ODE1

Math 331: Applied Mathematics III Lecture Notes
Department of Mathematics
Addis Ababa University
January 2, 2012
2
⃝January
c 2, 2012 Semu & Tilahun Draft
Table of Contents
Table of Contents 3
I Ordinary Differential Equations 5
1 Ordinary Differential Equations of the First Order 7

1.1 Basic Concepts and Ideas . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2 Separable Differential Equations . . . . . . . . . . . . . . . . . . . . . . 11
1.2.1 Equation Reducible to Separable Form . . . . . . . . . . . . . . 13
1.3 Exact Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . 19
1.4 Linear First Order Differential Equations . . . . . . . . . . . . . . . . . 27
1.5 *Nonlinear Differential Equations of the First Order . . . . . . . . . . . 31
1.5.1 The Bernoulli Equation . . . . . . . . . . . . . . . . . . . . . . 31
1.5.2 The Riccati Equation. . . . . . . . . . . . . . . . . . . . . . . . 32
1.5.3 The Clairuat Equation . . . . . . . . . . . . . . . . . . . . . . . 33
1.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2 Ordinary Differential Equations of The Second and Higher Order 37

2.1 Basic Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.2 General Solution of Homogeneous Linear ODEs . . . . . . . . . . . . . . 38
2.2.1 Reduction of Order . . . . . . . . . . . . . . . . . . . . . . . . 43
2.3 Homogeneous LODE with Constant Coefficients . . . . . . . . . . . . . 45
2.3.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.4 Nonhomogeneous Equations with Constant Coefficients . . . . . . . . . 49
2.4.1 The undetermined coefficient method . . . . . . . . . . . . . . . 51
2.4.2 Variation of Parameters . . . . . . . . . . . . . . . . . . . . . . 55
2.5 The Laplace Transform Method to Solve ODEs . . . . . . . . . . . . . . 58
2.6 The Cauchy-Euler Equation . . . . . . . . . . . . . . . . . . . . . . . . 63
⃝January
4 CONTENTS
2.7 *The Power Series Solution Method . . . . . . . . . . . . . . . . . . . . 66
2.7.1 Power series solution method . . . . . . . . . . . . . . . . . . . 67
2.7.2 Frobenius Method . . . . . . . . . . . . . . . . . . . . . . . . . 70
2.8 Systems of ODE of the First Order . . . . . . . . . . . . . . . . . . . . 75
2.8.1 Eigenvalue Method . . . . . . . . . . . . . . . . . . . . . . . . . 75
2.8.2 The Method of Elimination: . . . . . . . . . . . . . . . . . . . . 80
2.8.3 Reduction of higher order ODEs to a system of ODEs of the first
order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
2.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
2.10 Numerical Methods to Solve ODEs . . . . . . . . . . . . . . . . . . . . 85
2.10.1 Euler’s Method . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
2.10.2 Runge-Kutta Method . . . . . . . . . . . . . . . . . . . . . . . 87
2.11 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3 *Nonlinear ODEs and Qualitative Analysis 91

3.1 The Phase Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.2 Critical Points and Stability . . . . . . . . . . . . . . . . . . . . . . . . 93
3.2.1 Stability for linear systems . . . . . . . . . . . . . . . . . . . . . 96
3.2.2 Stability for nonlinear systems . . . . . . . . . . . . . . . . . . . 98
3.3 Stability by Lyapunav’s Method . . . . . . . . . . . . . . . . . . . . . . 105
3.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
⃝January
Part I
Ordinary Differential Equations
⃝January
Chapter 1
Ordinary Differential Equations of the

First Order
Part 1 of this material deals with equations that contain one or more derivatives of a
function of a single variable and such equations are called ordinary differential equations,
which can be used to model a phenomena of interest in the sciences, engineering, eco-
nomics, ecological studies, and other areas.
In the first section we will see the basic concepts and ideas and in the remaining sections
we will consider equations which involve the first derivative of a given independent variable
with respect to an independent variable, which are called Ordinary Differential Equations
of the First Order.
1.1 Basic Concepts and Ideas
The derivative 𝑑𝑦/𝑑𝑥 of a function 𝑦 = 𝑓 (𝑥) is itself another function 𝑓 ′ (𝑥) found by an
2
appropriate rule of differentiation. For example, the function 𝑦 = 𝑒𝑥 is differentiable on
2
the interval (−∞, ∞) and by the Chain Rule its derivative is 𝑑𝑦/𝑑𝑥 = 2𝑥𝑒𝑥 . If we replace
the right-hand expression of the last equation by the symbol y, the equation becomes
𝑑𝑦
= 2𝑥𝑦. (1.1)
𝑑𝑥
In differentiation, the problem was ”Given a function 𝑦 = 𝑓 (𝑥), find its derivative.”
⃝January
8 Ordinary Differential Equations of the First Order
Now, the problem we face here is ”If we are given an equation such as (1.1), is there some
way or method by which we can find the unknown function 𝑦 = 𝑓 (𝑥) that satisfy the
given equation, without prior knowledge how it was constructed?” These kind of problems
are the ones we are going to focus on in this part of the course.
Definition 1.1.1. An equation involving derivatives of one or more dependent variables
with respect to one or more independent variables is called a Differential Equation
( DE).
Example 1.1.1.
𝑑𝑦 𝑑3 𝑥 𝑑2 𝑥 ∂𝑣 ∂𝑣
+ 𝑦 = 𝑥, 3
+ 5 2 + 3𝑥 = sin 𝑡, + + 5𝑣 = 2. (1.2)
𝑑𝑥 𝑑𝑡 𝑑𝑡 ∂𝑠 ∂𝑡
are all Differential Equations.
Differential equations can be classified by their type, order, and in terms of linearity.
We will see these classifications before going to the solution concept.
Classification by Type
∙ If an equation contains only ordinary derivatives of one or more dependent variables

with respect to a single independent variable, then it is said to be an ordinary
differential equation (ODE).
For example,
)2
𝑑2 𝑦 𝑑3 𝑥 𝑑2 𝑥
(
𝑑𝑦 𝑑𝑦
+ 𝑦 = 𝑥, + 𝑥𝑦 =0 and + 5 + 3𝑥 = sin 𝑡
𝑑𝑥 𝑑𝑥2 𝑑𝑥 𝑑𝑡3 𝑑𝑡2
are all ordinary differential equations.
∙ If a function is defined in terms of two or more independent variables, the corre-

sponding derivative will be a partial derivative with respect to each independent
variable. An equation involving partial derivatives of one or more dependent vari-
ables of two or more independent variables is called a partial differential equation
(PDE).
For example,
∂ 2𝑢 ∂ 2𝑢 ∂𝑢 ∂𝑣 ∂𝑣
2
= 2
−2 and + + 5𝑣 = 2.
∂𝑥 ∂𝑡 ∂𝑡 ∂𝑠 ∂𝑡
are both partial differential equations.
In this part we will only consider the case of ordinary differential equations.
⃝January
1.1 Basic Concepts and Ideas 9
Classification by Order
The order of a differential equation (either ODE or PDE) is the order of the highest
derivative that appear in the equation. For example,
𝑑𝑦 𝑑2 𝑦 𝑑𝑦
4𝑥 +𝑦 =𝑥 2
+ 4 − 6𝑦 = 𝑒𝑥
𝑑𝑥 𝑑𝑥 𝑑𝑥
are first and second-order ordinary differential equations respectively.
The general 𝑛th −order ordinary differential equation in one dependent variable is given
by the general form
𝐹 (𝑥, 𝑦, 𝑦 ′ , 𝑦 ′′ , ..., 𝑦 (𝑛) ) = 0, (1.3)
where F is a real-valued function of 𝑛 + 2 variables 𝑥, 𝑦, 𝑦 ′ , 𝑦 ′′ , ..., 𝑦 (𝑛) .
Remark 1.1.2. For both practical and theoretical reasons we shall also make the assump-
tion hereafter that it is possible to explicitly solve the differential equation of the form
(1.3) uniquely for the highest derivative 𝑦 (𝑛) in terms of the remaining 𝑛 + 1 variables
𝑥, 𝑦, 𝑦 ′ , 𝑦 ′′ , ..., 𝑦 (𝑛−1) . Then the differential equation (1.3) becomes
𝑑𝑛 𝑦
= 𝑓 (𝑥, 𝑦, 𝑦 ′ , ..., 𝑦 (𝑛−1) ), (1.4)
𝑑𝑥𝑛
where 𝑓 is a real-valued continuous function and this is referred to as the normal form
of (1.3).
Example 1.1.2. The normal form of the first-order equation 4𝑥𝑦 ′ + 𝑦 = 𝑥 is

𝑥−𝑦
𝑦′ =
4𝑥
and the normal form of the second-order equation 𝑦 ′′ − 𝑦 + 6𝑦 = 0 is
𝑦 ′′ = 𝑦 ′ − 6𝑦.
The first order ordinary differential equation is generally expressed as:
𝐹 (𝑥, 𝑦, 𝑦 ′ ) = 0 or 𝑦 ′ = 𝑓 (𝑥, 𝑦).
For example, the differential equation 𝑦 ′ + 𝑦 = 𝑥 is equivalent to 𝑦 ′ + 𝑦 − 𝑥 = 0.

If 𝐹 (𝑥, 𝑦, 𝑦 ′ ) = 𝑦 ′ + 𝑦 − 𝑥, then the given differential equation becomes of the form
𝐹 (𝑥, 𝑦, 𝑦 ′ ) = 0.
⃝January
Classification by Linearity
An 𝑛𝑡ℎ -order ordinary differential equation (1.3) is said to be linear if F is linear in

𝑦, 𝑦 ′ , ..., 𝑦 (𝑛) . This means that an 𝑛𝑡ℎ -order linear ordinary differential equation is of
the form
𝑎𝑛 (𝑥)𝑦 (𝑛) + 𝑎𝑛−1 (𝑥)𝑦 (𝑛−1) + ⋅ ⋅ ⋅ + 𝑎1 (𝑥)𝑦 ′ + 𝑎0 (𝑥)𝑦 − 𝑏(𝑥) = 0, (1.5)
where 𝑎𝑛 (𝑥) ∕= 0.
If 𝑏(𝑥) ≡ 0, the equation (1.5) is called a homogeneous DE and otherwise it is called
nonhomogeneous.
Notation
We may equivalently use the notations

𝑑𝑛 𝑦
and 𝑦 (𝑛)
𝑑𝑥𝑛
interchangeably for the 𝑛𝑡ℎ −order derivative of 𝑦 with respect to 𝑥. Using this notation,
equation (1.5) can be equivalently written as
𝑑𝑛 𝑦 𝑑𝑛−1 𝑦 𝑑𝑦
𝑎𝑛 (𝑥) + 𝑎 𝑛−1 (𝑥) + ⋅ ⋅ ⋅ + 𝑎1 (𝑥) + 𝑎0 (𝑥)𝑦 = 𝑏(𝑥).
𝑑𝑥𝑛 𝑑𝑥𝑛−1 𝑑𝑥
Solution Concept
Consider the equation 𝑦 ′ + 2𝑥𝑦 = 0, which is a first order differential equation for the
2
unknown function 𝑦(𝑥). One can easily check that the function 𝑦(𝑥) = 𝑒−𝑥 satisfies the
2
given equation on (−∞, ∞) and we say that 𝑒−𝑥 is a solution for the given differential
equation.
Definition 1.1.3. Let ℎ(𝑥) be a real valued function defined on an interval [𝑎, 𝑏] and
having 𝑛𝑡ℎ order derivative for all 𝑥 ∈ (𝑎, 𝑏). If ℎ(𝑥) satisfies the 𝑛th order ODE (1.5) on
(𝑎, 𝑏), that is,
1. 𝐹 (𝑥, ℎ(𝑥), ℎ′ (𝑥), ℎ′′ (𝑥), . . . , ℎ(𝑛) (𝑥)) is defined for all 𝑥 ∈ (𝑎, 𝑏) and
2. 𝐹 (𝑥, ℎ(𝑥), ℎ′ (𝑥), ℎ′′ (𝑥), . . . , ℎ(𝑛) (𝑥)) = 0, for all 𝑥 ∈ (𝑎, 𝑏),
then 𝑦 = ℎ(𝑥) is called a ( an Explicit) solution of the ODE on [𝑎, 𝑏].
⃝January
1.2 Separable Differential Equations 11
Sometimes a solution of a differential equation may appear as an implicit function, i.e. the
solution can be expressed implicitly in the form: ℎ(𝑥, 𝑦) = 0, where ℎ is some continuous
function of 𝑥 and 𝑦, and such solution is called an Implicit Solution of the DE.
Example 1.1.3. Consider the differential equation 𝑦 ′′ + 𝑦 = 0.

Let ℎ(𝑥) = 2 sin 𝑥 + 3 cos 𝑥. Then if 𝑦 = ℎ(𝑥), 𝑦 ′ = ℎ′ (𝑥) = 2 cos 𝑥 − 3 sin 𝑥, 𝑦 ′′ =
ℎ′′ (𝑥) = −2 sin 𝑥 − 3 cos 𝑥 and 𝑦 ′′ + 𝑦 = (−2 sin 𝑥 − 3 cos 𝑥) + (2 sin 𝑥 + 3 cos 𝑥) = 0,
which means ℎ(𝑥) satisfies the given DE and hence it is an explicit solution.
Example 1.1.4. Consider the differential equation 𝑦𝑦 ′ = −𝑥.

Differentiating both sides of the equation 𝑥2 + 𝑦 2 − 1 = 0, (𝑦 > 0) with respect to 𝑥 we
get
𝑑 2 𝑑
(𝑥 + 𝑦 2 − 1) = (0).
𝑑𝑥 𝑑𝑥
That is,
𝑑𝑦
2𝑥 + 2𝑦 = 0,
𝑑𝑥
which is equivalent to the equation 𝑥 + 𝑦𝑦 ′ = 0 ⇐⇒ 𝑦𝑦 ′ = −𝑥.
Hence, 𝑥2 + 𝑦 2 − 1 = 0 is an implicit solution of the DE 𝑦𝑦 ′ = −𝑥 on (−1, 1), since
𝑦 > 0.
We are now in a position to descibe some rules that solve some differential equations.
There are different methods of solving differential equations and one method that works
for one DE may not work for another. In this chapter we will consider some of these
methods for solving ordinary differential equations of the first order.
1.2 Separable Differential Equations
Consider a differential equation

𝑑𝑦
= 𝑓 (𝑥). (1.6)
𝑑𝑥
Then 𝑑𝑦 = 𝑓 (𝑥)𝑑𝑥 and it can be solved by integration. If 𝑓 (𝑥) is a continuous function,
then integrating both sides of (1.6) gives
∫
𝑦 = 𝑓 (𝑥)𝑑𝑥 = 𝐺(𝑥) + 𝑐,
where G(x) is an antiderivative (indefinite integral) of 𝑓 (𝑥).
Example 1.2.1.
⃝January
∫𝑥
1. If 𝑦 ′ = 𝑥, then 𝑦(𝑥) = 0
𝑡𝑑𝑡 + 𝐶 = 12 𝑥2 + 𝐶
∫𝑥
2. If 𝑦 ′ = 𝑠𝑖𝑛(1 + 𝑥2 ), then 𝑦(𝑥) = 0
𝑠𝑖𝑛(1 + 𝑡2 )𝑑𝑡 + 𝐶. However, it is difficult to find
an explicit solution formula for this problem. (In such cases one may use numerical
methods to get approximate solutions.)
Many first-order ODEs can be reduced or transformed to the form
𝑔(𝑦)𝑦 ′ = 𝑓 (𝑥),
where 𝑔 and 𝑓 are continuous functions. Then, from elementary calculus we have:
𝑔(𝑦)𝑑𝑦 = 𝑓 (𝑥)𝑑𝑥.
Such type of equations are called separable equations. Integrating both sides we get:
∫ ∫
𝑔(𝑦)𝑑𝑦 = 𝑓 (𝑥)𝑑𝑥 + 𝑐
which is the general explicit solution of the given equation.
Example 1.2.2. Solve the DE 6𝑦𝑦 ′ + 4𝑥 = 0.
Solution:
The equation 6𝑦𝑦 ′ + 4𝑥 = 0 is equivalent to

𝑑𝑦
6𝑦 = −4𝑥
𝑑𝑥
and then 6𝑦𝑑𝑦 = −4𝑥𝑑𝑥. Integrating both sides,
∫ ∫
6𝑦𝑑𝑦 = (−4𝑥)𝑑𝑥,
gives
3𝑦 2 + 2𝑥2 = 𝐶,
which is an implicit solution of the given first order differential equation.
Example 1.2.3. Solve the DE 𝑦 ′ = 𝑦 2 𝑒2𝑥 .
First rewrite the equation as

𝑑𝑦
= 𝑦 2 𝑒2𝑥 .
𝑑𝑥
⃝January
If 𝑦 ∕= 0, this has the differential form
1
d𝑦 = 𝑒2𝑥 d𝑥,
𝑦2
where the variables have been separated. Integrating both sides we have
∫ ∫
1
𝑑𝑦 = 𝑒2𝑥 𝑑𝑥,
𝑦2
which implies
−1 𝑒2𝑥
= + 𝑐,
𝑦 2
where 𝑐 is a constant of integration. Then solve for 𝑦 to get
−2
𝑦(𝑥) = ,
(𝑒2𝑥+ 𝑐)
which is an explicit solution of the given first order differential equation.
Remark 1.2.1. It is recommended to write an explicit solution to the differential equation

when ever possible. However, sometimes solving for the dependent variable (in our case
𝑦) may not be possible. In those cases one can represent the final solution by an implicit
solution of the differential equation.
1.2.1 Equation Reducible to Separable Form
There are some differential equations which are not separable, but they can be transformed
to a separable form by simple change of variables. We will see some of the possible
substitutions hereunder.
A. Linear Substitution
Suppose we have a differential equation that can be written in the form:
𝑦 ′ = 𝑔(𝑎𝑥 + 𝑏𝑦 + 𝑐) (1.7)
Such an equation is not in general separable. However, if we set 𝑢 = 𝑎𝑥 + 𝑏𝑦 + 𝑐, we get

𝑑𝑢 𝑑𝑦
=𝑎+𝑏 .
𝑑𝑥 𝑑𝑥
Or
𝑑𝑦 1 𝑑𝑢 𝑎
= − .
𝑑𝑥 𝑏 𝑑𝑥 𝑏
⃝January
Thus (1.7) will be transformed into
1 𝑑𝑢 𝑎
− = 𝑔(𝑢),
𝑏 𝑑𝑥 𝑏
where 𝑢 and 𝑥 can be separated.
Example 1.2.4. Solve the differential equation 𝑦 ′ = (𝑥 + 𝑦)2 .
Solution:
Let 𝑢 = 𝑥 + 𝑦. Then 𝑢′ = 1 + 𝑦 ′ which implies 𝑦 ′ = 𝑢′ − 1. With this substitution the

equation 𝑦 ′ = (𝑥 + 𝑦)2 is equivalent to
𝑑𝑢
𝑢′ − 1 = 𝑢2 ⇐⇒ = 𝑢2 + 1.
𝑑𝑥
Then
𝑑𝑢
= 𝑑𝑥
𝑢2 +1
and integrate both sides, ∫ ∫
𝑑𝑢
= 𝑑𝑥
𝑢2 + 1
to get arctan 𝑢 = 𝑥 + 𝑐 for an arbitrary constant 𝑐. Substituting back 𝑢 = 𝑥 + 𝑦 in the
last equation gives us the general solution of the given DE to be arctan(𝑥 + 𝑦) = 𝑥 + 𝑐.
Example 1.2.5. Solve the differential equation (2𝑥 − 4𝑦 + 5)𝑦 ′ + 𝑥 − 2𝑦 + 3 = 0.
Solution
Let 𝑢 = 𝑥 − 2𝑦. Then, 𝑢′ = 1 − 2𝑦 ′ which implies 𝑦 ′ = 21 (1 − 𝑢′ ). Therefore, the equation

(2𝑥 − 4𝑦 + 5)𝑦 ′ + 𝑥 − 2𝑦 + 3 = 0 becomes (2𝑢 + 5) 12 (1 − 𝑢′ ) + 𝑢 + 3 = 0. Simplifying
this we get (2𝑢 + 5) − (2𝑢 + 5)𝑢′ + 2𝑢 + 6 = 0 which implies
( )
′ 2𝑢 + 5 𝑑𝑢
(2𝑢 + 5)𝑢 = 4𝑢 + 11 ⇐⇒ = 1.
4𝑢 + 11 𝑑𝑥
Then
4𝑢 + 10 1
𝑑𝑢 = 2𝑑𝑥 ⇐⇒ (1 − )𝑑𝑢 = 2𝑑𝑥.
4𝑢 + 11 4𝑢 + 11
Now we integrate both sides
∫ ∫
1
(1 − )𝑑𝑢 = 2𝑑𝑥
4𝑢 + 11
⃝January
and we get
1
𝑢− ln ∣4𝑢 + 11∣ = 2𝑥 + 𝑐1 .
4
But 𝑢 = 𝑥 − 2𝑦. Then substituting this in the above equation gives us
1
𝑥 − 2𝑦 − ln ∣4𝑥 − 8𝑦 + 11∣ = 2𝑥 + 𝑐1
4
for an arbitrary constant 𝑐1 , or equivalently 4𝑥 + 8𝑦 + ln ∣4𝑥 − 8𝑦 + 11∣ = 𝐶, where
𝐶 = −4𝑐1 .
B. Quotient Substitution
Suppose we have an equation that can be written in the form

𝑦
𝑦 ′ = 𝑔( ).
𝑥
Let us substitute
𝑦
𝑢= .
𝑥
Then
𝑑𝑢 𝑥𝑦 ′ − 𝑦 1 𝑦
= 2
= 𝑦′ − 2 .
𝑑𝑥 𝑥 𝑥 𝑥
This implies,
𝑦
𝑦 ′ = 𝑥𝑢′ + = 𝑥𝑢′ + 𝑢.
𝑥
Thus, the differential equation
𝑦
𝑦 ′ = 𝑔( )
𝑥
′
is reduced to the equation 𝑥𝑢 = 𝑔(𝑢) − 𝑢 which is equivalent to the differential equation
𝑑𝑥 𝑑𝑢
= .
𝑥 𝑔(𝑢) − 𝑢
Then by integrating we obtain a general solution.
Example 1.2.6. Solve 𝑥2 𝑦 ′ = 𝑥2 + 𝑥𝑦 + 𝑦 2 .
Solution
For 𝑥 ∕= 0, the differential equation 𝑥2 𝑦 ′ = 𝑥2 + 𝑥𝑦 + 𝑦 2 is equivalent to

𝑦 ( 𝑦 )2
𝑦′ = 1 + + .
𝑥 𝑥
⃝January
Let 𝑢 = 𝑥𝑦 . Then 𝑔(𝑢) = 1 + 𝑢 + 𝑢2 and we get, 𝑥𝑢′ = (1 + 𝑢 + 𝑢2 ) − 𝑢 = 1 + 𝑢2 , which
implies
𝑑𝑢 𝑑𝑥
2
= .
1+𝑢 𝑥
We then integrate ∫ ∫
𝑑𝑢 𝑑𝑥
=
1 + 𝑢2 𝑥
and get arctan 𝑢 = ln ∣𝑥∣ + 𝑐.
𝑦
Now substituting 𝑢 = 𝑥
gives us
𝑦
arctan( ) = ln ∣𝑥∣ + 𝑐 = ln ∣𝑥∣ + ln 𝑘 = ln 𝑘∣𝑥∣, for some 𝑘 > 0.
𝑥
That is,
𝑦
= tan(ln 𝑘∣𝑥∣)
𝑥
and solving for 𝑦 we get 𝑦(𝑥) = 𝑥 tan(ln 𝑘∣𝑥∣).
Example 1.2.7. Solve the DE: 2𝑥𝑦𝑦 ′ = 𝑦 2 − 𝑥2 .
Solution:
Divide both sides by 𝑥2 , for 𝑥 ∕= 0, to get

(𝑦) ( 𝑦 )2
′
2 𝑦 = − 1.
𝑥 𝑥
Let 𝑢 = 𝑥𝑦 . Then 𝑔(𝑢) = 12 (𝑢 − 𝑢1 ) and we get
1 1 −(𝑢2 + 1)
𝑥𝑢′ = (𝑢 − ) − 𝑢 =
2 𝑢 2𝑢
which implies
−2𝑢𝑑𝑢 𝑑𝑥
= .
1 + 𝑢2 𝑥
Then we integrate
−2𝑢𝑑𝑢
∫ ∫
𝑑𝑥
=
1 + 𝑢2 𝑥
and get ln(1 + 𝑢2 ) = − ln 𝑥 + 𝑐. This implies
𝐴
1 + 𝑢2 = 𝑒(− ln 𝑥+𝑐) = , for a constant A.
𝑥
𝑦
Now we substitute 𝑢 = to get
𝑥
𝑥2 + 𝑦 2 = 𝐴𝑥.
⃝January
Notice that the solution of each of the previous examples contain arbitrary constants. To
determine the constants in these solutions we need to impose some additional conditions.
For example, for the DE equation 6𝑦𝑦 ′ + 4𝑥 = 0, the equation 3𝑦 2 + 2𝑥2 = 𝐶 represents
an implicit solution for an arbitrary constant 𝐶. But if 𝑦(0) = 3 is given in addition, then
𝐶 = 27 and 3𝑥2 + 2𝑥 = 27 will be a specific solution of the given DE.
Definition 1.2.2. For the differential equation
𝐹 (𝑥, 𝑦, 𝑦 ′ , 𝑦 ′′ , . . . , 𝑦 (𝑛) ) = 0,
conditions of the form:
𝑦(𝑎) = 𝑦0 , 𝑦 ′ (𝑎) = 𝑦1 , . . . , 𝑦 (𝑛−1) (𝑎) = 𝑦(𝑛−1)
are called initial conditions (IC).

A Differential Equation 𝐹 (𝑥, 𝑦, 𝑦 ′ , . . . , 𝑦 (𝑛) ) = 0 together with Initial Conditions is called
an Initial Value Problem (IVP) or Cauchy’s problem.
Remark 1.2.3. The number of initial conditions necessary to determine a unique solution
equals the order of the differential equation.
Example 1.2.8. Solve the IVP 𝑦 ′′ + 𝑦 = 0, 𝑦(0) = 3 and 𝑦 ′ (0) = −4.
Solution:
First find the general solution with two unknown constants. Given
𝑦(𝑥) = 𝑐1 cos 𝑥 + 𝑐2 sin 𝑥,
since 𝑦(𝑥) satisfy the Differential Equation 𝑦 ′′ + 𝑦 = 0, it is a general solution. Now

𝑦(0) = 𝑐1 = 3 and 𝑦 ′ (𝑥) = −𝑐1 sin 𝑥 + 𝑐2 cos 𝑥 implies 𝑦 ′ (0) = 𝑐2 = −4. Hence the
particular solution of the equation is 𝑦(𝑥) = 3 cos 𝑥 − 4 sin 𝑥.
On the other hand, if some conditions are imposed at 𝑥 = 𝑎 and at 𝑥 = 𝑏, where 𝑎 and
𝑏 are some real numbers in the domain of 𝑦(𝑥), then the problem is called a Boundary-
Value Problem (BVP).
Remark 1.2.4. For a differential equation, initial conditions and/or boundary conditions
can also be given in mixed form. Total number of conditions that are required to solve
the problem uniquely is again equal to the order of the differential equation.
⃝January
Example 1.2.9. Suppose 𝑦 ′′ + 𝑦 = 0, 𝑦(0) = 3 and 𝑦( 𝜋2 ) = 5.
This is a boundary value problem with 𝑦(0) = 𝑐1 which implies 𝑐1 = 3 and 𝑦( 𝜋2 ) = 𝑐2 ,
which implies 𝑐2 = 5.
Hence, the particular solution of this BVP is
𝑦(𝑥) = 3 cos 𝑥 + 5 sin 𝑥.
Every differential equation may not have a solution, and some of them may also have
more than one solution functions satisfying initial conditions. Before we consider the so-
lution methods for differential equations we need to answer the following two fundamental
questions that arise in considering an initial-value problem and these are:
∙ Does a solution of the problem exist?
∙ If a solution exists, is it unique?
Getting answer for these questions is crucial before we try to find the solutions. The
following two theorems answer these questions by formulating conditions for existence
and uniqueness of an initial value ODE.
Theorem 1.2.5 (Existence).

If 𝑓 (𝑥, 𝑦) is a continuous function on some rectangular region 𝑅 in the 𝑥𝑦− plane con-
taining the point (𝑎, 𝑏) in its interior, then the problem
𝑦 ′ = 𝑓 (𝑥, 𝑦), with 𝑦(𝑎) = 𝑏 (1.8)
has at least one solution defined on some open interval of 𝑥 containing 𝑥 = 𝑎; i.e., for all
𝑥 such that ∣𝑥 − 𝑎∣ < 𝛿 for some 𝛿 > 0.
This theorem says that if the function defining the differential equation is continuous then
it is integrable in the neighbourhood of the initial point. However, it does not guarantee
the uniqueness of the solution obtained for such functions.
To see this, consider the nonlinear initial value problem,
𝑑𝑦 5
= 𝑦 2/5 ; 𝑦(0) = 0.
𝑑𝑥 3
Since 𝐹 (𝑥, 𝑦) = 35 𝑦 2/5 is a continuous function, the above theorem assures us that it has
a solution.
⃝January
1.3 Exact Differential Equations 19
You can verify that the function 𝑦(𝑥) = (𝑥 + 𝐶)5/3 satisfies the differential equation for
any constant 𝐶. And from the initial condition, 𝑦(0) = 0, we can determine the value of
𝐶 to be zero. Hence the function,
𝑦(𝑥) = 𝑥5/3
satisfies the differential equation as well as the initial condition.

On the other hand, since the right-hand-side of the differential equation vanishes at 𝑦 = 0.
Hence 𝑦(𝑥) ≡ 0 is also an equilibrium solution to the differential equation and it satisfies
the initial condition.
Thus the solution is not unique, there are in fact infinitely many solutions to the above
initial value problem for any 𝑎 > 0 as the function
{
0, 0 ≤ 𝑥 < 𝑎;
𝑦(𝑥) =
(𝑥 − 𝑎)5/3 , 𝑥 ≥ 𝑎.
satisfies the conditions in the problem. Solutions that fails to be unique are usually called
singular solutions to the differential equation.
To ensure uniqueness of a solution we need to have an additional condition in the above
theorem.
Theorem 1.2.6 (Uniqueness).

