Arindama Singh

MA2020 Classnotes
Differential Equations

© A.Singh
Contents

Syllabus v

1 First Order ODE 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Variables Separable . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Reducible to variables separable . . . . . . . . . . . . . . . . . 7
1.4 Exact Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.5 Integrating factors . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.6 Linear equations . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2 Second Order ODE 22

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2 The Wronskian . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3 Constant coefficients . . . . . . . . . . . . . . . . . . . . . . . 27
2.4 Euler-Cauchy Equation . . . . . . . . . . . . . . . . . . . . . . 33
2.5 Reduction of order . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.6 Non-homogeneous second order linear ODE . . . . . . . . . . . 38
2.7 Variation of parameters . . . . . . . . . . . . . . . . . . . . . . 39
2.8 Method of undetermined coefficients . . . . . . . . . . . . . . . 42

3 Series Solutions 47
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.2 Regular and singular points . . . . . . . . . . . . . . . . . . . . 48
3.3 Power series solution at an ordinary point . . . . . . . . . . . . 51
3.4 Series solution about a regular singular point . . . . . . . . . . . 55

4 Special Functions 63
4.1 Legendre polynomials . . . . . . . . . . . . . . . . . . . . . . . 63
4.2 Bessel Functions . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.3 Properties of 𝐽𝜈 and 𝐽𝑛 . . . . . . . . . . . . . . . . . . . . . . . 76
4.4 Sturm-Liouville problems . . . . . . . . . . . . . . . . . . . . . 78
4.5 Orthogonality . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

5 Partial Differential Equations 90

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.2 Lagrange method . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.3 Second order linear PDEs . . . . . . . . . . . . . . . . . . . . . 97

iii
iv
6 Separation of Variables 110
6.1 Modeling wave . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.2 D’ Alembert’s solution of wave equation . . . . . . . . . . . . . 111
6.3 Series solution of the wave equation . . . . . . . . . . . . . . . 113
6.4 One-dimensional heat flow . . . . . . . . . . . . . . . . . . . . 118
6.5 Laplace equation . . . . . . . . . . . . . . . . . . . . . . . . . 122

Index 129
Syllabus
Geometrical meaning of a first order differential equation: K 1.2
Separable differential equations: K 1.3
Exact differential equations, Integrating factors: K 1.4
Linear differential equation and Bernoulli differential equation: K 1.5
Linear differential equations of second and higher order Homogeneous linear equa-
tions of second order: K 2.1
Second order homogeneous equations with constant coefficients: K 2.2
Euler-Cauchy equation: K 2.5
Existence and uniqueness theory, Wronskian: K 2.6
Nonhomogeneous equations: K 2.7
Method of variation of parameters: K 2.10
Examples on Linear ODE of higher order: K 3.1-3.3
Power series method: K 5.1
Legendre’s equation, Legendre Polynomials: K 5.2
Rodrigue’s formula, Generating function, Recursion formula, Orthogonality and
Legendre series: K 5.2 (Problems 12,14)
Frobenius method : K 5.3, J 8.4
Bessel’s equation, Bessel’s function of first kind: K 5.4
Bessel’ s function of second kind, General solution of Bessel’ s equation: K 5.5
Generating function for Bessel’s functions of first kind: J 8.7
Sturm-Liouville Problems : J 8.10, K 11.5
Orthogonality of Bessel’s functions and Bessel expansion theorem: K 11.5-11.6
First order partial differential equations [Lagrange’s Method only] : J 18.2
Classification of second order partial differential equations : J 18.5-18.6
Wave equation- D’Alembert’s solution, Semi-infinite string problem, Method of
separation of variables for finite interval: K 12.2-12.4
Heat equation: K 12.6
Laplace equation in Cartesian and polar coordinates : K 12.10
Heat flow in rectangular and circular plates: J 18.10
Texts:
J. Alan Jeffrey, Advanced Engineering Mathematics, Harcourt/Academic Press,
2002.
K. Erwin Kreyszig, Advanced Engineering Mathematics, 10th Ed., John Wiley
& Sons, Inc. 2011.
References:
1. W.E. Boyce and R.C. DiPrima, Elementary Differential Equations, 7th Ed.,
John Wiely & Sons, 2002.

