial value problem 2y y′ = − + 2x x+1 y(0) = 2. eP (x) • If y ′ + p(t)y = g(t) is a differential equation such that y(t0 ) = y0 . A FIRST LOOK AT DIFFERENTIAL EQUATIONS 1. Solve the differential equation ( ) dy 2 sin x + y= 2 . Solve the initial value problem 2y y′ = − + ex 1+x y(0) = 1. y = (ex + 1)/(x + 1)2 . and p(t) and g(t) are continuous on the open interval I = (α. Integrating both sides. In this case. we get ( )′ eP (x) y = eP (x) (y ′ (x) + p(x)y(x) = eP (x) g(x).

we can determine that C = 1200 or t2 + 240t + 2400 x(t) = . Hint. and the tank contains x(80) = 70 kilograms of salt when the tank is full.5. y = sin x/(x + 2) 5. Supposed that water containing 0. FIRST-ORDER LINEAR EQUATIONS 59 4. we obtain t2 (120 + t)x = 120t + + C. How much salt is in the tank when the tank is full? Hint. and 5 liters per minute are removed from the bottom of the tank. dx = rate in − rate out dt x = 10(0. 2 Using the intial condition x(0) = 10. The volume of the tank is V = 600 + 5t. we have ( ) d dx dx 1 [(120 + t)x] = (120 + t) + x = (120 + t) + x = 120 + t. 120 + t Multiplying both sides of the differential equation by µ(t). A 600-gallon tank initially contains 200 liters of water containing 10 kilograms of salt.1) − 5 V x =1−5 600 + 5t x =1− . An integrating factor for this equation is given by (∫ ) 1 µ(t) = exp dt = 120 + t. Solve the initial value problem y cos x y′ = − + x+2 x+2 y(0) = 1.1 kilograms of salt flows into the top of the tank at a rate of 10 liters per minute. y = ex sec x 6. If x(t) is the amount of salt in the tank at time t. . 2t + 240 The tank is full at time t = 400/5 = 80.1. The water in the tank is kept well mixed. dt dt dt 120 + t Integrating both sides of this equation. we know that x(0) = 10. Hint. 120 + t The resulting equation dx 1 + x=1 dt 120 + t is a first order linear differential equation. We can model the amount of salt in the tank at time t with a differential equation. Solve the initial value problem ex y ′ = y tan x + cos x y(0) = 1.

If x(t) is the amount of solute at time t in Lake Baikal. Solve this equation 7 Levy. she retires and begins and begins withdrawing from the retirement fund at a rate of $3500 per month.5 meters below sea level. Hypothetically. The mill was constructed in 1966 on the shoreline of Lake Baikal and regularly discharged waste into the lake. Estimate ∆x during the time interval [t. Lake Baikal. Assume that C = Cin is the pollutant concentration flowing into Lake Baikal and Cout is the concentration of the outflow into the Angara River (measured in metric tons per cubic km). The fund grows at a rate of 3.60 CHAPTER 1. Clifford J. The plant was closed in November 2008 due to unprofitability. Lake Baikal’s waters are well-mixed and well-oxygenated in spite of its great depth.642 meters. A FIRST LOOK AT DIFFERENTIAL EQUATIONS 7. the Angara River.600 cubic kilometers and contains 20% of the world’s unfrozen fresh water. After 35 years. The lake has a depth of 1.100 cubic kilometers.5% per year compounded continuously. (b) From your estimate of ∆x in part (1). 8. One of the major polluters has been Baykalsk Pulp and Paper Mill. (November 8. (c) The equation that you found in part (2) is a first-order linear equation.7 Suppose that we wish to understand how the pollution level changes in Lake Baikal over a period of years. located in southern Siberia. The mill underwent a final bankruptcy in September 2013. .” The New York Times. If she does not make any deposits after she retires. Retrieved March 14. Let us assume that the lake is currently 6 times more polluted than these other sources of contaminants. and the bottom lies 1.html. 2014 from http://www. We wish to know how long it will take for the pollution level to reduce to half of its current level of lake.186. Pollution is an increasing concern in Lake Baikal. however. “Last Gasp for Factory Bequeathed by Soviets. but production resumed in January 2010. how long will her retirement fund last? [Hint: Divide the problem into two smaller problems—one that deals with the situation before retirement and one that deals with the problem after retirement. and rin = rout = r = 60 km3 /year be the rates of inflow from the numerous rivers that feed the lake and outflow to the Angara River. Lake Baikal has a volume of 23. In contrast. it is the world’s deepest lake. Although 544 rivers flow into Lake Baikal. the largest of the Great Lakes. Lake Superior.nytimes. 2010). 600 km3 . A manager in a communications company contributes $2400 per year into her retirement fund by making many small deposits throughout the year. so we can model this situation as a simple mixing problem.com/2010/11/09/world/europe/ 09baikal. The outflow of the lake is pretty constant at 60. (a) The volume of water in Lake Baikal is V = 23. where ∆t > 0 is small. but the longterm fate of the mill is still undecided. let us assume that the Baykalsk Pulp and Paper Mill has been responsible for most of the pollution in Lake Baikal for the last several decades.] Hint. Suppose that at t = 0 years the mill ceases operation and there are no longer any pollutants discharged into the lake from the mill although there are still other sources of pollution. then x0 = x(0) = 6CV is the initial amount of solute in the lake.4 cubic kilometers per year. is only the seventh largest lake in the world in terms of surface area. has a volume of only 12. write an initial value problem that describes the amount of pollutants in the lake at time t. there is only a single outlet. t + ∆t].

8 Polking p.5. (1. which we will revisit when we study second order linear differential equations. Bernoulli’s equation is an equation of the form y ′ = a(t)y + f (t)y n . show that the solution is [ ∫ ] y = A exp − p(t) dt . 9.1). Variation of Parameters. predict how many years it will take to reduce the pollution in Lake Baikal to half of its current level. Verify that the solution obtain in this manner agrees with the solution given in the proof of Theorem 1. 63 . show that we can transform Bernoulli’s equation into the linear equation z ′ = (1 − n)a(t)z + (1 − n)f (t).2) and deter- mine y. Then substitute for A(t) in equation (A. y ′ + p(t)y = g(t). (b) Solve the equation xy ′ + y = x4 y 3 . Consider the following method of solving the general linear equation of the first order.26) (c) Find A(t) from Equation (A. where A is a constant. Hint. (b) If g(t) is not identically zero. show that this solution is equivalent to the solution (∫ ) 1 y= µ(t)g(t) dt + C .1.8 (a) Using the substitution z = y 1−n . Hint. where n ̸= 0 or 1. (1. FIRST-ORDER LINEAR EQUATIONS 61 (d) Using part (3). This technique is know as variation of parameters.39. (1. 10.3).24) (a) If g(t) is identically zero.25) where A is now a function of t. That is. µ(t) where (∫ ) µ(t) = exp p(t) dt . Hint. assume that the solution is of the form [ ∫ ] y = A(t) exp − p(t) dt . Bernoulli’s equation is nonlinear and cannot be solved by the techniques that we have used to solve first order linear equations. By substituting for y in the given differential equation (A. show that A(t) must satisfy the condition [∫ ] ′ A (t) = g(t) exp p(t) dt .

If one solution y1 (t) of the Ricatti equation is known. v2 v v v which is just the first-order linear equation v ′ + [q(t) + 2r(t)y1 (t)]v = −r(t). v v v Therefore. (c) Find the solution to the Ricatti equation y ′ = cos t − y tan t + y 2 sec t given the particular solution y1 = sin t. then a more general solution containing an arbitrary constant can be found by substituting y = y1 (t) + 1/v(t) into equation (A. (d) Find the solution to the Ricatti equation y ′ = 2 − 3y + y 2 given the particular solution y1 ≡ 2.4) to find a first-order linear equation in v and t. Hint. ( ) ( ) ′ 1 1 2 y = p + q y1 + + r y1 + v v q 2ry 1 r = p + qy1 + + ry12 + + 2 v v v ′ q 2ry1 r = y1 + + + 2. Substituting into our original equation.62 CHAPTER 1. we obtain v′ v′ y ′ = y1′ − = p + qy1 + ry 2 1 − . v2 v2 On the other hand. which we can solve to find a general solution to the Ricatti equation.27) is known as the Ricatti equation and has some useful applications in control theory. then y ′ = y1′ − v ′ /v 2 . (b) 1 y =t+ C −t (c) 1 y(t) = + sin t C cos t − sin t . (a) Show that this first-order linear equation is v ′ + [q(t) + 2r(t)y1 (t)]v = −r(t). A FIRST LOOK AT DIFFERENTIAL EQUATIONS 11. v′ q 2ry1 r − = + + 2. The first-order nonlinear differential equation y ′ = p(t) + q(t)y + r(t)y 2 (1. (b) Find the solution to the Ricatti equation 1 1 y′ = − − y + y2 t2 t given the particular solution y1 = 1/t. (a) If y = y1 + 1/v.

EXISTENCE AND UNIQUENESS OF SOLUTIONS 63 (d) 1 y(t) = 2 + Cet −1 12. 1.5).5) is autonomous. Let us examine some consequences of the existence and uniqueness of solutions. Find this solution. The existence and uniqueness of solu- tions will prove to be very important—even when we consider applications of differential equations. there exists a unique solution u = u(t) for x′ = f (t. In other words.40 (Existence and Uniqueness Theorem). If the population drops too low. we can use the method of the previous problem to find a general solution to (A. x) : 0 ≤ |t − t0 | ≤ a.4). (1. Theorem 1. 1. we have already shown that a solution exists and is unique.1. (c) Equation (A. whales might not be able to find suitable mates and the species will eventually go extinct.28) dt N where P is the population of the whales at time t and N is the carrying capacity. x) and x(t0 ) = x0 is a linear differential equation. a species now very much on the endangered list. the case of the North Pacific right whale. Let x′ = f (t. (a) Find the equilibrium solutions of the this equation. we can find a solution using separation of vari- ables.1 The Existence and Uniqueness Theorem The following theorem tells us that solutions to first-order differential equations exist and are unique under certain reasonable conditions.8 Project—Modeling 1. If f and ∂f /∂x are continuous functions on the rectangle R = {(t.6 Existence and Uniqueness of Solutions If x′ = f (t. We will now take up the question of existence and uniqueness of solutions for all first-order differential equations. 0 ≤ |x − x0 | ≤ b} .5) is also a Ricatti equation (A. . (b) Since equation (A. the population will die out if it drops below a certain threshold.5. Since we know an equilibrium solution from part (1). We can model this with the following equation. For example. ( ) dP P =k 1− (P − aN ).6.6. x) and x(t0 ) = x0 on some interval |t − t0 | < h contained in the interval |t − t0 | < a. Find the general solution using the fact that we have a Ricatti equation and show that your solution agrees with the solution that you found in part (2) Hint. The constants k and a are positive with a < 1. x) have the initial condition x(t0 ) = x0 . Suppose that we have a population that not only grows logistically but also requires a minimum threshold population to survive.

Thus.42. a unique solution exists near t = 0. Suppose that y ′ = y 2 with y(0) = 1. Separating the variables. In the case that y(0) = −1. . 1−t Therefore. Example 1.43 (Solutions Curves Cannot Cross). The Existence and Uniqueness Theorem tells us that the integral curves of any differential equation satisfying the appropriate hy- pothesis. ∞). Separating the variables. the initial condition is satisfied and ( ) 2 3/2 y= t 3 is a solution for t ≥ 0. the solution is 1 y=− . Remark 1. there is no equilibrium solution at y = 0. y2 we see that 1 y=− t+C or 1 y= . 1) if y(0) = −1. we could take the point of intersection as the initial value for the differential equation. 3 If C = 0.41. Solutions are only guaranteed to exist on an open interval containing the initial value and are very dependent on the initial condition.64 CHAPTER 1. cannot cross. Since f (t. This is especially troubling if we are looking for equilibrium solutions. Consider the initial value problem y ′ = y 1/3 with y(0) = 0 and t ≥ 0. A FIRST LOOK AT DIFFERENTIAL EQUATIONS Example 1. 3 2/3 y =t+C 2 or ( )3/2 2 y= (t + C) . t+1 and a solution exists on (−1. The problem is that ∂ 1/3 1 −2/3 y = y ∂y 3 is not defined at y = 0. In this case. However. 3 y ≡ 0. we would no longer guaranteed unique solutions to a differential equation. a solution also exists on (−∞. Although y ′ = y 1/3 is an autonomous differential equation. we can find two additional solutions for t ≥ 0: ( )3/2 2 y=− t . y) = y 2 and ∂f /∂y = 2y are continuous everywhere. If the curves did cross. 1 dy = dt. y −1/3 y ′ = dt.

u(t)) is also continuous on I. the function u is a solution of the integral equation. t0 If we differentiate both sides of this equation. on some interval I containing t0 . we obtain ∫ t ∫ t u(t) − u(t0 ) = u′ (s) ds = f (s. we obtain u′ (t) = f (t. Theorem 1.44. that will converge to a function u(t) that is a solution to the integral equation ∫ t x(t) = x0 + f (s. u(s)) ds. .2 Picard Iteration It was Emile Picard (1856–1941) who developed the method of successive approximations to show the existence of solutions of ordinary differential equations. x(s)) ds. EXISTENCE AND UNIQUENESS OF SOLUTIONS 65 1. u(t)) and applying the Fundamental Theorem of Calculus. t0 the initial condition is fulfilled. x) x(t0 ) = x0 . t0 We define the first function of the sequence using the initial condition. {un (t)}.1. if and only if u is a solution to the integral equation ∫ t x(t) = x0 + f (s. t0 Proof. Integrating both sides of u′ (t) = f (t. One of the first steps towards understanding Picard iteration is to realize that an initial value problem can be recast in terms of an integral equation. To show the existence of a solution to the initial value problem x′ = f (t. we will construct a sequence of functions. assume that ∫ t u(t) = x0 + f (s. He proved that it is possible to construct a sequence of functions that converges to a solution of the differential equation. u(t)). The function u = u(t) is a solution to the initial value problem x′ = f (t.6. x) x(t0 ) = x0 . u0 (t) = x0 . the function F (t) = f (t.6. x) x(t0 ) = x0 . u(s)) ds = x0 . Conversely. Since u is continuous on I and f is continuous on R. x(s)) ds. u(s)) ds t0 t0 Since u(t0 ) = x0 . Suppose that u = u(t) is a solution to x′ = f (t. Since ∫ t0 u(t0 ) = x0 + f (s.

∫ t un+1 = x0 + f (s. ∫ t u1 (t) = 1 + ku0 (s) ds 0 ∫ t =1+ k ds 0 = 1 + kt. Example 1. We already know that the solution is x(t) = ekt . u0 (s)) ds. but let us see how this works by developing an example. 2 The next term in the sequence is (kt)2 (kt)3 u3 (t) = 1 + kt + + . . Furthermore. ∫ t u1 (t) = x0 + f (s. t0 Our goal is to show that un (t) → u(t) as n → ∞.45. We define the first few terms of our sequence {un (t)} as follows: u0 (t) = 1.66 CHAPTER 1. 2! 3! n! However. un (s)) ds. A FIRST LOOK AT DIFFERENTIAL EQUATIONS We derive the next function in our sequence using the right-hand side of the integral equa- tion. Consider the exponential growth equation. 2 2·3 and the nth term is ∫ t un (t) = 1 + 1 + kun−1 (s) ds 0 ∫ t ( ) (kt)2 (kt)3 (kt)n−1 =1+ k 1 + ks + + ··· + ds 0 2! 3! (n − 1)! (kt)2 (kt)3 (kt)n = 1 + kt + + + ··· + . ∫ t u2 (t) = 1 + ku1 (s) ds 0 ∫ t =1+ k(1 + ks) ds 0 (kt)2 = 1 + kt + . We will leave the proof of Picard’s Theorem to a series of exercises. dx = kx dt x(0) = 1. this is just the nth partial sum for the power series for u(t) = ekt . unique solution to our initial value problem. t0 Subsequent terms in the sequence can be defined recursively. we need to show that u is the continuous. which is what we expected.

40)? In each case. x(2) = 0 x−2 (f) y ′ = x tan y. EXISTENCE AND UNIQUENESS OF SOLUTIONS 67 1. since f (t.6. • The function u = u(t) is a solution to the initial value problem x′ = f (t. . (a) There exists a unique solution to y ′ = 4 + y 3 . y(1) = 0 √ (c) y ′ = y. 1). x) and x(t0 ) = x0 on some interval |t − t0 | < h contained in the interval |t − t0 | < a. x) have the initial condition x(t0 ) = x0 . justify your conclusion. y(0) = 0 1 (g) y ′ = y + 2t. y(0) = 1 √ (b) y ′ = y. x) : 0 ≤ |t − t0 | < a. if and only if u is a solution to the integral equation ∫ t x(t) = x0 + f (s. (a) y ′ = 4 + y 3 . there exists a unique solution u = u(t) for x′ = f (t. t0 • Existence and uniqueness of solutions is proved by Picard iteration. x) x(t0 ) = x0 . 1. This is of partic- ular interest since the proof actually tells us how to construct as sequence of functions that converge to our solution. Let x′ = f (t. ◦ Uniqueness is especially important when it comes to finding equilibrium solu- tions. ◦ Uniqueness of solutions tells us that the integral curves for a differential equation cannot cross. y)/∂y = 3y 2 are continuous at the point (0. x(s)) ds. x(0) = 2 x−2 t (e) x′ = . y(0) = 1.1. Which of the following initial value problems are guaranteed to have a unique solution by the Existence and Uniqueness Theorem (Theorem 1.3 Important Lessons • Existence and uniqueness of solutions of differential has important implications. In particular.6.6. ◦ Solutions are only guaranteed to exist locally. y) = 4 + y 3 and ∂f (t. y(1) = 1 t (d) x′ = . 0 ≤ |x − x0 | < b} . y(0) = 1 t Hint. If f and ∂f /∂x are continuous functions on the rectangle R = {(t.4 Exercises 1.

y(0) = 1. 2. Hence. Consider the initial value problem y ′ = 2ty + t y(0) = 1. (d) The Existence and Uniqueness Theorem does not apply to x′ = t/(x − 2). since f (t. y(1) = 0. (e) There exists a unique solution to x′ = t/(x − 2). y(0) = 0. √ √ (c) There exists a unique solution to y ′ = y. y) = y is not continuous at (1. 3. (g) The Existence and Uniqueness Theorem does not apply to y ′ = y/t + 2t. 1). x)/∂x = −t/(x − 2)2 are both continuous at the point (2. since f (t. y) = y/t + 2t is not continuous at (0. since f (x. t > t0 is a solution for the initial value problem. which helps explain the need for uniform continuity in the proof of the Existence and Uniqueness Theorem. since f (t. Hint. since f (t. (b) Show that { 0. Find an explicit solution to the initial value problem 1 y′ = (t − 1)(y + 1) y(0) = 1. [Hint: Make sure that the derivative of y(t) exists at t = t0 . Use your solution to determine the interval of existence. x = 1. 1). This is an example of a sequence of continuous functions that does not converge to a continuous function. 0). y)/∂y = x sec2 y are both continuous at the point (0. x) = t/(x − 2) is not continuous at (0. Hint. where t0 is any real number. x(0) = 2. is a solution to the initial value problem. (a) Show that the constant function. x(2) = 0. 2). ] (c) Explain why this example does not contradict the Existence and Uniqueness Theorem.68 CHAPTER 1. since √ f (t. Hint. x) = t/(x − 2) and ∂f (t. Let ϕn (x) = xn for 0 ≤ x ≤ 1 and show that { 0. Consider the initial value problem y ′ = 3y 2/3 y(0) = 0. y)/∂y = √ 1/(2 y ) are both continuous at the point (1. 4. 0). there exists an infinite number of solutions to the initial value problem. y(1) = 1. 0). t ≤ t0 y(t) = (t − t0 )3 . y) = y and ∂f (t. 5. y) = x tan y and ∂f (x. A FIRST LOOK AT DIFFERENTIAL EQUATIONS √ (b) The Existence and Uniqueness Theorem does not apply to y ′ = y. (f) There exists a unique solution to y ′ = x tan y. 0 ≤ x < 1 lim ϕn (x) = n→∞ 1. y(t) ≡ 0. .

. Hint. Show that there exists an M > 0 such that |u1 (t) − x0 | ≤ M |t − t0 |. 10. x1 ) − f (t. t0 Hint. Hint. 2. un (t)) − f (t. un−1 (t))| ≤ K|un (t) − un−1 (t)|. x2 ) in R. Define the sequence {un } by u0 (t) = x0 . Hint. 6. n = 1. t0 Use the result of the previous exercise to show that |f (t.6. x1 ) and (t. x2 )| ≤ K|x1 − x2 | for all (t. .1. x) : 0 ≤ |t − t0 | < a. Use the Fundamental Theorem of Calculus to show that the function u = u(t) is a solution to the initial value problem x′ = f (t. x(s)) ds. . 2 Hint. (b) Let ϕ0 (t) = 1 and use Picard iteration to find ϕn (t). prove the Existence and Uniqueness Theorem for first- order differential equations. 9. 8. 7. Hint. Show that KM |t − t0 |2 |u2 (t) − u1 (t)| ≤ . if and only if u is a solution to the integral equation ∫ t x(t) = x0 + f (s. . The following series of exercises. ∫ t un+1 = x0 + f (s. EXISTENCE AND UNIQUENESS OF SOLUTIONS 69 (a) Use the fact that y ′ = 2ty + t is a first-order linear differential equation to find a solution to the initial value problem. 0 ≤ |x − x0 | < b} . prove that there exists a K > 0 such that |f (t. . (c) Show that the sequence {ϕn (t)} converges to the exact solution that you found in part (a) as n → ∞. x) x(t0 ) = x0 . un (s)) ds. If ∂f /∂x is continuous on the rectangle R = {(t.

Use mathematical induction to show that K n−1 M |t − t0 |n |un (t) − un−1 (t)| ≤ . (b) Since ∫ t |u(t) − v(t)| − K |u(s) − v(s)| ds ≤ 0. A FIRST LOOK AT DIFFERENTIAL EQUATIONS 11. To show uniqueness. (a) Define10 ∫ t ϕ(t) = |u(s) − v(s)| ds. 14. Show that M ( K|h| ) ∞ ∑ ∞ M ∑ (K|t − t0 |)n |un (t) − un−1 (t)| ≤ ≤ e −1 . Any sequence of functions that converges uniformly. n=1 If we can show that this series converges absolutely. t0 Show that ∫ t |u(t) − v(t)| ≤ K |u(s) − v(s)| ds. Show that ϕ′ (t) = |u(t) − v(t)|. x(s)) ds. 10 A similar argument will work for t ≤ t0 . x0 ) and the boundary of the rectangle R. t0 then ϕ(t0 ) = 0 and ϕ(t) ≥ 0 for t ≥ t0 . we can view un (t) as a partial sum for the series ∞ ∑ u0 (t) + [un (t) − un−1 (t)].70 CHAPTER 1. must converge to a continuous function.9 Hint. we know that un (t) converges to a continuous function u(t) that solves our equation. Since |un (t) − un−1 (t)| → 0. then our sequence will converge to a function u(t). K n! K n=1 n=1 where h is the maximum distance between (t0 . 13. . 9 We must a theorem from advanced calculus here to ensure uniform continuity. Since un (t) = u1 (t) + [u2 (t) − u1 (t)] + · · · + [un (t) − un−1 (t)]. n! Hint. assume that u(t) and v(t) are both solutions to ∫ t x(t) = x0 + f (s. t0 Hint. 12. t0 we know that ϕ′ (t) − Kϕ(t) ≤ 0.

6. P1 = 160 (a sink) and P2 = 40 (a source). In this section we will investigate how the solutions of a differential equation vary as we change the value of a parameter. dx = kx dt has the growth rate parameter k. 1. BIFURCATIONS 71 Use this fact to show that d [ −Kt ] e ϕ(t) ≤ 0. dt 200 If we allowed fishing in our pond at a rate of 32 fish per year. dt 200 There are two equilibrium solutions for this equation. .7. then the fish will die out unless the pond is restocked or fishing is banned (Figure 1. then the equation became ( ) dP P =P 1− − 32. 1.7 Bifurcations Many of the equations that we have examined have a parameter. dt Conclude that ∫ t ϕ(t) = |u(s) − v(s)| ds = 0 t0 for t ≥ t0 or for all t ≥ t0 and u(t) = v(t).24 with the equation ( ) dP P =P 1− . which means that we actually have a family of differential equations. If the population of the pond falls below 40. For example.7. k and N .46).1. The logistic equation ( ) dP P = kP 1− dt N has two parameters.1 The Logistic Model with Harvesting Revisited Recall how we modeled logistic growth in a trout pond in Example 1.5 Project—Picard Iteration 1. Hint.

say H = 100.29) dt 200 for P .47). we must solve ( ) dP P =P 1− − 100 = 0 (1. Our differential equation now becomes ( ) dP P =P 1− − 100. √ 200 ± 2002 − 4 · 20000 √ P = = 100 ± −10000. dt 200 To determine the equilibrium solutions. Furthermore. A FIRST LOOK AT DIFFERENTIAL EQUATIONS 200 150 100 P(t) 50 0 0 2 4 6 8 10 t Figure 1.29) has no real solutions and that we have no equilbrium solutions.46: Harvesting with H = 32 Now let us see what happens when we allow more fishing in our pond.72 CHAPTER 1.5 t Figure 1. 2 which means that equation (1. This means that no matter how many fish are in the pond initially.47: Harvesting with H = 100 .5 2 2. This last equation can be rewritten as P 2 − 200P − 20000 = 0. Thus. dP /dt < 0 for all values of P . the trout population will eventually die out due to over fishing (Figure 1.5 1 1. 140 120 100 80 P(t) 60 40 20 0 0 0.

we will generalize our model. dt N If we allow harvesting at a constant rate H. BIFURCATIONS 73 Finally. 4 k . our model now becomes ( ) dP P = kP 1 − − H. we must solve ( ) dP P =P 1− − 50 = 0 dt 200 in order to determine any equilibrium solutions. Therefore. we have dP /dt < 0. P = 100 will be a node. If we will let ( ) P fH (P ) = kP 1 − − H. For values of P > 100 as well as values of P < 100. we will let H = 50.48: Harvesting with H = 50 To better understand what is happening. 200 150 P(t) 100 50 0 0 5 10 15 20 t Figure 1. and the number of fish in the pond will decrease (Figure 1. −2k 2 4 k The explanation of how our model behaves lies in the discriminant.48). N each equilibrium solution must satisfy fH (P ) = 0 or −kP 2 + kN P − HN = 0. Suppose that a population with a limited carrying capacity N is modeled with the logistic equation ( ) dP P = kP 1 − . we will first find the equilibrium solutions. In fact. We now obtain a single equilibrium solution at P = 50. N 2 HN − .7. our equilibrium solutions are given by √ √ −kN ± k 2 N 2 − 4kHN N N 2 HN P = = ± − . In this case. dt N To analyze our model.1.

74 CHAPTER 1. A FIRST LOOK AT DIFFERENTIAL EQUATIONS

If
N 2 HN
− <0
4 k
or, equivalently if H > kN /4, there are no equilibrium solutions and
dP
= fH (P ) < 0
dt
for all values of P . In particular, all solutions of dP /dt = fH (P ) tend towards negative
infinity as t → ∞. In this case, the population is doomed to extinction no matter how large
the initial population is. Since negative populations do not make sense, we say that the
population is extinct when P = 0.
On the other hand, if H < kN /4, we have equilibrium solutions at

N N 2 HN
P1 = + −
2 4 k
and √
N N 2 HN
P2 = − − .
2 4 k
The first equilibrium solution, P1 is a sink, while the second, P2 is a source.
Finally, if H = kN /4, then we will have exactly one equilibrium solution at P = N /2.
Although dP /dt < 0 for all P ̸= N /2, we see that P → N /2 as t → ∞ for all initial values
of P greater than N /2. For initial values of P less than N /2, solutions tend towards −∞
as t → ∞. Thus, the initial population of fish must be at least kN /4; otherwise, the fish
will go extinct.
In our example, we have a family of differential equations—one for each value of H,
( )
dP P
=P 1− − H. (1.30)
dt 200
A small change in H can have a dramatic effect on how the solutions of the differential
equation behave. Changing the value of H from 50 to 50.1 will doom the population of fish
to extinction no matter what the initial population is. As we increase the value of H, the
number of equilibrium solutions changes from two to one and then to none. This change
occurs exactly at H = 50. We say that a bifurcation occurs at H = 50 for equation
(1.30).

1.7.2 One-Parameter Families
Let us consider the equation
dx
= x2 − 4x + λ (1.31)
dt
as a family of differential equations indexed by the parameter λ. If we let fλ (x) = x2 −4x+λ,
then
dx
= fλ (x)
dt
is a called one-parameter family of differential equations. For each value of λ, we obtain
an autonomous differential equation, and for each value of λ, we have a different phase line
to examine.
For λ = 0, the differential equation
dx
= f0 (x) = x2 − 4x = x(x − 4),
dt

1.7. BIFURCATIONS 75

there is a sink at x = 0 and a source at x = 4 (Figure 1.49).

4

3

2 λ=0

1
x(t)
0

-1

-2

-3

-4
-4 -3 -2 -1 0 1 2 3 4
t
Figure 1.49: x′ = x2 − 4x + λ for λ = 0

For λ = 4, the differential equation
dx
= f4 (x) = x2 − 2x + 4 = (x − 2)2 ,
dt
we have exactly one equilibrium solution, a node at x = 2 (Figure 1.50).

6

5

4

3
x(t)

2

1 λ=4

0

-1

-2
-4 -3 -2 -1 0 1 2 3 4
t
Figure 1.50: x′ = x2 − 4x + λ for λ = 4

76 CHAPTER 1. A FIRST LOOK AT DIFFERENTIAL EQUATIONS

If λ = 8, then the differential equation

dx
= f8 (x) = x2 − 2x + 8
dt

has no equilibrium solutions (Figure 1.51).

6
5
4
3 λ=8
x(t)

2
1
0
-1
-2
-4 -3 -2 -1 0 1 2 3 4
t
Figure 1.51: x′ = x2 − 2x + λ for λ = 8

In fact, the number of equilibrium solutions for (1.31) changes at λ = 4. We say that
λ = 4 is a bifurcation value for the differential equation

dx
= fλ (x) = x2 − 4x + λ. (1.32)
dt

For λ < 4, we have two equilibrium solutions.


x=2± 4 − λ.

For values of λ > 1, there are no equilibrium solutions. We can record all of the information
for the various values in a graph called the bifurcation diagram. The horizontal axis is λ
and the vertical axis is x. Over each value of λ, we will plot the corresponding phase line.
The curve in the graph represents the various equilibrium solutions for the different values
of λ. The bifurcation diagram for equation (1.32) is a parabola (Figure 1.52). We have a
phase line for each value of λ.

1.7. BIFURCATIONS 77

x
5

4

3

2

1

λ
-1 1 2 3 4 5

-1

Figure 1.52: Bifurcation diagram for x′ = x2 − 4x + λ

Bifurcations for a one-parameter family of differential equations dx/dt = fλ (x) are, in
fact, rare. Let us consider a bifurcation where a sink changes to a source as we vary the
parameter λ. Suppose that for λ = λ0 , we have a sink at x0 . Then

dx
= fλ0 (x0 ) = 0.
dt

Furthermore, the graph of fλ0 (x) must be decreasing for x near x0 , since fλ0 (x) must be
postive for values of x < x0 and negative for values of x > x0 . In other words, fλ′ 0 (x) < 0 for
x near x0 with fλ′ 0 (x0 ) < 0, then for all λ1 sufficiently close to λ0 , the differential equation

dx
= fλ1 (x)
dt

must have sink at a point x = x1 very close to x0 . A similar situation hold if x0 is a source
and fλ′ 0 (x0 ) > 0. Thus, bifurcations can only occur when fλ0 (x0 ) = 0 and fλ′ 0 (x0 ) = 0.

Example 1.53. Now consider the one-parameter family

dy
= y 3 − αy = y(y 2 − α).
dt

We will have an equilibrium solution at zero for all values of α and two additional equilib-

rium solutions at ± α for α > 0. This type of bifurcation is a pitch fork bifurcation
(Figure 1.54).

78 CHAPTER 1. A FIRST LOOK AT DIFFERENTIAL EQUATIONS

x
1

0.5

λ
-1 -0.5 0.5 1

-0.5

-1

Figure 1.54: The bifurcation diagram for y ′ = y 3 − αy

Example 1.55. Let us find the bifurcation values of the one-parameter family
dy
= y(y − 2)2 + λ. (1.33)
dt
If gλ (y) = y(y − 2)2 + λ, then gλ′ (y) = 3y 2 − 8y + 4. The roots of gλ′ (y) = 0 are y = 2 and
y = 2/3. In order for λ to be a bifurcation value, we must have gλ (2) = λ = 0 or
32
gλ (2/3) = +λ=0
27
Thus, equation (1.33) has two bifurcation values, λ = −32/27 and λ = 0. The bifurcation
diagram for this one-parameter family is given in Figure 1.56.

x
3

2.5

2

1.5

1

0.5
λ
-3 -2 -1 1 2 3
-0.5

Figure 1.56: The bifurcation diagram for y ′ = y(1 − y)2 + λ

1.7.3 Important Lessons

• A one-parameter family of differential equations
dx
= fλ (x)
dt
has a bifurcation at λ = λ0 if a change in the number of equilibrium solutions occurs.

1.7. BIFURCATIONS 79

• Bifurcation diagrams are an effective way of representing the nature of the solutions
of a one-parameter family of differential equations.
• Bifurcations for a one-parameter family of differential equations dx/dt = fλ (x) are
rare. Bifurcations occur when fλ0 (x0 ) = 0 and fλ′ 0 (x0 ) = 0.

1.7.4 Exercises
1. Describe the phase line portraits for each of the following equations and how they depend
on the parameter λ. Draw the bifurcation diagram for each equation.
(a) y ′ = λy − sin y for λ > 2/π.
(b) y ′ = λy 2 − 1 for λ ∈ R.

Hint.
2. Outbreaks of the spruce budworm have been responsible for some major deforestations
in Canada and the United States. The equation
( x) x2
x′ = r 1 − x−c
K a + x2
has been used to describe the dynamics of spruce budworm populations, where the variable
x denotes the population or density of the insect [16]. One explanation that has been given
for the occurrence of outbreaks is based on the multiple bifurcations that occur with this
differential equation.
(a) If a = 0.01, c = 1, and K = 1, we have a family of differential equations parameterized
by r,
x2
x′ = rx(1 − x) − .
0.01 + x2
Solve the equation equation
x2
rx(1 − x) − =0
0.01 + x2
and plot the result in the xr-plane for 0 ≤ r < 1.
(b) To find the bifurcation diagram for the spruce budworm equation, reflect the graph
obtained in part (1) about the line r = x line.
(c) Estimate the two bifurcation values from your graph. Explain what happens to the
population as r increases. That is, when does an outbreak occur? What happens after
an outbreak?

Hint.
3. The differential equation
dy
= y − 4t + y 2 − 8yt + 16t2 + 4.
dt
is not autonomous, separable, or linear; however, we can solve this equation with a change
of variable.
(a) Transform this equation into a new differential equation of the form
du
= f (u)
dt
by letting u = y − 4t.

80 CHAPTER 1. A FIRST LOOK AT DIFFERENTIAL EQUATIONS

(b) Sketch the phase line for this new equation, u′ = f (u), and sketch several solutions.
(c) Find the solutions of the original differential equation that correspond to the equilib-
rium solutions of u′ = f (u). Graph these solutions in ty-plane. Also, sketch the graphs
of the solutions that you plotted in part (b).
(d) Solve the differential new equation and use this information to solve the original dif-
ferential equation.

Hint.

1.7.5 Project—Bifurcations

where t is the time measured in years and derived the system of differential equations dH = aH − bHL.1). the lynx (Section 1. berries. 2). y) dt to be those values of x and y such that f (x.1 Predator-Prey Systems Suppose that we have a predator-prey system consisting of a population of foxes (F ) and of rabbits (R). They will even eat fish and crabs. That is. F ) = (5. More specifically. and vegetables. including rabbits. the system is in balance and there is just enough prey to support a constant population of predators at the point (R. We have already derived a model that describes how a population of snowshoe hares and and how they interact with one of their primary predators. dt dF = −5F + RF. dt Just as in first-order differential equations. an equilibrium solution is a solution where neither x(t) or y(t) is changing. That is. 1 Foxes are omnivores. we define an equilibrium solution for a system of differential equations dx = f (x. dt dL = −cL + dHL. y) dt dy = g(x. as well as fruits.1.1 dR = 2R − RF. y) = 0 and g(x. We denoted the population of hares by H(t) and the population of lynx by L(t). y) = 0. Their diet consists of small mammals. 2 Systems of Differential Equations 2. 2. dt we have an equilibrium solution at R = 5 and F = 2.1 Modeling with Systems Many situations are best modeled with a system of differential equations rather than a single equation. 81 . we can examine the equilibrium solutions of a system.

then the system is no longer in balance. SYSTEMS OF DIFFERENTIAL EQUATIONS If the number of rabbits or foxes changes. if R = 1 and F = 1.1: A simple predator-prey system We can graph the solution to our system in a different manner—we can construct a parametric plot of our solution in the RF -plane. Thus. F (t)) at time t. then dR = 1. a point on the graph is given by (R(t). We can plot many solutions to our predator prey-system and even plot direction fields in the phase plane (Figure 2. We can view the solution curve of our system in the RF -plane in Figure 2. The RF -plane is called the phase plane for our system of differential equations and is analogous to the phase line that we used during our investigation of slope fields for autonomous differential equations. Notice that the solutions are periodic with the same period.3). F(t) 14 RF((tt)) 12 10 8 6 4 2 2 4 6 8 10 t Figure 2. If we assume that we have initial conditions R0 = R(0) = 1. F0 = F (0) = 1.2.82 CHAPTER 2. 2 You will find technology extremely useful when analyzing systems. Observe that a peak in the rabbit population is followed by a peak in the fox population. . we can apply a numerical algorithm to generate a solution for our system.2 The graphs of the solutions for R(t) and F (t) are given in Figure 2. dt dF = −4. R(t). dt and the rabbit population will be increasing while the fox population will decrease. For example.1. We will introduce Sage commands for analyzing systems of equations at the end of this section.

If we have initial values R0 = 1 and F0 = 1. dt where N is the carrying capacity. As a specific example.2.3: The phase plane for a predator-prey system We will now modify our system by assuming that the rabbit population will grow logis- tically if there are no predators present. . dt It is easy to see that we have two equilibrium solutions—one at (R. F ) = (0. dt dF = −cF + dRF. 0) and one at (R.1.2: A solution curve in the RF -plane 10 8 6 F(t) 4 2 0 0 2 4 6 8 10 12 14 R(t) Figure 2. MODELING WITH SYSTEMS 83 F(t) 8 6 4 2 0 2 4 6 8 10 12 14 R(t) Figure 2. 1). consider system ( ) dR R = 2R 1 − − RF. F ) = (5. dt 10 dF = −5F + RF. The system can now be written as dR = aR(1 − R/N ) − bRF. Our solutions now behave very differently from the assumption that the population of the prey grows exponentially.

5 0 0 2 4 6 8 R(t) Figure 2. If x = 0.5 2 F(t) 1.2 The Spring-Mass Model Revisited Recall the spring-mass model from Section 1. the solutions tend towards the equilibrium solution. 1) is an example of a stable equilibrium solution. the second-order differential equation for the spring-mass system is mx′′ = −bx′ − kx. then the spring is in a state of equilibrium (Figure 1.5 1 0. R(t).84 CHAPTER 2.5. The spring displacement is denoted by x. then the spring is stretched.4). The equilibrium solution (5. The phase plane for our modified predator-prey system is given in Figure 2.4: Solutions for a modified predator-prey system 3 2.1.5: Phase plane for a modified predator-prey system 2. In fact. the spring is compressed. If we use a dashpot to add damping. If x > 0. If x < 0. then the mass will oscillate. We have a mass lying on a flat surface that is attached to one end of a spring with the other end of the spring attached to a wall. SYSTEMS OF DIFFERENTIAL EQUATIONS then our solution is no longer periodic (Figure 2. F(t) 8 R(t) F(t) 7 6 5 4 3 2 1 2 4 6 8 10 t Figure 2.1. or mx′′ + bx′ + kx = 0.4). . If the surface is frictionless and we pull on the mass.

and k are all positive constants.3 Modeling Epidemics Systems of differential equations are very useful in epidemiology. Individuals enter the removed population by either recovering from the disease or dying. the equation becomes mv ′ + bv + kx = F (t).2. 1 b k v ′ = F (t) − v − x. with initial position x(0) = x0 and initial velocity x′ (0) = x′0 . and the 2015 ebola outbreak in west Africa. we can rewrite this equation as a system of first-order differential equations by letting v(t) = x′ (t). 2. In this case. we know that S(t) + I(t) + R(t) = N. we can represent the effect of the disease by the diagram S −→ I −→ R. including the bubonic plague. We will assume that each individual in the population falls into one of the following categories: S(t) = Susceptible individuals I(t) = Infected individuals R(t) = Removed individuals Susceptible individuals who do not yet have the disease and can catch the disease from infected individuals. Differential equations can be used to model various epidemics. MODELING WITH SYSTEMS 85 where m. x′ = v. we will consider a very simple situation. then the individual is immune to the disease. dS = −αSI dt . and birth do not play an important role. Since the population is closed. We will assume that we have a closed population of size N . We now have a system of first-order differential equations.1. We will also ignore any deaths that are not related to our disease. emigration. We can model how the disease acts with the following system of equations. To understand how we might model an epidemic. Schematically. Given the differential equation mx′′ + bx′ + kx = F (t). If an infected individual recovers. where immigration. AIDS. This model is called an SIR model. m m m with initial conditions x(0) = x0 v(0) = v0′ . b.1. A second-order differential equation can be restated in terms of a system of first order differential equations. influenza.

we can see how the susceptible and removed populations interact in an SIR epidemic in Figure 2. I(t) 100 S(t) I(t) 80 60 40 20 10 20 30 40 50 60 70 80 t Figure 2.005 and β = 0.6: The susceptible and removed populations for an SIR epidemic There are many questions associate with epidemic models. • Will there be an epidemic? • If there is an epidemic. Since d d [S(t) + I(t) + R(t)] = N = 0.6. dt Letting α = 0. S(t). what is the prevalence of the infection? • Can the disease be eradicated or controlled? • What is the effect of the population age structure? We will later explore how we can adjust the SIR model to address some of these issues.08. SYSTEMS OF DIFFERENTIAL EQUATIONS dI = αSI − βI dt dR = βI. .86 CHAPTER 2. dt dt we need only solve the system dS = −αSI dt dI = αSI − βI. dt We say that α is the rate of infection and β is the rate at which the infected are removed. an infected individual either dies or recovers after 1/β days. how many individuals will be infected? • Is there a period of time for which individuals are exposed to the disease but exhibit no symptoms and cannot infect others? • If the disease is endemic. That is.

y) = 0. dt can act very differently with respect to equilibrium solutions. y) dt dy = g(x. • We can graph solutions to the system dx = f (x. Consider the two predator-prey systems dH = aH − bHL. dt as parametric plots in the xy-plane. dt and dH = aH(1 − H/α) − bHL.2. y). y) dt to be those values of x and y such that f (x. dt dL = −cL + dHL. MODELING WITH SYSTEMS 87 2. The xy-plane is called the phase plane for our system of differential equations. dt dy = g(x. In the example of the second system. dt dy = g(x. y(t)) at time t.1. • Solution curves in the phase plane of dx = f (x. .4 Important Lessons • A second-order linear equation a(t)x′′ + b(t)x′ + c(t)x = g(t) can be written as a system of first-order equations by letting v(t) = x′ (t). y). A point on the graph is given by (x(t). y). • An equilibrium solution for a system of differential equations dx = f (x. y). solutions approach the equilibrium solution as t → ∞.1. dt dL = −cL + dHL. dt Solutions graph as closed orbits in our example of the first system. y) = 0 and g(x.

How would you modify the system to include this assumption? (d) Suppose the predators discover a second source of food that is limited in supply. (a) How would you modify this system to include the effect of hunting of the prey at a rate of α units of prey per unit of time? (b) How would you modify this system to include the effect of hunting of the predators at a rate proportional to the number of predators? (c) Suppose the predators discover a second. Hint. and that they move away if there are fewer than five times as many prey as predators. 4. SYSTEMS OF DIFFERENTIAL EQUATIONS 2. 2. (a) Consider the function y(t) = cos βt.1. Under what conditions on β is y(t) a solution to the differential equation? (b) What initial condition (t = 0) in the yv-plane corresponds to this solution? (c) In terms of k and m. (c) Using the result of the previous exercise.9xy. where x(t) is the population of the prey and y(t) is the population of the predator at time t. what is the period of this solution? (d) Sketch the solution curve in the yv-plane that corresponds to this solution. How would you modify the system to take this into account? Hint.88 CHAPTER 2. Consider the following predator-prey system.5 Exercises 1. x′ = 2x(1 − x/2) − 1. Consider the equation d2 k 2 + y=0 dt m for the motion of a simple harmonic oscillator. . A mass weighing 4 pounds stretches a spring 4 inches. How would you modify the system to take this into account? (f) Suppose that prey move out of an area at a rate proportional to the number of predators in the area.2xy y ′ = −y + 0. Under what conditions on β is y(t) a solution to the differential equation? (b) Formulate an initial-value problem that corresponds to the motion of this undamped mass-spring system if the mass is extended 1 foot from its rest position and released with no initial velocity. (a) Consider the function y(t) = cos βt. find the solution of this initial-value problem. Hint. How would you modify the system to include this assumption? (e) Suppose that the predators migrate to an area if there is more than five times as many prey in that area (x > 5y). unlimited source of food. 3. SEIR model Hint. but they still prefer to eat the primary prey when they can catch them.

(t.4) . we can divide each component by the length of the vector. x + y /5] p = plot_vector_field (F .1) ′ y = x + y/5 (2. t = var ( ' x␣y␣t ' ) F = [ -x . y ] . x .4).4) ) p Notice the the vectors are different lengths depending on their magnitudes.1 . x . x . -4 .k in P] . y . j] for i . -1] .4) First. ics = [0 . x . x + y /5] n = sqrt (F [0]^2 + F [1]^2) F_unit = [F [0]/ n .1. -4 . y) and then plot the points using the line command. x. y ] . step = 0.01) P We new have a numerical approximation of the solution to the system (2. F [1]/ n] p = plot_vector_field ( F_unit . It would be much more useful if we could display a graph of the solution. y . -4 . ' $y ( t ) $ ' ]) p Now suppose that we wish to plot solutions x(t) and y(t) to our system.y . (y . However.4) . end_points = 10 . t = var ( ' x␣y␣t ' ) F = [ -x . MODELING WITH SYSTEMS 89 2. (x .01) Q = [ [i .y . x + y /5] n = sqrt (F [0]^2 + F [1]^2) F_unit = [F [0]/ n . x) and (t. (y . (y . our approximation is just a very long list of points.1 .y . t = var ( ' x␣y␣t ' ) F = [ -x . y . x + y /5] P = desolve_system_rk4 (F . t = var ( ' x␣y␣t ' ) F = [ -x . step = 0. ivar = t . If we wish all of the vectors to be the same length.y .4) ) p We can also add axes labels to our plot. (x . We can use the Sage command desolve_system_rk4.y . [x .6 Project—Using Sage to Solve Systems We can use Sage to plot the solution of the system x′ = −x − y (2. t = var ( ' x␣y␣t ' ) F = [ -x . y .2. let us plot a direction field for our system.1)–(2. we get a list of triples. we grab pairs (t. x + y /5] P = desolve_system_rk4 (F .4) . -4 . (x .4) .j .3) y(0) = −1 (2. In fact.2) x(0) = 1 (2. ics = [0 . -1] . x . F [1]/ n] p = plot_vector_field ( F_unit . -4 . end_points = 10 . axes_labels =[ ' $x (t)$ ' . In the code below.1. -4 . y . y). ivar = t . [x .

axes_labels =[ ' $x ( t ) $ ' . x + y /5] P = desolve_system_rk4 (F . ivar = t .5) .k in P] p = line (Q .1. j] for i . legend_label = ' $x (t)$ ' . legend_color = ' red ' .y . legend_color = ' blue ' ) p To plot the solution in the xy–plane. (y . [x . end_points = 10 . color = ' blue ' .1 . -1] .j .5 . legend_label = ' $y ( t ) $ ' .k] for i . [x . ' $y ( t ) $ ' ]) p . ' $y ( t ) $ ' ] . ' $y ( t ) $ ' ] . -1.5 . y ] . thickness =3) p += arrow (Q[ int ( len (Q) /5) ]. fontsize =12) p += line (R .1.y . k] for i .1 .01) Q = [ [j . Q [ int ( len ( Q ) /5) + 1]) p Now let us add a vector field to our plot. y .y . axes_labels =[ ' $x (t)$ ' . and legend colors so that the plot makes more sense. x + y /5] P = desolve_system_rk4 (F . x . end_points = 10 . ics = [0 . t = var ( ' x␣y␣t ' ) F = [ -x .01) Q = [ [i .5) . -1.j . ivar = t . axes_labels =[ ' $t$ ' . -1] . y . [x .j . x + y /5] P = desolve_system_rk4 (F . we will need to select the second two entries in each triple. x . fontsize =12 .1 . x .j . fontsize =12) p We can adjust the thickness of the plot add an arrow to indicate the direction of the solution. end_points = 10 .k in P] p = line (Q . step = 0. ' $y ( t ) $ ' ] . [x . t = var ( ' x␣y␣t ' ) F = [ -x .k] for i .k] for i . thickness =3) p += arrow (Q[ int ( len (Q) /5) ]. x + y /5] P = desolve_system_rk4 (F .90 CHAPTER 2. axes_labels =[ ' $x ( t ) $ ' .01) Q = [ [j .k in P] p = line (Q .j . y ] . -1] . y ] . ivar = t . ivar = t .01) Q = [ [j .␣ y ( t ) $ ' ] . step = 0. y .k in P] p = line (Q) p += line (R) p We can also add colors. (x . y ] .k in P] R = [ [i .j . end_points = 10 . t = var ( ' x␣y␣t ' ) F = [ -x . step = 0. F [1]/ n] p += plot_vector_field ( F_unit . -1] . SYSTEMS OF DIFFERENTIAL EQUATIONS R = [ [i . k] for i . Q [ int ( len ( Q ) /5) + 1]) n = sqrt (F [0]^2 + F [1]^2) F_unit = [F [0]/ n . axes_labels =[ ' $x ( t ) $ ' . fontsize =12 . ics = [0 . y . ' $x ( t ) . x . color = ' red ' . ics = [0 .y . t = var ( ' x␣y␣t ' ) F = [ -x . axes labels. step = 0. ics = [0 .1 .k in P] p = line (Q .

x′ = v. this should not be too surprising.7. The Sage cell be- low can be used to make your own computations.8).5) dt dy = g(x.6) dt Equation (2.2.sagemath.2.org/doc/reference/calculus/sage/calculus/desolvers. we can rewrite this equation as the first order system. THE GEOMETRY OF SYSTEMS 91 To learn more about how to use Sage to solve systems of differential equations. (2. mx′′ + kx = 0.5) tells us how a solution curve changes in the x direction. (2.6) tells us how a solution curve changes in the y direction.2. After plotting only few vectors. y). while equation (2.1. The direction field is relatively easy to understand. . we can very quickly see that the vectors are tangent to circles at the origin (Figure 2. 2. Consider the differential equation for a simple harmonic oscillator that we developed in Section 1.2 The Geometry of Systems We can use direction fields in the phase plane to represent autonomous systems dx = f (x. 2. Since the solutions to the undamped harmonic oscillator x′′ + x = 0 are of the form x(t) = A cos t + B sin t for arbitrary constants A and B. see http:// www. y). If we assume that k and m are both equal to one and let x′ = v. v ′ = −x.1 Direction Fields Example 2.html.

9. 3 2 1 y(t) 0 -1 -2 -3 -3 -2 -1 0 1 2 3 x(t) Figure 2.10). The system x′ = x y′ = y gives us a direction field where the vectors point away from the origin (Figure 2.10: The direction field for x′ = x and y ′ = y .8: The direction field for a harmonic oscillator Let us examine some systems of equations with direction fields that are easily plotted. SYSTEMS OF DIFFERENTIAL EQUATIONS 3 2 1 v(t) 0 -1 -2 -3 -3 -2 -1 0 1 2 3 x(t) Figure 2. Example 2.92 CHAPTER 2.

11: The direction field for x′ = x and y ′ = y The system x′ = −x. however. we shall soon see that there are important differences between this direction field and the direction field of the previous system (Figure 2. y ′ = −y. also gives us a direction field where the vectors point towards the origin. THE GEOMETRY OF SYSTEMS 93 The system x′ = −x. y ′ = −5y.2. 3 2 1 y(t) 0 -1 -2 -3 -3 -2 -1 0 1 2 3 x(t) Figure 2. gives us a direction field where the vectors point towards the origin (Figure 2.11).2.11). .

dR = aR(1 − R/α) − bRF.94 CHAPTER 2. dt F ′ (t) −cF + dRF We can view the right side of the equation as a vector field. dt . In order to investigate the geometric properties of our system.12: The direction field for x′ = −x and y ′ = −5y 2. we will assume that the prey in our model has logistic growth. For each value of t. dt where N is the carrying capacity. Now we can write our predator-prey model as a single vector equation.2. x(t) corresponds to our solution curve (R(t). That is. ( ) ( ) dx R′ (t) aR(1 − R/N ) − bRF. F (t) This vector-valued function. we will rewrite our system in vector form.2 Modified Predator-Prey System Let us recall the modified predator-prey system that we developed in the last section. dt dF = −cF + dRF. F (t)) in the RF - plane. = = . SYSTEMS OF DIFFERENTIAL EQUATIONS 3 2 1 y(t) 0 -1 -2 -3 -3 -2 -1 0 1 2 3 x(t) Figure 2. we will use x(t) to denote the vector-valued function ( ) R(t) x(t) = . dt 10 dL = −5F + RF. The specific example that we examined was ( ) dH R = 2R 1 − − RF.

if we let (R. we have (R′ . dx ( x) = αx 1 − − βxy dt M dy ( y ) = γy 1 − − δxy.5 2 L(t) 1. We can model two competing species using the following system of first-order differential equations. Thus. dx = x(1 − x) − αxy dt dy = y(1 − y) − αxy. Suppose that x = x(t) and y = y(t) are two populations who compete for the same resources are governed by the following system of differential equations. For example. THE GEOMETRY OF SYSTEMS 95 The vector field form of this system is ( ) dx 2R(1 − R/10) − RF = . dt N The first term in each equation is the logistic growth of each species.13: A vector field for the predator-prey system 2.2. two species of fish may compete for the same food in a lake or cattle and sheep may compete for the same grazing land.13).2. The second term is how the species is affected by interacting with the competing species. . dt If we study how the two populations interact. F ) = (10. dt −5F + RF We can associate a vector in the RF -plane for each value of R and F . we will discover two very different cases depending on the value of the parameter α. we obtain an direction field for our system of differential equations (Figure 2. We can represent this vector in the phase plane by drawing an arrow in the proper direction. F ′ ) = (−100.5 0 0 2 4 6 8 H(t) Figure 2. 50). At this particular point.14. 3 2. the population of rabbits is falling rapidly while the number of foxes is climbing very quickly.2. 10).3 A Competing Species Model Suppose that x and y are the population of two distinct species that compete for the same resources. For example. Example 2.5 1 0.

7)–(2. y) = 0 are called the x-nullclines of the system and the values of x and y satisfying the equation dy/dt = g(x. dt We are interested in what happens as t → ∞.9) y = 1 − 2x.7) y = (1 − x)/2. dt If dx/dt = 0.10) are called nullclines.96 CHAPTER 2. (2.8) Similarly. we have y=0 (2.10) The lines in equations (2. y). (2. SYSTEMS OF DIFFERENTIAL EQUATIONS If we let α = 2. y) = 0 are called the y-nullclines of the system. dt then the values of x and y satisfying dx/dt = f (x. In general. if we are given the system dx = f (x. . if dy/dt = 0. First we will examine what happens when dx/dt = 0 and dy/dt = 0: dx = x(1 − x − 2y) dt dy = y(1 − y − 2x). then x=0 (2. y) dt dy = g(x. then there is a strong interaction between the two species and our system becomes dx = x(1 − x) − 2xy dt dy = y(1 − y) − 2xy.

1/3). If dx/dt = 0. There can be no right or left movement at this point in the phase portrait.8 1 1.15). 0. we can see that dx = −4.2.3) for example.2 dt dy = −4. 0).2. (0.2 1.5.4 0.4 0. we have no guarantee that the trajectory will continue in this direction. the equilibrium points for our particular system are (0. 0).4. dt Since both of these numbers are negative. the trajectory that crosses the R-nullcline at (0. and (1/3. the point (0. Furthermore.25) lies on the x-nullcline and dy (0. We will indicate the direction of horizontal motion by drawing a horizontal line with an arrow pointing to the right if dx/dt > 0 on the y-nullcline and an arrow pointing to the left if dx/dt < 0 on the y- nullcline.5. there can be no vertical motion on an y-nullcline. 1). then any solution crossing this x-nullcline must be moving vertically. 0.25) < 0. dt Therefore.5. 1. That is. we will examine the case when dx/dt = 0 or dy/dt = 0. we have an equilibrium solution whenever an x-nullcline intersects with a y-nullcline. let us determine what happens on the nullclines themselves.6 0.8 y(t) 0. 0.2 0 0 0. As we move along the x-nullcline. (1. We can indicate this fact along with the direction of vertical movement by drawing a vertical slash with an arrow on the x-nullcline (Figure 2.2 1 0.15). . If we are given the initial condition (1. For example. we can plot the x and y-nullclines to help us understand our system (Figure 2.4 1.4 x(t) Figure 2. THE GEOMETRY OF SYSTEMS 97 1.15: Nullclines for the case β = 2 In our example.25) is moving down. Before we proceed further with our analysis. Similarly. we can choose a representative point in each region to find how the direction field is oriented. Thus.03.2 0.6 0. Of course. the direction of this arrow can only change when we cross an y-nullcline. However. we can see that our initial trajectory is headed down and to the left—slightly more to the left than down.