If, in (1.2.5), 𝑓 (𝑥, 𝑦) is continuously differentiable, i.e., if the function
∂𝑓
∂𝑦
is continuous on the rectangle 𝑅, then the solution to the above equation (1.8) is unique
on some open interval containing 𝑥 = 𝑎.
Remark 1.2.7. The above condition for uniqueness can be weakened by using a condition
piecewise continuous instead of the condition that “ ∂𝑓
∂𝑦
is continuous”.
1.3 Exact Differential Equations
There are plenty of practical problems that are not separable, hence the above method
may not be applied on. Consider, for instance, the differential equation:
sin 𝑦
𝑦′ =
2𝑦 − 𝑥 cos 𝑦
⃝January
or equivalently
sin 𝑦𝑑𝑥 + (𝑥 cos 𝑦 − 2𝑦)𝑑𝑦 = 0.
Notice that the left hand side is the (total) differential of the function
𝐹 (𝑥, 𝑦) = 𝑥 sin 𝑦 − 𝑦 2 .
Recall that the total differential of a function 𝐹 (𝑥, 𝑦) of two variables is
∂𝐹 ∂𝐹
𝑑𝐹 (𝑥, 𝑦) = 𝑑𝑥 + 𝑑𝑦,
∂𝑥 ∂𝑦
for all (x,y) in the domain of 𝐹 .
Thus for 𝐹 (𝑥, 𝑦) = 𝑥 sin 𝑦 − 𝑦 2 , 𝑑𝐹 (𝑥, 𝑦) = sin 𝑦𝑑𝑥 + (𝑥 cos 𝑦 − 2𝑦)𝑑𝑦 = 0. This implies
that 𝐹 (𝑥, 𝑦) = 𝐶, that is, 𝑥 sin 𝑦 − 𝑦 2 = 𝐶, is the solution of the above DE.
Definition 1.3.1. The expression
𝑀 (𝑥, 𝑦)𝑑𝑥 + 𝑁 (𝑥, 𝑦)𝑑𝑦 = 0
is called an exact differential equation in some domain 𝐷 (an open connected set of
points) if there is a function 𝐹 (𝑥, 𝑦) such that
∂𝐹 ∂𝐹
= 𝑀 (𝑥, 𝑦) and = 𝑁 (𝑥, 𝑦),
∂𝑥 ∂𝑦
for all (𝑥, 𝑦) ∈ 𝐷.
If we can find a function 𝐹 (𝑥, 𝑦) such that
∂𝐹 ∂𝐹
= 𝑀 (𝑥, 𝑦) and = 𝑁 (𝑥, 𝑦),
∂𝑥 ∂𝑦
then the differential equation 𝑀 (𝑥, 𝑦)𝑑𝑥+𝑁 (𝑥, 𝑦)𝑑𝑦 = 0 is just 𝑀 (𝑥, 𝑦)𝑑𝑥+𝑁 (𝑥, 𝑦)𝑑𝑦 =
𝑑𝐹 = 0. But recall that, if 𝑑𝐹 = 0, then 𝐹 (𝑥, 𝑦) = constant. The equation 𝐹 (𝑥, 𝑦) = 𝑐,
where 𝑐 is an arbitrary constant, implicitly defines the general solution of the deferential
equation 𝑀 (𝑥, 𝑦)𝑑𝑥 + 𝑁 (𝑥, 𝑦)𝑑𝑦 = 0.
Now let us ask the following two fundamental questions. Given a Differential Equation
𝑀 (𝑥, 𝑦)𝑑𝑥 + 𝑁 (𝑥, 𝑦)𝑑𝑦 = 0
⃝January
1. How can we determine the existence of such a function 𝐹 (𝑥, 𝑦) ?
2. If it exists, how can we find it ?
The following theorem will answer the first question.
Theorem 1.3.2 (Test for Exactness). Let 𝑀 (𝑥, 𝑦), 𝑁 (𝑥, 𝑦), ∂𝑀
∂𝑦
and ∂𝑁
∂𝑥
be all continuous
functions within a rectangle 𝑅 (or some domain) in the 𝑥𝑦-plane. Then
𝑀 (𝑥, 𝑦)𝑑𝑥 + 𝑁 (𝑥, 𝑦)𝑑𝑦
is an exact differential in 𝑅 if and only if

∂𝑀 ∂𝑁
=
∂𝑦 ∂𝑥
every where in 𝑅.
Example 1.3.1. Consider the equation
𝑑𝑦 2𝑥𝑦 3 + 2
=− 2 2 .
𝑑𝑥 3𝑥 𝑦 + 8𝑒4𝑦
Then write
(2𝑥𝑦 3 + 2)𝑑𝑥 + (3𝑥2 𝑦 2 + 8𝑒4𝑦 )𝑑𝑦 = 0.
Let 𝑀 (𝑥, 𝑦) = 2𝑥𝑦 3 + 2 and 𝑁 (𝑥, 𝑦) = 3𝑥3 𝑦 2 + 8𝑒4𝑦 . Then
∂𝑀 ∂𝑁
= 6𝑥𝑦 = .
∂𝑦 ∂𝑥
Therefore, the given differential equation is exact.

1
(𝑦 ln 𝑦 − 𝑒−𝑥𝑦 )𝑑𝑥 + ( + 𝑥 ln 𝑦)𝑑𝑦 = 0.
𝑦
Let 𝑀 (𝑥, 𝑦) = 𝑦 ln 𝑦 − 𝑒−𝑥𝑦 and 𝑁 (𝑥, 𝑦) = 1

𝑦
+ 𝑥 ln 𝑦. Then ∂𝑀
∂𝑦
= ln 𝑦 + 𝑥𝑒−𝑥𝑦 + 𝑦
∂𝑁 ∂𝑀 ∂𝑁
and ∂𝑥
= ln 𝑦, which implies ∂𝑦
∕= ∂𝑥
. Therefore, the given differential equation is not
exact.
After knowing the exactness of a differential equation, the next question is ”How can we
solve the given equation?” The method for this is described here below.
⃝January
Suppose a differential equation 𝑀 (𝑥, 𝑦)𝑑𝑥 + 𝑁 (𝑥, 𝑦)𝑑𝑦 = 0 is exact. Then, there exists
a function 𝐹 (𝑥, 𝑦) such that
∂𝐹 ∂𝐹
𝑀= and 𝑁 = .
∂𝑥 ∂𝑦
∂𝐹
From 𝑀 = ∂𝑥
, we have (by integrating with respect to 𝑥)
∫
𝐹 (𝑥, 𝑦) = 𝑀 𝑑𝑥 + 𝐴(𝑦), (1.9)
where 𝐴(𝑦) is only a function of 𝑦 but constant with respect to 𝑥.

Now to determine 𝐴(𝑦) (the constant of integration), differentiate equation (1.9) with
respect to 𝑦 to get
∫
∂𝐹 ∂
= 𝑀 𝑑𝑥 + 𝐴′ (𝑦)
∂𝑦 ∂𝑦
which implies
∫ ∫
∂𝑀 ∂𝑀
𝑁 (𝑥, 𝑦) = 𝑑𝑥 + 𝐴′ (𝑦) and hence 𝐴′ (𝑦) = 𝑁 (𝑥, 𝑦) − 𝑑𝑥
∂𝑦 ∂𝑦
by exactness. Therefore,
∫ [ ∫ ]
∂𝑀
𝐴(𝑦) = 𝑁 (𝑥, 𝑦) − 𝑑𝑥 𝑑𝑦.
∂𝑦
Example 1.3.3. Solve the differential equation
sin 𝑦𝑑𝑥 + (𝑥 cos 𝑦 − 2𝑦)𝑑𝑦 = 0.
Solution
Let 𝑀 (𝑥, 𝑦) = sin 𝑦 and 𝑁 (𝑥, 𝑦) = 𝑥 cos 𝑦 − 2𝑦. Then 𝑀𝑦 = cos 𝑦 = 𝑁𝑥 . Since
𝑀, 𝑁, 𝑀𝑦 , 𝑁𝑥 are all continuous in R2 , the given equation is exact. Thus, there exists a
function 𝐹 (𝑥, 𝑦) such that
∂𝐹 ∂𝐹
= sin 𝑦 and = 𝑥 cos 𝑦 − 2𝑦,
∂𝑥 ∂𝑦
which implies 𝐹 (𝑥, 𝑦) = sin 𝑦𝑑𝑥 = 𝑥 sin 𝑦 + 𝐴(𝑦) and ∂𝐹 = 𝑥 cos 𝑦 + 𝐴′ (𝑦). That is,
∫
∂𝑦
𝑥 cos 𝑦 − 2𝑦 = 𝑥 cos 𝑦 + 𝐴′ (𝑦) which implies 𝐴′ (𝑦) = −2𝑦 and hence
∫
𝐴(𝑦) = −2𝑦𝑑𝑦 = −𝑦 2 + 𝐵.
⃝January
Therefore, 𝐹 (𝑥, 𝑦) = 𝑥 sin 𝑦 − 𝑦 2 + 𝐵 = constant, which implies
𝑥 sin 𝑦 − 𝑦 2 = 𝐶
determines 𝑦(𝑥) implicitly. □
(𝑥3 + 3𝑥𝑦 2 )𝑑𝑥 + (3𝑥2 𝑦 + 𝑦 3 )𝑑𝑦 = 0.
Solution
Step 1: Checking Exactness
Let 𝑀 (𝑥, 𝑦) = 𝑥3 + 3𝑥𝑦 2 and 𝑁 (𝑥, 𝑦) = 3𝑥2 𝑦 + 𝑦 3 . Then
∂𝑀 ∂𝑁
= 6𝑥𝑦 = .
∂𝑦 ∂𝑥
Therefore the given equation is exact.
Step 2: Finding Implicit Solution
Then to find 𝐹 (𝑥, 𝑦) we use

∫ ∫
1 3
𝐹 (𝑥, 𝑦) = 𝑀 𝑑𝑥 + 𝐴(𝑦) = (𝑥3 + 3𝑥𝑦 2 )𝑑𝑥 + 𝐴(𝑦) = 𝑥4 + 𝑥2 𝑦 2 + 𝐴(𝑦)
4 2
where 𝐴(𝑦) is a function of 𝑦 only. To find 𝐴(𝑦);
∂𝐹
= 3𝑥2 𝑦 + 𝐴′ (𝑦) = 𝑁 = 3𝑥2 𝑦 + 𝑦 3 ,
∂𝑦
which implies that 𝐴′ (𝑦) = 𝑦 3 and then 𝐴(𝑦) = 𝑦 3 𝑑𝑦 = 14 𝑦 4 + 𝐶.

∫
Therefore,
1 4 3 2 2 1 4
𝐹 (𝑥, 𝑦) = 𝑥 + 𝑥 𝑦 + 𝑦 +𝐶
4 2 4
1 4
= (𝑥 + 6𝑥 𝑦 + 𝑦 4 ) + 𝐶
2 2
(1.10)
4
⃝January
Step 3: Checking
Differentiate implicitly to check for 𝑦 ′ :
1
(4𝑥3 + 12𝑥𝑦 2 + 12𝑥2 𝑦𝑦 ′ + 4𝑦 3 𝑦 ′ ) = 0
4
which implies
𝑥3 + 3𝑥𝑦 2 + (3𝑥2 𝑦 + 𝑦 3 )𝑦 ′ = 0
and then
(𝑥3 + 3𝑥𝑦 2 )𝑑𝑥 + (3𝑥2 𝑦 + 𝑦 3 )𝑑𝑦 = 0.
Exercise 1.3.3. Solve each of the following differential equations.
1. (𝑦 + 𝑒𝑦 )𝑑𝑥 + 𝑥(1 + 𝑒𝑦 )𝑑𝑦 = 0; 𝑦 = 0 when 𝑥 = 1.

𝑑𝑦 2𝑥 + 1
2. = ; 𝑦(0) = 0.
𝑑𝑥 2𝑦 + 1
3. sin ℎ𝑥 cos 𝑦𝑑𝑥 = cos ℎ𝑥 sin 𝑦𝑑𝑦.
Integrating Factors
The differential equation 𝑦𝑑𝑥 + 2𝑥𝑑𝑦 = 0 is not exact. But if we multiply this equation
by 𝑦, the equation is changed to exact equation. That is,
𝑦 2 𝑑𝑥 + 2𝑥𝑦𝑑𝑦 = 0
is exact, since
∂𝑦 2 ∂(2𝑥𝑦)
= 2𝑦 = .
∂𝑦 ∂𝑥
Definition 1.3.4. If the differential equation 𝑀 (𝑥, 𝑦)𝑑𝑥 + 𝑁 (𝑥, 𝑦)𝑑𝑦 = 0 is not exact
but the differential equation
𝜇(𝑥, 𝑦)𝑀 (𝑥, )𝑑𝑥 + 𝜇(𝑥, 𝑦)𝑁 (𝑥, 𝑦)𝑑𝑦 = 0
is exact, then the multiplicative function 𝜇(𝑥, 𝑦) is called an integrating factor of the
DE.
(Of course 𝜇(𝑥, 𝑦) ∕= 0 so that the two equations are equivalent.)
⃝January
Example 1.3.5. Consider the differential equation
(3𝑦 + 4𝑥𝑦 2 )𝑑𝑥 + (2𝑥 + 3𝑥2 𝑦)𝑑𝑦 = 0.
Let 𝑀 (𝑥, 𝑦) = 3𝑦 + 4𝑥𝑦 2 and 𝑁 (𝑥, 𝑦) = 2𝑥 + 3𝑥2 𝑦. Then
∂𝑀 ∂𝑁
= 3 + 8𝑥𝑦 and = 2 + 6𝑥𝑦,
∂𝑦 ∂𝑥
which implies
∂𝑀 ∂𝑁
∕= .
∂𝑦 ∂𝑥
Hence the DE is not exact.
But if 𝜇(𝑥, 𝑦) = 𝑥2 𝑦 then 𝜇(𝑥, 𝑦)𝑀 𝑑𝑥 + 𝜇(𝑥, 𝑦)𝑁 𝑑𝑦 = 0 is exact, since
∂(𝜇(𝑥, 𝑦)𝑀 ) ∂(𝜇(𝑥, 𝑦)𝑁 )

= 6𝑥2 𝑦 + 12𝑥3 𝑦 2 = .
∂𝑦 ∂𝑥
□
Suppose we have a differential equation which is not exact but it can be made exact by
an integrating factor. Then we can ask the following fundamental questions.
1. How can we find the integrating factor 𝜇 ?
2. Given 𝜇, how can we solve the problem?
The method is described below.

Clearly 𝜇(𝑥, 𝑦) is any (non-zero) solution of the equation
∂ ∂
(𝜇𝑀 ) = (𝜇𝑁 ) (1.11)
∂𝑦 ∂𝑥
which is equivalent to the equation
𝜇𝑦 𝑀 + 𝜇𝑀𝑦 = 𝜇𝑥 𝑁 + 𝜇𝑁𝑥 .
This is a first-order partial differential equation in 𝜇. However the integrating factor 𝜇

can be found to be a function of 𝑥 alone 𝜇(𝑥) (or a function of 𝑦 alone 𝜇(𝑦)).
Then in this case equation (1.11) will be reduced to
𝑑𝜇 𝑑𝜇 𝑀𝑦 − 𝑁𝑥
𝜇𝑀𝑦 = 𝑁 + 𝜇𝑁𝑥 or equivalently = 𝜇( ) (1.12)
𝑑𝑥 𝑑𝑥 𝑁
⃝January
which is a separable differential equation.
This idea works correctly if the ratio
𝑀𝑦 − 𝑁𝑥
𝑁
is a function of 𝑥 only, that is,
𝑝(𝑥) = = a function of 𝑥.
𝑁
In this case ( )
𝑑𝜇 𝑀𝑦 − 𝑁𝑥
= 𝑑𝑥,
𝜇 𝑁
which implies
∫
𝑝(𝑥)𝑑𝑥
𝜇(𝑥) = 𝑒 .
If the quotient is not a function of 𝑥 alone, then the integrating factor 𝜇 can
𝑁
not be obtained using the above procedure, but we can try to find 𝜇 as a function of 𝑦
alone, 𝜇(𝑦).
Then when 𝜇(𝑦) is only a function of 𝑦, equation (1.11) will be reduced to
𝑑𝜇
𝑀 + 𝜇𝑀𝑦 = 𝜇𝑁𝑥
𝑑𝑦
which implies
( )
𝑑𝜇 𝑀𝑦 − 𝑁𝑥
= −𝜇 , which is a separable differential equation.
𝑑𝑦 𝑀
If the fraction
𝑀
is a function of 𝑦 alone, then
∫
𝜇(𝑦) = 𝑒− 𝑞(𝑦)𝑑𝑦
.
𝑑𝑥 + (3𝑥 − 𝑒−2𝑦 )𝑑𝑦 = 0. (1.13)
Let 𝑀 = 1 and 𝑁 = 3𝑥 − 𝑒−2𝑦 . Then ∂𝑀

∂𝑦
= 0 and ∂𝑁
∂𝑥
= 3 and hence ∂𝑀
∂𝑦
∕= ∂𝑁
∂𝑥
which
implies that the given differential equation is not exact.
Assume that the given equation has an integrating factor. But
𝑀𝑦 − 𝑁𝑥 0−3 −3𝑒2𝑦
= =
𝑁 3𝑥 − 𝑒−2𝑦 3𝑥𝑒2𝑦
⃝January
1.4 Linear First Order Differential Equations 27
which is not a function of 𝑥 alone. Hence obtaining 𝜇(𝑥) is not possible.
However,
𝑀𝑦 − 𝑁𝑥 0−3
= = −3
𝑀 1
can be considered as a function of 𝑦 alone. Therefore, it is possible to solve for 𝜇(𝑦) and
is given by
∫
𝜇(𝑦) = 𝑒− (−3)𝑑𝑦
= 𝑒3𝑦 .
Now to solve the problem in (1.13), multiplying the given equation by 𝜇(𝑦) = 𝑒3𝑦 we get
the equation
𝑒3𝑦 𝑑𝑥 + (3𝑥 − 𝑒−2𝑦 )𝑒3𝑦 𝑑𝑦 = 0,
which is an exact differential equation. Thus, there exists 𝐹 (𝑥, 𝑦) such that
∂𝐹 ∂𝐹
= 𝑒3𝑦 and = (3𝑥 − 𝑒−2𝑦 )𝑒3𝑦
∂𝑥 ∂𝑦
which implies that

∫ ∫
∂𝐹
𝐹 (𝑥, 𝑦) = 𝑑𝑥 = 𝑒3𝑦 𝑑𝑥 = 𝑥𝑒3𝑦 + 𝐴(𝑦).
∂𝑥
To determine 𝐴(𝑦) we use 𝐹 (𝑥, 𝑦) which is obtained above and differentiate it with respect
to 𝑦 and equate the result with (3𝑥 − 𝑒−2𝑦 )𝑒3𝑦 . Hence we have
∂𝐹
(3𝑥 − 𝑒−2𝑦 )𝑒3𝑦 = = 3𝑥𝑒3𝑦 + 𝐴′ (𝑦).
∂𝑦
Then 3𝑥𝑒3𝑦 − 𝑒𝑦 = 3𝑥𝑒3𝑦 + 𝐴′ (𝑦) which implies that

∫
′
𝐴 (𝑦) = −𝑒 and hence 𝐴(𝑦) = − 𝑒𝑦 𝑑𝑦 = −𝑒𝑦 + 𝐵.
𝑦
Therefore, 𝐹 (𝑥, 𝑦) = 𝑥𝑒3𝑦 − 𝑒𝑦 + 𝐵 = 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡. That means 𝑥𝑒3𝑦 − 𝑒𝑦 = 𝐶, where 𝐶 is

an arbitrary constant, defines the implicit solution of the Differential Equation in (1.13).
1.4 Linear First Order Differential Equations
Consider the general first-order linear differential equation
𝑎1 (𝑥)𝑦 ′ + 𝑎0 (𝑥)𝑦 = 𝑓 (𝑥), 𝑎1 (𝑥) ∕= 0 (1.14)
⃝January
By dividing both sides by 𝑎1 (𝑥) ∕= 0, we get 𝑦 ′ + 𝑝(𝑥)𝑦 = 𝑞(𝑥), where
𝑎0 (𝑥) 𝑓 (𝑥)
𝑝(𝑥) = and 𝑞(𝑥) = .
𝑎1 (𝑥) 𝑎1 (𝑥)
Here we assume that 𝑝(𝑥) and 𝑞(𝑥) are continuous.
There is a general approach to solve linear equations. To solve for 𝑦(𝑥) from the given
equation we start with the simplest case, when 𝑞(𝑥) = 0. That is, (1.14) becomes
𝑦 ′ + 𝑝(𝑥)𝑦 = 0. (1.15)
This problem is called a homogeneous version of (1.14). Now to solve (1.15) first we get
𝑦 ′ = −𝑝(𝑥)𝑦 and we divide both sides by 𝑦 and get
𝑦′
= −𝑝(𝑥).
𝑦
Then by integrating ∫ ∫
𝑑𝑦
=− 𝑝(𝑥)𝑑𝑥
𝑦
we get ∫
ln ∣𝑦∣ = − 𝑝(𝑥)𝑑𝑥 + 𝐶,
which implies
∫ ∫
∣𝑦∣ = 𝑒𝑐− 𝑝(𝑥)𝑑𝑥
= 𝐵𝑒− 𝑝(𝑥)𝑑𝑥
, for 𝐵 > 0.
Therefore,
∫
𝑦(𝑥) = 𝐴𝑒− 𝑝(𝑥)𝑑𝑥
, where 𝐴 is an arbitrary constant,
is a general solution of (1.15).
Example 1.4.1. Solve the following differential equations.
1. 𝑦 ′ + 2𝑥𝑦 = 0
2. (𝑥 + 2)𝑦 ′ − 𝑥𝑦 = 0
Solution:
𝑦′
1. If 𝑦 ′ + 2𝑥𝑦 = 0, then 𝑦
= −2𝑥. We integrate
∫ ∫
𝑑𝑦
= −2 𝑥𝑑𝑥
𝑦
2
to get ln ∣𝑦∣ = 𝑥2 + 𝐶 and hence 𝑦(𝑥) = 𝐴𝑒−𝑥 is the general solution.
⃝January
1.4 Linear First Order Differential Equations 29
2. If (𝑥 + 2)𝑦 ′ − 𝑥𝑦 = 0, then
𝑦′ 𝑥
= .
𝑦 𝑥+2
We integrate ∫ ∫
𝑑𝑦 𝑥
= 𝑑𝑥
𝑦 𝑥+2
𝐴
to get 𝑦(𝑥) = 𝐴𝑒(𝑥−2 ln(𝑥+2)) , or 𝑦(𝑥) = 𝑒𝑥 which is the general solution of
(𝑥 + 2)2
the given equation.
Now we want to solve the general first - order linear ordinary differential equation
𝑦 ′ + 𝑝(𝑥)𝑦 = 𝑞(𝑥) (1.16)
This can be done in two steps.
Step 1.
∫
Consider the homogeneous version of (1.16) and find the solution to be 𝑦ℎ (𝑥) = 𝐴𝑒− 𝑝(𝑥)𝑑𝑥
,
where ℎ indicate the general solution for the homogeneous part of the equation
Step 2.
To get the solution for the non-homogeneous part of the equation we vary the constant
𝐴 with different values of 𝑥.
Hence we assume that
∫
𝑦(𝑥) = 𝐴(𝑥)𝑒− 𝑝(𝑥)𝑑𝑥
(1.17)
is a solution for (1.16). Then (1.17) must satisfy (1.16). i.e.

∫ )′ ∫
𝐴(𝑥)𝑒− 𝑝(𝑥)𝑑𝑥 + 𝑝(𝑥) 𝐴(𝑥)𝑒− 𝑝(𝑥)𝑑𝑥 = 𝑞(𝑥),
( ( )
which implies
∫ ∫ ∫
𝐴′ (𝑥)𝑒− 𝑝(𝑥)𝑑𝑥
+ 𝐴(𝑥)(−𝑝(𝑥))𝑒− 𝑝(𝑥)𝑑𝑥
+ 𝑝(𝑥)𝐴(𝑥)𝑒− 𝑝(𝑥)𝑑𝑥
= 𝑞(𝑥).
Simplifying this gives us,

∫
𝐴′ (𝑥) = 𝑞(𝑥)𝑒 𝑝(𝑥)𝑑𝑥
.
⃝January
∫
𝐴′ (𝑥)𝑑𝑥 = 𝑞(𝑥)𝑒 𝑝(𝑥)𝑑𝑥 𝑑𝑥 to get
∫ ∫
Now integrate both sides
∫ ∫
𝐴(𝑥) = 𝑞(𝑥)𝑒 𝑝(𝑥)𝑑𝑥 𝑑𝑥 + 𝐶.
Hence the general solution of the non-homogeneous ODE (1.16) is given by:
∫ ∫
(∫ ∫
)
− 𝑝(𝑥)𝑑𝑥 − 𝑝(𝑥)𝑑𝑥 𝑝(𝑥)𝑑𝑥
𝑦(𝑥) = 𝐴(𝑥)𝑒 =𝑒 𝑞(𝑥)𝑒 𝑑𝑥 + 𝐶
∫ ∫
∫ ∫
− 𝑝(𝑥)𝑑𝑥 − 𝑝(𝑥)𝑑𝑥
= 𝐶𝑒 +𝑒 𝑞(𝑥)𝑒 𝑝(𝑥)𝑑𝑥 𝑑𝑥
= 𝑦ℎ (𝑥) + 𝑦𝑝 (𝑥)
Remark 1.4.1. It may not be necessary to memorize this long formula for 𝑦(𝑥). Instead,
we can carry out the following procedure.
Step 1. If the differential equation is linear, 𝑦 ′ + 𝑝(𝑥)𝑦 = 𝑞(𝑥), then first compute
∫
𝑝(𝑥)𝑑𝑥
𝑒 .
This is called an integrating factor for the linear equation.
Step 2. Multiply both sides of the differential equation by the integrating factor.
Step 3. Write the left side of the resulting equation as the derivative of the product of 𝑦
and the integrating factor. The integrating factor is designed to make this possible.
The right side is a function of just 𝑥.
Step 4. Integrate both sides of this equation with respect to 𝑥 and solve the resulting
equation for 𝑦, obtaining the general solution.
Example 1.4.2. Solve the differential equation 𝑦 ′ + 3𝑦 = 6.

The given equation is linear with 𝑝(𝑥) = 3 and 𝑞(𝑥) = 6.
Step 1. We compute the integrating factor

∫ ∫
𝑝(𝑥)𝑑𝑥 3𝑑𝑥
𝑒 =𝑒 = 𝑒3𝑥
Step 2. We multiply 𝑦 ′ + 3𝑦 = 6 by 𝑒3𝑥 to get 𝑦 ′ 𝑒3𝑥 + 3𝑦𝑒3𝑥 = 6𝑒3𝑥 .
Step 3. The above equation is equivalent to

𝑑(𝑦𝑒3𝑥 )
= 6𝑒3𝑥 .
𝑑𝑥
⃝January
1.5 *Nonlinear Differential Equations of the First Order 31
Step 4. We integrate
𝑑(𝑦𝑒3𝑥 )
∫ ∫
𝑑𝑥 = 6𝑒3𝑥
𝑑𝑥
and get 𝑦(𝑥)𝑒3𝑥 = 2𝑒3𝑥 + 𝐶. Then solve for 𝑦(𝑥) to get the general solution
𝑦(𝑥) = 𝐶𝑒−3𝑥 + 2 for an arbitrary constant 𝐶.
1.5 *Nonlinear Differential Equations of the First Or-

der
Some nonlinear differential equations can be reduced to linear form. In this section we
will consider three famous nonlinear equations: Bernoulli Equation, Riccati Equation
and Clairuat Equation
1.5.1 The Bernoulli Equation
A differential equation of the form
𝑦 ′ + 𝑝(𝑥)𝑦 = 𝑞(𝑥)𝑦 𝛼 ,
where 𝛼 is a constant, is called Bernoulli Equation.