v
vi
2. S.J. Farlow, Partial Differential Equations for scientists and Engineers, Dover,
2006.
3. N. Piskunov, Differential and Integral Calculus Vol. 1-2, Mir Publishers,
1974.
4. G.F. Simmons, Differential Equations, Tata McGraw-Hill publishing com-
pany, 1974.
1
First Order ODE

1.1 Introduction
A differential equation is an equation that involves derivatives of a variable that
depends on other independent variables. For instance,

𝑑𝑦 𝑑 3𝑦  𝑑𝑦  4
= 3𝑥 2 sin(𝑥 + 𝑦), +2 −𝑦 = 0
𝑑𝑥 𝑑𝑥 3 𝑑𝑥
are differential equations, where the variable 𝑦 is supposed to be the variable that
depends on the independent variable 𝑥. When the equation involves only one
independent variable, the equation is said to be an ordinary differential equation,
an ODE.
The order of an ODE is the order of the highest derivative of the dependent
variable. In the above equations, the first one is of first order and the second one is
of third order.
A solution of an ODE is a function which when replaces the dependent variable,
it is seen that the equation is satisfied. If the dependent variable is 𝑦 and the
independent variable is 𝑥 in an ODE of order 𝑘, then a solution of such an equation
is 𝑦 = 𝑦 (𝑥) which is 𝑘 times differentiable and which satisfies the given equation.
For example, 𝑦 (𝑥) = 2 sin 𝑥 − 31 cos(2𝑥) is a solution of the ODE

𝑑 2𝑦
+ 𝑦 = cos(2𝑥).
𝑑𝑥 2
(We also write 𝑦 0 for 𝑑𝑦/𝑑𝑥, 𝑦 (𝑛) for 𝑑 𝑛𝑦/𝑑𝑥 𝑛 etc.) This claim is verified as follows:
1 2
𝑦 = 2 sin 𝑥 − cos(2𝑥) ⇒ 𝑦 0 = 2 cos 𝑥 + sin(2𝑥)
3 3
4
⇒ 𝑦 00 = −2 sin 𝑥 + cos(2𝑥) = −𝑦 + cos(2𝑥).
3
Often an ODE comes with the restriction that the independent variable varies in a
particular subset of R. In that case, the domain of the dependent variable is assumed
to be that subset. For example, in the ODE

𝑥𝑦 0 + 𝑦 = 0, 𝑥 ≠0

1
2 MA2020 Classnotes

it is assumed that the domain of 𝑦 = 𝑦 (𝑥) is R \ {0}. In this case, the function
𝑦 (𝑥) = 1/𝑥 is a solution. Reason:
𝑑 (1/𝑥) 1  1 1
𝑦 = 1/𝑥 ⇒ 𝑥𝑦 0 + 𝑦 = 𝑥 + = 𝑥 − 2 + = 0 for 𝑥 ≠ 0.
𝑑𝑥 𝑥 𝑥 𝑥
It is easy to see that 𝑦 (𝑥) = 𝑐/𝑥 for any 𝑐 ∈ R is also a solution. In such a case, we
say that 𝑐 is an arbitrary constant.
Are there other types of solutions to this equation? Well, suppose, 𝑦 (𝑥) is a
solution of 𝑥𝑦 0 + 𝑦 = 0. Write 𝑧 (𝑥) = 𝑥𝑦. Then
𝑑𝑧
= 𝑥𝑦 0 + 1 · 𝑦 = 0 ⇒ 𝑧 (𝑥) = 𝑐 ⇒ 𝑥𝑦 = 𝑐 ⇒ 𝑦 = 𝑐/𝑥 .
𝑑𝑥
That is, any solution of 𝑥𝑦 0 + 𝑦 = 0 is in the form 𝑦 = 𝑐/𝑥 for 𝑥 ≠ 0.
Observe that a solution of an ODE need not be unique. However, if we have
another condition on the function 𝑦 (𝑥) such as 𝑦 (1) = 1, then substituting 𝑥 = 1 in
our solution 𝑦 = 𝑐/𝑥, we have

1 = 𝑦 (1) = 𝑐/1 = 𝑐.

We thus obtain the unique solution 𝑦 = 1/𝑥. The condition 𝑦 (1) = 1 is called an
initial condition for the ODE. In fact, when a condition on the dependent variable
is given by prescribing its value at a single point, it is called an initial condition. An
ODE with a given initial condition is called an initial value problem, an IVP.
It follows that 𝑦 = 𝑐/𝑥 = 1/𝑥 is the only solution to the initial value problem

𝑥𝑦 0 + 𝑦 = 0, 𝑦 (1) = 1, 𝑥 ≠ 0.

