Differential Equations II

(MATH 450)

Dr. Joseph K. Ansong

(Dept. of Mathematics, University of Ghana, Legon)

LECTURE NOTES
Math 450 : Differential Equations II J.K.A

2 c Dr. Joseph K. Ansong

Contents

1 Introduction 1
1.1 Definitions & Terminology . . . . . . . . . . . . . . . . . . . . 1
1.2 Approaches to Problems in ODE . . . . . . . . . . . . . . . . 3

2 First Order Differential Equations 5

2.1 Separable Equations . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Existence & Uniqueness of Solutions . . . . . . . . . . . . . . 7
2.3 Geometric Methods: first order equations . . . . . . . . . . . . 11
2.4 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.4.1 The Logistic Population Model . . . . . . . . . . . . . 21
2.4.2 The Logistic Model with Constant Harvesting & Bi-
furcation . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3 Linear Systems 25
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 Homogeneous Linear Systems . . . . . . . . . . . . . . . . . . 28

4 Planar Linear Systems 35

4.1 Geometric Methods for Planar Systems . . . . . . . . . . . . . 37
4.2 Phase Portraits . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.2.1 Real Distinct Eigenvalues . . . . . . . . . . . . . . . . 39
4.2.2 Complex Eigenvalues . . . . . . . . . . . . . . . . . . . 46
4.2.3 Repeated Eigenvalues: Case 1 . . . . . . . . . . . . . . 50
4.2.4 Repeated Eigenvalues: Case 2 . . . . . . . . . . . . . . 50
4.3 The Trace-Determinant Plane . . . . . . . . . . . . . . . . . . 57

5 The Matrix Exponential Function 63

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.2 The matrix exponential . . . . . . . . . . . . . . . . . . . . . . 64
5.2.1 Properties of the matrix exponential . . . . . . . . . . 65
5.2.2 Computation of eAt : Special Matrices . . . . . . . . . . 66

3
Math 450 : Differential Equations II J.K.A

5.3 Computing eAt : Generalized eigenvectors . . . . . . . . . . . . 69

5.3.1 Solving systems using generalized eigenvectors . . . . . 70
5.4 Computing eAt : The Putzer Algorithm . . . . . . . . . . . . . 76

6 Nonlinear autonomous systems 81

6.1 Linearization Theorem . . . . . . . . . . . . . . . . . . . . . . 81
6.2 Lyapunov functions . . . . . . . . . . . . . . . . . . . . . . . . 84

A Lipschitz Condition 89

B Linear Second Order ODEs 95

B.1 Homogeneneous Linear Equations . . . . . . . . . . . . . . . . 95
B.2 Non-homogeneous Equations . . . . . . . . . . . . . . . . . . . 97
B.2.1 The Method of undetermined coefficients . . . . . . . . 98

4 c Dr. Joseph K. Ansong

Chapter 1

Introduction

1.1 Definitions & Terminology

Mathematical models play a critial role in the sciences and engineering by
helping us gain a better understanding of real life phenomena. The mod-
els are generally simplified versions of the actual physical phenomena under
investigation. They are simplified because the parameters governing the nat-
ural phenomena is often not completely understood or may be too compli-
cated to be represented mathematically. The development of mathematical
models usually result in an equation or a set of equations specifying how
an unknown function (say φ(t)) changes with respect to a variable (say t).
Such an equation is referred to as a differential equation. For example,
suppose the variation in the population of mosquitoes in a certain village is
represented by the differential equation
dM
= kM, (1.1)
dt
where M is the number (population) of mosquitoes at time t, and k is a known
constant (obtained from observational or experimental data). Fortunately for
us, equation (1.1) can be solved easily using integration techniques learned
in your Calculus class. Re-writing the equation and integrating results in
dM
= kdt,
Z M Z
dM
= kdt,
M
ln(M ) = kt + C1 ,
M = ekt+C1 = eC1 ekt
M (t) = Cekt . (1.2)

1
Math 450 : Differential Equations II J.K.A

The integration constant C can be determined if we know the initial popu-

lation of mosquitoes. Equation (1.2) then helps us to predict the population
of mosquitoes at a future time. For instance, if initially the population of
mosquitoes is 1000 and k is −5, then we have
1000 = Ce5×0 = C,
=⇒ M (t) = 1000e−5t . (1.3)
Thus, our simple model may now be used to predict the decline in the pop-
ulation of mosquitoes in the village. A differential equation (e.g. equation
1.1) with the initial value specified (e.g. M = 1000 at t = 0; as in the above
example) is called an Initial Value Problem (IVP).
Notice that the solution to a differential equation is not a number but a
function (see equation 1.2), and contains arbitrary “constants of integration”.
The presence of these arbitrary constants implies that there is generally no
unique solution to a differential equation. The unknown function in a differ-
ential equation (e.g. M in equation 1.1) is called the dependent variable
and the others (e.g. t in equation 1.1) is called the independent variable.
It is usually clear from the equation which variable is dependent and which
is independent.

Definition 1. [Differential Equation]

A differential equation is an equation involving the rate of change of a
quantity. That is, it involves a function and its derivatives.

An ordinary differential equation (ODE) is a differential equation in

which only the derivatives of the unknown function with respect to one inde-
pendent variable appear in the equation. A partial differential equation
(PDE) is a differential equation in which partial derivatives of the unknown
function with respect to at least two independent variables appear in the
equation. Examples of ODEs are
dy
+ kt = 10,
dt
d2 y dy
2
+ a + by = x2 . (1.4)
dx dx
The following are some examples of PDEs
∂u2 ∂v
+ = 0,
∂x2 ∂x
∂φ ∂φ
+b + b = 0.
dx ∂y

2 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

The order of a differential equation is the order of the highest derivative

appearing in the equation. Examples of first order equations are
dx dy
+ x = 0, and + 3xy = 0,
dt dx
and the following are second order equations:
 3
d2 y d2 x dx
2
+ cy = 0, and 2
+ = 0.
dx dt dt
The constants appearing in a differential equation are called parameters;
for example the constants a and b in equation (1.4).
A first order ODE is called explicit if it can be written in the form
dx
= f (t, x), (1.5)
dt
for a real-valued function f of two variables.

1.2 Approaches to Problems in ODE

The following are typical guidelines to solving differential equations.
1. Is the problem well-posed? A problem is said to be well-posed if it
has a solution, if that solution is unique and if the solution depends
continuously on the initial data.

2. Are solutions stable? In order words, how does a given solution change
if the initial data is changed by a small amount?

3. Are there explicit solutions? It is important to know the types of

differential equations that can be solved exactly.

4. Can we find approximate solutions? If so, how accurate is the approx-

imate solution?

5. Can we gain any qualitative insights? Can you derive qualitative fea-
tures of the solution without actually having a solution?

6. Can we obtain numerical solutions? Several numerical routines are

available for solving certain types of differential equations. For exam-
ple MATLAB, SciPy, MATHEMATICA, etc. You need to check the
problems for well-posedness before proceeding numerically.

3 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

4 c Dr. Joseph K. Ansong

Chapter 2

First Order Differential

Equations

2.1 Separable Equations

Definition 2. [Separable Equations] A first order differential equation

dy
= f (x, y)
dx
is said to be separable if f (x, y) can be written as the product of a
function g(x) that depends on only x and a function h(y) that depends
on only y such that
dy
= h(y)g(x) (2.1)
dx

Method of Solution
To solve
dy
= h(y)g(x), (2.2)
dx
separate variables to get

dy
= g(x)dx,
h(y)

5
Math 450 : Differential Equations II J.K.A

where we assume that h(y) 6= 0. Now integrate both sides:

Z Z
dy
= g(x)dx.
h(y)
The result of the integration above yields the solution to the differential
equation.
Example 1. Solve the following differential equation
dy 1
= 3
dx xy
Solution.
dy 1
= 3.
dx xy
dx
=⇒ y 3 dy = ,
Z Zx
3 dx
=⇒ y dy = ,
x
1
=⇒ y 4 = ln(x) + C1 ,
4
4
=⇒ y = [ln(x) + 4C1 ] = [ln(x) + C],
y = [ln(x) + C]1/4
Example 2. Solve the following differential equations
dy
= y(2 + sin x)
dx
(2.3)
Solution.
dy
= y(2 + sin x),
Z dx Z
dy
= (2 + sin x)dx,
y
ln |y| = 2x − cos(x) + C1 ,
|y| = e2x−cos x+C1 = eC1 e2x−cos(x) = Ceex−cos x ,
where C > 0. So we get
y = Ke2x−cos(x) ,
where K is an arbitrary constant.
You should review other methods of solution of first order equations
from MATH 350, such as (1) exact equations, (2) changing variables (e.g.
Bernoulli equations), (3) solving Linear Equations using an integrating
factor, and so on.

6 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

2.2 Existence & Uniqueness of Solutions

Definition 3. [Solution of DE]

Let I be an interval and t0 ∈ I. We say that a differentiable function
x : I → R is a solution of the IVP
dx
= f (t, x), x(t0 ) = x0 , (2.4)
dt
dx
in the interval I if = f (t, x) for all t ∈ I and x(t0 ) = x0 .
dt

Theorem 1. [Picard-Lindelof Theorem (version 1)]

Consider the intervals IT = [t0 − T, t0 + T ] and Bd = [x0 − d, x0 + d] for
positive, real numbers T, d. Suppose that f : IT × Bd → R is continuous
∂f
and that its partial derivative : IT × Bd → R is also continuous.
∂x
Then there is a δ > 0 so that the initial value problem (2.4) has a unique
solution in the interval Iδ = [t0 − δ, t0 + δ].

W need to understand the importance of the theorem and how it can be used
to analyze first order differential equations.

Remark. Figure 2.1 illustrates the content of the Picard-Lindelof Theorem.

The conditions of the Theorem are satisfied in the larger rectangle IT × Bd ,
where f and ∂f /∂x are continuous. The smaller rectangle [t0 − δ, t0 + δ] ×
[x0 − d, x0 + d], with the cross-hatched region, is where the IVP is expected
to have a unique solution.
If a differential equation satisfies the hypotheses of Theorem 1,

1) we are assured that a solution to the IVP exists.

2) there is a unique solution to the IVP. That is, if we can find a solution,
then it is the only solution for the IVP. Uniqueness also implies that
there is only one solution curve that passes through the initial point
(t0 , x0 ). Thus, two solutions cannot cross anywhere in the rectangle.

3) the existence and uniqueness of the solution holds only in some neigh-
borhood (t0 − δ, t0 + δ). However, the theorem does not tell us the span
(2δ) of this neighborhood.

7 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Figure 2.1: Illustration of the Picard-Lindelof Theorem

4) The Theorem guarantees a solution for t close to t0 , but this solution

may not exist for all t ∈ R
5) Another formulation of the Picard-Lindelof theorem replaces the re-
quirement that ∂f /∂x exists and is continuous by a weaker condition
that f satisfies a Lipschitz condition with respect to its second argu-
ment. We first give some examples to help us understand the statement
of the Theorem.
Example 3. Consider the initial value problem
dx
= t2 − tx3 , x(1) = 6.
dt
Does Theorem 1 imply the existence of a unique solution?
Solution. From the equation, we have f (t, x) = t2 − tx3 and ∂f /∂x =
−3tx2 . In any rectangle containing the initial point (t0 , x0 ) = (1, 6), both

8 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

the function and it partial derivative are continuous, so the hypotheses of

Theorem 1 are satisfied. Hence the IVP has a unique solution in an interval
about t = 1 of the form (1 − δ, 1 + δ), where δ is some positive number.

Example 4. Consider the initial value problem

dx
= 3x2/3 , x(2) = 0.
dt
Does Theorem 1 imply the existence of a unique solution?

Solution. We have f (t, x) = 3x2/3 and ∂f /∂x = 2/x1/3 . Though the func-
tion f is continuous about the initial point (2, 0), its partial derivative ∂f /∂x
is not continuous or even defined at x = 0. So the hypotheses of Theorem 1
are not satisfied. Therefore we cannot use Theorem 1 to determine whether
the IVP has or does not have a unique solution.

Remark. Why did the problem violate the hypotheses of the theorem? First
notice that the equation can be solved via integration:

dx
= 3x2/3
dt
Z Z
−2/3
=⇒ x dx = 3dt

=⇒ 3x1/3 = 3t + C1
=⇒ x1/3 = (t + C)
=⇒ x = (t + C)3
Applying the initial condition results in the solution

x = (t − 2)3 .

Also notice that x(t) ≡ 0 is another solution to the IVP since it satisfies the
equation. So the IVP has two solutions and therefore cannot have a unique
solution as required by Theorem 1.

Example 5. Consider the initial value problem

dx
= 3x2/3 , x(2) = 1.
dt
Does Theorem 1 imply the existence of a unique solution?

9 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Solution. Example 5 is the same as example 4 but with a different initial

condition. Again, from the equation we have f (t, x) = 3x2/3 and ∂f /∂x =
2/x1/3 . We see now that both the function and its partial derivative ∂f /∂x
are continuous in any rectangle containing (2, 1) as long as x > 0. So the
hypotheses of Theorem 1 now hold. Thus, Theorem 1 implies that the IVP
now has a unique solution about t = 2.

Example 6. Consider the initial value problem

dx
= 2tx2 , x(0) = 1.
dt
Does Theorem 1 imply the existence of a unique solution? What is the
interval of existence of the solution?

Solution. We see that the function f and its partial derivative are continu-
ous, guaranteeing the existence of a unique solution. However, we will find
that the solution does not exist for all t ∈ R.
dx
= 2tx2 , x(0) = 1,
dt Z Z
dx
= 2tdt,
x2
1
− = t2 + C,
x
1
=⇒ x(t) = , (2.5)
1 − t2
after applying the initial condition. We note that the solution blows up (goes
to infinity) as t approached ±1, therefore the solution (as in Definition 3)
only exists in the interval −1 < t < 1.

The continuity of the function f (t, x) is sufficient for the existence of a

solution to the first order IVP (Peano’s Theorem). However, this does not
imply uniqueness. As demonstrated in Example (4), even though the func-
tion f (t, x) = 3x2/3 is continuous in the tx−plane, the IVP has at least two
solutions: x(t) ≡ 0 and x(t) = (t − 2)3 . Thus additional conditions are re-
quired on f (t, x) to ensure uniqueness. One such condition is the continuity
of ∂f /∂x as stated in Theorem 1. Another version of the Picard-Lindelof
Theorem requires the weaker condition that the variation of f relative to x
must remain bounded, that is, it must satisfy a so-called Lipschitz con-
dition (see Appendix ??). In this course we are going to stick to the first
version stated above.

10 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

2.3 Geometric Methods: first order equations

In this section, we will analyze the qualitative behaviour of solutions to first
order differential equations using the concept of direction fields and phase
lines. First consider the following simple example of a differential equation
whose analytical solution can be obtained.

Example 7.
dx
= αx, x(t0 ) = x0 , (2.6)
dt
where α is a parameter. Separating variables, the solution is given by

x(t) = Ceαt , (2.7)

where C = x0 e−αt is a contant obtained by applying the initial contidion. The

qualitative behaviour of the solutions depend on the sign of the parameter α
as described below

a) If α > 0, limt→∞ Ceαt = ∞ for C > 0 and limt→∞ Ceαt = −∞ for

C < 0.

b) If α = 0, we have the constant or equilibrium solutions given by

x(t) = Ceαt = constant.

c) If α < 0, limt→∞ Ceαt = 0.

Figure 2.2 displays the qualitative behaviour of the solutions. We find that
for α > 0 all nonzero solutions tend away from the equilibrium point as
t → ∞. On the other hand, for α < 0, all nonzero solutions tend toward
the equilibrium solution as t → ∞. So the sign of α plays a critical role
in determining the qualitative behaviour of the solutions. For instance, if
α > 0, we can replace α by another number with the same sign as α and the
qualitative nature of the solution does not change. However, if α = 0, then
the slightest change in α leads to a radical or drastic change in the behaviour
of solutions. Such a situation is what is called a bifurcation at α = 0 for
the family of differential equations dx/dt = αx.

