Documente Academic
Documente Profesional
Documente Cultură
Department of Mathematics
Addis Ababa University
January 2, 2012
2
⃝January
c 2, 2012 Semu & Tilahun Draft
Table of Contents
Table of Contents 3
⃝January
c 2, 2012 Semu & Tilahun Draft
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
⃝January
c 2, 2012 Semu & Tilahun Draft
Part I
⃝January
c 2, 2012 Semu & Tilahun Draft
Chapter 1
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.
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
c 2, 2012 Semu & Tilahun Draft
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
In this part we will only consider the case of ordinary differential equations.
⃝January
c 2, 2012 Semu & Tilahun Draft
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)
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).
𝑦 ′′ = 𝑦 ′ − 6𝑦.
⃝January
c 2, 2012 Semu & Tilahun Draft
10 Ordinary Differential Equations of the First Order
Classification by Linearity
where 𝑎𝑛 (𝑥) ∕= 0.
If 𝑏(𝑥) ≡ 0, the equation (1.5) is called a homogeneous DE and otherwise it is called
nonhomogeneous.
Notation
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),
⃝January
c 2, 2012 Semu & Tilahun Draft
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.
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.
Example 1.2.1.
⃝January
c 2, 2012 Semu & Tilahun Draft
12 Ordinary Differential Equations of the First Order
∫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.)
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
Solution:
gives
3𝑦 2 + 2𝑥2 = 𝐶,
⃝January
c 2, 2012 Semu & Tilahun Draft
1.2 Separable Differential Equations 13
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.
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
𝑦 ′ = 𝑔(𝑎𝑥 + 𝑏𝑦 + 𝑐) (1.7)
⃝January
c 2, 2012 Semu & Tilahun Draft
14 Ordinary Differential Equations of the First Order
Thus (1.7) will be transformed into
1 𝑑𝑢 𝑎
− = 𝑔(𝑢),
𝑏 𝑑𝑥 𝑏
where 𝑢 and 𝑥 can be separated.
Solution:
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.
Solution
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
c 2, 2012 Semu & Tilahun Draft
1.2 Separable Differential Equations 15
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
𝑑𝑥 𝑑𝑢
= .
𝑥 𝑔(𝑢) − 𝑢
Solution
⃝January
c 2, 2012 Semu & Tilahun Draft
16 Ordinary Differential Equations of the First Order
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∣).
Solution:
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
c 2, 2012 Semu & Tilahun Draft
1.2 Separable Differential Equations 17
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.
𝐹 (𝑥, 𝑦, 𝑦 ′ , 𝑦 ′′ , . . . , 𝑦 (𝑛) ) = 0,
Remark 1.2.3. The number of initial conditions necessary to determine a unique solution
equals the order of the differential equation.
Solution:
First find the general solution with two unknown constants. Given
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
c 2, 2012 Semu & Tilahun Draft
18 Ordinary Differential Equations of the First Order
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
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:
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.
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
c 2, 2012 Semu & Tilahun Draft
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 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.
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”.
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
c 2, 2012 Semu & Tilahun Draft
20 Ordinary Differential Equations of the First Order
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 .
∂𝐹 ∂𝐹
𝑑𝐹 (𝑥, 𝑦) = 𝑑𝑥 + 𝑑𝑦,
∂𝑥 ∂𝑦
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.
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
∂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
⃝January
c 2, 2012 Semu & Tilahun Draft
1.3 Exact Differential Equations 21
1. How can we determine the existence of such a function 𝐹 (𝑥, 𝑦) ?
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
every where in 𝑅.
𝑑𝑦 2𝑥𝑦 3 + 2
=− 2 2 .
𝑑𝑥 3𝑥 𝑦 + 8𝑒4𝑦
Then write
(2𝑥𝑦 3 + 2)𝑑𝑥 + (3𝑥2 𝑦 2 + 8𝑒4𝑦 )𝑑𝑦 = 0.
∂𝑀 ∂𝑁
= 6𝑥𝑦 = .
∂𝑦 ∂𝑥
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
c 2, 2012 Semu & Tilahun Draft
22 Ordinary Differential Equations of the First Order
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)
∫
∂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
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
c 2, 2012 Semu & Tilahun Draft
1.3 Exact Differential Equations 23
Therefore, 𝐹 (𝑥, 𝑦) = 𝑥 sin 𝑦 − 𝑦 2 + 𝐵 = constant, which implies
𝑥 sin 𝑦 − 𝑦 2 = 𝐶
Solution
∂𝑀 ∂𝑁
= 6𝑥𝑦 = .