A general first order ODE may be given by an equation using 𝑥, 𝑦, 𝑦 0, which would
then look like
ℎ(𝑥, 𝑦, 𝑦 0) = 0
for some specific expression ℎ(·, ·, ·). For simplicity, we may only consider equations
which can be solved for 𝑦 0; that is, an ODE in the form:

𝑦 0 = 𝑔(𝑥, 𝑦)

with a given domain, a subset of R, where 𝑥 varies. Geometrically, consider the

𝑥𝑦-plane. At a particular point (with an admissible 𝑥-value), say (𝑥 0, 𝑦0 ), the ODE
gives the value of 𝑦 0. That is, 𝑦 0 (𝑥 0 ) = 𝑔(𝑥 0, 𝑦0 ); it is a number which represents the
slope of the tangent to 𝑦 = 𝑦 (𝑥) at 𝑥 = 𝑥 0 . By varying 𝑥 throughout its domain and
with all possible values of 𝑦, the ODE prescribes slopes at each admissible point.
The set of all these slopes is called the direction field for the ODE.
By joining these slopes geometrically we may get many solution curves 𝑦 = 𝑦 (𝑥)
to the ODE. In general, we accept continuous curves in the 𝑥𝑦-plane as solutions to
First Order ODE 3
ODEs rather than functions 𝑦 = 𝑦 (𝑥). Once an initial value is prescribed, whenever
there exists a unique solution, we would obtain only one solution curve that passes
through the point (𝑥 0, 𝑦0 ).
The direction field for the ODE 𝑦 0 = 𝑥 + 𝑦 is plotted in the following figure. Also
plotted are three approximate solution curves passing through the points (0, 1),
(0, 0) and (0, −1), respectively.

It is a fact that even all initial value problems do not have unique solutions.
Anything can happen. There are IVPs having no solutions, having more than one
solutions, and there are IVPs having a unique solution. We will use the following
result without proof.

(1.1) Theorem (Existence-Uniqueness)

𝜕𝑔
Let 𝑔(𝑥, 𝑦) and be continuous in the rectangle 𝑅 : 𝑥 0 ≤ 𝑥 ≤ 𝑥 0 +𝑎, |𝑦 −𝑦0 | ≤ 𝑏.
𝜕𝑥
Compute 𝑀 = max{|𝑔(𝑥, 𝑦)| : (𝑥, 𝑦) ∈ 𝑅} and 𝛼 = min{𝑎, 𝑏/𝑀 }. Then, the IVP
𝑦 0 = 𝑔(𝑥, 𝑦), 𝑦 (𝑥 0 ) = 𝑦0 has a unique solution in the interval 𝑥 0 ≤ 𝑥 ≤ 𝑥 0 + 𝛼.

(1.2) Example
Consider the IVP: 𝑦 0 = sin(2𝑥)𝑦 1/3, 𝑦 (0) = 0.
(𝑥) = 0, the zero function.
It has a solution as 𝑦p
Verify that 𝑦 (𝑥) = ± 8/27 sin3 𝑥 are solutions of the same IVP.
Notice that 𝑓 (𝑥, 𝑦) = sin(2𝑥)𝑦 1/3 has no partial derivative at 𝑦 = 0.

1.2 Variables Separable

Sadly, all ODEs of the form 𝑦 0 = 𝑔(𝑥, 𝑦) cannot be solved since it would ask us to
integrate 𝑔(𝑥, 𝑦) with respect to 𝑥, where 𝑦 is an unknown function of 𝑥. A simpler
4 MA2020 Classnotes

case, which we may think of solving is when 𝑔(𝑥, 𝑦) is a function of 𝑥 alone. So,
we consider an ODE in the form

𝑦 0 = 𝑓 (𝑥).

Of course, we cannot even solve all equations in this form. For instance, we do not
know how to solve
2
𝑦0 = 𝑒 𝑥
since our data base for integrating algebraic expressions does not include such a
function. Knowing this fact very well, we will attempt solving first order ODEs; in
fact, whichever we can. In general, if we know how to integrate the function 𝑓 (𝑥),
we can get a solution of the ODE. In fact,
∫
0
𝑦 = 𝑓 (𝑥) ⇒ 𝑦 (𝑥) = 𝑓 (𝑥) 𝑑𝑥 .