SYSTEMS OF DIFFERENTIAL EQUATIONS We can also determine the basic direction of the solution curve by checking what happens at a point in each of the regions bounded by the nullclines.1) = 0. The system x′ = y − x2 y′ = x − 2 has an x-nullcline y = x2 and a y-nullcline x = 2.1. • Only species y survives and species x becomes extinct.1. By choosing a point in each of these regions and determining the direction of the slope field at that point. Example 2. 0. For example. The nullclines divide the plane into four basic regions (Figure 2. at the point (0.4. 0. dt Thus.98 CHAPTER 2.1) = 0. . • There are essentially equal numbers of species x and y.7 dt d y(0.16. There is no reason why our nullclines should be limited to straight lines. • Only species x survives and species y becomes extinct.1. 4). the general direction of any solution curve in this region is up and right (Figure 2.15). 1.3)? We see three possible scenarios if we follow the nullclines for large values of t.1) we find that d x(0. 0. These two nullclines intersect at (2.17). and the system has equilibrium solution.7. What happens to the initial condition (1. we can decide the direction of the slope field in the entire region.

x_max = input_box ( default = 1.2. y_null = checkbox ( False . step_size = (0.0) ) . x_start = input_box ( default = 1. x_min = input_box ( default = 0) .5) .2* x * y ) .3) . THE GEOMETRY OF SYSTEMS 99 8 6 y(t) 4 2 0 -4 -3 -2 -1 0 1 2 3 4 x(t) Figure 2.4 Plotting Direction Fields with Sage It is easy to plot direction fields using Sage.nullcline ␣ ( green ) " ) . You can evaluate the cell below to get a menu driven applet for drawing direction fields for systems of two differential equations.17: The system x′ = y − x2 and y ′ = x − 2 The only equilibrium solution of our system occurs at (2.y .001 . y_max = input_box ( default = 1. (5 . " Show ␣ \( x \) . x . 1. " Show ␣ \( y \) . t_start = input_box ( default = 0.t = var ( ' x .01 .y . 2.2.y ) . (0. x_null = checkbox ( False . y_start = input_box ( default = 1.x ) . We can plot the direction field for the system x′ = y − 2x. 4). g = input_box ( default = y *(1 .t ' ) @interact def _(f = input_box ( default = x *(1 .nullcline ␣ ( red ) " ) ) : n = sqrt (f ^2 + g ^2) .2.4) . 50) ) . y ′ = x − xy − x3 using the following Sage interact.5) . t_end = (10 .2* x * y ) . y_min = input_box ( default = 0) .0) .

then x(t) will approach on of the equilibrium points as t → ∞. y_start ]. then both dx/dt and dy/dt vanish. step = step_size ) Q = [ [j . y_max ) . y_max ) . pointsize =50) P = desolve_system_rk4 ([ f . ics = [ t_start . width = 1 .k in P ] p += line (Q . x_min . color = ' red ' . These intersections are the equilibrium points. Draw the curves where f (x. (x . x_min . (y .5 A Summary of Phase Plane Analysis We can use the following series of steps to summarize phase plane analysis for the nonlinear system dx = f (x. (x . color = ' green ' . x_max ) . g / n ) . (y . (y .j . y) dt dy = g(x. When x(t) lies on one of these curves. (x . y ] . ivar = t .k] for i . x_min . • Step 4. y_max )) p += point (( x_start . x_max ) . ymax = y_max ) pretty_print ( html (" <br ␣ / > " ) ) pretty_print ( html (r"$\ displaystyle \ frac { dx }{ dt } ␣ = ␣ % s$ " % latex (f))) pretty_print ( html (" <br ␣ / > " ) ) pretty_print ( html (r"$\ displaystyle \ frac { dy }{ dt } ␣ = ␣ % s$ " % latex (g))) 2. Q [ int ( len (Q ) ) . If x(t) is ever at one of these points. • Step 2. Draw horizontal slash marks on the y-nullclines to remind yourself that a trajectory crossing the nullcline can only do so if it is moving in a horizontal direction at the instant of crossing. y_start ) . dx/dt = 0. axes_labels = [ ' $x ( t ) $ ' . These curves are called the x-nullclines. These regions are always separated by x-nullclines. legend_label = ' \( x \) . Draw the curves where g(x. SYSTEMS OF DIFFERENTIAL EQUATIONS p = plot_vector_field ( ( f /n . g ] .2. Label the regions of the xy-plane where dx/dt < 0 and where dx/dy > 0. ' $y ( t ) $ ' ]) p += arrow (Q[ int ( len (Q)) . • Step 3. draw vertical slash marks on the x-nullclines to remind yourself that a trajectory crossing the nullcline can only do so if it is moving in a vertical direction at the instant of crossing. ymin = y_min . y_min . label the regions where dy/dt is positive and negative. y) = 0. arrowsize = 4) x_nullcline = implicit_plot (f . These curves are called the y-nullclines. dg/dt = 0. y_min .2] . y_min . . x_max ) . y) = 0. [x . legend_label = ' \( y \) . Likewise. Label the points where the x and y-nullclines intersect. xmax = x_max . end_points = t_end . x_start . dt • Step 1. This means that the trajectory stays at the equilibrium point for all time.1] .nullcline ' ) y_nullcline = implicit_plot (g . show ( xmin = x_min .100 CHAPTER 2. y). When a solution curve x(t) lies on one of these curves.nullcline ' ) if x_null : p += x_nullcline if y_null : p += y_nullcline p. If our system is going to tend towards as steady state.

. and down on those parts of the x-nullclines in the dy/dt < 0 regions. what happens to the two populations as t → ∞? Hint. the trajectory moves down and right. the trajectory moves up and left. ◦ If dx/dt > 0 and dy/dt < 0. (c) Sketch the phase plane for this system. For the following two systems of equations • Find the equilibrium points of the system. 2. (a) x′ = cos y y ′ = −x + y (b) x′ = y y ′ = x − x3 − y Hint. then both x(t) and y(t) are increasing and the trajectory moves up and right. draw arrows on the horizontal slash marks of the y-nullclines. say β = 1/2. • Briefly describe the behavior of typical solutions. 2. ◦ If dx/dt < 0 and dy/dt < 0. These arrows are pointing right on the parts of the y-nullclines in the dx/dt > 0 regions and left point on the parts in the dx/dt < 0 regions.3.2. (a) Find the x and y-nullclines for this system. ◦ If dx/dt < 0 and dy/dt > 0.Go back and put arrows on the vertical hash marks of the x null clines. • Step 6. Consider the competing species model dx = x(1 − x) − βxy dt dy = y(1 − y) − βxy. These arrows indicate whether the motion across the null cline is up or down. • Sketch the phase plane and direction field for each system using technology.4 and y(0) = 0. (b) Find all equilibrium points for this system. Likewise.2. THE GEOMETRY OF SYSTEMS 101 • Step 5. The arrows are up on the parts of the x-nullclines that are in the dy/dt > 0 region. (d) If the inital populations are given by x(0) = 2. the trajectory moves down and left.2. ◦ If dx/dt > 0 and dy/dt > 0. dt where the species interact weakly.6 Exercises 1.

Some diseases such as Ebola. -2* x + y] P = desolve_system_rk4 (F . step =0. Consider an epidemic that moves through an isolated population. and infectious individuals are imme- diately infectious. ' $y ( t ) $ ' ]) p .k] for i . -4 . • The length of the incubation period is negligible. The goal of this exercise is to examine the differences between the two. -4 . What happens to the solution curve as t → ∞? (c) What conclusions can you draw about the behavior of the two different epidemics? Hint. axes_labels =[ ' $x ( t ) $ ' .05.4) .0003S (2. (a) If α = 0.1I.4) . are extremely infectious with a mortaility rate of up to 90%. We assume that the constant of proportionality is α.12) dt The constant α is a measure of the relative infectivity of the disease. then dS = −αSI + 0. • Only a single individual is initially ill. thickness =2) n = sqrt (F [0]^2 + F [1]^2) F_unit = [F [0]/ n .k in P] p = line (Q . Suppose that S(0) = 30000 and I(0) = 1. SYSTEMS OF DIFFERENTIAL EQUATIONS 3.102 CHAPTER 2. y] . t = var ( ' x␣y␣t ' ) F = [ x + y .0003 per individual per year. axes_labels =[ ' $x (t)$ ' . fontsize =18 . draw the phase portrait. If S(t) is the number of susceptible and I(t) is the number of infected people. We will make the following assumptions about the epidemic.55 . • Infected individuals do not give birth. AIDS which is caused by the HIV virus. y ′ = −2x + y. has a much lower transmission rate.000001. y . What happens to the solution curve as t → ∞? (b) If α = 0. x . an infected individual dies after 10 days. draw the phase portrait. but susceptible individuals have a birth rate of 0.01) Q = [ [j . ' $y ( t ) $ ' ] . First let us plot the phase plane of the system without nullclines. Be sure to label all nullclines and equilibrium solutions.[x .j . ics =[0 . (y . • On the average.7 Project—Plotting Nullclines with Sage Let us use Sage to analyze the system x′ = x + y. ivar =t . Newborns are susceptible.0.0] . end_points =2 . (x . Suppose that S(0) = 1000 and I(0) = 1. F [1]/ n] p += plot_vector_field ( F_unit . (2. 2. • Individuals are infected at a rate proportional to the product of the number of infected and susceptible individuals. On the other hand.11) dt dI = αSI − 0. a viral hemorrhagic fever. Be sure to label all nullclines and equilibrium solutions.2.

gravity will become a factor and the crane may topple due to its weight. F [1]/ n] p += plot_vector_field ( F_unit . . color = ' green ' ) p += implicit_plot (F [1] . step =0. x . cranes are susceptible to winds and the end of the boom can move back and forth several feet even in moderate winds.55 .3 Numerical Techniques for Systems If we are unable to find an analytic solution to a first-order differential equation y ′ = f (t.4) .k] for i . ' $y ( t ) $ ' ]) p += implicit_plot (F [0] .6 We can add nullclines to our plot using the implicit_plot function from Sage. We can model the motion as a harmonic oscillator if the side-to-side motion is not too great. axes_labels =[ ' $x ( t ) $ ' . you may want to refer to Subsection 2. many of the numerical techniques that are used to solve a first-order equation can be extended to solve a system. ' $y ( t ) $ ' ] . -4 .3.k in P] p = line (Q . axes_labels =[ ' $x (t)$ ' . 2. fontsize =18 . ivar =t . NUMERICAL TECHNIQUES FOR SYSTEMS 103 If these Sage commands seem unfamiliar to you. y] . end_points =2 .4) .0] .4) .4) . thickness =2) n = sqrt (F [0]^2 + F [1]^2) F_unit = [F [0]/ n . -4 . there is no reason to expect that it will be any easier to solve a system of equations. color = ' red ' ) p The x-nullclines are plotted in green and the y-nullclines are in red. However. (y .[x . if the wind becomes too strong and the crane moves too much.0. y).1. (y .4) . 2. This is about the same height as a 50 story building. -4 .18).2. (x . However. At such heights. t = var ( ' x␣y␣t ' ) F = [ x + y . -4 .3. -4 . (x .1 Duffing’s Equation Large mobile cranes can reach up to 700 feet (Figure 2. (x . -2* x + y] P = desolve_system_rk4 (F . y . ics =[0 .j . (y .4) .01) Q = [ [j . -4 .

.

we must account for the effect of . t) position .3.19: Solutions to x′′ + 0. dt dv = −0.19. x(t).3x = 0 dt dt or the equivalent system. If the motion is larger. but it is easy to check that (√ ) (√ ) −t/20 119 −t/20 119 x(t) = c1 e cos t + c2 e sin t 20 20 is a solution to our equation. t ) + k * x == 0 . dt In the case of our swaying crane. the boom is bent. we will let x = x(t) be the displacement of the end of the boom from the perfect vertical position.3x = 0 Modeling the swaying motion of a crane with a harmonic oscillator might work only if the side-to-side motion is small. x ) velocity = diff ( position . the swaying of our crane’s boom might be described by an equation such as d2 x dx 2 + 0.2. When x(t) ̸= 0.t . t = var ( ' t ' ) x = function ( ' x ' )(t) b = 0. dx = v.2) + b * diff (x . dt We will learn how to solve such systems later.1x′ + 0.1 + 0. We show several solutions to our equation in Figure 2. NUMERICAL TECHNIQUES FOR SYSTEMS 105 dv = −qx − pv. show () The constants c1 and c2 can be determined if we know the initial position and initial velocity of the end of the crane’s boom. show () velocity . and the structure of the boom supplies a strong restorative force to bring everything back to a true vertical position. The Sage code for solving this system is given below. Thus.3 position = desolve ( diff (x .1v.3x − 0.1 k = 0. v(t) 3 x(t) v(t) 2 1 10 20 30 40 50 60 70 80 t -1 -2 Figure 2.

We can model this effect by setting the equation for our harmonic oscillator equal to a factor of x3 .02x3 . one of our main objectives should be finding and analyzing solutions to such a sys- tem. when x is large. The equation x′′ + 0.3x = 0. SYSTEMS OF DIFFERENTIAL EQUATIONS gravity in our model.1v − 0. dx =v (2. Consider the system dx = f (t. y) dt .3.106 CHAPTER 2.15) dt Thus. When x(t) is large. (2. We can rewrite equation (2.3x + 0.14) is nonlinear due the x3 term. (2.13) as a system of first-order differential equations by letting dx/dt = v. x. Thus. However. gravity will pull downward on that part of the crane will cause the crane to bend even further. we still need to numerically generate solutions in order to plot the phase plane.20).14) dt dv = −0. 4 2 v(t) 0 -2 -4 -4 -2 0 2 4 x(t) Figure 2.1 + 0. the term will contribute a great deal.1x′ + 0.02x3 is an example of Duffing’s equation. There is little hope to finding an analytic solution to such as system.02x3 .13) dt dt When x is small. d2 x dx 2 + 0. 2.2 Euler’s Method for Systems The system of equations (2. However. this forcing term will not contribute much to the motion of the building. We can use software to plot the phase plane in order to learn something about the solutions (Figure 2. part of the crane will not be above any other part of the crane.20: The phase plane for Duffing’s equation Let us see how we might find a numerical solution for a system.3x = 0.14)–(2.

.3. yk ). Figure 2. We can move in the direction of this tangent vector for time h in order to find the next point x1 .22 (Numerical Solutions for Duffing’s Equation). f = (f. . y2 . yk ) and fk = (f (xk . where h is the step size. dt . If we let xk = (xk . . We wish to find approximate values x1 . yk ). yn for the solution x(t) and y(t) at the points tk = t0 + kh. We can repeat this technique to generate an approximate solution curve in the phase plane.2. g(xk . . xk . yk ) yk+1 = yk + hg(tk . dy/dt). xn and y1 . . Suppose that dx = v. y). . n. y). g). yk )). x. We then calculate a new tangent vector f1 and then move along this new vector for a time step h to find x2 . xk .21: A solution curve for Duffing’s equation Example 2. dx/dt = (dx/dt. which is the tangent vector to the graph of the solution x(t) in the xy-plane (Figure 2. x) dt x(t0 ) = x0 where x = (x. We can rewrite our system in vector form dx = f(t. dt with initial conditions x(t0 ) = x0 and y(t0 ) = y0 . . k = 1. 2. . We are now in a position to calculate some solutions to Duffing’s equation.20). then Euler’s method now becomes xk+1 = xk + hfk or xk+1 = xk + hf (tk . . . x2 . The initial conditions are used to determine f0 . . y0 ). NUMERICAL TECHNIQUES FOR SYSTEMS 107 dy = g(t. and x0 = (x0 . .

x0 . 0. 1.23). SYSTEMS OF DIFFERENTIAL EQUATIONS dv = −0. we obtain and ( ) ( ) ( ) x1 x0 f (t0 . we can generate enough points to generate a graph of the solution (Figure 2. For small initial velocities. we will have a disaster. x(t) v(t) 3 2 1 t 10 20 30 40 50 -1 -2 Figure 2. Then (x0 . 30 x(t) v(t) 25 20 15 10 5 t 5 10 15 Figure 2.25.7653. x0 . solutions spiral towards the origin.24: Solutions for Duffing’s equation with initial conditions x(0) = 0 and v(0) = 1.1v. x0 . 0).23: Solutions for Duffing’s equation with initial conditions x(0) = 0 and v(0) = 1.02x3 − 0.23 ). If the crane sways too violently. we obtain a very different graph of the solution (Figure 2.108 CHAPTER 2. v0 ) = (0. Letting the step size be h = 0. However. y0 ) = (1. y0 ) = +h y1 y0 g(t0 . A slight change in the initial conditions can yield dramatically different solutions If we change the initial velokcity to v(0) = 1.1. dt with initial conditions (t0 .3x + 0. Using a numerical algorithm.7652 The surprising thing about Duffing’s equation is that it is so sensitive to initial con- ditions. y0 ) .7652). a larger initial velocity will send the solution in the phase plane away from the origin.7653 Example 2. Let us consider the system x′ = x − y + t y ′ = −x + y 2 − t2 x(0) = 1 y(0) = 0.

1000 1 0.1000 0. −0.5160 −0.3. y1 ) = +h y2 y1 g(t1 .1 1. yk ) h2 x′′k = +h + + ··· .6 2. however.1) 0 −1 ( ) 1. NUMERICAL TECHNIQUES FOR SYSTEMS 109 ( ) ( ) 1 1 = + (0. yk ) yk hg(tk . yk ) 2 yk′′ To get a more accurate approximation.1 Using this procedure.6097 −0.3 Taylor Series Methods Just as in the case of a single first-order differential equation.0000 −0.21). we can generate a list of values that will approximate the solution to our system (Table 2.1205 −0. x′ = x − y + t y ′ = −x + y 2 − t2 .5918 −1. we must compute x′′ .0990 3 0. y1 ) ( ) ( ) 1 1 = + (0.2. −0.3 1.1004 4 0.1990 −0.1 Similarly.1) 0 −1 ( ) 1.8338 0.9 3.0000 0. yk g(tk .1143 7 0.7 2.3324 −0.26: Euler’s approximation for a system of equations Notice that our system is not autonomous and depends on time.0990 2 0.2300 0.1143 0.4467 0.5 1.1000 0.1629 −0.3984 −0.5920 0. we cannot graph the phase plane of this system.2 1. Therefore. k tk xk hf (tk .8 2. In this case.7240 −0.2980 −0. xk .1542 Table 2.8310 0.2390 −0.1300 −0.3871 −0.1000 −0. If we return our example.3930 0. xk . we can take the first two terms of the Taylor series.4480 −0.1079 6 0.2818 0.26). xk .2833 −0. ( ) ( ) ( ′) ( ) xk+1 xk xk h2 x′′k = +h ′ + + ··· yk+1 yk yk 2 yk′′ ( ) ( ) ( ) xk f (tk .1033 5 0.9780 −0.1223 8 0.0 1. ( ) ( ) ( ) x2 x1 f (t1 .1318 9 0.4 1.0 3.7978 0.3.1991 −0.8462 −0. 2.1 = . x1 .5018 −0. we can think of Euler’s approximation as the first two terms of the Taylor series expansion.1 = . yk ) 0 0. we can graph a solution curve (Figure 2.1425 10 1. x1 . xk .

xk . we can see how this is done. • Uniqueness tells us that two solutions cannot start at the same place. If x0 is the initial value at time t0 . there exists a unique solution for the initial value problem on some open interval t0 − ϵ < t < t0 + ϵ for some ϵ > 0.110 CHAPTER 2. yk ) h2 x′′k = +h + yk+1 yk g(tk . 2. −x + y − 2xy + 2y 3 − 2t2 y − 3t Our algorithm now becomes clear. xk . • If we have an autonomous system x′ = f(x). SYSTEMS OF DIFFERENTIAL EQUATIONS x(0) = 1 y(0) = 0.3. Geometrically.4 A Word about Existence and Uniqueness Theorem 2. yk ) 2 yk′′ ( ) xk + h(xk − yk + tk ) + h2 (2xk − yk − yk2 + tk + t2k )/2 = . this algorithm requires us to compute some derivatives.27. • If we are interested in a certain system of differential equations. Let dx = f(t. see [12]. The consequences of the Existence and Uniqueness Theorem are much the same as they were for first-order differential equations. ( ) ( ) ( ) ( ) xk+1 xk f (tk . our solution does not depend on time. Thus. yk + h(−xk + yk2 − t2k ) + h2 (−xk + yk − 2xk yk + 2yk3 − 2t2k yk − 3tk )/2 Of course. 2. we obtain the same solution curve if we start at the same point even though we might start at different times.3. note that ( ′′ ) ′′ x x = y ′′ ( ) x′ − y ′ + 1 = −x′ + 2yy ′ − 2t ( ) (x − y + t) − (−x2 + y 2 − t2 ) = −(x − y + t) + 2y(−x + y 2 − t2 ) − 2t ( ) 2x − y − y 2 + t + t2 = . this implies that solution curves cannot cross. First. For a proof of existence and uniqueness for systems of differential equations.5 Important Lessons . x) dt be a system of differential equations such that f is continuously differentiable. it is very nice to know that a unique solution exists.

dx = v. yk )h. our solution does not depend on time. Geo- metrically. dt dt We can rewrite this equation as a first-order system. • If we have an autonomous system x′ = f(x). Thus. we can approximate a solution to the system by xk+1 = xk + f (xk . 2. dt with initial condition (x0 .2. Consider the system x′ = 2x y′ = y with initial conditions x(0) = 1 and y(0) = 3. we obtain the same solution curve if we start at the same point even though we might start at different times. dt dv = −qx − pv. or Runge-Kutta methods. NUMERICAL TECHNIQUES FOR SYSTEMS 111 • A damped harmonic oscillator can be described by the second-order equation d2 x dx 2 +p + qx = 0. • Provided certain conditions are met. yk )h yk+1 = yk + g(xk . Taylor series methods. dt dt is an example of a differential equation that is very sensitive to initial conditions. y0 ) and step size h. y) dt dy = g(x. • Uniqueness tells us that two distinct solutions cannot start at the same place. dt • Duffing’s equation. we are guaranteed unique local solutions to any system of first-order differential equations.3. We can extend this method to the improved Euler’s method.3 3 You will find Sage very useful for this exercise .3. • Many systems of equations cannot be solved analytically. y). d2 x dx 2 +p + qx = x3 .6 Exercises 1. Some of the following are consequences of existence and uniqueness. we can use numer- ical techniques such as Euler’s Method to find approximate solutions for the system. this implies that solution curves cannot cross. However. Given the system dx = f (x.

If we let x = x(t) be the amount of salt in Tank A at time t and y = y(t) be the amount of salt in Tank B at time t. we can describe the rate of change in each tank with a differential equation. (c) Use Euler’s method with step size ∆t = 0. Suppose that we have two tanks (A and B) between which a mixture of brine flows (Figure 2.4. and check how close the approximation is to the real solution when t = 2. 2. and t = 6. t = 4. dx x 5 = −500 · = − x. These problems often arise in applications—for example.5 to approximate this solution. 2. dt 300 3 dy x y y 5 5 = 500 · − 500 · − 500 · = x − y. Thus.3. Tank A contains 300 liters of water in which 100 kilograms of salt has been dissolved and Tank B contains 300 liters of pure water. 3et ) satisfies the initial value problem.1 Mixing Problems We will use linear systems of differential equations such as dx = ax + by dt dy = cx + dy dt to illustrate how we can use systems of differential equations to model how substances flow back and forth between two or more compartments. t = 4.112 CHAPTER 2. how chemicals move between tanks in a refinery or a brewery. or how pollutants move between a series of lakes or pounds.4 Solving Systems Analytically Mixing problems model how substances flow back and forth between two or more compart- ments. Systems of differential equations can prove very useful when it comes to modeling such situations. Hint. and t = 6. then we know that x(0) = 100 y(0) = 0 We know that the salt concentrations in the two tanks are x/300 kilograms per liter and y/300 kilograms per liter. and brine is pumped into Tank B from Tank A at the rate of 500 liters per hour.7 Project—Comparing Numerical Methods 2. dt 300 300 300 3 3 . SYSTEMS OF DIFFERENTIAL EQUATIONS (a) Show that x(t) = (e2t . All brine mixtures are well-stirred. (b) Use Euler’s method with step size ∆t = 0.28). we might want to model how greenhouse gases flow back and forth between different layers of the earth’s atomo- sphere[17]. and check how close the approximation is to the real solution when t = 2.1 to approximate this solution. Brine is also drained at a rate 500 liters of brine per hour from Tank B. (d) Discuss how and why the Euler approximations differ from the real solution. Fresh water is pumped into Tank A at the rate of 500 liters per hour.

28: Mixing example with two tanks We can now ask how we might solve the system of equations dx 5 = − x. if we have a system of the form dx = f (x). For example. dt 3 3 The task of solving the system dx = f (x.2. dt may be quite difficult or even impossible. dt then each equation is an autonomous first-order equation. we say the a partially coupled system is a system of the form dx = f (x). dt 3 dy 5 5 = x − y. y).4. y). dt dy = g(x. dt dy = g(x. we can solve this equation separately. Consider the system dx =x dt dy = x + y. Generalizing slightly. we can find solutions in certain cases. y). and substitute our solution into the second equation. we only need to solve two first-order equations. dt . Such a system is said to be decoupled. dt Since the first equation is an autonomous first-order equation in x. However. dt dy = g(y). To solve our system. SOLVING SYSTEMS ANALYTICALLY 113 500 l/hr 300 l 300 l 500 l/hr Tank A Tank B 500 l/hr Figure 2.

dt dt 3 3 Integrating both sides of this last equation gives us 500 e5t/3 y = t + c2 . dx/dt = x.114 CHAPTER 2. This equation has an integrating factor µ(t) = e−t . Thus. Revisiting the mixing problem that we posed at the beginning of this section. dt 3 3 Multiplying both sides of this last equation by the integrating factor µ(t) = e5t/3 yields d 5t/3 dy 5 500 (e y) = e5t/3 + e5t/3 y = . to obtain x = aet . Our second equation now becomes dy 5 5 500 −5t/3 5 = x− y = e − y. dt 3 3 3 3 This last equation is a first-order linear equation dy 5 500 −5t/3 + y= e . we can determine that c2 = 0. we have the following initial value problem. we can determine that c1 = 100 and x(t) = 100e−5t/3 . dt 3 dy 5 5 = x − y. 3 Using our initial condition y(0) = 0. the solution to our second equation is y(t) = atet + bet . 500 −5t/3 y= te 3 . Using this information in the second equation. SYSTEMS OF DIFFERENTIAL EQUATIONS We can easily solve the first equation. dx 5 = − x. dt 3 3 x(0) = 100. and ( ) d −t dy (e y) = µ(t) − y = aet µ(t) = a. Solving dx/dt = −(5/3)x is easy. We can quickly determine that x(t) = c1 e−5t/3 . we have dy − y = aet dt which is a first-order linear equation. y(0) = 0. dt dt Therefore. Applying the initial condition x(0) = 100.

y_start ]) sol [0].t) == g sol = desolve_system ([ de1 . the spring constant is 2.2. x_start = input_box ( default = 1) . show () pretty_print ( html (r"$\ displaystyle \ frac { dx }{ dt } ␣ = ␣ % s$ " % latex (f))) pretty_print ( html (r"$\ displaystyle \ frac { dy }{ dt } ␣ = ␣ % s$ " % latex (g))) 2. The following is a Sage interact that will solve the initial-value problem x′ = 3x − 4y y ′ = 2x + y x(0) = 1 y() = 1.4. however. We can use Sage to verify that the solution to our system is x(t) = e−t − e−2t .4.t) == f de2 = diff (y . However. de2 ] . show () sol [1]. where m = 1. . since we cannot use the same strategies that we used to solve partially- coupled systems. g = 2* x + y . t = var ( ' t ' ) x = function ( ' x ' )(t) y = function ( ' y ' )(t) @interact def _(f = 3* x . dt dv = −2x − 3v. x_start . dt Suppose that x(0) = 0 is the initial position of the mass and v(0) = 1 is the initial velocity. we can use Sage to solve linear systems for the moment. We will devote all of Chapter 3 to find an answer. We can rewrite this equation as a first-order linear system.4*y . ics =[ t_start .3 Harmonic Oscillators The equation d2 x dx 2 +3 + 2x = 0 dt dt is a specific case of a damped harmonic oscillator. t_start = input_box ( default = 0) . y ] . SOLVING SYSTEMS ANALYTICALLY 115 2. we can change the coefficients and initial value to be anything that we like. dx = v. y_start = input_box ( default = 1) ) : de1 = diff (x .2 Using Sage to Solve Linear Systems Solving linear systems such as x′ = ax + by y ′ = cx + dy is much more difficult.4. [x . and the damping constant is 3.

t = var ( ' t ' ) x = function ( ' x ' )(t) v = function ( ' v ' )(t) t_start = 0 x_start = 0 v_start = 1 de1 = diff (x . [x .29: An over-damped harmonic oscillator Now let us relax the damping and increase the spring constant on our harmonic oscillator.8 0. v ] . de2 ] . a spring-mass system that is modeled by this system of equations has so much damping that the mass will not oscillate.t) == v de2 = diff (v .2 t 1 2 3 4 5 Figure 2. Notice that our initial conditions have not changed.t) == -2* x . show () x(t). SYSTEMS OF DIFFERENTIAL EQUATIONS v(t) = −e−t + 2e−2t . This is an example of an over-damped harmonic oscillator (Figure 2. dt dv = −10x − 2v dt x(0) = 0 v(0) = 1. dt dt The corresponding linear system is dx = v.4 0. ics =[ t_start .29). v_start ]) sol [0]. x_start . In other words. t = var ( ' t ' ) . v(t) 1 x(t) = e −t − e −2t v(t) = − e −t + 2e −2t 0.3* v sol = desolve_system ([ de1 .116 CHAPTER 2. show () sol [1]. We again use Sage to solve our system.6 0. d2 x dx 2 +2 + 10x = 0.

8 0. v(t) 1 x(t) = (1−/t 3)e −t sin3t −t v(t) = e cos3t − (1/3)e sin3t 0. [x . Such a system can be solved by solving each equation inde- pendently.4 Figure 2.t) == -10* x .t) == v de2 = diff (v . v ] . x_start . dt dy = g(y) dt is said to be decoupled. 3 1 v(t) = e−t cos 3t − e−t sin 3t.30: An under-damped harmonic oscillator 2. de2 ] . v_start ]) sol [0].4.2 -0.2* v sol = desolve_system ([ de1 . This system is underdamped (Figure 2. dt . x(t).30).4 Important Lessons • A system of the form dx = f (x).2.6 0. 3 Notice that the system oscillates due to the sine and cosine terms in the solution. SOLVING SYSTEMS ANALYTICALLY 117 x = function ( ' x ' )(t) v = function ( ' v ' )(t) t_start = 0 x_start = 0 v_start = 1 de1 = diff (x . • A system of the form dx = f (x). show () The solution to our system is 1 x(t) = e−t sin t. ics =[ t_start .4 0.4.2 t 1 2 3 4 5 -0. show () sol [1].

• We can use Sage to solve systems of the form x′ = ax + by y ′ = cx + dy. Since the first equation is an autonomous first-order equa- tion in x. we can solve this equation separately.118 CHAPTER 2.4. 2. SYSTEMS OF DIFFERENTIAL EQUATIONS dy = g(x. . Hint.5 Exercises 1. • Sage is useful for investigating the behavior of harmonic oscillators. This is an exercise. y) dt a partially coupled system. and substitute our solution into the second equation.

1 Matrices and Systems of Linear Equations A short review of linear algebra and 2 × 2 matrices is useful at this point. (3. ( ) a b A= . Recall that any system of two equations in two variables. y ′ (t) cx(t) + dy(t) c d y(t) In other words.1) c d y cx + dy β We will denote the 2 × 2 coefficient matrix by A. can be written as a matrix equation ( )( ) ( ) ( ) a b x ax + by α = = . That is. cx + dy = β.1.1 Linear Algebra in a Nutshell Linear algebra and matrices provide a convenient notation for representing the 2 × 2 system dx = ax + by. dt where ( ′ ) ′ dx x (t) x = = . dt y ′ (t) 3. dt If we let ( ) ( ) a b x(t) A= and x(t) = . 3 Linear Systems 3. we can write our system as dx = Ax. ax + by = α.. c d y(t) then we can rewrite our system as ( ′ ) ( ) ( )( ) x (t) ax(t) + by(t) a b x(t) = = . dt dy = cx + dy. c d 119 .

they do lie on the same line. Then x = −(b/a)y. we will always get a line of solutions through the origin as long as at least one entry in our matrix is nonzero. z2 ) as ( ) ( ) ( ) z1 x1 y =α +β 1 . it is easy to find. x2 ) and y = (y1 .1. Let x = (x1 . The unique solution is given by ( ) ( )−1 ( ) x a b α = . x2 ) and y = (y1 . given two linearly independent vectors x = (x1 . To see why this is true. Theorem 3. since it will not play a role in our study of linear equations. x = 0 and y = 0. we can write z = (z1 . then we can write any vector in R2 as a unique linear combination of the two vectors.120 CHAPTER 3. LINEAR SYSTEMS If a solution for the system (3. the solutions to our system lie along a line through the origin. Equivalently. (ad − bc)y = 0. a Therefore. A unique solution will occur exactly when the matrix A is invertible (or nonsingular). and ( ) b −c y + dy = 0. we must solve the equations z1 = αx1 + βy1 1 We will not worry about the 2×2 zero matrix. Thus. then we either have no solution or infinitely many solutions. y2 ). Let us consider the special case ( ) ( ) x 0 A = . y2 ). Suppose that a ̸= 0.2 Linear Independence We say that two vectors x and y in R2 are linearly independent if they do not lie on the same line through the origin. ad − bc −c a The matrix A is invertible if and only if its determinant is nonzero. If det(A) = 0. we have exactly one solution. In fact. We leave the proof of the following theorem as an exercise. z2 x2 y2 where α and β are unique. On the other hand.1 3.1. . on the other hand. If. y c d β where ( ) 1 d −b A−1 = . y 0 If det(A) ̸= 0. if det(A) = 0. Since det(A) = ad − bc = 0. det(A) = ad − bc ̸= 0. That is. Then x and y are linearly independent if and only if ( ) x1 y1 det ̸= 0. we have infinitely many solutions. two vectors are linearly dependent if one vector is a multiple of the other. then the vectors are linearly dependent.1) exists. the variable y can assume any value and x = −(b/a)y. x2 y2 If we have a pair of linearly independent vectors in R2 .

3. LINEAR ALGEBRA IN A NUTSHELL 121 z2 = αx2 + βy2 for α and β.1. any two linearly independent vectors will form a basis for R2 . you should work through the exercises at the end of this section. then we can write z = 2v1 − 3v2 . 0 This matrix equation is certainly true if (x. 1) and v2 = (3. The constant λ is called an eigenvalue of A. 3. This will occur exactly when the determinant of ( ) a−λ b A − λI = c d−λ . Example 3. 1) and (−1. 1) form a basis for R2 . then we can write z = z1 e1 + z2 e2 . Indeed.3 Finding Eigenvalues and Eigenvectors A nonzero vector v is an eigenvector of A if Av = λv for some λ ∈ R. However. Let v1 = (2. However. z2 ).4. y) = (0. 0). v2 }. We say that the coordinates of z are (2. Example 3. −4). The vectors e1 = (1. If z = (−5.3. Letting ( ) ( ) a b x A= and v = ̸= 0. 2). x2 y2 Two vectors are said to be a basis for R2 if we can write any vector in R2 as a linear combination of these two vectors. −1) do not form a basis for R2 since these two vectors lie along the same line. 1 2 these vectors form a basis for R2 . we seek nonzero solutions to this system. 0) and e2 = (0. The vectors e1 and e2 are called the standard basis for R2 . this system has a unique solution since ( ) x1 y1 det ̸= 0. In matrix form this is ( )( ) ( ) ( )( ) ( )( ) a b x x a b x λ 0 x −λ = − c d y y c d y 0 λ y ( )( ) a−λ b x = c d−λ y ( ) 0 = . The vectors (1. Example 3. Since ( ) 2 3 det ̸= 0. c d y we have Ax = λx or Ax − λx = 0. By our arguments above.2.1. If 2 × 2 matrices and the rest of what we have described above make you nervous. if z = (z1 . −3) with respect to the basis {v1 .

y y If we let I denote the 2 × 2 identity matrix. In our example. . This system is equivalent to (1 − λ)x + 2y = 0 4x + (3 − λ)y = 0 which can be reduced to (1 − λ)x + 2y = 0 (λ2 − 4λ − 5)y = 0. ( ) 1−λ 2 det(A − λI) = det 4 3−λ = (1 − λ)(3 − λ) − 8 = λ2 − 4λ − 5 = (λ − 5)(λ + 1). Example 3. 4 3 To find the eigenvalues and eigenvectors for A.6. 0 1 we can rewrite this equation in the form ( ) ( ) x 0 (A − λI) = .2) y 0 We know that A − λI is a 2 × 2 matrix and that this system will only have nonzero solutions if det(A − λI) = 0. We summarize the results of this discussion in the following theorem. LINEAR SYSTEMS is zero. we will rewrite equation (3. λ = 5 or −1. Therefore.5. The roots of the characteristic polynomial of A are the eigenvalues of A. (3. Theorem 3.2) as x + 2y = λx 4x + 3y = λy. either λ = 5 or λ = −1 to obtain a nonzero solution. Suppose that we wish to find the eigenvalues and associated eigenvectors of ( ) 1 2 A= . c d−λ We say that det(A − λI) = λ2 − (a + d)λ + (ad − bc) is the characteristic polynomial of A. we must solve the equation ( ) ( ) x x A =λ . To see this from a different perspective. ( ) 1 0 I= . In this case ( ) a−λ b det(A − λI) = det = λ2 − (a + d)λ + (ad − bc).122 CHAPTER 3. Thus.

and the eigenvectors corresponding to this eigenvalue are the nonzero solutions of this equation. If ( ) ( ) 2 1−i 4i 1−i A= and B= . . 2−i 2 1 + 3i −2 − i find each of the following. LINEAR ALGEBRA IN A NUTSHELL 123 • If λ = 5. 2. for a matrix A. If ( ) ( ) 1 −2 4 1 A= and B = . If ( ) ( ) ( ) 3 −2 4 −2 A= .x = . 3. of a matrix A are the eigen- values of A. Given a specific eigenvalue. y 0 • If v1 and v2 are eigenvectors of two distinct real eigenvalues of a matrix A. (a) A + B (b) 3A − 2B (c) AB (d) BA Hint.1.1. a vector must be a nonzero multiple of (1. the eigenvectors associated with A are the nonzero solutions of the system of equations ( ) ( ) x 0 (A − λI) = .1. λ. 2) to be an eigenvector of A corresponding to λ = 5.4 Important Lessons • The roots of the characteristic polynomial. 3 1 −1 −2 find each of the following. (a) A + B (b) 2A − 3B (c) AB (d) BA (e) A−1 (f) B −1 Hint.3. 3. and y = . • If λ = −1. then v1 and v2 are linearly independent. then the corresponding eigenvectors are the nonzero multiples of (1.5 Exercises 1. −1). det(A − λI). 3. the first equation in the system becomes −2x + y = 0. 0 −1 1 3 find each of the following. That is.

( -2 . A = matrix ([[1 . 3] . [1 . x2 ) and y = (y1 . -2] Sage will also allow us find eigenvectors for each of the eigenvalues of A. −1 3 in the Sage cell below. -1]]) A . 4. 1) . -1]]) A [ 1 3] [ 1 -1] We can use the following command to find the eigenvalues of A. eigenvalues () [2 . LINEAR SYSTEMS (a) Ax (b) Ay (c) xT y (d) yT x Hint. eigenvectors_right () . 3. Let x = (x1 .1. = x2 y2 Hint. -1]]) A . [(1 . −1). 1) ] Thus. -1) ] . 1). [1 . 3]]) B . 1/3) and λ2 = −2 with eigenvector v2 = (1. 1/3) ].6 Project—Finding Eigenvalues and Eigenvectors Sage can be used to find eigenvalues and eigenvectors for a matrix A for now. 1] . 1 −1 Using Sage. Consider the matrix ( ) 1 3 A= . B = matrix ([[1 . we obtain a single repeated eigenvalue λ = 2 and only one eigenvector v = (1. Prove that x and y are linearly independent if and only if ( ) x1 y1 det ̸ 0. In the previous example. In our previous examples. we obtained two linearly independent eigenvalues allowing us to solve initial value problems given a general solution. [1 . the multiplicity is 1. we would enter the matrix A as follows. In the matrix ( ) 1 1 B= .124 CHAPTER 3. y2 ). eigenvectors_right () [(2 . 3] . A = matrix ([[1 . A = matrix ([[1 . 3] . There is a third entry in the Sage output which refers to the multiplicity of the eigenvalue.[ -1 . The multiplicity of this eigenvalue is 2. the matrix A has two eigenvalues: λ1 = 2 with eigenvector v1 = (1. [(1 .

3) dt dy = cx + dy.2 Planar Systems A first-order linear system of n equations and n variables is any system that can be written in the form dx1 = a11 (t)x1 (t) + · · · + a1n (t)xn (t) + f1 (t). −i) and v = (1. dt If each of the coefficients is constant and the functions fi vanish.3. dt . 1) . ..2. (3.7).4) dt 3. [(1 .2.1*I . (3. Tank A contains 300 liters of water in which 100 kilograms of salt . 1] . 1) ] 3. 1) ]. 2) ] We may also matrices such as ( ) 4 1 C= −1 4 has complex eigenvalues. eigenvectors_right () [(4 . C = matrix ([[4 . then we have a homoge- neous first-order linear system with constant coefficients. [(1 . PLANAR SYSTEMS 125 [(2 . [(1 . -1* I)]. Eigenvectors for λ and µ are u = (1. dx = ax + by. dt .1 Planar Systems and 2 × 2 Matrices We will use linear systems of differential equations to illustrate how we can use systems of differential equations to model how subtances flow back and forth between two or more compartments. dxn = an1 (t)x1 (t) + · · · + ann (t)xn (t) + fn (t).[ -1 . respectively. i).. (4 + 1* I . 1* I ) ] . dx1 = a11 x1 + · · · + a1n xn dt dx2 = a21 x1 + · · · + a2n xn . λ = 4 − i and µ = 4 + i. dt dx2 = a21 (t)x1 (t) + · · · + a2n (t)xn (t) + f2 (t). dt We will concentrate on 2 × 2 homogeneous first-order linear systems or planar systems for the time being. . dxn = an1 x1 + · · · + ann xn . Suppose that we have two tanks (A and B) between which a mixture of brine flows (Figure 3. 4]]) C .

and brine is pumped into Tank B from Tank A at the rate of 900 liters per hour. y ′ (t) cx(t) + dy(t) c d y(t) In other words.7: Mixing example with two tanks Matrix notation gives us a convenient way of representing the 2 × 2 system (3. c d y(t) then we can rewrite our system as ( ′ ) ( ) ( )( ) x (t) ax(t) + by(t) a b x(t) = = . If we let ( ) ( ) a b x(t) A= and x(t) = . LINEAR SYSTEMS has been dissolved and Tank B contains 300 liters of pure water. All brine mixtures are well-stirred. dt 300 300 300 500 l/hr 900 l/hr 300 l 300 l 400 l/hr Tank A Tank B 500 l/hr Figure 3.3)–(3. and an additional 500 liters of brine per hour is drained from Tank B. then we know that x(0) = 100 y(0) = 0 We know that the salt concentrations in the two tanks are x/300 kilograms per liter and y/300 kilograms per liter. Fresh water is pumped into Tank A at the rate of 500 liters per hour. we can describe the rate of change in each tank with a differential equation.126 CHAPTER 3. dx x y 4 = −900 · + 400 · = −3x + y. dt 300 300 3 dy x y y = 900 · − 400 · − 500 · = 3x − 3y. If we let x = x(t) be the amount of salt in Tank A at time t and y = y(t) be the amount of salt in Tank B at time t. Brine is also pumped back into Tank A from Tank B at the rate of 400 liters per hour. Thus. dt where ( ′ ) dx x (t) x′ = = . we can write our system as dx = Ax. dt y ′ (t) .3).

2. 0) is the unique equilibrium solution for the system.9.3. Then the function x(t) = eλt v0 is a solution of the system x′ = Ax. Now let us attack the problem of finding all of the solutions of the system x′ = Ax. If det(A) = 0.2. In this case. then the equilibrium solutions for the system form a straight line in R2 . y0 ) if ax0 + by0 = 0. In other words. We say that the solution x(t) = eλt v0 is a straight-line solution. . Theorem 3. We are now ready to state the results of our discussion in a theorem. we compute x′ (t) = λeλt v0 = eλt (λv0 ) = eλt (Av0 ) = A(eλt v0 ) = Ax(t). cx0 + dy0 = 0. A(αv0 ) = αAv0 = α(λv0 ) = λ(αv0 ). Let v0 be an eigenvector for the matrix A with associated eigenvalue λ. This is a very special case of course. Theorem 3. where A is not the zero matrix. the matrix A just sends the vector v0 to a vector on the same line through the origin. then (x. 2. then any nonzero multiple of v0 is also an eigenvector for A. To see that this is indeed the case. λv0 . y) = (0. PLANAR SYSTEMS 127 3.8. Suppose that we can find a nonzero vector v0 such that Av0 = λv0 for some real number λ. 1.2 Systems of Differential Equations A linear planar system x′ = ax + by y ′ = cx + dy has an equilibrium solution at (x0 . The vector eλt v0 lies on the same line for each value of t. If det(A) ̸= 0. The following proposition tells us exactly where to find the equilibrium solutions of a linear system with constant coefficients. Let dx = Ax dt be a 2 × 2 linear system. Note that if v0 is an eigenvector for A. however. we claim that x(t) = eλt v0 is a solution for our linear system if we can find such a vector. the key to solving a linear system x′ = Ax is to be able to find eigenvalues and eigenvectors for the matrix A.

then we can determine c1 and c2 by solving the linear system of equations 3c1 + c2 = 0 c1 − c2 = 1 to get c1 = 1/4 and c2 = −3/4. 1 −1 1 2 1 Similarly. −1) is an eigenvalue for A with associated eigenvector µ = −2. say x(0) = 0 and y(0) = 1. Consider the system x′ = x + 3y y ′ = x − y. If we are given initial conditions. Thus. −1 Since d d d (c1 x1 (t) + c2 x2 (t)) = c1 x1 (t) + c2 x2 (t) dt dt dt = c1 Ax1 (t) + c2 Ax2 (t) = A(c1 x1 (t) + c2 x2 (t)). 1) with associated eigenvalue λ = 2. where ( ) 1 3 A= . the solution to our initial value problem is 3 3 −2t x(t) = e2t − e 4 4 1 3 −2t y(t) = e2t + e . Thus. Thus.128 CHAPTER 3. 1 −1 The matrix A has an eigenvector u = (3. We can rewrite this system in matrix form as x′ = Ax. 1 and the solution ( ) −2t 1 x2 (t) = e . LINEAR SYSTEMS Example 3.10. we have two solutions for our system: the equilibrium solution at the origin. v = (1. 4 4 . since ( )( ) ( ) ( ) 1 3 3 6 3 Au = = =2 = λu. any linear combination of solutions to a linear system is also a solution. the solution ( ) 3 x1 (t) = e2t . a general solution to our system is ( ) ( ) 2t 3 −2t 1 x(t) = c1 e + c2 e 1 −1 or x(t) = 3c1 e2t + c2 e−2t y(t) = c1 e2t − c2 e−2t .

x2 (t) = e−5t v. ( −t ) ( ) 2e −2e−5t x(t) = c1 + c2 3e−t 3e−5t is a solution to our system. We now have the solution that we seek. 3 −3 It is easy to check that we have eigenvectors u = (2. Thus. Thus. x′ = Ax. Since any linear combination of solutions is also a solution. we have the following initial value problem. then x′1 = Ax1 x′2 = Ax2 . Using the initial values x(0) = 100 and y(0) = 0.3. x1 (t) = e−t u.2. then any linear com- bination of solutions to the linear system x′ = Ax is also a solution. for any two real numbers c1 and c2 . 3) with eigenvectors λ = −1 and µ = −5.11 (Principle of Superposition). x(t) = 50e−t + 50e−5t y(t) = 75e−t − 75e−5t . dt x(0) = 100. Theorem 3. If we write our system in matrix form. we have two solutions to our system. dt 3 dy = 3x − 3y. We state this result in the following theorem. y(0) = 0. d d d (c1 x1 (t) + c2 x2 (t)) = α x1 (t) + c2 x2 (t) dt dt dt = c1 Ax1 (t) + c2 Ax2 (t) = A(c1 x1 (t) + c2 x2 (t)). PLANAR SYSTEMS 129 If x1 (t) and x2 (t) are solutions to the linear system x′ = Ax. dx 4 = −3x + y. we can determine that c1 = 25 and c2 = −25. . then ( ) −3 4/3 A= . If A is a 2 × 2 matrix. respectively. Revisiting the mixing problem that we posed at the beginning of this section. 3) and v = (−2.

we know that v1 and v2 form a basis for R2 .3 Solving Linear Systems Our goal is to prove the following theorem. we obtain λ2 v1 = αλ2 v2 (3. then there exists α ̸= 0 such that v1 = αv2 . we know that x(t) = c1 eλ1 t v1 + c2 eλ2 t v2 .7).12. we can write x0 as a linear combination of v1 and v2 .13. Then we can write y(t) = f (t)v1 + g(t)v2 . x(0) = x0 .6) and (3. It remains to show that x(t) = c1 eλ1 t v1 + c2 eλ2 t v2 is the unique solution to the system x′ (t) = Ax(t).7) if we multiply both sides of (3.5) Multiplying both sides of this equation by A. λ1 and λ2 and eigenvectors v1 and v2 . (3. That is. Then v1 and v2 are linearly independent.5) by λ2 . . λ1 and λ2 . we have λ1 v1 = Av1 = αAv2 = αλ2 v2 . Proof. we can find c1 and c2 such that x0 = x(0) = c1 v1 + c2 v2 . y0 ). λ1 and λ2 . Suppose that A has a pair of distinct real eigenvalues. To show that this is the general solution.130 CHAPTER 3. Using (3. LINEAR SYSTEMS 3.13.2. respectively. this contradicts the assumption that λ1 and λ2 are distinct.6) One the other hand. If v1 and v2 are linearly dependent. g(0) = c2 . Then the general solution of the linear system x′ = Ax is given by x(t) = c1 eλ1 t v1 + c2 eλ2 t v2 . we must show that we can choose c1 and c2 to satisfy a given initial condition x0 = x(0) = (x0 . Theorem 3. By Lemma 3. Lemma 3. with associated eigenvectors v1 and v2 . with associated eigenvectors v1 and v2 . we can conclude that (λ1 − λ2 )v1 = α(λ2 − λ2 )v2 = 0v2 = 0. Let A be an 2 × 2 matrix with a pair of distinct real eigenvalues. is a solution to the linear system x′ = Ax. By the Principle of Superposition. Suppose that we have a linear system x′ = Ax such that A has a pair of distinct real eigenvalues. In other words. We can now proceed to the proof of the theorem. (3. However. where f (0) = c1 . Suppose that there is another solution y(t) such that y(0) = x0 .

. 3. f ′ (t) = λ1 f (t). Solve the following initial-value problem. On the other hand. 3. and proof our theorem is complete. g(t) = c2 eλ2 t . 2.4 Important Lessons • If v1 and v2 are eigenvectors of two distinct real eigenvalues of a matrix A. x′ = −x + 3y y ′ = −3x − y x(0) = 3 y(0) = 2.2. PLANAR SYSTEMS 131 Since y(t) is a solution to our system of equations. then v1 and v2 are linearly independent. If A has a pair of distinct real eigenvalues.3. Solve the following initial-value problem. • Let A be a 2 × 2 matrix. g(0) = c2 . with associated eigenvectors v1 and v2 . we have two first-order initial value problems. The solutions of these initial value problems are f (t) = c1 eλ1 t . Thus. then the general solution of the linear system x′ = Ax is given by x(t) = αeλ1 t v1 + βeλ2 t v2 . and g ′ (t) = λ2 g(t). f (0) = c1 . x′ = x + 2y y ′ = −x + 4y x(0) = 3 y(0) = 2. we know that Ay(t) = y′ (t) = f ′ (t)v1 + g ′ (t)v2 .2.2. Consequently. respectively. λ1 and λ2 . Ay(t) = f (t)Av1 + g(t)Av2 = λ1 f (t)v1 + λ2 g(t)v2 .5 Exercises 1. • The Principle of Superposition tells us that any linear combination of solutions to the linear system x′ = Ax is also a solution. y(t) = x(t).

x′ = −2x + y y ′ = −9x + 4y x(0) = 5 y(0) = 3. Consequently. c(t) d(t) Even though this system is not linear. g(t)). y(t)). . Solve the following initial-value problem. y(t) u2 (t) v2 (t) If A has constant entries. dx = Ax + f. (a) Find the general solution of the homogeneous system x′ = x + 3y y′ = x − y (b) Find a particular solution for x′ = x + 3y + (t − 3t2 ) y ′ = x − y + (1 + t − t2 ) (c) Find the solution of x′ = x + 3y + (t − 3t2 ) y ′ = x − y + (1 + t − t2 ) x(0) = 1 y(0) = −1. f(t) = (f (t). Consider the linear system x′ = x + 3y + (t − 3t2 ) y ′ = x − y + (2 − t + t2 ) x(0) = 1 y(0) = −1. and ( ) a(t) b(t) A= . Now. Let us assume that the solution of dx/dt = Ax is given by ( ) ( ) ( ) x(t) u1 (t) v1 (t) xh = = c1 + c2 . we are often able to find solutions to this system. we need to find the solution of the corresponding homogeneous system and one particular solution of the nonhomogeneous system. Show that ( ) x(t) + ϕ1 (t) xh + xp = y(t) + ϕ2 (t) is the general solution to the system. we can certainly determine this solution if A has distinct real eigenvalues. Consider the system of differential equations. suppose that xp = (ϕ1 (t). to solve a nonhomogeneous system of linear differential equations.132 CHAPTER 3. LINEAR SYSTEMS 3. 4. ϕ2 (t)) is a solution to dx/dt = Ax + f. dt where x(t) = (x(t). 5.

6. Prove that ( ) ( ) 1 λt λt t αe + βe 0 1 is the general solution of ( ) ′ λ 1 x = x.15). .2.3. Furthermore. 1 −1 The eigenvalues of A are λ = −2 or λ = 2 with eigenvectors u = (1. If c2 = 0. where ( ) 1 3 A= . then the general solution of the system is x(t) = c1 eλt u + c2 eµt v. If A has distinct real eigenvalues λ and µ with eigenvectors u and v. the straightline solutions must lines containing u and v (Figure 3.3. then all solutions will lie long the line in the xy-plane that contains the vector u.1 The Case λ1 < 0 < λ2 Example 3.2. PHASE PLANE ANALYSIS OF LINEAR SYSTEMS 133 Hint. Consider the system x′ = ax + y y ′ = 2ax + 2y.3 Phase Plane Analysis of Linear Systems In Section 3. we learned how to solve the system ( ) ( )( ) ( ) dx/dt a b x x = =A dy/dt c d y y provided the system has distinct real eigenvalues. For what values of a do you find a bifurcation (a change in the type of phase portrait)? Sketch typical phase portraits for a values of a above and below the bifurcation point. Assume that your solution must be of the form ( 2 ) a2 t + a1 t + a0 xp = b2 t2 + b1 t + b0 . Similarly. −1) and v = (3. we can use the general solution of such a system to find the straight-line solutions to the system. 1). respectively.6 Project 3.14.3. The system x′ = x + 3y y′ = x − y can be written in matrix form x′ = Ax. Therefore. This is called the method of undetermined coefficients. then all solutions will lie along the line in the xy-plane that contains the vector u. 3. 0 λ 3. if c1 = 0. 7. where a ∈ R. respectively.

our general solution is ( ) ( ) 1 0 x(t) = c1 eλ1 t + c2 eλ2 t . However. 0) and (0. 0 λ2 Since this is a decoupled system. It is easy to see that we can associate eigenvectors (1.15: Straightline solutions Let us consider the special case of the system x′ = Ax. respectively. where λ1 < 0 < λ2 and ( ) λ1 0 A= .134 CHAPTER 3. 1) to λ1 and λ2 . 0 1 . Thus. LINEAR SYSTEMS 3 2 1 y(t) 0 -1 -2 -3 -3 -2 -1 0 1 2 3 x(t) Figure 3. dt we already know how to find the solutions. dx = λ1 x dt dy = λ2 y. and our eigenvalues are λ1 and λ2 . we will find the eigenvalues of A. in keeping with the spirit of our investigation. The characteristic equation of A is (λ − λ1 )(λ − λ2 ) = 0.