If 𝛼 = 0, then the equation is linear and if 𝛼 = 1, then the equation is separable. We
have seen these two cases in the previous section.
For 𝛼 ∕= 1, use the change of variable 𝑢 = 𝑦 1−𝛼 . Then by differentiating with respect to
𝑥, we get 𝑢′ = (1 − 𝛼)𝑦 −𝛼 𝑦 ′ . But, from 𝑦 ′ + 𝑝(𝑥)𝑦 = 𝑞(𝑥)𝑦 𝛼 , we get 𝑦 ′ = 𝑞(𝑥)𝑦 𝛼 − 𝑝(𝑥)𝑦.
Then
𝑢′ = (1 − 𝛼)𝑦 −𝛼 𝑦 ′
= (1 − 𝛼)𝑦 −𝛼 (𝑞𝑦 𝛼 − 𝑝𝑦)
= (1 − 𝛼)(𝑞 − 𝑝𝑦 1−𝛼 )
= (1 − 𝛼)(𝑞 − 𝑝𝑢), since 𝑢 = 𝑦 1−𝛼
This implies that 𝑢′ + (1 − 𝛼)𝑝𝑢 = (1 − 𝛼)𝑞, which is a linear differential equation of first
order and hence we can solve it using one of the methods we have seen in the previous
sections.
Example 1.5.1. Solve the Bernoulli Equation 𝑦 ′ − 𝐴𝑦 = −𝐵𝑦 2 , where 𝐴 and 𝐵 are
positive constants. This equation is called Varhulst Equation.
⃝January
Solution
Since 𝛼 = 2, let 𝑢 = 𝑦 1−2 = 𝑦 −1 . Then 𝑢′ = −𝑦 −2 𝑦 ′ and substituting 𝐴𝑦 − 𝐵𝑦 2 for 𝑦 ′ we

get 𝑢′ = −𝑦 −2 (𝐴𝑦 − 𝐵𝑦 2 ) = 𝐵 − 𝐴𝑦 −1 = 𝐵 − 𝐴𝑢. Then we get the differential equation
𝑢′ + 𝐴𝑢 = 𝐵 which is equivalent to the equation 𝑑𝑢 − (𝐵 − 𝐴𝑢)𝑑𝑥 = 0. This implies
𝑑𝑢
= 𝑑𝑥
𝐵 − 𝐴𝑢
and we integrate ∫ ∫
𝑑𝑢
= 𝑑𝑥
𝐵 − 𝐴𝑢
and get 𝑢 = 𝐵
𝐴
+ 𝐶𝑒−𝐴𝑥 .
Therefore, the general solution of the original differential equation is
1 1
𝑦= = 𝐵
.
𝑢 𝐴
+ 𝐶𝑒−𝐴𝑥
1.5.2 The Riccati Equation.
A differential equation of the form
𝑦 ′ = 𝑝(𝑥)𝑦 2 + 𝑞(𝑥)𝑦 + 𝑟(𝑥)
is called Riccati equation. If 𝑝(𝑥) ≡ 0, then the equation is linear.
If we can obtain one particular solution 𝑠(𝑥) of the Riccati equation, then the change of
variables
1
𝑦 = 𝑠(𝑥) +
𝑧
transforms the Riccati equation in to a linear equation in 𝑥 and 𝑧. Then we find the
general solution of this linear equation and we use it to write the general solution of the
original Riccati equation.
Example 1.5.2. Solve the Riccati equation
1 2 1
𝑦′ = 𝑦 − 𝑦 + 1.
𝑥2 𝑥
⃝January
1.5 *Nonlinear Differential Equations of the First Order 33
Solution
By inspection one can see easily that 𝑦 = 𝑥 satisfies the above equation, and hence is one
solution of the equation.
1
Then use 𝑦 = 𝑥 + as a transformation function. Then we have
𝑧
𝑧′
𝑦′ = 1 − .
𝑧
Substituting this change of variables in to the equation we get
)2
𝑧′
( ( )
1 1 1 1
1− = 2 𝑥+ − 𝑥+ +1
𝑧 𝑥 𝑧 𝑥 𝑧
𝑧′ 1 1
⇒− 2 = + 2 2
𝑧 𝑧𝑥 𝑥 𝑧
′ 𝑧 1
⇒ 𝑧 = − − 2,
𝑥 𝑥
which is a linear first order equation in 𝑧.
Solving this last equation using the method in the previous section, we get
𝐶 − ln 𝑥
𝑧(𝑥) = , for arbitrary constant 𝐶
𝑥
𝑥
Hemce 𝑦(𝑥) = 𝑥 + is the general solution of the given Riccati equation. □
𝐶 − ln 𝑥
1.5.3 The Clairuat Equation
A nonlinear differential equation of the form
𝑦 = 𝑥𝑦 ′ + 𝑔(𝑦 ′ ),
is called a Clairuat equation.
To solve such equation, let us differentiate both sides of the equation with respect to 𝑥.
Then we get, 𝑦 ′ = 𝑦 ′ + 𝑥𝑦 ′′ + 𝑔 ′ (𝑦 ′ )𝑦 ′′ ,
which implies that 𝑦 ′′ (𝑥 + 𝑔 ′ (𝑦 ′ )) = 0 and hence 𝑦 ′′ = 0 or 𝑥 + 𝑔 ′ (𝑦 ′ ) = 0.
Solving 𝑦 ′′ = 0 gives us the general solution 𝑦 = 𝑎𝑥 + 𝑏 and solving 𝑥 + 𝑔 ′ (𝑦 ′ ) = 0 gives

us a singular solution.
Note that, a solution to a differential equation is said to be a singular solution, if the
initial value problem fails to have a unique solution.
⃝January
Example 1.5.3. Solve the Clairuat equation
1
𝑦 = 𝑥𝑦 ′ + .
𝑦′
Solution
Differentiating both sides with respect to 𝑥 to get
𝑦 ′′
𝑦 ′ = 𝑦 ′ + 𝑥′′ − .
(𝑦 ′ )2
This implies ( )
′′ 1
𝑦 𝑥− =0
(𝑦 ′ )2
and then solving 𝑦 ′′ = 0 gives us a general solution 𝑦 = 𝑎𝑥 + 𝑏 and solving
1
𝑥− =0
1 − (𝑦 ′ )2
√
gives us a singular solution. Then (𝑦 ′ )2 = 1
𝑥
which implies 𝑦 ′ = 1
√
± 𝑥
. Hence 𝑦 = 2 𝑥 + 𝑐
is a singular solution.
Remark 1.5.1. The general solution of the Clairuat equation is 𝑦 = 𝑎𝑥 + 𝑏. Therefore,

our main focus for such equation is in finding the singular solution.
1.6 Exercises
Exercise 1.6.1. Solve each of the following differential equations.
1. 𝑥𝑦 ′ + 𝑦 = 6𝑥2
2. 𝑥𝑦 ′ + 2𝑦 = 𝑥 + 2 with inital condition 𝑦(0) = 1.
⃝January
1.6 Exercises 35
This page is left blank intensionally.
⃝January
⃝January
Chapter 2
Ordinary Differential Equations of

The Second and Higher Order
A second-order differential equation is a differential equation containing a second derivative

of a dependent variable with respect to the independent variable but no higher derivative.
The theory of second-order differential equations is vast, and here we will focus on linear
second-order equations, which have many important uses. Most of the results are given for
a higher order ODEs and second order ODEs are special cases, but most of our examples
are for second order ODEs.
2.1 Basic Theory
In this section, we will focus on the general theory of linear ordinary differential equations
before we start to discuss about solving such problems.
Definition 2.1.1. A linear ordinary differential equation of order 𝑛 in the dependent

variable 𝑦 and independent variable 𝑥 is an equation which can be expressed in the form:
𝑎𝑛 (𝑥)𝑦 (𝑛) + 𝑎𝑛−1 (𝑥)𝑦 (𝑛−1) + ⋅ ⋅ ⋅ + 𝑎1 (𝑥)𝑦 ′ + 𝑎0 (𝑥) = 𝑓 (𝑥), (2.1)
where 𝑎𝑛 (𝑥) ∕≡ 0 and the functions 𝑎0 , . . . , 𝑎𝑛 are continuous real-valued functions of

𝑥 ∈ [𝑎, 𝑏].
The function 𝑓 (𝑥) is called the non-homogeneous term and all the points 𝑥𝜖 ∈ [𝑎, 𝑏] in
which 𝑎𝑛 (𝑥𝜖 ) = 0 are called singular points of the DE (2.1).
⃝January
38 Ordinary Differential Equations of The Second and Higher Order
If 𝑓 (𝑥) ≡ 0, then (2.1) is reduced to:
𝑎𝑛 (𝑥)𝑦 (𝑛) + 𝑎𝑛−1 (𝑥)𝑦 (𝑛−1) + ⋅ ⋅ ⋅ + 𝑎1 (𝑥)𝑦 ′ + 𝑎0 (𝑥) = 0 (2.2)
This equation is known as homogeneous Linear ODE of order 𝑛.
Example 2.1.1. The equation 𝑦 ′′ +3𝑥𝑦 ′ + 𝑥3 𝑦 = 𝑒𝑥 is a non homogeneous linear ordinary

differential equation of the 2nd order, whereas 𝑦 ′′ + 3𝑥𝑦 ′ + 𝑥3 𝑦 = 0 is a homogeneous
linear ordinary differential equation of the 2nd order.
Theorem 2.1.2 (Basic Existence Theorem for IVP). Consider the linear ODE given in
(2.1), where 𝑎0 , 𝑎1 , . . . , 𝑎𝑛−1 , 𝑎𝑛 and 𝑓 are continuous functions on the interval [𝑎, 𝑏] and
𝑎𝑛 (𝑥) ∕= 0, ∀𝑥 ∈ [𝑎, 𝑏]. Furthermore, let 𝑥0 be any point in [𝑎, 𝑏] and let 𝑐0 , 𝑐1 . . . 𝑐𝑛−1
be arbitrary real constants. Then there exists a unique solution function 𝑔(𝑥) of (2.1) on
[𝑎, 𝑏] satisfying the initial conditions,
𝑔(𝑥0 ) = 𝑐0 , 𝑔 ′ (𝑥0 ) = 𝑐1 , . . . , 𝑔 𝑛−1 (𝑥0 ) = 𝑐𝑛−1 .
2.2 General Solution of Homogeneous Linear ODEs
Consider the linear differential equation
𝑦 ′′ + 𝑦 = 0. (2.3)
Then, 𝑦1 = cos 𝑥 and 𝑦2 = sin 𝑥 are solutions of the differential equation (2.3). Let 𝑐1
and 𝑐2 be arbitrary constants. Then
𝑦 = 𝑐1 𝑦1 + 𝑐2 𝑦2 = 𝑐1 cos 𝑥 + 𝑐2 sin 𝑥
is also a solution of (2.3). Indeed, 𝑦 ′ = −𝑐1 sin 𝑥 + 𝑐2 cos 𝑥, and 𝑦 ′′ = −𝑐1 cos 𝑥 − 𝑐2 sin 𝑥
which implies that
𝑦 ′′ + 𝑦 = (−𝑐1 cos 𝑥 − 𝑐2 sin 𝑥) + (𝑐1 cos 𝑥 + 𝑐2 sin 𝑥) = 0
for all 𝑥. Therefore, any linear combination of the functions 𝑦1 = cos 𝑥 and 𝑦2 = cos 𝑥 is
a solution for the given differential equation.
This condition can be generalized for any homogenous linear differential equation in the
following theorem.
⃝January
2.2 General Solution of Homogeneous Linear ODEs 39
Theorem 2.2.1 (Linear Combination of Solutions). If 𝑦1 , 𝑦2 , . . . , 𝑦𝑘 are solutions of the
homogeneous linear ODE (2.1) and if 𝑐1 , 𝑐2 , . . . , 𝑐𝑘 are arbitrary constants, then the linear
combination
𝑘
∑
𝑦 = 𝑐1 𝑦1 + 𝑐2 𝑦2 + ⋅ ⋅ ⋅ + 𝑐𝑘 𝑦𝑘 = 𝑐𝑖 𝑦 𝑖
𝑖=1
is also a solution of (2.1). That is, any linear combination of solutions of a linear homo-
geneous differential equation is also a solution.
Definition 2.2.2 (Linearly Dependent and Linearly Independent Functions).
1. The functions 𝑓1 , 𝑓2 , . . . , 𝑓𝑛 are said to be Linearly Dependent (LD) on some interval

[𝑎, 𝑏] if there are constants 𝑐1 , 𝑐2 , . . . , 𝑐𝑛 , not all zero, such that
𝑐1 𝑓1 (𝑥) + 𝑐2 𝑓2 (𝑥) + ⋅ ⋅ ⋅ + 𝑐𝑛 𝑓𝑛 (𝑥) = 0 (2.4)
for all 𝑥 ∈ [𝑎, 𝑏].
2. If the relation (2.4) is true only when 𝑐1 = 𝑐2 = ⋅ ⋅ ⋅ = 𝑐𝑛 = 0, then the functions

𝑓1 , 𝑓2 , . . . , 𝑓𝑛 are said to be Linearly Independent (LI) on [𝑎, 𝑏].
Example 2.2.1. Examples of LD and LI functions.
1. The functions 𝑓1 (𝑥) = 𝑒𝑥 and 𝑓2 (𝑥) = 4𝑒𝑥 are Linearly Dependent on R since
−4𝑓1 (𝑥) + 𝑓2 (𝑥) = −4𝑒𝑥 + 4𝑒𝑥 = 0, for all 𝑥 ∈ R.
2. The functions
𝑓1 (𝑥) = 𝑒𝑥 , 𝑓2 (𝑥) = 𝑒−𝑥 , 𝑓3 (𝑥) = sinh 𝑥
are linearly dependent on R since

𝑒𝑥 − 𝑒−𝑥
𝑓3 (𝑥) = sinh 𝑥 =
2
and (1)𝑓1 (𝑥) + (−1)𝑓2 (𝑥) + (−2)𝑓3 (𝑥) = 0, ∀𝑥 ∈ R.
3. The two functions 𝑓1 (𝑥) = 𝑥 and 𝑓2 (𝑥) = 𝑥3 are Linearly Independent on R, since
for 𝑐1 , 𝑐2 ∈ R,
𝑐1 𝑓1 (𝑥) + 𝑐2 𝑓2 (𝑥) = 𝑐1 𝑥 + 𝑐2 𝑥3 = 0, ∀𝑥 ∈ R∖{0}
implies 𝑐1 = 0 and 𝑐2 = 0.
⃝January
The following theorem guarantees that any 𝑛th order Linear Homogenous Ordinary Dif-
ferential Equation has 𝑛 linearly independent solutions.
Theorem 2.2.3 (Existence of Linearly Independent Solutions for a LHODE). The Linear
Homogenous Differential Equation (LHODE) (2.2) always has 𝑛 Linearly Independent (LI)
solutions. Furthermore, if 𝑓1 (𝑥), 𝑓2 (𝑥), . . . , 𝑓𝑛 (𝑥) are 𝑛 LI solutions of (2.2), then every
solution of (2.2) can be expressed as a linear combination of these solution functions. i.e.
If 𝑦 is a solution for (2.2), then
𝑛
∑
𝑦(𝑥) = 𝑐𝑖 𝑓𝑖 (𝑥)
𝑖=1
for some 𝑐1 , ..., 𝑐𝑛 ∈ R..
Example 2.2.2. Consider the second order linear homogenous DE
𝑦 ′′ + 𝑦 = 0.
Then 𝑓1 (𝑥) = sin 𝑥, 𝑓2 (𝑥) = cos 𝑥 are LI solutions of the given equation. Then {sin 𝑥, cos 𝑥}
is the fundamental set of solutions of the given DE and hence the general solution of the
DE is given by 𝑦(𝑥) = 𝑐1 sin 𝑥 + 𝑐2 cos 𝑥, for constants 𝑐1 , 𝑐2 ∈ R.
Definition 2.2.4. If 𝑓1 (𝑥), 𝑓2 (𝑥), . . . , 𝑓𝑛 (𝑥) are 𝑛 linearly independent solutions of (2.2)
on [𝑎, 𝑏], then the set {𝑓1 (𝑥), 𝑓2 (𝑥), . . . 𝑓𝑛 (𝑥)} is called the Fundamental Set of Solu-
tions of (2.2) and the function
𝑓 (𝑥) = 𝑐1 𝑓1 (𝑥) + 𝑐2 𝑓2 (𝑥) + ⋅ ⋅ ⋅ + 𝑐𝑛 𝑓𝑛 (𝑥), 𝑥 ∈ [𝑎, 𝑏],
where 𝑐1 , 𝑐2 , . . . , 𝑐𝑛 are arbitrary constants is called a General Solution of (2.2) on [𝑎, 𝑏].
and each 𝑓1 , 𝑓2 , . . . , 𝑓𝑛 are called particular solutions.
Example 2.2.3. Consider the third order linear homogenous DE
𝑦 ′′′ − 2𝑦 ′′ − 𝑦 ′ + 2𝑦 = 0.
a) The functions 𝑒𝑥 , 𝑒−𝑥 , 𝑒2𝑥 are (particular) solutions (check!)
b) 𝑒𝑥 , 𝑒−𝑥 and 𝑒2𝑥 are LI (check!)
c) Therefore, the general solution of the given equation is given by:
𝑦(𝑥) = 𝑐1 𝑒𝑥 + 𝑐2 𝑒−𝑥 + 𝑐3 𝑒2𝑥 .
⃝January
There is a simple test to determine whether a given set of functions is linearly independent
or dependent on an open interval 𝐼 = (𝑎, 𝑏), for some real numbers 𝑎, 𝑏, by using the idea
of determinant of a matrix.
Definition 2.2.5. Let 𝑓1 (𝑥), 𝑓2 (𝑥), . . . , 𝑓𝑛 (𝑥) be 𝑛 real valued functions each of which
has an (𝑛 − 1)𝑡ℎ derivative on the interval [𝑎, 𝑏]. The determinant:

𝑓1 (𝑥) 𝑓2 (𝑥) ... 𝑓𝑛 (𝑥)

𝑓 ′ (𝑥) ′
𝑓2 (𝑥) ... 𝑓𝑛′ (𝑥)
1
W[f1 , f2 . . . , fn ] = = W(𝑥)

.. .. ..

. . .

(𝑛−1) (𝑛−1) (𝑛−1)
𝑓1 (𝑥) 𝑓2 (𝑥) . . . 𝑓𝑛 (𝑥)

is called the Wronskian of these 𝑛 functions.
Example 2.2.4. The function 𝑦1 (𝑥) = 𝑒2𝑥 and 𝑦2 (𝑥) = 𝑥𝑒4𝑥 are solutions of the second
order linear homogenous differential equation 𝑦 ′′ − 4𝑦 ′ + 4𝑦 = 0. Then the Wronskian,
W(𝑥) of 𝑦1 and 𝑦2 is

𝑒2𝑥 𝑥𝑒4𝑥
W(𝑥) = 2𝑥 4𝑥 = 𝑒4𝑥 + 2𝑥𝑒4𝑥 − 2𝑥𝑒4𝑥 = 𝑒4𝑥

2𝑒 𝑒 + 4𝑥𝑒4𝑥
Question: Are the two functions 𝑦1 (𝑥) = 𝑒2𝑥 and 𝑦2 (𝑥) = 𝑥𝑒4𝑥 linearly indepen-
dent?
The above question can be easily answered using the following theorem.
Theorem 2.2.6 (Wronskian Test for Linearly Independence). The 𝑛 functions 𝑓1 , 𝑓2 , . . . , 𝑓𝑛
are Linearly Independent on an interval [𝑎, 𝑏] if and only if the Wronskian of 𝑓1 , 𝑓2 , . . . , 𝑓𝑛
is different from zero for some 𝑥 ∈ [𝑎, 𝑏]. That is, 𝑓1 , 𝑓2 , . . . , 𝑓𝑛 are LI if and only if there
exists 𝑥 ∈ [𝑎, 𝑏] such that W(𝑥) ∕= 0.
Example 2.2.5. 1. Show that 𝑥 and 𝑥2 are Linearly Independent.
Solution
Consider the Wronskian of 𝑥 and 𝑥2 ,

𝑥 𝑥2
W(𝑥, 𝑥2 ) = = 2𝑥2 − 𝑥2 = 𝑥2

1 2𝑥
This implies W(𝑥, 𝑥2 ) = 𝑥2 ∕= 0, ∀𝑥 ∕= 0 and hence 𝑥 and 𝑥2 are LI.
2. Show that 𝑒𝑥 , 𝑒−𝑥 , 𝑒2𝑥 are Linearly Independent.
⃝January
Solution
Consider the Wronskian of 𝑒𝑥 , 𝑒−𝑥 and 𝑒2𝑥

𝑒−𝑥
𝑥
𝑒 𝑒2𝑥

W(𝑥) = 𝑒𝑥 −𝑒−𝑥 2𝑒2𝑥 ,

𝑒−𝑥 4𝑒2𝑥
𝑥
𝑒
which is equal to
𝑒𝑥 (−4𝑒𝑥 + 2𝑒𝑥 ) − 𝑒−𝑥 (4𝑒3𝑥 − 2𝑒3𝑥 ) + 𝑒2𝑥 (1 + 1)
= −2𝑒2𝑥 − 2𝑒2𝑥 + 2𝑒2𝑥
= −2𝑒2𝑥 ∕= 0, ∀𝑥 ∈ R.
Hence, 𝑒𝑥 , 𝑒−𝑥 and 𝑒2𝑥 are Linearly Independent.
3. Show that the functions
𝑓1 (𝑥) = 𝑒𝑥 , 𝑓2 (𝑥) = 𝑒−𝑥 , 𝑓3 (𝑥) = sinh 𝑥
are linearly dependent on R, since
𝑒𝑥 − 𝑒−𝑥
𝑓3 (𝑥) = sinh 𝑥 = .
2
Solution
Consider the Wronskian of 𝑒𝑥 , 𝑒−𝑥 and sinh 𝑥

𝑥 −𝑥
𝑒−𝑥 𝑒 −𝑒
𝑥
𝑒

2
𝑥 −𝑥

W(𝑥) = 𝑒𝑥 −𝑒−𝑥 𝑒 +𝑒 2
= 0, ∀𝑥 ∈ ℝ.

𝑥 −𝑥
𝑒−𝑥 𝑒 −𝑒
𝑥
𝑒 2

Hence 𝑒𝑥 , 𝑒−𝑥 and sinh 𝑥 are linearly dependent.
Remark 2.2.7. The Wronskian of 𝑛 solutions 𝑓1 , 𝑓2 , . . . , 𝑓𝑛 of the DE (2.2) is either

identically zero on [𝑎, 𝑏] or else is never zero on [𝑎, 𝑏]. That is, if 𝑓1 , 𝑓2 , . . . , 𝑓𝑛 are solutions
of the DE (2.2), then W(𝑥) = 0, ∀𝑥 ∈ [𝑎, 𝑏] if 𝑓1 , 𝑓2 , . . . , 𝑓𝑛 are LD. or W(𝑥) ∕= 0, ∀𝑥 ∈
[𝑎, 𝑏] if 𝑓1 , 𝑓2 , . . . , 𝑓𝑛 are LI.
⃝January
2.2.1 Reduction of Order
In the preceding section we saw that the general solution of a homogeneous linear second-
order differential equation
𝑦 ′′ + 𝑝(𝑥)𝑦 + 𝑞(𝑥)𝑦 = 0 (2.5)
is a linear combination of two functions 𝑦(𝑥) = 𝑐1 𝑦1 (𝑥) + 𝑐2 𝑦2 (𝑥), where 𝑦1 and 𝑦2 are
linearly independent solutions on some interval I.
In this method we can construct a second solution 𝑦2 of a homogeneous equation (2.5)

(even when the coefficients in (2.5) are variable) provided that we know a nontrivial so-
lution 𝑦1 of the DE. The basic idea described in this section is that equation (2.5) can be
reduced to a linear first-order DE by means of a substitution involving the known solution
𝑦1 . A second solution 𝑦2 of (2.5) is apparent after this first-order differential equation is
solved.
The method is described bellow.
Suppose that 𝑦1 denotes a nontrivial solution of (2.5) and that 𝑦1 is defined on an interval
I. We want to find a second solution 𝑦2 so that the set consisting of 𝑦1 and 𝑦2 is linearly
independent on I.
The quotient 𝑦2 /𝑦1 is nonconstant on I, that is,
𝑦2 (𝑥)
= 𝑢(𝑥)
𝑦1 (𝑥)
or 𝑦2 (𝑥) = 𝑢(𝑥)𝑦1 (𝑥). The function 𝑢(𝑥) can be found by substituting
𝑦2 (𝑥) = 𝑢(𝑥)𝑦1 (𝑥)
into the given differential equation.

Consider the derivatives 𝑦2′ = 𝑢′ 𝑦1 + 𝑢𝑦1′ and 𝑦2′′ = 𝑢′′ 𝑦1 + 2𝑢′ 𝑦1′ + 𝑢𝑦1′′ and substituting
these in (2.5) we get
(𝑢′′ 𝑦1 + 2𝑢′ 𝑦1′ + 𝑢𝑦1′′ ) + 𝑝(𝑥)(𝑢′ 𝑦1 + 𝑢𝑦1′ ) + 𝑞(𝑥)𝑢𝑦1 = 0
⃝January
and simplifying this gives us
𝑢′′ 𝑦1 + 𝑢′ (2𝑦1′ + 𝑝(𝑥)𝑦1 ) + 𝑢(𝑦1′′ + 𝑝(𝑥)𝑦1′ + 𝑞(𝑥)𝑦1 ) = 0.
But 𝑦1 , by assumption, is a solution of (2.5) and hence
𝑢(𝑦1′′ + 𝑝(𝑥)𝑦1′ + 𝑞(𝑥)𝑦1 ) = 0
which implies
𝑢′′ 𝑦1 + 𝑢′ (2𝑦1′ + 𝑝(𝑥)𝑦1 ) = 0.
This is a second order DE in 𝑢. Let 𝑢′ = 𝑧. Then 𝑢′′ = 𝑧 ′ . Using separation of variables

we get
𝑧′ −2𝑦1′
= −𝑝
𝑧 𝑦1
which is a first order DE and hence the name reduction of order and integrating and
∫
taking the constant zero gives us ln 𝑧 = −2 ln 𝑦1 − 𝑝𝑑𝑥. This implies
1 − ∫ 𝑝𝑑𝑥
𝑧= 𝑒 .
𝑦12
But 𝑧 = 𝑢′ . Then
1 − ∫ 𝑝𝑑𝑥
𝑢′ = 𝑒
𝑦12
and then ∫ ( )
𝑦2 1 − ∫ 𝑝𝑑𝑥
=𝑢= 𝑒 𝑑𝑥.
𝑦1 𝑦12
Therefore, the second solution for the given equation is
∫ ( )
1 − ∫ 𝑝𝑑𝑥
𝑦2 = 𝑦1 𝑒 𝑑𝑥.
𝑦12
Example 2.2.6. The function 𝑦1 (𝑥) = 𝑥 is a solution of the homogenous DE
(𝑥2 − 1)𝑦 ′′ − 2𝑥𝑦 ′ + 2𝑦 = 0.
Solve the given DE.
Solution
The given equation is equivalent to 𝑦 ′′ + 𝑝(𝑥)𝑦 ′ + 𝑞(𝑥)𝑦 = 0, where

−2𝑥 2
𝑝(𝑥) = and 𝑞(𝑥) = .
𝑥2 − 1 𝑥2 −1
⃝January
2.3 Homogeneous LODE with Constant Coefficients 45
If 𝑦2 is a second solution of the given equation then 𝑦2 (𝑥) = 𝑢(𝑥)𝑦1 (𝑥), where
∫ ( ) ) ∫ ( ) ∫
1 − ∫ 𝑥−2𝑥
(
𝑑𝑥 1 ln ∣𝑥 2 −1∣ 1 1
𝑢(𝑥) = 𝑒 2 −1
𝑑𝑥 = 𝑒 𝑑𝑥 = (1 − )𝑑𝑥 = 𝑥 + .
𝑥2 𝑥2 𝑥2 𝑥
Therefore, 𝑦1 (𝑥) = 𝑥 and 𝑦2 (𝑥) = 𝑢(𝑥)𝑦1 (𝑥) = 𝑥2 + 1 are two linearly independent
solutions of the given equation and hence the general solution of the given equation is
𝑦(𝑥) = 𝑐1 𝑥 + 𝑐2 (𝑥2 + 1), where 𝑐1 and 𝑐2 are constants. □
2.3 Homogeneous LODE with Constant Coefficients

Definition 2.3.1. A Differential Equation
𝑏𝑛 𝑦 (𝑛) + 𝑏𝑛−1 𝑦 (𝑛−1) + ⋅ ⋅ ⋅ + 𝑏1 𝑦 ′ + 𝑏0 𝑦 = 0, (2.6)
where 𝑏0 , 𝑏1 , . . . , 𝑏𝑛 are all real constants, is called a Homogenous Linear Differential

Equation of constant coefficients.
Let 𝑓 (𝑥) be any solution of (2.6) in [𝑎, 𝑏]. Then
𝑏𝑛 𝑓 (𝑛) (𝑥) + 𝑏𝑛−1 𝑓 (𝑛−1) (𝑥) + ⋅ ⋅ ⋅ + 𝑏1 𝑓 ′ (𝑥) + 𝑏0 𝑓 (𝑥) = 0 for all 𝑥 ∈ [𝑎, 𝑏].
Hence the derivatives of 𝑓 are linearly dependent since at least one of the coefficients
𝑏0 , 𝑏1 , . . . , 𝑏𝑛 is different from zero.
The simplest case with this property is a function 𝑓 such that
𝑓 (𝑘) (𝑥) = 𝑐𝑓 (𝑥), ∀𝑥 ∈ [𝑎, 𝑏]
for some constant 𝑐.