For example, the ODE 𝑦 0 = 𝑥 𝑟 for 𝑟 > −1 may be solved by taking

𝑥 𝑟 +1
∫
𝑦= 𝑥 𝑟 𝑑𝑥 = +𝐶 for an arbitrary constant 𝐶.
𝑟 +1

We can slightly generalize this method to solve most ODEs in the form

𝑔(𝑦)𝑦 0 = 𝑓 (𝑥) (1.2.1)

by using the differentials. Recall that 𝑑𝑦 = 𝑦 0 𝑑𝑥. Using this, we obtain

∫ ∫ ∫
0 0
𝑔(𝑦)𝑦 = 𝑓 (𝑥) ⇒ 𝑔(𝑦) 𝑑𝑦 = 𝑔(𝑦)𝑦 𝑑𝑥 = 𝑓 (𝑥) 𝑑𝑥 .

This amounts to the following formal manipulation:

∫ ∫
𝑑𝑦
𝑔(𝑦) = 𝑓 (𝑥) ⇒ 𝑔(𝑦) 𝑑𝑦 = 𝑓 (𝑥) 𝑑𝑥 ⇒ 𝑔(𝑦) 𝑑𝑦 = 𝑓 (𝑥) 𝑑𝑥 .
𝑑𝑥

Of course, we also add an arbitrary constant to the result of any one integral. This is
the reason, the ODE in (1.2.1) is called a variables separable ODE, and this method
is called the method of variables separable. The solution so obtained this way is
called the general solution of the ODE (1.2.1) since any solution can be put in this
form.

(1.3) Example
Find the general solutions to the following ODEs:
(a) 𝑦 0 = 𝑥 2 /𝑦 2 (b) 𝑦 0 = 1 + 𝑦 2 (c) 𝑦 0 = (𝑥 + 1)𝑒 −𝑥 𝑦 2 .
First Order ODE 5
(a) Separating the variables, we have 𝑦 2𝑦 0 = 𝑥 2 . Integrating,

𝑦3 𝑥 3
∫ ∫
2
𝑦 𝑑𝑦 = 𝑥 2 𝑑𝑥 ⇒ = + 𝐶 1 ⇒ 𝑦 3 = 𝑥 3 + 𝐶.
3 3
∫ ∫
𝑑𝑦
(b) 𝑦 0 =1 + 𝑦2 ⇒ (1 + 𝑦 2 ) −1𝑦 0 =1 ⇒ = 1 𝑑𝑥 ⇒ tan−1 𝑦 = 𝑥 + 𝐶
1 + 𝑦2
⇒ 𝑦 = tan(𝑥 + 𝐶).
∫ ∫
−2
(c) 𝑦 0 = (𝑥 + 1)𝑒 −𝑥 𝑦 2 ⇒ 𝑦 −2𝑦 0 = (𝑥 + 1)𝑒 −𝑥 ⇒ 𝑦 𝑑𝑦 = (𝑥 + 1)𝑒 −𝑥 𝑑𝑥
⇒ −𝑦 −1 = −(𝑥 + 2)𝑒 −𝑥 + 𝐶 ⇒ 𝑦 = [(𝑥 + 2)𝑒 −𝑥 − 𝐶] −1 .