∂𝑦 ∂𝑥
∂𝐹
= 3𝑥2 𝑦 + 𝐴′ (𝑦) = 𝑁 = 3𝑥2 𝑦 + 𝑦 3 ,
∂𝑦
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
c 2, 2012 Semu & Tilahun Draft
24 Ordinary Differential Equations of the First Order
Step 3: Checking
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.
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
is exact, then the multiplicative function 𝜇(𝑥, 𝑦) is called an integrating factor of the
DE.
⃝January
c 2, 2012 Semu & Tilahun Draft
1.3 Exact Differential Equations 25
Example 1.3.5. Consider the differential equation
∂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
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.
∂ ∂
(𝜇𝑀 ) = (𝜇𝑁 ) (1.11)
∂𝑦 ∂𝑥
𝜇𝑦 𝑀 + 𝜇𝑀𝑦 = 𝜇𝑥 𝑁 + 𝜇𝑁𝑥 .
𝑑𝜇 𝑑𝜇 𝑀𝑦 − 𝑁𝑥
𝜇𝑀𝑦 = 𝑁 + 𝜇𝑁𝑥 or equivalently = 𝜇( ) (1.12)
𝑑𝑥 𝑑𝑥 𝑁
⃝January
c 2, 2012 Semu & Tilahun Draft
26 Ordinary Differential Equations of the First Order
which is a separable differential equation.
This idea works correctly if the ratio
𝑀𝑦 − 𝑁𝑥
𝑁
is a function of 𝑥 only, that is,
𝑀𝑦 − 𝑁𝑥
𝑝(𝑥) = = a function of 𝑥.
𝑁
In this case ( )
𝑑𝜇 𝑀𝑦 − 𝑁𝑥
= 𝑑𝑥,
𝜇 𝑁
which implies
∫
𝑝(𝑥)𝑑𝑥
𝜇(𝑥) = 𝑒 .
𝑀𝑦 − 𝑁𝑥
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
𝑀𝑦 − 𝑁𝑥
𝑀
is a function of 𝑦 alone, then
∫
𝜇(𝑦) = 𝑒− 𝑞(𝑦)𝑑𝑦
.
⃝January
c 2, 2012 Semu & Tilahun Draft
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
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
⃝January
c 2, 2012 Semu & Tilahun Draft
28 Ordinary Differential Equations of the First Order
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,
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
c 2, 2012 Semu & Tilahun Draft
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
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)
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).
⃝January
c 2, 2012 Semu & Tilahun Draft
30 Ordinary Differential Equations of the First Order
∫
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 .
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.
⃝January
c 2, 2012 Semu & Tilahun Draft
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.
𝑦 ′ + 𝑝(𝑥)𝑦 = 𝑞(𝑥)𝑦 𝛼 ,
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
c 2, 2012 Semu & Tilahun Draft
32 Ordinary Differential Equations of the First Order
Solution
𝑑𝑢
= 𝑑𝑥
𝐵 − 𝐴𝑢
and we integrate ∫ ∫
𝑑𝑢
= 𝑑𝑥
𝐵 − 𝐴𝑢
and get 𝑢 = 𝐵
𝐴
+ 𝐶𝑒−𝐴𝑥 .
1 1
𝑦= = 𝐵
.
𝑢 𝐴
+ 𝐶𝑒−𝐴𝑥
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.
1 2 1
𝑦′ = 𝑦 − 𝑦 + 1.
𝑥2 𝑥
⃝January
c 2, 2012 Semu & Tilahun Draft
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
𝑦 = 𝑥𝑦 ′ + 𝑔(𝑦 ′ ),
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.
⃝January
c 2, 2012 Semu & Tilahun Draft
34 Ordinary Differential Equations of the First Order
Example 1.5.3. Solve the Clairuat equation
1
𝑦 = 𝑥𝑦 ′ + .
𝑦′
Solution
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.
1.6 Exercises
1. 𝑥𝑦 ′ + 𝑦 = 6𝑥2
⃝January
c 2, 2012 Semu & Tilahun Draft
1.6 Exercises 35
This page is left blank intensionally.