The question is, suppose we do not have the solutions given by equation
(2.7), can we still obtain a sketch or plot of the qualitative behaviour of
solutions, such as those displayed in Figure 2.2? The answer is yes. We are
going to explore the qualitative behaviour of solutions using the concept of
direction fields and phase lines as described below.

11 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Figure 2.2: (a) Solution curves from equation (2.7) for α > 0. (b) Same as
(a) but for α < 0.

Direction Field
In this section, we study how to use the direction field of a differential
equation to sketch the graphs of solutions of first order ODEs. The idea
is to obtain a qualitative view of the graph of solutions to a differential
equation without actually solving the equation. We first assume that we
have a family of solutions and calculate their direction field as outlined below.
After obtaining the direction field, we can then sketch the solutions of interest
depending on the initial condition(s).

12 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Definition 4. [Direction Field] The direction field of a differential equa-

tion
dx(t)
= f (t, x(t)) (2.8)
dt
is a map
 
1
V~ : R2 → R, V~ (t, x) = (2.9)
f (t, x)

Thus, V~ is a collection of all the tangent vectors to graphs of all possible

solutions of the ODE. The definition is derived from the idea that, at each
point in the plane, say (t∗ , x∗ ), the direction of the tangent vector at that
point is given by
 
d 1
T = dx ∗
dt
(t )

as illustrated in Figure below. But from the differential equation, we have

 
dx ∗ ∗ ∗ d 1
(t ) = f (t , x ), so that T = .
dt f (t∗ , x∗ )

Example 8. Sketch the direction field for equation (2.6), repeated here for
convenience:
dx
= αx.
dt
where α = 1. Hence sketch the solution satisfying the initial condition
(i) x(−2) = 0.5
Solution.    
1 1
V (t, x) == =
f (t, x) x
Below are a few of the vectors corresponding to V (t, x).
   
1 1
V (0, 0) = , V (0, 1) =
0 1
   
1 1
V (−2, 1) = , V (−1, 1) =
1 1
   
1 1
V (0, 2) = , V (0, 3) =
2 3

13 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

   
1 1
V (0, 4) = , V (1, 4) =
4 4

Notice that vectors on the same horizontal line have the same slope because
f is independent of t. Figure 2.3a displays the direction field using a couple
of points. The solution curve starting from the initial point (−2, 0.5) is
shown as the red solid curve. This curve is comparable to the solution curves
shown in Figure 2.2 (left panel for x > 0), which were obtained analytically.
Similarly, we can sketch the direction field corresponding to the case when
α = −1 to get the graph in Figure 2.3b with a solution curve starting from
about (−1.6, 4.5).

Figure 2.3: (a) Direction field for equation (2.9) with α = 1. (b) Same as (a)
but for α = −1.

14 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Example 9. Sketch the direction field for

dx
= t 2 + x2 .
dt
Hence sketch the solutions satisfying the initial conditions

(i) x(−1) = −1, (ii) x(−1) = 0.

Solution.
dx
= t2 + x2 ; so f (t, x) = t2 + x2
dt

   
1 1
V~ (t, x) = =
f (t, x) t + x2
2

Now we choose a couple of points to evaluate the vector field:

   
1 1
V (0, 0) = ; V (0, 1) =
0 1
   
1 1
V (1, 0) = ; V (1, 1) =
1 2
   
1 1
V (−1, 1) = ; V (2, 0) =
2 4

These vectors are sketched in Figure 2.4. Figure 2.5 is the same as Figure
2.4 but with additional vectors plotted to enable sketching of the the solution
curves starting at the initial conditions. For plotting purposes, it is usually
advisable to rescale the vectors by a suitable constant since we are mainly
interested in the direction of the vectors and not their magnitudes. Figure 2.6
shows the full direction field with the solution curves starting at the initial
values. Without normalizing the vectors Figure 2.6 severely lack clarity.

Example 10. Find all constant solutions (equilibrium points) of

dx
= x(x − 1)(x − 2). (2.10)
dt
Hence sketch the solutions satisfying the initial conditions

(i) x(0) = 0.5, (ii) x(0) = 1.5

15 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Figure 2.4: Direction field for a few points of the ODE dx/dt = t2 + x2 .

Solution. Constant solutions occur for f (t, x) = 0:

x(x − 1)(x − 2) = 0
=⇒ x = 0, 1, 2.
Note that f (t, x) does not depend on t; such equations are said to be au-
tonomous. The direction field for such equations are much easier to sketch
because the slopes on the same x−axis are the same since f is independent
of t. First draw the horizontal lines of the constant solutions, x(t) = 0, 1, 2.
The graphs of other solutions must not cross these horizontal lines. Besides
computing a few points using equation (2.9), the directions of the vectors
can also be determined in the following way
dx
x < 0, < 0, vectors point down: ↓
dt
dx
0 < x < 1, > 0, vectors point up: ↑
dt
dx
1 < x < 2, < 0, vectors point down ↓
dt
Figure 2.7 shows the direction field and the solution curves to the differ-
ential equation.

16 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Figure 2.5: Same as Figure 2.4 but with additional vectors and the solution
curves through the inital values sketched.

Phase Lines
Autonomous equations are of the form
dx
= f (x), (2.11)
dt
where the function f is independent of t. As mentioned previously, the direc-
tion field for such equations are much easier to sketch. However, the direction
field for autonomous equations contain much redundant information. Since
f does not depend on t, there is little point in sketching the arrows in the
(t, x)−plane. So we can drop the t−coordinate and draw arrows only on the
x−axis; this is called a phase line. These concept is illustrated using the
equation below:

dx
= −A(x − x1 )(x − x2 )(x − x3 )2 , (2.12)
dt
for A > 0 and x1 , x2 , x3 are numbers. So

f (t, x) = f (x) = −A(x − x1 )(x − x2 )(x − x3 )2 .

17 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Figure 2.6: Same as Figure 2.5 but with additional vectors and the solution
curves through the inital values plotted using MATLAB.

a) Constant or equilibrium solutions occur for

f (x) = 0 =⇒ x = x1 , x = x2 , x = x3 .

b) The slopes in the direction field are all identical along horizontal lines
as shown in Figure 2.8 (left panel).

c) New solutions can be generated from old ones by time shifting (i.e.
replacing x(t) with x(t − t0 )).

d) Equilibrium solution x = x2 is called a stable equilibrium, or sink,

because neighboring solutions are attracted to it as t → ∞. In other
words, an equilibrium point is called stable if, when you start close
enough to it, you stay close to it.

e) Equilibrium solutions that repel neighboring solutions are called sources.

So x = x1 is a source.

f) All other equilibria are called nodes; for example x = x3 .

g) Sources and nodes are unstable equilibria.

18 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Figure 2.7: Direction field for equation (2.10) together with the solution
curves; plotted with MATLAB.

Example 11. Sketch the direction field the following equation.

dx
= (x − 1)(x − 3)(x − 5), (2.13)
dt
Hence sketch the solutions satisfying
(i) x(0) = 0.9, (ii) x(0) = 1.2
(iii) x(0) = 4.5, (iv) x(0) = 5.1
Solution. The direction field and solutions curves are displayed in Figure
2.9 (left panel) and the phase line is shown on the right panel. We see that
the equilibrium solutions x = 1, 5 are sources while x = 3 is a sink.
Exercise. Find all constant solutions of
dx
+ x2 = 1
dt
and sketch the direction field for this equation. Hence sketch the solutions
satisfying
(i) x(0) = 1.1, (ii) x(0) = 0, (iii) x(0) = −1.1.

19 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Figure 2.8: [Left] Direction field for equation (2.12) together with some so-
lution curves. [Right] Phase line of the direction field on the left.

Figure 2.9: [Left] Direction field for equation (2.13) together with some so-
lution curves. [Right] Phase line of the direction field on the left.

20 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

2.4 Applications
2.4.1 The Logistic Population Model
The simple model
dx
= ax (2.14)
dt
is a simple example of a model for predicting population growth for a > 0.
The equation was briefly analyzed previously (see equation 2.6). It is called
the Malthusian model. x(t) measures the population of a species at time t.
It is based on the assumption that the rate of growth of the population is
proportional to the size of the population. It neglects several circumstances
of actual population growth, for example, actually populations cannot grow
without bound. Modifying the Malthusian model results in the logistic pop-
ulation model
dx  x
= ax 1 − (2.15)
dt N
where a is the rate of population growth and N > 0 is called the “carrying
capacity” or “ideal population”. The model assumes that, if the population
is small, the growth rate is Malthusian. But if the population grows too
large, the growth rate becomes negative. Normalizing (2.15) results in
dx̃
= ax̃(1 − x̃)
dt
where x̃ = x/N . Neglecting the tilde sign on x, we get
dx
= ax(1 − x). (2.16)
dt
Now x represents the fraction of the ideal population at time t (0 ≤ x ≤ 1).

Qualitative Analysis

dx
= ax(1 − x). (2.17)
dt
Let the right hand side be represented by fa (x) = ax(1 − x). At the equilib-
rium points fa (x) = 0. This implies that

x = 0, x = 1

21 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

are equilibrium points. So we have for

dx
x<0: <0
dt
dx
0<x<1: >0
dt
dx
x>1: <0
dt

Analysis from graphs of f (t, x)

The graph of f (x) = ax(1 − x) may also be used to analyze the qualative
behaviour of solutions of (2.17). By plotting the graph of f (x), we see that
(i) points where f (x) crosses the x−axis are the equilibrium solutions.
That is
f (x) = ax(1 − x) = 0 =⇒ x = 0, x = 1
are the equilibrium points.
(ii) for 0 < x < 1, we have f (x) > 0 (that is, the graph is above the
x−axis) which implies that slopes are positive (dx/dt > 0). Hence
solutions must increase in this region.
(iii) when x < 0 or x > 1, we have f (x) < 0 (that is, the graph is below the
x−axis) and so solutions must decrease.

Analytical approach
The behaviour of solutions determined above may also be obtained analyti-
cally. Consider
f (x) = ax(1 − x), a > 0.
=⇒ f 0 (x) = a − 2ax = a(1 − 2x)
Thus,
f 0 (0) = a > 0 and f 0 (1) = −a < 0
(i) f 0 (0) > 0 implies that slopes must increase through x = 0. That is,
slopes (dx/dt) are negative below x = 0 and positive above it. There-
fore x = 0 is a source.
(ii) f 0 (1) < 0 implies that slopes must decrease through x = 1. So slopes
are positive for x < 1 and negative for x > 1. Hence x = 1 is a sink.
This is illustrated in the Figure ... below

22 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Example 12. Consider the differential equation

dx
= g(x) = x − x3 .
dt
Find the equilibrium points. Draw the phase line and the solution curves of
the equation.
Solution.
g(x) = x − x3 = x(1 − x2 ) = x(1 − x)(1 + x)
For equilibrium solutions, we set g(x) = x(1 − x)(1 + x) = 0 to get

x = 0, 1, −1

as the equilibrium solutions or stationary (fixed) points. Now

g 0 (x) = 1 − 3x2 .

=⇒ g 0 (0) = 1 > 0
So the equilibrium solution x = 0 is a source.

g 0 (1) = −2 < 0,

implies that the fixed point x = 1 is a sink.

g 0 (−1) = −2 < 0

imples that the equilibrium solution x = −1 is also a sink. From the in-
formation above, we can draw the phase line, and subsequently the solution
curves of the differential equation as displayed in Figure...

2.4.2 The Logistic Model with Constant Harvesting &

Bifurcation
Here we want to analyze the logistic population model with the parameter
a = 1, and we introduce a new parameter, h, which represents the rate of
harvesting. This results in the equation
dx
= x(1 − x) − h. (2.18)
dt
We want to subject the equation above to the qualitative analysis learned so
far to determine the impact of harvesting on the population dynamics. Let

fh (x) = x(1 − x) − h = −x2 + x − h

23 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

For equilibrium solutions, we let fh (x) = 0 to get

x2 − x + h = 0
√
1 ± 1 − 4h
=⇒ x = .
2
We note the following points

(i) If 1 − 4h > 0 (i.e. 0 ≤ h < 1/4), we get two equilibrium solutions:

√ √
1 − 1 − 4h 1 + 1 − 4h
x = xl = , x = xr =
2 2

(ii) If 1 − 4h = 0 (i.e. h = 1/4), we get one equilibrium solution at

1
x= .
2

(iii) If 1 − 4h < 0 (i.e. h > 1/4), we get no real solutions (complex roots).

A graph of fh (x) as a function of x is shown in Figure ... for the different

regions of h. First note from the graph that, for h > 1/4, we have fh (x) < 0
irrespective of the value of x and so slopes (dx/dt) must decrease in this
region. Secondly, for h = 1/4, slopes decrease for all values of x except at
the single fixed point x = 1/2. Thirdly, when 0 ≤ h < 1/4, we have fh (x)
crossing the x−axis at the fixed points xl and xr , and solutions are similar
to the case without harvesting: slopes increase for xl < x < xr and decrease
for x < xl and x < xr .
The analysis above is succintly captured in a diagram called the bifur-
cation diagram. It is a plot of the dependent variable (x in this case) on
the vertical axis versus the parameter under investigation (h in this case)
in the horizontal axis. It is a plot of a series of phase lines for the different
values of h. Figure ... shows the bifurcation diagram for the case of constant
harvesting. We see that h = 1/4 is a critical number called the bifurcation
value. This is because for all values of h such that 0 ≤ h < 1/4, we get two
fixed points in which the solution xr is a sink and xl is a source. However, at
h = 1/4, the two fixed points appear to merge into a single fixed point with
slopes decreasing above and below x = 1/2. Also, just beyond h = 1/4, all
solutions decrease to negative infinity.
Ecologically, this implies that if the rate of harvesting is greater that h =
1/4, the specie would go extinct! That is why it is important to determine
bifurcation values for differential equations.

24 c Dr. Joseph K. Ansong

Chapter 3

Linear Systems

3.1 Introduction
Consider the linear system of the form
x1 = a11 (t)x1 + a12 (t)x2 + + a1n (t)xn + b1 (t)
x2 = a21 (t)x1 + a22 (t)x2 + + a2n (t)xn + b2 (t)
···
xn = an1 (t)x1 + an2 (t)x2 + + ann (t)xn + bn (t),
where the functions aij , 1 ≤ i, j ≤ n, bi , 1 ≤ i ≤ n are continuous real-valued
functions on the interval I. The equations above may be written in vector
equation form as
x0 (t) = A(t)x(t) + b(t) (3.1)
where
   
x1 x01
 x2   x02 
x= , x0 = 
   
.. .. 
 .   . 
xn x0n

   
a11 (t) · · · a1n (t) b1 (t)
A(t) =  .. .. .. b(t) =  ...  .
, (3.2)
   
. . .
an1 (t) · · · ann (t) bn (t)

Note that equation (3.1) also applies to nth order linear ODEs of the form
y (n) + pn−1 (t)y (n−1) + · · · + p0 (t)y = r(t)

25
Math 450 : Differential Equations II J.K.A

To see this, we let

xi (t) = y (i−1) (t), for t ∈ I, 1 ≤ i ≤ n

Then, with
 
x1
 x2 
x=
 
.. 
 . 
xn

we have that
   
0 1 0 ··· 0 0

 0 0 1 ··· 0 


 0 

A(t) =  .. .. ... ... .. ..
, b(t) = 
   
 . . .   . 