(1.4) Example
Find solutions to the following IVPs:
(a) 𝑦 0 = −2𝑥𝑦, 𝑦 (0) = 1.8 (b) 𝑒𝑦𝑦 0 = 𝑥 + 𝑥 3, 𝑦 (1) = 1.
∫ ∫
0 −1 0 −1
(a) 𝑦 = −2𝑥𝑦 ⇒ 𝑦 𝑦 = −2𝑥 ⇒ 𝑦 𝑑𝑦 = (−2𝑥) 𝑑𝑥 ⇒ log |𝑦| = −𝑥 2 + 𝐶 1
2 +𝐶 2 2
⇒ |𝑦| = 𝑒 −𝑥 1 ⇒ |𝑦| = 𝐶𝑒 −𝑥 ⇒ 𝑦 = ±𝐶𝑒 −𝑥 .
2
As 𝐶 is an arbitrary constant, which may be any real number, 𝑦 = 𝐶𝑒 −𝑥 .
2
Then 𝑦 (0) = 1.8 ⇒ 𝐶𝑒 0 = 1.8 ⇒ 𝐶 = 1.8. Hence, 𝑦 (𝑥) = 1.8 𝑒 −𝑥 .
𝑥2 𝑥4
∫ ∫
𝑦 0
(b) 𝑒 𝑦 = 𝑥 + 𝑥 ⇒ 𝑒 𝑑𝑦 = (𝑥 + 𝑥 3 ) 𝑑𝑥 ⇒ 𝑒𝑦 =
3 𝑦
+ +𝐶
2 4
𝑥2 𝑥4 
⇒ 𝑦 = log + +𝐶 for 𝐶 ≥ 0.
2 4
𝑦 (1) = 1 ⇒ 𝑒 1 = 21 + 14 + 𝐶 ⇒ 𝐶 = 𝑒 − 43 . Hence, the solution is
𝑥2 𝑥4 3
𝑦 (𝑥) = log + +𝑒 − .
2 4 4
Most often, the differential equation does not signal in any way that its solutions
are not defined at certain points. This can even happen for IVPs.

(1.5) Example
Solve the IVPs: (a) 𝑦 0 = 1 + 𝑦 2, 𝑦 (0) = 0 (b) 𝑦 0 = 1 + 𝑦 2, 𝑦 (0) = 1.
∫ ∫
(a) 𝑦 = 1 + 𝑦 ⇒ (1 + 𝑦 ) 𝑑𝑦 = 𝑑𝑥 ⇒ tan−1 𝑦 = 𝑥 + 𝐶 ⇒ 𝑦 = tan(𝑥 + 𝐶).
0 2 2 −1

𝑦 (0) = 0 ⇒ 0 = tan(𝐶) ⇒ 𝐶 = 0. Hence, the solution is 𝑦 = tan 𝑥.

This solution is not defined at 𝑥 = ±𝜋/2. Yet, the ODE does not signal anything
about this! The solution exists in (−𝜋/2, 𝜋/2).
(b) As in (a), 𝑦 = tan(𝑥 + 𝐶). 𝑦 (0) = 1 ⇒ 1 = tan 𝐶 ⇒ 𝐶 = 𝜋/4. So, the solution
is 𝑦 = tan(𝑥 + 𝜋/4). Again, this solution exists in (−3𝜋/4, 𝜋/4).
6 MA2020 Classnotes

(1.6) Example
Find the solution of the IVP 𝑦 0 = (1 + 𝑦)𝑥, 𝑦 (0) = −1.

𝑥2
∫ ∫
0 𝑑𝑦
𝑦 = (1 + 𝑦)𝑥 ⇒ = 𝑥 𝑑𝑥 ⇒ log |1 + 𝑦| = + 𝐶.
1 +𝑦 2
This solution is defined for 𝑦 ≠ −1. But the initial condition says otherwise. We
observe that 𝑦 (𝑥) = −1 is a solution. Due to our existence-uniqueness theorem,
𝑦 (𝑥) = −1 is the only solution of the IVP.

(1.7) Example
Find the solution to the IVP 𝑦𝑦 0 + (1 + 𝑦 2 ) sin 𝑥 = 0, 𝑦 (0) = 1.
𝑦𝑦 0 + (1 + 𝑦 2 ) sin 𝑥 = 0 ⇒ 𝑦 (1 + 𝑦 2 ) −1𝑦 0 = − sin 𝑥
∫ ∫
2𝑦 𝑑𝑦
⇒ = − 2 sin 𝑥 ⇒ log(1 + 𝑦 2 ) = 2 cos 𝑥 + 𝐶.
1 + 𝑦2
𝑦 (0) = 1 ⇒ log 2 = 2 + 𝐶 ⇒ 𝐶 = log 2 − 2.
So, log(1 + 𝑦 2 ) = 2 cos 𝑥 + log 2 − 2. Or,
2 (𝑥/2)
𝑦 2 = 𝑒 2 cos 𝑥−2+log 2 − 1 = 2𝑒 2(cos 𝑥−1) − 1 = 2𝑒 −4 sin − 1.
Since 𝑦 (0) > 0, we take the positive sign in the square root. That is,
p
2
𝑦 = 2𝑒 −4 sin (𝑥/2) − 1.
2 (𝑥/2)
This solution is defined for 2𝑒 −4 sin . However,
2 2 2 (𝑥/2)
2𝑒 −4 sin (𝑥/2) ≥1⇔ 𝑒 −4 sin (𝑥/2) ≥ 1/2 ⇔ 𝑒 4 sin ≤2
√
log 2
⇔ 4 sin2 (𝑥/2) ≤ log 2 ⇔ |𝑥/2| ≤ sin−1 2 .
√
log 2
That is, the solution exists in the interval (−𝑎, 𝑎), where 𝑎 = sin−1 2 .