⃝January
c 2, 2012 Semu & Tilahun Draft
36 Ordinary Differential Equations of the First Order
⃝January
c 2, 2012 Semu & Tilahun Draft
Chapter 2
In this section, we will focus on the general theory of linear ordinary differential equations
before we start to discuss about solving such problems.
⃝January
c 2, 2012 Semu & Tilahun Draft
38 Ordinary Differential Equations of The Second and Higher Order
If 𝑓 (𝑥) ≡ 0, then (2.1) is reduced to:
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,
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
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
c 2, 2012 Semu & Tilahun Draft
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.
1. The functions 𝑓1 (𝑥) = 𝑒𝑥 and 𝑓2 (𝑥) = 4𝑒𝑥 are Linearly Dependent on R since
2. The functions
𝑓1 (𝑥) = 𝑒𝑥 , 𝑓2 (𝑥) = 𝑒−𝑥 , 𝑓3 (𝑥) = sinh 𝑥
3. The two functions 𝑓1 (𝑥) = 𝑥 and 𝑓2 (𝑥) = 𝑥3 are Linearly Independent on R, since
for 𝑐1 , 𝑐2 ∈ R,
implies 𝑐1 = 0 and 𝑐2 = 0.
⃝January
c 2, 2012 Semu & Tilahun Draft
40 Ordinary Differential Equations of The Second and Higher Order
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
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
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.
𝑦 ′′′ − 2𝑦 ′′ − 𝑦 ′ + 2𝑦 = 0.
⃝January
c 2, 2012 Semu & Tilahun Draft
2.2 General Solution of Homogeneous Linear ODEs 41
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
⃝January
c 2, 2012 Semu & Tilahun Draft
42 Ordinary Differential Equations of The Second and Higher Order
Solution
which is equal to
= −2𝑒2𝑥 ∕= 0, ∀𝑥 ∈ R.
𝑒𝑥 − 𝑒−𝑥
𝑓3 (𝑥) = sinh 𝑥 = .
2
Solution
⃝January
c 2, 2012 Semu & Tilahun Draft
2.2 General Solution of Homogeneous Linear ODEs 43
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.
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.
𝑦2 (𝑥)
= 𝑢(𝑥)
𝑦1 (𝑥)
⃝January
c 2, 2012 Semu & Tilahun Draft
44 Ordinary Differential Equations of The Second and Higher Order
and simplifying this gives us
which implies
𝑢′′ 𝑦1 + 𝑢′ (2𝑦1′ + 𝑝(𝑥)𝑦1 ) = 0.
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
Solution
⃝January
c 2, 2012 Semu & Tilahun Draft
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. □
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
⃝January
c 2, 2012 Semu & Tilahun Draft
46 Ordinary Differential Equations of The Second and Higher Order
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).
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 .
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
c 2, 2012 Semu & Tilahun Draft
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 .
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
2. If further, the remaining roots are the distinct real roots 𝜆𝑘+1 , 𝜆𝑘+2 , . . . , 𝜆𝑛 , the
general solution of (2.6) will be:
⃝January
c 2, 2012 Semu & Tilahun Draft
48 Ordinary Differential Equations of The Second and Higher Order
Example 2.3.3.
𝑦 (4) − 5𝑦 ′′′ + 6𝑦 ′′ + 4𝑦 ′ − 8𝑦 = 0.
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
⃝January
c 2, 2012 Semu & Tilahun Draft
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
2.3.1 Exercises
1. 𝑦 ′′ + 𝑦 = 0.
2. 𝑦 ′′ − 6𝑦 ′ + 25𝑦 = 0.
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
⃝January
c 2, 2012 Semu & Tilahun Draft
50 Ordinary Differential Equations of The Second and Higher Order
where 𝑏𝑛 , ..., 𝑏0 are constants, which is called a nonhomogeneous differential equation with
constant coefficients. The following theorem is very important in such cases.
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).
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
c 2, 2012 Semu & Tilahun Draft
2.4 Nonhomogeneous Equations with Constant Coefficients 51
2.4.1 The undetermined coefficient method
Definition 2.4.4.
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
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.
Therefore, 𝑆 = {1, 𝑥, 𝑥2 , 𝑥3 }.
⃝January
c 2, 2012 Semu & Tilahun Draft
52 Ordinary Differential Equations of The Second and Higher Order
3. Let 𝑔(𝑥) = 2𝑥𝑒−𝑥 . 𝑔 is an UC function (as a product of UC function).
4. The function
1
𝑓 (𝑥) =
𝑥
is not a UC function.