 0 0 0 1   0 
−p0 (t) −p1 (t) −p2 (t) · · · −p(n−1)(t) r(t)
(3.3)
NOTE:
1) A matrix function is continuous on an interval I if and only if all of its
entries are continuous on I.
2) The matrix function A is called the companion (or coefficient) matrix
of the differential equation (3.1).
3) From the Picard-Lindelof theorem for systems of equations, the deriva-
tive matrix is given by

Df (t, x) = A(t).

Since A is assumed to be continuous, we expect an initial value problem

for an equation of the form (3.1) to have a unique solution in some
interval containing the initial point t0 . The theorem below gives an
even stronger result for linear systems.
Theorem 2. Assume that the n × n matrix function A and the n × 1 vector
function b are continuous on an interval I. Then the IVP

x = A(t)x + b(t), x(t0 ) = x0 ,

where t0 ∈ I and x0 is a given constant n × 1 vector, has a unique solution

that exists on the whole interval I.

26 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Example 13. Consider the linear sytem

x01 = x2 + t
x02 = −x1 + 1.
It can easily be shown that the functions x1 (t) = 2 + sin(t) and x2 (t) =
−t + cos(t) for t ∈ R is a solution on R of the system. Equivalently, the
vector function  
x1
x :=
x2
is a solution on R of the vector equation
   
0 0 1 t
x = x+ .
−1 0 1
Definition 5. A family of functions F defined on an interval I is said to be
a vector space or linear space provided whenever x, y ∈ F it follows that
for any constants α, β ∈ R
αx + βy ∈ F
That is
(αx + βy)(t) := αx(t) + βy(t), t∈I
Definition 6. If F and G are vector spaces of functions defined on an interval
I, then L : F −→ G is called a linear operator provided
L[αx + βy] = αL[x] + βL[y], (3.4)
for all α, β ∈ R, x, y ∈ F .
Example 14. Let F be the set of all n × 1 continuously differentiable vector
functions on an interval I and let G be the set of all n × 1 continuous vector
functions on an interval I and note that F and G are linear spaces. Define
L : F −→ G by
Lx(t) = x0 (t) − A(t)x(t),
for t ∈ I, where A is a given n × n continuous matrix function on I. Show
that L is a linear operator.
Solution. Let α, β ∈ R, let x, y ∈ F , and consider
L[αx + βy](t) = (αx + βy)0 (t) − A(t)(αx + βy)(t)
= αx0 (t) + βy 0 (t) − αA(t)x(t) − βA(t)y(t)
= α[x0 (t) − A(t)x(t)] + β[y 0 (t) − A(t)y(t)]
= αLx(t) + βLy(t)
= (αLx + βLy)(t),
for t ∈ I. So L[αx + βy] = αLx + βLy. Hence L is a linear operator.

27 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Remark. (1) Note that the differential equation (3.1) can be written in
the form
Lx = x0 (t) − A(t)x(t) = b
where L is the linear operator defined in (3.4), and is called a linear
vector differential equation.
(2) If b is not the trivial vector function (b 6= 0), then the equation

Lx = b

is called the nonhomogeneous linear vector differential equation. If

Lx = 0, the equation is called the homogeneous linear vector differ-
ential equation (LVDE).

3.2 Homogeneous Linear Systems

Consider the nonhomogeneous linear vector differential equation

x0 (t) = A(t)x(t) + b(t) (3.5)

To solve (3.5) we need to first solve its homogeneous LVDE:

x0 (t) = A(t)x(t) (3.6)

Theorem 3. Suppose there are exactly n constant n×1 vectors ψ1 , ψ2 , · · · , ψn

and M is the column matrix M = [ψ1 ψ2 · · · ψn ]. Then ψ1 , ψ2 , · · · , ψn are lin-
early dependent iff detM = 0.
Example 15. Consider the vectors
     
1 2 −4
ψ1 =  2  , ψ2 =  1  , ψ3 =  1  .
−3 −1 −3
Let  
1 2 −4
A= 2 1 1 
−3 −1 −3
Since detA = 0, the vectors are linearly dependent by the Theorem above.
Definition 7. Let the n × 1 vector functions φ1 , φ2 , · · · , φk be defined on an
interval I. The k vector functions are linearly dependent on I provided there
exists C1 , C2 , · · · , Ck ∈ R not all zero such that C1 Φ1 (t), C2 Φ2 (t), · · · , Ck Φk (t) =
0, ∀t ∈ I. Otherwise, the k vector functions are linearly independent on I.

28 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Note that the n × 1 vector functions φ1 , φ2 , · · · , φn are linearly indepen-

dent on an interval I provided that the only constants C1 , C2 , · · · , Ck that
satisfy the equation C1 φ1 (t), C2 φ2 (t), · · · , Ck φk (t) = 0, for all t ∈ I, are
C1 = C2 = · · · = Ck = 0.
Example 16. Show that the three vector functions φ1 , φ2 , φ3 defined by
   2   3 
t t t
φ1 (t) = , φ2 (t) = , φ3 (t) =
t t t

are linearly independent on any nondegenerate interval I (any interval con-

taining at least two points).
Solution. Assume c1 , c2 , c3 are constants such that

c1 φ1 + c2 φ2 + c3 φ3 = 0,

for all t ∈ I. Then

       
t t2 t3 0
c1 + c2 + c3 = ,
t t t 0
for all t ∈ I. This implies that

c1 t + c2 t2 + c3 t3 = 0,

for all t ∈ I. Taking three derivatives of both sides of the equation above
results in 6c3 = 0, giving c3 = 0. Letting c3 = 0 in the equation and taking
two derivatives of the resulting equation yields c2 = 0. It then follows that
c1 = 0. Hence the three vector functions are linearly independent on I.
Theorem 4. The linear vector differential equation (3.6) has n linearly in-
dependent solutions on I, and if Φ1 , Φ2 , · · · , Φn are n linearly independent
solutions on I, then

x = C1 Φ1 + C2 Φ2 + · · · + Cn Φn

for t ∈ I, where C1 , C2 , · · · , Cn are constants, is a general solution for (3.6).

Definition 8. Let A be a given n × n constant matrix and let x be a column
unknown n−vector.
(i) For any number λ, the vector equation

Ax = λx (3.7)

has the solution x = 0 called the trivial solution.

29 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

(ii) If λ0 is a number such that the vector equation (3.7) with λ replaced
by λ0 has a nontrivial solution x0 , then λ0 is called an eigenvalue of A
and x0 is called a corresponding eigenvector. (λ0 , x0 ) is an eigen pair
of A.

Remark. Assume λ is an eigenvalue of A, then equation (3.7) has a non-

trivial solution. Therefore,

(A − λI)x = 0

has a nontrivial solution. From linear algebra we get that the characteristic
equation
det(A − λI) = 0
is satisfied. If λ0 is an eigenvalue, then to find a corresponding eigenvector
we want to find a nonzero vector x so that Ax = λ0 x or, equivalently, (A −
λ0 I)x = 0.

Theorem 5. If λ0 , x0 is an eigen pair for the constant n × n matrix A, then

x(t) = eλ0 t x0 , t ∈ R defines a solution x of x0 (t) = Ax on R.

Proof. Let x(t) = eλ0 t x0 , then

x0 (t) = λ0 eλ0 t x0

x0 (t) = eλ0 t λ0 x0
= eλ0 t Ax0
= Aeλ0 t x0
= A(t)x
for t ∈ R.

Example 17. Solve the differential equation

 
0 0 1
x = x
−2 −3
 
0 1
Solution. The companion matrix is A = . The characteristic
−2 −3
equation of A is

−λ 1
det(A − λI) = =0
−2 −3 − λ

30 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

=⇒ λ2 + 3λ + 2 = (λ + 2)(λ + 1) = 0.
Thus, the eigen values are λ1 = −2, λ2 = −1. We find its corresponding
eigenvectors for λ1 = −2 :

(A − λ1 I)x = (A + 2I)x = 0.
    
2 1 x1 0
= =0
−2 −1 x2 0
   
x1 1
=
x2 −2
 
1
So −2, is an eigen pair for A.
−2
For λ2 = −1
(A − λ2 I)x = (A + I)x = 0.
    
1 1 x1 0
= =0
−2 −2 x2 0
   
x1 1
=
x2 −1
 
1
So −1, is an eigen pair for A. So the Theorem on solutions of LVDE,
−1
the vector functions Φ1 , Φ2 defined by
   
−2t 1 −t 1
Φ1 (t) = e , Φ2 (t) = e
−2 −1

are solutions in R. Since the vector functions Φ1 , Φ2 are linearly independent

on R, we have the general solution
   
−2t 1 −t 1
x(t) = C1 e + C2 e , t ∈ R.
−2 −1

Theorem 6. If x = U + iV is a complex vector-valued solution of (3.6),

where U, V are real vector-valued functions, then U, V are real vector-valued
solutions of (3.6).

Proof. Suppose x = U + iV is a complex vector-valued solution of (3.6).

Then
x0 (t) = U 0 (t) + iV 0 (t)
=⇒ A(t)x = A(t)[U (t) + iV (t)],

31 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

for t ∈ I. Equating real and imaginary parts, we have that

U 0 (t) = A(t)U (t),

V 0 (t) = A(t)V (t)

for t ∈ I.
Example 18. Solve the differential equation
 
0 3 1
x (t) = x.
−13 −3

Solution. The characteristic equation of the companion matrix is

λ2 + 4 = 0

Thus, the eigenvalues are

λ1 = 2i, λ2 = −2i.

For λ1 = 2i,
(A − 2iI)x = 0
    
3 − 2i 1 x1 0
=
−13 −3 − 2i x2 0
=⇒ (3 − 2i)x1 + x2 = 0
   
x1 1
=⇒ =
x2 −3 + 2i
 
1
Thus, 2i, is an eigen pair for A. Therefore, the vector function
−3 + 2i
Φ defined by  
2it 1
Φ(t) = e ,
−3 + 2i
 
1
= [cos(2t) + i sin(2t)] ,
−3 + 2i
   
cos(2t) sin(2t)
= +i ,
−3 cos(2t) + 2 sin(2t) 2 cos(2t) − 3 sin(2t)
is a solution.
By the Theorem above, the vector functions Φ1 , Φ2 defined by
 
cos(2t)
Φ1 =
−3 cos(2t) + 2 sin(2t)

32 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

 
sin(2t)
Φ2 =
2 cos(2t) − 3 sin(2t)
are real vector-valued solutions of the D.E. Since Φ1 , Φ2 are linearly indepen-
dent on R by the Theorem above, the general solution is given by

x(t) = C1 Φ1 (t) + C1 Φ2 (t)

   
cos(2t) sin(2t)
x(t) = C1 + C2
−3 cos(2t) + 2 sin(2t) 2 cos(2t) − 3 sin(2t)
for all t ∈ R.

Remark. If the matrix A has n linearly independent eigenvectors, then The-

orem 5 can be used to generate a general solution of x0 = Ax. However, it
is possible for an n × n constant matrix to have fewer than n linearly in-
dependent eigenvectors, for example, when the characteristic equation has
repeated eigenvalues. We would examine such cases later in the course.

Definition 9. The matrix differential equation of the linear vector differen-

tial equation (3.6) is defined as

X 0 = A(t)X (3.8)

where
   
x11 x12 ··· x1n x011 x012 ··· x01n
 x21 x22 ··· x2n  
0
x021 x022 ··· x02n 
X :=  and X := 
   
.. .. ..  .. .. .. 
 . . ··· .   . . ··· . 
xn1 xn2 ··· xnn xn1 x0n2
0
··· x0nn

are n × n matrix variables and A is a given n × n continuous matrix function

on an interval I.

Definition 10. A matrix function Φ is a solution of (3.8) on I provided Φ is

a continuously differentiable n×n matrix function on I and Φ0 (t) = A(t)Φ(t),
for t ∈ I.

Theorem 7. Suppose A is a continuous n × n matrix function on an interval

I and assume that Φ defined by

Φ(t) = [φ1 (t), φ2 (t), · · · , φn (t)], t ∈ I,

is the n × n matrix function with columns φ1 (t), φ2 (t), · · · , φn (t). Then Φ

is a solution of the matrix differential equation (3.8) on I iff each column

33 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

φi is a solution of the vector differential equation (3.6) on I for 1 ≤ i ≤ n.

Furthermore, if Φ is a solution of the matrix differential equation (3.8), then

x(t) = Φ(t)c

is a solution of the vector differential equation (3.6) for any constant n × 1

vector c.

Remark. The set of solutions {φ1 , φ2 , · · · , φn } that are linearly independent

on I is called a fundamental solution set for (3.6) on I, and the matrix
Φ(t) is called a fundamental matrix for (3.6).

Theorem 8 (Existence-Uniqueness Theorem). Assume A is a continuous

matrix function on an interval I. Then the IVP

X = A(t)X, X(t0 ) = X0 ,

where t0 ∈ I and X0 is an n × n constant matrix, has a unique solution X

that is a solution on the whole interval I.

34 c Dr. Joseph K. Ansong

Chapter 4

Planar Linear Systems

Consider the linear vector differential equation

x0 = Ax (4.1)
where A is a 2 × 2 constant matrix and x is a 2 × 1 constant vector. Suppose
that a real eigenvalue λ1 has multiplicity 2 and that v is the only eigenvector
corresponding to this eigenvalue. Then
Av = λ1 v
and
y1 (t) = eλ1 t v
is a solution of (4.1). To find a second linearly independent solution, we try
y2 (t) = teλ1 t v + eλ1 t w (4.2)
where w is a constant 2 × 1 vector. Subtituting (4.2) into (4.1), we get
eλ1 t v + λ1 teλ1 t v + λ1 eλ1 t w = A teλ1 t v + eλ1 t w

=⇒ eλ1 t (v + λ1 w) + λ1 teλ1 t v = Ateλ1 t v + Aeλ1 t w

Comparing coefficients results in v + λ1 w = Aw, so that
(A − λ1 I)w = v, (4.3)
and Av = λ1 v, giving rise to
(A − λ1 I)v = 0. (4.4)
Note that (4.4) means that λ1 is an eigenvalue corresponding to the eigen-
vector v which we already know. We find that our guess solution (4.2) is
actually the second solution to (4.1) if equation (4.3) is satisfied. A general
form of (4.3) would be derived later for a general linear system.

35
Math 450 : Differential Equations II J.K.A

Example 19. 1. Find the general solution of the linear system

x0 = Ax

where  
1 9
A= .
−1 −5

2. Solve the IVP

   
0 7 1 2
x (t) = x(t), x(0) = .
−4 3 −5
 
1 9
Solution. The companion matrix is A = . The characteristic
−1 −5
equation of A is

1−λ 9
det(A − λI) = = 0,
−1 −5 − λ

−(1 − λ)(5 + λ) + 9 = 0,
=⇒ 5 + λ − 5λ − λ2 − 9 = 0,
=⇒ λ2 + 4λ + 4 = 0,
=⇒ (λ + 2)2 = 0
Thus, the eigenvalues are λ1 = −2, λ2 = −2 (repeated). We find the corre-
sponding eigenvector to λ1 = −2 as

(A − λ1 I)x = (A + 2I)x = 0.
    
3 9 x1 0
=
−1 −3 x2 0
x1 + 3x2 = 0, =⇒ x2 = 1, x1 = −3
   
x1 −3
=⇒ =
x2 1
 
−3
So −2, is an eigen pair for [Link], one solution of the system is
1
 
−2t −3
y1 (t) = e .
1

36 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

To obtain a second linearly independent solution y2 (t), we let

 
−2t −3
y2 (t) = te + e−2t w
1

where 
−3
(A − λ1 I)w = .
1
    
3 9 w1 −3
=
−1 −3 w2 1
w1 + 3w2 = −1. w2 = 0 =⇒ w1 = −1.
   
w1 −1
=⇒ =
w2 0
   
−2t −3 −2t −1
∴ y2 (t) = te +e
1 0
 
−3t − 1
=⇒ e−2t
t
The general solution is given by
   
−2t −3 −2t −3t − 1
y = C1 e + C2 e .
1 t

4.1 Geometric Methods for Planar Systems

Consider the linear two dimensional autonomous system
dx
= ax + by = f (x, y)
dt
dy
= cx + dy = g(x, y) (4.5)
dt
where a, b, c, d ∈ R. System (4.5) may be written in the compact form

x0 (t) = Ax(t) (4.6)

where
 
a b
A= .
c d

Definition 11. Every solution of (4.6) may be written as a curve, say φ(t) =
(x(t), y(t)), in the plane. The solution curves are called trajectories or orbits.