Let 𝑓 (𝑥) = 𝑒𝜆𝑥 . Then 𝑓 𝑘 (𝑥) = 𝜆𝑘 𝑓 (𝑥) = 𝜆𝑘 𝑒𝜆𝑥 which implies 𝑐 = 𝜆𝑘 .
Thus we will look for the solution of (2.6) in the form 𝑦 = 𝑒𝜆𝑥 where the constant 𝜆 will
be chosen so that 𝑦 = 𝑒𝜆𝑥 does satisfy the equation (2.6).
Now insert 𝑦 = 𝑒𝜆𝑥 into (2.6) to get;
(𝑏𝑛 𝜆𝑛 + 𝑏𝑛−1 𝜆𝑛−1 + ⋅ ⋅ ⋅ + 𝑏1 𝜆 + 𝑏0 )𝑒𝜆𝑥 = 0.
Hence, if 𝑒𝜆𝑥 is a solution of the equation in (2.6), then 𝜆 should satisfy:
𝑏0 𝜆𝑛 + 𝑏1 𝜆𝑛−1 + ⋅ ⋅ ⋅ + 𝑏𝑛−1 𝜆 + 𝑏𝑛 = 0, (2.7)

since 𝑒𝜆𝑥 ∕= 0 for all 𝑥 ∈ R.
⃝January
Definition 2.3.2. The algebraic equation (2.7) is called an Auxiliary equation or char-
acteristic equation of the given differential equation in (2.6).
There are 3 different cases of the roots of (2.7).
Case 1. Distinct Real Roots
Suppose that (2.7) has 𝑛 distinct roots, 𝜆1 , 𝜆2 , . . . , 𝜆𝑛 where 𝜆𝑖 ∕= 𝜆𝑗 , for 𝑖 ∕= 𝑗. Then,

the solutions 𝑒𝜆1 𝑥 , 𝑒𝜆2 𝑥 , . . . , 𝑒𝜆𝑛 𝑥 are linearly independent. (Use the Wronskian to prove
this.)
If 𝜆1 , 𝜆2 , . . . , 𝜆𝑛 are the 𝑛 distinct real roots of (2.7), then the general solution of (2.6)
is: 𝑛
∑
𝜆1 𝑥 𝜆2 𝑥 𝜆𝑛 𝑥
𝑦(𝑥) = 𝑐1 𝑒 + 𝑐2 𝑒 + ⋅ ⋅ ⋅ + 𝑐𝑛 𝑒 = 𝑐𝑖 𝑒𝜆𝑖 𝑥 ,
𝑖=1
where 𝑐1 , 𝑐2 , . . . , 𝑐𝑛 are arbitrary constants.
Example 2.3.1. 1. For the differential equation 𝑦 ′′ − 3𝑦 ′ + 2𝑦 = 0, the characteristic

equation is: 𝜆2 − 3𝜆 + 2 = 0, and 𝜆1 = 2 and 𝜆2 = 1 are the two distinct real roots
of this characteristic equation. Hence, the general solution of the given differential
equation is 𝑦(𝑥) = 𝑐1 𝑒2𝑥 + 𝑐2 𝑒𝑥 .
2. For the differential equation 𝑦 ′′′ −4𝑦 ′′ +𝑦 ′ +6𝑦 = 0, the corresponding characteristic
equation is: 𝜆3 −4𝜆2 +𝜆+6 = 0 with distinct real roots 𝜆1 = 2, 𝜆2 = 3 and 𝜆3 = −1.
Therefore, the general solution of the give equation is 𝑦(𝑥) = 𝑐1 𝑒2𝑥 + 𝑐2 𝑒3𝑥 + 𝑐3 𝑒−𝑥 .
Case 2. Repeated Real Roots
To understand the situation let us consider the following example.
Example 2.3.2. Consider the DE 𝑦 ′′ − 6𝑦 ′ + 9𝑦 = 0. Then, its characteristic equation is

𝜆2 − 6𝜆 + 9 = 0, which implies (𝜆 − 3)2 = 0. Therefore, 𝜆1 = 𝜆2 = 3, which is a repeated
real root. One of the solutions of the given linear differential equation is 𝑒3𝑥 .
Let 𝑦1 (𝑥) = 𝑒3𝑥 . The given equation will have two linearly independent solutions and the
second solution can be found by using the method of reduction of order. Let 𝑦2 be
another solution so that 𝑦1 and 𝑦2 are linearly independent. Then 𝑦2 = 𝑢𝑦1 , where
∫ ( − ∫ −6𝑑𝑥 ) ∫ ( 6𝑥 ) ∫
𝑒 𝑒
𝑢(𝑥) = ( )2 𝑑𝑥 = 𝑑𝑥 = 1𝑑𝑥 = 𝑥.
𝑒3𝑥 𝑒6𝑥
⃝January
2.3 Homogeneous LODE with Constant Coefficients 47
Therefore 𝑦2 (𝑥) = 𝑥𝑒3𝑥 and 𝑦(𝑥) = 𝑐1 𝑒3𝑥 + 𝑐2 𝑥𝑒3𝑥 is a general solution for constants 𝑐1
and 𝑐2 .
Remark 2.3.3. Given a differential equation:
1. if the characteristic equation has double real root 𝜆, then 𝑒𝜆𝑥 and 𝑥𝑒𝜆𝑥 are two
linearly independent solutions and;
2. if the characteristic equation has triple root 𝜆, then the corresponding linearly in-
dependent solutions are 𝑒𝜆𝑥 , 𝑥𝑒𝜆𝑥 and 𝑥2 𝑒𝜆𝑥 .
Let us proof the first part of the above remark for a second order linear homogenous
differential equation.
If the given DE is 𝑎𝑦 ′′ + 𝑏𝑦 ′ + 𝑐𝑦 = 0, then its characteristic equation is 𝑎𝜆2 + 𝑏𝜆 + 𝑐 = 0
−𝑏
and then 𝜆 = 𝜆1 = 𝜆2 = 2𝑎
. One of the solution of the given DE is 𝑦1 = 𝑒𝜆𝑥 . Then we
can use the method of reduction of order to find a second solution 𝑦2 so that 𝑦1 and 𝑦2
are linearly independent.
The given equation is equivalent to
𝑏 𝑐
𝑦 ′′ + 𝑦 ′ + 𝑦 = 0
𝑎 𝑎
and 𝑦2 = 𝑢𝑦1 , where
∫ 𝑏 ∫ −𝑏 𝑥 ∫
𝑒− 𝑎 𝑑𝑥
∫ ( )
𝑒𝑎
𝑢= ( )2 𝑑𝑥 = = 1𝑑𝑥 = 𝑥,
𝑒𝜆𝑥 𝑒2𝜆𝑥
−𝑏
since 2𝜆 = 𝑎
and hence 𝑦2 = 𝑥𝑒𝜆𝑥 .
The following theorem is a generalization for the above remark.
Theorem 2.3.4.
1. If the characteristic equation (2.7) has the real root 𝜆 occurring 𝑘 times (𝑖.𝑒.𝜆1 =
𝜆2 = ⋅ ⋅ ⋅ = 𝜆𝑘 ) where 𝑘 ≤ 𝑛, then the part of the general solution for (2.6)
corresponding to this 𝑘 fold repeated root is
(𝑐1 + 𝑐2 𝑥 + 𝑐3 𝑥2 + ⋅ ⋅ ⋅ + 𝑐𝑘 𝑥𝑘−1 )𝑒𝜆𝑥
2. If further, the remaining roots are the distinct real roots 𝜆𝑘+1 , 𝜆𝑘+2 , . . . , 𝜆𝑛 , the
general solution of (2.6) will be:
𝑦(𝑥) = 𝑐1 𝑒𝜆𝑥 + 𝑐2 𝑥𝑒𝜆𝑥 + 𝑐3 𝑥2 𝑒𝜆𝑥 + ⋅ ⋅ ⋅ + 𝑐𝑘 𝑥𝑘−1 𝑒𝜆𝑥 + 𝑐𝑘+1 𝑒𝜆𝑘+1 𝑥 + ⋅ ⋅ ⋅ + 𝑐𝑛 𝑒𝜆𝑛 𝑥 .
⃝January
Example 2.3.3.
1. Consider the Differential Equation
𝑦 (4) − 5𝑦 ′′′ + 6𝑦 ′′ + 4𝑦 ′ − 8𝑦 = 0.
The corresponding characteristic equation is 𝜆4 − 5𝜆3 + 6𝜆2 − 8 = 0 and the roots

are 𝜆1 = 𝜆2 = 𝜆3 = 2, 𝜆4 = −1.
Therefore, the general solution is given by 𝑦(𝑥) = 𝑐1 𝑒2𝑥 + 𝑐2 𝑥𝑒2𝑥 + 𝑐3 𝑥2 𝑒2𝑥 + 𝑐4 𝑒−𝑥 ,
where 𝑐1 , 𝑐2 , 𝑐3 and 𝑐4 are constants.
2. Consider the Differential Equation 𝑦 ′′′ − 4𝑦 ′′ − 3𝑦 ′ + 18𝑦 = 0. The roots of the

characteristic equation are, 𝜆1 = 𝜆2 = 3 and 𝜆3 = −2 and hence the general
solution of the equation is: 𝑦(𝑥) = 𝑐1 𝑒3𝑥 + 𝑐2 𝑥𝑒3𝑥 + 𝑐3 𝑒−2𝑥 , where 𝑐1 , 𝑐2 and 𝑐3 are
constants.
Case 3. Conjugate Complex Roots
Suppose the equation (2.7) has a complex root 𝜆 = 𝑎 + 𝑖𝑏, where 𝑎, 𝑏 ∈ R. Then (we
¯ = 𝑎 − 𝑖𝑏 is also a root
know from the theory of algebraic equations that) the conjugate 𝜆
of (2.7) and the corresponding part of the general solution of (2.6) will be:
𝑘1 𝑒(𝑎+𝑖𝑏)𝑥 + 𝑘2 𝑒(𝑎−𝑖𝑏)𝑥 .
But 𝑒𝑎+𝑖𝑏 = 𝑒𝑎 𝑒𝑖𝑏 = 𝑒𝑎 (cos 𝑏 + 𝑖 sin 𝑏), (by applying Euler’s formula) and then
𝑘1 𝑒(𝑎+𝑖𝑏)𝑥 + 𝑘2 𝑒(𝑎−𝑖𝑏)𝑥 = 𝑘1 𝑒𝑎𝑥 (cos 𝑏𝑥 + 𝑖 sin 𝑏𝑥) + 𝑘2 𝑒𝑎𝑥 (cos 𝑏𝑥 − 𝑖 sin 𝑏𝑥)
= 𝑒𝑎𝑥 [(𝑘1 + 𝑘2 ) cos 𝑏𝑥 + 𝑖(𝑘1 − 𝑘2 ) sin 𝑏𝑥]
= 𝑒𝑎𝑥 (𝑐1 cos 𝑏𝑥 + 𝑐2 sin 𝑏𝑥),
where 𝑐1 = 𝑘1 + 𝑘2 and 𝑐2 = 𝑖(𝑘1 − 𝑘2 ) are arbitrary constants from the set of complex
numbers C.
¯ are each 𝑘 fold roots of (2.7), then the
On the other hand if 𝑎 + 𝑖𝑏 = 𝜆 and 𝑎 − 𝑖𝑏 = 𝜆
part of the general solution that corresponds to this root is
𝑒𝑎𝑥 [ 𝑐1 + 𝑐2 𝑥 + ⋅ ⋅ ⋅ + 𝑐𝑘 𝑥𝑘−1 cos 𝑏𝑥 + 𝑐𝑘+1 + 𝑐𝑘+2 𝑥 + ⋅ ⋅ ⋅ 𝑐2𝑘 𝑥𝑘−1 sin 𝑏𝑥].

( ) ( )
Example 2.3.4. Solve 𝑦 ′′ − 2𝑦 ′ + 10𝑦 = 0.
⃝January
2.4 Nonhomogeneous Equations with Constant Coefficients 49
Solution
The characteristic equation of the given equation is 𝜆2 −2𝜆+10 = 0 with roots 𝜆1 = 1+3𝑖
and 𝜆2 = 1 − 3𝑖. Then 𝑦1 = 𝑒1+3𝑖 and 𝑦2 = 𝑒1−3𝑖 are two independent solutions of the
given equation. Therefore, 𝑦 = 𝑐1 𝑦1 + 𝑐2 𝑦2 , where 𝑐1 and 𝑐2 are constants, is a general
solution of the given equation. That means
𝑦(𝑥) = 𝑒𝑥 (𝑐1 cos 3𝑥 + 𝑐2 sin 3𝑥).
2.3.1 Exercises
Exercise 2.3.5. Solve each of the following Differential Equation.
1. 𝑦 ′′ + 𝑦 = 0.
2. 𝑦 ′′ − 6𝑦 ′ + 25𝑦 = 0.
3. 𝑦 (4) − 4𝑦 ′′′ + 14𝑦 ′′ − 20𝑦 ′ + 25𝑦 = 0, where 𝜆1 = 𝜆2 = 1 + 2𝑖 and 𝜆3 = 𝜆4 = 1 − 2𝑖.
2.4 Nonhomogeneous Equations with Constant Coef-

ficients
Consider the equation

𝑑2 𝑥 𝑑𝑥
𝑚 + 𝑐 + 𝑘𝑥 = 𝐹 (𝑡). (2.8)
𝑑𝑡2 𝑑𝑡
which governs the displacement 𝑥(𝑡) of a mechanical oscillator. Here 𝐹 (𝑡) is the forcing
function and the equation is a non-homogeneous linear ODE with constant coefficients.
There are several practical problems which can be modeled in this form.
Recall that differential equations of the form
𝑏𝑛 (𝑥)𝑦 (𝑛) + ⋅ ⋅ ⋅ + 𝑏1 (𝑥)𝑦 ′ + 𝑏0 (𝑥)𝑦 = 𝑓 (𝑥), where 𝑓 (𝑥) ∕= 0 (2.9)
are called nonhomogeneous differential equations. In the previous sections we have seen
how to solve homogeneous differential equations. In this section we are going to see how
to solve differential equations of the form
𝑏𝑛 𝑦 (𝑛) + 𝑏𝑛−1 𝑦 (𝑛−1) + ⋅ ⋅ ⋅ + 𝑏1 𝑦 ′ + 𝑏0 𝑦 = 𝑓 (𝑥), (2.10)
⃝January
where 𝑏𝑛 , ..., 𝑏0 are constants, which is called a nonhomogeneous differential equation with
constant coefficients. The following theorem is very important in such cases.
Theorem 2.4.1 (Homogeneous-Nonhomogeneous Solution Relation). Consider the non-

homogeneous differential equation
𝑏𝑛 (𝑥)𝑦 (𝑛) + ⋅ ⋅ ⋅ + 𝑏1 (𝑥)𝑦 ′ + 𝑏0 (𝑥)𝑦 = 𝑓 (𝑥), where 𝑓 (𝑥) ∕≡ 0.
If 𝑓 (𝑥) ≡ 0, then the equation becomes a homogeneous equation.
1. If 𝑦1 and 𝑦2 are solutions of the nonhomogeneous equation on an interval I, then

𝑦1 − 𝑦2 is also a solution of the homogeneous equation in the interval I.
2. If 𝑦1 is a solution of the nonhomogeneous equation and 𝑦2 is a solution of the homo-

geneous equation in an interval I, then 𝑦1 + 𝑦2 is a solution of the nonhomogeneous
equation in the interval I.
The following remark follows directly from the theorem given above.
Remark 2.4.2. Suppose 𝑦ℎ (𝑥) denote the general solution of the homogeneous part and
𝑦𝑝 (𝑥) denote the particular solution of the DE. Then the general solution of (2.10) is
given by 𝑦(𝑥) = 𝑦ℎ (𝑥) + 𝑦𝑝 (𝑥).
Theorem 2.4.3 (Superposition Principle). If 𝑦ℎ (𝑥) is a general solution of the homo-

geneous part of (2.10) on an interval [𝑎, 𝑏] and 𝑦𝑝1 (𝑥), 𝑦𝑝2 (𝑥), . . . , 𝑦𝑝𝑘 (𝑥) are particular
solutions of (2.10) corresponding to 𝑓1 (𝑥), 𝑓2 (𝑥), . . . , 𝑓𝑘 (𝑥) respectively on the right hand
side, then the general solution of (2.10) where, 𝑓 (𝑥) = 𝑓1 (𝑥) + ⋅ ⋅ ⋅ + 𝑓𝑘 (𝑥) on [𝑎, 𝑏], is
𝑦(𝑥) = 𝑦ℎ (𝑥) + 𝑦𝑝1 (𝑥) + 𝑦𝑝2 (𝑥) + ⋅ ⋅ ⋅ + 𝑦𝑝𝑘 (𝑥).
The above result is called a Superposition Principle. It tells us that the response 𝑦𝑝 to
a superposition of inputs (the forcing functions 𝑓1 + 𝑓2 + ⋅ ⋅ ⋅ + 𝑓𝑘 ) is the superposition of
their individual outputs (𝑦𝑝1 , . . . , 𝑦𝑝𝑘 ).
We are going to use these results in solving nonhomogeneous differential equations with
constant coefficients in the coming sections.
⃝January
2.4.1 The undetermined coefficient method
Definition 2.4.4.
1. A function is called an undetermined coefficient function (UC function) if it is

either:
a) a function defined by (a linear combination) of the following
i) 𝑥𝑛 , 𝑛 = 0, 1, 2, . . . ,
ii) 𝑒𝑎𝑥 , where 𝑎 is any non-zero constant
iii) sin(𝑏𝑥 + 𝑐), where 𝑏, 𝑐 are constants, such that 𝑏 ∕= 0.
iv) cos(𝑏𝑥 + 𝑐), where 𝑏, 𝑐 are constants, such that 𝑏 ∕= 0.
or
b) a function which is defined as a finite product of two or more functions of the

above 4 types.
2. Let f be an UC function. The set S of functions consisting of 𝑓 and all the derivatives
of 𝑓 which are mutually LI UC functions is said to be the UC set of function f, if S
is a finite set and we shall denote it by S.
Example 2.4.1.
1. Let 𝑓 (𝑥) = 𝑥3 . Then f is UC function.
𝑓 ′ (𝑥) = 3𝑥2 , 𝑥2 is UC function.
𝑓 ′′ (𝑥) = 6𝑥, 𝑥 is UC function.
𝑓 ′′′ (𝑥) = 6, 1 is UC function.
Therefore, 𝑆 = {1, 𝑥, 𝑥2 , 𝑥3 }.
2. Let 𝑓 (𝑥) = sin(2𝑥). Then 𝑓 is an UC function.
𝑓 ′ (𝑥) = 2 cos(2𝑥), cos(2𝑥) is UC function.
𝑓 ′′ (𝑥) = −4 sin(2𝑥), sin(2𝑥) = 𝑓 (𝑥).
Therefore, 𝑆 = {sin(2𝑥), cos(2𝑥)}.
⃝January
3. Let 𝑔(𝑥) = 2𝑥𝑒−𝑥 . 𝑔 is an UC function (as a product of UC function).
𝑔 ′ (𝑥) = 2𝑒−𝑥 − 2𝑥𝑒−𝑥 and 𝑒−𝑥 , 𝑥𝑒−𝑥 are UC functions.
𝑔 ′′ (𝑥) = −4𝑒−𝑥 + 2𝑥𝑒−𝑥 and 𝑒−𝑥 , 𝑥𝑒−𝑥 are UC functions.
Therefore, 𝑆 = {𝑒−𝑥 , 𝑥𝑒−𝑥 }.
4. The function
1
𝑓 (𝑥) =
𝑥
is not a UC function.
We outline the method by using the following example.
Example 2.4.2. Consider the differential equation
𝑦 (4) − 𝑦 ′′ = 3𝑥2 − sin 2𝑥 (2.11)
1) First find the general solution to the homogeneous part
𝑦 (4) − 𝑦 ′′ = 0.
The characteristic equation of the given equation is 𝜆4 −𝜆2 = 0. Then 𝜆2 (𝜆2 −1) = 0
and hence 𝜆1 = 𝜆2 = 0 and 𝜆3 = 1, 𝜆4 = −1. Therefore, the general solution is:
𝑦ℎ (𝑥) = 𝑐1 + 𝑐2 𝑥 + 𝑐3 𝑒𝑥 + 𝑐4 𝑒−𝑥 .
2) The forcing function (nonhomogeneous term) is a combination of 𝑥2 and sin 2𝑥.
Next find the set of UC functions corresponding to the component functions
𝑓1 (𝑥) = 3𝑥2
with 𝑆1 = {𝑥2 , 𝑥, 1} and

𝑓2 (𝑥) = − sin 2𝑥
with 𝑆2 = {sin 2𝑥, cos 2𝑥}.

To find a particular solution 𝑦𝑝1 (𝑥) corresponding to 𝑓1 (𝑥), tentatively we seek it to be a
linear combination of the functions in 𝑆1 , i.e.
𝑦𝑝1 (𝑥) = 𝐴𝑥2 + 𝐵𝑥 + 𝐶,
where A,B and C are called the undetermined constants.
⃝January
∙ Check each term in 𝑦𝑝1 (𝑥) for duplication with terms in 𝑦ℎ (𝑥). Here the 𝐵𝑥 and C
terms are constant multiples of 𝑐2 𝑥 and 𝑐1 respectively.
∙ If there is any duplicate, then successively multiply each member of 𝑆𝑖 by the

lowest positive integral power of 𝑥, until (so that) the resulting revised set contains
no duplicate of the terms in the homogeneous (and previously found particular 𝑦𝑝𝑖 ’s)
solutions.
– 𝑦𝑝1 (𝑥) = 𝑥(𝐴𝑥2 + 𝐵𝑥 + 𝐶) = 𝐴𝑥3 + 𝐵𝑥2 + 𝐶𝑥 still a duplicate is there,
– 𝑦𝑝1 (𝑥) = 𝑥2 (𝐴𝑥2 + 𝐵𝑥 + 𝐶) = 𝐴𝑥4 + 𝐵𝑥3 + 𝐶𝑥2 , no more duplicate.
∙ Substitute the final revised form into the equation and determine the coefficients
A,B and C.
(4)
∙ 𝑦𝑝1 − 𝑦𝑝′′1 (𝑥) = 3𝑥2 which implies 24𝐴 − 12𝐴𝑥2 − 6𝐵𝑥 − 2𝐶 = 3𝑥2 .
Equating the coefficients of like terms we get:

⎧
⎨ −12𝐴 = 3


−6𝐵 = 0

⎩ 24𝐴 − 2𝐶 = 0

−1
This implies, 𝐴 = 4
, 𝐵 = 0 and 𝐶 = −3.
Therefore,
1
𝑦𝑝1 (𝑥) = − 𝑥4 − 3𝑥2 .
4
Next, we need to find 𝑦𝑝2 (𝑥) which corresponds to 𝑓2 (𝑥) = − sin 2𝑥. We seek 𝑦𝑝2 (𝑥)
to be a linear combination of the elements of 𝑆2 , that is,
𝑦𝑝2 (𝑥) = 𝐷 sin 2𝑥 + 𝐸 cos 2𝑥.
– Check for a duplicate both in 𝑦ℎ (𝑥) and 𝑦𝑝1 (𝑥). – No duplicate.
– Then substitute in 𝑦 (4) − 𝑦 ′′ = − sin 2𝑥 which implies
𝑦𝑝(4)
2
(𝑥) − 𝑦𝑝′′2 (𝑥) = − sin 2𝑥.
Hence,
(24 𝐷 sin 2𝑥 + 24 𝐸 cos 2𝑥) − (−22 𝐷 sin 2𝑥 − 22 cos 2𝑥) = − sin 2𝑥.
⃝January
Therefore, 20𝐷 sin 2𝑥 + 20𝐸 cos 2𝑥 = − sin 2𝑥 and then
1
20𝐷 = −1 ⇐⇒ 𝐷 = −
20
and 20𝐸 = 0. Therefore,
1
𝑦𝑝2 (𝑥) = − sin 2𝑥.
20
Hence the general solution of (2.11) is:
1 1
𝑦(𝑥) = 𝑦ℎ (𝑥)+𝑦𝑝1 (𝑥)+𝑦𝑝2 (𝑥) = 𝐶1 +𝐶2 𝑥+𝐶3 𝑒𝑥 +𝐶4 𝑒−𝑥 − 𝑥4 −3𝑥2 − sin 2𝑥.
4 20
Example 2.4.3. Solve 𝑦 ′′ − 2𝑦 ′ − 3𝑦 = 2𝑒−𝑥 − 10 sin 𝑥.

Let 𝐹 (𝑥) = 2𝑒−𝑥 − 10 sin 𝑥, 𝑓1 = 2𝑒−𝑥 , 𝑓2 = −10 sin 𝑥. Then 𝑆1 = {𝑒−𝑥 } and 𝑆2 =
{sin 𝑥, cos 𝑥}
∙ Solution of the homogeneous part 𝑦 ′′ − 2𝑦 ′ − 3𝑦 = 0.
Then 𝜆2 − 2𝜆 − 3 = 0 and hence 𝜆1 = 3, 𝜆2 = −1. Therefore, 𝑦ℎ (𝑥) = 𝐶1 𝑒3𝑥 +

𝐶2 𝑒−𝑥 .
∙ Particular solution corresponding to 𝑓1 (𝑥) = 2𝑒−𝑥 : 𝑦𝑝1 (𝑥) = 𝐵𝑒−𝑥 which duplicates
with 𝐶2 𝑒−𝑥 .
This implies, 𝑦𝑝1 (𝑥) = 𝐵𝑥𝑒−𝑥 , – no more duplicate.
∙ Insert this into 𝑦 ′′ − 2𝑦 ′ − 3𝑦 = 2𝑒−𝑥 to get
𝑦𝑝′′1 − 2𝑦𝑝′ 1 − 3𝑦𝑝1 = 2𝑒−𝑥 .
This implies, (−2𝐵𝑒−𝑥 + 𝐵𝑥𝑒−𝑥 ) − 2(𝐵𝑒−𝑥 − 𝐵𝑥𝑒−𝑥 ) − 3𝐵𝑥𝑒−𝑥 = 2𝑒−𝑥 .

Hence −4𝐵𝑒−𝑥 = 2𝑒−𝑥 ⇐⇒ 𝐵 = − 21 . Therefore,
1
𝑦𝑝1 (𝑥) = − 𝑥𝑒−𝑥 .
2
∙ Particular solution corresponding to 𝑓2 (𝑥) = −10 sin 𝑥.

Let 𝑦𝑝2 (𝑥) = 𝐷 sin 𝑥 + 𝐸 cos 𝑥 (No duplicate both in 𝑦ℎ and 𝑦𝑝1 ).
Then, 𝑦𝑝′′2 − 2𝑦𝑝′ 2 − 3𝑦𝑝2 = −10 sin 𝑥. This implies,
(−𝐷 sin 𝑥 − 𝐸 cos 𝑥) − 2(𝐷 cos 𝑥 − 𝐸 sin 𝑥) − 3(𝐷 sin 𝑥 + 𝐸 cos 𝑥) = −10 sin 𝑥.
⃝January
Simplifying this gives us, (2𝐸 − 4𝐷) sin 𝑥 + (−2𝐷 − 4𝐸) cos 𝑥 = −10 sin 𝑥.
Therefore, {
2𝐸 − 4𝐷 = −10
−2𝐷 − 4𝐸 = 0
20 10 20 10
which implies 𝐷 = and 𝐸 = . Then, 𝑦𝑝1 (𝑥) = sin 𝑥 + cos 𝑥.
3 3 3 3
Therefore, the general solution is
1 20 10 10
𝑦(𝑥) = 𝐶1 𝑒3𝑥 + 𝐶2 𝑒−𝑥 − 𝑥𝑒−𝑥 + sin 𝑥 + sin 𝑥 + cos 𝑥.
2 3 3 3
Exercise 2.4.5. Solve each of the following DEs.
1. 𝑦 ′′ − 𝑔𝑦 = 4 + 5 sinh 3𝑥
2. 𝑦 ′′ − 2𝑦 ′ + 2𝑦 = 2𝑥2 + 𝑒𝑥 + 2𝑥𝑒𝑥 + 4𝑒3𝑥
2.4.2 Variation of Parameters
The Undetermined Coefficient method is easier to apply, but works only for constant coef-
ficients and certain types of non-homogeneous terms (or forcing functions). If the forcing
𝑥+1
function is, for example, of the form 𝑓 (𝑥) = tan 𝑥 or 𝑓 (𝑥) = 2 , then both of them
𝑥 +1
are not UC functions. and hence we can not employ the method of undetermined coeffi-
cients in those cases. Hence, we need another method which works for more general set
of problems. In this subsection we will consider the method of Variation of Parameters
for a second order linear ordinary differential equation.
Consider the following second order linear differential equation.
𝑦 ′′ + 𝑏1 (𝑥)𝑦 ′ + 𝑏2 (𝑥)𝑦 = 𝑓 (𝑥), (2.12)
where 𝑏1 , 𝑏2 and 𝑓 are continuous functions. Suppose that the general solution for the
homogeneous part of (2.12) is
𝑦ℎ (𝑥) = 𝑐1 𝑦1 (𝑥) + 𝑐2 𝑦2 (𝑥).
Now we want to get a particular solution corresponding to 𝑓 (𝑥) and this can be done
by varying the constants, 𝑐1 and 𝑐2 with respect to 𝑥. If 𝑦𝑝 is a particular solution
corresponding to 𝑓 (𝑥), then
𝑦𝑝 (𝑥) = 𝑐1 (𝑥)𝑦1 (𝑥) + 𝑐2 (𝑥)𝑦2 (𝑥).
⃝January
We differentiate and substitute it in (2.12) to get
𝑦𝑝′′ (𝑥) + 𝑏1 (𝑥)𝑦𝑝′ + 𝑏2 (𝑥)𝑦𝑝 = 𝑓 (𝑥).
But 𝑦𝑝′ = 𝑐1 𝑦1′ + 𝑐2 𝑦2′ + 𝑐′1 𝑦1 + 𝑐′2 𝑦2 .

Since we are going to have only one equation with two variable functions 𝑐1 and 𝑐2 , we are
free to choose a condition which simplifies the equation. Therefore, we take the condition
𝑐′1 𝑦1 + 𝑐′2 𝑦2 = 0.
This will simplify the equation as
′′
𝑦𝑝′′ = 𝑐1 𝑦1′′ + 𝑐′1 𝑦1′ + 𝑐′2 𝑦2′ + 𝑐2 𝑦2
and after simplification, the equation (2.12) becomes
𝑐1 (𝑦1′′ + 𝑏1 𝑦1′ + 𝑏2 𝑦1 ) + 𝑐2 (𝑦2′′ + 𝑏1 𝑦2′ + 𝑏2 𝑦2 ) + 𝑐′1 𝑦1′ + 𝑐′2 𝑦2′ = 𝑓.
Since 𝑦1 and 𝑦2 are linearly independent solutions for the homogeneous part of equation
(2.12) we have the following system of equations:
{
𝑐′1 𝑦1′ + 𝑐′2 𝑦2′ = 𝑓
(2.13)
𝑐′1 𝑦1 + 𝑐′2 𝑦2 = 0,
which is a system of two algebraic equations in 𝑐′1 and 𝑐′2 . Then (2.13) has a unique
solution if the determinant of the coefficient matrix is non-zero, that is,

𝑦 (𝑥) 𝑦 (𝑥)
1 2
∕= 0.

′
𝑦1 (𝑥) 𝑦2′ (𝑥)
However, the above determinant is the Wronskian of the functions 𝑦1 and 𝑦2 . Since 𝑦1
and 𝑦2 are LI functions, then
W[y1 ,y2 ] (𝑥) ∕= 0.
Hence by Crammer’s rule we have:

0 𝑦
2
𝑓 𝑦2′

𝑊1 (𝑥)
c′1 (x) = =
𝑊 (𝑥) 𝑊 (𝑥)
⃝January
and
𝑦 0
1

𝑦1 𝑓 𝑊2 (𝑥)
c′2 (x) = =
𝑊 (𝑥) 𝑊 (𝑥)
By integrating both sides we will get:
[∫ ] [∫ ]
W1 (𝑥) W2 (𝑥)
𝑦𝑝 (𝑥) = 𝑑𝑥 𝑦1 (𝑥) + 𝑑𝑥 𝑦2 (𝑥).
W(𝑥) W(𝑥)
𝑦 ′′ − 4𝑦 = 8𝑥.
Solution
First solve the homogeneous equation 𝑦 ′′ − 4𝑦 = 0.