16: Saddle phase portrait Since λ1 < 0.14. The equilibrium point of such systems is called a saddle. If c2 ̸= 0. the straight-line solutions of the form c1 eλ1 t (1. For the system in Example 3. −1 and each solution on this line approaches the origin as t → ∞. the unstable line of solutions is ( ) 2t 3 x1 (t) = c1 e .18. Example 3. c2 eλ2 t ) as t → ∞. .17. we have x(t) → x2 (t) as t → −∞. Thus. the general solution to the system is ( ) ( ) 2t 3 −2t 1 x(t) = c1 e + c2 e . we have the phase portrait in Figure 3. we have x(t) → x1 (t) as t → ∞. and the y-axis is called the unstable line. These solutions approach zero as t → ∞. 0) lie on the x-axis. On the other hand. PHASE PLANE ANALYSIS OF LINEAR SYSTEMS 135 3 2 1 y(t) 0 -1 -2 -3 -3 -2 -1 0 1 2 3 x(t) Figure 3. The phase portrait for the system x′ = −x y′ = y is given in Figure 3. All other solutions ( ) ( ) λ1 t 1 λ2 t 0 x(t) = c1 e + c2 e 0 1 (with c1 . The x-axis is called the stable line.16. 1) line on the y-axis and approach infinity as t → ∞. 1 −1 If c1 ̸= 0. c2 ̸= 0) tend to infinity in the direction of the unstable line. the solutions c2 eλ2 t (0. since x(t) approaches (0.3. 1 Each solution tends away from the origin as t → ∞.3. By the Principle of Super- position. The stable line of solutions is given by ( ) 1 x2 (t) = c2 e−2t .

0 1 but unlike the case of the saddle. dx x (t) λ1 c1 e 1 λ1 c1 Since λ2 − λ1 > 0.18: Saddle phase portrait For the general case. we always have a stable line of solutions and an unstable line of solutions. To see how they solutions approach the origin. the solutions tend towards the origin tangentially to the y-axis (Figure 3. LINEAR SYSTEMS 3 2 1 y(t) 0 -1 -2 -3 -3 -2 -1 0 1 2 3 x(t) Figure 3. We say that the equilibrium point for this system is a sink. dy/dx. .2 The Case λ1 < λ2 < 0 Suppose λ1 < λ2 < 0 and consider the diagonal system ( ′ ) ( )( ) x (t) λ1 0 x(t) = . where A has eigenvalues λ1 < 0 < λ2 . the derivative.136 CHAPTER 3. must approach ±∞. we will compute dy/dx for c2 ̸= 0. All other solutions approach the unstable line as t → ∞ and the stable line as t → −∞. all solutions tend towards the origin as t → ∞. 3. Therefore. y ′ (t) 0 λ2 y(t) The general solution of this system is ( ) ( ) λ1 t 1 λ2 t 0 x(t) = c1 e + c2 e . If x(t) = c1 eλ1 t y(t) = c2 eλ2 t . provided c2 ̸= 0. then dy y ′ (t) λ2 c2 eλ2 t λ2 c2 (λ2 −λ1 )t = ′ = λ t = e .19).3.

Example 3. y) is given by dy λ1 c1 eλ1 t u2 + λ2 c2 eλ2 t v2 = dx λ1 c1 eλ1 t u1 + λ2 c2 eλ2 t v1 ( ) λ1 c1 eλ1 t u2 + λ2 c2 eλ2 t v2 e−λ2 t = λ1 c1 eλ1 t u1 + λ2 c2 eλ2 t v1 e−λ2 t λ1 c1 e(λ1 −λ2 )t u2 + λ2 c2 v2 = .19: Sink phase portrait Since λ1 < λ2 < 0. then we have the straight-line solution corresponding to the eigenvalue λ1 . respectively. all the solutions for this case (except those on the straight-line belonging to the dominant eigenvalue) tend toward the origin tangentially to the straight-line solution corresponding to the weaker eigenvalue. u2 ) and (v1 . respectively. If c2 = 0. Since the dominant eigenvalue is λ2 = −6. suppose that λ1 < λ2 < 0. Consider the system ( ′ ) ( )( ) x (t) −5 −2 x(t) = .20. v2 ). we say that λ1 is the dominant eigenvalue. To see what happens in the general case. . −1) more quickly (Figure 3.3. The x-coordinates of the solutions approach the origin much faster than the y-coordinates. y ′ (t) −1 −4 y(t) The eigenvalues of this system are λ1 = −6 and λ2 = −3 with eigenvectors v1 = (2. solutions tend towards the straightline solution containing the vector v2 = (1. PHASE PLANE ANALYSIS OF LINEAR SYSTEMS 137 3 2 1 y(t) 0 -1 -2 -3 -3 -2 -1 0 1 2 3 x(t) Figure 3.3. The general solution of our system is ( ) ( ) u1 v x(t) = c1 eλ1 t + c2 eλ2 t 1 . the eigenvectors associated with λ1 and λ2 are (u1 . Hence. λ2 . u2 v2 The slope of a solution curve at (x. λ1 c1 e(λ1 −λ2 )t u1 + λ2 c2 v1 This last expression tends toward the slope v2 /v1 of the eigenvector of λ2 (unless c2 = 0). 1) and v2 = (1.21). −1).

but the arrows are reversed (Figure 3.22).3 The Case λ1 > λ2 > 0 If λ1 > λ2 > 0.21: Source phase portrait 3. LINEAR SYSTEMS 4 3 2 1 y(t) 0 -1 -2 -3 -4 -4 -3 -2 -1 0 1 2 3 4 x(t) Figure 3. 3 2 1 y(t) 0 -1 -2 -3 -3 -2 -1 0 1 2 3 x(t) Figure 3. we say that the equilibrium point is a source.3. In this case.22: Source phase portrait .138 CHAPTER 3. we can regard our direction field as the negative of the direction field of the previous case. The general solution and the direction field are the same.

The general solution to the system x′ = Ax is ( ) ( ) λ1 t 1 λ2 t 0 x(t) = αe + βe . 0 λ2 then A has two distinct real eigenvalues.23. the equilibrium point of the system x′ = Ax is a source. • If ( ) λ1 0 A= . the equilibrium point of the system x′ = Ax is a saddle. 1). Consider the system ( ′ ) ( )( ) x (t) −5 −2 x(t) = . . 0 1 ◦ For the case λ1 < 0 < λ2 .3.24).3. Since the dominant eigenvalue is λ1 = 5. ◦ For the case λ1 < λ2 < 0. y ′ (t) −1 −4 y(t) The eigenvalues of this system are λ1 = 5 and λ2 = 1 with eigenvectors v1 = (3.3. PHASE PLANE ANALYSIS OF LINEAR SYSTEMS 139 Example 3. solutions are closer to the straightline solution containing the vector v2 = (3. the equilibrium point of the system x′ = Ax is a sink. dy/dt c d y y we can use the eigenvalues of A to find and classify the solutions of the system.24: Source phase portrait 3. −4) more as t → ∞ (Figure 3. respectively. −4) and v2 = (1. ◦ For the case 0 < λ1 < λ2 .4 Important Lessons • Given a system of linear differential equations ( ) ( )( ) ( ) dx/dt a b x x = =A . 4 3 2 1 y(t) 0 -1 -2 -3 -4 -4 -3 -2 -1 0 1 2 3 4 x(t) Figure 3.

4 3 2 1 y(t) 0 -1 -2 -3 -4 -4 -3 -2 -1 0 1 2 3 4 x(t) Figure 3.4 Complex Eigenvalues Consider the following system.1 Complex Eigenvalues Suppose that we have the system ( ) ( )( ) ( ) dx/dt 0 β x x = =A .3. Hint. and so we have imaginary eigenvalues ±iβ. 1 − i). respectively. LINEAR SYSTEMS 3.140 CHAPTER 3.3.4. However.5 Exercises 1. we . The roots of this polynomial are λ1 = −2 + i and λ2 = −2 − i with eigenvectors v1 = (1. 3. 1 + i) and v2 = (1.8) dy/dt −2 −1 y The characteristic polynomial of the system (3. It is clear from the phase portrait of the system that there are no straightline solutions (Figure 3. To find the eigenvector corresponding to λ = iβ.25). we would like to real solutions to a linear system with real coefficients. This is an exercise. ( ) ( )( ) dx/dt −3 1 x = (3.25: A system with no straightline solutions 3.8) is λ2 +4λ+5.6 Project 3. The characteristic polynomial of this system is det(A − λI) = λ2 + β 2 . dy/dt −β 0 y y where β ̸= 0.

4. we can find a complex eigenvector (1.3. Thus. Thus. both xRe (t) and xIm (t) are solutions to our system. 2 eiβt = cos βt + i sin βt. − sin βt cos βt Since x′Re (t) + ix′Im (t) = x′ (t) = Ax(t) = A(xRe (t) + ixIm (t) = AxRe (t) + iAxIm (t). since ( ) ( ) 1 0 xRe (0) = and xIm (0) = . . We claim that x(t) = c1 xRe (t) + c2 xIm (t) (3. Consequently. Moreover. That is. We can remedy the situation if we use Euler’s formula. COMPLEX EIGENVALUES 141 must solve the system ( )( ) ( ) −iβ β x 0 = . 0 1 we know that the appropriate linear combination of xRe (t) and xIm (t) will provide a solution to any initial value problem. If ( ) u(t) y(t) = v(t) 2 If you are unfamiliar with Euler’s formula. The problem is that we have a real system of differential equations and would like real solutions. ( ) iβt 1 x(t) = e i is a complex solution to the system x′ = Ax. −β −iβ y 0 however. try expanding both sides as a power series to check that we do indeed have a correct identity. we know that x′Re (t) = AxRe (t) and x′Im (t) = AxIm (t) by setting the real and imaginary parts equal. i).9) is a general solution to our system. this reduces to solving the equation iβx = βy. we must be able to write every solution of our system as a linear combination of xRe (t) and xIm (t). Let us rewrite our solution as ( ) iβt 1 x(t) = e i ( ) cos βt + i sin βt = i(cos βt + i sin βt) ( ) cos βt + i sin βt = − sin βt + i cos βt ( ) ( ) cos βt sin βt = +i − sin βt cos βt and consider the real and imaginary parts of the solution: ( ) ( ) cos βt sin βt xRe = and xIm = .

define f by f (t) = (u(t) + iv(t))eiβt . f (t) is a complex constant and f (t) = (u(t) + iv(t))eiβt = a + bi. Therefore. This type of system is called a center. LINEAR SYSTEMS is another solution to the system x′ = Ax. Consequently. We can now write u(t) + iv(t) = (a + ib)e−iβt . Therefore. then ( ′ ) ( )( ) ( ) ′ u (t) 0 β u(t) βv(t) y (t) = = = .26. Thus. u(t) = a cos βt + b sin βt v(t) = b cos βt − a sin βt. Example 3. ( ) ( ) u(t) a cos βt + b sin βt = v(t) b cos βt − a sin βt ( ) ( ) cos βt sin βt =a +b − sin βt cos βt = axRe (t) + bxIm (t).142 CHAPTER 3. u′ (t) = βv(t) and v ′ (t) = −βu(t). the general solution to the system is x(t) = c1 cos 2t + c2 sin 2t . v ′ (t) −β 0 v(t) −βu(t) In other words. The derivative of f is f ′ (t) = (u′ (t) + iv ′ (t))eiβt + iβ(u(t) + iv(t))eiβt = (βv(t) − iβu(t))eiβt + (iβu(t) + i2 βv(t))eiβt = 0. Notice that the solutions ( ) ( ) cos βt sin βt x(t) = c1 + c2 − sin βt cos βt are periodic with period 2π/β. Now. u(t) + iv(t) = (a + ib)e−iβt = (a + ib)(cos βt − i sin βt) = (a cos βt + b sin βt) + i(b cos βt − a sin βt). Therefore. Consider the initial value problem dx = 2y dt dy = −2x dt x(0) = 1 y(0) = 2. The eigenvalues of this system are λ = ±2i.

COMPLEX EIGENVALUES 143 y(t) = −c1 sin 2t + c2 cos 2t. The phase portrait is a circle of radius 2 about the origin (Figure 3. i) is an eigenvector for α + iβ. 3 2 1 y(t) 0 -1 -2 -3 -3 -2 -1 0 1 2 3 x(t) Figure 3.4. We can use the equation (α − (α + iβ))x + βy = 0 to show that (1.4. so the eigenvalues are λ = α ± iβ. The characteristic equation of A is λ2 − 2αλ + (α2 + β 2 ) = 0. the solution to our initial value problem is x(t) = cos 2t + 2 sin 2t y(t) = − sin 2t + 2 cos 2t. Using the initial conditions to solve for c1 and c2 . Therefore. we have a complex solution of the form ( ) 1 x(t) = e(α+iβ)t i . where ( ) α β A= −β α and α and β are nonzero real numbers.27).27: Phase portrait for a center 3.2 Spiral Sinks and Sources Now let us consider the system x′ = Ax.3.

Consider the initial value problem dx = −x/10 + y dt dy = −x − y/10 dt x(0) = 2 y(0) = 2. The eigenvalue λ = −1/10 + i has an eigenvector v = (1. i). Furthermore. − sin βt cos βt The eαt factor tells us that the solutions either spiral into the origin if α < 0 or spiral out to infinity if α > 0. xRe cannot be a multiple of xIm for all values of t. Thus. The complex solution of our system is ( ) (−1/10+i)t 1 x(t) = e i ( ) ( ) −t/10 cos t −t/10 sin t =e + ie − sin t cos t ( ) ( ) −t/10 cos t −t/10 sin t =e (cos t + i sin t) + ie (cos t + i sin t) − sin t cos t ( ) ( ) −t/10 − sin t + i cos t sin t 2 cos2 t + i cos t sin t = e−t/10 + e − cos t sin t − i sin2 t − cos t sin t + i cos2 t ( ) −t/10 (cos2 t − sin2 t) + 2i sin t cos t =e −2 sin t cos t + i(cos2 t − sin2 t) ( ) −t/10 cos 2t + i sin 2t =e − sin 2t + i cos 2t ( ) ( ) −t/10 cos 2t −t/10 sin 2t =e + ie − sin 2t cos 2t .144 CHAPTER 3. these solutions are linearly independent. The matrix ( ) −1/10 1 −1 −1/10 has eigenvalues λ = −1/10 ± i. both ( ) ( ) αt cos βt αt sin βt xRe (t) = e and xIm (t) = e − sin βt cos βt are real solutions to x′ = Ax. Example 3. Indeed.28. we have the general solution ( ) ( ) αt cos βt αt sin βt x(t) = c1 e + c2 e . respectively. LINEAR SYSTEMS ( ) ( ) αt cos βt αt sin βt =e + ie − sin βt cos βt = xRe (t) + ixIm (t). As before. In this case we say that the equilibrium points are spiral sinks and spiral sources.

29: Phase portrait for a spiral sink Example 3. − sin 2t cos 2t Applying our initial conditions.29). COMPLEX EIGENVALUES 145 The real and imaginary parts of this solution are ( ) ( ) −t/10 cos 2t −t/10 sin 2t xRe (t) = e and xIm (t) = e . Thus. − sin 2t The phase portrait of this solution indicates that we do indeed have a spiral sink (Fig- ure 3. 3 2 1 y(t) 0 -1 -2 -3 -3 -2 -1 0 1 2 3 x(t) Figure 3.30. The initial value problem dx = x/10 + y dt dy = −x + y/10 dt x(0) = 0 y(0) = 1/2. we can determine that c1 = 2 and c2 = 0. the solution to our initial value problem is ( ) −t/10 cos 2t x(t) = 2e .3. The matrix ( ) 1/10 1 −1 1/10 .4. we have the general solution ( ) ( ) −t/10 cos 2t −t/10 sin 2t x(t) = c1 e + c2 e . − sin 2t cos 2t respectively. hence.

If (a + d)2 − 4(ad − bc) < 0. Example 3. and we cannot apply the strategy that we used to determine the general solution in the case of distinct real roots.3 Solving Systems with Complex Eigenvalues Suppose that we have the linear system x′ = Ax. where ( ) −1 −2 A= .4. LINEAR SYSTEMS has an eigenvector (1. where ( ) a b A= . the solution to our initial value problem is ( ) 1 t/10 sin t x(t) = e . −i) with eigenvalue λ = 1/10 − i.146 CHAPTER 3. 2 cos t and he phase portrait is a spiral source (Figure 3. the complex solution is ( ) (1/10−i)t 1 x(t) = e . 3 2 1 y(t) 0 -1 -2 -3 -3 -2 -1 0 1 2 3 x(t) Figure 3. 4 3 . then the eigenvalues of A are complex. The system x′ = Ax.32. Thus. −i Following the procedure that we used in the previous example.31). c d The charactistic polynomial of A is p(λ) = λ2 − (a + d)λ + (ad − bc).31: Phase portrait for a spiral source 3.

the equilibrium point is a center. x(t) = e(1+2i)t v ( ) (1+2i)t 1 =e −1 − i ( ) t 2it 1 =ee −1 − i ( ) 1 = et (cos 2t + i sin 2t) −1 − i ( ) t 2 cos 2t + 2i sin 2t =e (−1 − i)(cos 2t + i sin 2t) ( ) t 2 cos 2t + 2i sin 2t =e (− cos 2t + sin 2t) + i(− cos 2t − sin 2t) ( ) ( ) t 2 cos 2t t 2 sin 2t =e + ie − cos 2t + sin 2t − cos 2t − sin 2t is a complex solution to our system. we have not shown that this method will work in all cases. Taking the real and imaginary parts of this solution. Although we have outlined a procedure to find the general solution of x′ = Ax if A has complex eigenvalues. −1 − i). Do this first without the use of technology. − cos 2t + sin 2t − cos 2t − sin 2t The nature of the equilibrium solution is determined by the factor eαt in the solution.3. λ = α ± iβ. We will do so in Section 3. If α > 0.4. Recalling that eiθ = cos θ + i sin θ. It is easy to show that an eigenvector for λ = 1 + 2i is v = (1.4 Important Lessons • If ( ) α β A= .4. Sketch the phase plane of each system and determine the nature of the equilibrium solution(s).5 Exercises 1. the equilibrium point is a spiral source.4. the equilibrium point is a spiral sink. the equilibrium point is a spiral sink. The general solution to the system x′ = Ax is ( ) ( ) cos βt sin βt x(t) = c1 eαt + c2 eαt . we obtain the general solution to our system ( ) ( ) t 2 cos 2t t 2 sin 2t x(t) = c1 e + c2 e . Solve each of the following linear systems for the given initial values. then check your answer with Sage. . If α > 0. If α < 0. −β α then A has two complex eigenvalues.6. 3. If α = 0. − sin βt cos βt If α < 0. 3. COMPLEX EIGENVALUES 147 The characteristic polynomial of A is λ2 − 2λ + 5 and so the eigenvalues are complex conjugates. the equilibrium point is a spiral source. λ = 1 + 2i and λ = 1 − 2i.

The strategy that we used in find the general solution to a system with distinct real eigenvalues will clearly have to be modified if we are to find a general solution to a system with a single eigenvalue. This there is a single straightline solution for this system (Figure 3. . LINEAR SYSTEMS (a) x′ = 2x + 2y y ′ = −4x + 6y x(0) = 2 y(0) = −3 (b) x′ = 2x − 6y y ′ = 2x + y x(0) = 2 y(0) = 1 (c) x′ = 2x + y y ′ = −x + 4y x(0) = 1 y(0) = −1 (d) x′ = 4x + 2y y ′ = −2x − y x(0) = 1 y(0) = −1 Hint.10) dy/dt −1 4 y The characteristic polynomial of the system (3. Hint.148 CHAPTER 3. 2. (3.33). if a 2 × 2 matrix A has complex eigenvalues.6 Project 3.10) is λ2 − 6λ + 9 and λ2 − 6λ + 9 = (λ − 3)2 .4.5 Repeated Eigenvalues Consider the following system ( ) ( )( ) dx/dt 2 1 x = . This polynomial has a single root λ = 3 with eigenvector v = (1. 3. then the solutions of x′ = Ax travel around the origin in a counterclockwise direction. Determine a computable condition that guarantees that. −1).

Thus. solutions to this system are of the form x(t) = αeλt v. dy/dt 0 λ y y The eigenvalues of A are both λ. . 0). any nonzero vector in R2 is an eigenvector for λ.5.5. 0 λ Again.33: A system with one straightline solution 3. Each solution to our system lies on a straight line through the origin and either tends to the origin if λ < 0 or away from zero if λ > 0. there is only one linearly independent eigenvector.3. we will rewrite the system as x′ = λx + y y ′ = λy. The simplest such case is ( ) ( )( ) ( ) dx/dt λ 0 x x = =A . 0 To find other solutions.1 Repeated Eigenvalues The remaining case the we must consider is when the characteristic equation of a matrix A has repeated roots. we have a single straight-line solution ( ) 1λt x1 (t) = αe . A more interesting case occurs if ( ) λ 1 A= . which we can take to be (1. Therefore. REPEATED EIGENVALUES 149 4 3 2 1 y(t) 0 -1 -2 -3 -4 -4 -3 -2 -1 0 1 2 3 4 x(t) Figure 3. Since Av = λv. however. both eigenvalues are λ.

If y ̸= 0. An eigenvector for λ is v = (1. . Notice that we have only one straightline solution (Figure 3.34. λ = −1. the solution of the second equation is y(t) = βeλt . Consequently.150 CHAPTER 3. 0 1 Example 3. a solution to our system is ( ) ( ) λt1 λt t αe + βe . the first equation becomes x′ = λx + βeλt . Applying the initial conditions x(0) = 1 and y(0) = 3. LINEAR SYSTEMS This is a partially coupled system. Thus. Therefore. the solution to our initial value problem is x(t) = e−t + 3te−t y(t) = 3e−t . the general solution to our system is x(t) = c1 e−t + c2 te−t y(t) = c2 e−t .35). which is a first-order linear differential equation with solution x(t) = αeλt + βteλt . 0). Consider the linear system x′ = −x + y y ′ = −y x(0) = 1 y(0) = 3. The matrix that corresponds to this system is ( ) −1 1 A= 0 −1 has a single eigenvalue.

Example 3. −4 1 Since the characteristic polynomial of A is λ2 − 6λ + 9 = (λ − 3)2 .5. we still need a second solution. If this is the situation. Example 3. then we actually have two separate cases to examine. 0 1 This is not two surprising since the system x′ = 2x y ′ = 2y is uncoupled and each equation can be solved separately.36. −2). but there are two linearly independent eigenvectors. depending one whether or not we can find two linearly independent eigenvectors. where ( ) 5 1 A= .5. 0) and v2 = (0. where ( ) 2 0 A= . Suppose we have the system x′ = Ax. we have only a single eigenvalue λ = 3 with eigenvector v1 = (1. 0 2 The single eigenvalue is λ = 2. Now let us consider the example x′ = Ax. In this case our solution is ( ) ( ) 2t 1 2t 0 x(t) = c1 e + c2 e .35: Phase portrait for repeated eigenvalues 3. v1 = (1. REPEATED EIGENVALUES 151 4 3 2 1 y(t) 0 -1 -2 -3 -4 -4 -3 -2 -1 0 1 2 3 4 x(t) Figure 3. This gives us one solution to our system. .37. x1 (t) = e3t v1 .2 Solving Systems with Repeated Eigenvalues If the characteristic equation has only a single repeated root. there is a single eigenvalue.3. 1). however.

152 CHAPTER 3. then the solutions tend towards the origin as t → ∞. 3. we will have a nodal sink. 3. we can take v2 = (1/2)w = (1/2. we cannot find a second linearly independent eigenvector and we need to obtain the second solution in a different manner.3 Important Lessons • If ( ) λ 1 A= . the solutions tend away from the origin. LINEAR SYSTEMS Since all other eigenvectors of A are a multiple of v.5. −4 −2 0 −4 Thus. 0). but we have only justified the case where A has distinct real eigenvalues. We must find a vector v2 such that (A − λI)v2 = v1 . 0 λ then A has one repeated real eigenvalue. where A has a repeated eigenvalue and any tow eigen vectors are linearly dependent. If it is negative.5. We then compute ( )( ) ( ) 2 1 1 2 (A − λI)w = (A − 3I)w = = = 2v1 . 0 µ −β α 0 λ 0 λ .5. 0 1 If λ < 0.6 Changing Coordinates In the beginning sections of this chapter. Notice that we have only given a recipe for finding a solution to x′ = Ax. our general solution is ( ) ( ) 3t 1 3t 1/2 + t x = c1 x1 + c2 x2 = c1 e + c2 e . 0). For λ > 0. . To do this we can start with any nonzero vector w that is not a multiple of v1 . The general solution to the system x′ = Ax is ( ) ( ) λt 1 λt t x(t) = αe + βe .6). we outlined procedures for solving systems of linear differential equations of the form ( ) ( )( ) ( ) dx/dt a b x x = =A dy/dt c d y y by determining the eigenvalues of A. 3. −2 −2t If the eigenvalue is positive. say w = (1. we will have a nodal source. we have considered the following special cases for A ( ) ( ) ( ) ( ) λ 0 α β λ 0 λ 1 . However. We will justify our procedure in the next section (Section 3. Hint.5 Project 3. and our second solution is ( ) 1/2 + t x2 = eλt (v2 + tv1 ) = e3t −2t Thus. .4 Exercises 1. .

1 Linear Maps A linear map or linear transformation on R2 is a function T : R2 → R2 that is defined by a matrix.6.6. Proof. A linear map T is invertible if and only if det T ̸= 0. where I is the identity transformation. we can certainly compute T −1 . In terms of matrices.3. If det T = 0. We will now show that we can transform any 2 × 2 system of first-order linear differential equations with constant coefficients into one of these special systems by using a change of coordinates.2 Changing Coordinates Suppose that we consider a linear system y′ = (T −1 AT )y (3. We will say that T : R2 → R2 is an invertible linear map if we can find a second linear map S such that T ◦ S = S ◦ T = I. y c d y When there is no confusion. then there are infinitely many nonzero vectors x such that T x = 0.6. On the other hand. where ( ) 1 0 I= . det T −c a Theorem 3. We write T −1 for the inverse matrix of T .11) . It is easy to check that the inverse of ( ) a b T = c d is ( ) −1 1 d −b T = . we are actually very close to providing a complete classification of all solutions. 3. this means that we can find a matrix S such that T S = ST = I. if the determinant is nonzero. That is. Then x = T −1 T x = T −1 0 = 0.38. at least in the 2 × 2 case. we will think of the linear map T : R2 → R2 and the matrix ( ) a b c d as interchangeable. CHANGING COORDINATES 153 Although it may seem that we have limited ourselves by attacking only a very small part of the problem of finding solutions for x′ = Ax. 0 1 is the 2 × 2 identity matrix. ( ) ( )( ) x a b x T = . which is a contradiction. 3. Suppose that T −1 exists and x ̸= 0 such that T x = 0.

Thus. Thus. If y(t) is a solution of (3. Indeed.154 CHAPTER 3. 1). we have (T −1 AT )ei = T −1 Avi = T −1 (λi vi ) = λi T −1 vi = λi ei for i = 1. respectively.6. A linear map T converts solutions of y′ = (T −1 AT )y to solutions of x′ = Ax. then T ei = vi for i = 1. y c d y 3. x′ (t) = (T y)′ (t) = T y′ (t) = T ((T −1 AT )y(t)) = A(T y(t)) = Ax(t).11). We can think of this in two ways. 2. We are now in a position to solve our problem of finding solutions of an arbitrary linear system ( ′) ( ) ( ) x a b x ′ = = . If e1 = (1. Therefore. the general solution of y′ = (T −1 AT )y is ( ) ( ) λ1 t 1 λ2 t 0 y(t) = αe + βe . the matrix T −1 AT is in canonical form. T −1 vi = ei for i = 1. 2. 2. 0 1 Hence. we claim that x(t) = T y(t) solves the equation x′ = Ax. ( ) −1 λ1 0 T AT = . Consequently. 0) and (0.3 Distinct Real Eigenvalues Consider the system x′ = Ax. 0 λ2 The eigenvalues of the matrix T −1 AT are λ1 and λ2 with eigenvectors (1. 1. The inverse of a linear map T takes solutions of x′ = Ax to solutions of y′ = (T −1 AT )y. respectively. 2. the general solution of x′ = Ax is ( ( ) ( )) 1 0 T y(t) = T αeλ1 t + βeλ2 t 0 1 . where A has two real. LINEAR SYSTEMS where T is an invertible matrix. 1). distinct eigenvalues λ1 and λ2 with eigenvectors v1 and v2 . 0) and e2 = (0. Let T be the matrix with columns v1 and v2 .

2. then T ei = vi for i = 1. CHANGING COORDINATES 155 ( ) ( ) λ1 t 1 λ2 t 0 = αe T + βe T 0 1 = αeλ2 t v1 + βeλ2 t v2 . 0) and (0. ( )( )( ) ( ) 1/3 1/3 1 2 1 1 5 0 T −1 AT = = . respectively. 1).41). T −1 vi = ei for i = 1. the general solution of y′ = (T −1 AT )y is ( ) ( ) 15t −t 0 y(t) = αe + βe .39. Consequently. 4 3 The eigenvalues of A are λ1 = 5 and λ2 = −1 and the associated eigenvectors are (1. 0) and e2 = (0. where ( ) 1/3 1/3 T −1 = .3. 1).6. where ( ) 1 2 A= . 2/3 −1/3 4 3 2 −1 0 −1 The eigenvalues of the matrix ( ) 5 0 0 −1 are λ1 = 5 and λ2 = −1 with eigenvectors (1. Thus. −1). 0 1 Hence. 2 −1 If e1 = (1. . 2) and (1. Example 3. the general solution of x′ = Ax is ( )( ( ) ( )) 1 1 5t 1 −t 0 T y(t) = αe + βe 2 −1 0 1 ( ) ( ) 1 1 = αe5t + βe−t 2 −1 The linear map T converts the phase portrait of the system y′ = (T −1 AT )y (Figure 3. 2. 2/3 −1/3 Thus. our matrix T is ( ) 1 1 . In this case. Suppose dx/dt = Ax.40) to the phase portrait of the system x′ = Ax (Figure 3. respectively.

LINEAR SYSTEMS 3 2 1 y(t) 0 -1 -2 -3 -3 -2 -1 0 1 2 3 x(t) Figure 3.4 Complex Eigenvalues Suppose the matrix ( ) a b A= c d .156 CHAPTER 3.40: Phase portrait for y′ = (T −1 AT )y 3 2 1 y(t) 0 -1 -2 -3 -3 -2 -1 0 1 2 3 x(t) Figure 3.6.41: Phase portrait for x′ = Ax 3.

Consequently. Av2 = (α + iβ)v2 .6. If v1 and v2 are not linearly independent. CHANGING COORDINATES 157 in system x′ = Ax has complex eigenvalues. Equating the real and imaginary parts. where a+d α= √2 4bc − (a − d)2 β= .43. However. we have A(v1 + iv2 ) = A(cv2 + iv2 ) = (c + i)Av2 . −β α Proof. v1 and v2 are linearly independent. one of the eigenvalues deter- mines the other. we find that Av1 = αv1 − βv2 . where v1 and v2 are real vectors. suppose that the eigenvalue λ = α + iβ has an eigenvector of the form v = v1 + iv2 . If λ = α + iβ is an eigenvalue of a real matrix A with β ̸= 0 and eigenvector the form v1 + iv2 . If v1 + iv2 . Let A be a real matrix with eigenvalue λ = α + iβ. Since v1 + iv2 is an eigenvector associated to the eigenvalue α + iβ. Thus. Now. if A is a real matrix with complex eigenvalues. this is a contradiction since the left-side of the equation says that we have real eigenvector while the right-side of the equation is complex. then there exists a matrix T such that ( ) −1 α β T AT = . then the vectors v1 and v2 are linearly independent. Proposition 3. A(v1 + iv2 ) = (α + iβ)(v1 + iv2 ) = (α + iβ)(c + i)v2 = (c + i)(α + iβ)v2 In other words. is an eigenvector for λ.42.3. Proof. Proposition 3. Then v = v1 − iv2 is an eigenvector for λ. where β ̸= 0. where v1 and v2 are real vectors. 2 The eigenvalues λ and λ are complex conjugates. the characteristic polynomial p(λ) = λ2 − (a + d)λ + (ad − bc) will have roots λ = α + iβ and λ = α − iβ. since Av = Av = λv = λv. On one hand. In this case. However. we have A(v1 + iv2 ) = (α + iβ)(v1 + iv2 ). then v1 = cv2 for some c ∈ R.

Similarly. Therefore. we can take T to be the matrix ( ) 2 0 T = . Thus. a center (α = 0). we have (T −1 AT )e1 = T −1 (αv1 − βv2 ) = αe1 − βe2 . center. 1/2 1 −1/2 −1 −1 1 −1/2 −1/2 . or source. −1 + i −1 1 respectively. 2 or equivalently x′ = Ax. where ( ) 0 1 A= . −1 1 Consequently. Example 3. ( )( )( ) ( ) 1/2 0 0 1 2 0 −1/2 1/2 T −1 AT = = . If T is the matrix with columns v1 and v2 . then T e1 = v1 T e2 = v2 . the phase portrait of x′ = Ax is equivalent to a sink. (T −1 AT )e2 = βe1 + αe2 . we have x′ = y 1 y ′ = − x − y. This particular equation might model a damped harmonic oscillator. or a spiral source (α > 0). Suppose that we wish to find the solutions of the second order equation 2x′′ + 2x′ + x = 0. If we rewrite this second-order equation as a first-order system. −β α The system y′ = (T −1 AT )y is in one of the canonical forms and has a phase portrait that is a spiral sink (α < 0).158 CHAPTER 3. LINEAR SYSTEMS Av2 = βv1 + αv2 . −1/2 −1 The eigenvalues of A are 1 1 − ±i . After a change of coordinates. Therefore.44. 2 2 The eigenvalue λ = (1 + i)/2 has an eigenvector ( ) ( ) ( ) 2 2 0 v= = +i . we can write the matrix T −1 AT as ( ) −1 α β T AT = .

3 2 1 y(t) 0 -1 -2 -3 -3 -2 -1 0 1 2 3 x(t) Figure 3. − cos(t/2) − sin(t/2) − sin(t/2) + cos(t/2) The phase portrait for this system is given in Figure 3. −β α The general solution to y′ = (T −1 AT )y is ( ) ( ) cos(t/2) sin(t/2) y(t) = c1 e−t/2 + c2 e−t/2 .46.45: Phase portrait for y′ = (T −1 AT )y The general solution of x′ = Ax is ( )[ ( ) ( )] 2 0 −t/2 cos(t/2) −t/2 sin(t/2) T y(t) = c1 e + c2 e −1 1 − sin(t/2) cos(t/2) ( )( ) ( )( ) 2 0 cos(t/2) 2 0 sin(t/2) = c1 e−t/2 + c2 e−t/2 −1 1 − sin(t/2) −1 1 cos(t/2) ( ) ( ) −t/2 2 cos(t/2) −t/2 2 sin(t/2) = c1 e + c2 e .3.6. . − sin(t/2) cos(t/2) The phase portrait of y′ = (T −1 AT )y is given in Figure 3. CHANGING COORDINATES 159 which is in the canonical form ( ) α β .45.

we have a much more efficient way of solving the system x′ = Ax. − cos(t/2) − sin(t/2) − sin(t/2) + cos(t/2) Taking the real and the imaginary parts of the last expression. the general solution of x′ = Ax is ( ) ( ) −t/2 2 cos(t/2) −t/2 2 sin(t/2) x(t) = c1 e + c2 e .6. . where ( ) 0 1 A= . Of course. −1/2 −1 Since A has eigenvalue λ = (−1 + i)/2 with an eigenvector v = (2. − cos(t/2) − sin(t/2) − sin(t/2) + cos(t/2) which agrees with the solution that we found by transforming coordinates. Then the characteristic polynomial of A is p(λ) = λ2 − (a + d)λ + (ad − bc). 3.5 Repeated eigenvalues Now suppose that A has a single real eigenvalue λ.47.160 CHAPTER 3. we can apply Euler’s formula and write the solution as x(t) = e(−1+i)t/2 v ( ) −t/2 it/2 2 =e e −1 + i ( ) −t/2 2 =e (cos(t/2) + i sin(t/2)) −1 + i ( ) ( ) −t/2 2 cos(t/2) −t/2 2 sin(t/2) =e + ie . −1 + i).46: Phase portrait of x′ = Ax Remark 3. then A has an eigenvalue λ = (a + d)/2. LINEAR SYSTEMS 3 2 1 y(t) 0 -1 -2 -3 -3 -2 -1 0 1 2 3 x(t) Figure 3.

49. If A has a single eigenvalue and a pair of linearly independent eigen- vectors. it must be the case that AT = (Au. So. then there exists a matrix T such that ( ) −1 λ 1 T AT = . Suppose that u and v are linearly indeendent eigenvectors for A. the system is uncoupled and is easily solved. CHANGING COORDINATES 161 Proposition 3. If v is an eigenvector for λ and any other eigenvector for λ is a multiple of v. det(T ) ̸= 0 and T is invertible. then ( ( ) ) ( ) α α A w+ v = Aw + Av β−λ β−λ ( ) α = αv + βw + λ v β−λ ( ) λ = βw + α 1 + v β−λ ( ) β−λ+λ = βw + α v β−λ ( ( ) ) α =β w+ v β−λ . otherwise. T e1 = u and T e2 = v. That is. Since u and v are linearly independent. Proposition 3. 0 λ Proof. We claim that β = λ. and let T be the matrix whose first column is u and second column is v. then A must be of the form ( ) λ 0 A= . λv) = λ(u. we would have a second linearly independent eigen- vector. we can solve each equation in the system x′ = λx y ′ = λy separately to obtain the general solution x = c1 eλt y ′ = c2 eλt . v) = λIT. or ( ) λ 0 A= . If w is another vector in R2 such that v and w are linearly independent. Suppose that A has a single eigenvalue λ.48. We can assume that α ̸= 0. If this were not the case. 0 λ In this case.3.6. then Aw can be written as a linear combination of v and w. 0 λ Proof. That is. Av) = (λu. Aw = αv + βw.

x′ = Ax is in canonical form after a change of coordinates. −2 −2t This solution agrees with the solution that we found in Example 3. Furthermore. we know that the general solution to the system ( ) ( )( ) dx/dt 3 1 x = dy/dt 0 3 y is ( ) ( ) 1 3t 3t t y(t) = c1 e + c2 e . −2 0 and ( )( )( ) ( ) −1 −1/2 2 5 1 1 1/2 3 1 T AT = = . α We now define T e1 = v and T e2 = u. where ( ) 5 1 A= . . Any other eigenvector for λ is a multiple of v. 0). the general solution to ( ) ( )( ) dx/dt 5 1 x = dy/dt −4 1 y is x(t) = T y(t) ( ) ( ) 3t 1 3t t = c1 e T + c2 e T 0 1 ( ) ( ) 1 1/2 + t = c1 e3t + c2 e3t . the matrix that we seek is ( ) 1 1/2 T = .162 CHAPTER 3. −2). 0 λ Therefore.37. Aw = αv + λw. 0 1 Therefore. Thus. Since AT = Au + Av = v + λu + λv ( ) λ 1 T = T (λe1 ) + T e1 + T (λe2 ) = v + λu + λv. If we choose w = (1. 0 λ we have ( ) −1 λ 1 T AT = . −4 −2 0 −2 0 So we can let u = (1/2)w = (1/2. Example 3. Consider the system x′ = Ax.3. then v and w are linearly independent. If we will let u = (1/α)w. 0). 1 1 −4 1 −2 0 0 3 From Section 3. LINEAR SYSTEMS and β would be an eigenvalue distinct from λ. ( ) ( ) ( ) ( ) ( ) 5 1 1 1 1 Aw = =2 +λ =2 +3 . Therefore. then λ Au = v + w = v + λu. −4 1 The characteristic polynomial of A is λ2 −6λ+9 = (λ−3)2 . we have only a single eigenvalue λ = 3 with eigenvector v = (1.50.

51: Mixing example with two tanks . • A linear map T converts solutions of y′ = (T −1 AT )y to solutions of x′ = Ax.7 The Trace-Determinant Plane Suppose that we have two tanks.3.6. that both have a volume of V liters and are both filled with a brine solution.51). This is an exercise.4. . .7. Suppose that pure water enters Tank A at a rate of rin liters per minute. 3.6. Tank A and Tank B.2–3.6. THE TRACE-DETERMINANT PLANE 163 In practice. . we find solutions to linear systems using the methods that we outlined in Sections 3. rin rA rB Tank A Tank B rout Figure 3. and a salt mixture enters Tank A from Tank B at a rate of rB liters per minute. 0 µ −β α 0 λ 0 λ 3. 3.7 Exercises 1. What we have demonstrated in this section is that those solutions are exactly the ones that we want. • A change of coordinates converts the system x′ = Ax to one of the following special cases. • The inverse of a linear map T takes solutions of x′ = Ax to solutions of y′ = (T −1 AT )y. ( ) ( ) ( ) ( ) λ 0 α β λ 0 λ 1 .8 Project 3. Brine also enters Tank B from Tank A at a rate of rA liters per minute.6 Important Lessons • A linear map T is invertible if and only if det T ̸= 0. Hint. brine is drained from Tank B at a rate of rout so that the volume in each tank is constant (Figure 3. Finally.

then our problem can be modeled with a linear system of two equations.52.7. dt V V V Furthermore. then the trace of A is λ1 + λ2 and det(A) = λ1 λ2 . Thus. LINEAR SYSTEMS If x(t) and y(t) are the amounts of salt in Tank A and Tank B.164 CHAPTER 3. we can rewrite the charac- teristic polynomial as det(A − λI) = λ2 − T λ + D. dt V V If we have initial conditions x(0) = x0 and y(0) = y0 . since the volume in Tank B is constant. ( ) a−λ b det(A − λI) = det = λ2 − (a + d)λ + (ad − bc). The quantity T = a + d is the sum of the diagonal elements of the matrix A. We call this quantity the trace of A and write tr(A). and we will have a stable equilibrium solution at (0. we need to derive the charac- teristic polynomial of A. Consequently.1 The Trace-Determinant Plane The key to solving the system ( ′) ( )( ) ( ) x a b x x = =A y′ c d y y is determining the eigenvalues of A. rA = rB + rout . c d−λ Of course. our system now becomes dx x y = −rA + rB dt V V dy x y = rA − rA . Indeed. 3. dx x y = rate in − rate out = −rA + rB dt V V dy x y y = rate in − rate out = rA − rB − rout . We can use the trace and determinant to establish the nature of a solution to a linear system. 2 2 Hence. 0). D = det(A) = ad − bc is the determinant of A. is it ever possible that the equilibrium solution is a spiral sink? One solution is provided by studying the trace-determinant plane. we can rewrite the characteristic polynomial as det(A − λI) = λ2 − T λ + D. Theorem 3. For example. The eigenvalues of A are now given by √ √ T + T 2 − 4D T− T 2 − 4D λ1 = and λ2 = . The proof follows from a direct computation. respectively. Determining the nature of the equilibrium solution is a more difficult question. Proof. . it is not too difficult to deduce that the amount of salt in each tank will approach zero as t → ∞. If a 2 × 2 matrix A has eigenvalues λ1 and λ2 . To find this eigenvalues. T = λ1 + λ2 and D = λ1 λ2 .

Thus. points on the parabola correspond to systems with repeated eigenvalues. we should be able to determine the phase portrait of a system x′ = Ax by simply examining the trace and determinant of A. points above the parabola (D > T 2 /4 or equivalently T 2 − 4D < 0) correspond to systems with complex eigenvalues. we have two distinct real eigenvalues. we have two complex eigenvalues.53). In fact. 2 we can immediately see that the expression T 2 −4D determines the nature of the eigenvalues of A. Therefore. • If T 2 − 4D > 0. Proof. The trace and determinant of a 2 × 2 matrix are invariant under a change of coordinates.52 tells us that we can determine the determinant and trace of a 2 × 2 matrix from its eigenvalues.53: The trace-determinant plane Theorem 3. Thus. It is straightforward to verify that det(AB) = det(A) det(B) and det(T −1 ) = 1/ det(T ) for 2 × 2 matrices A and B. we have repeated eigenvalues. Therefore. det(T ) A direct computation shows that tr(AB) = tr(BA). we can represent those systems with repeated eigenvalues by graphing the parabola D = T 2 /4 on the T D-plane or trace-determinant plane (Figure 3. det(T −1 AT ) = det(A) and tr(T −1 AT ) = tr(A) for any 2 × 2 matrix A and any invertible 2 × 2 matrix T . THE TRACE-DETERMINANT PLANE 165 Theorem 3. . If T 2 − 4D = 0 or equivalently if D = T 2 /4. • If T 2 − 4D < 0. and these eigenvalues are complex conjugates. we have repeated eigenvalues. That is. Since the eigenvalues of A are given by √ T ± T 2 − 4D λ= . • If T 2 − 4D = 0. tr(T −1 AT ) = tr(T −1 T A) = tr(A).3.7. and points below the parabola (D < T 2 /4 or equivalently T 2 − 4D > 0) correspond to systems with real eigenvalues.54. 1 det(T −1 AT ) = det(T −1 ) det(A) det(T ) = det(A) det(T ) = det(A). D D = T 2 /4 Complex Eigenvalues Repeated Eigenvalues T Distinct Real Eigenvalues Figure 3.

then we have a complex eigenvalues. The equilibrium points are spiral sinks and spiral sources. . If T > 0 we have a source. . or stay in a closed orbit if α = 0. . See Figure 3. 2 If T 2 − 4D < 0. D D = T 2 /4 Spiral Sink Center Spiral Source T Figure 3.55.166 CHAPTER 3. respectively. − sin βt cos βt The eαt factor tells us that the solutions either spiral into the origin if α < 0. or centers. The general solution to this system is ( ) ( ) αt cos βt αt sin βt x(t) = c1 e + c2 e . The eigenvalues of A are given by √ T ± T 2 − 4D λ= . LINEAR SYSTEMS Furthermore. we have a center.54. and the type of equilibrium point depends on the real part of the eigenvalue. spiral out to infinity if α > 0. If T = 0. . The sign of the real part is determined solely by T . we have a sink. −β α 0 µ 0 λ 0 λ The system ( ) ′ α β x = x −β α has eigenvalues λ = α ± iβ. where A is one of the following matrices: ( ) ( ) ( ) ( ) α β λ 0 λ 0 λ 1 . That is. each of the expression T 2 − 4D is not affected by a change of coordinates by Theorem 3. we only need to consider systems x′ = Ax. If T < 0. We will have three cases to consider if none of our eigenvalues are zero: • Both eigenvalues are positive (source). Suppose that we have a system ( ) ′ λ 0 x = x 0 µ with distinct eigenvalues λ and µ.55: D > T 2 /4 The situation for distinct real eigenvalues is a bit more complicated.

we have a saddle if T > 0 and D < 0. then T 2 − 4D > T 2 and √ T 2 − 4D > T √ Therefore. THE TRACE-DETERMINANT PLANE 167 • Both eigenvalues are negative (sink). If D < 0. we will have a nodal sink. • One eigenvalue is negative and the other is positive (saddle).56. That is. we have √ T 2 − 4D < T √ and the value of the second eigenvalue (T − T 2 − 4D )/2 is postive. Since we are considering the case T > 0. Then the eigenvalue (T − T 2 − 4D )/2 is always negative. In this case. then the eigenvalue √ T+ T 2 − 4D 2 is positive and we need only determine the sign of the second eigenvalue √ T− T 2 − 4D 2 If D < 0.3.7. telling √ us that any point in the fourth quadrant must correspond to a saddle. If D > 0. Our two eigenvalues are given by √ T± T 2 − 4D λ= . any point in the first quadrant below the parabola corresponds to a system with two positive eigenvalues and must correspond to a nodal source. we have one positive and one zero eigenvalue. suppose that T < 0. then T 2 − 4D < T 2 . 2 If T > 0. and we need to determine if other eigenvalue is positive or negative. . Therefore. the other eigenvalue (T − T 2 − 4D )/2 is positive. √ One the other hand. then T 2 − 4D < T and the second eigenvalue is negative. If D > 0. We summarize our findings in Figure 3.

we have sources if T > 0 and sinks if T < 0 (Figure 3. Since T 2 − 4D = 0.58.56: The trace-determinant plane for real and complex eigenvalues For repeated eigenvalues.57: D = T 2 /4 Example 3. LINEAR SYSTEMS D D = T 2 /4 Spiral Sink Center Spiral Source Nodal Sink Nodal Source T Saddle Figure 3. . the only eigenvalue is T /2. the analysis depends only on T .168 CHAPTER 3. The problem could be modeled by the system of equations dx x y = −rA + rB dt V V dy x y = rA − rA dt V V x(0) = x0 y(0) = y0 . Thus. Let us return to the mixing problem that we proposed at the beginning of this section. D D = T 2 /4 Spiral Sink Center Spiral Source Sink (stable line) Source (unstable line) Nodal Sink Nodal Source T Saddle Figure 3.57).

If a < 1/2. rA /V −rA /V Computing the trace and determinant of the matrix yields T = −2rA /V and D = (rA 2 − 2 rA rB )/V . thus. a bifurcation occurs at a = 1/2. 2 Thus. We are on the parabola if 1 T 2 − 4D = 4 − 8a = 0 or a= . where rA and rB are both positive.2 Parameterized Families of Linear Systems The trace-determinant plane is an example of a parameter plane. D D = T 2 /4 Spiral Sink Sink (stable line) Nodal Sink a T Saddle Figure 3. our sink becomes a saddle (Figure 3. Certainly. Finally.60).59.60: A one-parameter family of linear systems . we have a sink with real eigenvalues.7. If a > 1/2. Consider the system ( ′) ( ) x −2 a = Ax x. V V V2 Therefore. 3. if a < 0. since ( ) r2 − rA rB −2rA 2 4rA rB 4D − T 2 = 4 A 2 − =− < 0. We can adjust the entries of a matrix A and.7. THE TRACE-DETERMINANT PLANE 169 The matrix corresponding to this system is ( ) −rA /V +rB /V A= . T < 0 and 2 −r r rA A B rA (rA − rB ) rA rout D= 2 = 2 = > 0.3. any solution must be stable. change the value of the trace and the determinant. Example 3. we have a spiral sink. but D = det(A) = 2a. V V V2 we are below the parabola in the trace-determinant plane and know that our solution must be a nodal sink. Further more. y′ −2 0 The trace of A is always T = −2.

k/m) lies on the parabola. −k/m −b/m Thus. If (T. we need only concern ourselves with the second quadrant (Figure 3. If b = 0. we have an underdamped oscillator. D) = (−b/m. we have an undamped oscillator. k/m) lies below the parabola. k/m) lies above the parabola. T = −b and D = 3. If we rewrite this equation as a first-order system. for the harmonic oscillator T = −b/m and D = k/m. D) = (−b/m. b ≥ 0 is the damping coefficient.63. dt Thus. D D = T 2 /4 Under Damped Critically Damped Undamped Over Damped T Figure 3.170 CHAPTER 3.61: A one-parameter family for a harmonic oscillator If (T.61). D) = (−b/m. Now let us see what happens to our harmonic oscillator when we fix m = 1 and k = 3 and let the damping b vary between zero and infinity. we have ( ) ′ 0 1 x = x. We can rewrite our system as dx =y dt dy = −3x − by. Example 3. If (T. and k > 0 is the spring constant. we have a critically damped oscillator. If we use the trace-determinant plane to analyze the harmonic oscillator. dt dt where m > 0 is the mass. LINEAR SYSTEMS Recall that a harmonic oscillator can be modeled by the second-order equation d2 x dx m 2 +b + kx = 0.62. . we have an overdamped oscillator. We can see how the phase portrait varies with the parameter b in Figure 3.

Since the trace and the determinant of a matrix determine the eigenvalues of A. • If a 2×2 matrix A has eigenvalues λ1 and λ2 . Finally. If 0 < b < 2 √3. THE TRACE-DETERMINANT PLANE 171 D D = T 2 /4 Under Damped 3 Critically Damped Undamped Over Damped T Figure 3.63: The trace-determinant plane for varying damping The line D = 3 in the trace-determinant √ plane crosses the repeated eigenvalue parabola.64. the eigenvalues are real. the eigenvalues are complex with a nonzero real part—the underdamped case. Although the trace-determinant plane gives us a great deal of information about our system. we have √ purely imaginary eigenvalues. we can not determine everything from this parameter plane. and negative. This is the undamped harmonic oscillator. 3. If b = 0. then tr(A) is λ1 +λ2 and det(A) = λ1 λ2 . If b = 2 3. • The trace-determinant plane is determined by the graph of the parabola D = T 2 /4 on the T D-plane.7. For example. we can use the trace-determinant plane to parameterize the phase portraits of linear systems. Example 3. • The trace and determinant of a 2 × 2 matrix are invariant under a change of coordi- nates. In this case. D = T 2 /4 if b2 = 12 or when b = 2 3. . the matrices ( ) ( ) 0 1 0 −1 A= and B= −1 0 1 0 both have the same trace and determinant. √the eigenvalues are negative and repeated—the critically damped case. distinct.3.3 Important Lessons • The characteristic polynomial of a 2 × 2 matrix can be written as λ2 − T λ + D. but the solutions to x′ = Ax wind around the origin in a clockwise direction while those of x′ = Bx wind around in a counterclockwise direction. where T = tr(A) and D = det(A). we have the overdamped case. • The trace-determinant plane is useful for studying bifurcations. Points on the trace-determinant plane correspond to the trace and determinant of a linear system x′ = Ax.7. if b > 2 3.

172 CHAPTER 3. LINEAR SYSTEMS

3.7.4 Exercises
1. Consider the one-parameter family of linear systems given by
( ′) ( √ )( )
x √ a 2 + a/2 x
= .
y′ 2 − a/2 0 y

(a) Sketch the path traced out by this family of linear systems in the trace-determinant
plane as a varies.
(b) Discuss any bifurcations that occur along this path and compute the corresponding
values of a.

Hint.
2. Consider the two-parameter family of linear systems
( ′) ( )( )
x a b x
′ = .
y b a y

Identify all of the regions in the ab-plane where this system possesses a saddle, a sink, a
spiral sink, and so on.
Hint.

3.7.5 Project—

3.8 Linear Systems in Higher Dimensions
Suppose that we have two masses on a table, m1 and m2 , connected by three springs with
the outside springs connected to two walls (Figure 3.65), and the masses are free to move
horizontally. We will assume that the springs are uniform and all have the same spring
constant k. The horizontal displacements of the springs are denoted by x1 (t) and x2 (t) for
the masses m1 and m2 , respectively. Assuming that there is no damping, the only forces
acting on mass m1 at time t are those of left and middle springs. The force from the left
spring will be −kx1 while the force from middle spring will be k(x2 − x1 ). By Newton’s
Second Law of motion, we have

m1 x′′1 = −kx1 + k(x2 − x1 ).

Similarly, the only forces acting on the second mass, m2 , will come from middle and right
springs. Again using Newton’s Second Law,

m2 x′′2 = −k(x2 − x1 ) − kx2 .

x1 x2
wall wall
m1 m1

x1 = 0 x2 = 0

Figure 3.65: A double spring-mass system

3.8. LINEAR SYSTEMS IN HIGHER DIMENSIONS 173

We now have a system of two second-order linear equations

m1 x′′1 = −2kx1 + kx2
m2 x′′2 = kx1 − 2kx2 .

If we define x3 and x4 by x3 = x′1 and x4 = x′2 , we now have first-order linear system of
four equations,

x′1 = x3
x′2 = x4
k k
x′3 = −2 x1 + x2
m1 m1
k k
x′4 = x1 − 2 x2
m2 m2
We can represent this system in the matrix form x′ = Ax, where
 
0 0 1 0
 
 0 0 0 1
A= .
−2k/m1 k/m1 0 0
k/m2 −2k/m2 0 0

We will learn how to analyze and solve such systems in the next two sections.

3.8.1 Higher-Order Linear Systems
We can write the system

x′1 = a11 x1 + · · · a1n xn
x′2 = a21 x1 + · · · a2n xn
..
.
x′n = an1 x1 + · · · ann xn

in matrix form x′ = Ax, where
   
a11 a12 · · · a1n x1
 a21 a22 · · · a2n   x2 
   
A= . .. .. ..  and x =  . .
 .. . . .   .. 
an1 an2 · · · ann xn

The strategy for finding solutions to the system x′ = Ax is the same as for systems of
two equations. If λ is an eigenvalue of A with eigenvector v,

x(t) = eλt v

is a solution for our system. Indeed,

x′ (t) = λeλt v = eλt (λv) = eλt Av = A(eλt v) = Ax(t).

Example 3.66. The system

x′ = −5x − 8y − 2z

174 CHAPTER 3. LINEAR SYSTEMS

y ′ = 5x + 12y + 4z
z ′ = −11x − 19y − 5z

can be rewritten as
 ′   
x −5 −8 −2 x
x′ = y ′  =  5 12 4  y  = Ax.
z′ −11 −19 −5 z
We can compute the eigenvalues of A by finding the roots of its characteristic polynomial

det(A − λI) = −λ3 − 2λ2 − λ + 2 = (λ − 1)(λ + 1)(λ − 2).

Thus, the eigenvalues of A are λ = 2, λ = 1, and λ = −1. To find an eigenvector for λ = 2,
we must find a nontrivial solution for system of equations (A − 2I)x = 0,

−7x − 8y − 2z = 0
−5x + 10y + 4z = 0
−11x − 19y − 7z = 0.