37 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Remark. (a) The existence and uniqueness theorem guarantees that tra-
jectories do not cross.

(b) The qualitative behabiour of solutions is determined by plotting a few

trajectores.
Definition 12. The phase portrait or phase plane is a 2D figure showing
how the qualitative behaviour of system (4.5) as x and y vary with t.
dy
Definition 13. The direction field or vector field gives the gradient and
dx
direction vectors of the trajectories in the phase plane.
NOTE: The slope of the trajectories can be determined using the chain
rule
dy y0 g(x, y)
= 0 = (4.7)
dx x f (x, y)
and the direction of the vector field is given by x0 and y 0 at each poin in the
xy−plane. Equation (4.7) is called the phase plane equation.
dy
Definition 14. The contour lines for which is a constant are called
dx
isoclines.
Definition 15. A point (x0 , y0 ) where f (x0 , y0 ) = 0 and g(x0 , y0 ) = 0 is
called a critical point, or equilibrium point, of the system (4.5), and the
corresponding constant solution x(t) = x0 , y(t) = y0 is called an equilibrium
solution. The set of all critical points is called the critical point set.
Remark. The point (0, 0) is always an equilibrium point of (4.6). If both
eigenvalues of A are nonzero (so that A is invertible) then (0, 0) is the only
equilibrium point, but if one eigenvalue vanishes we have whole line of equi-
librium points.
We discuss the nature of these equilibria by first considering systems in
a canonical form where A is in one of the following forms;
   
λ1 0 α β
A1 = , A2 =
0 λ2 −β α
   
λ1 0 λ1 µ
A3 = , A4 =
0 λ1 0 λ1
where λ1,2 , α, β, and µ are real constants. A1 has two real distinct eigenvalues,
A2 has complex eigenvalues, and A3 and A4 have repeated eigenvalues.

38 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

4.2 Phase Portraits

Consider the planar system
x0 = Ax
We first consider the case of real and distinct eigenvalues.

4.2.1 Real Distinct Eigenvalues

The three consider to consider are

(a) λ1 < 0 < λ2

(b) λ1 < λ2 < 0

(c) 0 < λ1 < λ2

Case (a): The Saddle

 
λ1 0
A= , λ1 < 0 < λ2
0 λ2
Solving the characteristic equation, we have

|A − λI| = 0,

λ1 − λ 0
=⇒ =0
0 λ2 − λ
=⇒ (λ1 − λ)(λ2 − λ) = 0,
=⇒ λ = λ1 , λ2
Now,
AV0 = λV0
 
x
where V0 = is an eigenvector. For the eigenvector λ1 :
y
    
λ1 0 x1 x1
=⇒ = λ1
0 λ2 x2 x2

=⇒ λ1 x1 = λ1 x1 =⇒ x1 = 1,
where any value of x1 (not just x1 = 1) satisfies the equation; we chose x1 = 1
for simplicity. Also
λ2 x2 = λ1 x2 =⇒ x2 = 0,

39 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

 
1
since λ1 , λ2 are distinct. Thus, an eigenvector associated with λ1 is .
0
For λ2 , we have
    
λ1 0 x1 x1
=⇒ = λ2
0 λ2 x2 x2

=⇒ λ1 x1 = λ2 x1 =⇒ x1 = 0,
and
λ2 x2 = λ2 x2 =⇒ x2 = 1,
where we again
 chose x2 = 1 for simplicity. Thus, an eigenvector associated
0
with λ2 is .
1
Thus, the two solutions of the system are
   
λ1 t 1 λ2 t 0
X1 (t) = e , and X2 (t) = e .
0 1

Hence, the general solution is given by

X(t) = αX1 (t) + βX2 (t)

   
λ1 t 1 λ2 t 0
∴ X(t) = αe + βe ,
0 1
where α, β are constants.
 
1 λ1 t
Remark. Note that the solution X1 (t) = e lies along the x−axis
0
T
(i.e the direction
  of the eigenvector [1, 0] ) since y = 0, and because λ1 < 0,
0
X1 (t) → as t → ∞. This solution is called the stable line or stable
0
manifold (blue line in Figure 4.1). The solution X1 also corresponds
 to  the
0
general case when β = 0. Similarly, the solution X2 (t) = eλ2 t lies
1
along the y−axis because x = 0, and since λ2 > 0, X2 (t) → ∞ as t → ∞.
Solution X2 is called unstable line or unstable manifold. All other solutions
with α, β 6= 0 tend to ∞ in the direction of the unstable line since eλ2 t is
dominant. As t → −∞, solutions tend toward the stable line since eλ1 t
becomes dominant.

40 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Figure 4.1: A saddle.

Derivation of analytical curves

The type of orbits for a system may also be derived analytically as illustrated
below. Let

x(t) = αeλ1 t , y(t) = βeλ2 t

Then,

dx dy
= αλ1 eλ1 t , = βλ2 eλ2 t
dt dt
dy dy/dt βλ2 eλ2 t
=⇒ = =
dx dx/dt αλ1 eλ1 t
dy λ2 y
=⇒ =
dx λ1 x

41 c Dr. Joseph K. Ansong

 Math 450 : Differential Equations II J.K.A

Separating variables and integrating

dy λ2 dx
=
y λ1 x
λ2
=⇒ ln |y| = ln |x| + ln C
λ1
=⇒ ln |y| = ln |x|λ2 /λ1 + ln C
y
=⇒ ln | λ2 /λ1 | = ln C
x
=⇒ |y| = C |x|λ2 /λ1
where C > 0 is a constant of integration.

Thus, the nature of the orbits depend on λ1 and λ2 .

Remark.
(1) The Saddle: If λ1 6= λ2 , with different signs, we have
|y| = C|x|−|λ2 /λ1 | .
So the orbits are hyperbolic, as demonstrated above.
(2) The Node: If λ1 6= λ2 , but they have the same sign, then
|y| = C|x|λ2 /λ1
In this case the orbits are parabolic in nature.
(3) The Star: If λ1 = λ2 , then we have
y = Cx
The orbits are straight lines through the origin. The orbit is a stable star
for λ1 < 0 and unstable star for λ1 > 0.
Example 20 (Case 1: The Saddle). Let λ1 = −1, λ2 = 2. Then
C
|y| =
x2
The graph is shown in Figure (4.2a).
If λ1 = 1, λ2 = −2, then
C
|y| = C|x|−2 = .
|x|2
The graph is shown in Figure (4.2b).

42 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Figure 4.2: A saddle for (a) λ1 = −1, λ2 = 2, and (b) λ1 = 1, λ2 = −2.

Example 21 (Case 2: The Node). Let λ1 = −3, λ2 = −1. Then

C
|y| =
|x|1/3
The graph is shown in Figure (4.3a)
The equilibrium point is an stable node or a sink.
If λ1 = 1, λ2 = 2, then
|y| = C|x|2
The graph is shown in Figure (4.3b). The equilibrium point is an unstable
node or a source.

How do solutions approach (0, 0)?

How do solutions approach (0, 0) as t → ∞? In general, if λ1 < λ2 < 0 in
which λ1 has eigenvector (u1 , u2 )T and λ2 has eigenvector (v1 , v2 )T , then the
general solution is given by
   
λ1 t u1 λ2 t v1
X(t) = αe + βe .
u2 v2
The slope of the solution is given by
dy λ1 αeλ1 t u2 + λ2 βeλ2 t v2
= ,
dx λ1 αeλ1 t u1 + λ2 βeλ2 t v1
dy λ1 αe(λ1 −λ2 )t u2 + λ2 βv2 e−λ2 t
=⇒ = ×
dx λ1 αe(λ1 −λ2 )t u1 + λ2 βv1 e−λ2 t

43 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Figure 4.3: (a) A stable node for λ1 = −3, λ2 = −1, and (b) unstable node
for λ1 = 1, λ2 = 2.

dy λ1 αe(λ1 −λ2 )t u2 + λ2 βv2

=⇒ = .
dx λ1 αe(λ1 −λ2 )t u1 + λ2 βv1
 
dy v2
As t → ∞, → , which is the slope of the eigenvector λ2 . Hence all
dx v1
solutions tend to (0, 0) tangentially to the straight-line solution for λ2 (the
y−axis in this case). If 0 < λ2 < λ1 , we get the same phase portrait as before
but with the arrows reversed since
 
dy v2
→ , as t → −∞.
dx v1
If 0 < λ1 < λ2 , we have
dy λ1 αu2 + e(λ2 −λ1 )t
=
dx λ1 αu1 + e(λ2 −λ1 )t
and as t → −∞, we have  
dy u2
→
dx u1
which is the slope of the eigenvector λ1 . Hence, in this case, solutions are
tangential to the x−axis as t → −∞.
Example 22 (Case 3: The Star). Let λ1 = λ2 = −3, λ2 = −1. Then
y = Cx

44 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Figure 4.4: (a) A unstable star for λ1 = λ2 > 0, and (b) a stable star for
λ1 = λ2 < 0.

The graphs in Figure (4.4) depict stable and unstable stars.

Example 23. Solve the system

 
0 1 3
x (t) = Ax, where A =
1 −1
Solution. It is easy to show thatthe eigenvalues are λ = ±2. For λ1 = 2,
3
the corresponding eigenvector is . For λ2 = −2, the corresponding
  1
1
eigenvector is . Thus, the general solution is given by
−1
   
2t 3 −2t 1
x(t) = αe + βe .
1 −1

The graph is displayed in Figure (4.5).

Note that the system is given by
dx dy
= x + 3y, = x − y,
dt dt
dy x−y
=⇒ = ,
dx x + 3y

45 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

This is an exact equation, and may be solved to get the solution

1 3
xy − x2 + y 2 = C,
2 2
where C is a constant. The equation above may also be used to graph the
phase portrait for the system.

Figure 4.5: A saddle with λ1 = 2 and λ2 = −2.

4.2.2 Complex Eigenvalues

Consider the planar system

x0 (t) = Ax(t)

where
 
α β
A= .
−β α

The characteristic equation is

α−λ β
|A − λI| = 0 =⇒ = 0,
−β α − λ

=⇒ (α − λ)2 + β 2 = 0,

46 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

=⇒ α − λ = ±iβ
=⇒ λ = α ± iβ.
The eigenvector corresponding to α + iβ is given by
    
−iβ β x1 0
= ,
−β −iβ x2 0

=⇒ −iβx1 + βx2 = 0,
ix1 = x2 , =⇒ x1 = 1, x2 = i.
 
1
Hence is an eigenvector. Thus, a complex solution is given by
i
 
(α+iβ)t 1
X(t) = e ,
i
   
αt 1 cos βt + i sin βt
αt
= e (cos βt + i sin βt) =e ,
i − sin βt + i cos βt
   
αt cos βt αt sin βt
=e + ie ,
− sin βt cos βt
=⇒ X(t) ≡ XRe (t) + iXIm (t),
where XRe and XIm yield two real solutions. Hence, the general solution
becomes    
αt cos βt αt sin βt
X(t) = C1 e + C2 e .
− sin βt cos βt
Using the eigenvalue α − iβ should yield the same general solution. Try it.
Remark. From the general solution, we get the x, y coordinates as

x = C1 eαt cos βt + C2 eαt sin βt

y = −C1 eαt sin βt + C2 eαt cos βt

(i) If α = 0 (without the eαt term), solutions would wind periodically
around circles centered at the origin. To see this, we calculate

x2 + y 2 = (C1 cos βt + C2 sin βt)2 + (−C1 sin βt + C2 cos βt)2

=⇒ x2 + y 2 = C12 + C22 = R2 ,
which is the equation of a circle centered at (0, 0). The equilibrium
point in this case is called a center.

47 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

(ii) If α 6= 0, the term eαt causes solutions to spiral into the origin for α < 0
or away from the origin for α > 0. For the case α < 0, the equilibrium
point is a spiral sink, and it is a spiral source for α > 0. The spiral
sink is called a stable focus and the spiral source is unstable focus.
(iii) How are the circles or spirals traversed? To answer this question, we
need to re-write the system in polar coordinates:
 0    
x α β x
=
y0 −β α y
=⇒ x0 (t) = αx(t) + βy(t)
=⇒ y 0 (t) = −βx(t) + αy(t)
Let x = r cos θ and y = r sin θ. Then r(t)2 = x(t)2 + y(t)2 .
=⇒ 2rr0 = 2xx0 + 2yy 0
=⇒ rr0 = x(αx + βy) + y(−βx + αy),
=⇒ rr0 = αx2 + βxy − βxy + αy 2 ,
=⇒ rr0 = α(x2 + y 2 ) = αr2

∴ r0 = αr (4.8)
Notice that the solution to (4.8) is r = Ceαt , where C is a constant.
This also indicates the circular (for α = 0) and spiral (for α 6= 0) nature
of the solutions.
Now,
y(t)
tan θ(t) =
x(t)
Differentiating both sides yields,
xy 0 − yx0
sec2 θ · θ0 = .
x2
Since sec2 θ = 1 + tan2 θ, we get
r2 θ0 = xy 0 − yx0 = x(−βx + αy) − y(αx + βy),
=⇒ r2 θ0 = −βx2 + αxy − αxy − βy 2 ,
=⇒ r2 θ0 = −βr2 ,

∴ θ0 = −β. (4.9)
Hence

48 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

(a) If θ0 > 0 (i.e. β < 0), then the trajectories spiral counterclockwise
around the origin.

(b) If θ0 < 0 (i.e. β > 0), then we get clockwise trajectories around the
origin.

Figures (4.6) and (4.6) display various phase portraits for complex eigenval-
ues.

Figure 4.6: Spiral sink (stable focus) (a) α < 0, β > 0, and (b) α < 0, β < 0.
Spiral source (unstable focus) (c) α > 0, β < 0, and (d) α > 0, β > 0

49 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

4.2.3 Repeated Eigenvalues: Case 1

 
λ1 0
A=
0 λ1
|A − λI| = 0 =⇒ (λ1 − λ)2 = 0
=⇒ λ = λ1 , repeated.
For an eigenvector, we solve (A − λ1 I)V = 0
 
0 0
=⇒ V=0
0 0
Thus, every nonzero vector V is an eigenvector. So solutions are of the form
X(t) = αeλ1 t V, λ = λ1 .
Solutions are straight lines through the origin; tending to (0, 0) for λ < 0 and
away from (0, 0) for λ > 0. Alternatively, we could solve the system directly
x0 = λ1 x =⇒ x = C1 eλ1 t
y 0 = λ1 y =⇒ y = C2 eλ1 t
       
x λ1 t C1 λ1 t 1 λ1 t 0
=⇒ =e = C1 e + C2 e
y C2 0 1
The solution is referred to as a degenerate node or a star. Example
solutions curves are displayed in Figure 4.4.