Then the characteristic equation is 𝜆2 − 4 = 0, which implies 𝜆 = ±2. If 𝑦1 and 𝑦2 are
two linearly independent solutions of the equation 𝑦 ′′ − 4𝑦 = 0, then 𝑦1 = 𝑒2𝑥 , 𝑦2 = 𝑒−2𝑥 .
Therefore, the general solution of the homogeneous equation is 𝑦ℎ (𝑥) = 𝑐1 𝑒2𝑥 + 𝑐2 𝑒−2𝑥
and
𝑦
1 𝑦2 𝑒2𝑥 𝑒−2𝑥
W(x) = ′ = = −2 − 2 = −4,

′ −2𝑥
2𝑥
‘𝑦1 𝑦2 2𝑒 −2𝑒

0 𝑦2 0 𝑒2𝑥
W1 (x) = = = −8𝑥𝑒−2𝑥

′ 8𝑥 −2𝑒 −2𝑥
8𝑥 𝑦2
Therefore,
−8𝑥𝑒−2𝑥
∫ ∫ ∫
𝑊1 (𝑥)
𝑐1 (𝑥) = 𝑑𝑥 = 𝑑𝑥 = 2 𝑥𝑒−2𝑥 𝑑𝑥 = −𝑥𝑒−2𝑥 + 𝑒−2𝑥
𝑊 (𝑥) −4
and similarly
8𝑥𝑒2𝑥
∫ ∫ ∫
𝑊2 (𝑥)
𝑐2 (𝑥) = 𝑑𝑥 = 𝑑𝑥 = −2 𝑥𝑒2𝑥 𝑑𝑥 = 𝑥𝑒−2𝑥 − 𝑒−2𝑥 .
𝑊 (𝑥) −4
Therefore, 𝑦𝑝 (𝑥) = 𝑐1 (𝑥)𝑒2𝑥 + 𝑐2 (𝑥)𝑒−2𝑥 a particular solution and the general solution for
the problem is 𝑦(𝑥) = 𝑦ℎ (𝑥) + 𝑦𝑝 (𝑥).
Remark 2.4.6. This method looks easier when the integrands (or the quotients of the
Wronskian) are simple. However, it could be very difficult to get the particular solution
when the integrand is complicated.
⃝January
2.5 The Laplace Transform Method to Solve ODEs
In the previous sections we have discussed how to solve differential equations of the form:
𝑎𝑛 𝑦 (𝑛) + 𝑎𝑛−1 𝑦 (𝑛−1) + ⋅ ⋅ ⋅ + 𝑎1 𝑦 ′ + 𝑎0 𝑦 = 𝑓 (𝑥) (2.14)
by finding the general solutions and then evaluate the arbitrary constants in accordance
with the given initial conditions. However, the solution methods mainly dependent on
the structure of the forcing function 𝑓 (𝑥). Moreover, all the coefficients are assumed to
be constants. To address problems with more general forcing function and some form of
variable coefficients, we discuss the use of Laplace transform as possible alternative.
Definition 2.5.1 (Laplace Transform). The Laplace Transform of a function 𝑓 (𝑡), if it

exists, is denoted by ℒ{𝑓 (𝑡)} is given by,
∫ ∞
ℒ{𝑓 (𝑡)} = 𝑒−𝑠𝑡 𝑓 (𝑡)𝑑𝑡,
0
where 𝑠 is a real number called a parameter of the transform. For short we may write,
∫ ∞ ∫ ∞
−𝑠𝑡
ℱ(𝑠) to denote 𝑒 𝑓 (𝑡)𝑑𝑡. i.e., ℱ(𝑠) = 𝑒−𝑠𝑡 𝑓 (𝑡)𝑑𝑡.
0 0
Example 2.5.1. Find the Laplace Transform of the constant function 𝑓 (𝑡) = 1.
∫ ∞
ℒ{1} = 𝑒−𝑠𝑡 × 1𝑑𝑡.
0
Solution
∫ ∞
ℒ{1} = 𝑒−𝑠𝑡 × 1𝑑𝑡
0
∫ 𝑇
= lim 𝑒−𝑠𝑡 𝑑𝑡
𝑇 →∞ 0
𝑇
𝑒−𝑠𝑡
= lim
𝑇 →∞ −𝑠
0
[ ]
−1 −𝑠𝑇 1
= lim 𝑒 +
𝑇 →∞ 𝑠 𝑠
{
1
𝑠
; if 𝑠 > 0
=
∞; otherwise
1
Therefore, ℒ{1} = , if 𝑠 > 0.
𝑠
⃝January
2.5 The Laplace Transform Method to Solve ODEs 59
Table of some basic Laplace Transforms
Function (𝑓 (𝑡)) Laplace Trnsf. (𝐹 (𝑠))

1
1 𝑠
,𝑠 >0
𝑛 𝑛!
𝑡 ,𝑛 ∈ N ,𝑠 > 0
𝑠𝑛+1
1
𝑒𝑡𝑘 𝑠−𝑘
,𝑠 > 𝑘
𝑛!
𝑡𝑛 𝑒𝑘𝑡 (𝑠−𝑘)𝑛+1
,𝑠 > 𝑘
𝑘
sin(𝑘𝑡) 𝑠2 +𝑘2
,𝑠 > 0
𝑠
cos(𝑘𝑡) 𝑠2 +𝑘2
,𝑠 > 0
𝑘
sinh(𝑘𝑡) 𝑠2 −𝑘2
, 𝑠 > ∣𝑘∣
𝑠
cosh(𝑘𝑡) 𝑠2 −𝑘2
, 𝑠 > ∣𝑘∣
From the table above we have

1
ℒ{𝑒𝑎𝑡 } = for 𝑠 > 𝑎.
𝑠−𝑎
1
Thus the inverse operator applied on 𝑠−𝑎
will give us back the function 𝑒𝑎𝑡
1
i.e., ℒ−1 { } = 𝑒𝑎𝑡 for 𝑠 > 𝑎.
𝑠−𝑎
In general, ℒ−1 , the inverse Laplace Operator, is given by
∫ 𝛾+𝑖∞
−1 1
ℒ {𝐹 (𝑠)} = 𝐹 (𝑠)𝑒𝑠𝑡 𝑑𝑠,
2𝜋𝑖 𝛾−𝑖∞
(where 𝛾 is a positive real number), which is a complex improper integral.
Properties
Here below we state some important properties of the transform in a serious of theorems
without proof.
Theorem 2.5.2 (Linearity).
(a) If 𝑢(𝑡), 𝑣(𝑡) are functions and 𝛼, 𝛽 are any constants, then
ℒ{𝛼𝑢(𝑡) + 𝛽𝑣(𝑡)} = 𝛼ℒ{𝑢(𝑡)} + 𝛽ℒ{𝑣(𝑡)}.
(b) For any functions 𝑈 (𝑠), 𝑉 (𝑠) and any given scalars 𝛼, 𝛽, we have
ℒ−1 {𝛼𝑈 (𝑠) + 𝛽𝑉 (𝑠)} = 𝛼ℒ−1 {𝑈 (𝑠)} + 𝛽ℒ−1 {𝑉 (𝑠)}.
⃝January
Example 2.5.2. Evaluate the following transforms
1. ℒ{3𝑡 + 5𝑒−2𝑡 }.
2. ℒ{cos2 3𝑡}.
{ 2 }
−1 𝑠
3. ℒ (𝑠+1)3
.
Solutions
1. ℒ{3𝑡 + 5𝑒−2𝑡 }; Applying the liniarity property we get,
ℒ{3𝑡 + 5𝑒−2𝑡 } = 3ℒ{𝑡} + 5ℒ{𝑒−2𝑡 }

( ) ( )
1 1
= 3 2 +5
𝑠 𝑠+2
3 5 5𝑠2 + 3𝑠 + 6
= 2+ =
𝑠 𝑠+2 𝑠2 (𝑠 + 2)
2. ℒ{cos2 3𝑡}; Using half angle formula we can get
1 + cos 6𝑡 1 1
ℒ{cos2 3𝑡} = ℒ{ } = ℒ{ + cos 6𝑡}.
2 2 2
1 1
Then by linearity we have ℒ{cos2 3𝑡} = ℒ{1} + ℒ{cos 6𝑡}.
2 2
Then we are now able to read the transforms of 1 and cos 6𝑡 from the table and
get,
𝑠2 + 18
( ) ( )
2 1 1 1 𝑠
ℒ{cos 3𝑡} = + = .
2 𝑠 2 𝑠2 + 6 2 𝑠(𝑠2 + 36)
3. Using partial fractions we have
𝑠2 𝐴 𝐵 𝐶
3
= + 2
+
(𝑠 + 1) 𝑠 + 1 (𝑠 + 1) (𝑠 + 1)3
⇒ 𝑠2 = 𝐴(𝑠 + 1)2 + 𝐵(𝑠 + 1) + 𝐶 = 𝐴𝑠2 + (2𝐴 + 𝐵)𝑠 + (𝐴 + 𝐵 + 𝐶).
⇒ 𝐴 = 1, 𝐵 = −2, 𝐶 = 1.
Hence we can rewrite the inverse transform and apply linearity to get
𝑠2
{ } { }
−1 −1 1 2 1
ℒ = ℒ − +
(𝑠 + 1)3 𝑠 + 1 (𝑠 + 1)2 (𝑠 + 1)3
{ } { } { }
−1 1 −1 −2 −1 1
= ℒ +ℒ +ℒ
𝑠+1 (𝑠 + 1)2 (𝑠 + 1)3
1
= 𝑒−𝑡 − 2𝑡𝑒−𝑡 + 𝑡2 𝑒−𝑡
2
1 2 −𝑡
= (1 − 2𝑡 + 𝑡 )𝑒 .
2
⃝January
2.5 The Laplace Transform Method to Solve ODEs 61
The other important property that leads us to use the Laplace transform in solving ordinary
differential equation is how the transform performs on the derivative.
Theorem 2.5.3 (Transform of the derivative). Let 𝑓 (𝑡) be continuous and 𝑓 ′ (𝑡) be
piecewise continuous on some interval [0, 𝑡𝑜 ] for every finite 𝑡𝑜 , and let ∣𝑓 (𝑡)∣ < 𝐾𝑒𝑐𝑡 for
some constants 𝐾, 𝑇 , and 𝑐 and for all 𝑡 > 𝑇 . Then the transform ℒ{𝑓 ′ (𝑡)} exists for all
𝑠 > 𝑐 and
ℒ{𝑓 ′ (𝑡)} = 𝑠ℒ{𝑓 (𝑡)} − 𝑓 (0).
Example 2.5.3. Use the Laplace transform method to solve the initial-value problem.
𝑦 ′ + 2𝑦 = 0 with 𝑦(0) = 1.
Solution
Applying Laplace transform on both sides of the equation we have

ℒ{𝑦 ′ + 2𝑦} = ℒ{0} ⇒ ℒ{𝑦 ′ (𝑡)} + 2ℒ{𝑦(𝑡)}.
Now, letting ℒ{𝑦(𝑡)} := 𝑌 (𝑠), we get the algebraic equation,
1
𝑠𝑌 (𝑠) − 𝑦(0) + 2𝑌 (𝑠) = 0 ⇒ 𝑌 (𝑠) = .
𝑠+2
Therefore, reading from the transform table we get
1
𝑦(𝑡) = ℒ−1 {𝑌 (𝑠)} = ℒ−1 { } = 𝑒−2𝑡 .
𝑠+2
i.e., 𝑦(𝑡) = 𝑒−2𝑡 is the solution for the differential equation. ⊡
We can also use the Laplace method to solve higher order equations with constant coef-
ficients.
The following property of the transform, which is the continuouation of the above theorem,
is required.
Theorem 2.5.4. Let 𝑓 (𝑡) be continuous and 𝑓 (𝑛) (𝑡) be piecewise continuous on some
interval [0, 𝑡𝑜 ] for every finite 𝑡𝑜 , and let ∣𝑓 (𝑡)∣ < 𝐾𝑒𝑐𝑡 for some constants 𝐾, 𝑇 , and 𝑐
and for all 𝑡 > 𝑇 . Then we have
ℒ{𝑓 (𝑛) (𝑡)} = 𝑠𝑛 ℒ{𝑓 (𝑡)} − 𝑠𝑛−1 𝑓 (0) − 𝑠𝑛−2 𝑓 ′ (0) − ⋅ ⋅ ⋅ − 𝑓 (𝑛−1) (0).
Theorem 2.5.5 (First shifting theorem). If ℒ{𝑓 (𝑡)} = ℱ(𝑠) for 𝑅𝑒(𝑠) > 𝑏, then
ℒ{𝑒𝑎𝑡 𝑓 (𝑡)} = ℱ(𝑠 − 𝑎) for 𝑅𝑒(𝑠) > 𝑎 + 𝑏.
The proof of this theorem is easy to see using the definition.

Example 2.5.4. Find the Laplace transform for the function 𝑓 (𝑡) = 𝑒3𝑡 cos 4𝑡.
⃝January
Solution
𝑠
Recall that ℒ{cos 4𝑡} = .
𝑠2
+ 42
Then using the first shifting theorem we get
𝑠−3
ℒ{𝑒3𝑡 cos 4𝑡} = .
(𝑠 − 3)2 + 42
𝑠
Example 2.5.5. Find the inverse Laplace transform for the function ℱ(𝑠) = .
𝑠2 +𝑠+1
Solution
First let us rewrite the function ℱ(𝑠) as
𝑠 𝑠 𝑠 + 21 1
2
ℱ(𝑠) = = 1 2 3 = −
2
𝑠 +𝑠+1 (𝑠 + 2 ) + 4
(𝑠 + 12 )2 + 3
4
(𝑠 + 1 2
2
) + 3
4
and hence,
{ √ }
3
𝑠 + 21
{ } { }
𝑠 1
ℒ−1 = ℒ−1 − ℒ−1 √ 2
.
𝑠2 + 𝑠 + 1 (𝑠 + 12 )2 + 3
4 3 (𝑠 + 21 )2 + 3
4
Then using the first shifting theorem, we have,

{ } √ √
−1 𝑠 −𝑡/2 3𝑡 1 −𝑡/2 3𝑡
ℒ 2
=𝑒 cos −√ 𝑒 sin .
𝑠 +𝑠+1 2 3 2
Consider the general Laplace transform formula
∫ ∞
ℱ(𝑠) = 𝑒−𝑠𝑡 𝑓 (𝑡)𝑑𝑡.
0
Taking the derivative with respect to 𝑠 on both sides we get,
∫ ∞
′
ℱ (𝑠) = (−𝑡)𝑒−𝑠𝑡 𝑓 (𝑡)𝑑𝑡 = ℒ{−𝑡𝑓 (𝑡)}.
0
By further differentiating the above equation with respect to 𝑠, we get
ℱ ′′ (𝑠) = ℒ{𝑡2 𝑓 (𝑡)}.
In general we have
⃝January
2.6 The Cauchy-Euler Equation 63
Theorem 2.5.6 (Derivative of the transform). For a piecewise continuous function 𝑓 (𝑡)
and for any positive integer 𝑛, it holds that
ℒ{(−1)𝑛 𝑡𝑛 𝑓 (𝑡)} = ℱ (𝑛) (𝑠).
The formula in this theorem can be used to find transforms of functions of the form
𝑥𝑛 𝑓 (𝑥) when the Laplace transform of 𝑓 (𝑡) is known.
Exercise 2.5.7. Use the Laplace transform method to solve
𝑥𝑦 ′′ + (2𝑥 + 3)𝑦 ′ + (𝑥 + 3)𝑦 = 3𝑒−𝑥 ; 𝑦(0) = 0, 𝑦 ′ (0) = 1.
Remark 2.5.8. The main idea in using Laplace transform in solving ODEs is that, it
transforms the differential equation into an algebraic equation. Once the transformation
is completed, we seek for a solution to ℒ{𝑦(𝑡)} algebraically. Then the final step will be
to get back the value of 𝑦(𝑡) using the inverse Laplace transform.
2.6 The Cauchy-Euler Equation
In this section we are going to consider linear differential equations where the coefficients
are variables with some special forms.
Definition 2.6.1. The linear differential equation with variable coefficient of the form:
𝑎𝑛 𝑥𝑛 𝑦 (𝑛) + 𝑎𝑛−1 𝑥𝑛−1 𝑦 (𝑛−1) + ⋅ ⋅ ⋅ + 𝑎1 𝑥𝑦 ′ + 𝑎0 𝑦 = 𝐹 (𝑥) (2.15)
where 𝑎0 , 𝑎1 , . . . , 𝑎𝑛 are constants is called the Cauchy-Euler Equation.
Example 2.6.1. The linear differential equation 3𝑥2 𝑦 ′′ − 11𝑥𝑦 ′ + 2𝑦 = sin 𝑥 is a Cauchy-
Euler equation.
To solve Cauchy-Euler DEs first we reduce the given DE into a linear differential equation
with constant coefficients and solve the given equation with the methods derived in the
previous sections.
Theorem 2.6.2. The transformation 𝑥 = 𝑒𝑡 , 𝑡 ∈ R reduces the Cauchy-Euler DE to a

linear DE with constant coefficients.
⃝January
Let us consider the case when 𝑛 = 2. In this case the equation is:
𝑎2 𝑥2 𝑦 ′′ + 𝑎1 𝑥𝑦 ′ + 𝑎0 𝑦 = 𝐹 (𝑥) (2.16)
Let 𝑥 = 𝑒𝑡 . Then by solving for 𝑡 we get 𝑡 = ln 𝑥 for 𝑥 > 0 (or 𝑥 = −𝑒𝑡 if 𝑥 < 0) and
𝑑𝑦 𝑑𝑦 𝑑𝑡 1 𝑑𝑦
= . =
𝑑𝑥 𝑑𝑡 𝑑𝑥 𝑥 𝑑𝑡
and
𝑑2 𝑦 1 𝑑2 𝑦 𝑑𝑡 1 𝑑2 𝑦 𝑑𝑦
( ) ( ) ( )
1 𝑑 𝑑𝑦 𝑑𝑦 𝑑 1 1 𝑑𝑦
= + . = . − = 2 − .
𝑑𝑥2 𝑥 𝑑𝑥 𝑑𝑡 𝑑𝑡 𝑑𝑥 𝑥 𝑥 𝑑𝑡2 𝑑𝑥 𝑥2 𝑑𝑡 𝑥 𝑑𝑡2 𝑑𝑡
Substituting into (2.15) we get:
1 𝑑2 𝑦 𝑑𝑦
( )
2 1 𝑑𝑦
𝑎2 𝑥 2 2
− + 𝑎1 𝑥. + 𝑎2 𝑦 = 𝐹 (𝑒𝑡 ).
𝑥 𝑑𝑡 𝑑𝑡 𝑥 𝑑𝑡
This implies,
𝑑2 𝑦 𝑑𝑦
𝑎2 2
+ (𝑎1 − 𝑎2 ) + 𝑎0 𝑦 = 𝐹 (𝑒𝑡 ).
𝑑𝑡 𝑑𝑡
Then
𝑑2 𝑦 𝑑𝑦
2
𝐴2
+ 𝐴1 + 𝐴0 𝑦 = 𝐺(𝑡), (2.17)
𝑑𝑡 𝑑𝑡
where 𝐴2 = 𝑎2 , 𝐴1 = 𝑎1 − 𝑎2 , 𝐴0 = 𝑎0 and 𝐹 (𝑒𝑡 ) = 𝐺(𝑡), which is a second order linear
differential equation with constant coefficients.
Example 2.6.2. Solve each of the following DEs.
1. 𝑥2 𝑦 ′′ − 2𝑥𝑦 ′ + 2𝑦 = 0.
2. 3𝑥2 𝑦 ′′ − 11𝑥𝑦 ′ + 2𝑦 = sin 𝑥.
3. 𝑥2 𝑦 ′′ − 2𝑥𝑦 ′ + 2𝑦 = 𝑥3 .
Solution
1. Let 𝑥 = 𝑒𝑡 . Since 𝑎2 = 1, 𝑎1 = −2 and 𝑎0 = 2 we have 𝐴2 = 𝑎2 = 1, 𝐴1 = 𝑎1 −𝑎2 =

−3 and 𝐴0 = 𝑎0 = 2 which reduces the given equation to 𝑦 ′′ − 3𝑦 ′ + 2𝑦 = 0 which
is a homogenous second order linear differential equation with constant coefficients.
⃝January
2.6 The Cauchy-Euler Equation 65
Then the characteristic equation of the equation is 𝜆2 − 3𝜆 + 2 = 0 which has
eigenvalues 𝜆1 = 1, 𝜆2 = 2.
Therefore, the differential equation 𝑦 ′′ − 3𝑦 ′ + 2𝑦 = 0 has a general solution 𝑦(𝑡) =

𝑐1 𝑒𝑡 + 𝑐2 𝑒2𝑡 and since 𝑥 = 𝑒𝑡 the DE 𝑥2 𝑦 ′′ − 2𝑥𝑦 ′ + 2𝑦 = 0 has a general solution
𝑦(𝑥) = 𝑐1 𝑥 + 𝑐2 𝑥2 ,
where 𝑐1 and 𝑐2 are arbitrary constants.
2. Let 𝑥 = 𝑒𝑡 . Since 𝑎2 = 3, 𝑎1 = −11, 𝑎0 = 2, we have 𝐴2 = 𝑎2 = 3, 𝐴1 =

𝑎1 − 𝑎2 = −11 − 3 = −14, and 𝐴0 = 𝑎0 = 2, which reduce the given equation to
3𝑦 ′′ − 14𝑦 ′ + 2𝑦 = sin(𝑒𝑡 ) which is a DE with constant coefficients.
3. The given equation is transformed into 𝑦 ′′ − 3𝑦 ′ + 2𝑦 = 𝑒3𝑡 , with the substitution

𝑥 = 𝑒𝑡 , which is a DE with constant coefficients.
Example 2.6.3 (Application). Consider a mechanical oscillator.
a Figure here ... ...
Let 𝐹𝐺 = 𝑚𝑔, where 𝑚 is mass of the object on the spring and 𝑔 is the gravity and
∣𝐹ˆ𝑅 ∣ = 𝑘𝑥, (Hook’s law) where 𝑘 is the spring stiffness constant and 𝑥 is the the distance
moved by the mass 𝑚.

𝑑𝑥
If 𝐹𝐷 is a damping force for small velocity of mass, then ∣𝐹𝐷 ∣ = 𝐶 , where 𝐶 > 0 is
𝑑𝑡
called a damping constant.
Therefore, the final form of our governing equation of motion is of the form:
𝑚𝑥′′ + 𝐶𝑥′ + 𝑘𝑥 = 𝑚𝑔 = 𝐹 (𝑡).
If 𝐹 is a variable force then the problem will be a second order nonhomogeneous differential
equation, and can be solved using one of the previously discussed methods.
⃝January
2.7 *The Power Series Solution Method
Linear differential equations of the second and higher orders with variables coefficients are
usually difficult to solve, except for special cases (see the Cauchy-Euler Equation). In such
cases we give up on the prospect of finding solutions in closed form and seek solutions in
the form of finite series.
The method rests on the idea that any polynomial is differentiable to any order and finding
the integrals of polynomial functions are usually simple. Hence if we can represent all the
functions in the differential equation by polynomials we can readily find their integrals
and hence solve the differential equation. We will employ this fact in this section to solve
linear differential equations. Because of the difficulties in variable coefficients cases, we
will consider only the second order case here. Some functions can be represented by an
infinite sum of polynomials, which is called a power series. However, Not every functions
are representable in such forms.
You may recall that an infinite series of the form
∞
∑
𝑎𝑛 (𝑥 − 𝑥𝑜 )𝑛 = 𝑎0 + 𝑎1 (𝑥 − 𝑥𝑜 ) + 𝑎2 (𝑥 − 𝑥𝑜 )2 + ⋅ ⋅ ⋅ (2.18)
𝑛=0
is called a power series in (𝑥 − 𝑥𝑜 ) and it is said to converge at a point 𝑥 if the limit

𝑚
∑
lim 𝑎𝑛 (𝑥 − 𝑥𝑜 )𝑛
𝑚→∞
𝑛=0
exists and is a real number. Clearly the power series (2.18) always converges at 𝑥 = 𝑥𝑜 ,
which is the trivial case. If (2.18) converges on the interval (𝑥𝑜 − 𝑟, 𝑥𝑜 + 𝑟) ⊂ R for some
𝑟 > 0, then we say 𝑟 is a radius of convergence of the series. If the sum of the series is
∞ except at 𝑥 = 𝑥𝑜 , it is conventional to write 𝑟 = 0 as its radius of convergence, and if
the series converges for all 𝑥 ∈ R then we take 𝑟 = ∞.
If the power series (2.18) converges with radius of convergence 𝑟 > 0, the function
∞
∑
𝑓 (𝑥) = 𝑎𝑛 (𝑥 − 𝑥𝑜 )𝑛
𝑛=0
can be integrated and differentiated pointwise and hence it a derivative of all orders for
all 𝑥 such that ∣𝑥 − 𝑥𝑜 ∣ < 𝑟. Then from successive differentiation we can get that
𝑓 (𝑛) (𝑥𝑜 )
𝑎𝑛 = , for all 𝑛 = 0, 1, 2, . . .
𝑛!
⃝January
2.7 *The Power Series Solution Method 67
On the other hand, if 𝑓 (𝑥) is a function which has a derivative of all order for all 𝑥 such
that ∣𝑥 − 𝑥0 ∣ < 𝑟, with 𝑟 > 0, can 𝑓 (𝑥) be represented by a power series near 𝑥𝑜 ? The
answer is not always in the affirmative as there are some functions with no power series
expansions even though it could be infinitely differentiable. If it does, then we call such a
series a Taylor series expansion of 𝑓 at 𝑥𝑜 .
Recall that, the Taylor series of a given smooth function 𝑓 about a point 𝑥𝑜 is:
∞
∑ 𝑓 (𝑛) (𝑥𝑜 )
𝑇 𝑆 𝑓 ∣𝑥𝑜 = (𝑥 − 𝑥𝑜 )𝑛 ;
𝑛=0
𝑛!
If this series converges in some interval ∣𝑥 − 𝑥𝑜 ∣ < 𝑟, and is equal to 𝑓 , then we call 𝑓 is
analytic at 𝑥𝑜 and 𝑟 is the radius of convergence.
If 𝑓 is not analytic at 𝑥𝑜 , we call it is singular at 𝑥𝑜 .
2.7.1 Power series solution method
The method mainly uses the following theorem
Theorem 2.7.1 (Power Series Solution). If the functions 𝑝 and 𝑞 are analytic at a point
𝑐𝑜 , then every solution of the DE
𝑦 ′′ + 𝑝(𝑥)𝑦 ′ + 𝑞(𝑥)𝑦 = 0 (2.19)
is also analytic at 𝑐𝑜 , and can be found in the form

∞
∑
𝑦(𝑥) = 𝑎𝑛 (𝑥 − 𝑐𝑜 )𝑛 .
𝑛=0
Moreover, the radius of convergence of every solution is at least as large as the smaller of
the radii of convergence of 𝑇 𝑆 𝑝∣𝑐𝑜 and 𝑇 𝑆 𝑞∣𝑐𝑜 .
(𝑥 − 1)𝑦 ′′ + 𝑦 ′ + 2(𝑥 − 1)𝑦 = 0 (2.20)
on the interval [4, ∞) with initial conditions 𝑦(4) = 5 and 𝑦 ′ (4) = 0.
⃝January
Solution
The solution procedure in this approach is:
∙ Convert the problem to the form of equation (2.19) in the above Theorem
∙ Check analyticity of the coefficient functions 𝑝(𝑥) and 𝑞(𝑥) at the point 𝑥𝑜 (the
given initial point).
∙ Substitute into the equation (2.20) the general solution

∞
∑
𝑦(𝑥) = 𝑎𝑛 (𝑥 − 𝑥𝑜 )𝑛
𝑛=0
and determine the values of the coefficients 𝑎𝑛 , for each 𝑛 = 0, 1, 2, . . ..
1
Thus, when converted the problem takes the form, 𝑦 ′′ + 𝑦 ′ +2𝑦 = 0. The coefficient
𝑥−1
1
functions are 𝑝(𝑥) = and 𝑞(𝑥) = 2. Both of them are analytic except at 𝑥 = 1, and
𝑥−1
in particular they are analytic in the region sufficiently close to 𝑥 = 4, the initial point.
So, by the above theorem we seek a solution of the form:
∞
∑
𝑦(𝑥) = 𝑎𝑛 (𝑥 − 4)𝑛 ,
𝑛=0
where 𝑎𝑛 is to be determined for each natural number 𝑛. Now to determine the coeffi-
cients, substitute this 𝑦(𝑥) into the equation (2.20), with 𝑥 − 1 = 3 + (𝑥 − 4). Thus,
[3 + (𝑥 − 4)]𝑦 ′′ + 𝑦 ′ + 2[3 + (𝑥 − 4)]𝑦 = 0