It is easy to check that (2, −3, 5) is a solution. Similarly, we can determine that (5, −7, 13)
is an eigenvector for λ = 1 and (1, −1, 2) is an eigenvector for λ = −1. Thus, we have found
three solutions for the system x′ = Ax,
     
2 5 1
2t   t  −t 
x1 (t) = e −3 , x2 (t) = e −7 , and x3 (t) = e −1 .
5 13 2

The Principle of Superposition also holds for higher-order systems. If x1 (t) and x2 (t)
are solutions for x′ = Ax, then
c1 x1 (t) + c2 x2 (t)
is a solution for the system, since
d d d
[c1 x1 (t) + c2 x2 (t)] = c1 x1 (t) + c2 x2 (t)
dt dt dt
= c1 Ax1 (t) + c2 Ax2 (t)
= A[c1 x1 (t) + c2 x2 (t)].

Consequently,
     
2 5 1
x(t) = c1 e2t −3 + et −7 + c3 e−t −1
5 13 2
is a solution for our system. This is, in fact, the general solution for the system.
Although we shall not cover the notions of linear independence, canonical matrices, and
change of coordinates for Rn , the same ideas that we used for systems first-order linear
differential equations in R2 carry over to Rn .The necessary linear algebra is covered in any
good linear algebra course. Also, see Chapters 5 and 6 in [12].
Example 3.67. Let us see how the linear algebra works in the previous example. If we
form the matrix  
2 5 1
T = −3 −7 −1
5 13 2

3.8. LINEAR SYSTEMS IN HIGHER DIMENSIONS 175

from the eigenvectors of A, we can convert the system x′ = Ax to the system

y′ = (T −1 AT )y
   
1 −3 −2 −5 −8 −2 2 5 1
−1 1 1  5 12 4  −3 −7 −1 y,
4 1 −1 −11 −19 −5 5 13 2
 
2 0 0
= 0 1 0  y,
0 0 −1

which we can immediately solve:
     
1 0 0
2t   t  −t  
y(t) = c1 e 0 + c2 e 1 + c3 e 0 .
0 0 1

Multiplying our solution by T yields the general solution
     
2 5 1
x(t) = T y = c1 e2t −3 + et −7 + c3 e−t −1 ,
5 13 2

the solution to the system x′ Ax.

3.8.2 The Geometry of Solutions
In Section 3.6, we classified all of the geometry of the solutions for planar systems using the
trace-determinant plane. The geometry for linear systems in three variables is a bit more
complicated. For a system

x′ = a1 x + a1 y + a3 z
y ′ = b1 x + b2 y + b3 z
z ′ = c1 x + c2 y + c3 z,

our solutions curves live in R3 , and there is simply a lot more room to move around in
three dimensions than in two dimensions. The origin is still an equilibrium solution for a
system of linear differential equations in three variables. The origin is a stable equilibrium
solution if any solution x(t) approaches 0 = (0, 0, 0) as t → ∞; otherwise, 0 is an unstable
equilibrium solution. In the case of planar systems, an unstable solution is a nodal saddle,
a nodal source, a spiral source, or a source with a single unstable line. In the case of R3 , we
could have a stable line of solutions and an unstable plane of solutions. In this case, all
solutions of the system with initial condition lying on the stable line would approach the
origin as t → ∞, but all solutions with initial conditions that are a nonzero point on the
unstable plane would move away from the origin.
Example 3.68. In Example 3.66, we had solutions
     
2 5 1
x1 (t) = e2t −3 , x2 (t) = et −7 , and x3 (t) = e−t −1 .
5 13 2

The straight line through the origin and the point (1, −1, 2) is a stable line. That is, for
any initial condition x(0) = (x0 , y0 , z0 ) lying on this line, our solution will tend toward the

176 CHAPTER 3. LINEAR SYSTEMS

origin as t → ∞. On the other hand, the plane spanned by (2, −3, 5) and (5, −7, 13) is
unstable plane. Solutions on this plane move away from the origin as t → ∞. Of course,
(0, 0, 0) is an equlibrium solution for our system. We say that the origin is a saddle in this
example (Figure 3.69).

z-axis
(1, −1, 2)

T

Figure 3.69: A saddle in R3

Example 3.70. For the system

 
0 1 0
x′ = −1 0 0  x
0 0 −1

we have a very different type of unstable equilibrium solution. The eigenvalues of this matrix
are λ = ±i and λ = −1. Thus, a solution satisfying the initial condition x(0) = (x0 , y0 , z0 )
is given by

     
cos t sin t 0
   
x(t) = x0 − sin t + y0 cos t + z0 e −t  
0 .
0 0 1

If z0 = 0, then our initial condition is in the xy-plane and all of the solutions lie on circles
centered at the origin. On the other hand, if x0 = 0 and y0 = 0, we have a stable line of
solutions lying along the z-axis. In fact, each solution that does not lie on the stable line
lies on a cylinder in R3 given by x2 + y 2 = r2 for some constant r > 0. These solutions
spiral towards the circular solution of radius r in the xy-plane if z0 ̸= 0 (Figure 3.71).

3.8. LINEAR SYSTEMS IN HIGHER DIMENSIONS 177

Figure 3.71: A spiral center in R3

Example 3.72. For an example of a stable plane and an unstable line, let us consider the
system
 
−1 1 0
x′ = −1 −1 0 x = Ax.
0 0 1
The characteristic equation of the matrix A is

λ3 + λ2 − 2 = (λ − 1)(λ2 + 2λ + 2) = 0.

Thus, the eigenvalues of A are λ = 1 and λ = −1 ± i. Solving

[A − (−1 + i)]x = 0

gives us an eigenvector (1, i, 0) for λ = −1 + i. Since
     
1 1 0
 i  = 0 + i 1 ,
0 0 0

we will let v1 = (1, 0, 0) and v2 = (0, 1, 0). Since v3 = (0, 0, 1) is an eigenvector for λ = 1,
our system has solution
     
cos t sin t 0
x(t) = x0 e−t − sin t + y0 e−t cos t + z0 et 0 ,
0 0 1

where x(0) = (x0 , y0 , z0 ). If z0 = 0, our initial condition lies in the xy-plane and solution
curves sprial in towards the origin. Thus, we have a stable plane. On the other hand, if
x0 = 0 and y0 = 0 but z0 ̸= 0, then our solution approaches ±∞ as t → ∞. In this case,
the z-axis is an unstable line (Figure 3.73).

178 CHAPTER 3. LINEAR SYSTEMS

Figure 3.73: A spiral saddle in R3

For an example of a stable equilibrium solution at the origin, consider the system

 
λ1 0 0
x′ =  0 λ2 0  x,
0 0 λ3

where λ3 < λ2 < λ1 < 0. For an initial condition (x0 , y0 , z0 ) with at least one coordinate
nonzero, the corresponding solution tends towards the origin tangentially to the x-axis as
t → ∞ (Figure 3.74).

all solutions will approach the origin as t → ∞. Thus. z0 ) to be       cos t sin t 0 x(t) = x0 e−t − sin t + y0 e−t cos t + z0 e−t 0 . we will assume that m1 = m2 = k = 1.3. our matrix now becomes   0 0 1 0   0 0 0 1 A= . 0 0 −1 we obtain the solution satisfying the initial condition x(0) = (x0 . 3. −2k/m1 k/m1 0 0 k/m2 −2k/m2 0 0 To keep matter simple. y0 .74: A sink in R3 Changing the system that in Example3. 0 0 1 where x(0) = (x0 . In this case.8. −2 1 0 0 1 −2 0 0 . z0 ).8. y0 .3 The Double Spring-Mass Systems Revisited Let us return to our spring-mass system x′ = Ax. LINEAR SYSTEMS IN HIGHER DIMENSIONS 179 Figure 3. where   0 0 1 0    0 0 0 1 A= .72 to be   −1 1 0 x′ = −1 −1 0  x.

we can determine c1 = c2 = c3 = 0 and c4 = 2. respectively. We can find eigenvectors  √   √      i/ √3 −i/√ 3 −i i          −i/ 3  i/ 3  −i   i v1 =   . the general solution to our system is √ √ x(t) = c1 ei 3t v1 + c1 e−i 3t v2 + c1 eit v3 + c1 e−it v4 . v3 =   . we can rewrite the solution as  √   √      cos √3t − sin√ 3t cos t sin t          − cos √ 3t   sin 3t √   cos t   sin t  x(t) = c1  √  + c2  √  + c3   + c4  . √ this form of the solution is not very useful. λ4 = −i. Thus. 2 cos t 2 cos t and the two masses will oscillate with a frequency of one and an amplitude of two. LINEAR SYSTEMS The characteristic polynomial of A is det(A − λI) = λ4 + 4λ2 + 3 = (λ2 + 1)(λ2 + 3). By examining real and imaginary parts of ei 3t v1 and c1 eit v3 . . the solution to our initial value problem is   2 sin t    2 sin t  x(t) =  . We leave the details as an exercise.. however. x′n = an1 x1 + · · · ann xn . − 3 sin 3t − 3 cos 3t − sin t cos t √ √ √ √ 3 sin 3t 3 cos 3t − sin t cos t If we have the following initial conditions. √ Thus the eigenvalues of A are λ = ±i and λ = ±i 3. λ3 = i. v4 =   . Consequently. x1 (0) = 0 x2 (0) = 0 x′1 (0) = x3 (0) = 2 x′2 (0) = x4 (0) = 2.4 Important Lessons • x′1 = a11 x1 + · · · a1n xn x′2 = a21 x1 + · · · a2n xn . v2 =   .8.  −1   −1  1 1 1 1 1 1 √ √ corresponding to the eigenvalues λ1 = i 3.180 CHAPTER 3. 3. λ2 = −i 3.

. (a) Show that x1 (t) = e−bt/2a is a solution to ax′′ + bx′ + cx = 0.8.3. then c1 x1 (t) + c2 x2 (t) is a solution for the system.5 Exercises 1.  .   . If x1 (t) and x2 (t) are solutions for x′ = Ax. Find the general solution of the system x′1 = 6x1 + 9x2 + x3 + 4x4 x′2 = −x1 + x2 + x3 x′3 = 3x1 + 8x2 + 4x3 + 4x4 x′4 = −3x1 − 8x2 − 7x3 − 5x4 Hint. .8. where a ̸= 0 and b2 − 4ac = 0.. 2. . • The geometry for system in R3 is more complicated than the planar case.. x(t) = c1 e−bt/2a + c2 te−bt/2a is a general solution for ax′′ + bx′ + cx = 0. 3. we can solve the system x′ = Ax by finding eigenvalues and eigenvectors for A. . LINEAR SYSTEMS IN HIGHER DIMENSIONS 181 can be written in matrix form x′ = Ax. • The Principle of Superposition holds for higher-order systems. 3.. the solutions are usually characterized by stable lines or stable planes. Let ax′′ + bx′ + cx = 0. .  an1 an2 · · · ann xn • As in the case of R2 . where     a11 a12 · · · a1n x1  a21 a22 · · · a2n   x2      A= . . . Solve the system x′ = −5x + 15y − 4z y ′ = −x + 5y − z z ′ = 4x − 6y + 3z x(0) = −1 y(0) = 2 z(0) = 0 Hint. However. . Thus.. (b) Assume that y = v(t)x1 (t) = v(t)e−bt/2a is a solution to ax′′ + bx′ + cx = 0 and show that v(t) = c1 + c2 t.  and x =  ..

Determine all values of α.13) for t > 0 and find a second linearly independent solution to ((A. (a) Assume that x(t) = v(t)x1 (t) is a solution to x′′ + p(t)x′ + q(t)x = 0 and derive the equation x1 v ′′ + (2x′1 + px1 )v ′ = 0. Hint. Solve each of the following initial value problems. Suppose that x1 (t) is a solution (not identically zero) to the equation x′′ + p(t)x′ + q(t)x = 0. Reduction of Order. . if any. 5. 4. 6. (3. if any. LINEAR SYSTEMS Hint.11)). for which all nonzero solutions become unbounded as t → ∞. (c) Show that x1 (t) = 1/t is a solution to 2t2 x′′ + 3tx′ − x = 0 (3. (a) y ′′ − 2y ′ + 5y = 0 y(π/2) = 0 y ′ (π/2) = 2 (b) 9y ′′ − 12y ′ + 4y = 0 y(0) = 2 y ′ (0) = −1 (c) y ′′ + 8y ′ − 9y = 0 y(1) = 1 y ′ (1) = 0 (d) y ′′ + 2ay ′ + (a2 + 1)y = 0 y(0) = 1 y ′ (0) = 0 Hint.182 CHAPTER 3. Consider the equation y ′′ − (2α − 1)y ′ + α(α − 1)y = 0.10)) is a first-order linear differential equation in u. Also. for which all solutions tend toward zero as t → ∞. determine the values of α.12) (b) Let u = v ′ and show that ((A.

3. Hint: Show that dy dx dy = dt dt dx ( )2 2 d2 y dx d y d2 x dy = + . Hint.  1 x(t) = ce−t −1 4 is a solution to our system. However. −7 −15 −1 The characteristic polynomial of A is p(λ) = det(A − λI) = λ3 − 3λ2 + 3λ − 1 = (λ − 1)3 hence. −1. −1. . we can only find a single linearly independent eigenvector (1. this is not the general solution to the system.6 Project 3. we will need two other linearly independent solutions. 3. To construction a general solution to our system. there is only a single eigenvalue λ = 1.9 The Matrix Exponential Consider the linear system x′ = −17x − 46y − 7z y ′ = 8x + 21y + 3z z ′ = −7x − 15y − z The matrix associated with this system is   −17 −46 −7 A= 8 21 3 . Moreover. THE MATRIX EXPONENTIAL 183 Hint.8. An equation of this form is called an Euler equation.9. An important class of second-order linear differential equations are equations of the form t2 y ′′ + αty ′ + βy = 0. 4). 4). (a) Show that the substitution x = ln t transforms an Euler equation into an equation of constant coefficients. Thus. 7. dt2 dt dx2 dt2 dx (b) Solve the equation t2 y ′′ + 4ty ′ + 2y = 0. Euler Equations. We can only solve initial value problems where the initial condition lies on the line through the origin containing the vector (1. where t > 0 and α and β are real constants.

9. Thus. Recalling that the solution to the initial value problem x′ = kx x(0) = x0 is x(t) = x0 ekt . ∑ N 1 k SN = A k! k=0 converge. where A0 = I by convention. even a series whose individual terms are matrix expressions.1 The Exponential of a Matrix Our goal is to construct a solution to the initial value problem x′ = Ax x(0) = x0 . A series. Let us compute the exponential of ( ) s 0 A= . 0 t Actually. however. converges if and only if its partial sums. 0 et . this is quite easy. Example 3. the matrix exponential eA converges for all A. Although we shall not provide a proof. Each term makes sense in our definition since each is an expression of n × n matrices. We will define the exponential of A using the power series for et . ∑ 1 ∞ 1 2 1 eA = I + A + A + A3 + · · · = Ak . there are some issues surrounding the convergence of the power series. LINEAR SYSTEMS 3. 1 1 eA = I + A + A2 + A3 + · · · 2! 3! ( ) ( ) ( ) ( ) 1 0 s 0 1 s2 0 1 s3 0 = + + + + ··· 0 1 0 t 2! 0 t2 3! 0 t3 ( ) 1 + s + s2 /2! + s3 /3! + · · · 0 = 0 1 + t + t2 /2! + t3 /3! + · · · ( s ) e 0 = . we might guess that a solution to the initial value problem x′ = Ax x(0) = x0 has the form x(t) = etA x0 if we can make sense of the expression etA .184 CHAPTER 3. where A is an n × n matrix. 2! 3! k! k=0 where A is an n × n matrix.75.

dt Proof. Corollary 3.3. We simply need to differentiate t2 2 t3 3 etA = I + tA + A + A + ··· 2! 3! term by term. THE MATRIX EXPONENTIAL 185 Theorem 3. ( ) d tA d t2 2 t3 3 e = I + tA + A + A + · · · dt dt 2! 3! t 2 = A + tA2 + A3 + · · · ( 2! ) t2 2 t3 = A I + tA + A + + · · · 2! 3! = AetA . If x(t) = etA x0 .76.3 However.77. If we try to compute etA . . The problem is that matrix exponentials may not be so easy to compute. Example 3. The matrix ( ) 4 4 A= −9 −8 has repeated eigenvalues λ = −2.78. then d tA e = AetA . then t2 t3 etA = I + tA + A2 + A3 + · · · 2! 3! ( ) ( ) ( )2 ( )3 1 0 4 4 t2 4 4 t3 4 4 = +t + + + ··· 0 1 −9 −8 2! −9 −8 3! −9 −8 ( ) ( ) ( ) ( ) 1 0 4 4 t2 −20 −16 t3 64 48 = +t + + + ··· 0 1 −9 −8 2! −36 28 3! −108 −80 ( ) 1 + t − 20t2 /2! + 64t3 /3! + · · · 4t − 16t2 /2! + 48t3 /3! + · · · = . solving linear systems is simply a matter of computing matrix exponentials. Then x(t) = etA x0 is the solution to the initial value problem x′ = Ax x(0) = x0 Proof. 3 Since we are differentiating an infinite series.76. dt dt Thus. we still need to show that differentiating term by term is something that can be done. then d ( tA ) d ( tA ) x′ (t) = e x0 = e x0 = AetA x0 = Ax(t). If A is an n × n matrix. leave the details to a course in advanced calculus. Let A be an n × n matrix. −9t − 36t2 /2! − 108t3 /3! + · · · 1 − 8t + 28t2 /2! − 80t3 /3! + · · · It is not at all clear that this series converges to a matrix whose entries can be expressed in terms of elementary functions. The corollary follows immediately from Theorem 3. however.9. We will.

3. ∑n ( ) n n (A + B) = Ak B n−k . then eA eB = eA+B . Theorem 3. Proof. eA has inverse e−A . k k!(n − k)! . AeA = eA A. if AB = BA. Let λ be an eigenvalue of an n × n matrix A. ( ) 1 1 AeA = A I + A + A2 + A3 + · · · 2! 3! 1 1 = A + A2 + A3 + A4 + · · · ( 2! 3! ) 1 1 = I + A + A2 + A3 + · · · A 2! 3! = eA A. 2. Proving (2) is a only bit more complicated if we notice that the binomial expansion holds for matrices. LINEAR SYSTEMS Now let us see how we can use the matrix exponential to solve a linear system as well as invent a more direct way to compute the matrix exponential. we know that Av = λv. Since v is an eigenvalue for λ. Indeed. An v = A(An−1 v) = A(λn−1 v) = λn−1 Av = λn v.186 CHAPTER 3. ( ) t2 2 t3 3 e v = I + tA + A + A + · · · v tA 2! 3! t 2 t3 = v + tAv + A2 v + A3 v + · · · 2! 3! t 2 t3 = v + tλv + λ2 v + λ3 v + · · · ( 2! 3! ) t 2 t 3 = I + tλ + λ2 + λ3 + · · · v 2! 3! = eλt v. Using mathematical induction. To prove (1). Hence.79. Theorem 3. then etA v = eλt v. we can show that An has eigenvalue λn with associated eigenvector v. If v is an eigenvector for λ. Let A and B be n × n matrices. k k=0 where ( ) n n! = . The matrix exponential shares several properties with the function exponential function ex that we studied in calculus. we can simply expand both sides of the equality in a power series.80. Then 1. Proof.

9. we know that etA = et(λI+(A−λI)) = etλI et(A−λI) . since A and −A commute. −9 −6 and (A − λI)2 is the zero matrix. −9t −6t Our example suggests at the following proposition. The matrix A has repeated eigenvalue λ = −2. Now let us compute etA once again for ( ) 4 4 A= . We leave the proof of this proposition as an exercise. Simply expand each series out to see that this is true. Since the identity matrix I commutes with every matrix.75. . Part (3) follows directly from Part (2).81. −9 −8 First notice that A = λI + (A − λI). ( ) 6 4 A − λI = A + 2I = . Thus. Thus. ∑∞ 1 eA+B = (A + B)n n! n=0 ∞ ( n ( ) ) ∑ 1 ∑ n k n−k = A B n! k n=0 k=0 ∞ ( n ) ∑ ∑ 1 k n−k = A B k!(n − k)! n=0 k=0 (∞ )( ∞ ) ∑ 1 ∑ 1 = An Bn n! n! n=0 n=0 A B =e e .3. THE MATRIX EXPONENTIAL 187 providing AB = BA. etA = eλt et(A−λI) ( ) t2 t2 =e λt I + t(A − λI) + (A − λI) + (A − λI) + · · · 2 3 2! 3! ( 2 ) −2t t t2 =e I + t(A + 2I) + (A + 2I) + (A + 2I) + · · · 2 3 2! 3! = e−2t (I + t(A + 2I)) ( ) −2t 1 + 6t 4t =e . etA = et(λI+(A−λI)) = etλI et(A−λI) = eλt et(A−λI) . Consequently. We also know that eλtI = eλt I by Example 3. Example 3.

e2 = 1 .9.83. 20 52 8 Therefore. −7 −15 −1 This matrix has a single eigenvalue λ = 1.188 CHAPTER 3. We will use the standard basis vectors       1 0 0 e1 = 0 . where   −17 −46 −7 A= 8 21 3 .82.84. where   7 −1 2 A =  18 −2 6  . . We are now ready return to our original system x′ = Ax. It is easy to show that the only nonzero powers of A − λI = A − I are   −18 −46 −7 (A + I) =  8 20 3 −7 −15 −2   5 13 2 (A + I)2 = −5 −13 −2 . 2 k 2! k! Example 3. we simply compute etA v for three linearly independent vectors. LINEAR SYSTEMS Proposition 3. If A is an n × n matrix with a single eigenvalue λ. −7t + 10t 2 −15t + 26t 2 1 − 2t + 4t2 Now. −7t + 10t2 −15t + 26t2 1 − 2t + 4t2 3. e3 = 0 . ( ) tA t t2 2 e = e I + t(A + I) + (A + I) 2!   1 − 18t + 5t /2 −46t + 13t2 /2 2 −7t + t2 = et  8t − 5t2 /2 1 + 20t − 13t2 /2 3t − t2  .2 Generalized Eigenvalues Example 3. −9 2 −1 The characteristic polynomial of A is det(A − λI) = λ3 − 4λ2 + 5λ − 2 = (λ − 1)2 (λ − 2). then there exists nonnegative integer k < n such that ( ) t2 tk tA e =e λt I + t(A − λI) + (A − λI) + · · · + (A − λI) . Consider the system x′ = Ax. the general solution to our system is       1 − 18t + 5t2 /2 −46t + 13t2 /2 −7t + t2 x(t) = c1 e−t  8t − 5t2 /2  + c2 e−t 1 + 20t − 13t2 /2 + c3 e−t  3t − t2  . 0 0 1 Thus. to compute three linearly independent solutions for x′ = Ax.

−3) in this case. since it is the nullspace of A − I. THE MATRIX EXPONENTIAL 189 The eigenvalue λ1 = 1 has eigenvector v1 = (2. −3) and the eigenvalue λ2 = 2 has eigenvector v2 = (1. −9 2 −2 we have  0 1 2 (A − λI)2 = (A − I)2 = 0 3 6 . then (A−λ1 I)v = (A−I)v = 0. That is.9. which means that v is an eigenvector. v1 and v2 are linearly independent. Certainly. We wish to find a vector in the nullspace that is not a multiple of the vector v1 that is also in the nullspace of A − I. we must consider higher powers. 6. −1). Thus. 0 −1 −2 The nullspace of this matrix has dimension two. 6. The vector v = (0. −3 −1 Since λ1 = 1 has multiplicity two and we can find only one linearly independent eigenvector. Now our series truncates. 6. v must be a multiple of v1 = (2. −3) is in the nullspace of (A − I)2 . If we consider the exponential etA v = eλ1 t et(A−λ1 I) v ( ) t2 t2 =e λ1 t v + t(A − λ1 I)v + (A − λ1 I) v + (A − λ1 I) v + · · · 2 3 2! 3! where v. Since   6 −1 2 (A − λI) = (A − I) =  18 −3 6  . v1 = (2. we can find two linearly independent solutions in this case     2 1 x1 (t) = et  6  and x2 (t) = e2t  3  . we must look for vectors v such that (A − λ1 I)k v = (A − I)k v = 0. it is not possible to apply Proposition 3. 1. Since we already know that the eigenspace associated with this eigenvector has dimension one and is generated by v1 . x3 (t) = etA v = eλ1 t et(A−λ1 I) v = eλ1 t (v + t(A − λ1 I)v)     0 0 = eλ1 t 1 + t(A − I) 1 0 0   t = et 1 + 3t −t We now have a general solution for our system. 0) will do. 3. our goal is to choose v for which the series truncates. −3 −1 −t .3. Thus.82 in this case. If k = 1.       2 1 t x(t) = c1 et  6  + c2 e2t  3  + c3 et 1 + 3t .

9. 1. dt • Let A be an n × n matrix. 2. We now have a procedure for finding q linearly independent solutions corresponding to an eigenvalue λ of multiplicity q. 3. • Let A and B be n × n matrices. Theorem 3. . If λ = α + iβ has eigenvector z = x + iy.85. ◦ eA is nonsingular with inverse e−A . . then eA eB = eA+B . . 2! 3! k! k=0 • If A is an n × n matrix. ◦ if AB = BA. LINEAR SYSTEMS If λ is an eigenvalue of A and (A − λI)p v = 0 for some p ≥ 1. 3. . . Then ◦ AeA = eA A. vq } for the nullspace of (A − λI)p . 2. then v is called a generalized eigenvector of A. For each vj (j = 1.190 CHAPTER 3. Then there exists an integer p ≤ q such that the dimension of the nullspace of (A − λI)p is q. q). we can compute extra solutions by looking for vectors in the nullspace of (A − λI)p for p > 1. If v is an eigenvector for λ. . then set x = Re z and y = Im z. Suppose that λ is an eigenvalue of A with multiplicity q.3 Important Lessons • If A is an n × n matrix. Find a basis {v1 . then etA v = eλt v. we have the solution xj (t) = etA vj ( ) tp−1 =e λt vj + t(A − λI)vj + · · · + (A − λI) vj . . . We leave the proof of the theorem as an exercise in linear algebra. When eigenvalues have algebraic multiplicity greater than one. p−1 (p − 1)! This procedure works for complex eigenvalues as well as real. we define the exponential of A to be ∑ 1 ∞ 1 2 1 eA = I + A + A + A3 + · · · = Ak . . then d tA e = AetA . Find the smallest integer p such that the nullspace of (A − λI)p has dimension q. . Then x(t) = etA x0 is the solution to the initial value problem x′ = Ax x(0) = x0 • Let λ be an eigenvalue of an n × n matrix A. The following theorem tells us that this is always possible.

. . Thus. . (b) Assume that y = v(t)x1 (t) = v(t)e−bt/2a is a solution to ax′′ + bx′ + cx = 0 and show that v(t) = c1 + c2 t.9. where a ̸= 0 and b2 − 4ac = 0.3. p−1 (p − 1)! The procedure works for complex eigenvalues as well as real. (a) Show that x1 (t) = e−bt/2a is a solution to ax′′ + bx′ + cx = 0. 3. Solve the system x′ = −5x + 15y − 4z y ′ = −x + 5y − z z ′ = 4x − 6y + 3z x(0) = −1 y(0) = 2 z(0) = 0 Hint. 2. . then set x = Re z and y = Im z. THE MATRIX EXPONENTIAL 191 • If A is an n×n matrix with a single eigenvalue λ. we have the solution xj (t) = etA vj ( ) tp−1 =e λt vj + t(A − λI)vj + · · · + (A − λI) vj . Find the general solution of the system x′1 = 6x1 + 9x2 + x3 + 4x4 x′2 = −x1 + x2 + x3 x′3 = 3x1 + 8x2 + 4x3 + 4x4 x′4 = −3x1 − 8x2 − 7x3 − 5x4 Hint.9.4 Exercises 1. Let ax′′ + bx′ + cx = 0. ◦ Find a basis {v1 . . . If λ = α + iβ has eigenvector z. Then there exists an integer p ≤ q such that the dimension of the nullspace of (A − λI)p is q. . . q). 2! k! • Suppose that λ is an eigenvalue of A with multiplicity q. ◦ Find the smallest integer p such that the nullspace of (A − λI)p has dimension q. • The procedure for finding q linearly independent solutions corresponding to an eigen- value λ of multiplicity q is the following. then there exists nonnegative integer k < n such that ( ) t2 tk etA = eλt I + t(A − λI) + (A − λI)2 + · · · + (A − λI)k . vq } for the nullspace of (A − λI)p . x(t) = c1 e−bt/2a + c2 te−bt/2a is a general solution for ax′′ + bx′ + cx = 0. . 3. 2. . ◦ For each vj (j = 1.

for which all nonzero solutions become unbounded as t → ∞. Solve each of the following initial value problems.192 CHAPTER 3. for which all solutions tend toward zero as t → ∞. if any. if any. 6. determine the values of α. Also. Hint. (a) y ′′ − 2y ′ + 5y = 0 y(π/2) = 0 y ′ (π/2) = 2 (b) 9y ′′ − 12y ′ + 4y = 0 y(0) = 2 y ′ (0) = −1 (c) y ′′ + 8y ′ − 9y = 0 y(1) = 1 y ′ (1) = 0 (d) y ′′ + 2ay ′ + (a2 + 1)y = 0 y(0) = 1 y ′ (0) = 0 Hint. 4. (c) Show that x1 (t) = 1/t is a solution to 2t2 x′′ + 3tx′ − x = 0 (3. Suppose that x1 (t) is a solution (not identically zero) to the equation x′′ + p(t)x′ + q(t)x = 0. (a) Assume that x(t) = v(t)x1 (t) is a solution to x′′ + p(t)x′ + q(t)x = 0 and derive the equation x1 v ′′ + (2x′1 + px1 )v ′ = 0.14) (b) Let u = v ′ and show that ((A.15) for t > 0 and find a second linearly independent solution to ((A.10)) is a first-order linear differential equation in u. . LINEAR SYSTEMS Hint. Consider the equation y ′′ − (2α − 1)y ′ + α(α − 1)y = 0.11)). Determine all values of α. 5. Reduction of Order. (3.

THE MATRIX EXPONENTIAL 193 Hint. Hint: Show that dy dx dy = dt dt dx 2 ( )2 2 d y dx d y d2 x dy = + . An important class of second-order linear differential equations are equations of the form t2 y ′′ + αty ′ + βy = 0. 3. (a) Show that the substitution x = ln t transforms an Euler equation into an equation of constant coefficients. Euler Equations.5 Project .3.9. 7. Hint.9. dt2 dt dx2 dt2 dx (b) Solve the equation t2 y ′′ + 4ty ′ + 2y = 0. An equation of this form is called an Euler equation. where t > 0 and α and β are real constants.

Our new circuit is called an RLC circuit (Figure 4. a coil that generates a magnetic field. We will now add an inductor such as a solenoid. Thus. R + C E(t) I(t) − L Figure 4.1: An RLC Circuit 194 . b. We will first consider the case ax′′ + bx′ + cx = 0. An equation of this form is said to be homoge- neous with constant coefficients. We already know how to solve such equations since we can rewrite them as a system of first-order linear equations. a capacitor.1 Recall the RC circuits that we studied earlier (see Section 1. and a toaster or an electric stove is an example of something that might provide a resistance in a circuit. radios. and a resistor.1). where a. we can find the general solution of a homogeneous second-order linear differential equation with constant coeffi- cients by computing the eigenvalues and eigenvectors of the matrix of the corresponding system. A battery or generator is an example of a voltage source. 4 Second-Order Linear Equations 4. Inductor applications include transformers. 4.3).1. televisions. Capacitors store an electrical charge and are used in electronic flashes for cameras. power supplies.1 Homogeneous Linear Equations A differential equation of the form a(t)x′′ + b(t)x′ + c(t)x = g(t) is called a second-order linear differential equation. and c are constants and a ̸= 0. Such circuits contained a voltage source.

Inductance on the coil.1. and C = 1. we have the differential equation dI 1 L + RI + Q = E(t) (4. I(t). The following laws from physics govern how our circuit behaves. The impressed voltage. x′ = y c b y ′ = − x − y. the resistance R is measured in ohms.2) take the same form. L. is the rate at which a charge flows through this circuit and is measured in amperes or amps. (4.1) dt C or 1 LQ′′ + RQ′ + Q = E(t). The charge on the capacitor Q(t) at time t is measured in coulombs. dQ • I= .1. Applying Kirchhoff’s Second Law to our circuit. is measured in henrys. Solutions of a linear system x′ = Ax often include terms of the form ert . E(t).2) Since we can rewrite this equation as a system of first-order linear equations. Our equation becomes I ′′ + I ′ + I = E ′ (t) = cos t. We assign a direction to the current. we have 1 LI ′′ + RI ′ + I = E ′ (t). • The voltage drop across an inductor is L(dI/dt). and a current flowing in the opposite direction will be given negative values.2 Second-Order Linear Equations Suppose that we have a homogeneous second-order linear differential equation with constant coefficients. HOMOGENEOUS LINEAR EQUATIONS 195 Current. we might consider an RLC circuit with R = 1.1). a a we can find the general solution by computing the eigenvalues and eigenvectors of the matrix of the corresponding system. and the capacitance C is measured in farads. C Differentiating both sides of (4. C For example. 4. L = 1. • The voltage drop across a capacitor is Q/C. is measured in volts. This is an example of a second-order linear differential equation. the impressed voltage on the circuit is given by E(t) = sin(t). dt • The voltage drop across a resistor is IR (Ohm’s Law). It makes sense that solutions to equation (4. . ax′′ + bx′ + cx = 0. At t = 0 when both I(0) = 0 and I ′ (0) = Q(0) = 0.4. • In a closed circuit the impressed voltage is equal to the sum of the voltage drops in the rest of the circuit (Kirchhoff’s Second Law).

we can find the general solution for this equation by finding the roots of the quadratic polynomial aλ2 + bλ + c. we can substitute this expression into the left-hand side of (4.196 CHAPTER 4. Consider the equation x′′ + 7x′ + 10x = 0.3) to obtain d2 rt d 2 e + 7ert + 10ert = r2 ert + 7rert + 10ert dt dt = (r2 + 7r + 10)ert = (r + 5)(r + 2)ert .2. respectively. x(t) = c1 x1 (t) + c2 x2 (t) = c1 e−5t + c2 e−2t (4.4) is a solution to x′′ + 7x′ + 10x = 0. Applying the startegy in Example 4. we find that (r + 5)(r + 2) = 0 or r = −5 or −2. y(t) −5 −2 which agrees with (4. In general. 2a If b2 − 4ac > 0. where a ̸= 0. the solution to our system is ( ) ( ) ( ) x(t) −5t 1 −2t 1 = c1 e + c2 e . (4. Consequently. this is the general solution of our second-order equation since we have a one-to- one correspondence between the solutions of x′′ + 7x′ + 10x = 0 and the system x′ = y y ′ = −10x − 7y. The matrix associated with this system ( ) 0 1 A= −10 −7 has eigenvalues λ1 = −5 and λ2 = −2 with eigenvectors v1 = (1. Thus. −2).4). 2a 2a . Since ert is never zero.3) If we assume that a solution is of the form ert .2. suppose that ax′′ + bx′ + cx = 0. we have two solutions x1 (t) = e−5t and x2 (t) = e−2t . Indeed. SECOND-ORDER LINEAR EQUATIONS Example 4. −5) and v2 = (1. By the Principle of Superposition. we have real roots √ √ −b + b2 − 4ac −b − b2 − 4ac λ1 = and λ2 = . √ −b ± b2 − 4ac λ=− .

If b2 − 4ac > 0. (4. we have real roots √ √ −b + b2 − 4ac −b − b2 − 4ac λ1 = and λ2 = . To prove that equation (4. . the general solution to the system of differential equations x′ = Ax is ( ) ( ) ( ) ( ) x(t) x(t) λ1 t 1 λ2 t 1 x(t) = = = c1 e + c2 e . Again.5). Thus. respectively. 2a 2a We can find eigenvectors ( ) ( ) √ 1 1 v1 = = (−b + b2 − 4ac )/2a λ1 ( ) ( ) v1 = √ 1 = 1 (−b − b2 − 4ac )/2a λ2 for λ1 and λ2 . Substituting this function into our differential equation.5) is indeed the general solution to the second-order equation ax + bx′ + cx = 0. we can study the equivalent system of linear equations. If we let y = x′ .4. 2 Thus. r2 + 4r + 5 = 0 or √ −4 ± −4 r= = −2 ± i.1. As with the real case. −c/a −b/a The characteristic polynomial of A is b c p(λ) = det(A − λI) = λ2 + λ + . we will assume that our solution has the form x(t) = ert . y(t) x′ (t) λ1 λ2 which agrees with (4. Now let us solve the initial value problem x′′ + 4x′ + 5x = 0 x(0) = 1 x′ (0) = 1.3. we can find a solution x(t) = e(−2+i)t = e−2t eit = e−2t (cos t + i sin t). HOMOGENEOUS LINEAR EQUATIONS 197 and the solution to our second-order differential equation is x(t) = c1 eλ1 t + c2 eλ2 t . ′′ the corresponding linear system is x′ = Ax. a a The roots of p(λ) are the same as the roots of aλ2 + bλ + c.5) where c1 and c2 are arbitrary constants. where ( ) 0 1 A= . we find that 0 = x′′ + 4x′ + 5x = r2 ert + 4rert + 5ert = (r2 + 4r + 5)ert . Example 4.

As before. we first calculate x′ (t) = −2e−2t (c1 cos t + sin t) + e−2t (−c1 sin t + c2 cos t). Indeed. −c/a −b/a If b2 − 4ac < 0. respectively. 2a 2a The vector v = (1. Thus. α cos βt − β sin βt α sin βt + β cos βt . since x(t) = x1 (t) + ix2 (t) is a solution. Thus. the solution to our initial value problem is x(t) = e−2t cos t + 3e−2t sin t. the corresponding linear sytem is x′ = Ax. SECOND-ORDER LINEAR EQUATIONS The real and imaginary parts of our solution are x1 (t) = e−2t cos t x2 (t) = e−2t sin t.198 CHAPTER 4. the general solution to our equation is x(t) = c1 e−2t cos t + c2 e−2t sin t. Hence. Since the real part and the imaginary part of x(t) must both be zero. we can conclude that ax′′1 + bx′1 + cx1 = 0 and ax′′2 + bx′2 + cx2 = 0. a solution to our system of differential equations is ( ) x(t) x(t) = y(t) ( ) x(t) = x′ (t) ( ) (α+iβ)t 1 =e α + iβ ( ) αt 1 = e (cos βt + i sin βt) α + iβ ( ) ( ) cos βt sin βt = eαt + ieαt . To apply our initial conditions x(0) = 1 and x′ (0) = 1. and c1 = 1 and c2 = 3. where √ b 4ac − b2 α=− and β = . Therefore. We claim that both x1 (t) and x2 (t) are solutions to our differential equation. where ( ) 0 1 A= . α + iβ) is an eigenvector for λ. the eigenvalue of A are λ = α + iβ and λ = α − iβ. 0 = ax′′ + bx′ + cx = a(x1 + ix2 )′′ + b(x1 + ix2 )′ + c(x1 + ix2 ) = (ax′′1 + bx′1 + cx1 ) + i(ax′′2 + bx′2 + cx2 ). 1 = x(0) = c1 1 = x′ (0) = −2c1 + c2 .

the general solution to ax′′1 +bx′1 +cx1 = 0 is x(t) = c1 eαt cos βt + c2 eαt sin βt.3 Reduction of Order Given a second-order linear differential equation with constant coefficients. the general solution to x′′ + 2x′ + x = 0 is x(t) = c1 e−t + c2 te−t . 4. then x′ (t) = v(t)x′1 (t) + v ′ (t)x1 (t) = −v(t)e−t + v ′ (t)e−t and x′′ (t) = v ′′ (t)x1 (t) + 2v ′ (t)x′1 (t) + v(t)x′′1 (t) = v ′′ (t)e−t − 2v ′ (t)e−t + v(t)e−t . Since we already know that cx1 (t) is a solution to our differential equation. In order to find a general solution to x′′ + 2x′ + x = 0. and the second solution to our equation is x = te−t . x′′ + 2x′ + x = [v ′′ e−t − 2v ′ e−t + ve−t ] + 2[−ve−t + v ′ e−t ] + [ve−t ] = e−t v ′′ = 0.4. if x(t) = v(t)x1 (t) = v(t)e−t . In order to find a general solution to x′′ + 2x′ + x = 0. Example 4. HOMOGENEOUS LINEAR EQUATIONS 199 Taking the real and imaginary parts of x(t). Therefore. . x1 (t) = eαt cos βt and x2 (t) = eαt sin βt. we obtain two real solutions to the system. We leave it as an exercise to show that our solution agrees with the solution that we would obtain from solving the equivalent first-order linear system.4. We have covered the case where this equation has two distinct real solutions as well as when there are complex soluitons.1. Hence. we can assume that v(t) = t. we find x′′ + 2x′ + x = λ2 eλt + 2λeλt + eλt = eλt (λ + 1)2 = 0. Letting c1 = 1 and c2 = 0. If we choose eλt as our guess. but what if there is only a single real solution λ = −b/2a. Indeed. our strategy has been to solve the characteristic equation aλ2 + bλ + c = 0 to obtain two linearly independent solutions. Consider the equation x′′ + 2x′ + x = 0. Consequently.1. Thus. and v ′′ = 0. we will try to generalize this observation by replacing c with a nonconstant function v(t) and then try to determine v(t) so that v(t)x1 (t) is a solution to x′′ + 2x′ + x = 0. we must find a second solution that is not a multiple of x1 (t) = e−t . Therefore. v = c1 t + c2 . ax′′ + bx′ + cx = 0. we must find a second solution that is not a multiple of x1 (t) = e−t . λ = −1 and we have a solution x1 (t) = e−t .

• If b2 − 4ac < 0. where a ̸= 0 and b2 − 4ac > 0. x′ = ax + by y ′ = cx + dy. where λ = −b/2a. We did a great deal of work finding unique solutions to first-order linear systems of equations. the differential equation ax′′ + bx′ + cx = 0 has a general solution x(t) = c1 e−bt/2a + c2 te−bt/2a . That is. Since we can rewrite this equation as a system of first-order linear differential equations. SECOND-ORDER LINEAR EQUATIONS The technique that we have used in Example 4. We leave it as an exercise to show that this technique works in general. then the general solution is given by x(t) = c1 eλt + c2 teλt .4 Important Lessons • A second-order linear differential equation with constant coefficients is an equation of the form ax′′ + bx′ + cx = 0. We can guess the solution to this equation. we can determine the general solution to ax′′ + bx′ + cx = 0. we are guaranteed the existence and uniqueness of solutions. where α ± iβ are the roots of ar2 + br + c = 0. • Suppose that ax′′ + bx′ + cx = 0. .4 is called reduction of order. Our efforts are now rewarded. given a second-order linear differential equation ax′′ + bx′ + cx = 0 such that b2 − 4ac = 0. • If b2 − 4ac = 0.1. 4. the general solution to this differential equation is x(t) = c1 er1 t + c2 er2 t . Since each second-order homogeneous system with constant coefficients can be rewritten as a first-order linear system. the differential equation ax′′ + bx′ + cx = 0 has a general solution x(t) = c1 eαt cos βt + c2 eαt sin βt.200 CHAPTER 4. If the roots of ar2 + br + c are r1 and r2 .

2. Thus. (c) Show that x1 (t) = 1/t is a solution to 2t2 x′′ + 3tx′ − x = 0 (4.3) or an RLC circuit (Section 4.1) can be modeled with second-order linear differential equations. RLC circuits can also be modeled with provide another example of forcing.1.1. b is the damping coefficient. where m is the mass.4. (a) Show that x1 (t) = e−bt/2a is a solution to ax′′ + bx′ + cx = 0. FORCING 201 4. 4. where a ̸= 0 and b2 − 4ac = 0. g(t) and E ′ (t).7). is (measured in henrys). Such a term is called a forcing term.5 Exercises 1. x(t) = c1 e−bt/2a + c2 te−bt/2a is a general solution for ax′′ + bx′ + cx = 0. and F (t) = g(t) represents some external force applied to our system. C What is different about these two equations are the terms on the righthand side.6) is a first-order linear differential equation in u. R is the resistance (measured in ohms). 2. L. Indeed. . and the inductance. C is the capacitance (measured in farads). k is the spring constant.6 Project 4.2 Forcing Harmonic oscillators such as a spring-mass system (Subsection 1. Suppose that x1 (t) is a solution (not identically zero) to the equation x′′ + p(t)x′ + q(t)x = 0. we can model a spring-mass system with the equation mx′′ (t) + bx′ (t) + kx(t) = g(t).1. Reduction of Order. then the derivative of the impressed voltage (measured in volts). (a) Assume that x(t) = v(t)x1 (t) is a solution to x′′ + p(t)x′ + q(t)x = 0 and derive the equation x1 v ′′ + (2x′1 + px1 )v ′ = 0. If I(t) is the rate at which a charge flows through a circuit (measured in amperes or amps).6) (b) Let u = v ′ and show that (4. E(t). is the forcing term 1 LI ′′ + RI ′ + I = E ′ (t). Let ax′′ + bx′ + cx = 0. (b) Assume that y = v(t)x1 (t) = v(t)e−bt/2a is a solution to ax′′ + bx′ + cx = 0 and show that v(t) = c1 + c2 t.7) for t > 0 and find a second linearly independent solution to (4. (4.

(4. where yh is the general solution of the homogeneous equation x′′ + p(t)x′ + q(t)x = 0. let us derive some fundamental facts about second-order linear differential equations. we know that x′′1 + p(t)x′1 + q(t)x1 = g(t) x′′2 + p(t)x′2 + q(t)x2 = g(t).8). Therefore.9). Then xq − xp is a solution to the homogeneous equation x′′ + p(t)x′ + q(t)x = 0. In general. Theorem 4. We state this fact in the following theorem. Proof. these equations can be difficult to solve for an arbitrary function g(t). Our goal is to be able to solve such equations. We have already seen how examples of such equations arise when examining models of harmonic oscillators with forcing terms. . SECOND-ORDER LINEAR EQUATIONS 4. Since x1 and x2 are solutions of (4. We can use Theorem 4. suppose that xq is another solution to (4.9) can be written in the form x = xh + xp . Indeed. and xp is any solution of (4.1 Nonhomogeneous Equations A nonhomogeneous second-order linear differential equation is an equation of the form x′′ + p(t)x′ + q(t)x = g(t).2. d2 d 2 (x1 − x2 ) + p(t) (x1 − x2 ) + q(t)(x1 − x2 ) ( 2 dt dt) ( ) d x1 dx1 d2 x 2 dx2 = + p(t) + q(t)x1 − + p(t) + q(t)x2 dt2 dt dt2 dt = g(t) − g(t) = 0. Then x1 (t)−x2 (t) is a solution of the homogeneous linear differential equation x′′ + p(t)x′ + q(t)x = 0.8) has solutions x1 = x1 (t) and x2 = x2 (t).9).5. xq − xp = xh or xq = xh + xp .5 to derive the fact that the general solution to x′′ + p(t)x′ + q(t)x = g(t). Before we attempt to find solutions for some of the more common functions that might occur for g(t).202 CHAPTER 4. Thus. Suppose that x′′ + p(t)x′ + q(t)x = g(t) (4.

certain forcing functions often occur in practice.2.2 Forcing Terms The equation mx′′ + bx′ + kx = g(t) can be used to model a harmonic oscillator where forcing occurs.2. . In the case of the unforced damped harmonic oscillator. all solutions of a damped harmonic oscillator with nonzero damping are essentially the same for large values of t. where a force is applied periodically. 4. Some of the more important forcing functions are g(t) = e−at . In general. Thus. then x = xh + xp → xp as t → ∞.4. Then the general solution to x′′ + p(t)x′ + q(t)x = g(t) is x = xh + xp . b > 0. FORCING 203 Theorem 4. the solution is a sink. If x = xh + xp is the general solution to the equation. and k > 0. −q −p has trace of −p and determinant q. we will not be able to solve this equation explicitly for a given g(t). the harmonic oscillator is x′ = y y ′ = −qx − py. where the external force decreases exponentially over time. where p > 0 and q > 0. however.7. Theorem 4. In other words. g(t) = k. As a first-order linear system. The matrix corresponding to this system.6. and g(t) = cos ωt or g(t) = sin ωt. This leads us to the following conclusion. mx′′ + bx′ + kx = 0. ( ) 0 1 A= . we know that m > 0. That is. and xh be the general solution of the corresponding homogeneous equation x′′ + p(t)x′ + q(t)x = 0. we can rewrite this equation as x′′ + px′ + qx = 0. we know that any solution of the unforced equation tends toward the origin as t → ∞. Since tr(A) < 0 and det(A) > 0. where p = b/m and q = k/m are both positive. where a constant force is applied. Suppose x′′ + px′ + qx = g(t) has solution x = xh + xp . Let xp be a particular solution of the equation x′′ + p(t)x′ + q(t)x = g(t).

Let us solve the differential equation x′′ + 5x′ + 4x = 1. all solutions approach zero as t → ∞ (Figure 4. Therefore. the general solution is 1 x(t) = c1 e−t + c2 e−4t + . Suppose that we wish to solve x′′ + 5x′ + 4x = e−2t .11). we find that e−2t = x′′p + 5x′p + 4xp = 4Ae−2t − 10Ae−2t + 4Ae−2t = −2Ae−2t .9). x(t) dx/dt 2 1 2 4 6 8 10 t -1 Figure 4. 2 Again.10) a harmonic oscillator with a constant forcing function. We will use the method of undetermined coefficients to find a particular solution to (4. It is easy to check that the general solution to the homogeneous equation x′′ + 5x′ + 4x = 0 is xh = c1 e−t + c2 e−4t . It is reasonable to guess that a particular solution will have the form xp = Ae−2t . Hence. We already know the solution to the homogeneous equation.9: Solutions to x′′ + 5x′ + 4x = 1 Example 4. . all solutions approach the constant solution x = 1/4 (Figure 4.8. the solution that we seek is 1 x = c1 e−t + c2 e−4t − e−2t .2.11). (4.10) is given by xp = 1/4.11).11) This is the equation of a harmonic oscillator with a forcing function that decreases expo- nentially with time.3 The Method of Undetermined Coefficients Example 4.10. A particular solution to (4. (4. SECOND-ORDER LINEAR EQUATIONS 4. A = −1/2. Thus. Now let us consider a more complicated example.204 CHAPTER 4. 4 As t → ∞. Substituting this expression into (4.

Then x′p = −A sin t + B cos t x′′p = −A cos t − B sin t.2. FORCING 205 x(t) dx/dt 2 1 2 4 6 8 10 t -1 Figure 4. .4. We must solve the following system of equations to find a particular solution: 3A + 5B = 2 −5A + 3B = 0. takes the form xp = A cos t + B sin t. The general solution to our equation is 3 5 x = c1 e−t + c2 e−4t + cos t + sin t. 3 5 xp = cos t + sin t 17 17 is a particular solution to x′′ + 5x′ + 4x = 2 cos t.11: Solutions to x′′ + 5x′ + 4x = e−2t Example 4.13. we see that 2 cos t = x′′p + 5x′p + 4xp = (−A cos t − B sin t) + 5(−A sin t + B cos t) + 4(A cos t + B sin t) = (3A + 5B) cos t + (−5A + 3B) sin t. Substituting these expressions into the differential equation.12. Now let us examine what happens if we have a periodic forcing function. Let us assume that the particular solution to the equation x′′ + 5x′ + 4x = 2 cos t. Consequently. The solution of this system is A = 3/17 and B = 5/17. 17 17 The solutions to this equation are given in Figure 4.

Using this guess. d2 d e−t = 2 Ate−t + 5 Ate−t + 4Ate−t dt dt = A(−e−t − e−t + te−t ) + 5A(e−t − te−t ) + 4Ate−t = 3Ate−t .13: Solutions to x′′ + 5x′ + 4x = 2 cos t Example 4. 3 Solutions to the differential equation x′′ + 5x′ + 4x = e−t are given in Figure 4. Thus. Such a function must yield a multiple of e−t when differentiated. SECOND-ORDER LINEAR EQUATIONS x(t) dx/dt 2 1 2 4 6 8 10 t -1 Figure 4. As a final example. Our guess of xp (t) = Ae−t for a particular solution will no longer work since e−t is a solution to the homogeneous equation. consider the x′′ + 5x′ + 4x = e−t . The simplest such function is of the form xp = Ate−t .14. . A = 1/3 and our general solution is 1 x = c1 e−t + c2 e−4t + te−t . Recall that the solution to the homogeneous equation x′′ + 5x′ + 4x = 0 is xh = c1 e−t + c2 e−4t .206 CHAPTER 4.15. We must therefore consider a function of a different form.

Here s = 0. • Let xp be a particular solution of the equation x′′ + p(t)x′ + q(t)x = g(t).4 A Strategy We outline a general strategy for choosing xp for the Method of Undetermined Coefficients in Table 4. 1. • Suppose that x′′ + p(t)x′ + q(t)x = g(t) has solutions x1 = x1 (t) and x2 = x2 (t). 2 is the smallest integer that will ensure that no term in xp is a solution of the corresponding homogeneous equation. .16. g(t) xp Pn (t) = an + · · · + a1 t + a0 tn t (An t + · · · + A1 t + A0 ) s n Pn (t)eαt ts (An tn + · · · + A1 t + A0 )eαt { sin βt Pn (t)e αt ts eαt [(An tn + · · · + A1 t + A0 ) cos βt cos βt +(Bn tn + · · · + B1 t + B0 ) sin βt] Table 4. 4.15: Solutions to x′′ + 5x′ + 4x = e−t 4.16: Particular solutions of x′′ + px′ + qx = g(t).2.4. FORCING 207 x(t) dx/dt 2 1 2 4 6 8 10 t -1 Figure 4. Then x1 (t) − x2 (t) is a solution of the homogeneous linear differential equation x′′ + p(t)x′ + q(t)x = 0.2.2. Forced harmonic oscillators and RLC circuits provide good examples of nonhomoge- neous second-order linear differential equations.5 Important Lessons • A nonhomogeneous second-order linear differential equation is an equation of the form x′′ + p(t)x′ + q(t)x = g(t).

Hint. Hint. 235 Thus.2. Using the method of undetermined coefficients. Find the solution of the initial-value problem y ′′ + 6y ′ + 8y = cos 3t y(0) = 0 y ′ (0) = 0. • The Method of Undetermined Coefficients is useful for solving the equation x′′ + p(t)x′ + q(t)x = g(t). 235 Applying the initial conditions y(0) = 0 and y ′ (0) = 0.6 Exercises 1. Compute the solution of the initial value problem y ′′ + 4y = 3 cos 2t y(0) = y ′ (0) = 0. In particular. all solutions of the equation x′′ + p(t)x′ + q(t)x = g(t) approach xp (t) as t → ∞. y ′′ + 4y = 0 is yh = c1 cos 2t + c2 sin 2t.208 CHAPTER 4. SECOND-ORDER LINEAR EQUATIONS and xh be the general solution of the corresponding homogeneous equation x′′ + p(t)x′ + q(t)x = 0. we can find a particular solution 2 yp = (18 sin 3t − cos 3t). we have 16 −4t 18 −2t 2 y=− e + e + (18 sin 3t − cos 3t). the general solution is 2 y = yh + yp = c1 e−4t + c2 e−2t + (18 sin 3t − cos 3t). . when g is of the form { αt sin βt g(t) = Pn (t)e cos βt 4. if the solution to x′′ + p(t)x′ + q(t)x = 0 has a sink at the origin. The solution to the homogeneous equation y ′′ + 7y ′ + 10y = 0 is yh = c1 e−4t + c2 e−2t . 235 235 235 2. Then the general solution to x′′ + p(t)x′ + q(t)x = g(t) is x = xh + xp . The complex version of the equation is y ′′ + 4y = 3e2it . The solution to the homogeneous equation.