(1.8) Example
Find all solutions of 𝑦 0 = −𝑥/𝑦.
𝑦2 𝑥2
∫ ∫
0
𝑦 = −𝑥/𝑦 ⇒ 𝑦 𝑑𝑦 = − 𝑥 𝑑𝑥 ⇒ = − + 𝐶 1 ⇒ 𝑥 2 + 𝑦 2 = 𝐶.
2 2
In this case, we cannot find 𝑦 as a function of 𝑥. However, the solutions are solution
curves in the 𝑥𝑦-plane.

(1.9) Example
Solve the IVP (1 + 𝑒𝑦 )𝑦 0 = cos 𝑥, 𝑦 (𝜋/2) = 3.
∫ ∫
𝑦 0
(1 + 𝑒 )𝑦 = cos 𝑥 ⇒ (1 + 𝑒 ) 𝑑𝑦 = cos 𝑥 𝑑𝑥 ⇒ 𝑦 + 𝑒𝑦 = sin 𝑥 + 𝐶.
𝑦
First Order ODE 7
𝑦 (𝜋/2) = 3 ⇒ 3 + 𝑒 3 = 1 + 𝐶 ⇒ 𝐶 = 2 + 𝑒 3 . So, the solution is given by

𝑦 + 𝑒𝑦 = sin 𝑥 + 2 + 𝑒 3 .

Here, we cannot express 𝑦 in terms of 𝑥 explicitly. In general, we accept solutions

given implicitly.

1.3 Reducible to variables separable

Sometimes we use a suitable substitution so that a given ODE will become amenable
to the variables separable method. A specific case is when the ODE looks like

𝑦 0 = 𝑓 (𝑦/𝑥),

where the right hand side is a an expression depending directly on 𝑦/𝑥. In this case,
we substitute 𝑢 = 𝑦/𝑥. Then, 𝑢 = 𝑦/𝑥 ⇒ 𝑦 = 𝑢𝑥 ⇒ 𝑦 0 = 𝑢 0𝑥 + 𝑢. The ODE
becomes
∫ ∫
0 𝑑𝑢 𝑑𝑢 𝑑𝑥
𝑢 𝑥 + 𝑢 = 𝑓 (𝑢) ⇒ = 𝑓 (𝑢) − 𝑢 ⇒ = .
𝑑𝑥 𝑓 (𝑢) − 𝑢 𝑥

In fact, we do not remember the last formula. It only shows that the substitution
𝑦 = 𝑢𝑥 reduces the ODE to a case of variables separable.

(1.10) Example
Consider the ODE 2𝑥𝑦𝑦 0 = 𝑦 2 − 𝑥 2 .
𝑦2 − 𝑥 2 𝑦 𝑥
Here, 𝑦 0 = = − . Take 𝑦 = 𝑢𝑥 to get
2𝑥𝑦 2𝑥 2𝑦

𝑢 1 1 + 𝑢2 2𝑢 𝑑𝑢 1
𝑢 0𝑥 + 𝑢 = − ⇒ 𝑢 0𝑥 = − ⇒ = − .
2 2𝑢 2𝑢 1 + 𝑢 2 𝑑𝑥 𝑥
Integrating, we obtain
∫ ∫
2𝑢 𝑑𝑥 𝐶
2
𝑑𝑢 = − ⇒ log(1 + 𝑢 2 ) = − log |𝑥 | + 𝐶 ⇒ 1 + 𝑢 2 = .
1 +𝑢 𝑥 𝑥
Since 𝐶 is arbitrary, we write 𝐶/𝑥 instead of 𝐶/|𝑥 |.
Substituting back 𝑢 = 𝑦/𝑥, we get 1 + (𝑦/𝑥) 2 = 𝐶/𝑥 or, 𝑥 2 + 𝑦 2 = 𝐶𝑥 or,
 𝐶 2 𝐶2
𝑥− + 𝑦2 = .
2 4
8 MA2020 Classnotes

The solutions comprise a family of circles passing through the origin with center
on the 𝑥-axis.