∑∞ ∑∞
𝑛−2
⇒ [3 + (𝑥 − 4)] 𝑛(𝑛 − 1)𝑎𝑛 (𝑥 − 4) + 𝑛𝑎𝑛 (𝑥 − 4)𝑛−1
𝑛=2 𝑛=1
∞
∑
+2[3 + (𝑥 − 4)] 𝑎𝑛 (𝑥 − 4)𝑛 = 0
𝑛=0
∞
∑ ∞
∑
𝑛−2
⇒ 3𝑛(𝑛 − 1)𝑎𝑛 (𝑥 − 4) + 𝑛(𝑛 − 1)𝑎𝑛 (𝑥 − 4)𝑛−1 +
𝑛=2 𝑛=2
∞
∑ ∞
∑ ∞
∑
𝑛−1 𝑛
𝑛𝑎𝑛 (𝑥 − 4) + 6𝑎𝑛 (𝑥 − 4) + 2𝑎𝑛 (𝑥 − 4)𝑛+1 = 0
𝑛=1 𝑛=0 𝑛=0
Now to equate the two values in the last equation, we can treat 0 as an infinite polynomial
in (𝑥 − 4) with 0 coefficients corresponding to each degree of (𝑥 − 4). Hence we must use
⃝January
change of variables for the indices of the summations so that the values in the left-hand-
side are summable. That means, we should bring all the summands in the left-hand-side
to the same powers of (𝑥 − 4) under each summation in the last equation. Thus it follows
from last equation that,
∞
∑ ∞
∑
⇒ 3(𝑚 + 2)(𝑚 + 1)𝑎𝑚+2 (𝑥 − 4)𝑚 + (𝑚 + 1)𝑚𝑎𝑚+1 (𝑥 − 4)𝑚 +
𝑚=0 𝑚=0
∞
∑ ∞
∑ ∞
∑
𝑚 𝑚
(𝑚 + 1)𝑎𝑚+1 (𝑥 − 4) + 6𝑎𝑚 (𝑥 − 4) + 2𝑎𝑚−1 (𝑥 − 4)𝑚 = 0,
𝑚=0 𝑚=0 𝑚=0
with the assumption that 𝑎−1 = 0. Thus we can now bring the left-hand-side under a
single summation to get,
∞
∑
3(𝑚 + 2)(𝑚 + 1)𝑎𝑚+2 + (𝑚 + 1)2 𝑎𝑚+1 + 6𝑎𝑚 + 2𝑎𝑚−1 (𝑥 − 4)𝑚 = 0
[ ]
𝑚=0
which implies that 3(𝑚 + 2)(𝑚 + 1)𝑎𝑚+2 + (𝑚 + 1)2 𝑎𝑚+1 + 6𝑎𝑚 + 2𝑎𝑚−1 = 0 for all
𝑚 = 0, 1, . . ..
Hence solving for 𝑎𝑚+2 we get,
𝑚+1 2 2
𝑎𝑚+2 = − 𝑎𝑚+1 − 𝑎𝑚 − 𝑎𝑚−1 , and 𝑎−1 = 0,
3(𝑚 + 2) (𝑚 + 2)(𝑚 + 1) 3(𝑚 + 2)(𝑚 + 1)
for 𝑚 = 0, 1, 2, . . .
Now to determine each of the unknown coefficients 𝑎𝑚 we substitute each 𝑚 = 0, 1, 2, . . .,
in the last equation:
1 1 1
𝑚 = 0 : 𝑎2 = − 𝑎1 − 𝑎0 − 𝑎−1 = − 𝑎1 − 𝑎0
6 3 6 ( )
2 1 1 2 1 1 1
𝑚 = 1 : 𝑎3 = − 𝑎2 − 𝑎1 − 𝑎0 = − − 𝑎1 − 𝑎0 − 𝑎1 − 𝑎0
9 3 9 9 6 3 9
8 1
= − 𝑎1 + 𝑎0
27 9
1 1 1
𝑚 = 2 : 𝑎4 = − 𝑎3 − 𝑎2 − 𝑎1
4( 6 8 ) ( )
1 8 1 1 1 1
=− − 𝑎1 + 𝑎0 − − 𝑎1 − 𝑎0 − 𝑎1
4 27 9 6 6 8
5 5
= 𝑎1 + 𝑎0
108 36
.. .. ..
. . .
.. .. ..
. . .
⃝January
for arbitrary constants 𝑎1 and 𝑎0 . Therefore, the solution is given by,
( ) ( )
1 8 1
𝑦(𝑥) = 𝑎0 + 𝑎1 (𝑥 − 4) + − 𝑎1 − 𝑎0 (𝑥 − 4) + − 𝑎1 + 𝑎0 (𝑥 − 4)3 +
2
6 27 9
( )
5 5
+ 𝑎1 + 𝑎0 (𝑥 − 4)4 + ⋅ ⋅ ⋅
108 36
⎡ ⎤
⎢ 2 1 3 5 4
⎥
= 𝑎0 ⎢1 − (𝑥 − 4) + (𝑥 − 4) + (𝑥 − 4) + ⋅ ⋅ ⋅ ⎥+
⎣
| 9 {z 36 }
⎦
=𝑦1 (𝑥)
⎡ ⎤
⎢ 1 2 8 3 5 4
⎥
+𝑎1 ⎢ (𝑥 − 4) − (𝑥 − 4) − (𝑥 − 4) + (𝑥 − 4) ⋅ ⋅ ⋅ ⎥
⎣
| 6 27 {z 108 }
⎦
=𝑦2 (𝑥)
+ 𝑎0 𝑦1 (𝑥) + 𝑎1 𝑦2 (𝑥).
Now to determine the constants 𝑎0 𝑎1 we use the initial conditions 𝑦(4) = 5 and 𝑦 ′ (4) = 0
in the last equation, to get 𝑦(4) = 𝑎0 = 5 and 𝑦 ′ (4) = 𝑎1 = 0.
Therefore,
[ ]
12 3 5 4
𝑦(𝑥) = 5 1 − (𝑥 − 4) + (𝑥 − 4) + (𝑥 − 4) + ⋅ ⋅ ⋅ .
9 36
□
The limitations in this method are
1. in the solution, one can not calculate all the infinite coefficients of the series. How-
ever, it is possible to approximate the function to any required order using this
approach.
2. in the method, the coefficient functions of the differential equation are required to
be analytic near the initial point. This assumption can be relaxed in the following
subsection
2.7.2 Frobenius Method
Consider again a second order equation with variable coefficients,
ℎ(𝑥)𝑦 ′′ + 𝑝(𝑥)𝑦 ′ + 𝑞(𝑥)𝑦 = 0 (2.21)
⃝January
If ℎ(𝑥) ∕= 0 for some 𝑥 we can equivalently have
𝑝(𝑥) ′ 𝑞(𝑥)
𝑦 ′′ + 𝑦 + 𝑦 = 0, for ℎ(𝑥) ∕= 0. (2.22)
ℎ(𝑥) ℎ(𝑥)
If ℎ(𝑥) ∕= 0 for all 𝑥, applying simply the power series solution method gives us the
required result. But if ℎ(𝑥) = 0 for some 𝑥 the resulting equation will be different from
the original one at those points 𝑥 where ℎ(𝑥) = 0.
Definition 2.7.2.
1. A point 𝑥𝑜 is said to be an ordinary point of equation (2.21) if ℎ(𝑥𝑜 ) ∕= 0 and

𝑝(𝑥) 𝑞(𝑥)
,
ℎ(𝑥) ℎ(𝑥)
are analytic at 𝑥𝑜 . Otherwise, it is called a singular point of equation
(2.21).
2. A singular point 𝑥𝑜 is said to be a regular singular point of equation (2.21) if

the functions
𝑝(𝑥) 𝑞(𝑥)
and (𝑥 − 𝑥𝑜 )2
(𝑥 − 𝑥𝑜 )
ℎ(𝑥) ℎ(𝑥)
are analytic at 𝑥𝑜 . A non regular singular point is called an irregular singular
point of equation (2.21).
If equation (2.21) has a regular singular point at 𝑥𝑜 , then we use the power series:
∞
∑
𝑦(𝑥) = 𝑎𝑛 (𝑥 − 𝑥𝑜 )𝑛+𝑟
𝑛=0
as our initial guess of the solution function and determine the values of coefficients 𝑎𝑛 , 𝑛 =
0, 1, 2, . . . and the power 𝑟.
This last series is called a Frobenius series. Here we assume that 𝑎0 ∕= 0. Otherwise,
we can factor 𝑥 − 𝑥𝑜 ) and add the powers in to 𝑟.
Then differentiating and substituting this series for 𝑦(𝑥) in to the modified equation (2.22),
we will get,
∞
[ ∞
] ∞
∑ ∑ ∑
𝑚+𝑟−2 𝑚
𝑎𝑚 (𝑚 + 𝑟)(𝑚 + 𝑟 − 1)(𝑥 − 𝑥𝑜 ) + 𝑝𝑚 (𝑥 − 𝑥𝑜 ) 𝑎𝑚 (𝑚 + 𝑟)(𝑥 − 𝑥𝑜 )𝑚+𝑟−1
𝑚=0 𝑚=0 𝑚=0
[ ∞
] ∞
∑ ∑
+ 𝑞𝑚 (𝑥 − 𝑥𝑜 )𝑚 𝑎𝑚 (𝑥 − 𝑥𝑜 )𝑚+𝑟 = 0,
𝑚=0 𝑚=0
where ∞ ∞
𝑝(𝑥) ∑ 𝑞(𝑥) ∑
= 𝑝𝑚 (𝑥 − 𝑥𝑜 )𝑚 and = 𝑞𝑚 (𝑥 − 𝑥𝑜 )𝑚
ℎ(𝑥) 𝑚=0 ℎ(𝑥) 𝑚=0
⃝January
are the power series expansions, as they are analytic at 𝑥𝑜 .
Then combining the corresponding powers of (𝑥 − 𝑥𝑜 ) and equating the coefficients of
each powers of (𝑥 − 𝑥𝑜 ) to zero we will arrive at a system of equations in the variables
𝑎𝑚 and 𝑟. Taking only the coefficient for the constant part we have,
𝑎0 [𝑟(𝑟 − 1) + 𝑝0 𝑟 + 𝑞0 ] = 0 (2.23)
But since 𝑎0 ∕= 0 was assumed we have
𝑟(𝑟 − 1) + 𝑝0 𝑟 + 𝑞0 = 0, (2.24)
which is known as the indicial equation of the differential equation (2.21). Its roots, say
𝑟1 and 𝑟2 , define the two linearly independent solutions 𝑦1 (𝑥) and 𝑦2 (𝑥) of the differential
equation. Using these values of the roots for (2.24), we can determine the remaining
coefficients 𝑎𝑚 , 𝑚 = 0, 1, 2, 3, . . ..
This procedure is called the Frobenius Method.
Example 2.7.2. Use Frobenius method to solve
2𝑥2 𝑦 ′′ + 𝑥(2𝑥 + 1)𝑦 ′ − 𝑦 = 0
Solution
By dividing both sides by 2𝑥2 the problem can be rewritten as

1/2 + 𝑥 ′ −1/2
𝑦 ′′ + 𝑦 + 𝑦 = 0. (2.25)
𝑥 𝑥2
Then the coefficient functions satisfy the conditions for regular singularity of the point
𝑥 = 0; that is, 𝑥 × 1/2+𝑥
𝑥
= 1
2
+ 𝑥 and 𝑥2 × −1/2
𝑥2
= − 21 are both analytic. Hence we take
∞
∑
𝑦(𝑥) = 𝑎𝑛 𝑥𝑛+𝑟
𝑛=0
as our initial guess. Then substituting this expression in to the original equation we get
∞
∑ ∞
∑ ∞
∑
𝑛+𝑟 𝑛+𝑟+1
2(𝑛 + 𝑟)(𝑛 + 𝑟 − 1)𝑎𝑛 𝑥 + 2(𝑛 + 𝑟)𝑎𝑛 𝑥 + (𝑛 + 𝑟)𝑎𝑛 𝑥𝑛+𝑟
𝑛=0 𝑛=0 𝑛=0
∞
∑
− 𝑎𝑛 𝑥𝑛+𝑟 = 0.
𝑛=0
⃝January
After shifting the indices in the second summation, so that we have the same powers of
𝑥 in all summations, and assuming that 𝑎−2 = 𝑎−1 = 0, we get
[2𝑟(𝑟 − 1)𝑎0 + 𝑟𝑎0 − 𝑎0 ]𝑥𝑟

∑∞
+ [2(𝑚 + 𝑟)(𝑚 + 𝑟 − 1)𝑎𝑚 + 2(𝑚 + 𝑟 − 1)𝑎𝑚−1 + (𝑚 + 𝑟)𝑎𝑚 − 𝑎𝑚 ] 𝑥𝑚+𝑟 = 0
𝑚=1
∞
∑
𝑟
⇒ 𝑎0 [2𝑟(𝑟 − 1) + 𝑟 − 1]𝑥 + [(𝑚 + 𝑟 − 1)(2(𝑚 + 𝑟) + 1)𝑎𝑚 + 2(𝑚 + 𝑟 − 1)𝑎𝑚−1 ] 𝑥𝑚+𝑟 = 0.
𝑚=1
This last equation is true if each of the coefficients of 𝑥𝑚+𝑟 is zero for each 𝑚 = 0, 1, 2, . . .
Then we arrive at the system,
𝑎0 [2𝑟(𝑟 − 1) + 𝑟 − 1] = 0
(𝑚 + 𝑟 − 1)(2(𝑚 + 𝑟) + 1)𝑎𝑚 + 2(𝑚 + 𝑟 − 1)𝑎𝑚−1 = 0.
Assuming 𝑎0 ∕= 0 we get the indicial equation of the differential equation,

1 1
𝑟(𝑟 − 1) + 𝑟 − = 0,
2 2
1
with roots 𝑟1 = 1 and 𝑟2 = − . Then from the second equation from the above system
2
we get the recurrence relation,
[ ]
2
𝑎𝑚 = − 𝑎𝑚−1 , for all 𝑚 = 1, 2, 3, . . .
2(𝑚 + 𝑟) + 1
.
Now we substitute the values of 𝑟 to get some values of the coefficients.
Case 1 𝑟 = 𝑟1 = 1. Then we have

2 2
𝑎1 = − 𝑎0 = − 𝑎0
2×2+1 5
2 22
𝑎2 = − 𝑎1 = 𝑎0
2 × (2 + 1) + 1 7×5
2 23
𝑎3 = − 𝑎2 = − 𝑎0
2 × (3 + 1) + 1 9×7×5
.. .. ..
. . .
.. .. ..
. . . (
2𝑚
)
𝑚
𝑎𝑚 = (−1) 𝑎0
5 × 7 × 9 × ⋅⋅⋅
⃝January
Thus one Frobenius solution is
2𝑚
[ ( ) ]
2 2 4 3 8 4 𝑚 𝑚+1
𝑦1 (𝑥) = 𝑎0 𝑥 − 𝑥 + 𝑥 − 𝑥 + ⋅ ⋅ ⋅ + (−1) 𝑥 + ⋅⋅⋅
5 35 315 5 × 7 × 9 × ⋅⋅⋅
Case 2 𝑟 = 𝑟2 = − 21 . Then using similar procedure we get the second Frobenius solution
to be,
[ ]
−1/2 1 2
𝑦2 (𝑥) = 𝑥 1 − 𝑥 + 𝑥 + ⋅⋅⋅ .
2
Therefore, the general solution of the differential equation (2.25) on the interval (0, ∞)
is
𝑦(𝑥) = 𝑐1 𝑦1 (𝑥) + 𝑐2 𝑦2 (𝑥).
Note that, since the constant parts of the numerators of the coefficients in equation (2.25)
1
are 𝑝0 = 2
and 𝑞0 = − 12 , respectively, the indicial equation becomes,
1 1
𝑟(𝑟 − 1) + 𝑟 − = 0,
2 2
as we have seen above.
Remark 2.7.3. The general solution of the differential equation using Frobenius Method
depend upon the roots to the indicial equation,
𝑟(𝑟 − 1) + 𝑝0 𝑟 + 𝑞0 = 0,
which forces the coefficient of 𝑥𝑟 to be zero. In our solution procedure, we assumed that
the roots 𝑟1 and 𝑟2 are distinct. This is always the case when the difference 𝑟1 − 𝑟2 is
not an integer. However, if the difference is an integer (including 0) there could be a
possibility that the second Frobenius series solution does not exist. In fact most of the
times, the general solution will be only one Frobenius solution. The details of this case is
beyond the scope of this text material and hence omitted.
⃝January
2.8 Systems of ODE of the First Order 75
2.8 Systems of ODE of the First Order
A system of 𝑛 linear first-order equations in the 𝑛 unknowns 𝑥1 (𝑡), 𝑥2 (𝑡), . . . , 𝑥𝑛 (𝑡) is a

system that can be written in the form:
𝑥′1 = 𝑎11 (𝑡)𝑥1 + 𝑎12 (𝑡)𝑥2 + ⋅ ⋅ ⋅ + 𝑎1𝑛 (𝑡)𝑥𝑛 + 𝑓1 (𝑡)

𝑥′2 = 𝑎21 (𝑡)𝑥1 + 𝑎22 (𝑡)𝑥2 + ⋅ ⋅ ⋅ + 𝑎2𝑛 (𝑡)𝑥𝑛 + 𝑓2 (𝑡)
.. (2.26)
.
𝑥′𝑛 = 𝑎𝑛1 (𝑡)𝑥1 + 𝑎𝑛2 (𝑡)𝑥2 + ⋅ ⋅ ⋅ + 𝑎𝑛𝑛 (𝑡)𝑥𝑛 + 𝑓𝑛 (𝑡),
which is called the normal form. In vector form this system becomes:
X′ = AX + F(𝑡),
where
X = (𝑥1 , 𝑥2 , . . . , 𝑥𝑛 )𝑇 , A = (𝑎𝑖𝑗 )𝑛×𝑛 , and F(𝑡) = (𝑓1 (𝑡), 𝑓2 (𝑡), . . . , 𝑓𝑛 (𝑡))𝑇
The system (2.26) is called homogeneous if F(𝑡) ≡ 0, so that X′ = AX and if F(𝑡) ∕≡ 0

for some 𝑡, the system is nonhomogeneous.
Definition 2.8.1. A solution vector of the system of differential equation in (2.26) over
some interval 𝐼 is a a vector (𝑥1 (𝑡), 𝑥2 (𝑡), . . . , 𝑥𝑛 (𝑡))𝑇 whose entries are differentiable
functions that satisfies the system in (2.26) on the interval 𝐼.
In this section, we are going to see how to solve such systems of equations. Two methods
are going to be considered, the Eigenvalue Method and Elimination Method.
2.8.1 Eigenvalue Method
A. Homogeneous Systems with Constant Coefficients
Consider the system

y′ = Ay, (2.27)
with A = (𝑎𝑖𝑗 )𝑛×𝑛 be a constant matrix, that is, all the entries of A are constants.
Recall that, in the scalar case if 𝑦 ′ = 𝑘𝑦, then 𝑦 = 𝑐𝑒𝑘𝑡 , where 𝑐 is a constant (by
integration).
⃝January
Let y = x𝑒𝜆𝑡 , where x = (𝑥1 , 𝑥2 , . . . , 𝑥𝑛 )𝑇 . Substituting this into (2.27) we get:
𝜆x𝑒𝜆𝑡 = y′ = Ay = Ax𝑒𝜆𝑡 and 𝑒𝜆𝑡 ∕= 0.
This implies,
Ax = 𝜆x,
which is an eigenvalue problem. Once we find the eigenvalues 𝜆𝑖 and a corresponding

eigenvector 𝑥𝑖 , the general solution will be
y(𝑡) = 𝑐1 𝑥1 𝑒𝜆1 𝑡 + ⋅ ⋅ ⋅ + 𝑐𝑛 𝑥𝑛 𝑒𝜆𝑛 𝑡 ,
where 𝑐1 , ..., 𝑐𝑛 are constants.
Example 2.8.1. Solve each of the following systems of linear differential equations.
⎧
{
′

 𝑦1′ = 2𝑦1 + 𝑦2 + 𝑦3
𝑦1 = −3𝑦1 + 𝑦2 ⎨
1. 2. 𝑦2′ = 𝑦1 + 2𝑦2 + 𝑦3
𝑦2′ = 𝑦1 − 3𝑦2 
⎩ 𝑦 ′ = 𝑦 + 𝑦 + 2𝑦

3 1 2 3
Solution:
1. The system can be written as:

( ) ( )( )
𝑦1′ −3 1 𝑦1
=
𝑦2′ 1 −3 𝑦2
Let y(𝑡) = x𝑒𝜆𝑡 . Then the corresponding eigenvalue problem will be:
( )( ) ( )
−3 1 𝑥1 𝑥1
Ax = 𝜆x ⇐⇒ =𝜆
1 −3 𝑥2 𝑥2
which is equivalent to:

( )( ) ( )
−3 − 𝜆 1 𝑥1 0
=
1 −3 − 𝜆 𝑥2 0
which has characteristic equation

−3 − 𝜆 1
𝐴 − 𝜆𝐼 = 0 ⇐⇒ = (3 + 𝜆)2 − 1 = 0.

1 −3 − 𝜆
This implies 𝜆2 + 6𝜆 + 8 = 0 ⇐⇒ (𝜆 + 2)(𝜆 + 4) = 0. Therefore, the eigenvalues

are 𝜆1 = −2 and 𝜆2 = −4
⃝January
a) Now, let us find an eigenvector corresponding to 𝜆1 = −2.
( )( ) ( )
−3 − (−2) 1 𝑥1 0
=
1 −3 − (−2) 𝑥2 0
or equivalently,
( )( ) ( )
−1 1 𝑥1 0
= ⇐⇒ 𝑥1 − 𝑥2 = 0,
1 −1 𝑥2 0
which implies 𝑥1 = 𝑥2 and hence we have (𝑥1 , 𝑥2 )𝑇 = 𝑥1 (1, 1)𝑇 . Therefore,

the vector (1, 1)𝑇 is an eigenvector corresponding to 𝜆1 = −2.
b) Next, let us find an eigenvector corresponding to 𝜆2 = −4.
( )( ) ( ) ( )( ) ( )
−3 − (−4) 1 𝑥1 0 1 1 𝑥1 0
= ⇐⇒ =
1 −3 − (−4) 𝑥2 0 1 1 𝑥2 0
This implies 𝑥1 + 𝑥2 = 0 and hence 𝑥1 = −𝑥1 which implies (𝑥1 , 𝑥2 )𝑇 =

𝑥1 (1, −1)𝑇 . Therefore, the vector (1, −1)𝑇 is an eigenvector corresponding to
the eigenvalue 𝜆2 = −4
Hence, the general solution of the given system is
( ) ( )
1 1
y(𝑡) = 𝑐1 𝑒−2𝑡 + 𝑐2 𝑒4𝑡
1 −1
which is equivalent to
𝑦1 (𝑡) = 𝑐1 𝑒−2𝑡 + 𝑐2 𝑒−4𝑡

𝑦2 (𝑡) = 𝑐1 𝑒−2𝑡 − 𝑐2 𝑒−4𝑡
2. The given system is equivalent to the system

⎛ ⎞ ⎛ ⎞⎛ ⎞
′
𝑦 2 1 1 𝑦1
⎜ 1⎟ ⎜ ⎟⎜ ⎟
⎜𝑦 ′ ⎟ = ⎜1 2 1⎟ ⎜𝑦2 ⎟
⎝ 2⎠ ⎝ ⎠⎝ ⎠
′
𝑦3 1 1 2 𝑦3
That is y′ = Ay, where

⎛ ⎞ ⎛ ⎞ ⎛ ⎞
𝑦1′ 2 1 1 𝑦1
y′ = ⎜
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
′⎟, A =⎜ 𝑎𝑛𝑑 y =⎜
𝑦 1 2 1 ⎝𝑦2 ⎠ .
⎟ ⎟
⎝ 2⎠ ⎝ ⎠
𝑦3′ 1 1 2 𝑦3
⃝January
Let y(𝑡) = x𝑒𝜆𝑡 , y, x ∈ R3 . Then corresponding eigenvalue problem is AX = 𝜆X
with characteristic equation
A − 𝜆I = 0

which is equivalent to the equation

2 − 𝜆 1 1

1 2 − 𝜆 1 =0

2 − 𝜆

1 1
This equation is reduced to (𝜆 − 1)2 (𝜆 − 4) = 0. Therefore, 𝜆1 = 1, 𝜆2 = 4 are the

eigenvalues.
Eigenvector corresponding to the eigenvalue 𝜆1 = 1 can be found as follows.
⎛ ⎞⎛ ⎞ ⎛ ⎞ ⎛ ⎞⎛ ⎞ ⎛ ⎞
2−1 1 1 𝑥1 0 1 1 1 𝑥1 0
⎜ ⎟⎜ ⎟ ⎜ ⎟ ⎜ ⎟⎜ ⎟ ⎜ ⎟
⎜ 1
⎝ 2−1 1 ⎠ ⎝𝑥2 ⎠ = ⎝0⎠ ⇐⇒ ⎝1
⎟ ⎜ ⎟ ⎜ ⎟ ⎜ 1 1⎠ ⎝𝑥2 ⎠ = ⎝0⎟
⎟ ⎜ ⎟ ⎜
⎠
1 1 2−1 𝑥3 0 1 1 1 𝑥3 0
This implies, 𝑥1 + 𝑥2 + 𝑥3 = 0.
i.e., (𝑥1 , 𝑥2 , 𝑥3 )𝑇 = (−𝑥2 − 𝑥3 , 𝑥2 , 𝑥3 )𝑇 = 𝑥2 (−1, 1, 0)𝑇 + 𝑥3 (−1, 0, 1)𝑇 .
Similarly, eigenvector corresponding to 𝜆2 = 4 is obtained to be (1, 1, 1)𝑇 .
Thus, the general solution will be
⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞
𝑦1 (𝑡) −1 −1 1
⎜ ⎟ ⎜ ⎟ 𝑡 ⎜ ⎟ 𝑡 ⎜ ⎟ 4
⎜𝑦2 (𝑡)⎟ = 𝑐1 ⎜ 1 ⎟ 𝑒 + 𝑐2 ⎜ 0 ⎟ 𝑡𝑒 + 𝑐3 ⎜1⎟ 𝑒 𝑡.
⎝ ⎠ ⎝ ⎠ ⎝ ⎠ ⎝ ⎠
𝑦3 (𝑡) 0 1 1
B. Nonhomogeneous Systems
Suppose a non homogeneous system of linear ODEs is given by:
y′ = Ay + F(𝑡), with F(𝑡) ∕≡ 0.
The general solution for this system takes the form y = yℎ + y𝑝 , where yℎ is the general
solution of the corresponding homogeneous system y′ = Ay and y𝑝 is any particular
solution of y′ = Ay + F.
Now to find a particular solution vector y𝑝 we use the method of Undetermined Coeffi-
cients. As in the scalar case, first assuming that y𝑝 has the same general form as 𝐹 and
then find the constants. We will illustrate the method by the following example.
⃝January
Example 2.8.2. Solve the following system of DEs.
𝑦1′ = 𝑦1 + 𝑓1 (𝑡)
𝑦2′ = 6𝑦1 − 𝑦2 + 𝑓2 (𝑡),
where
( ) ( ) ( ) ( ) ( ) ( ) ( )
𝑓1 (𝑡) 2 𝑓1 (𝑡) 2 𝑓1 (𝑡) 2 0
a) = , b) = 𝑒2𝑡 and c) = sin 𝑡+
𝑓2 (𝑡) 1 𝑓2 (𝑡) 1 𝑓2 (𝑡) 1 1
Solution
The system is y′ = Ay + F(𝑡), where

( )
1 0
A= .
6 −1
Then, the corresponding homogeneous system 𝑦 ′ = 𝐴𝑦 has characteristic equation

1 − 𝜆 0
= 1 − 𝜆2 = 0,

6 −1 − 𝜆
with eigenvalues 𝜆 = ±1.
An eigenvector corresponding to the eigenvalue 𝜆 = 1 is (1, 3)𝑇 and an eigenvector
corresponding to the eigenvalue 𝜆 = −1 is (0, 1)𝑇 and hence the solution to homogenous
system is given by ( ) ( )
1 0
y ℎ = 𝑐1 𝑒𝑡 + 𝑐2 𝑒−𝑡 .
3 1
Then for each given case, we are going to find y𝑝
a) Since F(𝑡) = (2, 1)𝑇 which is a constant vector, then y𝑝 will take the form (𝑝1 , 𝑝2 )𝑇
which is also a constant vector. Now, substituting into the equation y′ = Ay + F
we have: ( ) ( )( ) ( )
0 1 0 𝑝1 2
= +
0 6 −1 𝑝2 1
This implies that 𝑝1 + 2 = 0 and 6𝑝1 − 𝑝2 + 1 = 0 and solving this gives us 𝑝1 = −2
and 𝑝2 = −11. Therefore, the particular solution is yp =(−2, −11)𝑇 = −(2, 11)𝑇
and the general solution is
( ) ( ) ( )
1 𝑡 0 −𝑡 2
y(𝑡) = 𝐶1 𝑒 + 𝐶2 𝑒 − .
3 1 11
⃝January
b) Here we have
( )
𝑝1
F(𝑡) = p𝑒2𝑡 = 𝑒2𝑡 .
𝑝2
This implies
( ) ( ) ( ) ( )
2 2𝑝1 𝑝1 2
2p𝑒2𝑡 = Ap𝑒2𝑡 + 𝑒2𝑡 ⇐⇒ = +
1 2𝑝2 6𝑝1 − 𝑝2 1
Since 𝑒𝑡 ∕= 0, by simplifying the given equation we get 2𝑝1 = 𝑝1 + 2, 2𝑝2 =