442 √ where ϕ = π + arctan(−21). 4. we obtain 4 4 y=√ sin 3t + ϕ = √ cos(3t − θ). We include a factor of t. What is the amplitude. to be linearly independent on an open interval I = (a. a = 3/(4i) = −3i/4. Substituting yc and yc′′ into the differential equation. and −4 −2 + 42i A= = . . f (t) and f (t) are linearly dependent. If we assume that the complex solution has form yc = Ae3it . f (t) and g(t). FORCING 209 Assume that the particular solution to this equation is yc = ate2it . 4 Applying the initial conditions y(0) = y ′ (0) = 0. yc = Ae3it = 4ei(3t+ϕ) / 442.4. Use the complex method to find a particular solution to y ′′ + 7y ′ + 10y = −4 sin 3t. Otherwise. Suppose that f (t) and g(t) are solutions to the homogeneous linear equation y ′′ + py ′ + qy = 0. 4 we have 3 yp = t sin 2t. we find that c1 = c2 = 0 and 3 y = t sin 2t. the period.2. and the frequency of your solution? Hint. The amplitude is 4/ 442. then yc′′ + 7yc′ + 10yc = A(1 + 21i)e3it = −4e3it . 442 442 √ where θ = ϕ − π/2 = π/2 + arctan(−21). since e2it is a solution to the homogeneous equation. 1 + 21i 221 In polar form. Using the imaginary part of this solution. b) if there do not exist nonzero constants c1 and c2 such that c1 f (t) + c2 g(t) = 0 for all t ∈ I. We define two functions. 4 Therefore the general solution is 3 y = yh + yp = c1 cos 2t + c2 sin 2t + t sin 2t. and the frequency is 3/2π. 4 3. Thus. we have yc′′ + 4yc = 4a(i − t)e2it + 4ate2it = 4aie2it = 3e2it . Taking the real part of 3 yc = − it(cos 2t + i sin 2t). the period is 2π/3. Write your solution as a function of cosine. 4 iϕ A = |A|eiϕ = √ e . Thus.

dt g g g2 and f /g is constant on the interval (c. d). f and g are linearly dependent. Abel’s Theorem. Then one function is a multiple of the other. y2 ](t) = c exp − dt = c exp − ln t = ct−3/2 . g) = 2 =− = 0. ( ) f (t) g(t) W (f. d) and g does not vanish on this interval. b). Assume that g(t0 ) ̸= 0 for some t0 in (a. Since g is differentiable. f (t) g ′ (t) for all t in (a. If g = 0. Thus. b).210 CHAPTER 4. b). Hint. where p and q are continuous on an open interval I = (a. y2 ](t) = c exp − p(t) dt . it must also be continuous and there is some interval (c. b) such that t0 ∈ (c. f ′ (t) = cg ′ (t). b) if and only if W [f. for some constant c that depends on y1 and y2 but not on t. ( ) d f f ′g − f g′ W (f. If y1 and y2 are solutions of the homogeneous equation y ′′ + p(t)y ′ + q(t)y = 0. Since f and cg are both solutions to the differential equation y ′′ + py ′ + qy = 0 and have the same initial condition. f (t) = cg(t) for all t ∈ (a. g](t) ≡ 0. Therefore. SECOND-ORDER LINEAR EQUATIONS Show that f (t) and g(t) are linearly dependent on an interval I = (a. g)(t) = det ′ = 0. Thus. d) contained in (a. 5. (a) Use Abel’s Theorem to find the Wronskian of 2t2 y ′′ + 3ty ′ − y = 0 up to a constant multiple. we know that y1′′ + p(t)y1′ + q(t)y1 = 0 . (b) Prove Abel’s Theorem. b) by the existence and uniqueness theorem. f (t0 ) = cg(t0 ) and f (t′0 ) = cg ′ (t0 ). show that ( ∫ ) W [y1 . then 0f = g and the two functions are linearly dependent. Hint. (a) We can rewrite 2t2 y ′′ + 3ty ′ − y = 0 as 3 ′ 1 y ′′ + y − 2 y = 0. 2t 2t Since p(t) = 1/2t. say f (t) = cg(t). Abel’s Theorem tells us that ( ∫ ) ( ) 3 3 W [y1 . 2t 2 (b) Since y1 and y2 are solutions to our differential equation. Consequently. Suppose that that f (t) and g(t) are linearly dependent on an interval I = (a. f ′ (t) g ′ (t) Conversely. suppose that ( ) f (t) g(t) W (f. where t > 0. b). g)(t) = det = f (t)g ′ (t) − f ′ (t)g(t) = cg(t)g ′ (t) − cg ′ (t)g(t) = 0.

y ′′ + p(t)y ′ + q(t)y = f (t). y2 )(t) = y1 y2′ − y1′ y2 . y2 ](t) y1 (t)f (t) u′2 (t) = . The Method of Variation of Parameters.13) if we know know that the general solution to the homogeneous equation y ′′ +p(t)y ′ +q(t)y = 0 is yh = c1 y1 + c2 y2 .13) is ∫ t ∫ t y2 (s)f (s) y1 (s)f (s) yp = −y1 (t) ds + y2 (t) ds. then W ′ = y1 y2′′ − y1′′ y2 . 1 2y ](s) t0 W [y1 . y2 ](s) ∫ t y1 (s)f (s) u2 (t) = ds t0 W [y1 . y2 ](s) for any point t0 in I. This equation is separable with solution ( ∫ ) W (t) = c exp − p(t) dt . (b) Show that −y2 (t)f (t) u′1 (t) = W [y1 . (c) If p. In this problem we will describe another method of finding a particular solution to a nonhomogeneous equation. where u′1 (t)y1 (t) + u′2 (t)y2 (t) = 0.13) has the form yp (t) = u1 (t)y1 (t) + u2 (t)y2 (t).12) If W (t) = W (y1 . and equation (A. Multiplying the first equation by y2 and the second equation by y1 and subtracting. Substitute yp into the left-hand side of (A. and f are continuous on an interval I.13) to show that u′1 (t)y1′ (t) + u′2 (t)y2′ (t) = f (t). y2 ](t) = y1 (t)y2′ (t) − y2 (t)y1′ (t) is the Wronskian of y1 and y2 .4. 6. (4. t0 W [y . a particular solution to (A. Consequently. W [y1 . (4. (a) Assume that a particular solution of (A. FORCING 211 y2′′ + p(t)y2′ + q(t)y2 = 0. y2 ](t) where W [y1 .12) becomes W ′′ + p(t)W = 0. show that ∫ t y2 (s)f (s) u1 (t) = − ds t0 W [y1 . y2 ](s) . q. we obtain (y1 y2′′ − y1′′ y2 ) + p(t)(y1 y2′ − y1′ y2 ) = 0.2.

2 Integrating. 2 . Substituting these expressions into equation (A. we have yp′′ + pyp′ + qyp = (u′1 y1′ + u1 y1′′ + u′2 y2′ + u2 y2′′ ) + p(u1 y1′ + u2 y2′ ) + q(u1 y1 + u2 y2 ) = u1 [y1′′ + py1′ + qy1 ] + u2 [y2′′ + py2′ + qy2 ] + u′1 y1′ + u′2 y2′ = u′1 y1′ + u′2 y2′ = f (t). we obtain −y2 (t)f (t) u′1 (t) = W [y1 . W [y1 . we obtain u1 (t) = −3 sin t 3 u2 (t) = ln | csc t − cot t| + 3 cos t.13). for u′1 and u′2 . SECOND-ORDER LINEAR EQUATIONS (d) Find the general solution of the differential equation y ′′ + 4y = 3 csc t. y2 ](t) (c) Integrate the two equations from part (2). (d) The general solution to the homogeneous equation y ′′ + 4y = 0 is yh = c1 cos 2t + c2 sin 2t. (b) If we solve the system u′1 (t)y1 (t) + u′2 (t)y2 (t) = 0 u′1 (t)y1′ (t) + u′2 (t)y2′ (t) = f (t). Hint. By part (2) u′1 (t) = −3 cos t 3 u′2 (t) = csc t − 3 sin t. then yp′ = u′1 y1 + u1 y1′ + u′2 y2 + u2 y2′ = u1 y1′ + u2 y2′ yp′′ = u′1 y1′ + u1 y1′′ + u′2 y2′ + u2 y2′′ . To find a particular solution. y2 ](t) y1 (t)f (t) u′2 (t) = . (a) If yp = u1 y1 + u2 y2 . assume that the solution has the form yp = u1 (t) cos 2t + u2 (t) sin 2t.212 CHAPTER 4.

SINUSOIDAL FORCING 213 Therefore. the alternative is to become an expert in trigonometric identities . functions that are periodic are especially important. The period for each of these two functions is 2π/ω and the frequency is ω/2π.3.3. ∫ T 1 g(t) dt = 0.4. That is. yp (t) = u1 (t)y1 (t) + u2 (t)y2 (t) [ ] 3 = −3 sin t cos 2t + ln | csc t − cot t| + 3 cos t sin 2t. These two functions share the additional property that their average value is zero. eiβt = cos βt + i sin βt to derive a particular solution. we can use Euler’s formula.7 Project 4. 4. The most familiar periodic functions are g(t) = sin ωt and g(t) = cos ωt. 2 4.1 Complexification Given a second-order linear differential equation ax′′ + bx′ + cx = A cos ωt + B sin ωt. That is. we will assume that our particular solution has the form xc = xRe + ixIm and use the properties of complex numbers.3 Sinusoidal Forcing If we consider different forcing functions g(t) for the equation x′′ + px′ + qx = g(t). Recall that a function g(t) is periodic if g(t + T ) = g(t) for all t and some fixed constant T . T 0 We say that sinusoidal forcing occurs in the differential equation x′′ + px′ + qx = A cos ωt + B sin ωt. 2 and the general solution is y = yh + yp [ ] 3 = c1 cos 2t + c2 sin 2t − 3 sin t cos 2t + ln | csc t − cot t| + 3 cos t sin 2t.2.1 1 If complex numbers make you uncomfortable.

the general solution is 12 1 x = xh + xp = c1 e−5t + c2 e−t − cos 2t + sin 2t. we will obtain a particular solution 12 1 xp = − cos 2t + sin 2t. x′′ + 6x′ + 5x = 0.17. if we can find a complex solution. we can use the method of undetermined coefficients and assume that the solution has the form xp = A cos 2t + B sin 2t. Now let us assume that our solution has the form xc = Ae2it . Equating the real and imaginary parts of this equation. SECOND-ORDER LINEAR EQUATIONS Example 4. is xh = c1 e−5t + c2 e−t . we can find a solution to x′′ + 6x′ + 5x = sin 2t simply by examining the imaginary part of the solution. (4.214 CHAPTER 4.14) will approach the particular solution as t → ∞. Let us consider the equation x′′ + 6x′ + 5x = sin 2t. we obtain x′′Re + 6x′Re + 5xRe = cos 2t x′′Im + 6x′Im + 5xIm = sin 2t . If we carry out the appropriate calculations.14) using complex numbers. Then x′′c + 6x′c + 5xc = −4Ae2it + 12Aie2it + 5Ae2it = (1 + 12i)Ae2it = (1 + 12i)A(cos 2t + i sin 2t).18. we obtain x′′Re + 6x′Re + 5xRe = cos 2t x′′Im + 6x′Im + 5xIm = sin 2t.14) The solution to the corresponding homogeneous equation. To find a particular solution. Example 4. 145 145 Notice that all solutions of (4. then d2 d d2 d 2 xc + 6 x c + 5x c = 2 (xRe + ixIm ) + 6 (xRe + ixIm ) + 5(xRe + ixIm ) dt dt dt dt = e2it = cos 2t + i sin 2t. Now let us solve (4. If we assume that the equation has a complex solution of the form xc = xRe + ixIm . Thus. Equating the real and imaginary parts of this equation. 145 145 Thus.

Let us rewrite A in polar form. The amplitude of our solution is 1/ 145 and the period is π (Figure 4. . 145 145 2 145 where ϕ ≈ 3.4877.3. Since 1 |A| = √ .19.4. the complex solution to x′′ + 6x′ + 5x = e2it is xc = Ae2it ( ) 1 12 = − i · (cos 2t + i sin 2t) 145 145 ( ) ( ) 1 12 12 1 = cos 2t + sin 2t + i − cos 2t + sin 2t . 1 iθ 2it 1 i(2t+θ) xc = Ae2it = √ e e =√ e . Thus. 145 145 Our particular solution is the imaginary part of xc . 1 1 ( π) 1 xp (t) = √ sin(2t + θ) = √ cos 2t + θ − =√ cos (2t − ϕ) .20). Example 4. √ We say that ϕ is the phase angle of our solution.17. 145 145 145 145 The imaginary part of this function is 12 1 xIm = − cos 2t + sin 2t. 145 we know that 1 iθ A= √ e . Therefore. 145 where θ = arctan(−12) ≈ −1. 145 145 which is the particular solution that we have been seeking.2 Qualitative Analysis We can use the complex solution of ax′′ + bx′ + cx = A cos ωt + B sin ωt to analyze the qualitative behavior of solutions. We discovered that the complex solution of x′′ + 6x′ + 5x = e2it to be xc = Ae2it . SINUSOIDAL FORCING 215 and we immediately see that 1 1 12 A= = − i.3. where A = (1 − 12i)/145. our general solution agrees with what we found in Example 4. 4. 1 + 12i 145 145 Therefore.058451.

the direction field changes with time. Example 4. Since the direction field changes with time.15). Consequently.216 CHAPTER 4.15) The solution to the homogeneous equation x′′ + 2x′ + 17x = 0 is xh = c1 e−t cos 4t + c2 e−t sin 4t The complex version of this equation is x′′ + 2x′ + 17x = −2e3it . Consider the harmonic oscillator that is modeled by the differential equa- tion x′′ + 2x′ + 17x = −2 sin 3t. In other words. solu- tions can cross each other in the xy-plane without violating the Existence and Uniqueness Theorem. is x′ = y y ′ = −qx − py + g(t). SECOND-ORDER LINEAR EQUATIONS x(t) 0. 05 t π 2π 3π 4π 5π −0. but also on the time t. This is a nonautonomous system. and we will use the Method of Undetermined Coefficients and assume that we can find a particular solution of the form xc = Ae3it . Thus.21. 05 Figure 4. y) value at different times can follow different paths. we find that (8 + 6i)Ae3it = −2e3it . Substituting xc into equation (4. (4. y).20: Steady state solution to x′′ + 6x′ + 5x = sin 2t The corresponding first order system for the differential equation x′′ + px′ + qx = g(t). xc is a solution if −2 4 3 A= =− + i 8 + 6i 25 25 . and the tangent vector of a solution curve in the phase plane depends not only on the position (x. two solutions with the same (x.

25 25 Thus.15) is 3 4 x(t) = c1 e−t cos 4t + c2 e−t sin 4t + cos 3t − sin 3t.4. Notice how the solution curve can intersect itself.3. we must solve the linear system 3 c1 + =0 25 9 −4c1 − c2 − = 0. respectively. We can quickly determine that 12 9 x′ (t) = c1 e−t (4 cos 4t − sin 4t) + c2 e−t (− cos 4t − 4 sin 4t) − cos 3t − sin 3t 25 25 To solve this initial value problem. 25 25 25 25 The imaginary part of this function is the solution that we seek. x′ (0) = 0 Since y = x′ (t). x(t) 0.22: Solution to x′′ + 2x′ + 17x = −2 sin 3t. 1 π 2π 3π t −0.22.23). 2 Figure 4. 2 0. 40 80 40 20 The graph of our solution is given in Figure 4. 25 25 Now suppose that x(0) = 0 and x′ (0) = 0. the . x(0) = 0. The restoring force and damping are proportional to x and y = x′ . 3 4 xp = cos 3t − sin 3t. When x and y are close to the origin. the general solution to (4. 1 −0. we can now graph the solution curve in the phase plane (Figure 4. 25 We obtain c1 = −3/25 and c2 = 3/25. and the solution to our initial value problem is 3 −2t 3 3 3 x(t) = − e cos 4t + e−2t sin 4t + cos 2t − sin 2t. SINUSOIDAL FORCING 217 We have ( ) 4 3 xc (t) = − + i (cos 3t + i sin 3t) 25 25 ( ) ( ) 4 3 3 4 = − cos 3t − sin 3t + i cos 3t − sin 3t .

2 -0.218 CHAPTER 4.23: Phase Plane for x′′ + 2x′ + 17x = −2 sin 3t. Thus.2 -0.4 0. we can solve the linear system 3 c1 + =2 25 9 −4c1 − c2 − = 2. the external force overcomes the damping and pushes the solution away from the origin.24. suppose we have initial conditions x(0) = 2 and x′ (0) = 2.2 x -0. 25 to obtain c1 = 47/25 and c2 = 109/100. 25 100 40 20 The graph of our solution is given in Figure 4.2 -0. y 0.4 -0. x′ (0) = 0 On the other hand. SECOND-ORDER LINEAR EQUATIONS external force is as large or larger than the restoring and damping forces.1 0.6 Figure 4.6 0. x(0) = 0.1 0. In this part of the xy-plane. solution to our initial value problem is 47 −t 109 −t 3 3 x(t) = e cos 4t + e sin 4t + cos 2t − sin 2t. .

SINUSOIDAL FORCING 219 x(t) 2.25).25: Phase Plane for x′′ + 2x′ + 17x = −2 sin 3t. 5 π 2π 3π t −0. x′ (0) = 2 4. • We can use Euler’s formula and complexification to solve the equation x′′ + px′ + qx = g(t). 0 0. 0 1. x(0) = 2. the solutions in the xy-plane tend to spiral towards the origin and are similar to the solutions of the unforced equation. x′ (0) = 2 If we examine the phase plane for this solution (Figure 4.3. . These average value of each of these functions is zero. 0 Figure 4.3 Important Lessons • The functions sin ωt and g(t) = cos ωt are periodic with period 2π/ω and frequency ω/2π. y 2 -1 -0. 5 1. we see that the initial damping and restoring forces are much larger than the external force.5 0.4.5 1 1.5 2 x -2 -4 -6 Figure 4.3. Thus.24: Solution to x′′ + 2x′ + 17x = −2 sin 3t. x(0) = 2. if we are far from the origin. 5 −1.

Therefore.youtube.4 Exercises 1. • If we write the equation x′′ + px′ + qx = g(t). 4.4 Forcing and Resonance Resonance happens when the natural frequency and the forcing frequency of an undamped harmonic oscillator are the same. While a certain amount of damping occurs in all physical systems. 4. and ϕ is the phase angle. In this case the direction field changes with time. Furthermore. the external force is as large or larger than the restoring and damping forces. For example. the external force overcomes the damping and pushes the solution away from the origin. it is possible for an trained singer shatter a champaign glass by simply singing the correct note (see https://www. solutions can cross each other without violating the Existence and Uniqueness Theorem. There are many real world examples of resonance. x′ = y y ′ = −qx − py + g(t). Shattering a glass by singing the correct note is can be modeled by an equation of an undamped harmonic os- cillator with sinusoidal forcing. Compute the solution of the initial value problem y ′′ + 4y = 3 cos 2t y(0) = y ′ (0) = 0.3. • If we are far from the origin. the damping is often so small that it makes sense to use an undamped harmonic oscillator. and two solutions with the same (x. SECOND-ORDER LINEAR EQUATIONS where the forcing function g(t) is sin ωt or cos ωt. we can use complex numbers to express or solution in the form x(t) = A cos(ωt − ϕ). In this part of the xy-plane. Hint. .5 Project 4. the solutions in the xy-plane tend to spiral towards the origin and are similar to the solutions of the unforced equation. where A is the amplitude of the solution. When x and y are close to the origin.com/watch?v=IZD8ffPwXRo). y) value at different times can follow different paths. we obtain a nonautonomous system.220 CHAPTER 4.3. ω/2π is the frequency of the solution. as a first-order system.

4.4. The collapse occurred when a column of soldiers marched in cadence over the bridge. When the bridge began oscillating. a video of tourists crossing the bridge revealed the opposite. This pedestrian bridge. which opened to the public in June 2000. However. The engineers did not envision this to be a problem since tourists do not generally march in time. FORCING AND RESONANCE 221 4. if people were to walk in cadence. The Millennium Bridge. the first new bridge to span the Thames River in London in over 100 years. Studies by designers found that the bridge experienced high amplitude horizontal oscillations in response to horizontal forcing at a rate of one cycle per second. The frequency of the force was approximately equal to the natural frequency of the bridge.youtube. Typically. England in 1831. Thus. they would could set up strong horizontal forcing that would place a destructive load on the bridge. the bridge collapsed when large oscillations occurred. For this reason soldiers are ordered to break cadence whenever they cross a bridge. Figure 4.26). setting up a periodic force of rather large amplitude. is a modern example of how resonance can effect a bridge (Figure 4. people walk at a rate of two steps per second.26: The Millennium Bridge Consider the case of an undamped harmonic oscillator with a sinusoidal forcing function. Thus.4.1 Resonance Resonance was responsible for the collapse of the Broughton suspension bridge near Manch- ester. so the time between two successive steps of the left foot is about one second. people tended to walk in cadence in order to keep their balance (https://www.com/watch?v=gQK21572oSU). . was quickly closed after the bridge experienced high amplitude horizontal oscillations during periods of high traffic.

17). we cannot assume that our solution has the form aeiω0 t . we can see that the natural frequency of the harmonic oscillator is ω0 .17) In this case. we must have A a= . Substituting xc and x′′c into the left-hand side of (4. Equivalently. Therefore. 2iω0 Thus. If the two frequencies are close but not equal. we will have two very different situations.16) Since A cos ω0 t is a solution to the homogeneous equation x′′ + ω02 x = 0. we will see the phenomenon of beats. (4. Depending on whether or not the frequency of the forcing term ω is equal to the natural frequency. then they will re-enforce one another and we will have resonance.16) has the form a cos ω0 t + b sin ω0 t. If the two frequencies are equal. In order for xc = ateiω0 t to be a solution to (4. x′′ + ω02 x = A cos ω0 t. SECOND-ORDER LINEAR EQUATIONS We know that we can model this situation with the second-order linear differential equation x′′ + ω02 x = A cos ωt Since the homogeneous solution to this equation is xh (t) = c1 cos ω0 t + c2 sin ω0 t. x′c = a(1 + iω0 t)eiω0 t x′′c = a(2iω0 − ω02 t)eiω0 t . we cannot assume that a particular solution to equation (4. we have x′′c + ω02 xc = a(2iω0 − ω02 t)eiω0 t + ω02 ateiω0 t = 2aiω0 eiωt .222 CHAPTER 4. our solution to the complex form of the differential equation is A xc = teiω0 t 2iω0 Ai iω0 t =− te 2ω0 Ai =− t(cos ω0 t + i sin ω0 t) 2ω0 A A = t sin ω0 t − i t cos ω0 t. We wish to investigate what happens when the driving frequency ω is equal to the natural frequency and when it is not. 2ω0 2ω0 .16). Let us examine the case where the forcing frequency and the natural frequency of the oscillator are the same. (4. we will look for a complex solution of the form xc = ateiω0 t for the complex equation that corresponds to (4.17). if we use the complex method. x′′ + ω02 x = Aeiω0 t . This is the frequency of the oscillations if there is no forcing term.

4 1 a= = − i. we have a particular solution 1 xp (t) = t sin 12t. we have x′′c + 144xc = a(24i − 144t)e12it + 144ate12it = 24aie12it . the general solution to x′′ + 144x = 4 cos 12t is 1 x(t) = xh (t) + xp (t) = c1 cos 12t + c2 sin 12t + t sin 12t. We must assume that the solution has the form xc (t) = ate12it . 2ω0 Example 4. FORCING AND RESONANCE 223 The real part of xc . since ae12it is a solution to the homogeneous equation. 6 Applying the initial conditions. 6 6 6 6 Taking the real part of our complex solution.27. If we substitute xc and x′′c into the left-hand side of our differential equation. This type of behavior is called resonance. 24i 6 and our complex solution is 1 1 1 1 − ie12it = − i(cos 12t + i sin 12t) = sin 12t − i cos 12t. where A = 4 and ω0 = 12. Consequently.28. 6 Thus. Now let us consider the initial-value problem x′′ + 144x = 4 cos 12t x(0) = 0 x′ (0) = 0.16). the amplitude increases with time. we have x′′c = a(24i − 144t)e12it . To find a particular solution to x′′ + 144x = 4 cos 12t.4. As before. A xp (t) = t sin ω0 t. This growth is due to the fact that the frequency of the forcing term is equal to the natural frequency of the oscillator. we will use the complex method and try to find a particular solution to x′′ + 144x = 4e12it . The solution to the homogeneous equation x′′ + 144x = 0 is xh (t) = c1 cos 12t + c2 sin 12t. the solution to the initial-value problem is 1 x(t) = t sin 12t. 6 The graph of this solution is given in Figure 4. our general solution is A x(t) = xh (t) + xp (t) = c1 cos ω0 t + c2 sin ω0 t + t sin ω0 t. both c1 = 0 and c2 = 0.4. 2ω0 is a particular solution to (4. Since the force pulls and pushes at a frequency equal to the natural frequency of the oscillator. . Thus. Notice that our solution grows with time. Therefore.

The tuner will strike a tuning fork which vi- brates at the correct frequency. we find x′′c + ω02 xc = −aω 2 eiωt + ω02 aeiωt = a(ω02 − ω 2 )eiωt .com/watch?v=pRpN9uLiouI. 0 Figure 4. watch the video https://www.224 CHAPTER 4.19) If we assume out solution has the form xc = aeiωt and substitute xc into the left-hand side of (4. a piano tuner uses this phenomenon to make certain that a particular string is correctly tuned. we will first consider the equation. 0 1. we must have A a= . For example. the beats will go away. 0 −2. ω ̸= ω0 . (4. for xc = aeiωt to be a solution for x′′ + ω02 x = Aeiωt . x′′ + ω02 = Aeiωt . Therefore.28: The solution to x′′ + 4x = 4 sin 2t. 4. To see another example of beats. then we have the phenomenon of beats. we can look for a particular solution to the complex of the form of equation (4. ω02 − ω2 . Next. SECOND-ORDER LINEAR EQUATIONS x(t) 2. x′ (0) = 0. This modulation gives rise to beats in the tone that are readily audible.youtube. When the string is properly tuned.2 Beats or the Case ω ̸= ω0 If the forcing frequency and natural frequency are not equal but close. (4.19). the tuner hits the poorly tuned piano string.18). To understand the situation where the forcing and natural frequencies differ by a small amount.18) The solution to the homogeneous equation x′′ + ω02 = 0 is xh (t) = c1 cos ω0 t + c2 sin ω0 t.4. x′′ + ω02 = A cos ωt. 0 t π 2π 3π 4π −1. x(0) = 0. Since the forcing frequency is not equal to the natural frequency.

23) ′ x (0) = 0.21) 2ωδ Example 4.18). ω02 − ω2 Thus.29. To understand how the superposition of the two frequencies works. sin ωt whose amplitude varies according to |(A/2ωδ) sin δt|. we have a superposition to two oscillations of different frequencies.24) . and A sin δt x(t) = sin ωt.18) is A x(t) = xh (t) + xp (t) = c1 cos ω0 t + c2 sin ω0 t + cos ωt. the general solution to (4. Hence the solution to our initial value problem is A x(t) = (cos ωt − cos ω0 t). In the initial-value problem x′′ + 144x = 4 cos(13t) (4. ω0 = ω + δ. we have a sine function.4. we can easily determine that c1 = −A/(ω02 − ω 2 ) and c2 = 0.22) x(0) = 0 (4. In this case. (4. (4. ω02 − ω2 Now let us examine the case where the motion starts at equilibrium. (4.20) ω02 − ω2 Thus. That is. 2ωδ Thus. we will let ω0 + ω ω= 2 be the mean frequency and ω0 − ω δ= 2 be the half difference. FORCING AND RESONANCE 225 The real part of xc is a particular solution to (4. we know that A x(t) = (cos ωt − cos ω0 t) ω02− ω2 A = (cos[(ω − δ)t] − cos[(ω + δ)t]) (ω + δ) − (ω − δ)2 2 A = [cos(ωt − δt) − cos(ωt + δt)] 4ωδ A = [cos(ωt) cos(δt) + sin(ωt) sin(δt) − cos(ωt) cos(δt) + sin(ωt) sin(δt)] 4ωδ A = [2 sin(ωt) sin(δt)] 4ωδ A sin δt = sin ωt.4. A xp (t) = cos ωt. we will investigate what happens when x(0) = 0 and x′ (0) = 0. Since ω = ω − δ.

we can view the solution (4.226 CHAPTER 4. SECOND-ORDER LINEAR EQUATIONS we have A = 4. The factor sin δt = sin(−t/2) in (4. 1 −0. δ = −1/2 and ω = 25/2. 1 t π 2π 3π 4π 5π 6π 7π 8π 9π 10 π −0. 3 0.30: Solution to the initial-value problem (4.21) as a fast oscillation with a frequency ω and amplitude .21) oscillates very slowly in comparison to sin(ω) = sin(25t/2). 2 −0. and ω = 13. Thus. 2 0.22) In our example. 3 Figure 4. 25 The graph of this solution is given in Figure 4. The solution to this equation is 4 x(t) = (cos 12t − cos 13t). x(t) 0. ω0 = 12. The two frequencies are almost equal and interfere with each other.30 and illustrate the phenomenon of beats.

.

.

( ).

.

A sin δt .

.

8 .

.

.

= .

sin t .

. (4.25) .

2ωδ .

.

25 2 .

x′′ + 2cx′ + ω02 x = A cos ωt.30. which oscillates much more slowly.3 Forced Damped Harmonic Motion Since perpetual motion does not really exist.4. we should consider what happens when we add some damping to our harmonic oscillator. 3 0. 1 −0.26) . x(t) 0.25) onto the graph in Figure 4. (4. 2 −0. If we superimpose the graph of (4. 2 0. we obtain an envelope of the faster oscillation.31: Envelope for the initial-value problem (4. 3 Figure 4.22) 4. 1 x(t) |(A/92πωδ)sinδt10 t π 2π 3π 4π 5π 6π 7π 8π | π −0.

130 It is easy to see that all solutions approach xp (t) as t → ∞ no matter what the initial conditions may be.4. it would be much more useful if we would write the particular solution in the form xp (t) = K cos(4t − ϕ) for some constants K and ϕ rather than a linear combination of cos 4t and sin 4t. (4.29) The characteristic polynomial of (4. 130 The general solution for (4. 3 xp (t) = (−7 cos 4t + 4 sin 4t). we will look for a solution of the form xc = ae4it to the equation x′′ + 2x′ + 2x = 3e4it (4.29). 3 3 3 a= = =− (7 + 4i). Substituting xc into the lefthand side of (4. (4.28). we can determine a particular solution by taking the real part of xc . To find a particular solution for the equation (4. which has solution xh (t) = c1 e−t cos t + c2 e−t sin t. Therefore.28) and then set xp (t) to the real part of the solution. FORCING AND RESONANCE 227 Example 4. Returning to the general case. the associated homogeneous equation for (4.32. P (λ) = λ2 + 2cλ + ω02 . Let us consider the differential equation x′′ + 2x′ + 2x = 3 cos 4t. we obtain x′′c + 2x′c + 2xc = (−14 + 8i)ae4it = P (4i)ae4it = 3e4it .26) is x′′ + 2cx′ + ω02 x = 0. P (4i) −14 + 8i 130 and the solution to the complex equation is 3 xc = − (7 + 4i)e4it 130 3 =− (7 + 4i)(cos 4t + i sin 4t) 130 3 = [(−7 cos 4t + 4 sin 4t) + i(−4 cos 4t − 7 sin 4t)].27) is 3 x(t) = xh (t) + xp (t) = c1 e−t cos t + c2 e−t sin t + (−7 cos 4t + 4 sin 4t). where P (λ) = λ2 + 2λ + 2 is the characteristic polynomial of x′′ + 2x′ + 2x = 0. However. 130 Thus.4.27).27) The corresponding homogeneous equation is x′′ + 2x′ + 2x = 0. (4.30) .

32. Equivalently. let us write P (iω) = (iω)2 + 2c(iω) + ω02 = (ω02 − ω 2 ) + 2icω.31) and then set xp equal to the real part of our solution. we will look for a particular solution to (4. 2cω Therefore. P (iω) R .31). where P is the characteristic polynomial (4. we can write the transfer function as 1 1 H(iω) = = e−iϕ .30). equation (4. and A iωt xc (t) = aeiωt = e = H(iω)Aeiωt . (4. then the solution to the homogeneous equation (4. Substituting xc into the right-hand side of (4. we know that sin ϕ > 0. we shall look for a solution of the form xc = aeiωt to the equation x′′c + 2cx′c + ω02 xc = Aeiωt . SECOND-ORDER LINEAR EQUATIONS has roots √ λ = −c ± c2 − ω02 . 0 < ϕ < π. As in Example 4.26) by examining the equivalent complex differential equation. Thus. where √ R= (ω02 − ω 2 )2 + 4c2 ω 2 and ϕ is the angle defined by the equations ω02 − ω 2 cos ϕ = √ . (ω02 − ω 2 )2 − 4c2 ω 2 2cω sin ϕ = √ . P (iω) We say that H(λ) = 1/P (λ) is the transfer function. P (iω) where 1 H(iω) = . ( 2 ) −1 ω0 − ω 2 ϕ = ϕ(ω) = cot . √ where β = ω02 − c2 . we obtain x′′c + 2cx′c + ω02 xc = [(iω)2 + 2c(iω) + ω02 ]aeiωt = P (iω)aeiωt . in polar form. If our harmonic oscillator is undedamped (c < ω0 ).228 CHAPTER 4. Thus.31) becomes P (iω)aeiωt = Aeiωt . P (iω) = Reiϕ = R(cos ϕ + i sin ϕ).29) is xh = e−ct (c1 cos βt + c2 sin βt). That is. (ω02 − ω 2 )2 − 4c2 ω 2 Since 2cω > 0. Let us examine the transfer function more closely to see if we can use it to write our particular solution in a more useable form. First.

Let us examine the steady-state solution of the harmonic oscillator in Example 4. Since xh has the factor e−ct . our particular solution is xp (t) = Re(xc (t)) = G(ω)A cos(ωt − ϕ). R (ω0 − ω 2 )2 + 4c2 ω 2 and we will rewrite the transfer function as H(iω) = G(ω)e−iϕ(ω) .2481. As we can see.32.4142 If we choose the initial conditions so that c1 = 0 and c2 = 1.4. It is now clear that particular solution has the same frequency as the forcing term. The phase is ϕ = ϕ(4) ≈ 2.0620. and A = 3.4. xp is out of phase with the driving force by the amount ( ) −1 ω02 − ω 2 ϕ = ϕ(ω) = cot . We have already determined that the second-order linear differential equation x′′ + 2x′ + 2x = 3 cos 4t has solution 3 x(t) = xh (t) + xp (t) = c1 e−t cos t + c2 e−t sin t + (−7 cos 4t + 4 sin 4t). Example 4. 130 √ The natural frequency is ω0 = 2 ≈ 1.32). our transient term is xh (t) = e−ct sin βt = e−t sin t. Thus. The solution and the steady-state solution are given in Figure 4. ω = 4. For this reason. the amplitude of the steady state solution is G(4) · 4 ≈ 0. (4. the gain is G(4) ≈ 0. the homogeneous part of the solution quickly decays to zero as t → ∞.32) Taking the real part of (4.33. FORCING AND RESONANCE 229 We define the gain to be 1 1 G(ω) = =√ 2 . transient solutions can be quite large when compared to steady-state solutions. the solution to x′′c + 2cx′c + ω02 xc = Aeiωt is xc (t) = H(iω)Aeiωt = G(ω)Aei(ωt−ϕ) . xh is called the transient term while xp is called the steady-state term. Since c = 1. .6224.34. 2cω The general solution to x′′ + 2cx′ + ω02 x = A cos ωt is x(t) = xh (t) + xp (t) = e−ct (c1 cos(ηt) + c2 sin(ηt)) + G(ω)A cos(ωt − ϕ). In addition.

computers should be left on except in the case that they are not being used for a period of several days. 4 0. For example. where the phase angle ϕ is ( ) ω02 − ω 2 ϕ = ϕ(ω) = cot−1 2cω and the gain G is given by 1 G(ω) = √ . a light bulb usually burns out when a large transient flows through an already weakened bulb. A particular solution is given by xp (t) = Re(xc (t)) = G(ω)A cos(ωt − ϕ).4. xp (t) = G(ω)A cos(ωt − ϕ). usually occurs at the time the light bulb is turned out or turned off. For this reason. Large transient currents are particularly harmful to the hard drive in a computer. 3 0.4 Important Lessons • A forced damped harmonic oscillator. (1 − s2 )2 + D2 s2 where ω = sω0 and D = 2c/ω0 . .230 CHAPTER 4. 1 t π 2π 3π −0. 2 0. SECOND-ORDER LINEAR EQUATIONS x(t) 0. This. 2 Figure 4. Large transient currents can be destructive in electrical circuits. √ where η = ω02 − c2 . 4. x′′ + 2cx′ + ω02 x = A cos ωt has homogeneous solution xh = e−ct (c1 cos ηt + c2 sin ηt). 5 xh + xp xp 0. 1 −0. (ω02 − ω 2 )2 + 4c2 ω 2 • The amplitude and phase of the steady-state solution.34: A forced damped harmonic oscillator and the steady-state solution. are determined by 1 ω02 G = √ .

4. FORCING AND RESONANCE 231 • In an undamped harmonic oscillators with a sinusoidal forcing term. x′′ + 2cx′ + ω02 x = A cos ωt has homogeneous solution xh = e−ct (c1 cos ηt + c2 sin ηt). determine the following. √ where η = ω02 − c2 . Hint. • If the driving frequency is not equal to the natural frequency (ω ̸= ω0 ). we can observe a phenomenon called resonance.4. where the phase angle ϕ is ( ) −1 ω02 − ω 2 ϕ = ϕ(ω) = cot 2cω and the gain G is given by 1 G(ω) = √ . (b) The frequency of the rapid oscillations. As the fre- quency of the forcing term approaches the natural frequency of the equation. we can see a phenomenon called beats. we can see the resonance phenomenon. .4.5 Exercises 1. • If the driving frequency is equal to the natural frequency. we have a particular solution A xp (t) = t sin ω0 t 2ω0 In this case. We can use the equation x′′ + ω02 x = A cos ωt to model an undamped harmonic oscillator with sinusoidal forcing. ω0 − ω 2 If the two frequencies are close. (c) Use parts (1) and (2) to sketch a graph of a typical solution. (a) The frequency of the beats. • A forced damped harmonic oscillator. A particular solution is given by xp (t) = Re(xc (t)) = G(ω)A cos(ωt − ϕ). the natural fre- quency of the solution interacts with the frequency of the forcing term. we have a particular solution A xp (t) = 2 (cos ωt − cos ω0 t). (ω02 − ω 2 )2 + 4c2 ω 2 4. Given the equation y ′′ + 10y = sin 3t.

(c) Set q = 1 and plot M (p. p. q). where 1 G(ω) = √ . The natural frequency is fixed in x′′ + 2cx′ + ω02 x = A cos ωt and D = 2c/ω0 is proportional to the damping constant. q) = √ . Will it be harder or easier to break the glass when it is half full of water? (c) Commercials for a maker of recording tape show a crystal glass being broken by the sound of a recording of an opera singer’s voice. SECOND-ORDER LINEAR EQUATIONS 2. p. q) denote the maximum value of A(ω. ∂A (a) Compute . q) as a function of ω. and q.232 CHAPTER 4. These new constants. (q − ω 2 )2 + p2 ω 2 That is. Compute an expression for M (p. let M (p. (d) Explain why M (p. 1 G = 2√ ω0 (1 − s2 )2 + D2 s2 or 1 ω02 G = √ . For large t. (1 − s )2 + D2 s2 2 This expression shows us how the gain varies as s = ω/ω0 varies. Will the singer have to sing a higher note or a lower note to break an identical glass that is half full of water? (b) Suppose that both notes are within the opera singer’s range. (ω02 − ω 2 )2 + 4c2 ω 2 Now let s = ω/ω0 and D = 2c/ω0 . every solution of y ′′ + py ′ + qy = cos ωt oscillates with angular frequency ω and amplitude A given by 1 A(ω. p. Hint. measure the ratio of the driving frequency to the natural frequency and the effect of the damping force. Let us examine the amplitude and phase of the steady-state solution. Thus. ∂ω (b) For fixed p and q. the amplitude of A is a function of the parameters ω. 4. . Hint. q) as a function of p. q) is proportional to 1/p as p → 0. respectively. (a) Suppose that an opera singer can break a glass by singing a particular note. s and D. 3. Is this a good test of the quality of the tape? Why or why not? Hint. xp (t) = G(ω)A cos(ωt − ϕ).

A. Figure 4. The bridge failed castrophically at 11:10 a. On the morning of November 7. A.6 Project—The Tacoma Narrows Bridge The History of the Tacoma Narrows Bridge On July 1.4.m. these vertical oscillations were as large as three feet. the bridge began undulating up and down with motions as large as three feet. Later that morning. For a long time. “Galloping Gertie”.35: The Tacoma Narrows Bridge Construction on first Tacoma Narrows Bridge. 1940.36). the bridge experienced vertical oscillations of several feet. people driving bridge experienced vertical oscillations of several feet. On the morning of November 7. the bridge began to experience torsional oscillations as great as 23 degrees. FORCING AND RESONANCE 233 4. was com- pleted on July 1. we shall soon see the the story is more complicated. the bridge began to experience torsional oscillations as large as 23 degrees. with the center span falling into Puget Sound below.4. From the first day of operation.35). . Even on the first day of operation. The bridge received the nickname “Galloping Gertie” for its wild behavior. Later.4. 1940. However. The bridge finally came crashing down at 11:10 a.m. the Tacoma Narrows Bridge was completed and opened to traffic (Fig- ure 4.K. the collapse of the Tacoma Narrows Bridge was attributed to resonance. the bridge was opened to traffic (Figure 4.

Imagine a mass suspended by both a spring and a rubber band.234 CHAPTER 4. “If you build the exact same bridge exactly as before.3 2 The Jet Propulsion Laboratory. If the collapse of the bridge was due to resonance. The rubber band acts like a spring when it is stretched. 3 For the arguments for and against resonance. However. exactly as before. During sentencing at his trial. and one’s first guess as to the reason for the collapse of the bridge would be resonance. One of the insurance policies had been written by a local travel agent who pocketed the premium and failed to report the $800. The road bed of the suspension bridge was hung from vertical cables attached to other cables strung between the two towers. but there is no restoring force if the oscillator is in a compressed position (Figure 4. he pointed out that bridge officials had planned to cancel all policies in one more week after which his crime would have never been discovered.2 sent a telegram to the governor stating.” Theodore Von Kármán.000 policy to his company. the forcing frequency of a forced harmonic oscillator must be close to its natural frequency.” The Design of the Narrows Bridge The Tacoma Narrows bridge is a suspension bridge like the Golden Gate bridge. consult the references at the end of this section. After the collapse of the Tacoma Narrows bridge.37).36: The Tacoma Narrows Bridge Video There are many interesting stories connected with the first Tacoma Narrows bridge. it will fall into the exact same river exactly as before. the governor of the State of Washington announced. If we think of the cables as long springs. director of the Guggenheim Aeronautical Laboratory at the California Institute of Technology and a member of the board of inquiry into the collapse. it is tempting to model the oscillations of the road bed with an harmonic oscillator. One explanation might be the fact that cables are not true springs. SECOND-ORDER LINEAR EQUATIONS Figure 4. . but the wind (the forcing) simply does not behave like this. “We are going to build the exact same bridge. the answer may not be so simple. They act more like rubber bands. managed by the California Institute of Technology for the National Aeronautics and Space Administration grew out of the Guggenheim Aeronautical Laboratory.

all of the solutions eventually approach the steady-state solution.4. Even though the transient part of the solution can differ a great deal depending on the initial conditions. (4. 0).34) Our first observation is that this equation is a nonlinear due to the y + term.37: A spring-rubber band system The Spring-Rubber Band Model Suppose that y(t) is vertical displacement of the mass the spring-rubber band model (Fig- ure 4. In Figure 4. k1 is the spring constant. Thus.33) in the more familiar form my ′′ + by ′ + k1 y + k2 y + = F (t). We can rewrite equation (4. FORCING AND RESONANCE 235 spring rubber band m Figure 4.33) where y + = max(y. b is the damping coefficient.37). if the rubber band is stretched. (4. and y ′ (0) = 3. and F (t) is the forcing term. m is the mass. y ′ (0) = 1. This may effect how sensitive the system is to initial conditions. y + (t) = 0.4. but has no effect on the system if the rubber band is slack. Here. Of course. the rubber band is stretched. A linear harmonic oscillator is not sensitive to initial conditions. suppose we consider the harmonic oscillator y ′′ + 2y ′ + 2y = sin t and examine the behavior of the solutions for different initial conditions. We will also assume that we have a damped system. For example. where the displacement is positive in the downward direction and negative in the upward direction. the term my ′′ is just an application of Newton’s Second law of Motion. we give the solutions for y(0) = 0 with y ′ (0) = −4.38. we can model the motion of the mass with the differential equation my ′′ = −by ′ − k1 y − k2 y + + F (t). the rubber band acts as a spring with spring coefficient k2 . . The term k2 y + corresponds to the rubber band in our system and can be explained as follows: { y(t). the rubber band is slack or tight but not stretched. If this is the case.

This is a nonautonomous (and nonlinear) system and can be very sensitive to initial con- ditions.38: Solutions of y ′′ + 2y ′ + 2y = sin t Let us return to our spring-rubber band model and let us assume that m = 1. To see how this works in practice. Or in the case of the Tacoma Narrows Bridge. If we find a numerical solution to this system with y(0) = 0 and v(0) = 4. . k2 = 4. a large initial velocity such as a gust of wind on the bridge. where { k1 y if y < 0 h(y) = (k1 + k2 )y if y ≥ 0. there will always be tension on the cables of the suspension bridge.39). SECOND-ORDER LINEAR EQUATIONS Figure 4. If we keep λ small and assume that both y(0) and y ′ (0) are small. µ = 4. we only have to change the initial velocity of the system slightly to v(0) = 4.01.764. However.34) as a system of first-order equations. the system becomes y′ = v v ′ = −h(y) − 0. we choose constants λ = 0.236 CHAPTER 4. We can rewrite equation (4.1.765 in order to obtain large amplitude solutions that do not die out over time. However. and b = 0. we obtain low amplitude solutions (Figure 4. y′ = v v ′ = −k1 y − k2 y + − bv + F (t) or x′ = v v ′ = −h(y) − bv + F (t). might cause the cables to go slack.01v + 10 + 4 sin 0.1t where { 13x if x < 0 h(y) = 17x if x ≥ 0. there will always be tension on the rubber band. k1 = 13. Let us assume that the forcing term is F (t) = 10 + λ sin µt.

• βy is the force provided by the material of the bridge pulling the bridge back to y = 0.4. • αy ′ is the damping term. Just think of how a flag flutters in a wind. . • F (t) is the periodic force provided by the wind.4. where • y is the height of the road bed at the center of the bridge. We assume that α is small since suspension bridges are relatively flexible structures.765 The Bridge Model We can now model the bridge by my ′′ + αy ′ + βy + γy + = −gm + F (t). • γy + is the force provided by the pull of the cables. Yes. • m is the mass of the bridge. • −gm is the force provided by gravity.39: Initial conditions y(0) = 0 with y ′ (0) = 4. FORCING AND RESONANCE 237 Figure 4.764 and y ′ (0) = 4. wind does provide a periodic force.

we have 3k θ′′ = −δθ′ + cos θ[(y − l sin θ)+ − (y + l sin θ)+ ] + F (t) ml k y ′′ = −δy ′ − [(y − l sin θ)+ + (y + l sin θ)+ ] + g. m If we assume that the cables never lose tension. 1940. Adding forcing and damping terms. The extension of one spring is given by (y −l sin θ)+ and (y +l sin θ)+ in the other. we see the road bed suspended from two cables—one on each side of the road. we obtain the equations 1 2 ′′ ml θ = kl cos θ[(y − l sin θ)+ − (y + l sin θ)+ ] 3 y ′′ = −k[(y − l sin θ)+ + (y + l sin θ)+ ] + mg Since the force exerted by the springs (cables) is not perpendicular to the rod but at an angle θ. 2 dt 6 dt Let L = T − V and put ( ) d δL δL = dt δθ′ δθ and ( ) d δL δL = dt δy ′ δy where δθ′ and δy ′ are small damping terms. 6 dt The potential due to gravity is −mgy. SECOND-ORDER LINEAR EQUATIONS The Model for Torsion We have still not accounted for the large torsional oscillations that occurred on November 7. If a spring with spring constant k is extended by a distance y. cos θ must appear in the first equation. this is a reasonable assumption. Thus. the potential energy is ky 2 /2. If we look down the bridge (Figure 4. The bar is free to move vertically and rotate about its center of mass. Thus. we can view a cross-section of the bridge as a rod suspended by cables or springs on each end. where + y = max(y. . then (y − l sin θ)+ = y − l sin θ 4 Consult your local physicist.36). then its kinetic energy is given by4 ( ) 1 2 dθ 2 ml . The total potential energy is given by ( )( )2 k ( )2 V = (y − l sin θ)+ + (y + l sin θ)+ − mgy. 0). Since the cables are relatively long compared to the length of the roadbed. We are making the assumption that the cables remain vertical. If a rod of mass m and length 2l rotates about its center of gravity with angular velocity dθ/dt.238 CHAPTER 4. 2 and the total kinetic energy is given by ( ) ( ) m dy 2 1 2 dθ 2 T = + ml . Let θ be the angle of the rod from the horizontal. the day that the Tacoma Narrows bridge collapsed.

41 Figure 4.40: Solution of the linear model for torsion . If the torsional oscillations are very small. The linear model θ′′ = −0.01θ′ − (2. FORCING AND RESONANCE 239 (y + l sin θ)+ = y + l sin θ and we end up with the uncoupled equations 6k θ′′ = −δθ′ − cos θ sin θ + F (t) m 2k y ′′ = −δy ′ − y + g.4.4) cos θ sin θ + (1. then we can assume that sin θ ≈ θ and cos θ ≈ 1. m The first equation models the torsional motion while the second models the vertical motion.2) sin(0.06t) behave very differently under the initial conditions θ(0) = 1.4)θ + (1.06t) and the nonlinear model θ′′ = −0.2) sin(0.2 and θ′ (0) = 0.4.40 and 4. See Fig- ures 4. However. or 6k θ′′ = −δθ′ − θ + F (t) m 2k y ′′ = −δy ′ − y + g. the torsional oscillations for the Narrow bridge were as large as 23 degrees.01θ′ − (2. m each of which is which is linear.

240 CHAPTER 4.41: Solution of the nonlinear model for torsion . SECOND-ORDER LINEAR EQUATIONS Figure 4.

we explored how to solve linear systems. 241 . for a given modeling problem. we have an equilibrium solution.1 Linearization In Chapter 3. we can ask many questions that may be answered without finding an explicit solution for the associated system of differential equations. y) = 0. y0 ) = g(x0 . we obtain a linear system. 5 Nonlinear Systems 5. y0 ) such that f (x0 . y) y ′ (t) = g(x. For example. Translating the pair of intersection lines to the origin. it may not be possible to find solutions for a nonlinear system in terms of elementary functions. what are the equilibrium points? Are the equilibrium points stable? Do closed solution curves exist in the phase plane? 5.1 Equilibrium Solutions One of the most useful methods of determining the nature of an equilibrium solution for a given nonlinear system is to approximate the nonlinear system with a linear system. However. y) dt dy = g(x. y) is a point (x0 . In general. when a curve defined by dx/dt = f (x. and we can apply everything that we learned about such systems in Chapter 3.1. However. y) dt when an x-nullcline intersects a y-nullcline. y) = 0 intersects a curve defined dy/dt = g(x. if we are asked to solve a nonlinear system such as dx = x − 3y + xy 2 dt dy = 2x − 4y − x2 y. An equilibrium solution for a first-order system of differential equations x′ (t) = f (x. an equilibrium solution occurs for the linear system dx = f (x. That is. y0 ) = 0. More specifically. dt we are faced with a much more difficult task. Since the x and y-nullclines are simply curves in the xy-plane. we can approximate them locally by intersecting straight lines.

when a curve defined by dx/dt = f (x.2. If we have the initial conditions x(0) = 1 and y(0) = 1. Consider the system dx = x − 3y + xy 2 dt dy = 2x − 4y − x2 y. y) = 0 intersects a curve defined dy/dt = g(x. y) = 0. we obtain a linear system. 0). Since the x and y-nullclines are simply curves in the xy-plane. NONLINEAR SYSTEMS Notice that neither x nor y is changing at this point. we can see that the solution tends toward the origin as t → ∞. If we have the initial conditions. Of course the interesting question is what happens if our initial conditions are close to an equilibrium solution. an equilibrium solution occurs for the linear system dx = f (x.242 CHAPTER 5. However. y) dt dy = g(x. x(0) = x0 and y(0) = y0 . The phase portrait for this system is given in Figure 5. dt we can quickly conclude that the only equilibrium solution to the system is (0. then the solution to the system is x(t) = x0 and y(t) = y0 . it is unclear from the phase portrait if the solution curves of all initial value problems with initial conditions near the origin tend towards the equilibrium solution as t → ∞. Example 5.1. That is. we can approximate them locally by intersecting straight lines. More specifically. or is there a combination of the two? One of the most useful methods of determining the nature of an equilibrium solution for a given nonlinear system is to approximate the nonlinear system with a linear system. dt From dx = x − 3y + xy 2 = 0 dt dy = 2x − 4y − x2 y = 0. Translating the pair of intersection lines to the origin. Do solutions tend toward the equilibrium solution. . y) dt when an x-nullcline intersects a y-nullcline. and we can apply everything that we learned about such systems in Chapter 3. away from the equilibrium solution. we have an equilibrium solution.

2 that we can model two competing species using the following system of first-order differential equations. For example. 2) and v = (1.3 (A Competing Species Model). Of course. Suppose that x and y are the population of two distinct species that compete for the same resources.1. ( ) 1 −3 A= . This indeed suggests that solutions near the origin tend towards the origin as t → ∞.2: Phase Portrait for x′ = x − 3y + xy 2 and 2x − 4y − x2 y Since the nonlinear terms of the system do not contribute much towards dx/dt and dy/dt for values of x and y near zero. dt The matrix for this linear system. dx = x − 3y dt dy = 2x − 4y. Example 5. dx ( x) = αx 1 − − βxy dt M . The general solution for this linear system is x(t) = 3c1 e−t + c2 e−2t y(t) = 2c1 e−t + c2 e−2t . we say that the equilibrium solution is stable. 2 −4 has eigenvalues λ = −1 and µ = −2 with eigenvectors u = (3.5. Recall from Section 2. In this case. if we are given an initial condition such as x(0) = −0. respectively.5 and y(0) = 1. 1). we can determining the nature of the equilibrium solution by examining the system consisting of only linear terms on the right-hand side of the equation. two species of fish may compete for the same food in a lake or sheep and cattle competing for the same grazing land. LINEARIZATION 243 2 1 y(t) 0 -1 -2 -2 -1 0 1 2 x(t) Figure 5.

1). If we decide on an appropriate change of variables. then du d dx = (x − 1) = . (5. (0. 0). The second term in each equation tells how each species is affected by interacting with the competing species. let us consider the following system. we can translate the entire system so that this equilibrium solution is at the origin. 3). 4 3 y(t) 2 1 0 0 1 2 3 4 x(t) Figure 5. dt N The first term in each equation is the logistic growth of each species.3) dt . 0).1) and (5.4: A Competing Species Model Let us analyze what happens at the equilibrium solution (1. dt dt dt Equations (5.244 CHAPTER 5.4. More specifically. 1). and (1. dx ( x) = 2x 1 − − xy (5.1) dt 2 dy ( y) = 3y 1 − − 2xy. (5. and some solution curves for this system in Figure 5. NONLINEAR SYSTEMS dy ( y) = γy 1 − − δxy.2) now become du = −u − v − u2 − uv. dt dt dt dv d dy = (y − 1) = .2) dt 3 It is easy to see that the four equilibrium solutions: (0. (2. If we let u=x−1 v = y − 1. the phase plane. We can view the direction field.

(5.4) are du = −u − v.4142t .1213e−2. and equations (5.10) As t → ∞ notice that both x(t) and y(t) become very large and tend away from the origin. LINEARIZATION 245 dv = −2u − v − 2uv − v 2 .4142t (5. We can determine the local nature of the equilibrium solution by examining the eigenvalues of the matrix ( ) −1 −1 A= .0859e−2.4142t + 0.8) become x(t) = 2. (5. 1) is unstable.4142 with eigenvectors u = (1. one can quickly determine that c1 = 3/2 + 2 ≈ 2. ( ) dy x3 x5 = −y − sin x = −y − x − + − ··· . We can expand sin x into a power series.4) and (5.9142e0.4142). (5. The solution to the linear system is now ( ) ( ) λt √1 µt √1 x(t) = c1 e + c2 e . dx =y dt dy = −y − sin x. 1.4) dt As before we consider only the linear part of equations (5.6) dt The idea is that the linear part is a good local approximation to the original equations much like a tangent line is a good local approximation to a smooth function in calculus. That is. dt Certainly. 2 ) ≈ (1. −1. dt . If systems such as the following can be ap- proximated by linear systems.1.5. 0). 2 − 2 or x(t) = c1 eλt + c2 eµt (5. this system can be approximated by the linear system dx =y dt dy = −y − x.4142t − 0. then the solution curve will approach the equilibrium solution as t → ∞. respectively. −2 −1 √ √ The eigenvalues of A √ are λ = −1 + 2 = 0.4142) and v = (1.4142 √ and µ = −1 − 2 ≈ −2. We can conclude that the equilibrium solution (1.7) √ √ y(t) = c1 2eλt − c2 2eµt . this sytems has an equilibrium solution at (0.9) y(t) = 4. − 2 ) ≈ (1. (5.8) If we have √ initial conditions x(0) =√3 and y(0) = 4. (5.9142 and c2 = 3/2 − 2 ≈ 0.7) and (5.0859.1213e0.5 (A Nonpolynomial Example).5) dt dv = −2u − v. tend away from the equilibrium solution as t → ∞ if one population begins with a slight advantage over the other. If neither popluation has an initial advantage over the other. Example 5. dt 3! 5! Thus.

(5.8.7: Phase portrait of x′ = y and y ′ = −y − x Example 5.246 CHAPTER 5.11) dt dy = x3 − y.6: Phase portrait of x′ = y and y ′ = −y − sin x 4 2 y(t) 0 -2 -4 -4 -2 0 2 4 x(t) Figure 5. For example. NONLINEAR SYSTEMS 4 2 y(t) 0 -2 -4 -4 -2 0 2 4 x(t) Figure 5. consider the system dx = x2 − 4x − y + 4 (5.12) dt .