Another type of ODEs can be reduced to variables separable form. They are
equations of the form
𝑑𝑦 𝑎𝑥 + 𝑏𝑦 + 𝑐
= 0 .
𝑑𝑥 𝑎 𝑥 + 𝑏 0𝑦 + 𝑐 0
Here, 𝑎, 𝑏, 𝑐, 𝑎0, 𝑏 0, 𝑐 0 are some real numbers. We consider two cases.
𝑎 𝑏
Case 1: Suppose the coefficients of 𝑥 and 𝑦 are in ratio. That is, 0 = 0 .
𝑎 𝑏
In this case, the ODE is in the form

𝑑𝑦 𝑎𝑥 + 𝑏𝑦 + 𝑐
= .
𝑑𝑥 𝑚(𝑎𝑥 + 𝑏𝑦) + 𝑐 0

We substitute 𝑢 = 𝑎𝑥 + 𝑏𝑦 so that 𝑢 0 = 𝑎 + 𝑏𝑦 0 and the ODE is reduced to

𝑢 +𝑐
𝑢 0 = 𝑎 + 𝑏𝑦 0 = 𝑎 + 𝑏 .
𝑚𝑢 + 𝑐 0
Here, the variables are separated.
𝑎 𝑏
Case 2: Suppose the the coefficients of 𝑥 and 𝑦 are not in ratio. That is, 0 ≠ 0 .
𝑎 𝑏
In this case, we shift both the independent and dependent variables; that is,
we take 𝑥 = 𝑋 + ℎ and 𝑦 = 𝑌 + 𝑘 for some constants ℎ, 𝑘 to be determined suitably.
With this change of variables, we have

𝑑𝑌 𝑑𝑦 𝑎𝑥 + 𝑏𝑦 + 𝑐 𝑎𝑋 + 𝑏𝑌 + 𝑎ℎ + 𝑏𝑘 + 𝑐
= = 0 = .
𝑑𝑋 𝑑𝑥 𝑎 𝑥 + 𝑏 0𝑦 + 𝑐 0 𝑎0𝑋 + 𝑏 0𝑌 + 𝑎0ℎ + 𝑏 0𝑘 + 𝑐 0

The trick is to take ℎ, 𝑘 in such a way that the last expression is simplified. So, we
take
𝑎ℎ + 𝑏𝑘 + 𝑐 = 0, 𝑎0ℎ + 𝑏 0𝑘 + 𝑐 0 = 0. (1.3.1)
Then, the ODE is simplified to

𝑑𝑌 𝑎𝑋 + 𝑏𝑌 𝑎 + 𝑏 (𝑌 /𝑋 )
= 0 0
= 0 0 = 𝑓 (𝑌 /𝑋 ).
𝑑𝑋 𝑎 𝑋 + 𝑏 𝑌 𝑎 + 𝑏 (𝑌 /𝑋 )

Now, the earlier method of substituting 𝑌 = 𝑢𝑋 will separate the variables. Then,
solution curves will be obtained in the form 𝑔(𝑋, 𝑌 ) = 0, or, 𝑔(𝑥 − ℎ, 𝑦 − 𝑘) = 0.

(1.11) Example
Solve the ODE (𝑥 + 𝑦 − 1)𝑦 0 = 2𝑥 + 2𝑦 + 1.
First Order ODE 9
Here, the coefficients of 𝑥, 𝑦 are in ratio. We substitute 𝑢 = 𝑥 + 𝑦 so that 𝑢 0 = 1 + 𝑦 0,
and the ODE is reduced to
2𝑢 + 1 𝑑𝑢 2𝑢 + 1 3𝑢
𝑢0 − 1 = ⇒ = 𝑢0 = +1= .
𝑢−1 𝑑𝑥 𝑢−1 𝑢−1
Integrating, we get
∫ ∫
𝑢−1 𝑢 1
𝑑𝑢 = 𝑑𝑥 ⇒ − log |𝑢 | = 𝑥 + 𝐶 1 .
3𝑢 3 3
Substituting 𝑢 = 𝑥 + 𝑦 and simplifying we obtain
𝑦 − 2𝑥 − 𝐶 2 = log |𝑥 + 𝑦| ⇒ 𝐶𝑒𝑦−2𝑥 = 𝑥 + 𝑦
for an arbitrary constant 𝐶. This gives the solution curves of the ODE.