13
6𝑝1 − 𝑝2 + 1 and solving for 𝑝1 and 𝑝2 gives us 𝑝1 = 2 and 𝑝2 = .
2
c) For the given DE we have
( ) ( )
2 0
F(𝑡) = sin 𝑡 + 𝑒𝑡
1 1
Then we get
( ) ( ) ( )
𝑝1 𝑞1 𝑟1
y𝑝1 = sin 𝑡 + cos 𝑡, 𝑎𝑛𝑑 𝑦 𝑝2 = 𝑒𝑡
𝑝2 𝑞2 𝑟2
and solve for the constants.
2.8.2 The Method of Elimination:
In some cases it could be preferable to consider a higher order differential equation in

stead of a system of first order equations (especially when the characteristic polynomial
is easier to solve). It is possible to transform a system of 𝑛 first order linear ODE to an
𝑛th order linear ODE. The transformation requires the idea of the differential operators.
𝑑𝑛
The operator , which is denoted by 𝐷𝑛 , is called a differential operator and the
𝑑𝑥𝑛
natural number 𝑛 is the power of the operator.
Example 2.8.3.
1. 𝐷2 (𝑥3 + 3𝑥) = 𝐷(3𝑥2 + 3) = 6𝑥.
2. 𝐷2 (2𝑥3 − 2𝑥2 + 3) = 12𝑥 − 4.
⃝January
A linear combination of differential operators of the form
𝑎𝑛 𝐷𝑛 + 𝑎𝑛−1 𝐷𝑛−1 + ⋅ ⋅ ⋅ + 𝑎1 𝐷 + 𝐷0 ,
where 𝑎0 , 𝑎1 , . . . , 𝑎𝑛 are constants is called an 𝑛th order polynomial operator and is denoted
𝑃 (𝐷) and
𝑑𝑛 𝑦 𝑑𝑦
𝑃 (𝐷)𝑦 = (𝑎𝑛 𝐷𝑛 + ⋅ ⋅ ⋅ + 𝑎1 𝐷 + 𝑎1 )𝑦 = 𝑎𝑛 + ⋅ ⋅ ⋅ + 𝑎 1 + 𝑎0 𝑦
𝑑𝑥𝑛 𝑑𝑥
Example 2.8.4.
1. 𝑦 ′′ + 3𝑦 ′ − 𝑦 = 0 implies (𝐷2 + 3𝐷 − 1)𝑦 = 0
2. 𝑦 ′′′ − 4𝑦 ′ = cos 𝑥 implies (𝐷3 − 4𝐷)𝑦 = cos 𝑥.
Definition 2.8.2.
1. Two polynomial operators 𝑃1 (𝐷) and 𝑃2 (𝐷) are equal if and only if 𝑃1 (𝐷)𝑦 =
𝑃2 (𝐷)𝑦 for all functions 𝑦.
2. The sum 𝑃1 (𝐷) + 𝑃2 (𝐷) is obtained by first expressing 𝑃1 and 𝑃2 as linear combi-
nations of the operator D and adding the coefficients of like powers of D
3. The product 𝑃1 (𝐷)𝑃2 (𝐷) is obtained by using the operator 𝑃2 (𝐷) followed by
𝑃1 (𝐷), i.e.
[𝑃1 (𝐷)𝑃2 (𝐷)]𝑦 = 𝑃1 (𝐷)[𝑃2 (𝐷)𝑦].
Example 2.8.5. Let us illustrate the sum and product of operators
1. If 𝑃1 (𝐷) = 3𝐷2 + 7𝐷 − 5, 𝑃2 (𝐷) = 𝐷3 + 6𝐷2 − 2𝐷 − 3, then
𝑃1 (𝐷) + 𝑃2 (𝐷) = 𝐷3 + 𝑔𝐷2 + 5𝐷 − 8.
2. If 𝑃1 (𝐷) = 2𝐷 + 3, 𝑃2 (𝐷) = 𝐷 − 5, then
𝑃1 (𝐷)𝑃2 (𝐷) = (2𝐷 + 3)(𝐷 − 5) = 2𝐷2 − 7𝐷 − 15
⃝January
Basic Properties
If 𝑃 (𝐷) a differential operator, 𝑦1 , 𝑦2 and 𝑦 are functions and 𝑐 is a constant then:
a. 𝑃 (𝐷)(𝑦1 + 𝑦2 ) = 𝑃 (𝐷)𝑦1 + 𝑃 (𝐷)𝑦2 and
b. 𝑃 (𝐷)(𝑐𝑦) = 𝑐𝑃 (𝐷)𝑦.
To solve any system of linear ODE with constant coefficients by elimination method, we
first write each equation using polynomial operators and treat the operators as simple
constants and solve the system using linear algebraic solution methods.
Example 2.8.6. Solve the following systems.
𝑦1′ − 𝑦2 = 𝑥2 𝑦1′ − 2𝑦1 + 2𝑦2′ = 2 − 4𝑒2𝑥

1. 2.
𝑦2′ + 4𝑦1 = 𝑥 2𝑦1′ − 3𝑦1 + 3𝑦2′ − 𝑦2 = 0.
Solution
1. The system is equivalent to
𝐷𝑦1 − 𝑦2 = 𝑥2 𝐷𝑦1 − 𝑦2 = 𝑥2
} {
⇐⇒
𝐷𝑦2 + 4𝑦1 = 𝑥 4𝑦1 + 𝐷𝑦2 = 𝑥
To eliminate 𝑦2 , apply 𝐷 on the first equation. Then the equation is equivalent to:
𝐷2 𝑦1 − 𝐷𝑦2 = 2𝑥
4𝑦1 + 𝐷𝑦2 = 𝑥
Adding the two equations gives 𝐷2 𝑦1 + 4𝑦1 = 3𝑥 ⇐⇒ (𝐷2 + 4)𝑦1 = 3𝑥 and

the characteristic equation for the homogenous part is 𝜆2 + 4 = 0, which implies
𝜆 = ±2𝑖 and 𝑦1ℎ = 𝐶1 cos 2𝑥 + 𝐶2 sin 2𝑥
Then using undetermined constants we get: 𝑦1𝑝 = 𝐴𝑥 + 𝐵 which implies
3
𝐴 = ,𝐵 = 0
4
Therefore, 𝑦1 = 𝐶1 cos 2𝑥 + 𝐶2 sin 2𝑥 + 34 𝑥 and from 4𝑦1 + 𝐷𝑦2 = 𝑥 we have:
𝑦2′ = 𝑥 − 4𝑦1 = 𝑥 − 4𝐶1 cos 2𝑥 − 4𝐶2 sin 2𝑥 − 3𝑥,
which implies 𝑦2′ = −4𝐶1 cos 2𝑥 − 4𝐶2 sin 2𝑥 − 2𝑥.
⃝January
By integrating both sides we get
𝑦2 = −2𝐶1 sin 2𝑥 + 2𝐶2 cos 2𝑥 − 𝑥2 + 𝐶3 .
Then substituting 𝑦1 and 𝑦2 into the first equation we obtain 𝐶3 = 43 . Therefore,
3
𝑦1 = 𝐶1 cos 2𝑥 + 𝐶2 sin 2𝑥 + 𝑥
4
and
3
𝑦2 = −2𝐶1 sin 2𝑥 + 2𝐶2 cos 2𝑥 − 𝑥2 +
4
Remark 2.8.3. It is possible, and may be easier, to use Crammer’s rule in solving
such non homogeneous equation systems.
2. The system is equivalent to
𝐷𝑦1 − 2𝑦1 + 2𝐷𝑦2 = 2 − 4𝑒2𝑥 (𝐷 − 2)𝑦1 + 2𝐷𝑦2 = 2 − 4𝑒2𝑥

⇐⇒
2𝐷𝑦1 − 3𝑦1 + 3𝐷𝑦2 − 𝑦2 = 0 (2𝐷 − 3)𝑦1 + (3𝐷 − 1)𝑦2 = 0
And this can be written in matrix form as

( )( ) ( )
𝐷−2 2𝐷 𝑦1 2 − 4𝑒2𝑥
=
2𝐷 − 3 3𝐷 − 1 𝑦2 0
Then by Crammer’s rule

2 − 4𝑒2𝑥 2𝐷

0 3𝐷 − 1 (3𝐷 − 1)(2 − 4𝑒2𝑥 )
𝑦1 = =
𝐷−2
2𝐷 (𝐷 − 2)(3𝐷 − 1) − (2𝐷 − 3)2𝐷

2𝐷 − 3 3𝐷 − 1
This implies [⟨3𝐷2 − 7𝐷 + 2⟩ − ⟨4𝐷2 − 6𝐷⟩]𝑦1 = (3𝐷 − 1)(2 − 4𝑒2𝑥 ). Simplifying

this gives us (−𝐷2 − 𝐷 + 2)𝑦1 = −12𝑥2𝑒2𝑥 − 2 + 4𝑒2𝑥 and then −𝑦1′′ − 𝑦1′ + 2𝑦1 =
−20𝑒2𝑥 − 2 which is reduced to a second order linear DE in 𝑦1 (Solve this equation.)
and
𝐷 − 2 2 − 4𝑒2𝑥

2𝐷 − 3 0 (2𝐷 − 3)(2 − 4𝑒2𝑥 )
𝑦2 = =
𝐷−2
2𝐷 (−𝐷2 − 𝐷 + 2)

2𝐷 − 3 3𝐷 − 1
⃝January
This implies
(−𝐷2 − 𝐷 + 2)𝑦2 = −8(2𝑒2𝑥 ) − 6 + 12𝑒2𝑥
and then −𝑦2′′ − 𝑦2′ + 2𝑦2 = −4𝑒2𝑥 − 6 which is reduced to a second order linear DE
in 𝑦2 (Solve this equation.)
Therefore, the solution is:
𝑦1 = 𝐶1 𝑒−2𝑥 + 𝐶2 𝑒𝑥 + 5𝑒2𝑥 − 1
𝑦2 = −𝐶1 𝑒−2𝑥 + 21 𝐶2 𝑒𝑥 − 𝑒2𝑥 + 3
2.8.3 Reduction of higher order ODEs to a system of ODEs of

the first order
In the previous sections we used the characteristics equation to solve higher order ODEs.
The characteristic equations are polynomials of degree 𝑛, where 𝑛 is the order of the ODE.
However, solving polynomials is a challenging task when the degree gets larger. Because
of the techniques developed in linear algebra to reduce matrices, it is preferable to solve
eigenvalue problems when the order of the ODE is higher.
There is an equivalence between an 𝑛th order linear ODE and a system of 𝑛 ODEs of first
order. In subsection 2.8.2 we have seen how to transform a system of 𝑛 first order ODEs
to an 𝑛th order linear ODE.
It is also possible to convert a higher order equation into a system of first order equations
using new variable definitions. To see this, consider a homogeneous 𝑛th order linear ODE:
𝑎𝑛 𝑦 (𝑛) + 𝑎𝑛−1 𝑦 (𝑛−1) + ⋅ ⋅ ⋅ + 𝑎1 𝑦 ′ + 𝑎0 𝑦 = 0, where 𝑡 is the independent variable.
Then redefine the variables as follows,
𝑥1 (𝑡) = 𝑦(𝑡)
𝑥2 (𝑡) = 𝑦 ′ (𝑡)
𝑥3 (𝑡) = 𝑦 ′′ (𝑡)
.. .
. = ..
𝑥𝑛 (𝑡) = 𝑦 (𝑛−1) (𝑡)
Then the equation will be equivalent to the system
⃝January
2.9 Exercises 85
𝑥′1 (𝑡) = 𝑥2 (𝑡)

𝑥′2 (𝑡) = 𝑥3 (𝑡)
𝑥′3 (𝑡) = 𝑥4 (𝑡)
.. .
. = ..
𝑥′𝑛−1 (𝑡) = 𝑥𝑛 (𝑡)
𝑎𝑛−1 𝑎𝑛−2 𝑎1 𝑎0
𝑥′𝑛 (𝑡) = − 𝑥𝑛 − 𝑥𝑛−1 − ⋅ ⋅ ⋅ − 𝑥2 − 𝑥1
𝑎𝑛 𝑎𝑛 𝑎𝑛 𝑎𝑛
Or in matrix notation:
𝑋 ′ = 𝐴𝑋,
where the coefficient matrix is

⎛ ⎞
0 1 0 0 ⋅⋅⋅ 0 0
⎜ ⎟
⎜ 0
⎜ 0 1 0 ⋅⋅⋅ 0 0 ⎟
⎟
⋅⋅⋅
⎜
⎜ 0 0 0 1 0 0
⎟
⎟
𝐴=⎜ ⎜ .. .. .. .. .. .. ..
⎟,
⎜ . . . . . . .
⎟
⎟
⎜ ⎟
⎜ 0
⎝ 0 0 0 ⋅⋅⋅ 1 0 ⎟
⎠
𝑎𝑛−4 𝑎𝑛−3
𝑎0 𝑎1
− 𝑎𝑛 − 𝑎𝑛 − 𝑎𝑛 − 𝑎𝑛 ⋅⋅⋅ − 𝑎𝑛−2
𝑎𝑛
− 𝑎𝑛−1
𝑎𝑛
which is the so called the companion matrix of the 𝑛th degree characteristic equation
of the differential equation. Such matrices have special future in matrix theory and the
eigenvalue problem could be solved by employing Jordan form of the matrix.
2.9 Exercises
2.10 Numerical Methods to Solve ODEs
Analytic solutions (in closed form), if there is any possibility to determine, are the best
solutions for any differential equation problem. However, it could be impossible to ana-
lytically solve many practical problems. Then the next step will be to get approximate
solution using infinite series methods. But still a great many DE problems that model real
world situation, and most of nonlinear differential equation problems are too difficult to
solve them analytically even with the help of infinite series approximations. However, an
⃝January
approximate solution can be obtained using quantitative methods, by “discretizing” the
problem. We start with the simplest setting and proceed to a method that best approx-
imates the dependent variable for some discrete values of the independent value, with a
reasonable speed of convergence.
2.10.1 Euler’s Method
Consider a first order initial - value problem:
𝑦 ′ = 𝑓 (𝑥, 𝑦); 𝑦(𝑎) = 𝑏. (2.28)
△𝑦 △𝑦
Since 𝑦 ′ (𝑥) = lim△𝑥→0 , we can approximate 𝑦 ′ by the ratio
△𝑥 △𝑥
Hence we have
△𝑦 ≃ 𝑓 (𝑥, 𝑦)△𝑥 (2.29)
Here, we are looking for an approximate solution for the problem (2.28) on the interval
[𝑎, 𝑑], where 𝑑 is any real number greater than 𝑎 including infinity. To start the proce-
dure, we partition the interval [𝑎, 𝑑] into some points, 𝑥𝑜 , 𝑥1 , 𝑥2 , . . . , 𝑥𝑛−1 , 𝑥𝑛 , 𝑥𝑛+1 , . . .,
where 𝑥𝑜 = 𝑎. Then, let us denote the 𝑦 values at those discretized points of 𝑥
as 𝑦𝑜 , 𝑦1 , 𝑦2 , . . . , 𝑦𝑛−1 , 𝑦𝑛 , 𝑦𝑛+1 , . . . corresponding to the specified values of 𝑥, where
𝑦𝑜 = 𝑦(𝑥𝑜 ) is the initial value. Now let △𝑥 = ℎ denote a constant increment in 𝑥,
called the step size.
Then by taking the first order approximate value of 𝑦(𝑥) from (2.29) we have:
𝑦1 = 𝑦𝑜 + 𝑓 (𝑥𝑜 , 𝑦𝑜 )ℎ
𝑦2 = 𝑦1 + 𝑓 (𝑥1 , 𝑦1 )ℎ; 𝑥 1 = 𝑥𝑜 + ℎ
.. ..
. .
In general, the nth iteration will be
𝑦𝑛+1 = 𝑦𝑛 + 𝑓 (𝑥𝑛 , 𝑦𝑛 )ℎ; 𝑥𝑛 = 𝑥𝑛−1 + ℎ, 𝑛 = 0, 1, 2, . . .
This iterative method is known as Euler’s method. This method is also known as the
tangent-line method because the first straight line segment of the approximate solution
is tangent to the exact solution 𝑦(𝑥) at the point (𝑎, 𝑏), and each subsequent segment
emanating from (𝑥𝑛 , 𝑦𝑛 ) is tangent to the solution curve through that point.
⃝January
2.10 Numerical Methods to Solve ODEs 87
Example 2.10.1. Consider the initial value problem
𝑦 ′ = sin 𝑦 + 𝑦, 𝑦(0) = 0.
Let us try to find approximate solution of this equation using Euler’s method.
..
.
..
.
..
.
Since Euler’s method is based on first order approximation, it may work for sufficiently
small step size ℎ, which makes the iterative scheme very slow. Moreover, when ℎ is
very small the total roundoff error (which may depend on the machine as the computing
machine can only carry a finite number of digits and rounds off the remaining digits then
after) will be significant in size. So, Euler’s method is not only slow in convergence but
also it results in a very rough approximation of the solution.
2.10.2 Runge-Kutta Method
In the Euler’s method we were approximating the curve between the points (𝑥𝑛 , 𝑦𝑛 ) and
(𝑥𝑛+1 , 𝑦𝑛+1 ), by a line with slope 𝑓 (𝑥𝑛 , 𝑦𝑛 ). But the curve may have a different slope in
between these two points and the variation could be significant if the distance between
the two points (the step size) is large. So, if the step size ℎ is not sufficiently small the
difference between the curve for the exact solution and the curve for the approximate
solution could be significantly big. On the other hand taking the step size ℎ to be very
small, can increase the number of iteration so that it will be impractical to take that long
time to arrive at the approximate solution. To improve this limitations in Euler’s method, it
1( )
is better to take the average slopes at the two end points, is 𝑓 (𝑥𝑛 , 𝑦𝑛 )+𝑓 (𝑥𝑛+1 , 𝑦𝑛+1 )
2
instead of simply taking 𝑓 (𝑥𝑛 , 𝑦𝑛 ) alone. But since the value of 𝑦𝑛+1 is not yet determined
at this point of iteration, we estimate it by using Euler’s method as 𝑦𝑛+1 = 𝑦𝑛 +𝑓 (𝑥𝑛 , 𝑦𝑛 )ℎ.
Hence we have
1
𝑦𝑛+1 = 𝑦𝑛 + (𝑘1 + 𝑘2 ),
2
where 𝑘1 = ℎ𝑓 (𝑥𝑛 , 𝑦𝑛 ), 𝑘2 = ℎ𝑓 (𝑥𝑛+1 , 𝑦𝑛 + 𝑘1 )
which is called the Runge-Kutta method of second order (in honer of the two German
mathematicians, Carl D. Runge and M. Wilhelm Kutta of the early 20th century). Actually
⃝January
this method is equivalent to the second order approximation of 𝑦(𝑥) if we assume that 𝑓
is continuous and its first- and second-order partial derivatives are also continuous.
The Runge-Kutta method works by using a weighted average of slopes in the basic Euler
formula to estimate 𝑦(𝑥𝑜 + ℎ) [or in general 𝑦(𝑥𝑘 + ℎ).]
The fourth-order Runge-Kutta method is given by
1
𝑦𝑛+1 = 𝑦𝑛 + ℎ (𝑚1 + 2𝑚2 + 2𝑚3 + 𝑚4 )
6
where 𝑚1 = 𝑓 (𝑥𝑛 , 𝑦𝑛 )
1 1
𝑚2 = 𝑓 (𝑥𝑛 + ℎ, 𝑦𝑛 + ℎ𝑚1 )
2 2
1 1
𝑚3 = 𝑓 (𝑥𝑛 + ℎ, 𝑦𝑛 + ℎ𝑚2 )
2 2
𝑚4 = 𝑓 (𝑥𝑛 + ℎ, 𝑦𝑛 + ℎ𝑚3 )
This method is surprisingly accurate for values of ℎ < 1.
Example 2.10.2. Solve the initial value problem
𝑦 ′ = sin 𝑦 𝑦(0) = 0,
given in the previous example and compare the corresponding results for a given step size,
ℎ = 0.5
Remark 2.10.1. Note that the above methods are given only for for ordinary differential
equations of the first order. However, the same technique can be applied to a system of
ODEs of the first order by simply applying the same procedure for each equation in the
system. Therefore, a higher order ODE can be solved numerically by first converting it to
a system of ODEs of first order and then solving the system numerically.
2.11 Exercises
⃝January
2.11 Exercises 89
This page is left blank intensionally.
⃝January
⃝January
Chapter 3
*Nonlinear ODEs and Qualitative

Analysis
The theory of differential equations has been studied for more than 200 years now and
the case of linear ODEs seems fairly complete. However, very little is known about the
general nature of nonlinear differential equations. Nonlinear equations are rarely solvable
analytically. So, a totally different approach, a qualitative theory, was founded by Henri
Poincaré around the late 1880’s. He sidestepped the search for solutions altogether and
tries to answer fundamental questions about the qualitative and topological behaviour of
solution trajectories of the differential equations. The theory is based upon the phase plane
analysis and the singular points of the equation. We will survey some of the main topics
and methods in this regard. Our aim in this chapter is to develop qualitative methods to
determine properties of solutions without having them explicitly at hand. These properties
are meant to tell us the way how all the trajectories (solution curves) behave near to some
points.
3.1 The Phase Plane

Consider the following second order equation which models the motion of a particle of
unit mass moving on the 𝑥-axis with a force 𝑓 (𝑥(𝑡), 𝑥′ (𝑡)) acting on it:
𝑑2 𝑥
( )
𝑑𝑥
= 𝑓 𝑥, (3.1)
𝑑𝑡2 𝑑𝑡
In this equation the values of 𝑥 and 𝑥′ (𝑡) (the position and velocity of the particle)
characterize the state of the system at each instant of time, hence are called the phases
⃝January
92 *Nonlinear ODEs and Qualitative Analysis
of the system. The plane of the variables 𝑥 and 𝑥′ (𝑡) is known as the phase-plane, and
the set of solutions of the equation depicted in this plane (for different initial conditions)
is called the phase plane diagram.
𝑑𝑥
If we introduce a variable 𝑦 = , then equation (3.1) can be equivalently written as the
𝑑𝑡
system
𝑑𝑥
⎧

⎨ =𝑦
𝑑𝑡 (3.2)
⎩ 𝑑𝑦 = 𝑓 (𝑥, 𝑦).

𝑑𝑡
By regarding 𝑡 as a parameter, a solution of this system is a pair of functions (𝑥(𝑡), 𝑦(𝑡))
defining a curve in the 𝑥𝑦-plane (which is the phase plane). Thus we will be interested
in the overall picture formed by these curves in the phase plane. Using the phase plane
method we can partially solve solve the problem for 𝑥′ (𝑡). We obtain the 𝑥𝑡 solution
graphically from the phase plane diagram by following the procedure below.
Draw the graph of the solution to 𝑦 = 𝑥′ (𝑡) on the phase plane (the 𝑥𝑥′ plane).
graph here ...
Let 𝑃𝑜 be the point which corresponds to 𝑡 = 0 and let 𝑃1 corresponds to the time
𝑡 = △𝑡1 . The average value of 𝑦 = 𝑥′ in the interval 0 < 𝑡 < △𝑡1 can be estimated.
𝑑𝑥 △𝑥1 △𝑥1
Since 𝑦 = , we can estimate 𝑦 1 ≈ and △𝑡1 ≈ . Then we plot the point
𝑑𝑡 △𝑡1 𝑦1
(△𝑡1 , 𝑥1 ) on the 𝑥𝑡 plane.
△𝑥2
Similarly, the point 𝑃2 corresponds to the time 𝑡 = △𝑡1 + △𝑡2 with △𝑡2 ≈ . Then
𝑦2
the point (△𝑡1 + △𝑡2 , 𝑥2 ) can be plotted on the 𝑥𝑡 plane. This process can continue in
a similar fashion until all possible points are sketched. To make sure that the curve we
⃝January
3.2 Critical Points and Stability 93
have drawn is unique and is a solution for the differential equation problem, the following
theorem is helpful.
Theorem 3.1.1 (Existence and Uniqueness).

Let 𝑃 (𝑥, 𝑦, 𝑡) and 𝑄(𝑥, 𝑦, 𝑡) be functions such that each of their partial derivatives
𝑃𝑥 , 𝑃𝑦 , 𝑄𝑥 , 𝑄𝑦 exist and continuous in some neighbourhood of the point (𝑥𝑜 , 𝑦𝑜 , 𝑡𝑜 ) in
Cartesian 𝑥, 𝑦, 𝑡 -coordinate space. Then the initial-value problem
𝑑𝑥
= 𝑃 (𝑥, 𝑦); 𝑥(𝑡𝑜 ) = 𝑥𝑜
𝑑𝑡
𝑑𝑦
= 𝑄(𝑥, 𝑦); 𝑦(𝑡𝑜 ) = 𝑦𝑜
𝑑𝑡
has a solution 𝑥(𝑡), 𝑦(𝑡) on some open interval of 𝑡 containing 𝑡 = 𝑡𝑜 and that solution is
unique.
In the above theorem, if the solutions 𝑥(𝑡), 𝑦(𝑡) are not both constant functions, then
the pair (𝑥(𝑡), 𝑦(𝑡)) defines a curve in the phase plane and this curve is called the path
(or trajectory) of the system described by the differential equation. It is also possible to
verify that at most one path passes through each point of the phase plane. The direction
of increasing 𝑡 along a given path is the same for all solutions. the solution paths do not
intersect one another on the phase plane except at some points, say (𝑥0 , 𝑦0 ), where both
𝑃 and 𝑄 vanish at the same time. That is,
𝑃 (𝑥0 , 𝑦0 ) = 0, and 𝑄(𝑥0 , 𝑦0 ) = 0.
Such points (𝑥0 , 𝑦0 ) are called critical points and at such a point the unique solution for
the system is the constant solution 𝑥(𝑡) = 𝑥0 and 𝑦(𝑡) = 𝑦0 .
Note that a constant solution does not define a path, and therefore no path goes through
a critical point.
3.2 Critical Points and Stability
Consider a general form of an autonomous nonlinear system in two variables as in (3.2)
𝑑𝑥
= 𝑃 (𝑥, 𝑦)
𝑑𝑡
𝑑𝑦
= 𝑄(𝑥, 𝑦) (3.3)
𝑑𝑡
⃝January
If we assume that 𝑦 is dependent on 𝑥, we can equivalently get
𝑑𝑦 𝑑𝑦/𝑑𝑡 𝑄(𝑥, 𝑦)
= =
𝑑𝑥 𝑑𝑥/𝑑𝑡 𝑃 (𝑥, 𝑦)
Any point (𝑥𝑜 , 𝑦𝑜 ) at which both 𝑃 and 𝑄 vanish simultaneously is called a critical (or
singular or equilibrium) point of the system (3.3).
At a singular point (𝑥𝑜 , 𝑦𝑜 ), since 𝑥˙ = 0 and 𝑦˙ = 0, a particular solution of equation (3.3)
is simply the constant values 𝑥(𝑡) = 𝑥𝑜 , 𝑦(𝑡) = 𝑦𝑜 . Note that a constant solution does
not define a path, and therefore no path goes through a critical point.
An equilibrium point 𝑋𝑜 = (𝑥𝑜 , 𝑦𝑜 ) of system (3.3) is said to be stable if motions (or
trajectories) that start sufficiently close to 𝑋𝑜 remain close to 𝑋𝑜 , otherwise it is called
unstable. This statement can be reformulated mathematically as in the following defini-
tion.
Definition 3.2.1. Let 𝑑(𝑃1 , 𝑃2 ) denote the distance between any two points 𝑃1 = (𝑥1 , 𝑦1 )
and 𝑃2 = (𝑥2 , 𝑦2 ) and let 𝑃 (𝑡) = (𝑥(𝑡), 𝑦(𝑡)) denote the representative point in the phase
plane corresponding to system (3.3). Then
(i) a singular (or an equilibrium) point 𝑋𝑜 = (𝑥𝑜 , 𝑦𝑜 ) is stable if for any given 𝜖 > 0,
there is a 𝛿 > 0 such that
𝑑(𝑃 (0), 𝑋𝑜 ) < 𝛿 ⇒ 𝑑(𝑃 (𝑡), 𝑋𝑜 ) < 𝜖, ∀𝑡 > 0
Otherwise, the equilibrium point 𝑋𝑜 is called unstable.
(ii) a singular point 𝑋𝑜 is called asymptotically stable if motions (or trajectories) that
start out sufficiently close to 𝑋𝑜 not only stay close to 𝑋𝑜 but actually approach
𝑋𝑜 as 𝑡 → ∞
i.e., ∃𝛿 > 0 s.t. 𝑑(𝑃 (𝑜), 𝑋𝑜 ) < 𝛿 ⇒ lim 𝑑(𝑃 (𝑡), 𝑋𝑜 ) = 0

𝑡→∞
Singular points of a system can also be classified further depending on the style of motions
of the trajectories around it.
Definition 3.2.2. A singular point is called:
1) a Center if it is surrounded by closed orbits (paths) corresponding to periodic

motions. A center is stable but not asymptotically stable.
⃝January
Figure 3.1: A center with elliptical orbits (always stable)
Figure 3.2: Stable and unstable Spirals
2) a Focus (or Spiral) if all trajectories around 𝑋𝑜 ‘focus’ towards (or outward) it as
𝑡 → ∞. A focus can be asymptotically stable or unstable.
3) a Node if there are infinitely many trajectories entering (or leaving) the point 𝑋𝑜 .
There are two types of nodes, Proper or Improper nodes, where each could be
stable or unstable.
Figure 3.3: Proper and improper nodes with respective stability
4) a Saddle if all trajectories (paths) approach to 𝑋𝑜 in one direction and move away
from it in the other direction. A saddle is always unstable.
⃝January
Figure 3.4: A saddle point (always unstable)
The two straight-line trajectories through the saddle (along which the flow is at-
tracted and repelled) are called the stable and unstable manifolds respectively.
In many practical problems people are usually interested in the stability of equilibrium
points, especially for nonlinear coupled systems. That means, if we take an initial point
near to an equilibrium point 𝑋𝑜 = (𝑥𝑜 , 𝑦𝑜 ), does the point (𝑥(𝑡), 𝑦(𝑡)) on the solution
curve (trajectory) remain near 𝑋𝑜 ?
To identify properties of the system that can characterize the type and stability of equi-
librium points we need to investigate solutions of linear systems first and then we apply
those results also for nonlinear systems.
3.2.1 Stability for linear systems
Consider a system of linear equations,
ẋ(𝑡) = 𝐴x(𝑡), (3.4)
where 𝐴 is a constant 𝑛 × 𝑛 matrix and x(𝑡) is a vector function of 𝑛 components.

Then from the previous chapter we know that the solution of the system depends on
the eigenvalues of the coefficient matrix 𝐴. Suppose 𝐴 has 𝑛 distinct real eigenvalues,
𝜆1 , 𝜆2 , . . . , 𝜆𝑛 . Hence the general solution of (3.4) is given by,
x(𝑡) = 𝑐1 𝑒𝜆1 𝑡 v1 + 𝑐2 𝑒𝜆2 𝑡 v2 + ⋅ ⋅ ⋅ + 𝑐𝑛 𝑒𝜆𝑛 𝑡 v𝑛 , (3.5)
where the vectors v𝑘 is an eigenvector of 𝐴 corresponding to the eigenvalue 𝜆𝑘 for each

𝑘 = 1, 2, . . . , 𝑛.
It is clear that x(𝑡) ≡ 0, where x(𝑡0 ) = 0 for initial time 𝑡0 , is also a solution of the
system (3.4) (the trivial solution). Now to see the relationship between the 0 solution
⃝January
and the general solution of the linear system when 𝑡 increases, let us consider the limit,
lim x(𝑡) = lim 𝑐1 𝑒𝜆1 𝑡 v1 + 𝑐2 𝑒𝜆2 𝑡 v2 + ⋅ ⋅ ⋅ + 𝑐𝑛 𝑒𝜆𝑛 𝑡 v𝑛

[ ]
𝑡→∞ 𝑡→∞
( ) ( )
= lim 𝑒𝜆1 𝑡 𝑐1 v1 + ⋅ ⋅ ⋅ + lim 𝑒𝜆𝑛 𝑡 𝑐𝑛 v𝑛 ,
𝑡→∞ 𝑡→∞
Since the vectors 𝑐𝑘 v𝑘 are independent of 𝑡, for all 𝑘 = 1, . . . , 𝑛, the limit for each
exponential function either vanishes or explodes. Thus if all the eigenvalues 𝜆𝑘 ’s are
positive real numbers, the limit of the exponential functions tends to infinity, and if all the
eigenvalues are negative real numbers each limit approaches asymptotically to 0. Hence,
the general solution of the system converges to 0 if all the eigenvalues are negative real
numbers and diverges if some of the eigenvalues are positive.
On the other hand if some of the eigenvalues are complex, 𝜆𝑘 = 𝑎𝑘 + 𝑖𝑏𝑘 , for some
𝑘 = 1, . . . , 𝑛, then the corresponding solution is of the form
x𝑘 (𝑡) = 𝑒𝑎𝑘 𝑡 𝑐1𝑘 cos 𝑏𝑘 𝑡 + 𝑐2𝑘 sin 𝑏𝑘 𝑡 v𝑘 .