4.2.4 Repeated Eigenvalues: Case 2

 
λ1 1
A=
0 λ1
|A − λI| = 0 =⇒ (λ1 − λ)2 = 0
=⇒ λ = λ1 , repeated.
For an eigenvector, we solve (A − λ1 I)V = 0
  
0 1 v1
=⇒ =0
0 0 v2
=⇒ v2 = 0
 
1
So is an eigenvector, and a solution to the system is given by
0
 
λ1 t 1
X1 (t) = αe .
0

50 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

For a second linearly independent solution to the system, we can solve the
system directly:
x0 = λ 1 x + y
y 0 = λ1 y =⇒ y = C2 eλ1 t
=⇒ x0 = λ1 x + C2 eλ1 t
=⇒ x0 − λ1 x = C2 eλ1 t
This is a linear first order differential equation which can be solved using an
integrating factor, given by
R
µ = e− λ1 dt
= e−λ1 t
d −λ1 t

=⇒ e x = C2
dt
=⇒ e−λ1 t x = C2 t + C1
=⇒ x = c2 teλ1 t + C1 eλ1 t
Thus,    
x C1 eλ1 t + C2 teλ1 t
=
y C2 eλ1 t
     
x λ1 t 1 λ1 t t
∴ = C1 e + C2 e .
y 0 1
This is, in fact, the general solution. It may be written as
      
λ1 t 1 λ1 t 1 0
X(t) = C1 e + C2 e t + .
0 0 1

Alternatively,

X(t) = C1 eλ1 t V1 + C2 eλ1 t [tV1 + V2 ] , (4.10)

   
1 0
where V1 = and V2 = .
0 1
Remark. (i) If λ1 < 0, each term of the solution tends to zero as t → ∞.
Note that
t t
lim teλ1 t = lim −λ1 t = lim |λ1 |t ,
t→∞ t→∞ e t→∞ e

and by applying Lopital’s rule, we have

1
lim = 0.
t→∞ |λ1 |e|λ1 |t

51 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

(ii) If λ1 > 0, solutions tend away from (0, 0).

(iii)
dy C2 λ1
= (4.11)
dx C2 + C2 λ1 t + C1 λ1
dy
Thus, → 0 as t → ∞ for nonzero C2 , λ1 . This implies that slopes
dx
tend to be flat as t → ∞. That is, solutions tend toward
 or  away from
1
the origin in a direction tangent to the eigenvector . Alterna-
0
tively, note from equation (4.10) that as t → ∞, the dominant
 term is
1
X ≈ C2 eλ1 t tV1 . So solutions are tangent to V1 = . Notice that
0
equation (4.11) may also be written as
dy C 2 λ1
= , (4.12)
dx C2 + C2 ln Cy2 + C1 λ1

where we used y = C2 eλ1 t . This may be solved by separating variables

to obtain y as a function of x. If λ1 < 0, the equilibrium is a stable
improper node, and if λ1 > 0, it is a unstable improper node.
Figure 4.7 shows some solution curves for the case of repeated eigen-
values, as explained above, with λ = −1. Also note that the solution
may be written in the form
 
C1 1 y
x=y + ln
C2 λ1 C2

and may be used to draw the graph in Geogebra as in Figure 4.7.

Direction of Trajectories
Remark. In general

(a) Trajectories in these cases always emerge from (or move into) the origin
in a direction that is parallel to the eigenvector.

(b) They often start in one direction before turning around and moving off
into the other direction.

(c) The directions in which they move are opposite depending on which
side of the trajectory corresponding to the eigenvector we are on.

52 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Figure 4.7: Stable improper node (using λ = −1).

(d) As the trajectory move away from the origin it should start becoming
parallel to the trajectory corresponding to the eigenvector.

The direction of the trajectories for the case of repeated eigenvalues can
be a little confusing. One approach is to compute the direction of vectors
using the coefficient matrix at a few representative points in the plane. For
example, suppose
x0 = Ax
 
−1 1
where A = . Then λ1 = −1.
0 −1     
−1 1 0 1
At the point (0, 1), we get the vector = , which
0 −1 1 −1
is pointing in the southeast direction.     
−1 1 0 −1
At the point (0, −1), we get the vector = ,
0 −1 −1 1
which is pointing in the northwest direction. In this particular case, these
two test should be enough to get the direction of the trajectories. Other
calculations are shown below:     
−1 1 1 0
At the point (1, 1), we get the vector = , which
0 −1 1 −1
is pointing downward.

53 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

    
−1 1 −1 0
At the point (−1, −1), we get the vector = ,
0 −1 −1 1
which is pointing upward.     
−1 1 −1 2
At the point (−1, 1), we get the vector = ,
0 −1 1 −1
which is pointing to the southeast.

Example 24. Find the general solution of the following system, and sketch
the phase portraits.  
0 7 1
x = x
−4 3

Solution. The eigenvalues are λ = 5, 5. For λ = 5, we have

    
2 1 x1 0
=
−4 −2 x2 0

=⇒ 2x1 + x2 = 0 =⇒ x2 = −2x1
 
1
Thus, V = is an eigenvector. For a second eigenvector, we need to
−2
solve
(A − λI)w = V
    
2 1 w1 1
=⇒ =
−4 −2 w2 −2
2w1 + w2 = 1 =⇒ w2 = 1 − 2w1

w1 = 0 =⇒ w2 = 1
 
0
Thus, is an eigenvector, and the general solution is given by
1
      
5t 1 5t 1 5t 0
X(t) = C1 e + C2 te +e
−2 −2 1

To sketch the phase portrait, we will start in the same way that we do for the
real, distinct eigenvalue
 case. We first sketch the trajectory that is parallel to
1
the eigenvector , and since the eigenvalue is positive the trajectory
−2
will be moving away from the origin. The rest of the trajectories are sketched
as explained in the remarks above. The graph is displayed in Figure 4.8.

54 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Figure 4.8: Unstable improper node.

Remark. To sketch the graph using Geogebra, the parametric equations

may be written in the form
 
C1 1 y + 2x
x = (y + 2x) + ln
C2 5 C2
Using different values for C1 and C2 yields the graph for this equation.
The direction of the vectors can be deduced from the these computations
    
7 1 0 1
=
−4 3 1 3

So vectors are pointing upwards (north-east direction) across the positive y

axis.     
7 1 0 −1
=
−4 3 −1 −3
So vectors are pointing downwards (south-west direction) across the negative
y axis.
Example 25. Find the general solution of the following system, and sketch
the phase portraits.  
0 −7 1
x = x
−4 −3

55 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Solution.
−7 − λ 1
|A − λI| = 0 =⇒ =0
−4 −3 − λ
(7 + λ)(3 + λ) + 4 = 0,
=⇒ λ2 + 10λ + 25 = 0,
=⇒ (λ + 5)2 = 0,
∴ λ = −5, −5.
For λ = −5, we have
    
−2 1 x1 0
=
−4 2 x2 0

=⇒ −2x1 + x2 = 0 =⇒ x2 = 2x1
 
1
Thus, V = is an eigenvector. So one solution is
2
 
−5t 1
X1 = e .
2

A second solution is given by

 
−5t 1
X2 = te + e−5t w,
2

where
(A − λI)w = V.
    
−2 1 w1 1
=⇒ =
−4 2 w2 2
−2w1 + w2 = 1 =⇒ w2 = 1 + 2w1
w1 = 0 =⇒ w2 = 1.
 
0
Thus, is an eigenvector, and so
1
   
−5t 1 −5t 0
X2 = te +e .
2 1

The general solution is given by

      
−5t 1 −5t 1 0
X(t) = C1 e + C2 e t +
2 2 1

56 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

For purposes of graphing with a software, we may write the components of

the solution as
x = C1 e−5t + C2 te−5t ,
y = 2C1 e−5t + C2 e−5t (2t + 1).
Eliminating C1 results in

y − 2x 1 y − 2x
e−5t = , t = − ln
C2 5 C2

Hence,  
C1 1 y − 2x
x = (y − 2x) − ln .
C2 5 C2
The graph is displayed in Figure 4.9.

Figure 4.9: Stable improper node.

4.3 The Trace-Determinant Plane

Consider  
a b
c d

57 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

For eigenvalues |A − λI| = 0,

a−λ b
=⇒ = 0,
c d−λ

=⇒ (a − λ)(d − λ) − bc = 0,

=⇒ λ2 − (a + d)λ + (ad − bc) = 0.

The determinant of A is D = Det(A) = ad−bc, and the trace is T = Tr(A) =
a + d. Thus,
λ2 − T λ + D = 0,
√
T ± T 2 − 4D
=⇒ λ± = .
2
So the eigenvalues of A are

1 √ 1 √
λ+ = (T + T 2 − 4D and λ− = (T − T 2 − 4D.
2 2

Remark. (i) The following equations show the relationship between the
eigenvalues, and the trace and determinant of A.
√ √
T + T 2 − 4D T − T 2 − 4D
λ+ + λ− = +
2 2
T T
= + =T
2 2
∴ T = λ+ + λ− .

 √  √ 
T+ T 2 − 4D T− T 2 − 4D
λ+ λ− =
2 2
1 2
T − T 2 + 4D = D


=
4
∴ D = λ+ λ−

(ii) Knowing D and T tells us the eigenvalues of A and therefore virtu-

ally everything about the portraits of x0 (t) = Ax. The advantage of
using this approach is that we do not have to directly compute the
eigenvalues of A before determining the qualitative nature of the phase
portraits. This is because D and T can be obtained directly from A.
The limitations of this approach would become clear shortly.

58 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

(iii) In the trace-determinant plane, a matrix with trace T and determinant

D corresponds to the point with coordinates (T, D). The location of
the point in the T D−plane then determines the geometry of the phase
portrait.

(iv) The sign of the discriminant T 2 − 4D tells us that the eigenvalues are

(a) complex with nonzero imaginary part if T 2 − 4D < 0.

(b) real and distinct if T 2 − 4D > 0.
(c) real and repeated if T 2 − 4D = 0.

Now consider the parabola T 2 − 4D = 0 or D = T 2 /4. We are going to

analyze various cases for the geometry of phase portraits based on the sign
of the discrimant.

Case 1: T 2 − 4D < 0
This case yields complex eigenvalues so the phase portraits are spirals or
a center. Because the real part of the eigenvalues is T /2, we obtain the
following scenarios:

(a) a spiral sink if T < 0,

(b) a spiral source if T > 0,

(c) a center if T = 0.

Case 2: T 2 − 4D > 0
In this case both eigenvalues are real, and result in the following scenarios.

(a) If D < 0, we get a saddle. This is because, since D = λ+ λ− , the

eigenvalues must have different signs for D < 0. Eigenvalues with
opposite signs give rise to saddles.

(b) If D > 0 and T < 0 then we get a sink (this could be a nodal sink
or star sink). Recall that D = λ+ λ− and T = λ+ + λ− . Now D > 0
implies that both eigenvalues must either be positive or both must be
negative, but since T < 0, both eigenvalues must be negative. That is
the reason for the sink in this case.

(c) If D > 0 and T > 0, then we get a source, since both eigenvalues are
positive.

59 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Case 3: D = 0, T 6= 0
In this case we get one zero eigenvalue.

Case 4: D = 0, T = 0
Both eigenvalues vanish in this case. The different cases are summarized in
the TD-plane in Figure with sample plots in the various regions/sections.

Figure 4.10: TD plane with representative plots.

Remark. (i) There are infinitely many different matrices for each point
on the T D−plane. This is because the T D−plane is actually a 2D
representation of a 4D space, since the 2 × 2 matrices are determined
by four parameters.

60 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

(ii) The T D−plane is the analog of the bifurcation diagram for planar
systems. For example, crossing the parabola T 2 − 4D = 0 or the
T −axis results in the phase portrait undergoing a major change in its
geometry.

(iii) Without computing eigenvalues, we can still obtain a lot of information

from D and T . For example, if D < 0, we know immediately that the
geometry is a saddle at the origin. And if both D and T are positive,
we get a source at the origin.

(iv) One limitation with using only D and T to analyze the geometry of pla-
nar systems is that we do not have the luxury of the actual eigenvalues
to help obtain accurate sketches of the phase portraits. For instance,
even though we obtain a saddle for D < 0, we cannot determine the
direction of trajectories from D and T .

61 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

62 c Dr. Joseph K. Ansong

Chapter 5

The Matrix Exponential

Function

5.1 Introduction
Consider the homogeneous system
x0 (t) = Ax(t), (5.1)
where x is a vector and A is a matrix. Following the approach for linear
equations, we assume a solution of the form
x(t) = ert u (5.2)
where r is a constant and u is a constant vector. Substitute (5.2) into (5.1)
to get
rert u = Aert u = ert Au
=⇒ ru = Au
∴ (A − Ir)u = 0. (5.3)
This implies that:
(i) Equation (5.2) is a solution to (5.1) iff (5.3) is satisfied. The r0 s are the
eigenvalues and the nonzero u0 s are called the eigenvectors. We have
already encountered this method in the previous chapters.
(ii) If a matrix A is not symmetric, it is possible for A to have a repeated
eigenvalue but not to have two linearly independent eigenvectors. For
example, the system with
 
1 −1
A= ,
4 −3

63
Math 450 : Differential Equations II J.K.A

has repeated eigenvalues

  r1 = r2 = −1 but all the eigenvectors are of
1
the form u = s . In the case of a 2 × 2 matrix, a second linearly
2
independent eigenvector w can be found by solving

(A − Ir1 )w = u; s = 1,

as seen previously. A general procedure for finding a general solution

of system (5.1) (irrespective of the order of A) is by using the matrix
exponential and generalized eigenvectors.

5.2 The matrix exponential

Motivation: Recall that a first order equation of the form
dx
x0 = = αx,
dt
where α is a constant, has a general solution: x(t) = Ceαt , where C is a
constant value. Thus, we expect a system of equations of the form

x0 (t) = Ax(t) (5.4)

where A is a constant n × n matrix to have a solution of the form

x(t) = eAt · C

where eAt is called the matrix exponential, and C is a constant vector. Hence-
forth, we let A = A, x = x and C = C.
Definition 16. Suppose A is an n × n constant matrix. Then
t2 tn
eAt = I + At + A2 + · · · An + · · · (5.5)
2! n!
for t ∈ R and I is the identity matrix. That is
∞
At
X tn
e = An .
n=0
n!

Similar to the scalar exponential eat , the series (5.5) converges for all t;
but the proof is beyond the scope of this course. The computation of eAt is
more involved if A is not a special type of matrix. We would consider the
special cases before tackling the general case, after outlining some properties
of the matrix exponential:

64 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

5.2.1 Properties of the matrix exponential

Theorem 9. Let A and B be n × n constant matrices and r, s and t be real
(or complex) numbers. Then
(a) eA0 = e0 = I
(b) eA(t+s) = eAt · eAs

−1
(c) eAt = e−At

(d) e(A+B)t = eAt · eBt , provided that AB = BA.

(e) erIt = ert · I
d At
(f) e = AeAt
dt

−1
Remark. (i) eAt = e−At implies that for any A, the matrix eAt has
an inverse for all t. And the inverse is obtained by replacing t by −t.
(ii)
2 3 n
 
d At
d 2t 3t nt
e = I + At + A +A + ··· + A + ···
dt dt 2! 3! n!
t2 tn−1
= A + A2 t + A3 + · · · + An + ···
2! (n − 1)!
t2 tn−1
= A[I + At + A2 + · · · + An−1 + ···]
2! (n − 1)!
d At

=⇒ e = AeAt .
dt
This implies that x(t) = eAt is a solution to the matrix differential
equation
x0 (t) = Ax(t).
Since eAt is invertible, it implies that the columns of eAt are linearly
independent solutions of (5.1). So eAt is a fundamental matrix of (5.1).
Theorem 10. [eAt is a fundamental matrix] If A is an n × n constant
matrix, then the columns of eAt form a fundamental solution set for the
system
x0 (t) = Ax(t).
Therefore eAt is a fundamental matrix for the system, and a general solution
is x(t) = eAt C.
We now return to the question of how eAt can be computed when A is a
special type of matrix, after which we consider the general case.

65 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

5.2.2 Computation of eAt : Special Matrices

The two special matrices to be considered are the diagonal matrix and the
nilpotent matrix.