9: Phase portrait for the system (5. (5.14) dt Our new system has u and v-nullclines v = u(u − 2) and v = u(u2 + 3u + 3). all solution curves starting sufficiently close to (1. 1) is a stable equilibrium solution. Furthermore. LINEARIZATION 247 The x and y-nullclines of this system are the curves y = x2 −4x+4 and y = x3 .11)–(5.5. respectively. 1) will approach the equilibrium solution as t → ∞ (Figure 5. . it appears that (1.9). 3 2 y(t) 1 0 -1 -1 0 1 2 3 x(t) Figure 5.1. Notice that we have simply moved the phase portrait of the original system so that our equilibrium solution is now at the origin.13) dt dv u(u2 + 3u + 3) − v. respectively. From Figure 5. we can approximate the u and v-nullclines by their tangent lines v = −2u and v = 3u.9. it appears that we are approximating our original system with a linear system. respectively. and we obtain the system du = u(u − 2) − v (5. simply translates the entire system to the origin. 1). Since the two nullclines intersect only at (1. we have a single equilibrium solution. That is.12) Making the substitution u = x − 1 and v = y − 1. From the phase plane.

y0 )v 1 1 + fxx (x0 .10: Phase portrait for the system (5. y0 )v 2 + · · · . f and its linear approximation . y0 )uv + fyy (x0 . du d dx = (x − x0 ) = = f (x. dv/dt = dy/dt = g(x0 + u. In order to approximate the nonlinear system with a linear system. y0 )u2 + fxy (x0 . y0 + v). In order to move the equilibrium solution to the origin. y) = f (x0 + u. That is.14) To determine the general case. Under this change of coordinates. y0 + v). y0 )u + fy (x0 . then (u. dt dt dt Similarly. we will introduce new variables u = x − x0 v = y − y0 .16) near each equilibrium point (x0 . NONLINEAR SYSTEMS 2 1 v(t) 0 -1 -2 -2 -1 0 1 2 u(t) Figure 5. y) (5. The idea is to move the equilibrium solution to the origin with a change of coordinates and then approximate the nonlinear system with a linear system.15) ′ y (t) = g(x. y) (5. we will use a Taylor series in two variables. y0 ).248 CHAPTER 5. y0 + v) = f (x0 . we can write f (x0 + u. v) will be close to the origin. Of course.13)–(5. If (x. y0 ) + fx (x0 . we obtain a fairly accurate approximation of f provided we are near the equilibrium solution. we must approximate a nonlinear system x′ (t) = f (x. y) is close to the equilibrium solution (x0 . y0 ) with a linear system. 2 2 If we only use the linear terms of the Taylor series.

15)–(5. y0 ) v The matrix ( ) fx (x0 .16) now becomes du = f (x0 . y0 ) fy (x0 . we have ( ) ( )( ) du/dt fx (x0 . y0 ) + fx (x0 . 1). y0 ) is the Jacobian matrix of the system. A solution curve might behave quite differently if it is far away from the equilibrium solution. y0 ) fy (x0 . 1). ( ) −2 −1 J= . y0 )v. y0 ) J= gx (x0 . y0 )u + gy (x0 . y0 ) gy (x0 . y0 ) 2x0 − 4 −1 J= = gx (x0 . . We can determine the sink to be spiral or nodal by examining whether (T. y0 ) is an equilibrium solution. if D = det(J) > 0 and T = tr(J) < 0. y0 )v dt dv = g(x0 . dt The Jacobian matrix of this system is ( ) ( ) fx (x0 .11. dt Since (x0 . dt If we write our linearization in matrix form. y0 )v. y0 )u + gy (x0 . y0 ) fy (x0 . y0 )u + fy (x0 . y0 ) + gx (x0 . y0 ) gy (x0 .8. y0 ) u = . y0 )v dt dv = gx (x0 . the constant terms vanish in each equation and du = fx (x0 . The linearization of the system of equations (5. 3 −1 √ Since J has eigenvalues λ = (−3 ± i 11)/2.5. y0 ) gy (x0 . dv/dt gx (x0 . y0 )u + fy (x0 . It is important to note that linearization only tells us the local story. we considered the system dx = x2 − 4x − y + 4 dt dy = x3 − y. LINEARIZATION 249 may be quite different for values far away from the equilibrium solution.1. For example. the equilibrium solution will act as a spiral sink for initial values close to (1. y0 ) 3x20 −1 For the equilibrium solution (1. D) lies above or below the parabola T 2 = 4D in the trace-determinant plane. we have a sink. In Example 5. We can now classify equilibrium solutions of nonlinear systems by examining the eigenvalues of the Jacobian matrix of the system or by using the trace-determinant plane. Example 5.

2 0.1 -0.12: When linearization fails .2 x(t) Figure 5. dt The origin is an equilibrium solution. ( ) 0 1 A= . the solution curves of the linear system are simply circles about the origin. 0. For example. However. but this linearization of this system vanishes.2 -0. it might well be the case that the linear terms vanish in the nonlinear system.12). solutions spiral out slowly from the origin (Figure 5.2 -0.1 0. This system has no periodic solutions. Thus. First. −1 0 are λ = ±i.250 CHAPTER 5. NONLINEAR SYSTEMS 5.2 When Linearization Fails There are at least two cases when linearization does not give us the information that we seek. and the linearization of this system is dx =y dt dy = −x. In the nonlinear system.1 0 0. the system dx = xy dt dy = −x2 + xy dt has an equilibrium solution at the origin.1.1 y(t) 0 -0. A more subtle example is the system dx = y − (x2 + y 2 )x dt dy = −x − (x2 + y 2 )y. the nonlinear system acts quite differently. dt The eigenvalue of the matrix corresponding to the linear system.

4 Exercises 1. 0)? Justify your answer. gx (x0 . y0 ) fy (x0 . y0 ) is the Jacobian matrix of the system. y0 ) gy (x0 . We make the following additional assump- tions about our model.1. y) y ′ (t) = g(x. A solution curve might behave quite differently if it is far away from the equilibrium solution. ( ) fx (x0 . y0 ) J= . • Linearization only tells us how solutions behave near the equilibrium point.1. Which two systems have phase portraits with the same “local picture” near (0. . he or she is immune. Let us consider an epidemic model for a city. If an individual recovers. LINEARIZATION 251 5. y) near each equilibrium point (x0 .3 Important Lessons • We can approximate a nonlinear system x′ (t) = f (x. All new born babies are susceptible to the disease. • The city has a constant birth rate rate of α persons per day. linearization can fail. Consider the following three systems (a) x′ = 3 sin x + y y ′ = 4x + cos y − 1 (b) x′ = −3 sin x + y y ′ = 4x + cos y − 1 (c) x′ = −3 sin x + y y ′ = 4x + 3 cos y − 3 All three systems have an equilibrium solution at (0. We can classify nonlinear systems by exam- ining the Jacobian matrix of the system and using the trace-determinant plane. 5.1. The matrix of our linear approximation. 2. • In some cases. • Infected people either recover or die after a certain number of days.5. Hint. 0). y0 ) with a linear system by using a Taylor series approximation for f and g.

γ We are interested in the behavior of solutions near (S0 . If dS = −αSI + β = 0 dt dI = −γI + αSI = 0.1. We will now consider the system dx = x + y − x3 dt dy = −0. This will occur at γ S0 = α β I0 = . The x-component of the direction field . The system itself was very difficult to analyze due to its highly nonlinear nature. then the disease will become endemic to the population.5x. where a and b are constants satisfying a certain condition and I is a parameter. Similar models have been constructed to model the systolic and diastolic movements of the heart. If solutions approach this equilbrium point. dt The constant α is determined by the probability of an interaction between a susceptible individual and an infected individual. x represents the excitability of the system and y represents a combination of other forces that return the system to rest. The x-nullcline for this system is y = −x + x3 . dt then both the susceptible and infected populations do not change. but it did earn the two scientists a Nobel Prize for medicine in 1963 (the year after Watson and Crick won their prize). I0 ). NONLINEAR SYSTEMS If we let S(t) be the number of susceptible individuals at time t and I(t) be the number of infected individuals at time t. In these equations. our assumptions give rise to the following system of differential equations. dS = −αSI + β dt dI = −γI + αSI.5 Project—Modeling the Nervous System Hodgkin and Huxley developed a model for the firing of nerve cells in the 1950s. Hint. Fitzhugh and Nagumo managed to simplify the Hodgkin-Huxley model to two equations x′ = y + x − x3 /3 + I y ′ = −x + a − by. and γ is the rate at which infected individuals are removed from the population.252 CHAPTER 5. The model was a system of four differential equations that described the electrochemical transmission of neuronal signals along cell membranes in giant squid. 5. dt which has been used to study nerve cells and is similar to the Fitzhugh-Nagumo model.

To the right of the y-nullcline. Thus. The y-nullcline of the system is the y-axis.13). 2 1 y(t) 0 -1 -2 -2 -1 0 1 2 x(t) Figure 5. we have dy/dt < 0.5x. the x and y-nullclines divide the plane into four regions and all solutions must circulate in a clockwise manner about the origin (Figure 5. HAMILTONIAN SYSTEMS 253 above this nullcline is positive.2 Hamiltonian Systems An undamped harmonic oscillator.13: Modeling nerve cells The origin is the only equilibrium solution of the system. we can conclude that the origin is a spiral source.5. m . however.2. Below the nullcline the x-component of the direction field is negative. hence. and dy/dt > 0 to the left of the y-nullcline. A solution that passes close to the origin tends to spiral outward. can be written as the system y′ = v k v′ = − y. solutions that are far from the origin spiral inward. 5. dt to determine the nature of the equilibrium solution. We shall later see that we must have a closed periodic solution to this system by the Poincaré-Bendixson Theorem. we can examine the linearization dx =x+y dt dy = −0. Since the eigenvalues of this linear system are (1 ± i)/2. my ′′ + ky = 0.

dv/dy. dy mv we obtain the solution 1 1 mv 2 + ky 2 = C. Léon Foucault. 0 2 For this reason.254 CHAPTER 5.14) to demonstrate that the earth actually rotated on its axis. used a pendulum (Figure 5. y) dt dy = g(x.17) tells us that energy is conserved. NONLINEAR SYSTEMS Now suppose that (y(t). That is.17) is the kinetic energy function of the harmonic oscillator 1 K = mv 2 . the French physicist. v(t)) is a solution curve in the yv-plane. (5. y) Using separation of variables to solve the equation dv ky =− . dx f (x. y).1 The Nonlinear Pendulum While pendulums have long been used in clocks to keep time. dy dt dy dy/dt mv In general for the system dx = f (x. We will calculate the slope of the solution curve.17) 2 2 where C is a constant. . but Foucault’s experiment give the first simple proof of this phenomena. Using the fact from calculus that the derivative of an inverse function is d −1 1 f (x) = ′ . 2 while the second term is the potential energy function ∫ y 1 U= ks ds = ky 2 . the sum of the potential energy and the kinetic energy is constant. The first term of (5. we call the function 1 1 E = K + U = mv 2 + ky 2 2 2 the total energy function of the harmonic oscillator. they have also been used to measure gravity as well as used in early seismometers to measure the effect of earthquakes. In 1851. dt we have dy g(x. Equation (5.2. y) = . 5. dx f (x) we have dv dv dt dv/dt ky = · = =− . One of the more interesting uses of a pendulum has been to measure the rotation of the earth. The fact that the earth rotates had been known for a long time.

dt2 m dt L . Newton’s second law tells us that d2 θ dθ mL = −bL − mg sin θ dt2 dt or d2 θ b dθ g + + sin θ = 0.16) The component of the acceleration that points along the direction of motion of the bob is d2 θ L .15). dt2 We can take the force due to friction to be proportional to the velocity. Thus. −L cos θ(t)).2. We will assume that θ(0) is when the bob is in the vertical position. dt where b ≥ 0. which we will measure in a counterclockwise direction. θ L m Figure 5. HAMILTONIAN SYSTEMS 255 Figure 5. the length of the velocity vector (Figure 5.15: The nonlinear pendulum There are two forces acting on the pendulum—gravity and friction.5. and the velocity of the bob is L(dθ/dt).14: Foucault Pendulum Clock—California Academy of Sciences Let us consider a pendulum made of a light rod of length L called the arm of the pendulum with a mass mon the end of the rod called the bob (Figure 5. The position of the bob is given by θ(t). We will ignore the mass of the arm in our system. dθ −bL . The position of the bob at time t is (L sin θ(t).

2. More specifically.16: Forces on the nonlinear pendulum If there is no damping for our pendulum. Example 5. dθ Lv we know that the energy function for the pendulum is 1 g E(θ. dt L Since dv g sin θ =− . 5. we say that we have an ideal pendulum.2 Hamiltonian Systems The ideal pendulum and the undamped harmonic oscillator are examples of Hamiltonian systems. y) dt ∂y dy ∂H =− (x. dt m L θ L mg sin θ mg Figure 5. a Hamiltonian system is a system of the form dx ∂H = (x. In this case.256 CHAPTER 5. dθ =v dt dv g = − sin θ. y). v) is constant on the ideal pendulum. dt ∂x where H : R2 → R is a smooth function. v) = v 2 − cos θ. 2 L Of course. E(θ.17. NONLINEAR SYSTEMS As a system. The system dx = −2x − 3y 2 dt dy = −3x2 + 2y dt . we can model the pendulum as dθ =v dt dv b g = − v − sin θ.

5.19. y) dt ∂y dy ∂H =− (x. y(t)) be a solution curve for the system. Theorem 5. ∂x Example 5. the Hamiltonian function is 1 1 E = K + U = mv 2 + ky 2 . . dt ∂x be a Hamiltonian system. ∂y A change of variables (y = my) tells us that this system is equivalent to the system y′ = v k v′ = − y. v) = v 2 − cos θ. 2 2 In this case. y) = x3 − 2xy − y 3 ∂H = −2x − 3y 2 ∂y ∂H = −(−3x2 + 2y). Let dx ∂H = (x. Let (x(t). y). That is.2. Then H is constant along any solution curve. H is just the energy function. m For the ideal pendulum ∂E dθ = =v ∂v dt ∂E dv g − = = − sin θ. In the case of the harmonic oscillator and ideal pendulum. 2 L The following theorem tells the the importance of Hamiltonian systems.18. For the harmonic oscillator. the solution curves of the system are simply the level sets of the Hamiltonian function. Using the chain rule dH ∂H dx ∂H dy = + dt ∂x dt ∂y dt ( ) ( ) ∂H ∂H ∂H ∂H = + − ∂x ∂y ∂y ∂x = 0. ∂θ dt L where 1 g E(θ. where H : R2 → R. Proof. HAMILTONIAN SYSTEMS 257 is Hamiltonian since for H(x. ∂E = y ′ = mv ∂v ∂E − = v ′ = −ky.

the equilibrium points of the Hamiltonian system occur at the critical points of H (where the partials of H vanish). v) = v 2 − cos θ. since the total energy function 1 g H(θ.2. this is quite easy since we know the vector field of the system.5 -1.5 1 1. In Figure 5. For example.5 x(t) Figure 5.20. dt L is Hamiltonian. If we set g/L = 1. while cosine curves correspond to a pendulum that . ∂x in Figure 5. Assuming that the Hamiltonian function H is not constant on any open set in R2 . v) = E(θ. we simply need to plot the level curves.18) 2 L is the Hamiltonian function for the system. However.5 -1 -1.258 CHAPTER 5. then the solution curves of the system are just the level curves of (5. 1.18). we can see the solution curves of ∂H = −2x − 3y 2 ∂y ∂H = −(−3x2 + 2y). H(x. (5. The solutions of the system live on these level sets. the closed ellipses correspond to the normal motion of a pendulum.5 1 0.5 -1 -0.20: Solution curves of x′ = x − 3y 2 and y ′ = −y 5.5 0 0. NONLINEAR SYSTEMS Theorem 5. y) = C.21. and all we need to do is find the direction of the solution curve.3 The Ideal Pendulum Revisited The ideal pendulum dθ =v dt dv g = − sin θ.19 tells us how to draw the solution curves in the phase plane without solving the system.5 y(t) 0 -0. Furthermore.

y) ∂y ∂H = −g(x. Example 5. 4 3 2 1 v(t) 0 −1 −2 −3 −4 −4 π −3 π −2 π −π 0 π 2π 3π 4π θ(t) Figure 5. y). y) such that ∂H = f (x. HAMILTONIAN SYSTEMS 259 always rotates in the same direction. y) dt dy = g(x. The curves that join the equilibrium points correspond to the pendulum that rotates exactly to the top of the arc and then rotates back in the other direction. dt we wish to find a function H(x.21: The ideal pendulum Hamiltonian systems are rather rare. ∂x ∂y .22.5. we need a test to see if it is indeed Hamiltonian. ∂x ∂x∂y ∂y∂x ∂y Thus. We already know that the system dx = −2x − 3y 2 dt dy = −3x2 + 2y dt is Hamiltonian. y). Indeed. ∂x If such a function exists.2. ∂f ∂g =− . Given a system. ∂x ∂y if our system is Hamiltonian. then ∂f ∂2H ∂2H ∂g = = =− . For the system dx = f (x. ∂f ∂g =− = −2.

y) dt dy = g(x. y). y) dy + ϕ(x). y) dy − g(x.19) is true on a rectangle R. y) = (x. If ∂H f (x. y) = f (x. NONLINEAR SYSTEMS The condition that ∂f ∂g =− . y) dy + ϕ(x). we will show how we constructed the Hamiltonian function given in Example 5. y) dt dy = g(x. 1. ∂x Example 5. y). then the system is Hamiltonian on R and ∫ H(x. y) − f (x.260 CHAPTER 5. ∂x ∂y is also sufficient for dx = f (x. y) = f (x. then ∂f ∂g =− .17 for the system dx = −2x − 3y 2 dt . ∂x ∂x Now we can determine ϕ(x) by solving the first-order differential equation ∫ ∂ ϕ′ (x) = −g(x. If equation (5. y) dy + ϕ′ (x) = −g(x. y). y) = f (x. y). As an example. y). Theorem 5. dt to be a Hamiltonian system. ∂y then ∫ H(x. If the system is Hamiltonian. ∂x We summarize our results in the following theorem. dt where f and g are continuously differentiable. Differentiating H with respect to x. (5. where (∫ ) ∂ ϕ′ (x) = − f (x.19) ∂x ∂y 2. we can construct a Hamiltonian function if ∂f /∂x = −∂g/∂y. y) dy. That is. Let dx = f (x.23.24. tells us that ∫ ∂H ∂ (x.

If we choose C = 0. dt Since ∂f ∂g = x2 cos x cos y − 2x cos y sin x = − ∂x ∂y our system is Hamiltonian. ∫ ∫ H(x. ∫ ∂ ϕ′ (x) = −g(x. Consequently. As a second example. y) dy + ϕ(x) = (−2x − 3y 2 ) dy + ϕ(x) = −2xy − y 3 + ϕ(x). y) = f (x. y) = −2xy − y 3 + ϕ(x) = −2xy − y 3 + x3 . Letting C = 0. To find a Hamiltonian. .25.26. and ϕ(x) = x3 + C. our Hamiltonian function is H(x. ∫ ′ ∂ ϕ (x) = −g(x. y) − f (x. y) dy ∂x = −(x2 cos x sin y + 2x(sin x sin y − 1)) − (2x sin x sin y − x2 cos x sin y) = 2x. our Hamiltonian function is H(x. where C is any constant. suppose dx = −x2 cos y sin x dt dy = x2 cos x sin y + 2x(sin x sin y − 1). Thus. HAMILTONIAN SYSTEMS 261 dy = −3x2 + 2y dt First. y) = f (x. and ϕ(x) = x2 + C. Solution curves for this system are given in Figure 5. Example 5. y) − f (x. y) = x2 − x2 sin x sin y = x2 (1 − sin x sin y).2. y) dy + ϕ(x) = (−x2 cos y sin x) dy + ϕ(x) = −x2 sin x sin y + ϕ(x). y) dy ∂x ∂ = −(−3x2 + 2y) − (−2xy − y 3 ) ∂x = 3x2 − 2y + 2y = 3x2 .5. we compute ∫ ∫ H(x.

17.2. 2/3). Since solution curves must follow contours for the Hamiltonian function H(x.262 CHAPTER 5. y0 ) Hyx (x0 . we might guess that (0. y0 ) If we let α = Hxy (x0 . y0 ). y0 ) . Referring to the phase portrait for the system (Figure 5. dx = −2x − 3y 2 dt dy = −3x2 + 2y dt has equilibrium solutions at (0. we will compute the Jacobian matrix of the system at (x0 . y) = − (x. y) = x3 − 2xy − y 3 . y0 ) fy (x0 . y0 ) Hyy (x0 . y0 ) is an equilibrium solution for the system dx ∂H = f (x. 0) is a saddle and (−2/3. y). spiral sinks do not make sense. gx (x0 .25 5. 0) and (−2/3. y0 ) −Hxy (x0 .26: Solution curves for Example 5. y0 ) = Hyx (x0 . y0 ) J= = . ( ) ( ) fx (x0 . NONLINEAR SYSTEMS 6 4 2 y(t) 0 -2 -4 -6 -6 -4 -2 0 2 4 6 x(t) Figure 5. dt ∂x In order to determine the nature of the equilibrium solution. Let us examine the possible types of equilibrium solutions for a Hamiltonian system.20). y0 ) gy (x0 .2/3) is either a center or a spiral sink. y) = (x. y0 ) β = Hyy (x0 . y) dt ∂y dy ∂H = g(x. Suppose that (x0 .4 Equilibrium Solutions of Hamiltonian Systems The system that we studied in Example 5. y0 ) −Hxx (x0 .

2/3) is ( ) −2 −4 J= . a Hamiltonian system has no spiral sinks or sources. .2. dt m dt L or the system dθ =v dt dv b g = − v − sin θ. dt ∂x where H : R2 → R2 is a smooth function. γ −α The characteristic polynomial of this matrix is λ2 − α2 − βγ. we have eigenvalues λ = ±2. our equi- librium solution is a saddle. solution curves near (−2/3. Returning to Example 5.5. 5. our matrix has eigenvalues √ λ = ± α2 + βγ. In particular. then J becomes ( ) α β . 0). The Jacobian matrix corresponding to (x0 . Therefore. we have two purely imaginary eigenvalues. • If α2 + βγ < 0.= . y0 ). −Hxx (x0 . • If α2 + βγ = 0.2. and we have the following possibilities. the Jacobian matrix is ( ) ( ) Hyx (x0 . we have two real eigenvalues of opposite signs. dt m L • A Hamiltonian system is a system of the form dx ∂H = (x. y0 ) Hyy (x0 .5 Important Lessons • A pendulum can be modeled by the equation d2 θ b dθ g 2 + − sin θ = 0. • If α2 + βγ > 0. y0 ) −6x0 2 For the equilibrium solution (0. −4 2 √ Since J has eigenvalues λ = ±2 5 i. which tells us that the origin is a nodal saddle.17. y0 ) −Hxy (x0 . y0 ) −2 −6y0 J= . y) dt ∂y dy ∂H =− (x. 2/3) should look like closed orbits. HAMILTONIAN SYSTEMS 263 γ = −Hxx (x0 . y0 ) = (−2/3. y). Therefore. the only eigenvalue is zero. Therefore.

NONLINEAR SYSTEMS • If dx ∂H = (x.2.3 More Nonlinear Mechanics The equation for the nonlinear pendulum with damping is d2 θ b dθ g 2 + + sin θ = 0. the solution curves of a Hamiltonian system are the level sets of H.21) dt m L . ∂x • A Hamiltonian system has no spiral sinks or sources. ∂x ∂y • If fx = −gy is true on a rectangle R. y). (5. 5. then ∂f ∂g =− . y). then H is constant along any solution curve. y) = f (x.6 Projects Possible project 5. y) dt dy = g(x. then the system dx = f (x. y) dy + ϕ(x). y). dt m dt L As a system. y) dt dy = g(x. y). y) dy − g(x.20) dt dv b g = − v − sin θ. • If the system dx = f (x. In particular.264 CHAPTER 5. dt is Hamiltonian on R and ∫ H(x. where (∫ ) ′ ∂ ϕ (x) = − f (x. y) dt ∂y dy ∂H =− (x. dt ∂x is a Hamiltonian system. we can model the pendulum as dθ =v (5. dt is Hamiltonian.

5.3. MORE NONLINEAR MECHANICS 265

This system is not Hamiltonian since
∂v
= 0,
∂θ
but ( )
∂ b g b
− − v − sin θ = ̸= 0.
∂v m L m
We shall have to invent new techniques to analyze such systems.

5.3.1 The Nonlinear Pendulum and Damping
To find the nullclines of the nonlinear pendulum, let

=v=0
dt
dv b g
= − v − sin θ = 0.
dt m L
The θ-nullcline is v = 0, while the v-nullcline is
mg
v=− sin θ.
bL
It follows that the equilibrium solutions for the nonlinear pendulum occur at
(θ, v) = (0, 0), (±π, 0), (±2π, 0), . . . .
This makes sense since the pendulum should not move if the bob is initially hanging down-
ward (θ = 2πn) or is at the very top or the very bottom of a swing (θ = (2n + 1)π). Since
our first goal is to determine the nature of each equilibrium solution, we will compute the
Jacobian of the system (5.20)–(5.21). This is just
( )
0 1
.
−(g/L) cos θ −b/m
At the equilibrium solutions (θ, v) = (0, 0), (±2π, 0), (±4π, 0), . . ., the pendulum is hanging
downward, and the Jacobian matrix becomes
( )
0 1
J1 = .
−g/L −b/m
On the other hand, if (θ, v) = (±π, 0), (±3π, 0), (±5π, 0), . . ., the pendulum is at the top of
its swing, and the Jacobian matrix is
( )
0 1
J2 = .
g/L −b/m
We will first examine the case where the pendulum is hanging downward. The charac-
teristic polynomial of J1 is
b g
λ2 + λ + .
m L
Thus, the eigenvalues of J1 are
√( )
b b 2 g
λ=− ± − . (5.22)
2m 2m L
We can analyze the nature of these eigenvalues by examining the sign of
( )
b 2 g
∆= − .
2m L

266 CHAPTER 5. NONLINEAR SYSTEMS

• If ∆ < 0, the eigenvalues of the Jacobian are complex. Since the real part of (5.22) is
negative, these equilibrium solutions are spiral sinks.

• If ∆ > 0, we have two distinct real eigenvalues. Since
( ) ( )
b 2 g b 2
− < ,
2m L 2m
we know that √(
( ) )2
b b g
> − .
2m 2m L
Thus, both of our eigenvalues must be negative. Therefore, we have a nodal sink.

• If ∆ = 0, we have a single real negative eigenvalue. Thus, we also have a sink.

Consequently, if we assume that b is small, then ∆ < 0 and we will only have spiral sinks
(Figure 5.27).

4
3
2
1
v(t)

0
−1

−2

−3

−4
−3 π −2 π −π 0 π 2π 3π
θ(t)
Figure 5.27: Nullclines of the damped pendulum

Now let us consider the type of equilibrium solutions that we will obtain when the pendu-
lum is standing upright. These solutions will occur at (θ, v) = (±π, 0), (±3π, 0), (±5π, 0), . . ..
The characteristic polynomial of the Jacobian matrix J2 at these points is
b g
λ2 + λ− ;
m L
hence, the eigenvalues of J2 are
√( )2
b b g
λ=− ± + . (5.23)
2m 2m L

Furthermore, we have distinct real eigenvalues since
( )
b 2 g
+ > 0.
2m L

5.3. MORE NONLINEAR MECHANICS 267

In fact, the eigenvalues will have opposite sign
√( )
b b 2 g
λ1 = − + + >0
2m 2m L
√( )
b b 2 g
λ2 = − − + < 0.
2m 2m L

Thus, the equilibrium solutions are saddles.

5.3.2 Lyapunov Functions
The function
1 g
H(θ, v) = v 2 − cos θ
2 L
is a Hamiltonian function for the ideal pendulum

=v
dt
dv g
= − sin θ,
dt L
since dH/dt = 0. However, if b > 0, then our system


=v
dt
dv b g
= − v − sin θ.
dt m L
has damping and
dH ∂H dθ ∂H dv
= +
dt ∂θ dt ∂v dt( )
(g ) b g
= sin θ v + v − v − sin θ
L m L
b 2
= − v ≤ 0.
m
Thus, H is decreasing whenever v ̸= 0. Hence solution curves in the θv-plane cross the level
sets of H moving from larger to smaller H values.
We can now devise a strategy for sketching the phase plane of the damped pendulum. If
b/m and v are both small, the value of H decreases slowly along the solutions (Figure 5.27).
The function H in the case of the damped pendulum is an example of a Lyapunov
function. Specifically, a function L(x, y) is called a Lyapunov function for the system

dx
= f (x, y)
dt
dy
= g(x, y)
dt
if for every solution of the system, (x(t), y(t)), that is not an equilibrium solution of the
system,
dL
≤ 0,
dt
with strict inequality except possible for a discrete set of ts.

268 CHAPTER 5. NONLINEAR SYSTEMS

As an example, let us return to the damped harmonic oscillator

dy
=v
dt
dv
= −qy − pv.
dt

If p = 0, then
1 q
H(y, v) = v 2 + y 2
2 2
is a Hamiltonian function for our system. Recall that we also call H the energy function of
the system. However, if p > 0 and (y(t), v(t)) is a solution for our system, we have

dH ∂H dy ∂H dy
(y, v) = +
dt ∂y dt ∂v dt
= (qy)v + v(−qy − pv)
= −pv 2 ≤ 0.

Consequently, H(y(t), v(t)) decreases at a nonzero rate (except when v = 0), and H is a
Lyapunov function. The level sets of H are ellipses in the yv-plane. As H decreases, the
energy dissipates and the ellipses become spiral sinks.
It is easy to see that the system

dx
= y + αx(x2 + y 2 ) (5.24)
dt
dy
= −x + αy(x2 + y 2 ) (5.25)
dt

has an equilibrium solution at the origin no matter what the value of α is. The Jacobian
of this system is
( )
3αx2 + αy 2 1 + 2αxy
J= .
−1 + 2αxy αx2 + 3αy 2

Since our equilibrium solution is the origin,
( )
0 1
J=
−1 0

and the linearization of our system at the origin is

dx
=y
dt
dy
= −x.
dt

Since the eigenvalues of
( )
0 1
J= ,
−1 0

are ±i, the linearization has a center at the origin. The phase plane consists of circles about
the origin (Figure 5.28). Notice that the linearization does not depend on α.

5.3. MORE NONLINEAR MECHANICS 269

3

2

1

y(t)
0

-1

-2

-3
-3 -2 -1 0 1 2 3
x(t)
Figure 5.28: Solution curves for the linearization.

Now let us consider what happens to system (5.24)–(5.25) if we consider different values
of α. If α = 5, the situation is quite different that the linearization of our system. A
solution curve spirals out from the origin as t → ∞ (Figure 5.29). As t → −∞, the solution
curve spirals back into the origin, but it seems to stop before actually reaching the origin. If
α = −5 on the other hand, we seem to have the opposite behavior with the solution curves
spiraling into the origin as t → ∞. As before, the solutions do not seem to reach the origin
(Figure 5.30).

270 CHAPTER 5. NONLINEAR SYSTEMS

3

2

1

y(t)
0

-1

-2

-3
-3 -2 -1 0 1 2 3
x(t)
Figure 5.29: Solution curves α = 5.

3

2

1
y(t)

0

-1

-2

-3
-3 -2 -1 0 1 2 3
x(t)
Figure 5.30: Solution curves for α = −5.

Suppose that (x(t), y(t)) is a solution to the nonlinear system. The function

r(t) = x(t)2 + y(t)2

is the distance of a point on the solution curve to the origin in the xy-plane. To see how r
changes as t → ±∞, we can compute the derivative of r. Actually, it is easier to work with

5.3. MORE NONLINEAR MECHANICS 271

the equation r(t)2 = x(t)2 + y(t)2 . Thus,
d 2 d
r = (x(t)2 + y(t)2 )
dt dt
dx dy
= 2x + 2y
dt dt
= 2x(y + αx(x2 + y 2 )) + 2y(−x + αy(x2 + y 2 ))
= 2α(x2 + y 2 )2
= 2αr4 .

Since (r2 )′ = 2rr′ , we have
r′ = αr3 (5.26)
for r ̸= 0.
Equation (5.26) is separable, and it is quite easy to determine the solution as
1
r(t) = √ .
C − 2αt
However, we do not need to know this solution to determine the nature of the equilibrium
solution at the origin. If α > 0 and t → −∞, equation (5.26) tells us that r(t) → 0. Thus,
any solution to the system (5.24)–(5.25) we have a spiral sink at the origin if α = −5. Even
though linearization fails to tell us the nature of the equilibrium solution at the origin, we
were able to determine the nature of the equilibrium solution with further analysis.
We will now try to exploit what we have learned from our last example and from
Hamiltonian systems to see if it is possible to analyze more general systems. If we consider
solutions, (x(t), y(t)), of the system
dx
= f (x, y)
dt
dy
= g(x, y),
dt
we might ask how a function V (x, y) varies along the solution curve. We already have an
answer if our system is Hamiltonian, and V is the corresponding Hamiltonian function. In
this case dV /dt = 0. In general, we know that
dV ∂V dx ∂V dy
(x(t), y(t)) = +
dt ∂x dt ∂y dt
∂V ∂V
= f (x, y) + g(x, y).
∂x ∂y
Thus, if we let
∂V ∂V
V̇ = f (x, y) + g(x, y),
∂x ∂y
we know that
dV
(x(t), y(t)) = V̇ (x(t), y(t)).
dt
Thus, V is increasing along a solution curve if V̇ (x, y) > 0 and decreasing along a solution
curve if V̇ (x, y) < 0. Our example suggests that we can determine this information without
finding the solution.
Recall that the gradient of a function V : R2 → R is
( )
∂V ∂V
∇V = , .
∂x ∂y

That is. y0 ) is in S. where the point x0 = (x0 . 611. any solution that starts near the equilibrium solution will stay near the equilibrium solution. Albert Boggess. Let V be a continuously differentiable function defined on a neighborhood U of (x0 . y) graphs as a surface in R3 and V (x. y) is going downhill. y0 ). then (x0 . and David Arnold. respectively. Thus. Prentice Hall. then (x0 . is the directional derivative of V in the direction of F. if V̇ (x. y0 ) is a asymptotically stable equilibrium solution or a sink. If V̇ is negative semidefinite on U . m m then the energy function. 2 2 is positive definite. Upper Saddle River. y0 ) that is positive definite with minimum at (x0 . y) dt has an equilibrium solution at (x0 . y) = . y) then V̇ = ∇V · F. 2001. 2. we know that the solution curve at (x.272 CHAPTER 5. Similarly. g(x. 1 1 E(y. y) dt dy = g(x. if V̇ (x. y0 ) is a stable equilibrium solution. Similarly. If V̇ is negative definite on U . the solution curve is traveling uphill. NONLINEAR SYSTEMS If ( ) f (x. y) in R3 is increasing. y) = constant gives the contour lines or level curves of the surface in the xy-plane. Suppose that the system dx = f (x. We say that V is positive definite if V (x) > 0 for all x in S. we know that V is increasing in the direction of the direction of the vector field F and the elevation of the solution curve through (x. y) F(x.31. Differential Equations. and V is positive semidefinite if V (x) ≥ 0. . Theorem 5. NJ. That is.2 Now suppose that V is a real-valued function defined on a set S in the xy-plane. 1 See Figures 1 and 2 in John Polking. where x ̸= x0 . y) > 0. y0 ). 2 The argument that we have made here also works in higher dimensions. 1. Let us use this new information about V to obtain information about equilibrium solu- tions of our system. v) = mv 2 + ky 2 . For example.1 We also know that the gradient of V points in the direction that V is increasing the fastest and that the gradient is orthogonal to the level curves of V . We do know that V (x. p. y) < 0. we say that V is negative definite and negative semidefinite if V (x) < 0 and V (x) ≤ 0. if we consider the system for a harmonic oscillator y′ = v k b v′ = − y − v.

with an isolated minimum at the origin. Lyapunov functions are defined on a domain U . Also.5. y) + (x. where linearization only tells us what happens on a small neighborhood around the equilibrium solution. y) = − − + 8. If we compare The- orem 1 to using linearization to determine stability of an equilibrium solution. The gradient of S is ( ) ∂S ∂S ∇S = . y) = ∂x ∂S g(x. y) dt is a gradient system if ∂S f (x.31 tells us that the origin is a stable equilibrium point. dt The function V (x. Unfortunately. y) ∂x ∂y = 2x(y + αx(x2 + y 2 )) + 2y(−x + αy(x2 + y 2 )) = 2α(x2 + y 2 )2 . We can compute ∂V ∂V V̇ = (x. ∂y For example. y)f (x. we will find that we can apply this result where linearization fails. y) dt dy = g(x. dx = y + αx(x2 + y 2 ) dt dy = −x + αy(x2 + y 2 ). If α < 0. where x2 x4 y 2 S(x. y)g(x.3. the system dx = x − x3 dt dy = −y dt is a gradient system. ∂x ∂y The system dx = f (x. 2 4 2 . there are no general ways of finding Lya- punov functions.31 is called a Lyapunov function. then V̇ is negative definite on R2 . Theorem 5.3 Gradient Systems Let S be a real-valued function on the xy-plane. y) = . The function V in Theorem 5. . y) = x2 + y 2 is positive definite on R2 .3. MORE NONLINEAR MECHANICS 273 Recall our example. 5.

spiral sinks. β = Sxy . If (x(t). suppose that dx ∂S = dt ∂x dy ∂S = dt ∂y is a gradient system. dt However. the eigenvalues are 1 1√ λ = (α + β) ± (α − γ)2 + 4β 2 . dS ∂S dx ∂S dy = + = (x − x3 )2 + y 2 ≥ 0. S increases at every point on the solution curves except at the equilibrium points. a gradient system can have no spiral sources. That is. In general. Since ( )2 ( )2 dS ∂S dx ∂S dy ∂S ∂S (x(t). . Let us see what this means in terms of the linearization of the system. g) is ( ) ( ) fx fy Sxx Sxy J= = . β γ where α = Sxx . x2 + y 2 is a Lyapunov function for the system dx = −x + y dt dy = −x − y. NONLINEAR SYSTEMS Now let us see what happens on the solution curves of this gradient system. 2 2 Since we have real eigenvalues. The Jacobian matrix of (f. the matrix J must have the form ( ) α β J= . gx gy Syx Syy Since Sxy = Syx . and γ = Syy . so this system cannot be a gradient system. y(t)) = + = + ≥ 0. y(t)). The characteristic polynomial of J is λ2 − (α + γ)λ + αγ − β 2 . For example. dt ∂x dt ∂y dt ∂x ∂y we know that S increases on every solution to the system except at the critical points of S(x(t). Therefore. S increases at point on the solution curve where the gradient of S is nonzero. or centers. the origin is a spiral sink. y(t)) is a solution curve.274 CHAPTER 5. dt ∂x dt ∂y dt Thus.

where x ̸= x0 . • The system dx = f (x. where x ̸= x0 . y0 ) is a stable equilibrium solution. The function V is called a Lyapunov function.4 Important Lessons • The equation for the nonlinear pendulum with damping dθ =v dt dv b g = − v − sin θ. any solution that starts near the equilibrium solution will stay near the equilibrium solution. then (x0 . y0 ) is a asymptotically stable equilibrium solution or a sink. ◦ We say that V is negative definite if V (x) < 0 for all x in S. That is. y) = . y) = ∂x ∂S g(x. y) dt has an equilibrium solution at (x0 .3. If V̇ is negative definite on U . Let V be a continuously differentiable function defined on a neighborhood U of (x0 . The function H for this system is an example of a Lyapunov function. ◦ We say that V is negative semidefinite if V (x) ≤ 0 for all x in S. y0 ).3. y0 ) is in S and V (x0 ) = 0. y0 ). where H is the Hamiltonian function for the ideal pendulum. ◦ We say that V is positive definite if V (x) > 0 for all x in S. • Suppose that the system dx = f (x. MORE NONLINEAR MECHANICS 275 5. y0 ) that is positive definite with minimum at (x0 . 1. y) dt dy = g(x. We can use these results to analyze the behavior of equilibrium solutions where lin- earization fails. We have no general methods for finding Lyapunov functions. 2. y) dt is a gradient system if ∂S f (x. ◦ We say that V is positive semidefinite if V (x) ≥ 0 for all x in S. ∂y . y) dt dy = g(x. then (x0 . dt m L can be analyzed by examining dH/dt. If V̇ is negative semidefinite on U . • Let V be a real-valued function defined on a set S in the xy-plane such that the point x0 = (x0 .5.

Fortunately. who founded the firm Pei Cobb Freed & Partners with famed architect I.5 Exercises 1. the architects and engineers thought thought that the problem was caused by the building’s tendancy to sway excessively in high winds. Hint. However.3 The building suffered many problems during construction. 5. 5. or centers. y(t)).3. spiral sinks. a gradient system has no spiral sources. New England’s tallest building.” 3 The John Hancock Tower is officially known by its street address. and the building was nicknamed the “Plywood Palace. Initially. Since ( )2 ( )2 dS ∂S ∂S (x(t). y(t)) = + ≥ 0. Since the eigenvalues of a gradient system are real. they later determined that the falling-window problem was caused by the air space between the double-paned windows and pressure differentials between the interior and exterior of the building. In early 1972 when construction was still underway.32). 790-foot mirror-glass John Hancock Tower. .6 Project—Tuned-Mass Dampers The 60-story. no one was injured. During the repairs. The problem was solved by replacing all of the windows with single sheets of fully tempered glass.3. NONLINEAR SYSTEMS where S be a real-valued function on the xy-plane. the most notorious of which was how the original glass windows were attached. dt ∂x ∂y S increases on every solution to the system except at the critical points of S(x(t). Pei. In all more than 100 of the building’s windows suffered the same fate.276 CHAPTER 5. 200 Clarendon. This is an exercise. (Figure 5. was completed in 1976 and was designed by Henry N. Cobb. one of the 500- pound windows popped out of the building and committed suicide on the sidewalk below.M. the windows were replaced with plywood.

This is the floor of concrete that you see near the top of the structure (Figure 5. In fact. To remedy than problem.3. office workers in the upper stories of the building complained of motion sickness. In addition. we will consider a mass-spring system connected to a fixed wall. Engineers determined that the building’s natural sway period was dangerously close to the period of its torsion.5. the engineers installed a pair of tuned mass dampeners on the 58th floor of the building. 1.500 tons of steel braces were installed to keep the building from falling over in a high wind. To construct a simple model of a tuned-mass damper. We .32). MORE NONLINEAR MECHANICS 277 Figure 5. The mass is allowed to slide freely with no damping. The building actually twisted as it was swaying back and forth.32: The John Hancock Tower Another flaw in the design of the building resulted in an extreme amount of swaying.

NONLINEAR SYSTEMS can think of this mass as the top of the building moving back and forth. x2 . a Hamiltonian system is a system of 2n equations of the form dxi ∂H = dt ∂yi . p1 . If we let L1 and L2 be the natural lengths of the springs. 2m1 2m2 2 L Indeed. If we “tune” the mass-damper by choosing the appropriate spring. (5.4 4 In general. dt ∂x2 We can show that a solution to the system.278 CHAPTER 5. we can rewrite this system as a system of four first-order linear equations. x2 (t). it is easy to check that dx1 ∂H = dt ∂p1 dx2 ∂H = dt ∂p2 dp1 ∂H =− dt ∂x1 dp2 ∂H =− . we can derive the Hamiltonian function p21 p2 k1 k2 H(x1 . we can remove energy from the system and the motion of the first mass (the building) will decrease.27) dt2 d2 x 2 m2 2 = −k2 (x2 − x1 − L2 ). we attach a second mass by using a spring and a dashpot. p2 (t)). dx1 p1 = dt m1 dx2 p2 = dt m2 dp1 = −k1 (x1 − L1 ) + k2 (x2 − x1 − L2 ) dt dp2 = −k2 (x2 − x1 − L2 ). To this first mass. the first spring will exert a force −k1 (x1 − L1 ). is constant on H. p2 ) = + 2 + (x1 − L1 )2 + (x2 − x1 − L2 )2 . (x1 (t). The second spring pushes the two masses in opposite directions according to ±k2 (x2 − x1 − L2 ). p1 (t). This second mass is called the tuned mass-damper. Thus. We will first model our system without a dashpot.28) dt If we let dx1 p1 = m1 dt dx2 p2 = m2 dt be the momenta of the two masses. we can use Newton’s second law to derive two equations d2 x 1 m1 = −k1 (x1 − L1 ) + k2 (x2 − x1 − L2 ) (5. dt If we consider the kinetic and potential energy of both masses and springs.

We want to choose b fairly large so that there is a rapid loss of energy. We. remember that m2 is sitting on top of a very tall building. Thus.28) become ( ) d2 x1 p2 p1 m1 = −k (x 1 1 − L 1 ) + k (x 2 2 − x 1 − L 2 ) + b − dt2 m2 m1 2 ( ) d x2 p2 p1 m2 2 = −k2 (x2 − x1 − L2 ) − b − . We should.. . Let us consider example. Moreover. MORE NONLINEAR MECHANICS 279 Now suppose that we add a damping term to our system. energy decreases whenever the second mass is moving relative to the first. Suppose that m1 = 1. Finally. i. ( ) ( ) dx2 dx1 p2 p1 ±b − = ±b − . . We should choose m2 large enough so that this mass oscillates with respect to m1 . then the tuned mass-damper (m2 ) located on one of the top floors of the building will start to move relative to the building and energy will be removed from the system by the dampers. . (5. .27) and (5. . . If we choose k2 so that the second spring is in resonance with the first. k2 and b. dH/dt < 0 whenever the distance between the two masses is changing. We can show that H is constant on solution curves of the system. . n. . xn . If the initial conditions for our system are x1 (0) = 10 dyi ∂H =− dt ∂xi for i = 1. we wish to choose m2 to be large so that we guaranteed that this mass will oscillate with respect to m1 . where H(x1 . if the wind or an earthquake starts our building (m1 ) swaying back and forth. Indeed. dt m2 m1 and our new first-order system is dx1 p1 = dt m1 dx2 p2 = dt m2 ( ) dp1 p2 p1 = −k1 (x1 − L1 ) + k2 (x2 − x1 − L2 ) + b − dt m2 m1 ( ) dp2 p2 p1 = −k2 (x2 − x1 − L2 ) − b − . Thus.1. we will have relatively large oscillations of the second mass and a large loss of energy. .5. are free to choose m2 . then the strong damper will almost serve as a rigid connection between m1 and m2 .29) dt m2 m1 dt dt Equation (5. we should choose k2 to maximize the rate at which the second mass oscillates with respect to the first and to maximize the oscillations of m2 . . Thus. Of course.29) tells us that dH/dt ≤ 0.e. We will also choose m2 to be 0. dt m2 m1 Hamiltonian. dt dt m2 m1 Then equations (5. the magnitude of dH/dt is large. Therefore. ( )2 ( )2 dH p2 p1 dx2 dx1 = −b − = −b − . . 2. k1 = 1. the oscillations of the first mass force the second at its resonance frequency. y1 . If we choose m2 two small. however.05 and b = 0. however. yn is a real-valued differentiable function on R2n such that H is nonconstant on every open ball in R2n . m1 and k1 involve the building and are set by the architects and engineers.3. .

5.5 If we choose k2 ≈ 0. CA. Robert L. an equilibrium solution might shift. we will choose k2 to tune our system. If such a change occurs in the system through an adjustment of a parameter. we say a bifurcation has occurred. p.5 for various values of k2 . Brooks/Cole. NONLINEAR SYSTEMS x2 (0) = 0 p1 (0) = 0 p2 (0) = 0. Hall. For example. and Glen R. we√have a single equilibrium solution at the origin if α = 0 and two equilibrium solutions.280 CHAPTER 5. However. the system dx = α + x2 dt dy = −y dt has no equilibrium solution if α > 0 (Figure 5. To do this we will plot the time t that it takes for the amplitude of the oscillations of m1 to reach and stay below 2. .42 in Paul Blanchard. However.1 Bifurcations The simplest bifurcations occur when an equilibrium solution appears. (± −α. 2002.33). if α < 0. disappears. then we might reasonably expect the phase portrait to change slightly. y).4. Differential Equations. 5 See Figure 5. then we can minimize the time to be t ≈ 120. dt we may see closed orbits and various types of equilibrium solutions. but we would not expect new equilibrium solutions to appear or a spiral source to turn into a spiral sink. Second edition. Devaney.4 The Hopf Bifurcation If we consider the phase portrait of a system dx = f (x.05. 5. 0). a closed orbit might expand or contract. this may be exactly what happens. For example. or a spiral source might tighten up. y) dt dy = g(x. or splits into two or more equilibrium solutions. 499. If we make a slight change in the system. Pacific Grove.

This is not an elementary equilibrium point such a saddle. On the right half of the xy-plane. then J has eigenvalues 0 and −1. while on the left half it resembles a sink (Figure 5. 0 −1 If α = 0. . y) = .5. THE HOPF BIFURCATION 281 10 5 y(t) 0 -5 -10 -10 -5 0 5 10 x(t) Figure 5.34). or a spiral source. We call this type of equilibrium solution a saddle-node.4. a sink.33: Solutions for α > 0 The Jacobian of the system dx = α + x2 dt dy = −y dt is ( ) 2x 0 J(x. the equilibrium solution resembles a saddle.

5 -0. x x = α . since both eigenvalues are negative.5 Bifurcation Point α -1.35: Bifurcation diagram . Stable Equilibrium p Figure 5. The two equilibrium solutions move in opposite directions as α decreases. we have a nodal sink. NONLINEAR SYSTEMS 10 5 y(t) 0 -5 -10 -10 -5 0 5 10 x(t) Figure 5. Saddle Point p 1 0.5 -1 x = − α .282 CHAPTER 5. 0 −1 0 −1 In the first case. then ( √ ) ( √ ) 2 α 0 −2 α 0 J= or .5 -1 -0. In the second case.35.5 1 1. We can summarize what happens for various values of α with a bifurcation diagram such as the one in Figure 5. we have a saddle since the eigenvalues are real and of opposite sign.5 0.34: Solutions for α < 0 If α < 0.

4. 2 1 y(t) 0 -1 -2 -2 -1 0 1 2 x(t) Figure 5. we have a spiral sink. and then to a spiral source (α > 0). In the case of the nonlinear system. the origin changes to a center (α = 0). while solutions outside of the orbit spiral inward (Figure 5. The origin is stable equilibrium solution with solutions spiraling into the origin very slowly. something quite different happens at α = 0. dt The origin is an equilibrium solution for all values of α. As α increases past zero. In addition. The linearization of our system is dx = αx + 5y dt dy = −5x + αy.36: Solutions for α < 0 . If α < 0. As α moves from negative values to positive values. a closed orbit of radius α develops. the origin is still a spiral source (Figure 5. the origin is still a spiral sink for α > 0 (Figure 5.2 The Hopf Bifurcation Now let us consider an entirely different type of bifurcation by examining the system dx = αx + 5y − x(x2 + y 2 ) dt dy = −5x + αy − y(x2 + y 2 ).37).4. we can easily determine the nature of of our linearization.36). Solutions inside of this closed orbit spiral out towards the orbit. However. dt Since the eigenvalues of this system are α ± 5i.38). THE HOPF BIFURCATION 283 5. If α > 0.5. the equilibrium solution destabilizes and √ becomes a source.

If we make the substitution x = r cos θ and y = r sin θ.284 CHAPTER 5. NONLINEAR SYSTEMS 2 1 y(t) 0 -1 -2 -2 -1 0 1 2 x(t) Figure 5. we will rewrite our system in polar coordinates.37: Solutions for α = 0 2 1 y(t) 0 -1 -2 -2 -1 0 1 2 x(t) Figure 5. our nonlinear system can be rewritten as dr = αr − r3 dt .38: Solutions for α > 0 To see exactly what happens as α passes zero and becomes positive.

. y) dt parameterized by α. (5. b(α0 ) ̸= 0. r′ = 0. the origin is a sink since αr − r3 < 0 for all r > 0. We also know that r′ > 0 for 0 < r < α and r′ < 0 √ √ if r > α. y) dt dy = gα (x. dt If α < 0. and b(0) = 5. Example 5. If we let α0 = 0. when r = α.4. All nonzero solutions spiral towards this orbit as t → ∞. the origin is still an equilibrium solution. and that the equilibrium point is asymptotically stable. Thus. Suppose that (x0 (α). a′ (0) = 1. No new equilibrium solutions arise. y0 (α)) is a equilibrium solution for the family of systems dx = fα (x. This equation can be written as the system dx =y (5. and the Jacobian matrix for the system evaluated at this equilibrium has eigenvalues a(α) ± ib(α). The origin is the only equilibrium solution to Van der Pol’s equation. the hypothesis of the Hopf Bifurcation Theorem are satisfied and we are guaranteed a period solution surrounding our equilibrium solution.40 (Van der Pol’s equation). Springer-Verlag. So the circle of radius α is a periodic solution with the trajectory moving clockwise since θ′ = −5 < 0. a′ (α0 ) > 0.30) dt dy = −x + x(1 − x2 )y. If we examine the system consisting of only linear terms on the right-hand side of the equation. but a periodic solution develops as α passes through the bifurcation value.39 (Hopf Bifurcation). we have a(α0 ) = 0.31) dt The phase portrait for Van der Pol’s equation is given in Figure 5. 0) with eigenvalues α ± 5i. New York. THE HOPF BIFURCATION 285 dθ = −5. is x′′ − x(1 − x2 )x′ + x = 0. When α > 0. Theorem 5. In this case all solutions tend towards the origin as t → ∞. Van der Pol’s equation. dx =y dt dy = −x + y dt 6 For a proof and description of the Hopf Bifurcation Theorem see C. Then there exists an α1 such that the system has a periodic solution encircling the equilibrium solution for α 0 < α < α 1 .5. Assume that at some α = α0 . Chicone. √ √ Moreover. Ordinary Differential Equations with Applications. 1999.41. then a(0) = 0. a simple nonlinear equa- tion having applications in electrical engineering and mathematical biology. This type of bifurcation is called a Hopf bifurcation. and one might guess that this solution acts like a spiral source by examining the phase portrait.6 Recall that our example system dx = αx + 5y − x(x2 + y 2 ) dt dy = −5x + αy − y(x2 + y 2 ) dt has an equilibrium at (0.

. dt 0. This suggests that the origin looks like a spiral source at least locally. −1 1 √ has eigenvalues λ = (1 ± i 3)/2. We might model this satiable predator-prey scenario by the equations dx cxy = (a − bx)x − dt d + αx dy f xy = −ey + .3 + αx dy xy = −0. . Let us choose the constants a. f and α are constants. 0). an increase in the prey population has little effect on the interaction terms in our model. NONLINEAR SYSTEMS we might get a better sense of what is happening. Here. .30)–(5.3 + αx . dt d + αx where x(t) is the prey population at time t and y(t) is the predator population at time t. . ( ) 0 1 A= . If this is the case. . f so that our system becomes dx xy = (1 − x)x − dt 0.5y + . the Jacobian matrix for (5. 4 3 2 1 y(t) 0 -1 -2 -3 -4 -4 -3 -2 -1 0 1 2 3 4 x(t) Figure 5. a.286 CHAPTER 5. we will assume that a predator’s appetite is satiated when food is abundant. . −1 − 2x0 y0 1 − x20 which agrees with A for (x0 . . . y0 ) = (0. Indeed. In our model.41: Van der Pol’s equation Let us examine how a bifurcation might occur in a predator-prey model. The matrix for this linear system.30) is ( ) 0 1 J= .

9. If α = 0.6 0.44).5.42: Predator-prey model with α = 0 . we have a periodic trajectory.2 0 0 0.4 0.4. THE HOPF BIFURCATION 287 The constant α is know as the satiation constant.8 0. all trajectories inside the population quadrant spiral asymptotically towards a single stable equilibrium point (Figure 5. we have a very similar phase portrait with solutions spiraling in towards a stable equilibrium solution (Figure 5. If α = 1. Let us see examine the phase portrait for several values of α.43). if α = 0.8 1 x(t) Figure 5.35. 1 0.6 y(t) 0.42).2 0. In this case. The larger the value of α. the more quickly a predator’s appetite is satiated as the prey population increases. then there is no satiation effect.4 0. However. All other trajectories are pulled toward this attracting periodic orbit (Figure 5.

2 0.2 0.43: Predator-prey model with α = 1.8 0.6 y(t) 0.6 0.44: Predator-prey model with α = 0.8 1 x(t) Figure 5.6 y(t) 0.4 0.6 0.4 0.35 1 0. NONLINEAR SYSTEMS 1 0.8 1 x(t) Figure 5.4 0.2 0 0 0.288 CHAPTER 5.9 .4 0.2 0 0 0.8 0.

and that the equilibrium point is asymptotically stable.85.6 0.5. • A satiable predator-prey scenario can be modeled by the equations dx cxy = (a − bx)x − dt d + αy dy f xy = −ey + .2 0. For example. the amplitude of the periodic solution begins to decrease. The x-coordinate of our equilibrium solution is 0.3 Bifurcation Point sink 0.5.3 Important Lessons • If we make a slight change in the system. At this point the equilibrium solution destabilizes and spawns an attracting periodic orbit. However. at α ≈ 0.5α and we can summarize our findings with a bifurcation diagram (Figure 5. but we would not expect new equilibrium solutions to appear or a spiral source to turn into a spiral sink. y) dt parameterized by α.5 0.15 x= . or a spiral source might tighten up. • Suppose that (x0 (α).4 Figure 5.4 0. a closed orbit might expand or contract. an equilibrium solution might shift.4 source 0. THE HOPF BIFURCATION 289 x 0. • A Hopf bifurcation if an attracting periodic solution encircling an equilibrium so- lution develops as α passes through the bifurcation value.45).2. and the Jacobian matrix for the system evaluated at this equilib- rium has eigenvalues a(α) ± ib(α). However. a′ (α0 ) > 0. dt d + αy . y) dt dy = gα (x.2 1. the equilibrium point re-stabilizes and the periodic solution is absorbed. For α > 1.6 sink 0. 5. Assume that at some α = α0 . At α ≈ 1.4. we say a bifurcation has occurred. As α continues to increase. If such a change occurs in the system through an adjustment of a parameter. the amplitude of this periodic orbit increases. Then there exists an α1 such that the system has a periodic solution encircling the equilib- rium solution for α0 < α < α1 . then we might reasonably expect the phase portrait to change slightly. we have a(α0 ) = 0.8 1 1. 1 − 0.2 Bifurcation Point α 0 0.4. b(α0 ) ̸= 0. y0 (α)) is a equilibrium solution for the family of systems dx = fα (x.45: Bifurcation diagram for the satiable predator-prey model As α continues to increase a Hopf bifurcation occurs at α ≈ 0. we only have a stable equilibrium solution in the first quadrant.2. this may be exactly what happens.