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 (non- homogeneous term) is a combination of 𝑥2 and sin 2𝑥.
𝑓1 (𝑥) = 3𝑥2
⃝January
c 2, 2012 Semu & Tilahun Draft
2.4 Nonhomogeneous Equations with Constant Coefficients 53
∙ 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.
∙ 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 .
−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,
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
c 2, 2012 Semu & Tilahun Draft
54 Ordinary Differential Equations of The Second and Higher Order
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
1 1
𝑦(𝑥) = 𝑦ℎ (𝑥)+𝑦𝑝1 (𝑥)+𝑦𝑝2 (𝑥) = 𝐶1 +𝐶2 𝑥+𝐶3 𝑒𝑥 +𝐶4 𝑒−𝑥 − 𝑥4 −3𝑥2 − sin 2𝑥.
4 20
∙ Particular solution corresponding to 𝑓1 (𝑥) = 2𝑒−𝑥 : 𝑦𝑝1 (𝑥) = 𝐵𝑒−𝑥 which duplicates
with 𝐶2 𝑒−𝑥 .
1
𝑦𝑝1 (𝑥) = − 𝑥𝑒−𝑥 .
2
(−𝐷 sin 𝑥 − 𝐸 cos 𝑥) − 2(𝐷 cos 𝑥 − 𝐸 sin 𝑥) − 3(𝐷 sin 𝑥 + 𝐸 cos 𝑥) = −10 sin 𝑥.
⃝January
c 2, 2012 Semu & Tilahun Draft
2.4 Nonhomogeneous Equations with Constant Coefficients 55
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
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.
where 𝑏1 , 𝑏2 and 𝑓 are continuous functions. Suppose that the general solution for the
homogeneous part of (2.12) is
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
⃝January
c 2, 2012 Semu & Tilahun Draft
56 Ordinary Differential Equations of The Second and Higher Order
We differentiate and substitute it in (2.12) to get
𝑐′1 𝑦1 + 𝑐′2 𝑦2 = 0.
′′
𝑦𝑝′′ = 𝑐1 𝑦1′′ + 𝑐′1 𝑦1′ + 𝑐′2 𝑦2′ + 𝑐2 𝑦2
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.
0 𝑦
2
𝑓 𝑦2′
𝑊1 (𝑥)
c′1 (x) = =
𝑊 (𝑥) 𝑊 (𝑥)
⃝January
c 2, 2012 Semu & Tilahun Draft
2.4 Nonhomogeneous Equations with Constant Coefficients 57
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
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
c 2, 2012 Semu & Tilahun Draft
58 Ordinary Differential Equations of The Second and Higher Order
2.5 The Laplace Transform Method to Solve ODEs
In the previous sections we have discussed how to solve differential equations of the form:
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.
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
c 2, 2012 Semu & Tilahun Draft
2.5 The Laplace Transform Method to Solve ODEs 59
Table of some basic Laplace Transforms
Properties
Here below we state some important properties of the transform in a serious of theorems
without proof.
(a) If 𝑢(𝑡), 𝑣(𝑡) are functions and 𝛼, 𝛽 are any constants, then
(b) For any functions 𝑈 (𝑠), 𝑉 (𝑠) and any given scalars 𝛼, 𝛽, we have
⃝January
c 2, 2012 Semu & Tilahun Draft
60 Ordinary Differential Equations of The Second and Higher Order
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
⇒ 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
c 2, 2012 Semu & Tilahun Draft
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
⃝January
c 2, 2012 Semu & Tilahun Draft
62 Ordinary Differential Equations of The Second and Higher Order
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
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
∫ ∞
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
In general we have
⃝January
c 2, 2012 Semu & Tilahun Draft
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
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.
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.
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:
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.
⃝January
c 2, 2012 Semu & Tilahun Draft
64 Ordinary Differential Equations of The Second and Higher Order
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
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.
1. 𝑥2 𝑦 ′′ − 2𝑥𝑦 ′ + 2𝑦 = 0.
3. 𝑥2 𝑦 ′′ − 2𝑥𝑦 ′ + 2𝑦 = 𝑥3 .
Solution
⃝January
c 2, 2012 Semu & Tilahun Draft
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.