The Case of a Diagonal Matrix

If A is an n × n diagonal matrix with r1 , r2 , · · · , rn down its main diagonal,
then eAt is the diagonal matrix with er1 t , er2 t , · · · , ern t down its main diagonal:
   
r1 0 · · · 0 er1 t 0 · · · 0
 0 r2 · · · 0   0 er2 t · · · 0 
At
A =  .. .. . . =⇒ e = ..  .
   
. . .. ..
. ..   ..
 
 . . . . . 
rn t
0 0 · · · rn 0 0 ··· e

Example 26. Consider the matrix

 
s 0
A= .
0 g

Now,
t2 tn
eAt = I + At + A2 + · · · + An + · · ·
2! n!
    2 
2 s 0 s 0 s 0
A = · =
0 g 0 g 0 g2
 2    3 
3 2 s 0 s 0 s 0
A =A A= =
0 g2 0 g 0 g3
..
.
 n 
n s 0
A = .
0 gn

Thus
       
At 1 0 st 0 s2 t2 /2! 0 sn tn /n! 0
e = + + 2 2 + ··· + n n + ···
0 1 0 gt 0 g t /2! 0 g t /n!
 
1 + st + s2 t2 /2! + · · · + sn tn /n! + · · · 0
=
0 1 + gt + g t /2! + · · · + g n tn /n! + · · ·
2 2
 st 
e 0
=
0 egt

66 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Example 27. Consider the system

 
0 −2 0
x (t) = x(t),
0 3

where vector x has components [x1 , x2 ]. The matrix exponential is given by

e−2t 0
 
At
e = .
0 e3t

Thus, the solution of the system is given by

e−2t 0 e−2t C1 e−2t

        
At C1 0
x=e ·C = = C1 + C2 = .
0 e3t C2 0 e3t C2 e3t

Remark. Also find the solution of the same system by (a) finding the eigen
pairs (b) solving the individual equations directly. Did you get the same
solution as above?

The Case of a Nilpotent Matrix

Definition 17. A square matrix B such that B k = 0 for some positive
integer k is called a nilpotent matrix.

If B is nilpotent, then B k = B k+1 = · · · = 0. Note that B k+1 = B k · B =

0 · B = 0. Thus,

t2 tk−1
eBt = I + Bt + B 2 + · · · + B k−1
2! (k − 1)!

(the matrix exponential series is now truncated because B is nilpotent). If r

is a scalar then, since A = rI + A − rI, we have

eAt = erIt e(A−Ir)t = ert · e(A−rI)t (5.6)

So if B = A − rI is nilpotent then we get a finite representation for eAt .

Before employing this idea to solving systems, we need the Cayley-Hamilton
Theorem:

Theorem 11. [Cayley-Hamilton Theorem] Every n × n constant matrix

satisfies its characteristic equation.

67 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

 
2 3
Example 28. Let A = . Then the characteristic polynomial equa-
4 1
tion for A is given by
2−r 3
P (r) = = r2 − 3r − 10 = 0.
4 1−r

The Cayley-Hamilton Theorem implies that A2 − 3A − 10I = 0. Check to

see if this is true.
We return to equation (5.6). If A has a characteristic equation of the
form P (r) = (r − r1 )n , that is, A has one eigenvalue r1 of multiplicity n,
then by the Cayley-Hamilton Theorem, (A − r1 I)n = 0, and so A − r1 I is
nilpotent. So from (5.6), we get
n−1
 
At r1 t n−1 t
e =e I + (A − Ir1 )t + · · · + (A − Ir1 )
(n − 1)!
Example 29. Find the fundamental matrix eAt for the system x0 (t) = Ax,
where  
2 1 1
A= 1 2 1 .
−2 −2 −1
Solution.  
2−r 1 1
P (r) = |A − rI| =  1 2−r 1 
−2 −2 −1 − r
= −r3 + 3r2 − 3r + 1
=⇒ P (r) = −(r − 1)3 .
So r = 1 is an eigenvalue of A, with multiplicity 3. By the Cayley-Hamilton
Theorem, (A − I · 1)3 = 0, so (A − I) is nilpotent for the integer k=3. Thus,
2
 
At t (A−I)t t 2t
e = e .e = e I + (A − I)t + (A − I)
2!
Now,  
1 1 1
A−I = 1 1 1 
−2 −2 −2
  
1 1 1 1 1 1
(A − I)2 =  1 1 1  1 1 1 
−2 −2 −2 −2 −2 −2

68 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A



0 0 0
=⇒ (A − I)2 =  0 0 0 
0 0 0
(Note that (A − I)2 = 0 is purely coincidental, it has nothing to do with
nilpotency). Thus,
    
 1 0 0 1 1 1 
eAt = et  0 1 0  +  1 1 1 t
0 0 1 −2 −2 −2
 
 t 
e (1 + t) tet tet
= tet et (1 + t) tet .
t t t
−2te −2te e (1 − 2t)
A general procedure for computing eAt if nilpotency does not hold is given
in the next few sections. The first approach we want to consider is by using
generalized eigenvectors. An alternative method, which would be considered
later, is referred to as the Putzer algorithm.

5.3 Computing eAt: Generalized eigenvectors

Lemma 1. Let X(t) and Y (t) be two fundamental matrices for the same
system x0 (t) = Ax. Then there exists a constant matrix C such that X(t) =
Y (t)C.
Now suppose X(t) is a fundamental matrix of x0 (t) = Ax, then
eAt = X(t)C
At t = 0 we have C = X −1 (0)
∴ eAt = X(t)X −1 (0). (5.7)
How do we find X(t)? We explore this question in the following. Note that
the columns of a fundamental matrix have the form eAt u where u is a vector.
So from (5.6), we have
eAt = ert e(A−rI)t
=⇒ eAt u = ert e(A−rI)t u
tk
 
=⇒ eAt u = ert k
u + t(A − rI)u + · · · + (A − rI) u + · · ·
k!
Note that if (r, u) is an eigenpair for the system, then (A − rI)u = (A −
rI)2 u = · · · = 0, so eAt u = ert u is a solution. We expect (A − rI)k u = 0 for
some nontrivial vector u and some positive constant k.

69 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Definition 18. [Generalized Eigenvectors] Let A be an n×n constant matrix

and r be an eigenvalue of A. A nontrivial vector u that satisfies

(A − rI)k u = 0

for some positive integer k is called a generalized eigenvector associated

with r.

Remark. If the characteristic polynomial for A is of the form

P (r) = (r1 − r)m1 · · · (rk − r)mk

where the ri0 s are distinct eigenvalues of A and mi is the multiplicity of the
eigenvalue ri , then for each i there are mi linearly independent eigenvectors
associated with ri . If u is a generalized eigenvector associated with ri , then

(A − ri I)mi u = 0.

5.3.1 Solving systems using generalized eigenvectors

Procedure: For a fundamental solution set for the system

x0 (t) = Ax,

(a) Compute the characteristic polynomial

P (r) = |A − rI|

and find the distinct eigenvalues r1 , · · · rk .

(b) For each eigenvalue ri , find mi linearly independent generalized eigen-

vectors, where mi is the multiplicity of ri .

(c) Use the n linearly independent generalized eigenvectors obtained in (b)

to compute the n linearly independent solutions to x0 (t) = Ax of the
form
t2
 
At rt 2
e u = e u + t(A − rI)u + (A − rI) u + · · · (5.8)
2!

where r is an eigenvalue and u is a corresponding generalized eigenvec-

tor. If r has multiplicity mi , then the series in (5.8) reduces to the first
mi terms.

70 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Example 30. Using generalized eigenvectors, compute eAt for the system
x0 (t) = Ax, where  
1 1
A= .
−1 3
Hence, find the solution of the system.
Solution. We first solve |A − rI| = 0 such that

1−r 1
=0
−1 3 − r

=⇒ (1 − r)(3 − r) + 1 = 0
=⇒ 3 − r − 3r + r2 + 1 = 0, =⇒ r2 − 4r + 4 = 0
=⇒ (r − 2)2 = 0.
Therefore r = 2, with multiplicity 2. For the eigenvector corresponding to
r = 2, we solve
(A − 2I)U = 0
  
−1 1 u1
=⇒ =0
−1 1 u2
=⇒ −u1 + u2 = 0, u1 = u2
 
1
∴ U1 =
1
is an eigenvector. So a solution is
 
2t 1
X1 = e .
1

For a second linearly independent eigenvector, we solve

(A − 2I)2 u = 0.

Now,     
2 −1 1 −1 1 0 0
(A − 2I) = = .
−1 1 −1 1 0 0
    
0 0 u1 0
=⇒ =
0 0 u2 0
Thus, any vector U2 , linearly independent of U1 , will suffice. So we choose
 
0
U2 = .
1

71 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Therefore a second solution is given by

X2 = e2t [U2 + t(A − 2I)U2 ]

     
2t 0 −1 1 0
=e +t
1 −1 1 1
   
0 1
= e2t +t
1 1
 
t
∴ X2 = e2t .
1+t

Using the two solutions, X1 and X2 , we obtain the fundamental matrix

 2t 
e te2t
X(t) = .
e2t te2t + e2t
   
1 0 −1 1 0
X(0) = =⇒ X (0) = .
1 1 −1 1
Thus,
eAt = X(t)X −1 (0),
  
e2t te2t 1 0
= ,
e2t te2t + e2t −1 1
  
2t 1 t 1 0
=e ,
1 t+1 −1 1
 
At 2t 1−t t
∴e =e .
−t 1 + t
So a solution to the system
 
0 1 1
x (t) = x
−1 3

is given by
   
At 2t 1−t 2t t
x(t) = e · C = C1 e + C2 e .
−t 1+t

Example 31. Find the fundamental matrix eAt for the system
 
1 0 0
0
x (t) = Ax, A = 1 3 0 .
0 1 1

72 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Solution.
1−r 0 0
P (r) = |A − rI| = 1 3−r 0 ,
0 1 1−r
 
3−r 0
= (1 − r) = (1 − r)(3 − r)(1 − r),
1 1−r
= (1 − r)2 (3 − r) = −(r − 1)2 (r − 3).
Therefore, the eigenvalues are r = 1 with multiplicity 2, and r = 3 with
multiplicity 1. For r = 1, we need two linearly independent eigenvectors.
Let’s represent them by U1 and U2 . We start with

(A − rI)U1 = 0, =⇒ (A − I)U1 = 0
  
0 0 0 u1
=⇒  1 2 0   u2  = 0
0 1 0 u3
=⇒ u1 + 2u2 = 0
But from the last row, we have

u2 = 0. =⇒ u1 = 0.

Thus, we get the generalized eigenvector:

   
u1 0
∴ U1 =  u2  =  0  ,
u3 1

where we chose u3 = 1. So one solution is given by

 
0
rt t t
X1 = e U1 = e U1 = e 0 .
1

There is a multiplicity of 2 so we next solve

(A − rI)2 U2 = 0. =⇒ (A − I)2 U2 = 0
 
0 0 0 0 0 0
=⇒  1 2 0   1 2 0  U2 = 0
0 1 0 0 1 0

73 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

  
0 0 0 u1
 2 4 0   u2  = 0
1 2 0 u3
=⇒ u1 + 2u2 = 0.
In general, we can let u3 = s, and u2 = v, so that
     
−2v −2 0
U2 =  v  = v  1  + s  0  .
s 0 1

We choose s = 0 and v = 1 to get the linearly independent eigenvector

 
−2
U2 =  1  .
0

Note that if we had chosen s = 1 and v = 0, we get the vector U1 . Now

using U2 , we get the solution

X2 = eAt U2 = et [U2 + t(A − I)U2 ]

    
−2 0 0 0 −2
=⇒ X2 = et  1  + tet  1 2 0   1 
0 0 1 0 0
   
−2 0
= et  1  + tet  0 
0 1
 
−2et
∴ X2 (t) =  et  .
tet
For the eigenvector r = 3, we solve

(A − 3I)U3 = 0
  
−2 0 0 u1
=⇒  1 0 0   u2 
0 1 −2 u3
=⇒ −2u1 = 0 =⇒ u1 = 0.
u2 − 2u3 = 0

74 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Let u3 = s =⇒ u2 = 2s. Thus

  
0 0
U3 = 2s = s 2 
  
s 1
We choose s = 1 to get the eigenvector
 
0
U3 =  2  .
1
Thus, a third linearly independent solution to the system is given by
   
0 0
X3 (t) = e3t U3 = e3t  2  =  2e3 t  .
1 e3t
From the three solutions, X1 , X2 and X3 , we get the fundamental matrix
 
0 −2et 0
X(t) =  0 et 2e3t  .
et tet e3t
Thus,
eAt = X(t)X −1 (0).
Now,  
0 −2 0
X(0) =  0 1 2  .
1 0 1
−2 0
det(X(0)) = = −4.
1 2
The cofactor matrix of X(0) is
 
1 2 0 2 0 1
 + − +
 0 1 1 1 1 0 

 −2 0 0 0 0 −2 
Cof actorX(0) =   − + − ,
 0 1 1 1 1 0 

 −2 0 0 0 0 −2 
+ − +
1 2 0 2 0 1
 
1 2 −1
= 2 0 −2  .
−4 0 0

75 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

The transpose of the Cofactor of X(0) is

 
1 2 −4
 2 0 0 .
−1 −2 0
Thus, the inverse of X(0) is given by
  1
− 4 − 12 1
 
1 2 −4
1
X −1 (0) = −  2 0 0  =  − 12 0 0 
4 1 1
−1 −2 0 4 2
0
Hence,  
et 0 0
eAt = − 12 et + 21 e3t e3t 0 .
1 t 1 t 1 3t 1 t 1 3t
− 4 e − 2 te + 4 e − 2 e + 2 e et

5.4 Computing eAt: The Putzer Algorithm

Theorem 12. Let λ1 , λ2 , · · · , λn be the (not necessarily distinct) eigenvalues
of the matrix A. Then
n−1
X
At
e = Pk+1 (t)Mk
k=0
k
Q
where M0 := I, Mk := (A − λi I), for 1 ≤ k ≤ n and the vector function
i=1
P defined by  
P1 (t)
 P2 (t) 
P (t) =  , for t ∈ R,
 
..
 . 
Pn (t)
is the solution of the initial value problem (IVP)
   
λ1 0 0 ··· 0 1
 1 λ2 0
 ··· 0  

 0 

0
P (t) =  0 1 λ3
 ··· 0   P, P (0) = 
 0 .

 .. .. . . .. .  ..
. .. 
 
 . . .  . 
0 0 0 · · · λn 0
Example 32. Use the Putzer algorithm to find eAt for
 
1 1
A=
−1 3

76 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Solution.
|A − λI| = 0 =⇒ λ2 − 4λ + 4 = 0
=⇒ (λ − 2)2 = 0, =⇒ λ1 = λ2 = 2.
1
X
At
e = Pk+1 (t)Mk
0

=⇒ eAt = P1 (t)M0 + P2 (t)M1 · · · · · · (∗)

 
1 0
M0 = I = .
0 1
 
−1 1
M1 = A − 2I = .
−1 1
 
P1 (t)
Now, let P (t) = and solve the IVP
P2 (t)
   
0 2 0 1
P (t) = P (t), P (0) =
1 2 0
 0    
P1 (t) 2 0 P1 (t)
=⇒ =
P20 (t) 1 2 P2 (t)
Thus, we get the two equations:

P10 (t) = 2P1

P20 (t) = P1 + 2P2 .

Solving the first equation, we get P1 = e2t C1 . But P1 (0) = 1, =⇒ C1 = 1.
Thus
P1 (t) = e2t .
The second equation becomes

P20 (t) = e2t + 2P2

=⇒ P20 − 2P2 = e2t ,

which is a linear first order equation with integrating factor
R
−2dt
µ=e = e−2t .