Consider the system of differential equations x′ = x(−x − y + 1) y ′ = y(−ax − y + b). 4. b > 0. y) = + − 2 2 3 is a Lyapunov function for the system. (c) What can you conclude about the phase portrait of the system from (1) and (2)? Hint. sketch the level sets of S. . y ≥ 0. Suppose that this system is only defined for x. Consider the system x′ = y y y ′ = −x − + x2 . . Let G(x. (c) How many dead fish are there and where are they? (d) Using the results from part (2). Suppose that the smell of a bunch of dead fish in the region −2 ≤ x ≤ 2. 3. The constant α is know as the satiation constant. Here. 2. .4. (a) What is the gradient system with vector field given by the gradient of G? (b) Sketch the graph of G and the level sets of G. y) = x3 − 3xy 2 . 4 (a) Verify that the function x2 y 2 x3 L(x. (b) Sketch the level sets of L. f and α are constants.290 CHAPTER 5. where a and b are parameters with a. NONLINEAR SYSTEMS where x(t) is the prey population at time t and y(t) is the predator population at time t. 5. −2 ≤ y ≤ 2 is given by the function x4 + y 4 S(x. a. . (c) Sketch the phase portrait of the gradient system in (1).4 Exercises 1. . the more quickly a predator’s appetite is satiated as the prey population increases. (e) Why is this model not realistic for large values of x or y? Hint. 4 (a) What is the gradient system who vector field is the gradient of S? (b) Use Sage to graph the phase portrait of the system. Hint. (a) Use nullclines to sketch the phase portrait for this system for several values of a and b. y) = x2 + y 2 − − 3x2 y 2 + 100. The larger the value of α.

dt where σ. initial values are never precisely known. y) = (0. In addition. and β = 8/3. Edward N. The Lorenz equations can be written as dx = −σx + σy dt dy = ρx − y − xz dt dz = −βz + xy. our system dx = −10x + 10y dt . However. ρ. Lorenz. Hint. and β are constants. Hint. Since real data always has some inherent uncertainty. When he analyzed the system. when he let σ = 10. Show that the quadratic polynomial V (x. an MIT professor. and we may have little success modeling real world phenomena. solutions can stay in a bounded region of the three dimensional version of the phase plane and wind through the region along an incredibly convoluted path. Use the positive definite function V (x. 0) if and only if a > 0 and ac − b2 > 0.5. 0). For certain values of σ. y) = (0. the trajectories are extremely sensitive to initial conditions. THE HOPF BIFURCATION 291 (b) Determine the values of a and b at which a bifurcation occurs. Lorenz discovered that the trajectories of the solutions were incredibly convoluted and effectively unpredictable for certain parameters.4. y) = ax2 + 2bxy + cy 2 is positive definite with minimum at (x. he found that the trajectories of the solutions were incredibly convoluted and effectively unpredictable for certain parameters.5 Project—The Lorenz Equations The Lorenz Equations In 1963. Hint. 5. 6.4. published a paper on his research in mete- orology. Lorenz noticed that the system behaved strangely. 5. Thus. however. Lorenz had developed a simplified system to model certain weather-related phenomena. y) = x2 + y 2 to argue that the system x′ = −x − 2y 2 y ′ = 2xy − y 3 has an asymptotically stable equilibrium solution at (x. and β. ρ. Using differential equations. x(0) = x0 y(0) = y0 z(0) = z0 . The Existence and Uniqueness Theorem for systems of differential equations guarantees a unique solution for each set of initial conditions. There is much more freedom to move around in three dimensions than there is in two. ρ = 28.

(x. z) = 28 − z −1 −x  . ( ) 8 F(x. − z + xy . (6 2. 27). That is. y. 28x − y − xz. so we will have to use a numerical algorithm to approximate solutions (Figure 5. 0. NONLINEAR SYSTEMS dy = 28x − y − xz dt dz 8 = − z + xy. z) = −10x + 10y. the Jacobian is   −10 10 0 J =  28 −1 0 . 6 2. 0. 3 and the equilibrium solutions occur exactly when this vector field is zero. y. −6 2.292 CHAPTER 5. The Equilibrium Solutions If we wish to understand the nature of the equilibrium solutions of the Lorenz system. − z + xy = (0. z) is an equilibrium solution if ( ) 8 F(x. 0 0 −8/3 . y. and (−6 2. 27) are the equilibrium solutions to the Lorenz system for the parameters that we have chosen. 28x − y − xz. y.46). With the exception of these equilibrium solutions it is probably impossible to find solutions for the system. 0). The Jacobian matrix of the Lorenz system is   −10 10 0 J(x. dt 3 defines a vector field in R3 .46: The solution curves for two systems with nearly equal initial conditions. it makes sense to linearize the system. z) = −10x + 10y. 3 √ √ √ √ It is easy to check that (0. y x −8/3 At the origin. Figure 5. 0).

dt 3 This system decouples into two systems. That is. Therefore. the equilibrium solution (−6 2. it begins to spiral away. the trajectory come under the influence of the saddle at the origin and either returns to repeat it previous pattern or moves over to the other equilibrium solution (Figure 5.2i.8 λ2 ≈ 0. Now let us consider √ what √ happens at the other two equilibrium solutions. then spiral slowly away along the √ plane corresponding √ to the complex eigenvalues.2i λ3 ≈ 0. A √ trajectory approaches one of the two equilibrium solutions (±6 2. . we have the Jacobian matrix   −10 10 0√ J = 1 −1 . −6 2. ±6 2. Now that we have some idea of how the solutions of the Lorenz system behave near the equilibrium solutions. the origin is a sink in the z-direction and a saddle in the xy-plane. 27). 6 2. √ √ −6 2 6 2 6 2 −8/3 The eigenvalues of this matrix are λ1 ≈ −13. For the equilibrium point (6 2.5. Similarly. dt and dz 8 = − z. dt 3 Consequently. THE HOPF BIFURCATION 293 Thus. When the trajectory gets close enough to the equilibrium solutions. we can start to understand global behavior √ of the system.4. When the spiral is large enough. 27) along the straight line corresponding to the negative eigenvalue. solutions approach this equilibrium point along the direction of the eigenvalue of the negative eigenvector.094 + 10. 27) is a spiral saddle. dx = −10x + 10y dt dy = 28x − y.47). the linear system that approximates the Lorenz system at the origin is dx = −10x + 10y dt dy = 28x − y dt dz 8 = − z.094 − 10. this equilibrium point is a spiral saddle that has a line of solutions that tend toward the equilibrium point as t increases and a plane of solutions that spiral towards the equilibrium point as t decreases.

If V (x. 2.48. Assuming this. 5.294 CHAPTER 5. 3. y. Use part (4) to show that the ellipsoid is invariant. Solution. z) ≤ 3br2 /m} . Show that V̇ = −2a[rx2 + y 2 + b(z − r)2 − br2 ]. z) = rx2 + ay 2 + a(z − 2r)2 . y. Exercise 5. y. It is true that (z − r)2 ≥ (z − 2r)2 − r2 . 1/a. Consider the Lorenz system x′ = −ax + ay y ′ = rx − y − xz z ′ = −bz + xy. we shall show that the solutions of the system are bounded. Use parts (1) and (2) to show that V̇ ≤ −2maV + 4abr2 . NONLINEAR SYSTEMS Figure 5. where m is the smallest of the three numbers 1.47: Solutions in R3 . Use part (3) to show that V̇ ≤ −2abr2 < 0 everywhere outside of the ellipsoid { } R = (x. . and b/2a. z)|V (x. and that every solution curve ends up inside of R. 1. show that rx2 + y 2 + b(z − r)2 ≥ mV − br2 . Hint. 4. In this exercise.

6 The Laplace Transform 6. C The function E ′ (t) can have discontinuities. For example. . In other words. As a second example. this method will even work when g is a discontinuous function. let us consider a population of fish that is governed by exponential growth. Surprisingly. 295 . However.      0 1/2 < t < 1. we might only allow harvesting during the first half of the year.      .   0 3/2 < t < 2. the Laplace transform. the idea is to use the Laplace transform to change the differential equation into an equation that can be solved algebraically and then transform the algebraic solution back into a solution of the differential equation. allows us to change a differential equation to an algebraic equation. H(t) = r 1 ≤ t ≤ 3/2. dP = kP dt P (0) = P0 . the voltage in the circuit can be periodically turned on and off.    r 0 ≤ t ≤ 1/2. and suppose that we wish to determine the effects of seasonal fishing.1 The Laplace Transform Consider the electrical circuit governed by the differential equation 1 LI ′′ + RI ′ + I = E ′ (t). We do not yet have the tools to analyze differential equations possessing discontinuous terms. harvesting will not be continuous. For example.. an integral transform. provided the discontinuities are not too bad. The previous methods that we have used to solve second order linear differential equations may not apply here. Given an initial-value problem ay ′′ + by ′ + cy = g(t) y(0) = y0 y ′ (0) = y0′ .

the integral ∫ ∞ et e−st dt 2 0 does not converge. b] if f satisfies the following conditions. 0 s2 for s > 0.1 Definition of the Laplace Transform We shall define the Laplace transform of a function f (t) by ∫ ∞ L(f )(s) = F (s) = e−st f (t) dt. Then for any real number s. • Let f (t) = k. The Laplace transform of does not always exist.1. If c ∈ [a. let us compute several Laplace transforms. 2. ∫ b ∫ b t2 −st e e dt = et(t−s) dt. 0 s−a for s > a. Thus. does exist in many cases.296 CHAPTER 6. before we investigate these properties. However. We also assume that limt→a+ f (t) and limt→b− f (t) exist. We will show that the Laplace transform of a function exists if the function does not grow too quickly and has no bad discontinuities. 0 s for s > 0. . even for functions that are infinitely 2 differentiable. THE LAPLACE TRANSFORM 6. let f (t) = et . • If f (t) = eat . the Laplace transform is ∫ ∞ 1 L(f )(s) = te−st dt = . • For f (t) = t. 0 0 and the integrand is greater then one for t > s. b] is a discontinuity. Then ∫ ∞ k L(f )(s) = ke−st dt = . b]. then the Laplace transform of f is ∫ ∞ 1 L(f )(s) = e−st eat dt = . 0 provided the integral converges. For example. The Laplace transform has many nice properties. where k is a constant. The Laplace transform. There are a finite number of discontinuities of f on [a. We say that a function f is piecewise continuous on an interval [a. however. then the one-sided limits lim f (t) and lim f (t) t→c− t→c+ both exist. 1. especially with respect to the deriva- tive of f .

b] for all b > 0. Theorem 6. ∞) does not grow too quickly as t → ∞. ∫ ∞ L(f )(s) = F (s) = e−st f (t) dt 0 exists.1. ∞). That is. The square wave function { 1.2. this is the reason why Laplace transforms are so useful when trying to solve initial value problems. Transforms of functions having the linearity property are called linear operators.1. Laplace transforms forms behave very nicely with respect to derivatives. The proof of this statement follows directly from the definition of the Laplace transform and the properties of integrations. THE LAPLACE TRANSFORM 297 We say that a function is piece-wise continuous on an interval [0. exponentially bounded function defined [0. We can evaluate the Laplace transform of y ′ by using integration by parts. We also wish to be certain that a function f (t) defined on [0. then the continuities are not too bad.2 Properties of the Laplace Transform One of the most important properties of the Laplace transform is linearity. the graph of f must lie between the curves y = M eat and y = −M eat . ∞). In fact. A function f (t) is exponentially bounded on [0. ∫ ∞ L(y ′ )(s) = y ′ (t)e−st dt 0 [ . ∞) if there exist constants M ≥ 0 and a such that |f (t)| ≤ M eat . t > 1. exponentially bounded function and assume that y ′ is also exponentially bounded. ∞) if it is piecewise continuous on the interval [0. 0 ≤ t ≤ 1 f (t) = 0. The types of discontinuities that we are de- scribing are sometimes called jump discontinuities. then the Laplace transform of f . where α. Let y = y(t) be a piecewise continuous. In other words. Then for large values of s L(y ′ )(s) = sL(y)(s) − y(0) = sY (s) − y(0). If a function is piecewise continuous. Theorem 6. where f (t + 2) = f (t) is an example of a piecewise continuous function defined on [0.6. ∞).1. 6. where Y (s) is the Laplace transform of y. for all t in [0. Proof. L(αf + βg)(s) = αL(f )(s) + βL(g)(s). If f is a piecewise continuous. β ∈ R provided the Laplace transforms of f and g exist.

b ∫ b ] −st .

−st = lim e y(t).

+ s y(t)e dt b→∞ 0 0 [ ] = lim e−sb y(b) − y(0) + sL(y)(s) . b→∞ .

for all t in [0. • The Laplace transform is a linear operator. ∫ ∞ L(f )(s) = F (s) = e−st f (t) dt 0 exists. ∞). In general. then the Laplace transform of f . Thus. • If f is a piecewise continuous.e.. Since. Let y = y(t) and y ′ (t) be piecewise continuous. exponentially bounded functions and assume that y ′′ is exponentially bounded. The right-hand side of this inequality as b → ∞ for s > a. 6. • Let y = y(t) be a piecewise continuous. exponentially bounded function and assume that y ′ is also exponentially bounded. Then for large values of s L(y ′′ )(s) = s2 L(y)(s) − sy(0) − y ′ (0) = s2 Y (s) − sy(0) − y ′ (0). where Y (s) is the Laplace transform of y. Then for large values of s L(y ′ )(s) = sL(y)(s) − y(0) = sY (s) − y(0). there exist constants M ≥ 0 and a such that |y(t)| ≤ M eat . Theorem 6. THE LAPLACE TRANSFORM We claim that limb→∞ e−sb y(b) = 0.3. where Y (s) is the Laplace transform of y. then L(y (k) )(s) = sk L(y)(s) − sk−1 y(0) − · · · − sy (k−2) (0) − y (k−1) (0) = sk Y (s) − sk−1 y(0) − sk−1 y(0) − · · · − sy (k−2) (0) − y (k−1) (0).1. ∞). β ∈ R provided the Laplace transforms of f and g exist. i. exponentially bounded function defined [0. . where α. |e−sb y(b)| ≤ M e−(s−a)b . y is exponentially bounded. Thus.3 Important Lessons • Many initial value problems have discontinuous forcing terms. L(αf + βg)(s) = αL(f )(s) + βL(g)(s). 0 provided the integral converges.298 CHAPTER 6. • The Laplace transform of a function f (t) by ∫ ∞ L(f )(s) = F (s) = e−st f (t) dt. L(y ′ )(s) = sL(y)(s) − y(0). exponentially bounded functions and y (k) is piecewise continuous. The Laplace transform also behave nicely with respect to higher order derivatives. if y = y(t) and all of its derivatives up to order k − 1 are piecewise continuous.

(b) Prove that Γ(x + 1) = xΓ(x) and deduce that Γ(n + 1) = n! for n ∈ N. If f is a piecewise continuous. prove that the Laplace transform of f . exponentially bounded functions and y (k) is piecewise continuous. where Y (s) is the Laplace transform of y.1. (a) Let y = y(t) and y ′ (t) be piecewise continuous. prove that L(y (k) )(s) = sk L(y)(s) − sk−1 y(0) − · · · − sy (k−2) (0) − y (k−1) (0) = sk Y (s) − sk−1 y(0) − sk−1 y(0) − · · · − sy (k−2) (0) − y (k−1) (0). 4. (a) te−2t (b) sin at (c) cos at (d) eat sin bt 2. exponentially bounded functions and assume that y ′′ is exponentially bounded. then L(y (k) )(s) = sk L(y)(s) − sk−1 y(0) − · · · − sy (k−2) (0) − y (k−1) (0) = sk Y (s) − sk−1 y(0) − sk−1 y(0) − · · · − sy (k−2) (0) − y (k−1) (0). 3. 0 (a) Show that Γ(1) = 1.6. THE LAPLACE TRANSFORM 299 • Let y = y(t) and y ′ (t) be piecewise continuous. if y = y(t) and all of its derivatives up to order k−1 are piecewise continuous. where Y (s) is the Laplace transform of y. Find the Laplace transform of each of the following functions. In general. exponentially bounded functions and assume that y ′′ is exponentially bounded. (c) Show that the Laplace transform of f (t) = tα is Γ(α + 1) L(f )(s) = . ∫ ∞ L(f )(s) = F (s) = e−st f (t) dt 0 exists. . then L(tn )(s) = n!/sn+1 . Then for large values of s L(y ′′ )(s) = s2 L(y)(s) − sy(0) − y ′ (0) = s2 Y (s) − sy(0) − y ′ (0). exponentially bounded functions and y (k) is piecewise continuous. Prove that L(y ′′ )(s) = s2 L(y)(s) − sy(0) − y ′ (0) = s2 Y (s) − sy(0) − y ′ (0) for large values of s. sα+1 If n is a positive integer.4 Exercises 1. exponentially bounded function defined [0. (b) If y = y(t) and all of its derivatives up to order k − 1 are piecewise continuous.1. ∞). Define the gamma function to be ∫ ∞ Γ(x) = e−t tx−1 dt. 6.

5 5 Of course. Let f and g be two piecewise continuous exponentially bounded functions with Laplace transforms. If we take the Laplace transform of both sides of the equation y ′′ − y ′ − 6y = 0. Consider y ′′ − y ′ − 6y = 0 y(0) = 2 y ′ (0) = −1. s−a and the solution that we seek is 3 7 y(t) = e3t + e−2t . which we will denote by L−1 (F (s))(t). we get 2s − 3 3/5 7/5 Y (s) = = + . as the function f (t) whose unique Laplace transform is F (s). Then f (t) = g(t) for all t ≥ 0. Since the Laplace transform of eat is 1/(s − a). respectively. where Y (s) is the Laplace transform of y. there are much easier methods that we can use if we would like to solve this particular initial value problem. F (s) and G(s). Theorem 6. we know that ( ) −1 1 L = eat . We can now solve our initial value problem. It now makes sense to define the inverse Laplace transform of a function F (s). THE LAPLACE TRANSFORM 6. we can use Laplace transforms to help us solve initial value problems. We state the following theorem without proof. we can recover the original function. but it is nice to see that the two methods agree. If we know the Laplace transform of a function. Suppose that F (s) = G(s) for all s > c. we have L(y ′′ − y ′ − 6y) = L(y ′′ ) − L(y ′ ) − 6L(y) = [s2 Y (s) − sy(0) − y ′ (0)] − [sY (s) − y(0)] − 6Y (s) = [s2 Y (s) − 2s + 1] − [sY (s) − 2] − 6Y (s) = L(0) = 0.300 CHAPTER 6. The idea is to convert an initial value problem into an algebraic equation. the inverse Laplace transform is linear. L−1 (αF + βG) = αL−1 (F ) + βL−1 (G). Furthermore. Solving for Y (s). (s − 3)(s + 2) s−3 s+2 where we have used partial fractions to get the last expression.1 The Inverse Laplace Transform As we mentioned in the previous section. where c is some positive number.2 Solving Initial Value Problems 6. and recall that L(y ′ )(s) = sL(y)(s) − y(0) = sY (s) − y(0) L(y ′′ )(s) = s2 L(y)(s) − sy(0) − y ′ (0) = s2 Y (s) − sy(0) − y ′ (0). .4.2.

s > 0 sinh at a/(s2 − a2 ).2.2. we need either tables or computer software such as Sage so that we can easily find Laplace transforms and inverse Laplace transforms. we need to be able to handle discontinuous functions. s > 0 eat 1/(s − a). A function of this form is called a step function or Heaviside function. s > 0 sin at a/(s2 + a2 ). s With this new information. named for the British engineer Oliver Heaviside. we define the function { 0 t<c uc (t) = 1 t ≥ c. s > a tn .2 A Table of Laplace Transforms To use the method of Laplace transforms effectively. If c ≥ 0. s > 0 uc (t)f (t − c) e−cs F (s) ect f (t) F (s − c) f (ct) (1/c)F (s/c).3 Discontinuous Functions In order for the Laplace transform to be useful in solving initial value problems. s > 0 cos at s/(s2 + a2 ). s > a uc (t) e−cs /s. n ∈ N n!/sn+1 . n ∈ N n!/(s − a)n+1 . SOLVING INITIAL VALUE PROBLEMS 301 6. s > a tn eat .6. c > 0 δ(t − c) s F (s) − sf (0) − f ′ (0) 2 f ′ (t) s2 F (s) − sf (0) − f ′ (0) f ′′ (t) s2 F (s) − sf (0) − f ′ (0) Table 6. s > a eat cos bt (s − a)/[(s − a)2 + b2 ]. s > |a| cosh at s/(s2 − a2 ). we can now solve initial value problems such as y ′ = −y + u3 (t) . s > 0 ta . a > −1 Γ(a + 1)/sa+1 . The Laplace transform of uc is given by ∫ ∞ L(uc (t))(s) = uc (t)e−st dt ∫0 c ∫ ∞ = uc (t)e−st dt + uc (t)e−st dt ∫0 ∞ c = e−st dt c e−cs = . s > |a| eat sin bt b/[(s − a)2 + b2 ]. f (t) = L−1 (F (s)) F (s) = L(f (t)) 1 1/s.2.5: Table of Laplace Transforms 6.

Let us compute the Laplace transform of g. s(s + 1) s s−1 Hence. s Using the fact that y(0) = 2 and solving for Y (s). then ∫ ∞ ∫ ∞ −s(u+c) −sc L(g) = f (u)e dt = e f (u)e−su dt = e−sc L(f ). then L(uc (t)f (t − c)) = e−sc F (s). 0 0 In other words. we obtain e−3s sY (s) − y(0) = −Y (s) + . s+1 however. ( ) ( ) ( ) ( ) −1 e−3s −1 e−3s −1 e−3s −1 e−3s L =L −L = u3 (t) − L . consider the new function g(t) = uc (t)f (t − c). Notice that we can rewrite the equation y ′ = −y + u3 (t) as { −y t<3 y′ = −y + 1 t ≥ 3. if L(f ) = F (s). s + 1 s(s + 1) Therefore. THE LAPLACE TRANSFORM y(0) = 2. To obtain the function g from f . Returning to our example. we shift the graph of the f to the right by c units and let g(t) = 0 for 0 ≤ t < c. ( ) ( ) −1 2 −1 e−3s y(t) = L +L .302 CHAPTER 6. we will now show how we can use this new information to compute the inverse Laplace transform of e−3s . we get 2 e−3s Y (s) = + . we can use partial fractions to obtain 1 1 1 = − . s(s + 1) s s+1 s+1 . s+1 s(s + 1) The inverse Laplace transform of the first term is ( ) −1 2 L = 2e−t . ∫ ∞ ∫ ∞ −st L(g) = g(t)e dt = f (t − c)e−st dt. Given a function f (t). further investigation is needed to determine the inverse Laplace transform of the second term. If we take the Laplace transform of both sides of y ′ = −y + u3 (t). 0 c If we make the substitution u = t − c. s(s + 1) First.

we obtain [s2 Y (s) − sy(0) − y ′ (0)] + 2[sY (s) − y(0)] + 5Y (s) = L(h). where h(t) is given by { 5 t<7 h(t) = 0 t ≥ 7. Substituting the initial conditions and solving for Y . we have 3s −s s Y (s) = = 2 + . +1 s2 where L(y)(s) = Y (s). That is. ( ) y(t) = 2e−t + u3 (t) 1 − e−(t−3) . the table is suddenly returned to the level position. then the Laplace transform of g(t) is e−3s L(g) = e−3s L(e−t ) = . y ′′ + 2y ′ + 5y = h(t) y(0) = y ′ (0) = 0. where L(y)(s) = Y (s). s+1 Thus. Substituting the initial conditions and evaluating the Laplace trans- form on the right. where the mass is one unit. s s .2.2.4 Forced Harmonic Oscillators Consider the forced harmonic oscillator y ′′ + 4y = 3 cos t y(0) = y ′ (0) = 0. the spring constant is 5. Taking the Laplace transform of both sides of y ′′ + 2y ′ + 5y = h(t). SOLVING INITIAL VALUE PROBLEMS 303 If g(t) = u3 (t)e−(t−3) . At time t = 7. (s2 + 4)(s2 + 1) s + 4 s2 + 1 where the last expression was obtained using partial fractions. Taking the Laplace transform of both sides of the equation y ′′ + 4y = 3 cos t. We may consider this to be a mass-spring system sliding on a table. 6. Taking the inverse Laplace transform. we have our solution ( ) ( ) −s s y(t) = L−1 + L −1 = − cos 2t + cos t. When t < 7 the table is tilted so that gravity provides a force of 5 units when stretching the spring. s2 + 4 s2 + 1 Now let us consider a harmonic oscillator with discontinuous forcing. we have ( ) 1 e−7s 2 (s + 2s + 5)Y (s) = 5 − . and the damping coefficient is 2. we obtain L(y ′′ ) + 4L(y) = 3L(cos t) or 3s s2 Y (s) − sy(0) − y ′ (0) + 4Y (s) = . h(t) = 5(1 − u7 (t)).6.

we can recover the original function using the inverse Laplace transform of a function L−1 (F (s))(t). . we have 5 5e−7s Y (s) = − s(s2 + 2s + 5) s(s2 + 2s + 5) and ( ) 5 5e−7s y = L−1 − . we can rewrite the first term as 5 1 s+2 = − 2 . s(s2 + 2s + 5) s(s2 + 2s + 5) Using partial fractions. s+2 s+2 = s2 + 2s + 5 (s + 1)2 + 4 s+1 1 = + (s + 1)2 + 4 (s + 1)2 + 4 s+1 1 2 = + . s2 + 2s + 5 2 2 and ( ) ( ) ( ) −1 5 −1 1 s+2 −t 1 L 2 =L − 2 =1−e cos 2t + sin 2t . s(s2 + 2s + 5) s s + 2s + 5 The inverse Laplace transform of 1/s is 1. (s + 1)2 + 4 2 (s + 1)2 + 4 Consequently. THE LAPLACE TRANSFORM Solving for Y (s).304 CHAPTER 6. 2 2 6.5 Important Lessons • If we know the Laplace transform F (s) of a function. To find the inverse Laplace transform of the second term. ( ) ( ) −1 s+2 −t 1 −t −t 1 L =e cos 2t + e sin 2t = e cos 2t + sin 2t . s(s + 2s + 5) s s + 2s + 5 2 We can compute the inverse Laplace transform of 5e−7s s(s2 + 2s + 5) using the Heaviside function u7 (t) and the inverse Laplace transform that we just calculated to obtain ( ) ( ( )) −1 5e−7s −(t−7) 1 L = u7 (t) 1 − e cos 2(t − 7) + sin 2(t − 7) . s(s2 + 2s + 5) 2 Therefore.2. we complete the square of the denominator. the solution to our original initial value problem is ( ) ( ( )) −t 1 −(t−7) 1 y(t) = 1 − e cos 2t + sin 2t − u7 (t) 1 − e cos 2(t − 7) + sin 2(t − 7) .

then L(uc (t)f (t − c)) = e−sc F (s).3. The Laplace transform of uc is e−cs L(uc (t))(s) = . For example. • If c ≥ 0.3.2. 0 ≤ t < 5 and t ≥ 20.1 Impulse Forcing Impulse forcing is the term used to describe a very quick push or pull on a system. g(t) = u5 (t) − u20 (t) = 0. s • If L(f ) = F (s). dt2 dt where g(t) is a function that is very large in a very short time interval. consider the equation for a damped harmonic oscillator d2 y dy + 2 + 26y = g(t). Find the solution of the initial value problem 2y ′′ + y ′ + 2y = g(t) y(0) = y ′ (0) = 0 where g(t) is defined by { 1. L−1 (αF + βG) = αL−1 (F ) + βL−1 (G).6. 2. we define the Heaviside function to be { 0 t<c uc (t) = 1 t ≥ c.6 Exercises 1.3 Delta Functions and Forcing 6. DELTA FUNCTIONS AND FORCING 305 • The inverse Laplace transform is linear. say |t − t0 | < τ and zero otherwise. 5 ≤ t < 20. 6. such as the blow of a hammer or the force of an explosion. The integral ∫ t0 +τ I(τ ) = g(t) dt t0 −τ or since g(t) is zero outside of the interval |t − t0 | < τ ∫ ∞ I(τ ) = g(t) dt −∞ . Find the solution of the initial value problem 2y ′′ + y ′ + 2y = 0 y(0) = y ′ (0) = 0 using the Laplace transform. 6.

−∞ Of course. τ →0 Assuming that t0 − τ > 0. The “function” δ is an example of what is known as a generalized function. the Dirac delta function. δ(t). δ(t − t0 ) = 0. We call δ. In other words. otherwise. however. ∫ ∞ δ(t) dt = 1. Thus. We can define a unit impulse at a point t0 by considering the function δ(t − t0 ). Now let us realize the forcing function over shorter and shorter time intervals with τ getting closer and closer to zero. In particular. δ(t) has the properties δ(t) = 0. It is easy to see that I(τ ) = 1 in this case.306 CHAPTER 6. In this case. τ →0 We can use this information to define the unit impulse function. assume that t0 = 0 and { 1/2τ. More specifically. exponentially bounded function. we study no such function in calculus. THE LAPLACE TRANSFORM measures the strength or impulse of the forcing function g(t). we find that I(τ ) = 1 in all cases. t ̸= t0 . −τ < t < τ g(t) = dτ (t) = 0. the Laplace transform of dτ (t − t0 ) is ∫ ∞ L(dτ (t − t0 )) = e−st dτ (t − t0 ) dt 0 ∫ t0 +τ = e−st dτ (t − t0 ) dt t0 −τ ∫ t0 +τ 1 = e−st dt 2τ t0 −τ . but is zero for all values of t other than zero. to be the “function” that imparts an impulse of magnitude one at t = 0. we can define its Laplace transform as the limit of the Laplace transform of dτ (t) as τ → 0. ∫ ∞ δ(t − t0 ) dt = 1. lim I(τ ) = 1. assume that t0 > 0 and L(δ(t − t0 )) = lim L(dτ (t − t0 )).3.2 The Laplace Transform of the Dirac Delta Function Even though the Dirac delta function is not a piecewise continuous. −∞ 6. t ̸= 0. lim dτ (t) = 0 τ →0 for t ̸= 0.

1 −st .

.

t=t0 +τ = e .

2sτ t=t0 −τ 1 −st0 sτ = e (e − e−sτ ) 2sτ .

6. Solving for Y (s). L(δ(t − t0 )) = e−st0 . We can think of this as a damped harmonic oscillator that is struck by a hammer at time t = 4. τ →0 sτ τ →0 s Thus. s2 + 2s + 26 s2 + 2s + 26 The inverse Laplace transform of Y (s) is ( ) ( ) s+2 e−4s y=L 2 +L 2 s + 2s + 26 s + 2s + 26 ( ) ( ) s+2 1 5e−4s =L + L (s + 1)2 + 25 5 (s + 1)2 + 25 1 1 = e−t cos 5t + e−t sin 5t + u4 (t)e−(t−4) sin(5(t − 4)). We can extend this result to allow t0 = 0.3. sτ We can use l’Hospital’s rule to evaluate (sinh sτ )/sτ as τ → 0. Let Y (s) = L(y)(s) and take the Laplace transform of both sides of the differential equation to obtain s2 Y (s) − sy(0) − y ′ (0) + 2(sY (s) − y(0)) + 26Y (s) = L(δ4 )(s) or s2 Y (s) − s + 2sY (s) − 2 + 26Y (s) = e−4s . sinh sτ s cosh sτ lim = lim = 1. t0 →0 t0 →0 Let us now solve the initial value problem d2 y dy 2 + 2 + 26y = δ4 (t) dt dt y(0) = 1 y ′ (0) = 0. DELTA FUNCTIONS AND FORCING 307 sinh sτ −st0 = e . 5 5 . by lim L(δ(t − t0 )) = lim e−st0 = 1. we have s+2 e−4s Y (s) = + .

• We define the unit impulse function. but is zero for all values of t other than zero.3.308 CHAPTER 6. the Dirac delta function. consider the equation for a damped harmonic oscillator d2 y dy 2 + p + qy = g(t). δ(t) has the properties δ(t) = 0. We call δ.6: Solution to y ′′ + 2y ′ + 26y − δ4 (t) It is important to notice that we are using the Dirac delta function like an ordinary function. The integral ∫ t0 +τ I(τ ) = g(t) dt t0 −τ or since g(t) is zero outside of the interval |t − t0 | < τ ∫ ∞ I(τ ) = g(t) dt −∞ measures the strength or impulse of the forcing function g(t). 6. δ(t). . say |t − t0 | < τ and zero otherwise. such as the blow of a hammer or the force of an explosion. dt dt where g(t) is a function that is very large in a very short time interval. ∫ ∞ δ(t) dt = 1. −∞ The “function” δ is an example of what is known as a generalized function. This requires some rigorous mathematics to justify that we can actually do this. In other words. t ̸= 0.3 Important Lessons • Impulse forcing is the term used to describe a very quick push or pull on a system. to be the “function” that imparts an impulse of magnitude one at t = 0. For example. THE LAPLACE TRANSFORM Figure 6.

In this case. ∫ ∞ δ(t − t0 ) dt = 1. • Associativity: f ∗ (g ∗ h) = (f ∗ g) ∗ h.1 ConvolutionApplying the Convolution Theorem If f and g are two piecewise continuous exponentially bounded functions. we can define a unit impulse at a point t0 by considering the function δ(t − t0 ).4. by lim L(δ(t − t0 )) = lim e−st0 = 1. −∞ • The Laplace transform of the Dirac delta function is L(δ(t − t0 )) = e−st0 . CONVOLUTION 309 • Similarly. 6. • Distribution: f ∗ (g + h) = f ∗ g + f ∗ h. We can extend this result to allow t0 = 0. • Commutivity: f ∗ g = g ∗ f . 6. dt dt where g(t) is a function that is very large in a very short time interval. then we define the convolution product of f and g to be ∫ t ∫ t (f ∗ g)(t) = f (t − τ )g(τ ) dτ = f (τ )g(t − τ ) dτ.3. t ̸= 0. or d2 y dy 2 + p + qy = g(t).4 Exercises 1.6.4. t0 →0 t0 →0 • We can use the Dirac delta function to solve initial value problems such as d2 y dy 2 + 2 + 26y = δ4 (t) dt dt y(0) = 1 y ′ (0) = 0. . Find the solution of the initial value problem 2y ′′ + y ′ + 2y = δ(t − 5) y(0) = y ′ (0) = 0.4 Convolution 6. 0 0 The convolution product has many properties similar to those of ordinary multiplication. where δ(t) is the unit impulse function. δ(t − t0 ) = 0.

the inverse Laplace transform of H(s) is ∫ t at − sin at h(t) = (t − τ ) sin aτ dτ = . Using the Convolution Theorem. Reversing the order of integration. All of these properties can be proven using the definition of convolution and the properties of integration. The inverse Laplace transform of 1/s2 is t. Consider the IVP y ′′ + 4y = g(t) .310 CHAPTER 6. Then ∫ t (f ∗ 1)(t) = cos(t − τ ) · 1 dτ = sin t. 0 Also. THE LAPLACE TRANSFORM • 0 ∗ f = f ∗ 0 = 0. we have ∫ ∞ (∫ ∞ ) ∫ ∞ (∫ t ) −st −st g(τ ) e f (t − τ ) dt dτ = e f (t − τ )g(τ ) dτ dt. Then F (s)G(s) = L(f ∗ g)(s) for s > a. 0 a2 We can also use the Convolution Theorem to solve initial value problems. Proof. no multiplicative identity.7. s (s + a2 ) s s + a2 instead of using partial fractions. There is.7) to calculate the inverse Laplace transform of a 1 a H(s) = 2 2 = 2· 2 . For example. In other words. Let us use the Convolution Theorem (Theorem 6. and suppose that L(f )(s) = F (s) and L(g)(s) = G(s) for s ≥ a > 0. however. Theorem 6. suppose that f (t) = cos t. 0 τ where ξ = t − τ is the change of variable. f ∗ 1 ̸= f . Let f and g be two piecewise continuous exponentially bounded functions. it may not be the case that f ∗ f is a nonnegative function. 0 τ 0 0 However. One important property of the Laplace transform is how convolution products behave. is sin at. 0 then ∫ ∞ ∫ ∞ −sξ F (s)G(s) = e f (ξ) dξ e−sτ g(τ ) dτ 0 ∫ ∞ (∫ ∞ 0 ) −st = g(τ ) e f (t − τ ) dt dτ. this last expression is just the Laplace transform of f ∗ g. and the inverse Laplace transform of a/(s2 + a2 ). If ∫ ∞ F (s) = e−sξ f (ξ) dξ ∫0 ∞ G(s) = e−sτ g(τ ) dτ.

Taking the Laplace transform of both sides of the differential equation and applying the initial conditions. we first write 1 Ψ(s) = 2 G(s) = H(s)G(s). where (as + b)y0 + ay1 Φ(s) = as2 + bs + c G(s) Ψ(s) = 2 . where G(s) is the Laplace transform of g(t). we have 3s − 1 G(s) s 1 2 1 2 Y (s) = 2 + 2 =3 2 − + G(s). we get ∫ t 2 1 y = 3 cos 2t − sin 2t + sin 2(t − τ )g(τ ) dτ.4. Taking the Laplace transform of both sides of the differ- ential equation and using the initial conditions. y = ϕ(t) + ψ(t). we obtain s2 Y (s) − 3s + 1 + 4Y (s) = G(s). s +4 s +4 s + 4 2 s + 4 2 s2 + 4 2 The last term corresponds to the forcing term of our differential equation. as + bs + c Therefore. b. and c. 2 2 0 It is possible to write a solution for the initial value problem ay ′′ + by ′ + cy = g(t) y(0) = y0 y ′ (0) = y1 using the Convolution Theorem. as + bs + c . Observe that ϕ(t) is the solution to the initial value problem ay ′′ + by ′ + cy = 0 y(0) = y0 y ′ (0) = y1 . Taking the inverse Laplace transform of both sides and applying the Convolution Theorem. the function ϕ(t) is easy to find. Once we have values for a.6. CONVOLUTION 311 y(0) = 3 y ′ (0) = −1. To find ψ(t). we have (as2 + bs + c)Y (s) − (as + b)y0 − ay1 = G(s) or Y (s) = Φ(s) + Ψ(s). Solving for Y (s). where ϕ(t) = L−1 (Φ(s)) and ψ(t) = L−1 (Ψ(s)).

Consequently. THE LAPLACE TRANSFORM where H(s) = 1/(as2 + bs + c). g(t) = δ(t) and ψ(s) = H(s).4. however. Using convolution of functions. For this reason. find the solution to the initial value problem y ′′ + 4y ′ + 13y = g(t) y(0) = −5. 6. 0 Let us consider the case where G(s) = 1. ◦ Commutivity: f ∗ g = g ∗ f ◦ Distribution: f ∗ (g + h) = f ∗ g + f ∗ h ◦ Associativity: f ∗ (g ∗ h) = (f ∗ g) ∗ h ◦ 0∗f =f ∗0=0 There is. and g(t) is a piecewise continuous function. This means that h(t) is a solution to the initial value problem ay ′′ + by ′ + cy = δ(t) y(0) = y0 y ′ (0) = y1 .3 Exercises 1. • Let f and g be two piecewise continuous exponentially bounded functions. 6. . h(t) is sometimes called the impulse response of the system. Then F (s)G(s) = L(f ∗ g)(s) for s > a. it may not be the case that f ∗ f is a nonnegative function. • It is possible to write a solution for the initial value problem ay ′′ + by ′ + cy = g(t) y(0) = y0 y ′ (0) = y1 . 0 0 • The convolution product has many properties similar to those of ordinary multiplica- tion. y ′ (0) = 2.4. no multiplicative identity. f ∗ 1 ̸= f .2 Important Lessons • If f and g are two piecewise continuous exponentially bounded functions. using the Convolution Theorem. then we define the convolution production of f and g to be ∫ t ∫ t (f ∗ g)(t) = f (t − τ )g(τ ) dτ = f (τ )g(t − τ ) dτ. then ∫ t −1 ψ(t) = L (H(s)G(s)) = h(t − τ )g(τ ) dτ. Also. In other words. If h(t) is the inverse Laplace transform of H(s).312 CHAPTER 6. and suppose that L(f )(s) = F (s) and L(g)(s) = G(s) for s ≥ a > 0.

We can modify the logistic growth model to account for a threshold population.9 Exercises 1. 4. That is. the species is endanger of extinction. The black rhino.000 around 1900. Verify that y(t) = 3e−2t is a solution to the differential equation y ′ = −2y y ′ = −6e−2t = −2(3e−2t ) = −2y 2. (a) For what values of P is the rhino population increasing? What can be said about the value of dP /dt for these values of P ? (b) For what values of P is the rhino population decreasing? What can be said about the value of dP /dt for these values of P ? (c) For what values of P is the rhino population in equilibrium? What can be said about the value of dP /dt for these values of P ? (d) Find a differential equation that models the population of rhinos at time t. We can determine that C = 3. y(t) = 3e−t is a solution to the initial value problem. where P changes at a rate proportional to the square root of P . Since y ′ + y = −3e−t + 3e−t = 0. (a) The population is increasing if dP /dt > 0 and 1000 < P < 20000. if the population of a species drops below a certain level. once the most numerous of all rhinoceros species and now critically endangered.000 rhinos. competing species. native to eastern and southern Africa. habitat changes. Suppose the this minimum or threshold population for the black rhino is 1000 animals and that remaining habitant in Africa will support no more that 20. Verify that y(t) = Ce−t is a solution to the differential equation y ′ + y = 0. Because of hunting. and most of all illegal poaching. √ dP /dt = k P 5. 313 . the number of black rhinos today is estimated to be below 3000. Write a differential equation that models a population P . Consider the case of the black rhinoceros. Determine C if y(0) = 3. the animals may not be able to find suitible mates and the black rhino will become extinct. was estimated to have a population of about 100. There must be a minimum population for the species to continue.1. If the wild population becomes too low. A Hints and Solutions to Selected Exercises 1.

2. where c is any constant and t0 ∈ I. 8. Given the equation x′ + px = q(t).1 using Sage. Consider the following predator-prey systems of differential equations dx x xy =− + . Solve each of the following differential equations. Solve each of the following equations in Exercise 1. (c) x = 3/(3 − t3 ). dy (a) = 8 − 2x dx dy x+1 (b) = dx y dx (c) = x2 t2 . Rewriting the differential equation as x′ + px − q(t) = 0 and using the fact that ∫ t ′ −at −at x (t) = −aCe − ae eas b(s) ds + b(t).2.8 Exercises 1. 1. t0 we see that ∫ t ′ −at −at x (t) + ax(t) − b(t) = −aCe − ae eas b(s) ds + b(t) t0 ∫ t −at −at + aCe + ae eas b(s) ds − b(t) t0 = 0. show that ∫ t −at −at x(t) = Ce +e eas b(s) ds t0 is a solution of this equation. where p is a constant and q(t) is a continuous function defined on an interval I. . y(0) = −1 y+1 (a) y = −x2 + 8x + C. x(0) = 1 dt t2 y − y (d) y ′ = . HINTS AND SOLUTIONS TO SELECTED EXERCISES 6. 2.314 APPENDIX A.8. dt 2 2+y dy xy = y(1 − y) − . dt 2+y (a) Which equation models the prey population and which equation models the predator population? (b) How does the prey population grow if there are no predators present? (c) What happens if there are a lot of prey present? Think about the limit of the interaction term as the number of prey becomes very large.

source. where the graph of f (y) is given below. and {ai }ki=0 is a sequence satisfying t a0 ≥ − s .4. the error estimates. (b) Use Euler’s method with step size h = 0. 315 (a) x = var ( ' x ' ) y = function ( ' y ' )(x) de = diff (y . 2. In this series of exercises. 3. What happens if we increase the harvest rate to 100 in Example 1.? What should be our strategy to maintain a viable population in the trout pond? 1. and the actual error. (c) Sketch the graph of the approximate and exact solutions.33). Your solution should include a table of approximate values of the dependent variable the exact values of the dependent variable. Make sure that your approximations are accurate to four decimal places.3. (b) Use part (1) and geometric series to prove the following statement: If s and t are positive real numbers.7 Exercises 2.5 to approximate the solution to the initial value problem on the interval [0. (d) Use the error bound theorem (Theorem 1. (a) Use Taylor’s Theorem to show that for all x ≥ −1 and any positive m.5 Exercises 1. (a) Find the exact solution of the initial value problem. or node. 2] Your solution should include a table of approximate values of the dependent variable as well as the exact values of the dependent variable. Make sure that your approximations are accurate to four decimal places.33) to estimate the error at each approxi- mation. Consider the differential equation y ′ = f (y). we will prove the error bound theorem for Euler’s method (Theorem 1. Identify any equilibrium solutions and classify each equilibrium solution as a sink.24.2* x desolve (de . 0 ≤ (1 + x)m ≤ emx . x) == 8 . Consider the differential equation dy = 3y − 1 dt with initial value y(0) = 2. y) -x ^2 + _C + 8* x 1.

316 APPENDIX A. HINTS AND SOLUTIONS TO SELECTED EXERCISES

ai+1 ≤ (1 + s)ai + t, for i = 0, 1, 2, . . . , k,

then ( )
t t
ai+1 ≤ e(i+1)s + a0 − .
s s

(c) When i = 0, y(t0 ) = Y0 = α and the theorem is true. Use Euler’s method and Taylor’s
Theorem to show that

h2 ′′
|yi+1 − Yi+1 | ≤ |yi − Yi | + h|f (ti , yi ) − f (ti , Yi )| + |y (ξi )|
2

where ξi ∈ (ti , ti+1 ) and yk = y(tk ).

(d) Show that
M h2
|yi+1 − Yi+1 | ≤ |yi − Yi |(1 + hL) + .
2

(e) Let aj = |yj − Yj | for j = 0, 1, . . . , N , s = hL, and t = M h2 /2 and apply part (2) to
show ( )
M h2 M h2
|yi+1 − Yi+1 | ≤ e (1+i)hL
|y0 − Y0 | + − .
2hL 2hL
and derive that
M h ( (ti+1 −a)L )
|yi+1 − Yi+1 | ≤ e −1
2L
for each i = 0, 1, . . . , N − 1.

Hints for part (2):

• For fixed i show that

ai+1 ≤ (1 + s)ai + t
≤ (1 + s)[(1 + s)ai−1 + t] + t
≤ (1 + s){(1 + s)[(1 + s)ai−2 + t] + t} + t
..
.
≤ (1 + s)i+1 a0 + [1 + (1 + s) + (1 + s)2 + · · · + (1 + s)i ]t.

• Now use a geometric sum to show that
( )
t t t
ai+1 ≤ (1 + s) i+1
a0 + [(1 + s)i+1 − 1] = (1 + s)i+1 + a0 − .
s s s

• Apply part (1) to derive
( )
t t
ai+1 ≤ e (i+1)s
+ a0 − .
s s

317

1.5.7 Exercises
2. Solve the initial value problem

2y
y′ = − + 2x
x+1
y(0) = 2.

3x4 + 8x3 + 6x2 + 12
y=
6(x + 1)2
3. Solve the initial value problem

2y
y′ = − + ex
1+x
y(0) = 1.

y = (ex + 1)/(x + 1)2
4. Solve the initial value problem
y cos x
y′ = − +
x+2 x+2
y(0) = 1.

y = sin x/(x + 2)
5. Solve the initial value problem

ex
y ′ = y tan x +
cos x
y(0) = 1.

y = ex sec x
6. A 600-gallon tank initially contains 200 liters of water containing 10 kilograms of salt.
Supposed that water containing 0.1 kilograms of salt flows into the top of the tank at a rate
of 10 liters per minute. The water in the tank is kept well mixed, and 5 liters per minute
are removed from the bottom of the tank. How much salt is in the tank when the tank is
full?
If x(t) is the amount of salt in the tank at time t, we know that x(0) = 10. The volume
of the tank is V = 600 + 5t. We can model the amount of salt in the tank at time t with a
differential equation,

dx
= rate in − rate out
dt
x
= 10(0.1) − 5
V
x
=1−5
600 + 5t
x
=1− .
120 + t
The resulting equation
dx 1
+ x=1
dt 120 + t

318 APPENDIX A. HINTS AND SOLUTIONS TO SELECTED EXERCISES

is a first order linear differential equation. An integrating factor for this equation is given
by (∫ )
1
µ(t) = exp dt = 120 + t.
120 + t
Multiplying both sides of the differential equation by µ(t), we have
( )
d dx dx 1
[(120 + t)x] = (120 + t) + x = (120 + t) + x = 120 + t.
dt dt dt 120 + t
Integrating both sides of this equation, we obtain

t2
(120 + t)x = 120t + + C.
2
Using the intial condition x(0) = 10, we can determine that C = 1200 or

t2 + 240t + 2400
x(t) = .
2t + 240
The tank is full at time t = 400/5 = 80, and the tank contains x(80) = 70 kilograms of salt
when the tank is full.
7. A manager in a communications company contributes $2400 per year into her retirement
fund by making many small deposits throughout the year. The fund grows at a rate of
3.5% per year compounded continuously. After 35 years, she retires and begins and begins
withdrawing from the retirement fund at a rate of $3500 per month. If she does not make
any deposits after she retires, how long will her retirement fund last? [Hint: Divide the
problem into two smaller problems—one that deals with the situation before retirement and
one that deals with the problem after retirement.]
8. Lake Baikal, located in southern Siberia, is only the seventh largest lake in the world in
terms of surface area; however, it is the world’s deepest lake. The lake has a depth of 1,642
meters, and the bottom lies 1,186.5 meters below sea level. Lake Baikal has a volume of
23,600 cubic kilometers and contains 20% of the world’s unfrozen fresh water. In contrast,
Lake Superior, the largest of the Great Lakes, has a volume of only 12,100 cubic kilometers.
Although 544 rivers flow into Lake Baikal, there is only a single outlet, the Angara River.
The outflow of the lake is pretty constant at 60.4 cubic kilometers per year.
Pollution is an increasing concern in Lake Baikal. One of the major polluters has been
Baykalsk Pulp and Paper Mill. The mill was constructed in 1966 on the shoreline of Lake
Baikal and regularly discharged waste into the lake. The plant was closed in November
2008 due to unprofitability, but production resumed in January 2010. The mill underwent
a final bankruptcy in September 2013, but the longterm fate of the mill is still undecided.1
Suppose that we wish to understand how the pollution level changes in Lake Baikal over
a period of years. Hypothetically, let us assume that the Baykalsk Pulp and Paper Mill
has been responsible for most of the pollution in Lake Baikal for the last several decades.
Suppose that at t = 0 years the mill ceases operation and there are no longer any pollutants
discharged into the lake from the mill although there are still other sources of pollution.
Let us assume that the lake is currently 6 times more polluted than these other sources of
contaminants. We wish to know how long it will take for the pollution level to reduce to
half of its current level of lake. Lake Baikal’s waters are well-mixed and well-oxygenated in
spite of its great depth, so we can model this situation as a simple mixing problem.
1
Levy, Clifford J. (November 8, 2010). “Last Gasp for Factory Bequeathed by Soviets.” The New
York Times. Retrieved March 14, 2014 from http://www.nytimes.com/2010/11/09/world/europe/
09baikal.html.

319

(a) The volume of water in Lake Baikal is V = 23, 600 km3 , and

rin = rout = r = 60 km3 /year

be the rates of inflow from the numerous rivers that feed the lake and outflow to the
Angara River. Assume that C = Cin is the pollutant concentration flowing into Lake
Baikal and Cout is the concentration of the outflow into the Angara River (measured
in metric tons per cubic km). If x(t) is the amount of solute at time t in Lake Baikal,
then
x0 = x(0) = 6CV
is the initial amount of solute in the lake. Estimate ∆x during the time interval
[t, t + ∆t], where ∆t > 0 is small.

(b) From your estimate of ∆x in part (1), write an initial value problem that describes
the amount of pollutants in the lake at time t.

(c) The equation that you found in part (2) is a first-order linear equation. Solve this
equation

(d) Using part (3), predict how many years it will take to reduce the pollution in Lake
Baikal to half of its current level.

9. Variation of Parameters. Consider the following method of solving the general linear
equation of the first order,
y ′ + p(t)y = g(t). (A.1)

(a) If g(t) is identically zero, show that the solution is
[ ∫ ]
y = A exp − p(t) dt ,

where A is a constant.

(b) If g(t) is not identically zero, assume that the solution is of the form
[ ∫ ]
y = A(t) exp − p(t) dt , (A.2)

where A is now a function of t. By substituting for y in the given differential equation
(A.1), show that A(t) must satisfy the condition
[∫ ]

A (t) = g(t) exp p(t) dt . (A.3)

(c) Find A(t) from Equation (A.3). Then substitute for A(t) in equation (A.2) and
determine y. Verify that the solution obtain in this manner agrees with the solution
given in the proof of Theorem 1.39. That is, show that this solution is equivalent to
the solution (∫ )
1
y= µ(t)g(t) dt + C ,
µ(t)
where (∫ )
µ(t) = exp p(t) dt .

320 APPENDIX A. HINTS AND SOLUTIONS TO SELECTED EXERCISES

This technique is know as variation of parameters, which we will revisit when we study
second order linear differential equations.
10. Bernoulli’s equation is an equation of the form

y ′ = a(t)y + f (t)y n ,

where n ̸= 0 or 1. Bernoulli’s equation is nonlinear and cannot be solved by the techniques
that we have used to solve first order linear equations.2

(a) Using the substitution z = y 1−n , show that we can transform Bernoulli’s equation
into the linear equation

z ′ = (1 − n)a(t)z + (1 − n)f (t).

(b) Solve the equation xy ′ + y = x4 y 3 .

11. The first-order nonlinear differential equation

y ′ = p(t) + q(t)y + r(t)y 2 (A.4)

is known as the Ricatti equation and has some useful applications in control theory. If one
solution y1 (t) of the Ricatti equation is known, then a more general solution containing an
arbitrary constant can be found by substituting y = y1 (t) + 1/v(t) into equation (A.4) to
find a first-order linear equation in v and t, which we can solve to find a general solution
to the Ricatti equation.

(a) Show that this first-order linear equation is v ′ + [q(t) + 2r(t)y1 (t)]v = −r(t).

(b) Find the solution to the Ricatti equation

1 1
y′ = − − y + y2
t2 t
given the particular solution y1 = 1/t.

(c) Find the solution to the Ricatti equation

y ′ = cos t − y tan t + y 2 sec t

given the particular solution y1 = sin t.

(d) Find the solution to the Ricatti equation

y ′ = 2 − 3y + y 2

given the particular solution y1 ≡ 2.

(a) If y = y1 + 1/v, then y ′ = y1′ − v ′ /v 2 . Substituting into our original equation, we
obtain
v′ v′
y ′ = y1′ − 2 = p + qy1 + ry12 − 2 .
v v
2
Polking p. 63

321

On the other hand,
( ) ( )
′ 1 1 2
y = p + q y1 + + r y1 +
v v
q 2ry 1 r
= p + qy1 + + ry12 + + 2
v v v
q 2ry r
= y1′ + +
1
+ 2.
v v v
Therefore,
v′ q 2ry1 r
− 2
= + + 2,
v v v v
which is just the first-order linear equation

v ′ + [q(t) + 2r(t)y1 (t)]v = −r(t).

(b)
1
y =t+
C −t

(c)
1
y(t) = + sin t
C cos t − sin t

(d)
1
y(t) = 2 +
Cet−1

12. Suppose that we have a population that not only grows logistically but also requires a
minimum threshold population to survive. For example, the case of the North Pacific right
whale, a species now very much on the endangered list. If the population drops too low,
whales might not be able to find suitable mates and the species will eventually go extinct.
In other words, the population will die out if it drops below a certain threshold. We can
model this with the following equation,
( )
dP P
=k 1− (P − aN ), (A.5)
dt N

where P is the population of the whales at time t and N is the carrying capacity. The
constants k and a are positive with a < 1.

(a) Find the equilibrium solutions of the this equation.

(b) Since equation (A.5) is autonomous, we can find a solution using separation of vari-
ables. Find this solution.

(c) Equation (A.5) is also a Ricatti equation (A.4). Since we know an equilibrium solution
from part (1), we can use the method of the previous problem to find a general solution
to (A.5). Find the general solution using the fact that we have a Ricatti equation and
show that your solution agrees with the solution that you found in part (2)

322 APPENDIX A. HINTS AND SOLUTIONS TO SELECTED EXERCISES

1.6.4 Exercises
1. Which of the following initial value problems are guaranteed to have a unique solution
by the Existence and Uniqueness Theorem (Theorem 1.40)? In each case, justify your
conclusion.

(a) y ′ = 4 + y 3 , y(0) = 1

(b) y ′ = y, y(1) = 0

(c) y ′ = y, y(1) = 1
t
(d) x′ = , x(0) = 2
x−2
t
(e) x′ = , x(2) = 0
x−2
(f) y ′ = x tan y, y(0) = 0
1
(g) y ′ = y + 2t, y(0) = 1
t

(a) There exists a unique solution to y ′ = 4 + y 3 , y(0) = 1, since f (t, y) = 4 + y 3 and
∂f (t, y)/∂y = 3y 2 are continuous at the point (0, 1).