𝑦(𝑥) = 𝑐1 𝑥 + 𝑐2 𝑥2 ,
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:
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
c 2, 2012 Semu & Tilahun Draft
66 Ordinary Differential Equations of The Second and Higher Order
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
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
c 2, 2012 Semu & Tilahun Draft
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 .
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
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 .
⃝January
c 2, 2012 Semu & Tilahun Draft
68 Ordinary Differential Equations of The Second and Higher Order
Solution
∙ 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).
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,
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
c 2, 2012 Semu & Tilahun Draft
2.7 *The Power Series Solution Method 69
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
c 2, 2012 Semu & Tilahun Draft
70 Ordinary Differential Equations of The Second and Higher Order
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
⃝January
c 2, 2012 Semu & Tilahun Draft
2.7 *The Power Series Solution Method 71
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.
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
c 2, 2012 Semu & Tilahun Draft
72 Ordinary Differential Equations of The Second and Higher Order
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)
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.
Solution
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
c 2, 2012 Semu & Tilahun Draft
2.7 *The Power Series Solution Method 73
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
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.
⃝January
c 2, 2012 Semu & Tilahun Draft
74 Ordinary Differential Equations of The Second and Higher Order
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
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
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
c 2, 2012 Semu & Tilahun Draft
2.8 Systems of ODE of the First Order 75
2.8 Systems of ODE of the First Order
which is called the normal form. In vector form this system becomes:
X′ = AX + F(𝑡),
where
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.
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
c 2, 2012 Semu & Tilahun Draft
76 Ordinary Differential Equations of The Second and Higher Order
Let y = x𝑒𝜆𝑡 , where x = (𝑥1 , 𝑥2 , . . . , 𝑥𝑛 )𝑇 . Substituting this into (2.27) we get:
This implies,
Ax = 𝜆x,
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:
Let y(𝑡) = x𝑒𝜆𝑡 . Then the corresponding eigenvalue problem will be:
( )( ) ( )
−3 1 𝑥1 𝑥1
Ax = 𝜆x ⇐⇒ =𝜆
1 −3 𝑥2 𝑥2
⃝January
c 2, 2012 Semu & Tilahun Draft
2.8 Systems of ODE of the First Order 77
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
⃝January
c 2, 2012 Semu & Tilahun Draft
78 Ordinary Differential Equations of The Second and Higher Order
Let y(𝑡) = x𝑒𝜆𝑡 , y, x ∈ R3 . Then corresponding eigenvalue problem is AX = 𝜆X
with characteristic equation
A − 𝜆I = 0
⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞
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
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
c 2, 2012 Semu & Tilahun Draft
2.8 Systems of ODE of the First Order 79
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
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
c 2, 2012 Semu & Tilahun Draft
80 Ordinary Differential Equations of The Second and Higher Order
b) Here we have
( )
𝑝1
F(𝑡) = p𝑒2𝑡 = 𝑒2𝑡 .
𝑝2
This implies
( ) ( ) ( ) ( )
2 2𝑝1 𝑝1 2
2p𝑒2𝑡 = Ap𝑒2𝑡 + 𝑒2𝑡 ⇐⇒ = +
1 2𝑝2 6𝑝1 − 𝑝2 1
Then we get
( ) ( ) ( )
𝑝1 𝑞1 𝑟1
y𝑝1 = sin 𝑡 + cos 𝑡, 𝑎𝑛𝑑 𝑦 𝑝2 = 𝑒𝑡
𝑝2 𝑞2 𝑟2
Example 2.8.3.
⃝January
c 2, 2012 Semu & Tilahun Draft
2.8 Systems of ODE of the First Order 81
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.
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].
⃝January
c 2, 2012 Semu & Tilahun Draft
82 Ordinary Differential Equations of The Second and Higher Order
Basic Properties
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.
Solution
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
3
𝐴 = ,𝐵 = 0
4
Therefore, 𝑦1 = 𝐶1 cos 2𝑥 + 𝐶2 sin 2𝑥 + 34 𝑥 and from 4𝑦1 + 𝐷𝑦2 = 𝑥 we have:
⃝January
c 2, 2012 Semu & Tilahun Draft
2.8 Systems of ODE of the First Order 83
By integrating both sides we get
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.
and
𝐷 − 2 2 − 4𝑒2𝑥
2𝐷 − 3 0 (2𝐷 − 3)(2 − 4𝑒2𝑥 )
𝑦2 = =
𝐷−2
2𝐷 (−𝐷2 − 𝐷 + 2)
2𝐷 − 3 3𝐷 − 1
⃝January
c 2, 2012 Semu & Tilahun Draft
84 Ordinary Differential Equations of The Second and Higher Order
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
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:
𝑥1 (𝑡) = 𝑦(𝑡)
𝑥2 (𝑡) = 𝑦 ′ (𝑡)
𝑥3 (𝑡) = 𝑦 ′′ (𝑡)
.. .