Thus,
d −2t

e P2 = 1, =⇒ e−2t P2 = t + C2
dt
77 c Dr. Joseph K. Ansong
Math 450 : Differential Equations II J.K.A

=⇒ P2 = te2t + C2 e2t .
But P2 (0) = 0 =⇒ C2 = 0. Thus,

P2 (t) = te2t .

Hence      
P1 e2t 2t 1
P = = =e .
P2 te2t t
From equation (*), we have
   
At 2t 1 0 −1 1 2t
e =e + te
0 1 −1 1
   
2t 1 0 −t t
=e +
0 1 −t t
 
2t 1−t t
=e .
−t 1 + t
Note that we also solved the problem above using generalized eigenvectors.
Compare the two methods. We can find the solution of the system x0 = Ax
if we need to; the solution is given by
   
At 2t 1−t 2t t
x(t) = e c = c1 e + c2 e .
−t 1+t

Example 33. Use the Putzer algorithm to solve the system

 
2 0 0
x0 (t) =  1 2 0  x.
1 0 3

Solution.
 
2−λ 0 0
|A − λI| = 0 =⇒  1 2−λ 0  = 0.
1 0 3−λ

=⇒ (λ − 2)2 (λ − 3) = 0.
=⇒ λ1 = λ2 = 2, λ3 = 3
are eigenvalues. By the Putzer algorithm
1
X
At
e = Pk+1 (t)Mk
0

78 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

=⇒ eAt = P1 (t)M0 + P2 (t)M1 + P3 (t)M2 , · · · · · · · · · (∗)

where  
1 0 0
M0 = I =  0 1 0  .
0 0 1
 
0 0 0
M1 = (A − λ1 I) =  1 0 0  .
1 0 1
M2 = (A − λ1 I)(A − λ2 I)
    
0 0 0 0 0 0 0 0 0
=⇒ M2 =  1 0 0   1 0 0  =  0 0 0  .
1 0 1 1 0 1 1 0 1
 
P1 (t)
Now, let P (t) = P2 (t)  and solve the IVP

P3 (t)
 0      
P1 2 0 0 P1 1
 P20  =  1 2 0   P2  , P (0) =  0  .
P30 0 1 3 P3 0

Thus, we get the three IVPs below, which need to be solved sequentially:

P10 = 2P1 , P1 (0) = 1

P20 = P1 + 2P2 , P2 (0) = 0

P30 = P2 + 3P3 , P3 (0) = 0.
Solving the first equation and applying the initial condition results in

P1 = e2t .

The second equation becomes

P20 = P1 + 2P2 = e2t + 2P2 , p2 (0) = 0

d
=⇒ P20 − 2P2 = e2t =⇒ P2 e−2t = 1


dt
=⇒ P2 e−2t = t + C =⇒ P2 = te2t + Ce2t
P2 (0) = 0 =⇒ C = 0
∴ P2 = te2t .

79 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

The third equation becomes

P30 = P2 + 3P3 = te2t + 3P3 , P3 (0) = 0.

Solving this results in

P3 (t) = −te2t − e2t + e3t .

Thus, from (*), we get

     
1 0 0 0 0 0 0 0 0
eAt = e2t  0 1 0  + te2t  1 0 0  + (−te2t − e2t + e3t )  0 0 0 
0 0 1 1 0 1 1 0 1
 
e2t 0 0
∴ eAt =  te2t e2t 0  .
e3t − e2t 0 e3t
Hence, the solution of the system is given by
 
c1
x(t) = eAt c = eAt  c2 
c3
     
e2t 0 0
∴ x(t) = c1  te2t  + c2  e2t  + c3  0  .
e3t − e2t 0 e3t

80 c Dr. Joseph K. Ansong

Chapter 6

Nonlinear autonomous systems

Our focus here is to analyze the stability of nonlinear autonomous systems

near fixed (or equilibrium) points. It is rare for one to be able to write down
explicit solutions to nonlinear ODEs. However, if equilibrium solutions to the
nonlinear system can be obtained, it is often possible to use the techniques of
linearization and the ideas developed for linear systems to analyze solutions
to the nonlinear sytem near the fixed points.

6.1 Linearization Theorem

We first start with two-dimensional autonomous systems and later general-
ize the idea to n-dimensional systems. Consider the nonlinear autonomous
system

x˙1 = f1 (x1 , x2 ), (6.1)

x˙2 = f2 (x1 , x2 ),

where f1 , f2 : R2 → R are continuously differentiable functions. Let x∗ =

(x∗1 , x∗2 ) be a fixed point of (6.1) such that f1 (x∗1 , x∗2 ) = 0 = f2 (x∗1 , x∗2 ). We
seek linear approximations to (6.1) for x close to x∗ so we let

x1 = x∗1 + ξ1 , and x2 = x∗2 + ξ2 , (6.2)

where we assume that the ξi (i = 1, 2) are small (|ξi |  1). By carrying out
a 2-dimensional Taylor expansion, we get
∂f1 ∗ ∂f1 ∗
f1 (x) = f1 (x∗ ) + ξ1 (x ) + ξ2 (x ) + higher order terms,
∂x1 ∂x2
∂f1 ∗ ∂f2 ∗
=⇒ f1 (x) ≈ ξ1 (x ) + ξ2 (x ),
∂x1 ∂x2

81
Math 450 : Differential Equations II J.K.A

since ξi are small and f1 (x∗ ) = 0. Similarly, we get

∂f2 ∗ ∂f2 ∗
f2 (x) ≈ ξ1 (x ) + ξ2 (x ).
∂x1 ∂x2
Note from (6.2) that x˙1 = ξ˙1 and x˙2 = ξ˙2 , so system (6.1) becomes the linear
system
∂f1 ∗ ∂f1 ∗
ξ˙1 = ξ1 (x ) + ξ2 (x ),
∂x1 ∂x1
∂f2 ∗ ∂f2 ∗
ξ˙2 = ξ1 (x ) + ξ2 (x ).
∂x1 ∂x2
This linear system can be written in the from
ẋ = Ax, (6.3)
where the Jacobian matrix at x∗ is given by
 
∂f1 ∗ ∂f1 ∗
(x ) ∂x2 (x )
A =  ∂x1 , (6.4)
∂f2 ∗ ∂f2 ∗
∂x1
(x ) ∂x2 (x )
and  
ξ1
x= .
ξ2

This linear system is called the linearized equation of (6.1) and it is the equa-
tion which describes the behaviour of the sytem near the fixed point. We
would like to know if solutions of the nonlinear system (6.1) and the linear
system (6.3) behave similarly, at least close to the fixed point x∗ . So the
stability of the equilibrium points of the nonlinear system is now reduced to
analyzing the behaviour of the linearized system.

In n−dimensional systems, the nonlinear equations may be written as

x0 = f (x), x ∈ Rn (6.5)
where f : Rn → Rn , and the Jacobian matrix is defined as
 
∂f1 ∂f1 ∂f1
· · · ∂xn
 ∂x1 ∂x2 
 ∂f2 ∂f2 ∂f2 
· · ·
Df (x∗ ) =  ∂x1 ∂x2 ∂x2 
 
 .. .. .. ..  . (6.6)
 . . . . 
 
∂fn ∂fn ∂fn
∂x1 ∂x2
· · · ∂xn

82 c Dr. Joseph K. Ansong

 Math 450 : Differential Equations II J.K.A

The linearization theorem stated below shows that, depending on the

eigenvalues of the Jacobian matrix, there is some correspondence in qualita-
tive behaviour between the nonlinear and the linearized system.

Theorem 13. [Linearization Theorem]

Let λ and µ be eigenvalues of the Jacobian matrix (6.4). If Re(λ) 6=
0, Re(µ) 6= 0 and λ 6= µ, then the qualitative behaviour of solutions of
the nonlinear system (6.1) near the equilibrium point x∗ = (x∗1 , x∗2 ) is
the same as the behaviour of the linear system (6.3) near the equilibrium
point (0, 0).

Remark. The equilibrium points to which the theorem applies are sta-
ble/unstable nodes, stable/unstable spiral points and saddle (hyperbolic)
points. In the other cases not mentioned in the theorem, the behaviour of
the fixed point may remain the same as in the linear system or may behave
completely different from the linear case! The following examples demon-
strate the application of the linearization theorem.
Example 34. Find the equilibrium points and determine their nature for
the system
ẋ = 2y + xy, (6.7)
ẏ = x + y.
Hence plot a possible phase diagram of the system.
Solution. We determine the equilibrium point (x, y) from the simultaneous
equations
2y + xy = 0,
x + y = 0.
From the first equation, we find that y = 0 or x = −2, and the from the
second equation we get y = −x. So the equilibrium points are (0, 0) and
(−2, 2). We consider each equilibrium point.
Case 1: Equilibrium point (0, 0).

The Jacobian matrix at (0, 0) for the corresponding linear system is

given by
 
0 2
A= .
1 1

83 c Dr. Joseph K. Ansong

 Math 450 : Differential Equations II J.K.A

The eigenvalues
 of A are λ1 =  −1 and λ2 = 2 with corresponding
−2 1
eigenvectors and . Thus, (0, 0) is a saddle point for the
1 1
linear system, and hence, by the linearization theorem, also a saddle
point for the nonlinear system.

Case 2: Equilibrium point (−2, 2).

The Jacobian matrix at (−2, 2) for the corresponding linear system

is given by
 
2 0
A= .
1 1

The eigenvalues
  of Aareλ1 = 1 and λ2 = 2 with corresponding eigen-
0 1
vectors and . Thus, (−2, 2) is an unstable node for the
1 1
linearized system, and hence, by the linearization theorem, also for the
nonlinear system. A possible phase diagram of the linearized system is
shown in Figure (6.1).

6.2 Lyapunov functions

The linearization theorem described in the previous section only applies to
certain systems of equations. When linearization theorem does not apply,
Lyapunov functions may be used to determine the stability of the equi-
librium point. The main advantage of this approach is that stability can
be discussed without any prior knowledge of the solutions. The following
example illustrates the method. Consider the nonlinear system
p
ẋ = y − x x2 + y 2 (6.8)
p
ẏ = −x − y x2 + y 2 .

The linearized equation at (0, 0) is given by the system

 
0 1
ẋ = Ax, with A = .
−1 0

The Jacobian matrix at (0, 0) has eigenvalues ±i. Hence (0, 0) is a center for
the linearized system, and the Linearization Theorem cannot be applied to

84 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Figure 6.1: A phase diagram of the linearized system for equation (6.7).

draw any conclusions about the stability of the nonlinear system at (0, 0).
From (6.8), we get
p
ẋx = xy − x2 x2 + y 2 ,
p
ẏy = −xy − y 2 x2 + y 2 ,

and by adding the two equations we get

p
ẋx + ẏy = −(x2 + y 2 ) x2 + y 2 ,
=⇒ ẋx + ẏy = −(x2 + y 2 )3/2 .

85 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

d 2
Note that (x + y 2 ) = 2xẋ + 2y ẏ, therefore
dt
d 2
(x + y 2 ) = 2xẋ + 2y ẏ = −2(x2 + y 2 )3/2 < 0. (6.9)
dt
So the function V (x, y) = x2 + y 2 is a strictly decreasing function of t.
Since V (x, y) measures the separation from the origin, we deduce that x(t)
gets closer to (0, 0) as t increases. Thus we deduce that (0, 0) is a stable
equilibrium point of (6.8). This conclusion can be proved in this particular
example by integrating (6.9) to
1
x2 (t) + y 2 (t) = ,
(t + c)2
showing that V tends to 0 as t → ∞. The technique above is often useful in
establishing whether an equilibrium point is stable or not, and the function
V = x2 + y 2 can be replaced by any function with similar properties.
The following definitions and theorem are helpful in analyzing the stabil-
ity, asymptotic stability, and unstability of the trivial solution (or equilibrium
point) of nonlinear autonomous systems. Consider the nonlinear system

ẋ = f (x), (6.10)

where the function f = (f1 , · · · , fn ) and its partial derivatives are assumed
to be continuous in an open set Ω ∈ Rn containing the origin. Let V (x) be
a scalar continuous function defined in Ω and V (0) = 0.

Definition 19. V (x) is said to be positive definite in Ω if and only if

V (x) > 0 for x 6= 0, x ∈ Ω.

Definition 20. V (x) is said to be positive semidefinite in Ω if V (x) ≥ 0

(with equality only at certain points) for all x ∈ Ω.

Definition 21. V (x) is said to be negative definite (negative semidefi-

nite) in Ω if and only if −V (x) is positive definite (positive semidefinite)
in Ω.

Example 35. The function V (x1 , x2 ) = c1 x21 + c2 x22 , where c1 > 0, c2 > 0 is
positive definite in Ω = R2 .

86 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Example 36. The function V (x1 , x2 , x3 ) = c1 x21 + c2 x22 , where c1 > 0, c2 > 0
is positive semidefinite in Ω = R3 since it vanishes on the x3 −axis.

Theorem 14. Consider an autonomous system ẋ = f (x) with an iso-

lated equilibrium point x∗ . Let Ω be an open set containing x∗ and let
V : Ω → R be a smooth function such that

(1) V (x∗ ) = 0 and V (x) > 0 for all x ∈ Ω \ {x∗ },

(2) For every solution x of the system taking values in Ω, if

(i) d
dt
(V (x(t))) ≤ 0, then x∗ is a stable equilibrium point,
(ii) d
dt
(V (x(t))) > 0, then x∗ is an unstable equilibrium point,
(iii) d
dt
(x(t))) < 0, then x∗ is an asymptotically stable equilib-
(V
rium point.

Remark. Note that by the chain rule, we have

dV (x) ∂V (x) dx1 ∂V (x) dxn

= V ∗ (x) = + ··· + ,
dt ∂x1 dt ∂xn dt
n
X ∂V (x)
= fi (x) = gradV (x) · f (x) (6.11)
i=1
∂xi

Thus, the derivative of V (x) with respect to t along the solution x(t) of the
nonlinear system is known, although we do not have the explicit solution.
In general, it is difficult to find a Lyapunov function for a given system
of equations. It is sometimes possible to guess a Lyapunov function from
geometrical or physical considerations.

Example 37. For the system

x˙1 = x2 + x1 (r2 − x21 − x22 ),

x˙2 = −x1 + x2 (r2 − x21 − x22 ),

consider the positive definite function V (x1 , x2 ) = x21 + x22 in Ω = R2 . We

analyze the stability of the system about (0, 0). Computing the derivative of
V gives (after some manipulations)

dV (x)
V∗ = = −2(x21 + x22 )(x22 + x22 − r2 ).
dt

87 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

If r = 0, we see that V ∗ is negative definite, and hence the trivial solution

(equilibrium point at (0, 0)) is asymptotically stable. If r 6= 0 then V ∗ is
positive definite in the region x21 + x22 < r2 , hence the trivial solution is
unstable.

Example 38. Consider the differential equation

ẋ = −xy 4 ,
ẏ = yx4 .

We choose the positive definite function V (x, y) = x4 + y 4 in Ω = R2 . Here,

we find that V ∗ ≡ 0 and hence the equilibrium point (0, 0) is stable.

Example 39. For the following system, construct a Lyapunov function of

the form c1 u21 + c2 u22 to determine whether the trivial solution is stable,
asymptotically stable, or unstable

u˙1 = 2u1 u2 + u31 ,

u˙2 = −u21 + u52 .

Solution. Let V (u1 , u2 ) = c1 u21 + c2 u22 . Then

∂V ∂u1 ∂V ∂u2
V∗ = +
∂u1 ∂t ∂u2 ∂t
= 2c1 u1 u̇1 + 2c2 u2 u̇2
= 2c1 u1 (2u1 u2 + u31 ) + 2c2 u2 (−u21 + u52 )
= 4c1 u21 u2 + 2c1 u41 − 2c2 u21 u2 + 2c2 u62
= u2 u21 (4c1 − 2c2 ) + 2(c1 u41 + c2 u62 )

Let 4c1 − 2c2 = 0, then c2 = 2c1 and

V ∗ = 2(c1 u41 + c2 u62 )

= 2c1 (u41 + 2u62 ).