( )
Again the convergence of the solution x𝑘 (𝑡) depends on the real part of the eigenvalues.
That is, if Re(𝜆𝑘 ) < 0 the solution converges to 0, otherwise it diverges as the amplitudes
increases indefinitely. To formulate a general characterization for the stability of the
solution of a linear system, we need the result of the following Lemma.
Lemma 3.2.3. Let 𝛼, 𝛽 ∈ R and 𝑘 ∈ N ∪ {0}, and let the function 𝑓 (𝑡) be defined by
𝑓 (𝑡) = 𝑡𝑘 𝑒𝛼𝑡 cos 𝛽𝑡 or 𝑓 (𝑡) = 𝑡𝑘 𝑒𝛼𝑡 sin 𝛽𝑡.
Then
i) lim 𝑓 (𝑡) = 0 if, and only if 𝛼 < 0.

𝑡→∞
ii) There exists a constant 𝐶, such that ∣𝑓 (𝑡)∣ ≤ 𝐶 for all 𝑡 ≥ 0 if, and only if either
𝛼 < 0 or 𝛼 = 0 and 𝑘 = 0.
This Lemma indicates that exponential decays always cancel out polynomial growths.
Thus, if the real parts of all eigenvalues are negative, the system has a vanishing general
solution in the limiting case, whether the eigenvalues are repeating or not. This justifies
the following theorem.
⃝January
Theorem 3.2.4. A system of first order differential equation
ẋ(𝑡) = 𝐴x(𝑡)
with constant coefficient matrix 𝐴, has asymptotically stable zero solution if, and only if
Re(𝜆) < 0 for all eigenvalue 𝜆 of 𝐴.
If there is an eigenvalue 𝜆, with Re(𝜆) > 0, then the zero solution is unstable.
Therefore, the stability of a zero solution depend on the sign of the real parts of all the
eigenvalues of the coefficient matrix 𝐴.
Example 3.2.1. Consider the following the system of differential equations

𝑑𝑥
= 2𝑥 − 6𝑦 + 𝑧,
𝑑𝑡
𝑑𝑦
= 3𝑥 − 3𝑦 − 𝑧,
𝑑𝑡
𝑑𝑧
= 3𝑥 − 𝑦 − 3𝑧.
𝑑𝑡
⎛ ⎞
2 −6 1
⎜ ⎟
The coefficient matrix is then, 𝐴 = ⎝ 3 −3 −1 ⎟ ⎠ with eigenvalues 𝜆1 = −2, 𝜆2 =
⎜
3 −1 −3
√ √
−1 + 𝑖 6, and 𝜆3 = −1 − 𝑖 6.
Since the real parts of the⎛eigenvalues
⎞ ⎛ are −2
⎞ and −1, both negative, by the above
𝑥(𝑡) 0
⎜ ⎟ ⎜ ⎟
⎝ 𝑦(𝑡) ⎠ ≡ ⎝ 0 ⎠ is asymptotically stable.
theorem the zero solution, ⎜ ⎟ ⎜ ⎟
𝑧(𝑡) 0
3.2.2 Stability for nonlinear systems
To study the type and stability of coupled nonlinear systems, we approximate the non-
linear system (equation (3.3)) by its linear terms in the Taylor series expansion in the
neighborhood of each singular point.
Consider again the system:
𝑑𝑥
= 𝑃 (𝑥, 𝑦)
𝑑𝑡
𝑑𝑦
= 𝑄(𝑥, 𝑦)
𝑑𝑡
⃝January
The first order Taylor series approximation of the two functions gives us,
𝑃 (𝑥, 𝑦) ≈ 𝑃𝑥 (𝑥𝑜 , 𝑦𝑜 )(𝑥 − 𝑥𝑜 ) + 𝑃𝑦 (𝑥𝑜 , 𝑦𝑜 )(𝑦 − 𝑦𝑜 )

𝑄(𝑥, 𝑦) ≈ 𝑄𝑥 (𝑥𝑜 , 𝑦𝑜 )(𝑥 − 𝑥𝑜 ) + 𝑄𝑦 (𝑥𝑜 , 𝑦𝑜 )(𝑦 − 𝑦𝑜 )
Now from this approximation, since the coefficients of the linear terms are constants,
letting 𝑎 = 𝑃𝑥 (𝑥𝑜 , 𝑦𝑜 ), 𝑏 = 𝑃𝑦 (𝑥𝑜 , 𝑦𝑜 ), 𝑐 = 𝑄𝑥 (𝑥𝑜 , 𝑦𝑜 ) and 𝑑 = 𝑄𝑦 (𝑥𝑜 , 𝑦𝑜 ) we have
𝑋˙ = 𝑎𝑋 + 𝑏𝑌
𝑌˙ = 𝑐𝑋 + 𝑑𝑌, (3.6)
where 𝑋 = 𝑥 − 𝑥𝑜 and 𝑌 = 𝑦 − 𝑦𝑜 are translates of the variables by the equilibrium point

(𝑥𝑜 , 𝑦𝑜 ).
The above process is called a linearization process and we now study closely the ap-
proximated linear system for the stability of its equilibrium point 0.
System (3.6) can be rewritten as
( ) ( )( )
𝑋˙ 𝑎 𝑏 𝑋
=
𝑌˙ 𝑐 𝑑 𝑌
Clearly (0, 0) is a critical point for the linear system (3.6) [and hence the point (𝑥𝑜 , 𝑦𝑜 ) is
a critical point for the system (3.3)].
( )
𝑎 𝑏
Let 𝜆1 and 𝜆2 be the two eigenvalues of the coefficient matrix .
𝑐 𝑑
Then the nature of the critical point (0, 0) of the system (3.6) depends upon the nature
of the eigenvalues 𝜆1 and 𝜆2 of the coefficient matrix, as we have seen in the previous
section.
1. If 𝜆1 and 𝜆2 are real, unequal and of the same sign, then the critical point (0, 0) of
the linear system (3.6) is a node.
– If, in addition, both 𝜆1 and 𝜆2 are positive, then the critical point is an unstable
node.
– If, both 𝜆1 and 𝜆2 are negative, then the critical point is a stable node.
2. If 𝜆1 and 𝜆2 are real and of opposite sign, then the critical point (0, 0) of the linear
system (3.6) is a saddle point.
⃝January
3. If 𝜆1 and 𝜆2 are real and equal, then the critical point (0, 0) of the linear system
(3.6) is a node.
– If, in addition, 𝜆1 = 𝜆2 < 0, then it is a stable node and if 𝜆1 = 𝜆2 > 0, then

it is an unstable node.
– If, 𝑎 = 𝑑 ∕= 0 and 𝑏 = 𝑐 = 0, then it is a proper node, otherwise an improper
node.
4. If 𝜆1 and 𝜆2 are complex conjugates with the real part not zero, then the equilibrium
point (0, 0) of the linear system (3.6) is a focus or spiral.
– If, in addition, the real part is negative, then the critical point is a stable focus.
– If, the real part is positive then it is an unstable focus.
5. If 𝜆1 and 𝜆2 are pure imaginary, then the equilibrium point (0, 0) of the linear system
(3.6) is a center.
A center is always stable even though it is not asymptotically stable.
In general for two dimensional system

( of differential
) equations since the associated coef-
𝑎 𝑏
ficient matrix has the form, 𝐴 = whose characteristics equation will be
𝑐 𝑑
𝜆2 − (𝑎 + 𝑑)𝜆 + (𝑎𝑑 − 𝑏𝑐) = 𝜆2 − tr(𝐴)𝜆 + det(𝐴),
the following graph summarizes the stability analysis of equilibrium points.
Remark 3.2.5. In the above process some properties of the functions 𝑃 and 𝑄 were
simply assumed without additional clarifications. But we need the following remark.
1. In the linearization process it was assumed that
– the constants 𝑎, 𝑏, 𝑐, and 𝑑 are real numbers;

– the functions 𝑃 and 𝑄 have continuous first partial derivatives in the neigh-
borhood of the critical points.
The above two requirements will be met, if the Jacobian

∂(𝑃, 𝑄)
∕= 0.
∂(𝑥, 𝑦) (𝑥𝑜 ,𝑦𝑜 )
⃝January
Figure 3.5: Stability regions for two dimensional Linear Systems.
2. The constant terms in the linearized system are missing because 𝑃 (𝑥𝑜 , 𝑦𝑜 ) =
𝑄(𝑥𝑜 , 𝑦𝑜 ) = 0.
The nature of the equilibrium points of the nonlinear system (3.3) can be determined
from that of the linearized system (3.6) as in the following Theorems.
Theorem 3.2.6 (Poincaré’s Result). The classification of all singular points of the non-
linear system (3.3) correspond in both type and stability with the results obtained by
considering the linearized system (3.6) except for a center and a proper node.
In these exceptional cases
(i) a center of the linearized system could be either a focus or a center for the nonlinear
system.
(ii) a proper node could also be either a spiral or a node for the nonlinear system (3.3).
To determine these exceptional cases one requires to study further the original nonlinear
system itself.
Example 3.2.2. The pair of differential equations

1
𝑥˙ = 𝑥 − 𝑥𝑦 (3.7)
2
𝑦˙ = −2𝑦 + 𝑥𝑦, 𝑥, 𝑦 ≥ 0, (3.8)
⃝January
occur in a study of interacting populations. Find the equilibrium points and determine
their nature.
Solution
Here the coupled functions are

1 1
𝑃 (𝑥, 𝑦) = 𝑥 − 𝑥𝑦 = 𝑥( − 𝑦)
2 2
𝑄(𝑥, 𝑦) = −2𝑦 + 𝑥𝑦 = 𝑦(𝑥 − 2).
But 𝑃 (𝑥, 𝑦) = 0 if either 𝑥 = 0 or 𝑦 = 21 . Similarly 𝑄(𝑥, 𝑦) = 0 if either 𝑦 = 0 or 𝑥 = 2.

Hence, (0, 0) and (2, 21 ) are the only critical points for the system.
Now to linearize the nonlinear system, we need the following partial derivatives.
∂𝑃 1 ∂𝑃
= −𝑦 = −𝑥
∂𝑥 2 ∂𝑦
∂𝑄 ∂𝑄
=𝑦 = 𝑥 − 2.
∂𝑥 ∂𝑦
Then we will characterize the two critical points separately.
1 At the critical point (0, 0).

The linearized equation of the system (3.7) at (0, 0) is given by
( ) ( )( )
1
𝑥˙ 2
0 𝑥
=
𝑦˙ 0 −2 𝑦
1
Then the eigenvalues are 𝜆1 = 2
and 𝜆2 = −2. Since the two eigenvalues are real
and with opposite signs, the equilibrium point is a saddle point, which is unstable.
By Poincarés result, the point has the same property also for the original nonlinear
system.
2 At the critical point (2, 12 ).

In this case the linearized equation of the system (3.7) will be
( ) ( )( )
𝑥˙ 0 −2 𝑥−2
= 1
𝑦˙ 2
0 𝑦 − 12
Then the eigenvalues for the coefficient matrix are solutions of the characteristics
equation, 𝜆2 + 1 = 0, which are 𝜆1,2 = ±𝑖, – pure imaginary.
⃝January
Hence the point (2, 12 ) is a center for the linear system. But since a center is one
of the exceptional cases in the Poincaré’s Theorem, we need to check whether it is
a center or a spiral for the original nonlinear system. To check this case we need to
investigate the original equations further, by assuming 𝑦 is dependent on 𝑥. Hence
we have
( )( )
𝑑𝑦 𝑄(𝑥, 𝑦) 𝑦(𝑥 − 2) 𝑦 𝑥−2
= = 1 = 1
𝑑𝑥 𝑃 (𝑥, 𝑦) ( 2 − 𝑦)𝑥 −𝑦 𝑥
∫ (1 ) 2
−𝑦
) ∫ (
2 𝑥−2
⇒ 𝑑𝑦 = 𝑑𝑥
𝑦 𝑥
1
⇒ ln 𝑦 − 𝑦 = 𝑥 − 2 ln 𝑥 + 𝑐, for some arbitrary constant 𝑐,
2
which represents infinitely many curves (depending on the value of 𝑐) on the phase-
plane. Among them consider a solution curve which passes through the point (2, 1).
Then the last equation becomes
1
ln 1 − 1 = 2 − 2 ln 2 + 𝑐 ⇒ 𝑐 = −3 + 2 ln 2.
2
Hence the equation of the particular trajectory that passes through the point (2, 1)
is given by
1
ln 𝑦 − 𝑦 = 𝑥 − 2 ln 𝑥 − 3 + 2 ln 2.
2
If the critical point (2, 21 ) is a center, then a straight line through this point must
intersect with the trajectory exactly twice; otherwise the critical point will be a
spiral. To this end, consider now the intersection of this last curve with the vertical
line 𝑥 = 2, which pass through the critical point (2, 12 ). Then substituting 𝑥 = 2 in
the equation of the trajectory, to get the intersection points, we will get
1
ln 𝑦 − 𝑦 = 2 − 2 ln 2 − 3 + 2 ln 2 = −1.
2
1
⇒ ln 𝑦 = 𝑦−1
2
⇒ ln 𝑦 = 2(𝑦 − 1) ⇒ 𝑦 = 𝑒2(𝑦−1) .
Then by setting 𝑧 = 𝑦 and 𝑧 = 𝑒2(𝑦−1) and draw the graph we can see that there
only two intersection points between the two curves.
Hence the critical point (2, 12 ) is a center also for the original nonlinear system. □
The above procedure (the linearization process) can also be used to study a second order
nonlinear ODE. This can be done by using substitution of variables 𝑦 = 𝑥,
˙ which imply
that 𝑦˙ = 𝑥¨. This will result in a nonlinear system of two first order equations.
⃝January
Figure 3.6: intersection points between the graphs of 𝑧 = 𝑦 and 𝑧 = 𝑒2(𝑦−1) .
However, if such an equation has no term in 𝑥,

˙ the following theorem assures us that the
exceptional cases will be waived.
Theorem 3.2.7. If the nonlinear equation 𝑥¨ + 𝑓 (𝑥) = 0 has a singular point in the 𝑥𝑥˙
plane (phase plane), where the linearized system indicates a center or a proper node, the
nonlinear equation also has the same property.
Example 3.2.3. The equation 𝑥¨ + 𝜖𝑥˙ 3 + 𝑥 = 0 models a harmonic oscillator with cubic
damping - that is, with a damping term proportional to the velocity cubed. Find the
critical point(s) and determine their nature.
Solution
˙ then we have 𝑦˙ = −𝑥 − 𝜖𝑦 3 . With this change of variables the second order

Let 𝑦 = 𝑥,
equation can be rewritten as the following system.
𝑥˙ = 𝑦
𝑦˙ = −𝑥 − 𝜖𝑦 3
Then it is easy to check that (0, 0) is the only critical point for the system. Now to
determine the nature of the critical point, we linearize the system at (0, 0) to get,
( ) ( )( )
𝑥˙ 0 1 𝑥
= with eigenvalues of the coefficient matrix are 𝜆1,2 = ±𝑖.
𝑦˙ −1 0 𝑦
Hence the critical point is a center for the linear system. However since the equation
contains a velocity term (𝑥),
˙ the above theorem does not hold. Hence we have to check
⃝January
3.3 Stability by Lyapunav’s Method 105
whether this critical point is a center or a focus for the nonlinear equation. It is to the
reader to verify using a similar approach as in the previous example that the critical point
(0, 0) is a stable focus for the original nonlinear equation. □
Example 3.2.4. Locate the singular points of the differential equation

1
𝑥¨ + 𝑥 − 𝑥2 = 0
4
and determine their nature.
Solution
Let 𝑦 = 𝑥.
˙ Then the equivalent system of first order equations is
𝑥˙ = 𝑦
1 2
𝑦˙ = 𝑥 −𝑥
4
Then the singular points are (0, 0) and (4, 0). To determine the nature of each equilibrium
points we need to linearize the system at each of the critical points and study their
nature. Since the nonlinear equation has no velocity term, Poincarés result hold without
exceptions. The linearized systems are:
1 At the point (0, 0)

( ) ( )( )
𝑥˙ 0 1 𝑥
= with the corresponding eigenvalues 𝜆1,2 = ±𝑖.
𝑦˙ −1 0 𝑦
Hence the critical point (0, 0) is a center for the linear system as well as for the
original equation.
2 At the point (4, 0):

( ) ( )( )
𝑥˙ 0 1 𝑥−4
= with the corresponding eigenvalues of the coeffi-
𝑦˙ 1 0 𝑦
cient matrix 𝜆1,2 = ±1.
Hence the singular point (4, 0) is a saddle point for the original equation. □
3.3 Stability by Lyapunav’s Method
The linearization technique is a very good approach to determine stability of nonlinear

coupled systems provided that it gives a conclusive characterization. However, as can be
⃝January
seen in Poincarés Theorem (Theorem (3.2.6)), there are cases where stability property
in linearized system does not go along with the nonlinear system. In such cases the
procedure rests on reconsidering the nonlinear system using some systematic approach,
which is not so easy to determine (if it is possible at all). Thus we need to find another
method which can help us to conclude the stability of the critical points in the those
exceptional cases as well. To this end if we can find a decreasing function of 𝑡, with
the same set of variables which has a local minimum at that equilibrium point, then the
point is a stable equilibrium. This was generalized by the nineteenth century Russian
Mathematician, Alexander M. Lyapunov.
Definition 3.3.1. A Lyapunov function for the first order autonomous system ẋ = 𝐹 (x)
is a continuous real-valued function 𝐿(x) that is non-decreasing on all solutions x(𝑡)), of
the system. That means,
𝐿(x(𝑡)) ≤ 𝐿(x(𝑡𝑜 )) for all 𝑡 > 𝑡𝑜
𝐿 is said to be a strict Lyapunov function if it satisfies the strict inequality
𝐿(x(𝑡)) < 𝐿(x(𝑡𝑜 )) for all 𝑡 > 𝑡𝑜
whenever x(𝑡) is a non-equilibrium solution to the system.
Remark 3.3.2. One can conclude from the conditions of the above definition that the
Lyapunov function has a constant value at the equilibrium solutions of the system.
Given a system of ODEs 𝐹 (x) and a function 𝐿(x), how can we check whether 𝐿(x)
represents a Lyapunov function of a system? This question can be answered by considering
the following observation, where x ∈ R2 . Let 𝒞 = x(𝑡) = (𝑥(𝑡), 𝑦(𝑡)) represents a path of
the solution for the system ẋ = 𝐹 (x), and consider the function 𝐿(x) which is continuous
and has continuous first partial derivative in a region containing the path 𝒞. Then we
have,
𝑑𝐿 ∂𝐿 𝑑𝑥 ∂𝐿 𝑑𝑦
= +
𝑑𝑡 ∂𝑥 𝑑𝑡 ∂𝑦 𝑑𝑡
∂𝐿 ∂𝐿
= 𝑃 (𝑥, 𝑦) + 𝑄(𝑥, 𝑦)
∂𝑥 ∂𝑦
= ∇𝐿(x) ⋅ 𝐹 (x),
𝑑𝑥 𝑑𝑦
where, 𝑑𝑡
= 𝑃 (𝑥, 𝑦) = 𝐹1 and 𝑑𝑡
= 𝑄(𝑥, 𝑦) = 𝐹2 represents the system.
⃝January
3.3 Stability by Lyapunav’s Method 107
Then 𝐿 is a Lyapunov function, then it must be decreasing on the path 𝒞 for all 𝑡 > 𝑡𝑜 ,
which implies that its time derivative must be non-positive. This is summarized in the
following Proposition.
Proposition 3.3.3. A continuously differentiable function 𝐿(x) is a Lyapunov function

for the system ẋ = 𝐹 (x) if, and only if it satisfies the Lyapunov inequality:
𝑑
𝐿(x(𝑡)) = ∇𝐿(x) ⋅ 𝐹 (x) ≤ 0 for all x(𝑡).
𝑑𝑡
Moreover, the above inequality is strict for 𝐹 (x) ∕= 0 if, and only if 𝐿(x) is a strictly
Lyapunov function.
Thus we can use the Lyapunov function technique to decide the stability of equilibria of
a system as in the following Theorem.
Theorem 3.3.4 (Lyapunov). Consider the system
ẋ = 𝐹 (x) (3.9)
with equilibrium point x𝑜 . Let 𝒰 be an open set containing x𝑜 and let 𝐿 : 𝒰 → R be a

smooth function such that
a) x𝑜 is a minimum point for 𝐿,

𝑑
b) (𝐿(x(𝑡))) ≤ 0 whenever x is a solution of system (3.9).
𝑑𝑡
Then x𝑜 is a stable equilibrium point.
Remark 3.3.5. In condition (b) of the above theorem,
𝑑
1. if (𝐿(x(𝑡))) < 0 for every solution x(𝑡) of system (3.9), then the equilibrium
𝑑𝑡
point x𝑜 is asymptotically stable.
𝑑
2. if (𝐿(x(𝑡))) > 0 for every solution x(𝑡) of system (3.9), then the equilibrium
𝑑𝑡
point x𝑜 is unstable.
Example 3.3.1. Consider the coupled nonlinear system,

{ √
𝑥˙ = 𝑦 − 𝑥 𝑥2 + 𝑦 2
√
𝑦˙ = −𝑥 − 𝑦 𝑥2 + 𝑦 2 .
⃝January
One can easily verify that (0, 0) is an equilibrium point, which is a center for the linearized
system. Thus we can not draw a conclusion about its stability for the original nonlinear
system. Now, to apply the Lyapunov method, define,
𝐿(𝑥, 𝑦) = 𝑥2 + 𝑦 2 .
Then, since
√ √
∇𝐿⋅𝐹 = (2𝑥, 2𝑦)⋅(𝑦−𝑥 𝑥2 + 𝑦 2 , −𝑥−𝑦 𝑥2 + 𝑦 2 ) = −2(𝑥2 +𝑦 2 )3/2 < 0 ∀(𝑥, 𝑦) ∕= (0, 0),
clearly 𝐿 is a Lyapunov function. Moreover the point (0, 0) is a minimum point of 𝐿.

Then by Lyapunov Theorem since
𝑑 𝑑
𝐿(𝑥, 𝑦) = (𝑥2 + 𝑦 2 ) = 2𝑥𝑥˙ + 2𝑦 ⋅ 𝑦 = −2(𝑥2 + 𝑦 2 )3/2 < 0 ∀(𝑥, 𝑦) ∕= (0, 0),
𝑑𝑡 𝑑𝑡
the equilibrium point (0, 0) is asymptotically stable for the nonlinear system. Hence it is
a stable spiral for the nonlinear system. □
The main difficulty in Lyapunov’s method is finding the Lyapunov function. There is no
general procedure of finding this function and one has to rely on heuristics in guessing the
possible function 𝐿 that can satisfy the conditions in Lyapunov’s Theorem.
3.4 Exercises
1. Locate the equilibrium points

{ and determine their nature for
{ each of the following
2
𝑥˙ = 𝑥 + 4𝑦 − 𝑥 𝑥˙ = sin 𝑥 − 4𝑦
systems a) b)
𝑦˙ = 6𝑥 − 𝑦 + 2𝑥𝑦 𝑦˙ = sin 2𝑥 − 5𝑦
2. The equation of a damped pendulum is given by
𝑥¨ + 𝑘 𝑥˙ + 𝑛2 sin 𝑥 = 0, 𝑘 > 0 and for some natural number 𝑛.
Find the equilibrium points (with respect to 𝑘) and determine their nature.
⃝January

ODE1

Încărcat de

Informații document

Descriere originală:

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

ODE1

Încărcat de

Drepturi de autor:

Formate disponibile

Math 331: Applied Mathematics III Lecture Notes

I Ordinary Diﬀerential Equations 5

1 Ordinary Diﬀerential Equations of the First Order 7

2 Ordinary Diﬀerential Equations of The Second and Higher Order 37

3 *Nonlinear ODEs and Qualitative Analysis 91

Ordinary Diﬀerential Equations

Ordinary Diﬀerential Equations of the

1.1 Basic Concepts and Ideas

∙ If an equation contains only ordinary derivatives of one or more dependent variables

∙ If a function is deﬁned in terms of two or more independent variables, the corre-

where F is a real-valued function of 𝑛 + 2 variables 𝑥, 𝑦, 𝑦 ′ , 𝑦 ′′ , ..., 𝑦 (𝑛) .

Example 1.1.2. The normal form of the ﬁrst-order equation 4𝑥𝑦 ′ + 𝑦 = 𝑥 is

The ﬁrst order ordinary diﬀerential equation is generally expressed as:

𝐹 (𝑥, 𝑦, 𝑦 ′ ) = 0 or 𝑦 ′ = 𝑓 (𝑥, 𝑦).

For example, the diﬀerential equation 𝑦 ′ + 𝑦 = 𝑥 is equivalent to 𝑦 ′ + 𝑦 − 𝑥 = 0.

An 𝑛𝑡ℎ -order ordinary diﬀerential equation (1.3) is said to be linear if F is linear in

𝑎𝑛 (𝑥)𝑦 (𝑛) + 𝑎𝑛−1 (𝑥)𝑦 (𝑛−1) + ⋅ ⋅ ⋅ + 𝑎1 (𝑥)𝑦 ′ + 𝑎0 (𝑥)𝑦 − 𝑏(𝑥) = 0, (1.5)

We may equivalently use the notations

then 𝑦 = ℎ(𝑥) is called a ( an Explicit) solution of the ODE on [𝑎, 𝑏].

Example 1.1.3. Consider the diﬀerential equation 𝑦 ′′ + 𝑦 = 0.

Example 1.1.4. Consider the diﬀerential equation 𝑦𝑦 ′ = −𝑥.

1.2 Separable Diﬀerential Equations

Consider a diﬀerential equation

where G(x) is an antiderivative (indeﬁnite integral) of 𝑓 (𝑥).

Many ﬁrst-order ODEs can be reduced or transformed to the form

which is the general explicit solution of the given equation.

Example 1.2.2. Solve the DE 6𝑦𝑦 ′ + 4𝑥 = 0.

The equation 6𝑦𝑦 ′ + 4𝑥 = 0 is equivalent to

which is an implicit solution of the given ﬁrst order diﬀerential equation.

Example 1.2.3. Solve the DE 𝑦 ′ = 𝑦 2 𝑒2𝑥 .

First rewrite the equation as

Remark 1.2.1. It is recommended to write an explicit solution to the diﬀerential equation

1.2.1 Equation Reducible to Separable Form

Suppose we have a diﬀerential equation that can be written in the form:

Such an equation is not in general separable. However, if we set 𝑢 = 𝑎𝑥 + 𝑏𝑦 + 𝑐, we get

Example 1.2.4. Solve the diﬀerential equation 𝑦 ′ = (𝑥 + 𝑦)2 .

Let 𝑢 = 𝑥 + 𝑦. Then 𝑢′ = 1 + 𝑦 ′ which implies 𝑦 ′ = 𝑢′ − 1. With this substitution the

Example 1.2.5. Solve the diﬀerential equation (2𝑥 − 4𝑦 + 5)𝑦 ′ + 𝑥 − 2𝑦 + 3 = 0.

Let 𝑢 = 𝑥 − 2𝑦. Then, 𝑢′ = 1 − 2𝑦 ′ which implies 𝑦 ′ = 21 (1 − 𝑢′ ). Therefore, the equation

Suppose we have an equation that can be written in the form

Then by integrating we obtain a general solution.

Example 1.2.6. Solve 𝑥2 𝑦 ′ = 𝑥2 + 𝑥𝑦 + 𝑦 2 .

For 𝑥 ∕= 0, the diﬀerential equation 𝑥2 𝑦 ′ = 𝑥2 + 𝑥𝑦 + 𝑦 2 is equivalent to

Example 1.2.7. Solve the DE: 2𝑥𝑦𝑦 ′ = 𝑦 2 − 𝑥2 .

Divide both sides by 𝑥2 , for 𝑥 ∕= 0, to get

Deﬁnition 1.2.2. For the diﬀerential equation

conditions of the form:

𝑦(𝑎) = 𝑦0 , 𝑦 ′ (𝑎) = 𝑦1 , . . . , 𝑦 (𝑛−1) (𝑎) = 𝑦(𝑛−1)

are called initial conditions (IC).

Example 1.2.8. Solve the IVP 𝑦 ′′ + 𝑦 = 0, 𝑦(0) = 3 and 𝑦 ′ (0) = −4.

𝑦(𝑥) = 𝑐1 cos 𝑥 + 𝑐2 sin 𝑥,

since 𝑦(𝑥) satisfy the Diﬀerential Equation 𝑦 ′′ + 𝑦 = 0, it is a general solution. Now

𝑦(𝑥) = 3 cos 𝑥 + 5 sin 𝑥.

∙ Does a solution of the problem exist?

∙ If a solution exists, is it unique?

Theorem 1.2.5 (Existence).

𝑦 ′ = 𝑓 (𝑥, 𝑦), with 𝑦(𝑎) = 𝑏 (1.8)

satisﬁes the diﬀerential equation as well as the initial condition.

Theorem 1.2.6 (Uniqueness).

1.3 Exact Diﬀerential Equations

Recall that the total diﬀerential of a function 𝐹 (𝑥, 𝑦) of two variables is

for all (x,y) in the domain of 𝐹 .