. = ..
𝑥𝑛 (𝑡) = 𝑦 (𝑛−1) (𝑡)
⃝January
c 2, 2012 Semu & Tilahun Draft
2.9 Exercises 85
Or in matrix notation:
𝑋 ′ = 𝐴𝑋,
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
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
c 2, 2012 Semu & Tilahun Draft
86 Ordinary Differential Equations of The Second and Higher Order
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.
△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
c 2, 2012 Semu & Tilahun Draft
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.
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
c 2, 2012 Semu & Tilahun Draft
88 Ordinary Differential Equations of The Second and Higher Order
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 )
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
c 2, 2012 Semu & Tilahun Draft
2.11 Exercises 89
This page is left blank intensionally.
⃝January
c 2, 2012 Semu & Tilahun Draft
90 Ordinary Differential Equations of The Second and Higher Order
⃝January
c 2, 2012 Semu & Tilahun Draft
Chapter 3
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.
⃝January
c 2, 2012 Semu & Tilahun Draft
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).
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
c 2, 2012 Semu & Tilahun Draft
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.
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,
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.
𝑑𝑥
= 𝑃 (𝑥, 𝑦)
𝑑𝑡
𝑑𝑦
= 𝑄(𝑥, 𝑦) (3.3)
𝑑𝑡
⃝January
c 2, 2012 Semu & Tilahun Draft
94 *Nonlinear ODEs and Qualitative Analysis
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
(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 → ∞
Singular points of a system can also be classified further depending on the style of motions
of the trajectories around it.
⃝January
c 2, 2012 Semu & Tilahun Draft
3.2 Critical Points and Stability 95
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.
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
c 2, 2012 Semu & Tilahun Draft
96 *Nonlinear ODEs and Qualitative Analysis
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.
⃝January
c 2, 2012 Semu & Tilahun Draft
3.2 Critical Points and Stability 97
and the general solution of the linear system when 𝑡 increases, let us consider the limit,
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
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
Then
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
c 2, 2012 Semu & Tilahun Draft
98 *Nonlinear ODEs and Qualitative Analysis
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.
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
c 2, 2012 Semu & Tilahun Draft
3.2 Critical Points and Stability 99
The first order Taylor series approximation of the two functions gives us,
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)
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
c 2, 2012 Semu & Tilahun Draft
100 *Nonlinear ODEs and Qualitative Analysis
3. If 𝜆1 and 𝜆2 are real and equal, then the critical point (0, 0) of the linear system
(3.6) is a 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.
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.
⃝January
c 2, 2012 Semu & Tilahun Draft
3.2 Critical Points and Stability 101
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.
⃝January
c 2, 2012 Semu & Tilahun Draft
102 *Nonlinear ODEs and Qualitative Analysis
occur in a study of interacting populations. Find the equilibrium points and determine
their nature.
Solution
Then the eigenvalues for the coefficient matrix are solutions of the characteristics
equation, 𝜆2 + 1 = 0, which are 𝜆1,2 = ±𝑖, – pure imaginary.
⃝January
c 2, 2012 Semu & Tilahun Draft
3.2 Critical Points and Stability 103
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
c 2, 2012 Semu & Tilahun Draft
104 *Nonlinear ODEs and Qualitative Analysis
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
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
c 2, 2012 Semu & Tilahun Draft
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. □
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:
⃝January
c 2, 2012 Semu & Tilahun Draft
106 *Nonlinear ODEs and Qualitative Analysis
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,
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
c 2, 2012 Semu & Tilahun Draft
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.
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.
ẋ = 𝐹 (x) (3.9)
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.
⃝January
c 2, 2012 Semu & Tilahun Draft
108 *Nonlinear ODEs and Qualitative Analysis
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),
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
Find the equilibrium points (with respect to 𝑘) and determine their nature.
⃝January
c 2, 2012 Semu & Tilahun Draft