Let c1 > 0, then c2 > 0 and so V = c1 u21 + c2 u22 = c1 (u21 + 2u22 ) > 0 for
(u1 , u2 ) 6= (0, 0). Also, V ∗ > 0 for c1 > 0 and therefore the equilibrium point
(0, 0) is unstable.

Remark. Note that the case c1 < 0, and therefore c2 < 0, is unacceptable
since this violates requirement (1) in the Theorem 14.

88 c Dr. Joseph K. Ansong

Appendix A

Lipschitz Condition

Definition 22. [Lipschitz Condition] The function f (x, y) is said to

satisfy a uniform Lipschitz condition in any domain D if

|f (x, y1 ) − f (x, y2 )| ≤ L|y1 − y2 | (A.1)

for all (x, y1 ), (x, y2 ) in the domain D. The nonegative constant L is

called the Lipschitz constant.

Example 40. Let f (x, y) = 3y + 2. Then

|f (x, y2 ) − f (x, y1 )| = |(3y2 + 2) − (3y1 + 2)|

= |3y2 − 3y1 |
= 3|y2 − y1 |.
∴ f is Lipschitz with L = 3.
Example 41. Show that f (x, y) = x2 y + x2 ex satisfies a Lipschitz condition
on the interval D = {(x, y)|1 ≤ x ≤ 2; −2 ≤ y ≤ 5}. We have
2 2
|f (x, y1 ) − f (x, y2 )| = y1 + x2 ex − y2 − x2 ex
x x
2
= |y1 − y2 |
x
≤ 2|y1 − y2 |.
The inequality holds by applying the bounds on x. Thus, f satisfies a Lips-
chitz condition on D with L = 2.

89
Math 450 : Differential Equations II J.K.A

Example 42. Let f (x, y) = xy 2 . In general

|f (x, y2 ) − f (x, y1 )| = |xy12 − xy22 |

= |x||y1 + y2 ||y1 − y2 |
is not bounded by any constant times |y1 − y2 |, so f is not Lipschitz on the
domain R × R. However, f is Lipschitz on any rectangle [a, b] × [c, d] since
we have
x|y1 + y2 | ≤ 2 · max{|a|, |b|} · max{|c|, |d|}
on R.

Theorem 15. [Picard-Lindelof Theorem (version 2)]

Consider the intervals IT = [t0 − T, t0 + T ] and Bd = [x0 − d, x0 + d] for
positive, real numbers T, d. Suppose that f : IT × Bd → R is continuous
and satisfies a Lipschitz condition on IT ×Bd → R. Then there is a δ > 0
so that the initial value problem
dx
= f (t, x), x(t0 ) = x0 , (A.2)
dt
has a unique solution in the interval Iδ = [t0 − δ, t0 + δ].

Remark. a) There are examples of functions that are Lipschitz but not
differentiable, as for example f (y) = |y|, and Theorem 15 applies for
such functions.

b) If the function f (x, y) is differentiable with respect to y, then it is

easy to compute the Lipschitz constant. The following theorem gives
a simple test for a function to be Lipschitz with respect to y.

Theorem 16. []
Let the domain D be convex and the function f (x, y) be differentiable
with respect to y in D. Then for the Lipschitz condition to be satisfied,
it is necessary and sufficient that

∂f (x, y)
Sup ≤ L. (A.3)
D ∂y

90 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Definition 23. [Convex Set] A set D ⊃ R2 is said to be convex if

whenever (x1 , y1 ) and (x2 , y2 ) belong to D and λ ∈ [0, 1], the point
((1 − λ)x1 + λx2 , (1 − λ)y1 + λy2 ) also belongs to D.

Geometrically, a set is convex means that a line segment connecting (x1 , y1 )

and (x2 , y2 ) is in D whenever (x1 , y1 ) and (x2 , y2 ) belong to D. For example
the set D = {(x, y), a ≤ x ≤ b; −∞ < y < ∞} is convex.
Proof. [Theorem 16]
Suppose equation (A.3) holds. Since f (x, y) is differentiable with respect to
y and D is convex, for all (x, y1 ) and (x, y2 ) ∈ D the mean value theorem
gives
∂f (x, y ∗ ) f (x, y1 ) − f (x, y2 )
=
∂y y1 − y2
where y ∗ ∈ [y2 , y1 ].

∂f (x, y ∗ ) f (x, y1 ) − f (x, y2 )

=⇒ = ≤L
∂y y1 − y2

by applying equation (A.3). Now suppose equation (2.2) holds:

|f (x, y1 ) − f (x, y2 )| ≤ L|y1 − y2 |

f (x, y1 ) − f (x, y2 )
=⇒ ≤L
y1 − y2
Taking the limit as y2 → y1 on both sides, we get

f (x, y1 ) − f (x, y2 ) ∂f (x, y1 )

lim = ≤ L. 2
y2 →y1 y1 − y2 ∂y1

Example 43. Show that f (x, y) = 1+x sin(xy) satisfies a Lipschitz condition
on 0 ≤ x ≤ 2 and −∞ < y < ∞.
Solution. The set {0 ≤ x ≤ 2 and − ∞ < y < ∞} is convex.
∂f
= x2 cos(xy)
∂y

∂f
=⇒ = x2 |cos(xy)| ≤ x2
∂y
≤4
So f is Lipschitz with L = 4.

91 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Alternatively, we could use the Mean Value Theorem on y:

f (x, y1 ) − f (x, y2 ) ∂f (x, c)

= = x2 cos(xc)
y1 − y2 ∂y

=⇒ |f (x, y1 ) − f (x, y2 )| = x2 cos(xc) |y1 − y2 |

≤ 22 |y1 − y2 | = 4|y1 − y2 |.
dy
Example 44. Consider the IVP dx = y 2/3 , y(0) = y0 ; such that f (x, y) =
2/3
y . It is easy to check that if y0 = 0, the IVP has no unique solution.
What is the problem? We can check to see if f satisfies a Lipschitz condition
around y = 0.

f (x, y) − f (x, 0) |y 2/3 | 1

= = 1/3 → ∞ as y → 0.
y−0 |y| |y |

So there is no Lipschitz constant in any interval containing zero. Note that

if y0 > 0, then on the interval J = (y0 /2, ∞), ∂f /∂y is bounded. This is
because
∂f 2
= y −1/3
∂y 3
is decreasing in y on J so that

∂f (x, y) 2 2
= ≤ ≡ L.
∂y 3|y 1/3 | 3 (y0 /2)1/3

Theorem 17. [Integral Equation] Let f (x, y) be continuous in the do-

main D, then any solution of the IVP
dy
= f (x, y), y(x0 ) = y0 , (A.4)
dx
is also a solution of the integral equation
Z x
y(x) = y0 + f (t, y(t))dt, (A.5)
x0

and conversely.

Proof. Let
dy
= f (x, y(x)),
dx
92 c Dr. Joseph K. Ansong
Math 450 : Differential Equations II J.K.A

Z x Z
dy
=⇒ dt = xf (t, y(t))dt
x0 dt x0
Z
=⇒ y(x) − y(x0 ) = xf (t, y(t))dt
x0
Z
∴ y(x) = y0 + xf (t, y(t))dt
x0

Conversely, if y(x) is any solution of (A.5), then y(x0 ) = y0 . Since f (x, y) is

continuous, we can differentiate (A.5) to get

dy
= f (x, y(x)).
dx

93 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

94 c Dr. Joseph K. Ansong

Appendix B

Linear Second Order ODEs

B.1 Homogeneneous Linear Equations

Consider the equation

ay 00 (t) + by 0 (t) + cy(t) = f (t). (B.1)

If f (t) ≡ 0, we get what is called a homogeneous ordinary differential equa-

tions such that

ay 00 (t) + by 0 (t) + cy(t) = 0. (B.2)

The following approach is used to obtain a general solution to (B.2):

a) Assume a solution of the form y = ert and substitute into (B.2) to get
the characteristic or auxilliary equation:

ar2 + br + c = 0. (B.3)

b) Find the roots of (B.3) to get

√
−b +b2 − 4ac
r1 = ,
√2a
−b − b2 − 4ac
r2 = .
2a

c) If b2 − 4ac > 0, the roots r1 and r2 are real and distinct. Then the
solutions are y1 = er1 t and y2 = er2 t . Combining the solutions gives the
general solution
y = C1 er1 t + C2 er2 t
where C1 and C2 are unknown constants.

95
Math 450 : Differential Equations II J.K.A

d) If b2 − 4ac = 0, then r1 and r2 are real repeated roots, and the solutions
are given by y1 = er1 t and y2 = ter1 t . The general solution is then given
by
y = C1 er1 t + C2 ter1 t

d) If b2 − 4ac < 0, then r1 and r2 are complex conjugate roots: r1 = α + iβ

and r2 = α−iβ. The solutions are y1 = eαt cos(βt), and y2 = eαt sin(βt),
and the general solution is

y = C1 eαt cos(βt) + C2 eαt sin(βt).

Example 45. Find the general soltution to the following equations

a) y 00 − y 0 − 2y = 0

b) y 00 + 8y 0 + 16y = 0

c) y 00 − 6y 0 + 10y = 0

Solution. a)
y 00 − y 0 − 2y = 0
Substitute y = ert into the equation to get

r2 − r − 2 = 0

=⇒ (r − 2)(r + 1) = 0
=⇒ r = 2, −1
S we get
y = C1 e2t + C2 e−t

b)
y 00 + 8y 0 + 16y = 0
=⇒ r2 + 8r + 16 = 0,
=⇒ (r + 4)2 = 0,
=⇒ r = −4, −4;
a repeated root.
=⇒ y = C1 e−4t + C2 te−4t

96 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

c)
y 00 − 6y 0 + 10y = 0
=⇒ r2 − 6r + 10 = 0,
√
6 ± 36 − 40
=⇒ r =
2
6 ± 2i
=⇒ r =
2
=⇒ 3 ± i
=⇒ α = 3, β = 1.
Thus, the general solution is given by

y = C1 e3t cos(t) + C2 e3t sin(t).

Example 46. Solve the IVP

y 00 + 2y 0 − 8y = 0,
y 0 (0) = −12. (B.4)

Solution.
y 00 + 2y 0 − 8y = 0,
=⇒ r2 + 2r − 8 = 0
=⇒ (r + 4)(r − 2) = 0
=⇒ r = −4, 2
=⇒ y = C1 e−4t + C2 e2t
Applying the initial conditions:

y(0) = 3 =⇒ C1 + C2 = 3

y 0 (0) = −12 =⇒ −4C1 + 2C2 = −12

=⇒ C1 = 3, andC2 = 0.
=⇒ y = 3e−4t .

B.2 Non-homogeneous Equations

If f (t) 6= 0 in equation (B.1), then the equation is said to be non-homogeneous.
There are two main analytical methods for solving non-homogeneous equa-
tions: The method of undetermined coefficients and the method of variation
of parameters.

97 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

B.2.1 The Method of undetermined coefficients

First determine the solution to the homogeneous equation (say yh ) as dis-
cussed in the previous section. Depending on the form of the non-homogeneous
term f (t), there different ways of obtaining the particular solution (say
yp ) to the nonhomogeneous equation.

(a) To find a particular solution to the differential equation

ay 00 + by 0 + cy = ktm ert ,

assume the following form for the particular solution:

yp (t) = ts (Am tm + · · · + A1 t + A0 )ert

with

(i) s = 0 if r is not a root of the characteristic equationd

(ii) s = 1 if r is a root of the characteristic equation
(iii) s = 2 if r is a double root of the characteristic equation

(b) To find a particular solution to the differential equation

ay 00 + by 0 + cy = ktm eαt cos(βt)

or
ay 00 + by 0 + cy = ktm eαt sin(βt)
use the following form for the particular solution

yp (t) = ts (Am tm + · · · + A1 t + A0 )eαt cos(βt)+

ts (Bm tm + · · · + B1 t + B0 )eαt sin(βt) (B.5)

with

(i) s = 0 if α + iβ is not a root of the characteristic equation.

(ii) s = 1 if α + iβ is a root of the characteristic equation.

Example 47. Find the particular solution of the following equations

(a) y 00 − y = −11t + 1

(b) 2y 00 + y = 9e2t

(c) y 00 − y = t sin(t)

98 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

Solution. (a)
y 00 − y = −11t + 1
The characteristic equation is given by

r2 − 1 = 0 =⇒ r = 1, −1.

So the homogeneous solution is given by

yh = C1 et + C2 e−t .

Let the form of the particular solution be

yp = At + B

=⇒ yp0 = A
=⇒ yp00 = 0
Substituting into the original equation results in

−yp = −11t + 1

=⇒ −At − B = −11t + 1
Comparing coefficients gives

A = 11, B = −1.

So the particular solution is given by

yp = 11t − 1

Note that the general solution of the nonhomogeneous equation is given

by y = yh + yp :
y = C1 et + C2 e−t + 11t − 1.

(b)
y 00 − y = t sin(t)
=⇒ r2 − 1 = 0 =⇒ r = ±1.
The form of the particular solution is

yp = (At + B) cos(t) + (Ct + D) sin(t)

=⇒ yp0 = (A + D + Ct) cos(t) + (−At − B + C) sin(t)

99 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

=⇒ yp00 = (−At − B + 2C) cos(t) + (−2A − D − Ct) sin(t)

Substitute into the original equation to get

(−2At − 2B + 2C) cos(t) + (−2A − 2D − 2Ct) sin(t) = t sin(t)

Comparing coefficients yields:

1
−2C = 1 =⇒ C = −
2
−2A = 0 =⇒ A = 0
−2A − 2D = 0 =⇒ D = 0
1
−2B + 2C = 0 =⇒ B = C = −
2
Thus,
1 1
yp = − cos(t) − t sin(t).
2 2
(c)
2y 00 + y = 9e2t
2r2 + 1 = 0
i
=⇒ r = ± √
2
Let
yp = Ae2t
=⇒ yp0 = 2Ae2t
=⇒ yp00 = 4Ae2t
Sub into original equation to get

8Ae2t + Ae2t = 9e2t

=⇒ 9A = 9, =⇒ A = 1
Thus
yp = e2t .

Example 48. Find the form of the particular solution to

y 00 + 2y 0 − 3y = f (t)

where f (t) is

100 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

(a) 7 cos(3t) (b) 2tet sin(t)

(c) t2 cos(πt) (d) 5e−3t
(e) 3tet (f) t2 et
Solution.
y 00 + 2y 0 − 3y = 0
=⇒ r2 + 2r − 3 = 0
(r + 3)(r − 1) = 0
=⇒ r = 1, −3.
(a) f (t) = 7 cos(3t)
yp = A cos(3t) + B sin(3t)

(b) f (t) = 2tet sin(t)

yp = (At + B)et cos(t) + (Ct + D)et sin(t)

(c) f (t) = t2 cos(πt)

yp = (At2 + Bt + C) cos(πt) + (Dt2 + Et + F ) sin(πt)

(d) f (t) = 5e−3t

Note that r = −3 is a root of the auxilliary equation, so
yp = Ate−3t .

(e) f (t) = 3tet

Note that r = 1 is a root of the auxilliary equation so
yp = t(At + B)et

(f) f (t) = t2 et
yp = t(At2 + Bt + C)et

Example 49. Find the form of the particular solution to

y 00 − 2y 0 + y = f (t)
where f (t) is the same as those in example (48).

s (B.6)

101 c Dr. Joseph K. Ansong

Math 450 : Differential Equations II J.K.A

102 c Dr. Joseph K. Ansong