(b) The Existence and Uniqueness Theorem does not apply to y ′ = y, y(1) = 0, since

f (t, y) = y is not continuous at (1, 0).
√ √
(c) There exists a unique solution to y ′ = y, y(1) = 1, since f (t, y) = y and

∂f (t, y)/∂y = 1/(2 y ) are both continuous at the point (1, 1).

(d) The Existence and Uniqueness Theorem does not apply to x′ = t/(x − 2), x(0) = 2,
since f (t, x) = t/(x − 2) is not continuous at (0, 2).

(e) There exists a unique solution to x′ = t/(x − 2), x(2) = 0, since f (t, x) = t/(x − 2)
and ∂f (t, x)/∂x = −t/(x − 2)2 are both continuous at the point (2, 0).

(f) There exists a unique solution to y ′ = x tan y, y(0) = 0, since f (x, y) = x tan y and
∂f (x, y)/∂y = x sec2 y are both continuous at the point (0, 0).

(g) The Existence and Uniqueness Theorem does not apply to y ′ = y/t + 2t, y(0) = 1,
since f (t, y) = y/t + 2t is not continuous at (0, 1).

2. Find an explicit solution to the initial value problem
1
y′ =
(t − 1)(y + 1)
y(0) = 1.

Use your solution to determine the interval of existence.
3. Consider the initial value problem

y ′ = 3y 2/3
y(0) = 0.

Let ϕn (x) = xn for 0 ≤ x ≤ 1 and show that { 0. x = 1. t0 7. ] (c) Explain why this example does not contradict the Existence and Uniqueness Theorem. there exists an infinite number of solutions to the initial value problem. Use the Fundamental Theorem of Calculus to show that the function u = u(t) is a solution to the initial value problem x′ = f (t. 0 ≤ |x − x0 | < b} . Consider the initial value problem y ′ = 2ty + t y(0) = 1. which helps explain the need for uniform continuity in the proof of the Existence and Uniqueness Theorem. (c) Show that the sequence {ϕn (t)} converges to the exact solution that you found in part (a) as n → ∞. where t0 is any real number. Hence. 6. 5. x) : 0 ≤ |t − t0 | < a. (b) Show that { 0. x) x(t0 ) = x0 . y(t) ≡ 0. prove the Existence and Uniqueness Theorem for first-order differential equations. if and only if u is a solution to the integral equation ∫ t x(t) = x0 + f (s. 4. If ∂f /∂x is continuous on the rectangle R = {(t. The following series of exercises. (a) Use the fact that y ′ = 2ty + t is a first-order linear differential equation to find a solution to the initial value problem. This is an example of a sequence of continuous functions that does not converge to a continuous function. x1 ) − f (t. (b) Let ϕ0 (t) = 1 and use Picard iteration to find ϕn (t). prove that there exists a K > 0 such that |f (t. 0 ≤ x < 1 lim ϕn (x) = n→∞ 1. t ≤ t0 y(t) = (t − t0 )3 . 323 (a) Show that the constant function. is a solution to the initial value problem. [Hint: Make sure that the derivative of y(t) exists at t = t0 . x2 )| ≤ K|x1 − x2 | . t > t0 is a solution for the initial value problem. x(s)) ds.

∫ t un+1 = x0 + f (s. 2 11. .324 APPENDIX A. x0 ) and the boundary of the rectangle R. x(s)) ds. we know that un (t) converges to a continuous function u(t) that solves our equation. t0 14. then our sequence will converge to a function u(t). Since un (t) = u1 (t) + [u2 (t) − u1 (t)] + · · · + [un (t) − un−1 (t)]. n = 1. we can view un (t) as a partial sum for the series ∞ ∑ u0 (t) + [un (t) − un−1 (t)]. un−1 (t))| ≤ K|un (t) − un−1 (t)|. . 10. x1 ) and (t. . Use mathematical induction to show that K n−1 M |t − t0 |n |un (t) − un−1 (t)| ≤ . t0 Use the result of the previous exercise to show that |f (t. 9. 2. n=1 If we can show that this series converges absolutely. HINTS AND SOLUTIONS TO SELECTED EXERCISES for all (t. 3 We must a theorem from advanced calculus here to ensure uniform continuity. must converge to a continuous function. . Any sequence of functions that converges uniformly. K n! K n=1 n=1 where h is the maximum distance between (t0 . x2 ) in R. . 8. Since |un (t) − un−1 (t)| → 0. Show that there exists an M > 0 such that |u1 (t) − x0 | ≤ M |t − t0 |. Show that KM |t − t0 |2 |u2 (t) − u1 (t)| ≤ . un (s)) ds. To show uniqueness. n! 12. Define the sequence {un } by u0 (t) = x0 . t0 Show that ∫ t |u(t) − v(t)| ≤ K |u(s) − v(s)| ds. Show that M ( K|h| ) ∞ ∑ ∞ M ∑ (K|t − t0 |)n |un (t) − un−1 (t)| ≤ ≤ e −1 . un (t)) − f (t. assume that u(t) and v(t) are both solutions to ∫ t x(t) = x0 + f (s.3 13.

and K = 1. One explanation that has been given for the occurrence of outbreaks is based on the multiple bifurcations that occur with this differential equation. (b) Since ∫ t |u(t) − v(t)| − K |u(s) − v(s)| ds ≤ 0. x2 x′ = rx(1 − x) − .4 Exercises 1. 0. t0 then ϕ(t0 ) = 0 and ϕ(t) ≥ 0 for t ≥ t0 . c = 1.01 + x2 Solve the equation equation x2 rx(1 − x) − =0 0. Outbreaks of the spruce budworm have been responsible for some major deforestations in Canada and the United States. (b) y ′ = λy 2 − 1 for λ ∈ R. we have a family of differential equations parameterized by r. Show that ϕ′ (t) = |u(t) − v(t)|. Describe the phase line portraits for each of the following equations and how they depend on the parameter λ. 4 A similar argument will work for t ≤ t0 . (a) If a = 0. . (a) y ′ = λy − sin y for λ > 2/π.01 + x2 and plot the result in the xr-plane for 0 ≤ r < 1. where the variable x denotes the population or density of the insect [16]. t0 we know that ϕ′ (t) − Kϕ(t) ≤ 0. 2. Use this fact to show that d [ −Kt ] e ϕ(t) ≤ 0. dt Conclude that ∫ t ϕ(t) = |u(s) − v(s)| ds = 0 t0 for t ≥ t0 or for all t ≥ t0 and u(t) = v(t).7.01. 325 (a) Define4 ∫ t ϕ(t) = |u(s) − v(s)| ds. 1. The equation ( x) x2 x′ = r 1 − x−c K a + x2 has been used to describe the dynamics of spruce budworm populations. Draw the bifurcation diagram for each equation.

How would you modify the system to include this assumption? . (c) Find the solutions of the original differential equation that correspond to the equi- librium solutions of u′ = f (u). Consider the following predator-prey system.1. x′ = 2x(1 − x/2) − 1. HINTS AND SOLUTIONS TO SELECTED EXERCISES (b) To find the bifurcation diagram for the spruce budworm equation. or linear.326 APPENDIX A. (b) Sketch the phase line for this new equation. That is. The differential equation dy = y − 4t + y 2 − 8yt + 16t2 + 4. however. Explain what happens to the population as r increases.2xy y ′ = −y + 0.5 Exercises 1. when does an outbreak occur? What happens after an outbreak? 3. unlimited source of food. but they still prefer to eat the primary prey when they can catch them. (a) Transform this equation into a new differential equation of the form du = f (u) dt by letting u = y − 4t. (d) Solve the differential new equation and use this information to solve the original differential equation. (a) How would you modify this system to include the effect of hunting of the prey at a rate of α units of prey per unit of time? (b) How would you modify this system to include the effect of hunting of the predators at a rate proportional to the number of predators? (c) Suppose the predators discover a second. where x(t) is the population of the prey and y(t) is the population of the predator at time t. dt is not autonomous. sketch the graphs of the solutions that you plotted in part (b). u′ = f (u). and sketch several solutions. reflect the graph obtained in part (1) about the line r = x line. separable. How would you modify the system to include this assumption? (d) Suppose the predators discover a second source of food that is limited in supply. Graph these solutions in ty-plane. (c) Estimate the two bifurcation values from your graph. 2. we can solve this equation with a change of variable. Also.9xy.

dt where the species interact weakly. (c) Using the result of the previous exercise. find the solution of this initial-value problem. Under what conditions on β is y(t) a solution to the differential equation? (b) What initial condition (t = 0) in the yv-plane corresponds to this solution? (c) In terms of k and m. SEIR model 2. Consider the competing species model dx = x(1 − x) − βxy dt dy = y(1 − y) − βxy. Under what conditions on β is y(t) a solution to the differential equation? (b) Formulate an initial-value problem that corresponds to the motion of this undamped mass-spring system if the mass is extended 1 foot from its rest position and released with no initial velocity. 327 (e) Suppose that the predators migrate to an area if there is more than five times as many prey in that area (x > 5y). 4.4 and y(0) = 0. 3. How would you modify the system to take this into account? 2. A mass weighing 4 pounds stretches a spring 4 inches. (a) Consider the function y(t) = cos βt. How would you modify the system to take this into account? (f) Suppose that prey move out of an area at a rate proportional to the number of predators in the area. and that they move away if there are fewer than five times as many prey as predators.6 Exercises 1. (b) Find all equilibrium points for this system. what is the period of this solution? (d) Sketch the solution curve in the yv-plane that corresponds to this solution. what happens to the two populations as t → ∞? . (a) Consider the function y(t) = cos βt.2.3. (c) Sketch the phase plane for this system. (a) Find the x and y-nullclines for this system. Consider the equation d2 k + y=0 dt2 m for the motion of a simple harmonic oscillator. (d) If the inital populations are given by x(0) = 2. say β = 1/2.

6) dt dI = αSI − 0.0003 per individual per year. • On the average. but susceptible individuals have a birth rate of 0.7) dt The constant α is a measure of the relative infectivity of the disease.000001. Some diseases such as Ebola. Consider an epidemic that moves through an isolated population. • Infected individuals do not give birth. (a) x′ = cos y y ′ = −x + y (b) x′ = y y ′ = x − x3 − y 3. • Only a single individual is initially ill. (A. AIDS which is caused by the HIV virus. and infectious individuals are im- mediately infectious. We assume that the constant of proportionality is α. We will make the following assumptions about the epidemic. What happens to the solution curve as t → ∞? (b) If α = 0. an infected individual dies after 10 days. • Individuals are infected at a rate proportional to the product of the number of infected and susceptible individuals. • Sketch the phase plane and direction field for each system using technology. Suppose that S(0) = 1000 and I(0) = 1. draw the phase portrait. Suppose that S(0) = 30000 and I(0) = 1. Be sure to label all nullclines and equilibrium solutions. • The length of the incubation period is negligible.05. For the following two systems of equations • Find the equilibrium points of the system. • Briefly describe the behavior of typical solutions. Be sure to label all nullclines and equilibrium solutions. On the other hand. a viral hemorrhagic fever.1I.328 APPENDIX A. If S(t) is the number of susceptible and I(t) is the number of infected people. (a) If α = 0. Newborns are susceptible.0003S (A. The goal of this exercise is to examine the differences between the two. HINTS AND SOLUTIONS TO SELECTED EXERCISES 2. are extremely infectious with a mortaility rate of up to 90%. draw the phase portrait. then dS = −αSI + 0. What happens to the solution curve as t → ∞? (c) What conclusions can you draw about the behavior of the two different epidemics? . has a much lower transmission rate.

5 Exercises 1. (b) Use Euler’s method with step size ∆t = 0. (a) A + B (b) 2A − 3B (c) AB (d) BA (e) A−1 (f) B −1 2.5 (a) Show that x(t) = (e2t . If ( ) ( ) 2 1−i 4i 1−i A= and B= . and check how close the approximation is to the real solution when t = 2. (a) A + B (b) 3A − 2B (c) AB 5 You will find Sage very useful for this exercise . 3et ) satisfies the initial value problem. 3.5 to approximate this solution. (d) Discuss how and why the Euler approximations differ from the real solution.3. t = 4. This is an exercise.6 Exercises 1. and check how close the approximation is to the real solution when t = 2. (c) Use Euler’s method with step size ∆t = 0. 2. 2−i 2 1 + 3i −2 − i find each of the following. 3 1 −1 −2 find each of the following. 329 2. If ( ) ( ) 1 −2 4 1 A= and B = . and t = 6. Consider the system x′ = 2x y′ = y with initial conditions x(0) = 1 and y(0) = 3.5 Exercises 1. and t = 6.1 to approximate this solution.1.4. t = 4.

5 Exercises 5.x = . Prove that x and y are linearly independent if and only if ( ) x 1 y1 det ̸ 0. If ( ) ( ) ( ) 3 −2 4 −2 A= . (a) Ax (b) Ay (c) xT y (d) yT x 4. (a) Find the general solution of the homogeneous system x′ = x + 3y y′ = x − y (b) Find a particular solution for x′ = x + 3y + (t − 3t2 ) y ′ = x − y + (1 + t − t2 ) (c) Find the solution of x′ = x + 3y + (t − 3t2 ) y ′ = x − y + (1 + t − t2 ) x(0) = 1 y(0) = −1. Assume that your solution must be of the form ( 2 ) a2 t + a1 t + a0 xp = b2 t2 + b1 t + b0 . Let x = (x1 . x2 ) and y = (y1 . . y2 ).330 APPENDIX A.2. Consider the linear system x′ = x + 3y + (t − 3t2 ) y ′ = x − y + (2 − t + t2 ) x(0) = 1 y(0) = −1. This is called the method of undetermined coefficients. HINTS AND SOLUTIONS TO SELECTED EXERCISES (d) BA 3. = x 2 y2 3. 0 −1 1 3 find each of the following. and y = .

3. Sketch the phase plane of each system and determine the nature of the equilibrium solution(s).5.4.5 Exercises 1. then the solutions of x′ = Ax travel around the origin in a counterclockwise direction.4 Exercises 1. Solve each of the following linear systems for the given initial values. if a 2×2 matrix A has complex eigenvalues. (a) x′ = 2x + 2y y ′ = −4x + 6y x(0) = 2 y(0) = −3 (b) x′ = 2x − 6y y ′ = 2x + y x(0) = 2 y(0) = 1 (c) x′ = 2x + y y ′ = −x + 4y x(0) = 1 y(0) = −1 (d) x′ = 4x + 2y y ′ = −2x − y x(0) = 1 y(0) = −1 2. 331 3.5 Exercises 1. Do this first without the use of technology. Determine a computable condition that guarantees that. This is an exercise.3. .7 Exercises 1. 3. 3. This is an exercise. then check your answer with Sage.6.

Thus. Consider the one-parameter family of linear systems given by ( ′) ( √ )( ) x a 2 + a/2 x ′ = √ . (b) Assume that y = v(t)x1 (t) = v(t)e−bt/2a is a solution to ax′′ + bx′ + cx = 0 and show that v(t) = c1 + c2 t. 2.8. a sink. and so on. Consider the two-parameter family of linear systems ( ′) ( )( ) x a b x ′ = . Find the general solution of the system x′1 = 6x1 + 9x2 + x3 + 4x4 x′2 = −x1 + x2 + x3 x′3 = 3x1 + 8x2 + 4x3 + 4x4 x′4 = −3x1 − 8x2 − 7x3 − 5x4 3. y 2 − a/2 0 y (a) Sketch the path traced out by this family of linear systems in the trace-determinant plane as a varies. Let ax′′ + bx′ + cx = 0. HINTS AND SOLUTIONS TO SELECTED EXERCISES 3.4 Exercises 1.332 APPENDIX A. y b a y Identify all of the regions in the ab-plane where this system possesses a saddle. (a) Show that x1 (t) = e−bt/2a is a solution to ax′′ + bx′ + cx = 0.5 Exercises 1. x(t) = c1 e−bt/2a + c2 te−bt/2a is a general solution for ax′′ + bx′ + cx = 0. where a ̸= 0 and b2 − 4ac = 0. .7. a spiral sink. Solve the system x′ = −5x + 15y − 4z y ′ = −x + 5y − z z ′ = 4x − 6y + 3z x(0) = −1 y(0) = 2 z(0) = 0 2. (b) Discuss any bifurcations that occur along this path and compute the corresponding values of a. 3.

Reduction of Order. Also. 333 4.9) for t > 0 and find a second linearly independent solution to ((A. (A. if any. . (a) y ′′ − 2y ′ + 5y = 0 y(π/2) = 0 y ′ (π/2) = 2 (b) 9y ′′ − 12y ′ + 4y = 0 y(0) = 2 y ′ (0) = −1 (c) y ′′ + 8y ′ − 9y = 0 y(1) = 1 y ′ (1) = 0 (d) y ′′ + 2ay ′ + (a2 + 1)y = 0 y(0) = 1 y ′ (0) = 0 6.11)). Suppose that x1 (t) is a solution (not identically zero) to the equation x′′ + p(t)x′ + q(t)x = 0. Determine all values of α. (c) Show that x1 (t) = 1/t is a solution to 2t2 x′′ + 3tx′ − x = 0 (A. if any.10)) is a first-order linear differential equation in u. determine the values of α. Consider the equation y ′′ − (2α − 1)y ′ + α(α − 1)y = 0. for which all solutions tend toward zero as t → ∞.8) (b) Let u = v ′ and show that ((A. (a) Assume that x(t) = v(t)x1 (t) is a solution to x′′ + p(t)x′ + q(t)x = 0 and derive the equation x1 v ′′ + (2x′1 + px1 )v ′ = 0. for which all nonzero solutions become unbounded as t → ∞. Solve each of the following initial value problems. 5.

Find the general solution of the system x′1 = 6x1 + 9x2 + x3 + 4x4 x′2 = −x1 + x2 + x3 x′3 = 3x1 + 8x2 + 4x3 + 4x4 x′4 = −3x1 − 8x2 − 7x3 − 5x4 3. HINTS AND SOLUTIONS TO SELECTED EXERCISES 7. Euler Equations. (a) Show that x1 (t) = e−bt/2a is a solution to ax′′ + bx′ + cx = 0. Thus. Hint: Show that dy dx dy = dt dt dx 2 ( )2 2 d y dx d y d2 x dy = + 2 . Let ax′′ + bx′ + cx = 0. where a ̸= 0 and b2 − 4ac = 0. . where t > 0 and α and β are real constants. Solve the system x′ = −5x + 15y − 4z y ′ = −x + 5y − z z ′ = 4x − 6y + 3z x(0) = −1 y(0) = 2 z(0) = 0 2. dt2 dt dx2 dt dx (b) Solve the equation t2 y ′′ + 4ty ′ + 2y = 0.334 APPENDIX A. (b) Assume that y = v(t)x1 (t) = v(t)e−bt/2a is a solution to ax′′ + bx′ + cx = 0 and show that v(t) = c1 + c2 t.9. x(t) = c1 e−bt/2a + c2 te−bt/2a is a general solution for ax′′ + bx′ + cx = 0. 3.4 Exercises 1. (a) Show that the substitution x = ln t transforms an Euler equation into an equation of constant coefficients. An equation of this form is called an Euler equation. An important class of second-order linear differential equations are equations of the form t2 y ′′ + αty ′ + βy = 0.

Determine all values of α. Suppose that x1 (t) is a solution (not identically zero) to the equation x′′ + p(t)x′ + q(t)x = 0. (A. Also.10) (b) Let u = v ′ and show that ((A. Consider the equation y ′′ − (2α − 1)y ′ + α(α − 1)y = 0. (a) Assume that x(t) = v(t)x1 (t) is a solution to x′′ + p(t)x′ + q(t)x = 0 and derive the equation x1 v ′′ + (2x′1 + px1 )v ′ = 0.11) for t > 0 and find a second linearly independent solution to ((A. Reduction of Order. . determine the values of α. for which all solutions tend toward zero as t → ∞.11)). (a) y ′′ − 2y ′ + 5y = 0 y(π/2) = 0 y ′ (π/2) = 2 (b) 9y ′′ − 12y ′ + 4y = 0 y(0) = 2 y ′ (0) = −1 (c) y ′′ + 8y ′ − 9y = 0 y(1) = 1 y ′ (1) = 0 (d) y ′′ + 2ay ′ + (a2 + 1)y = 0 y(0) = 1 y ′ (0) = 0 6. (c) Show that x1 (t) = 1/t is a solution to 2t2 x′′ + 3tx′ − x = 0 (A. if any. Solve each of the following initial value problems. for which all nonzero solutions become unbounded as t → ∞.10)) is a first-order linear differential equation in u. 335 4. if any. 5.

we can find a particular solution 2 yp = (18 sin 3t − cos 3t). An important class of second-order linear differential equations are equations of the form t2 y ′′ + αty ′ + βy = 0. Find the solution of the initial-value problem y ′′ + 6y ′ + 8y = cos 3t y(0) = 0 y ′ (0) = 0. Compute the solution of the initial value problem y ′′ + 4y = 3 cos 2t y(0) = y ′ (0) = 0. (a) Show that the substitution x = ln t transforms an Euler equation into an equation of constant coefficients.2. dt2 dt dx2 dt dx (b) Solve the equation t2 y ′′ + 4ty ′ + 2y = 0. The solution to the homogeneous equation y ′′ + 7y ′ + 10y = 0 is yh = c1 e−4t + c2 e−2t .336 APPENDIX A. we have 16 −4t 18 −2t 2 y=− e + e + (18 sin 3t − cos 3t). y ′′ + 4y = 0 is yh = c1 cos 2t + c2 sin 2t. . 235 Thus.6 Exercises 1. HINTS AND SOLUTIONS TO SELECTED EXERCISES 7. 235 Applying the initial conditions y(0) = 0 and y ′ (0) = 0. 235 235 235 2. where t > 0 and α and β are real constants. The solution to the homogeneous equation. Euler Equations. An equation of this form is called an Euler equation. the general solution is 2 y = yh + yp = c1 e−4t + c2 e−2t + (18 sin 3t − cos 3t). 4. Hint: Show that dy dx dy = dt dt dx ( )2 2 d2 y dx d y d2 x dy = + 2 . Using the method of undetermined coefficients.

and the frequency is 3/2π. 4 we have 3 yp = t sin 2t. The amplitude is 4/ 442. 442 √ where ϕ = π + arctan(−21). since e2it is a solution to the homogeneous equation. the period is 2π/3. 4 3. Substituting yc and yc′′ into the differential equation. What is the amplitude. Taking the real part of 3 yc = − it(cos 2t + i sin 2t). 4 Applying the initial conditions y(0) = y ′ (0) = 0. 4 iϕ A = |A|eiϕ = √ e . Thus. we obtain 4 4 y=√ sin 3t + ϕ = √ cos(3t − θ). and −4 −2 + 42i A= = . b) if there do not exist nonzero constants c1 and c2 such that c1 f (t) + c2 g(t) = 0 . 442 442 √ where θ = ϕ − π/2 = π/2 + arctan(−21). then yc′′ + 7yc′ + 10yc = A(1 + 21i)e3it = −4e3it . Thus. We define two functions. Using the imaginary part of this solution. Assume that the particular solution to this equation is yc = ate2it . We include a factor of t. we have yc′′ + 4yc = 4a(i − t)e2it + 4ate2it = 4aie2it = 3e2it . Write your solution as a function of cosine. the period. Use the complex method to find a particular solution to y ′′ + 7y ′ + 10y = −4 sin 3t. and the frequency of your solution? If we assume that the complex solution has form yc = Ae3it . 337 The complex version of the equation is y ′′ + 4y = 3e2it . 1 + 21i 221 In polar form. f (t) and g(t). we find that c1 = c2 = 0 and 3 y = t sin 2t. 4. 4 Therefore the general solution is 3 y = yh + yp = c1 cos 2t + c2 sin 2t + t sin 2t. to be linearly independent on an open interval I = (a. a = 3/(4i) = −3i/4. yc = Ae3it = 4ei(3t+ϕ) / 442.

dt g g g2 and f /g is constant on the interval (c. for some constant c that depends on y1 and y2 but not on t. Thus. HINTS AND SOLUTIONS TO SELECTED EXERCISES for all t ∈ I. then 0f = g and the two functions are linearly dependent. it must also be continuous and there is some interval (c. Then one function is a multiple of the other. Consequently. show that ( ∫ ) W [y1 . f (t) = cg(t) for all t ∈ (a. ( ) f (t) g(t) W (f. Thus. f and g are linearly dependent. Show that f (t) and g(t) are linearly dependent on an interval I = (a. 5. f ′ (t) = cg ′ (t). b). g)(t) = det ′ = 0.338 APPENDIX A. Therefore. b) by the existence and uniqueness theorem. If y1 and y2 are solutions of the homogeneous equation y ′′ + p(t)y ′ + q(t)y = 0. g)(t) = det = f (t)g ′ (t) − f ′ (t)g(t) = cg(t)g ′ (t) − cg ′ (t)g(t) = 0. b). 2t 2t Since p(t) = 1/2t. ( ) d f f ′g − f g′ W (f. Abel’s Theorem. Suppose that f (t) and g(t) are solutions to the homogeneous linear equation y ′′ + py ′ + qy = 0. g) = 2 =− = 0. (b) Prove Abel’s Theorem. Suppose that that f (t) and g(t) are linearly dependent on an interval I = (a. (a) We can rewrite 2t2 y ′′ + 3ty ′ − y = 0 as 3 ′ 1 y ′′ + y − 2 y = 0. f (t) and f (t) are linearly dependent. f ′ (t) g ′ (t) Conversely. d) contained in (a. y2 ](t) = c exp − dt = c exp − ln t = ct−3/2 . Abel’s Theorem tells us that ( ∫ ) ( ) 3 3 W [y1 . d) and g does not vanish on this interval. g](t) ≡ 0. say f (t) = cg(t). Since f and cg are both solutions to the differential equation y ′′ + py ′ + qy = 0 and have the same initial condition. where p and q are continuous on an open interval I = (a. suppose that ( ) f (t) g(t) W (f. If g = 0. d). b). where t > 0. Otherwise. Since g is differentiable. f (t0 ) = cg(t0 ) and f (t′0 ) = cg ′ (t0 ). b) if and only if W [f. (a) Use Abel’s Theorem to find the Wronskian of 2t2 y ′′ + 3ty ′ − y = 0 up to a constant multiple. b) such that t0 ∈ (c. 2t 2 . f (t) g ′ (t) for all t in (a. Assume that g(t0 ) ̸= 0 for some t0 in (a. b). y2 ](t) = c exp − p(t) dt .

y2 )(t) = y1 y2′ − y1′ y2 . we obtain (y1 y2′′ − y1′′ y2 ) + p(t)(y1 y2′ − y1′ y2 ) = 0.12) becomes W ′′ + p(t)W = 0. . In this problem we will describe another method of finding a particular solution to a nonhomogeneous equation. y ′′ + p(t)y ′ + q(t)y = f (t). 6. (A. (b) Show that −y2 (t)f (t) u′1 (t) = W [y1 . W [y1 . y2 ](t) y1 (t)f (t) u′2 (t) = . where u′1 (t)y1 (t) + u′2 (t)y2 (t) = 0. (A. 339 (b) Since y1 and y2 are solutions to our differential equation.12) If W (t) = W (y1 . (a) Assume that a particular solution of (A. then W ′ = y1 y2′′ − y1′′ y2 . and equation (A. y2 ](t) = y1 (t)y2′ (t) − y2 (t)y1′ (t) is the Wronskian of y1 and y2 . y2 ](t) where W [y1 . Multiplying the first equation by y2 and the second equation by y1 and subtracting.13) has the form yp (t) = u1 (t)y1 (t) + u2 (t)y2 (t). Substitute yp into the left-hand side of (A.13) to show that u′1 (t)y1′ (t) + u′2 (t)y2′ (t) = f (t).13) if we know know that the general solution to the homogeneous equation y ′′ +p(t)y ′ +q(t)y = 0 is yh = c1 y1 + c2 y2 . The Method of Variation of Parameters. This equation is separable with solution ( ∫ ) W (t) = c exp − p(t) dt . we know that y1′′ + p(t)y1′ + q(t)y1 = 0 y2′′ + p(t)y2′ + q(t)y2 = 0.

. (a) If yp = u1 y1 + u2 y2 . W [y1 .340 APPENDIX A. (b) If we solve the system u′1 (t)y1 (t) + u′2 (t)y2 (t) = 0 u′1 (t)y1′ (t) + u′2 (t)y2′ (t) = f (t). To find a particular solution. we obtain −y2 (t)f (t) u′1 (t) = W [y1 . q. y2 ](s) for any point t0 in I. Consequently. Substituting these expressions into equation (A. 1 2y ](s) t0 W [y1 . y2 ](s) ∫ t y1 (s)f (s) u2 (t) = ds t0 W [y1 . y2 ](t) y1 (t)f (t) u′2 (t) = . we have yp′′ + pyp′ + qyp = (u′1 y1′ + u1 y1′′ + u′2 y2′ + u2 y2′′ ) + p(u1 y1′ + u2 y2′ ) + q(u1 y1 + u2 y2 ) = u1 [y1′′ + py1′ + qy1 ] + u2 [y2′′ + py2′ + qy2 ] + u′1 y1′ + u′2 y2′ = u′1 y1′ + u′2 y2′ = f (t). show that ∫ t y2 (s)f (s) u1 (t) = − ds t0 W [y1 . then yp′ = u′1 y1 + u1 y1′ + u′2 y2 + u2 y2′ = u1 y1′ + u2 y2′ yp′′ = u′1 y1′ + u1 y1′′ + u′2 y2′ + u2 y2′′ . HINTS AND SOLUTIONS TO SELECTED EXERCISES (c) If p. y2 ](t) (c) Integrate the two equations from part (2). t0 W [y . a particular solution to (A. assume that the solution has the form yp = u1 (t) cos 2t + u2 (t) sin 2t.13) is ∫ t ∫ t y2 (s)f (s) y1 (s)f (s) yp = −y1 (t) ds + y2 (t) ds. (d) The general solution to the homogeneous equation y ′′ + 4y = 0 is yh = c1 cos 2t + c2 sin 2t. and f are continuous on an interval I.13). y2 ](s) (d) Find the general solution of the differential equation y ′′ + 4y = 3 csc t. for u′1 and u′2 .

Will the singer have to sing a higher note or a lower note to break an identical glass that is half full of water? (b) Suppose that both notes are within the opera singer’s range. yp (t) = u1 (t)y1 (t) + u2 (t)y2 (t) [ ] 3 = −3 sin t cos 2t + ln | csc t − cot t| + 3 cos t sin 2t. determine the following. (b) The frequency of the rapid oscillations. 341 By part (2) u′1 (t) = −3 cos t 3 u′2 (t) = csc t − 3 sin t. 2 Integrating. (a) The frequency of the beats. 2 Therefore.4 Exercises 1.5 Exercises 1. 2. Given the equation y ′′ + 10y = sin 3t. 4. Compute the solution of the initial value problem y ′′ + 4y = 3 cos 2t y(0) = y ′ (0) = 0.3. 2 4. (c) Use parts (1) and (2) to sketch a graph of a typical solution. (a) Suppose that an opera singer can break a glass by singing a particular note. 2 and the general solution is y = yh + yp [ ] 3 = c1 cos 2t + c2 sin 2t − 3 sin t cos 2t + ln | csc t − cot t| + 3 cos t sin 2t. Will it be harder or easier to break the glass when it is half full of water? . we obtain u1 (t) = −3 sin t 3 u2 (t) = ln | csc t − cot t| + 3 cos t.4.

Compute an expression for M (p.342 APPENDIX A. These new constants. 1 G = 2√ ω0 (1 − s2 )2 + D2 s2 or 1 ω02 G = √ . The natural frequency is fixed in x′′ + 2cx′ + ω02 x = A cos ωt and D = 2c/ω0 is proportional to the damping constant. ∂A (a) Compute . 4. let M (p. measure the ratio of the driving frequency to the natural frequency and the effect of the damping force. q). Let us examine the amplitude and phase of the steady-state solution. (q − ω 2 )2 + p2 ω 2 That is. p. For large t. ∂ω (b) For fixed p and q. (d) Explain why M (p. Consider the following three systems (a) x′ = 3 sin x + y y ′ = 4x + cos y − 1 . (ω02 − ω 2 )2 + 4c2 ω 2 Now let s = ω/ω0 and D = 2c/ω0 . p.1. (c) Set q = 1 and plot M (p. Thus. respectively. the amplitude of A is a function of the parameters ω. s and D. p. 5. (1 − s )2 + D2 s2 2 This expression shows us how the gain varies as s = ω/ω0 varies. q) as a function of p. Is this a good test of the quality of the tape? Why or why not? 3. q) is proportional to 1/p as p → 0. q) = √ . q) denote the maximum value of A(ω. xp (t) = G(ω)A cos(ωt − ϕ).4 Exercises 1. HINTS AND SOLUTIONS TO SELECTED EXERCISES (c) Commercials for a maker of recording tape show a crystal glass being broken by the sound of a recording of an opera singer’s voice. and q. every solution of y ′′ + py ′ + qy = cos ωt oscillates with angular frequency ω and amplitude A given by 1 A(ω. q) as a function of ω. where 1 G(ω) = √ .

3. • Infected people either recover or die after a certain number of days. • The city has a constant birth rate rate of α persons per day. dt The constant α is determined by the probability of an interaction between a susceptible individual and an infected individual. If dS = −αSI + β = 0 dt dI = −γI + αSI = 0. then the disease will become endemic to the population. If we let S(t) be the number of susceptible individuals at time t and I(t) be the number of infected individuals at time t. Let us consider an epidemic model for a city. 2. We make the following additional as- sumptions about our model. I0 ).5 Exercises 1. γ We are interested in the behavior of solutions near (S0 . 5. All new born babies are susceptible to the disease. 343 (b) x′ = −3 sin x + y y ′ = 4x + cos y − 1 (c) x′ = −3 sin x + y y ′ = 4x + 3 cos y − 3 All three systems have an equilibrium solution at (0. he or she is immune. and γ is the rate at which infected individuals are removed from the population. 0). our assumptions give rise to the following system of differential equations. If an individual recovers. If solutions approach this equilbrium point. This will occur at γ S0 = α β I0 = . This is an exercise. . dS = −αSI + β dt dI = −γI + αSI. Which two systems have phase portraits with the same “local picture” near (0. 0)? Justify your answer. dt then both the susceptible and infected populations do not change.

sketch the level sets of S. (e) Why is this model not realistic for large values of x or y? 4. −2 ≤ y ≤ 2 is given by the function x4 + y 4 S(x. y ≥ 0. y) = + − 2 2 3 is a Lyapunov function for the system. (c) What can you conclude about the phase portrait of the system from (1) and (2)? 2. Suppose that the smell of a bunch of dead fish in the region −2 ≤ x ≤ 2. (b) Sketch the level sets of L. 4 (a) What is the gradient system who vector field is the gradient of S? (b) Use Sage to graph the phase portrait of the system. Consider the system x′ = y y y ′ = −x − + x2 .344 APPENDIX A. (c) Sketch the phase portrait of the gradient system in (1). where a and b are parameters with a. Let G(x. 4 (a) Verify that the function x2 y 2 x3 L(x. y) = x2 + y 2 − − 3x2 y 2 + 100. . HINTS AND SOLUTIONS TO SELECTED EXERCISES 5. Consider the system of differential equations x′ = x(−x − y + 1) y ′ = y(−ax − y + b). (c) How many dead fish are there and where are they? (d) Using the results from part (2).4. (b) Determine the values of a and b at which a bifurcation occurs.4 Exercises 1. (a) Use nullclines to sketch the phase portrait for this system for several values of a and b. 3. Suppose that this system is only defined for x. (a) What is the gradient system with vector field given by the gradient of G? (b) Sketch the graph of G and the level sets of G. y) = x3 − 3xy 2 . b > 0.

345 5. 0) if and only if a > 0 and ac − b2 > 0. Use the positive definite function V (x. y) = ax2 + 2bxy + cy 2 is positive definite with minimum at (x. Show that the quadratic polynomial V (x. y) = (0. 0). y) = (0. y) = x2 + y 2 to argue that the system x′ = −x − 2y 2 y ′ = 2xy − y 3 has an asymptotically stable equilibrium solution at (x. 6. .

Secondarily. unlimited in duration. The “Document”. ethical or political position regarding them. to use that work under the conditions stated herein. We recommend this License principally for works whose purpose is instruction or reference. 2007. This License is a kind of “copyleft”. 0. 1. philosophical. 2001. regardless of subject matter or whether it is published as a printed book. 2002. Inc. because free software needs free documentation: a free program should come with manuals providing the same freedoms that the software does. A “Modified Version” of the Document means any work containing the Document or a portion of it. but changing it is not allowed. that contains a notice placed by the copyright holder saying it can be distributed under the terms of this License.3. (Thus. royalty-free license. or with modifications and/or translated into another language. a Secondary Section may not explain any mathematics. if the Document is in part a textbook of mathematics. PREAMBLE The purpose of this License is to make a manual. below.) The relationship could be a matter of historical connection with the subject or with related matters. either copied verbatim. with or without modifying it. which means that derivative works of the document must themselves be free in the same sense. commercial. while not being considered responsible for modifications made by others. and is addressed as “you”. in any medium. it can be used for any textual work. textbook. Everyone is permitted to copy and distribute verbatim copies of this license document. or of legal. A “Secondary Section” is a named appendix or a front-matter section of the Document that deals exclusively with the relationship of the publishers or authors of the Document to the Document’s overall subject (or to related matters) and contains nothing that could fall directly within that overall subject. modify or distribute the work in a way requiring permission under copyright law. this License preserves for the author and publisher a way to get credit for their work. refers to any such manual or work. 346 . We have designed this License in order to use it for manuals for free software. You accept the license if you copy. It complements the GNU General Public License. which is a copyleft license designed for free software. Such a notice grants a world-wide. B GNU Free Documentation License Version 1. 3 November 2008 Copyright © 2000. But this License is not limited to software manuals. or other functional and useful document “free” in the sense of freedom: to assure everyone the effec- tive freedom to copy and redistribute it. APPLICABILITY AND DEFINITIONS This License applies to any manual or other work. either commercially or noncommercially. Any member of the public is a licensee. 2008 Free Software Foundation.

as Front-Cover Texts or Back-Cover Texts. or “History”. “Dedications”. “Endorsements”. PostScript or PDF produced by some word processors for output purposes only. provided that this License. has been arranged to thwart or discourage subsequent modification by readers is not Transparent. A copy that is not “Transparent” is called “Opaque”. (Here XYZ stands for a specific section name mentioned below. 347 The “Invariant Sections” are certain Secondary Sections whose titles are designated. and that you add no other conditions whatsoever to those of this License. The “Title Page” means. SGML or XML for which the DTD and/or processing tools are not generally available. for a printed book. PostScript or PDF designed for human modifica- tion. VERBATIM COPYING You may copy and distribute the Document in any medium. represented in a format whose specification is available to the general public. LaTeX input format. “Title Page” means the text near the most prominent appearance of the work’s title. The Document may contain zero Invariant Sections. For works in formats which do not have any title page as such. in the notice that says that the Document is released under this License. in the notice that says that the Document is released under this License. the material this License requires to appear in the title page. An image format is not Transparent if used for any substantial amount of text. the copyright notices. A copy made in an otherwise Transparent file format whose markup. and the machine-generated HTML.) To “Preserve the Title” of such a section when you modify the Document means that it remains a section “Entitled XYZ” according to this definition. You may not use technical measures to obstruct or control the reading or further copying of . Opaque formats include proprietary formats that can be read and edited only by proprietary word proces- sors. and a Back-Cover Text may be at most 25 words. These Warranty Disclaimers are considered to be included by reference in this License. Examples of suitable formats for Transparent copies include plain ASCII without markup. A section “Entitled XYZ” means a named subunit of the Document whose title either is precisely XYZ or contains XYZ in parentheses following text that translates XYZ in another language. SGML or XML using a publicly available DTD. A “Transparent” copy of the Document means a machine-readable copy. preceding the beginning of the body of the text. or absence of markup. The “Cover Texts” are certain short passages of text that are listed. and that is suitable for input to text formatters or for automatic translation to a variety of formats suitable for input to text formatters. the title page itself. A Front-Cover Text may be at most 5 words. Examples of transparent image formats include PNG. plus such following pages as are needed to hold. either commercially or noncommercially. and standard-conforming simple HTML. and the license notice saying this License applies to the Document are reproduced in all copies. as being those of Invariant Sections. XCF and JPG. that is suitable for revising the document straightforwardly with generic text editors or (for images composed of pixels) generic paint programs or (for drawings) some widely available drawing editor. The “publisher” means any person or entity that distributes copies of the Document to the public. The Document may include Warranty Disclaimers next to the notice which states that this License applies to the Document. legibly. Texinfo input format. such as “Acknowledgements”. If the Document does not identify any Invariant Sections then there are none. 2. If a section does not fit the above definition of Secondary then it is not allowed to be designated as Invariant. but only as regards disclaiming warranties: any other implication that these Warranty Disclaimers may have is void and has no effect on the meaning of this License.

or state in or with each Opaque copy a computer-network location from which the general network-using public has access to download using public-standard network protocols a complete Transparent copy of the Document. In addition. and from those of previous versions (which should. State on the Title page the name of the publisher of the Modified Version. numbering more than 100. one or more persons or entities responsible for authorship of the modifications in the Modified Version. free of added material. and continue the rest onto adjacent pages. if it has fewer than five). 4. 2. as long as they preserve the title of the Document and satisfy these conditions. you may accept compensation in exchange for copies. . you must do these things in the Modified Version: 1. unless they release you from this requirement. It is requested. You may also lend copies. you must take reasonably prudent steps. MODIFICATIONS You may copy and distribute a Modified Version of the Docu- ment under the conditions of sections 2 and 3 above. to ensure that this Transparent copy will remain thus accessible at the stated location until at least one year after the last time you distribute an Opaque copy (directly or through your agents or retailers) of that edition to the public. 3. However. together with at least five of the principal authors of the Document (all of its principal authors. under the same conditions stated above. Copying with changes limited to the covers. Use in the Title Page (and on the covers. and you may publicly display copies. as authors. COPYING IN QUANTITY If you publish printed copies (or copies in media that commonly have printed covers) of the Document. clearly and legibly. you must enclose the copies in covers that carry. thus licensing distribution and modification of the Modified Version to whoever possesses a copy of it. when you begin distribution of Opaque copies in quantity. GNU FREE DOCUMENTATION LICENSE the copies you make or distribute. that you contact the authors of the Document well before redistributing any large number of copies. You may add other material on the covers in addition. and the Document’s license notice requires Cover Texts. If you publish or distribute Opaque copies of the Document numbering more than 100. List on the Title Page. If you use the latter option. provided that you release the Modified Version under precisely this License. If the required texts for either cover are too voluminous to fit legibly. all these Cover Texts: Front-Cover Texts on the front cover.348 APPENDIX B. You may use the same title as a previous version if the original publisher of that version gives permission. with the Modified Version filling the role of the Doc- ument. be listed in the History section of the Document). 3. if there were any. If you distribute a large enough number of copies you must also follow the conditions in section 3. to give them a chance to provide you with an updated version of the Document. as the publisher. 4. if any) a title distinct from that of the Document. and Back-Cover Texts on the back cover. you must either include a machine-readable Transparent copy along with each Opaque copy. you should put the first ones listed (as many as fit reasonably) on the actual cover. Preserve all the copyright notices of the Document. The front cover must present the full title with all words of the title equally prominent and visible. can be treated as verbatim copying in other respects. Both covers must also clearly and legibly identify you as the publisher of these copies. but not required.

to the end of the list of Cover Texts in the Modified Version. Add an appropriate copyright notice for your modifications adjacent to the other copyright notices. Include. and preserve in the section all the substance and tone of each of the contributor acknowledgements and/or dedications given therein. If there is no section Entitled “History” in the Document. Preserve the network location. Section numbers or the equivalent are not considered part of the section titles. Such a section may not be included in the Modified Version. 13. These titles must be distinct from any other section titles. and publisher of the Document as given on its Title Page. provided it contains nothing but en- dorsements of your Modified Version by various parties — for example. Preserve in that license notice the full lists of Invariant Sections and required Cover Texts given in the Document’s license notice. then add an item describing the Modified Version as stated in the previous sentence. 10. Preserve all the Invariant Sections of the Document. 11. or if the original publisher of the version it refers to gives permission. and publisher of the Modified Version as given on the Title Page. given in the Document for public access to a Transparent copy of the Document. Preserve the section Entitled “History”. Preserve its Title. 349 5. in the form shown in the Addendum below. year. 9. You may add a passage of up to five words as a Front-Cover Text. if any. You may omit a network location for a work that was published at least four years before the Document itself. a license notice giving the public permission to use the Modified Version under the terms of this License. Delete any section Entitled “Endorsements”. create one stating the title. 7. 15. 8. and a passage of up to 25 words as a Back-Cover Text. 14. Only one passage of Front-Cover Text and one of Back-Cover Text may be added . Preserve any Warranty Disclaimers. 12. 6. add their titles to the list of Invariant Sections in the Modified Version’s license notice. Include an unaltered copy of this License. To do this. statements of peer review or that the text has been approved by an organization as the authoritative definition of a standard. Do not retitle any existing section to be Entitled “Endorsements” or to conflict in title with any Invariant Section. You may add a section Entitled “Endorsements”. These may be placed in the “History” section. For any section Entitled “Acknowledgements” or “Dedications”. you may at your option designate some or all of these sections as invariant. year. unaltered in their text and in their titles. authors. new authors. and likewise the network locations given in the Document for previous versions it was based on. If the Modified Version includes new front-matter sections or appendices that qualify as Secondary Sections and contain no material copied from the Document. and add to it an item stating at least the title. Preserve the Title of the section. immediately after the copyright notices.

If there are multiple Invariant Sections with the same name but different contents. TRANSLATION Translation is considered a kind of modification. but you may replace the old one. COMBINING DOCUMENTS You may combine the Document with other docu- ments released under this License. forming one section Entitled “History”. on explicit permission from the previous publisher that added the old one. the Document’s Cover Texts may be placed on covers that bracket the Document within the aggregate. In the combination. 7. and list them all as Invariant Sections of your combined work in its license notice. 6. or else a unique number. unmodified. and multiple identical Invariant Sections may be replaced with a single copy. You may extract a single document from such a collection. or the electronic equivalent of covers if the Document is in electronic form. but you . in or on a volume of a storage or distribution medium. and follow this License in all other respects regarding verbatim copying of that document. 8. The author(s) and publisher(s) of the Document do not by this License give permission to use their names for publicity for or to assert or imply endorsement of any Modified Version. and any sections Entitled “Dedications”. under the terms defined in section 4 above for modified versions. You must delete all sections Entitled “Endorsements”.350 APPENDIX B. the name of the original author or publisher of that section if known. The combined work need only contain one copy of this License. in parentheses. provided that you follow the rules of this License for verbatim copying of each of the documents in all other respects. you may not add another. so you may dis- tribute translations of the Document under the terms of section 4. 5. make the title of each such section unique by adding at the end of it. If the Cover Text requirement of section 3 is applicable to these copies of the Document. previously added by you or by arrangement made by the same entity you are acting on behalf of. provided you insert a copy of this License into the extracted document. GNU FREE DOCUMENTATION LICENSE by (or through arrangements made by) any one entity. and replace the individual copies of this License in the various documents with a single copy that is included in the collection. is called an “aggregate” if the copyright resulting from the compilation is not used to limit the legal rights of the compilation’s users beyond what the individual works permit. provided that you include in the combination all of the Invariant Sections of all of the original documents. Replacing Invariant Sections with translations requires special permission from their copyright holders. then if the Document is less than one half of the entire aggregate. Make the same adjustment to the section titles in the list of Invariant Sections in the license notice of the combined work. Otherwise they must appear on printed covers that bracket the whole aggregate. and that you preserve all their Warranty Disclaimers. you must combine any sections Entitled “History” in the various original documents. COLLECTIONS OF DOCUMENTS You may make a collection consisting of the Document and other documents released under this License. If the Document already includes a cover text for the same cover. and distribute it individually under this License. likewise combine any sections Entitled “Acknowledgements”. this License does not apply to the other works in the aggregate which are not themselves derivative works of the Document. AGGREGATION WITH INDEPENDENT WORKS A compilation of the Document or its derivatives with other separate and independent documents or works. When the Document is included in an aggregate.

or “His- tory”. “Dedications”. that proxy’s public statement of acceptance of a version permanently authorizes you to choose that version for the Document. Each version of the License is given a distinguishing version number. FUTURE REVISIONS OF THIS LICENSE The Free Software Foundation may publish new. If the Document specifies that a proxy can decide which future versions of this License can be used. you have the option of following the terms and conditions either of that specified version or of any later version that has been published (not as a draft) by the Free Software Foundation. receipt of a copy of some or all of the same material does not give you any rights to use it. as well as future copyleft versions of that license published by that same organization. if you cease all violation of this License. 11. A “Massive Multiauthor Collaboration” (or “MMC”) contained in the site means any set of copyrightable works thus published on the MMC site. or distribute it is void. and (b) permanently. then your license from a particular copyright holder is reinstated (a) provisionally. your license from a particular copyright holder is reinstated permanently if the copyright holder notifies you of the violation by some reasonable means. California. modify. However. If a section in the Document is Entitled “Acknowledgements”. if the copyright holder fails to notify you of the violation by some reasonable means prior to 60 days after the cessation. Moreover. TERMINATION You may not copy. See Copyleft. In case of a disagreement between the translation and the original version of this License or a notice or disclaimer. the original version will prevail. If the Document does not specify a version number of this License. this is the first time you have received notice of violation of this License (for any work) from that copyright holder. 10. and all the license notices in the Document. sublicense. 9. If the Document specifies that a particular numbered version of this License “or any later version” applies to it. A public wiki that anybody can edit is an example of such a server. You may include a translation of this License. If your rights have been terminated and not permanently reinstated. “CC-BY-SA” means the Creative Commons Attribution-Share Alike 3. and any Warranty Disclaimers. a not-for-profit corporation with a principal place of business in San Francisco. provided that you also include the original English version of this License and the original versions of those notices and disclaimers. or distribute the Document except as expressly provided under this License. modify. . revised versions of the GNU Free Documentation License from time to time. and you cure the violation prior to 30 days after your receipt of the notice. sublicense. the requirement (section 4) to Preserve its Title (section 1) will typically require changing the actual title. Termination of your rights under this section does not terminate the licenses of parties who have received copies or rights from you under this License. RELICENSING “Massive Multiauthor Collaboration Site” (or “MMC Site”) means any World Wide Web server that publishes copyrightable works and also provides prominent facilities for anybody to edit those works. Any attempt otherwise to copy. and will automatically terminate your rights under this License. 351 may include translations of some or all Invariant Sections in addition to the original ver- sions of these Invariant Sections. Such new versions will be similar in spirit to the present version.0 license pub- lished by Creative Commons Corporation. you may choose any version ever published (not as a draft) by the Free Software Foundation. unless and until the copyright holder explic- itly and finally terminates your license. but may differ in detail to address new problems or concerns.

with no Invariant Sections.” line with this: with the Invariant Sections being LIST THEIR TITLES. If you have Invariant Sections without Cover Texts. 2008. as part of another Document. we recommend releasing these examples in parallel under your choice of free software license. If your document contains nontrivial examples of program code. distribute and/or modify this document under the terms of the GNU Free Documenta- tion License. or some other combination of the three. and (2) were thus incorporated prior to November 1. replace the “with… Texts. no Front-Cover Texts. GNU FREE DOCUMENTATION LICENSE “Incorporate” means to publish or republish a Document. merge those two alternatives to suit the situation. . with the Front-Cover Texts being LIST. An MMC is “eligible for relicensing” if it is licensed under this License. and with the Back-Cover Texts being LIST. (1) had no cover texts or invariant sections. in whole or in part. to permit their use in free software. The operator of an MMC Site may republish an MMC contained in the site under CC- BY-SA on the same site at any time before August 1. include a copy of the License in the document and put the following copyright and license notices just after the title page: Copyright © YEAR YOUR NAME Permission is granted to copy.3 or any later version published by the Free Software Foundation. Version 1. Front-Cover Texts and Back-Cover Texts. provided the MMC is eligible for relicensing. and subse- quently incorporated in whole or in part into the MMC. and no Back- Cover Texts. and if all works that were first published under this License somewhere other than this MMC. A copy of the license is included in the section entitled “GNU Free Documentation License”. such as the GNU General Public License.352 APPENDIX B. If you have Invariant Sections. ADDENDUM: How to use this License for your documents To use this License in a document you have written. 2009.

College Mathematics Journal 31(2000) 338–346. [3] Robert L. & Robert L. Second edition. CA. & an Introduction to Chaos. Nicholson.215–244. Eighth edition. Fifth edi- tion. F. [4] William E. [11] Elton. San Diego. Springer. Sage for Undergraduates. Essential Mathematical Biology. Elementary Differential Equations with Boundary Value Problems. third edition. “Deterministic nonperiodic flow. Brooks/Cole. Shammas. 2004. [7] Nicholas Britton. Burden & Douglas Faires. 2005. Upper Saddle River. Second edition. John Wiley & Sons. Pacific Grove. C. Fourth edition. Large amplitude periodic oscillations in suspension bridges: some new connections with nonlinear analysis. 2004. NJ. Elementary Differential Equations and Boundary Value Problems. 2015. D. [5] Brauer. Humphreys and R. [15] Edward N. CA. [8] Richard L. 2006. New York. Differential Equations. Differential Equations and Their Applications: An Introduction to Applied Mathematics. Boyce & Richard C. 2005. 353 . Springer. 2004. Pacific Grove. New York. 11(1942). Lorenz. 130–141. Lazer and P. Eighth edition Brooks/- Cole. Differential Equations: A Modeling Per- spective. Robert L. Academic Press. Differential Equations. Diprima. John Wiley & Sons. McKenna. “The ten year cycle in the numbers of lynx in Canada.” Journal of Atomospheric Sci- ence 20(1963). [9] Ward Cheney & David Kincaid. & M. Springer-Verlag.” Journal of Animal Ecology. Prentice Hall. Stephen Smale. SIAM Review 32(1990) 537– 578. pp. Borrelli & Courtney S. 2003. Penney. Castillo-Chávez. 2004. New York. Dynamical Systems. CA. New York. Hall. Fifth edition. Hirsch. 2001. Coleman. Pacific Grove. Devaney. [12] Morris W. C. Provi- dence. [14] A. [10] C. [6] Martin Braun. Mathematical Models in Population Biology and Epidemiology. [2] Paul Blanchard. Henry Edwards & David E. Devaney. Numerical Analysis. Numerical Mathematics and Computing. Bard. American Mathematical Society. [13] L. pp. New York. B Readings and References [1] Gregory V. Springer Undergraduate Series. 1992. Finding unpredictable behavior in a simple ordinary differential equation. Brooks/Cole. S. & Glen R. Texts in Applied Mathematics 40. J. & C.

second edition. “Modeling Viral Infections” in An Introduction to Mathematical Modeling in Physiology. S. [21] John Polking. & J..” Nature 367(1994). Large torsional oscillations in suspension bridges visited again: vertical forcing creates torsional response. [18] P. [20] Perelson. “Better protection of the ozone layer. & Crawford S. & David Arnold. Cambridge University Press. N. pp. 315–332. 2008. Modeling Differential Equations in Biology. [17] K. McKenna. “Qualitative analysis of insect outbreak systems: the spruce budworm and forest. Prentice Hall. & P. Holling. American Mathematical Society. pp. [22] Clifford Henry Taubes. Large torsional oscillations in suspension bridges revisited: fixing an old approximation. Albert Boggess. Cell Biology. second edition. B. Nelson. W. Upper Saddle River. Sze. W. NJ. American Mathe- matical Monthly 108(2001) 738–745. [19] P. J. McKenna and Cillian Ò Tuama.354 READINGS AND REFERENCES [16] Donald Ludwig. American Mathematical Monthly 106(1999) 1–18. 2002. Malcolm. . A. Cambridge. U. Differential Equations with Boundary Value Problems. and Immunology. Prather. 2006. Jones. Providence. J.” The Journal of Animal Ecology (1978).K. Dixon D. 505–508.

77 Kirchhoff’s Second Law. 43 sink. 16 parameter. 42 logistic equation. 113 initial condition. 3 Lipschitz condition. 54 diagram. 125 Hooke’s Law. Index bifurcation. 33 mixing problems. 30 phase plane. 85 systems of differential equations. 24 bifurcation value. 113 direction field. 71 separable. 28 node. 16 autonomous. 34. 76 integrating factor. 76 pitch fork. 74 nullcline. 91 harmonic oscillator. 53 first-order linear. 23 particular solution. 125 initial value problem. 7 first-order linear. 39 solution curves. 35 second-order. 34 direction field. 81 over-damped. 81 slope field. 2 decoupled. 53 Newton’s law of cooling. 30 Euler’s method. 20 exact. 74 integral equation. 6 equilibrium solution. 84 SIR model. 125 partially coupled. 3 differential equation. 65 equilibrium solution. 19 one-parameter family. 106 Picard iteration. 35 Runge-Kutta method. 44 source. 2 improved. 6 homogeneous first-order linear. 35 RC circuit. 96 order. 2 solution. 16 phase line. 2 vector field. 2 autonomous. 16. 82 Duffing’s equation. 44 system exponential growth. 29 carrying capacity. 35 stable. 91 general solution. 2 integral curves. 2 planar. 94 355 .

356 INDEX .

.This book was authored in MathBook XML.