(1.12) Example
Solve the ODE (3𝑦 − 7𝑥 + 7) + (7𝑦 − 3𝑥 + 3)𝑦 0 = 0.
7𝑥 − 3𝑦 − 7
The ODE is 𝑦 0 = . The coefficients of 𝑥, 𝑦 in both linear expressions
−3𝑥 + 7𝑦 + 3
are not in ratio. So, we substitute 𝑥 = 𝑋 + ℎ, 𝑦 = 𝑌 + 𝑘. Equation 1.3.1 gives
7ℎ − 3𝑘 − 7 = 0 = −3ℎ + 7𝑘 + 3. Solving these, we get ℎ = 1, 𝑘 = 0. That is, we
take 𝑥 = 𝑋 + 1, 𝑦 = 𝑌 so that the ODE is reduced to
    𝑑𝑌 𝑑𝑌 7𝑋 − 3𝑌 7 − 3(𝑌 /𝑋 )
3𝑌 −7(𝑋 +1)+7 + 7𝑌 −3(𝑋 +1)+3 =0 ⇒ = = .
𝑑𝑋 𝑑𝑋 −3𝑋 + 7𝑌 −3 + 7(𝑌 /𝑋 )
𝑑𝑌 𝑑𝑢 7 − 3𝑢
Substitute 𝑌 = 𝑢𝑋 so that = 𝑋 +𝑢 = . This gives
𝑑𝑋 𝑑𝑋 −3 + 7𝑢
𝑑𝑢 7 − 3𝑢 7 − 3𝑢 + 3𝑢 − 7𝑢 2 7 − 7𝑢 2
𝑋 = −𝑢 = = .
𝑑𝑋 −3 + 7𝑢 −3 + 7𝑢 −3 + 7𝑢
∫ ∫
7𝑢 − 3 𝑑𝑋
Separating the variables, we obtain 𝑑𝑢 = .
7 − 7𝑢 2 𝑋
7𝑢 − 3 1 2𝑢 3 1 3 1
Now, 2
=− 2 + − . Then the above gives
7 − 7𝑢 2 𝑢 − 1 14 𝑢 − 1 14 𝑢 + 1
1 3 3
− log |𝑢 2 − 1| + log |𝑢 − 1| − log |𝑢 + 1| = log |𝑋 | + 𝐶 1 .
2 14 14
Taking exponential of both sides and simplifying we get
𝐶 2 |𝑋 | = |𝑢 − 1| −2/7 |𝑢 + 1| −5/7 .
Substituting 𝑢 = 𝑌 /𝑋, 𝑋 = 𝑥 − 1, 𝑌 = 𝑦 and simplifying we obtain
(𝑦 + 𝑥 − 1) 5 (𝑦 − 𝑥 + 1) 2 = 𝐶
for some arbitrary constant 𝐶. This gives the solution curves.
10 MA2020 Classnotes

1.4 Exact Equations

Sometimes observing simple identities about the differentials help in solving ODEs.
For instance, consider the ODE

𝑥𝑦 0 + 𝑦 − 2𝑥 = 0.
𝑑 (𝑥𝑦)
Notice that = 𝑥𝑦 0 + 𝑦. Then the ODE can be solved as follows:
𝑑𝑥
∫ ∫
𝑑 (𝑥𝑦)
= 2𝑥 ⇒ 𝑑 (𝑥𝑦) = 2𝑥 𝑑𝑥 ⇒ 𝑥𝑦 = 𝑥 2 + 𝐶.
𝑑𝑥
Using differentials, the ODE can be written as

𝑥 𝑑𝑦 + 𝑦 𝑑𝑥 − 2𝑥 𝑑𝑥 = 0.

This can be solved as

∫ ∫
2
𝑑 (𝑥𝑦) − 𝑑 (𝑥 ) = 0 ⇒ 𝑑 (𝑥𝑦) − 𝑑 (𝑥 2 ) = 𝐶 ⇒ 𝑥𝑦 − 𝑥 2 = 𝐶.

In fact, we will write an ODE of the first order such as 𝑥𝑦 0 + 𝑦 − 2𝑥 = 0 as

𝑥 𝑑𝑦 + (𝑦 − 2𝑥) 𝑑𝑥 = 0

using the differentials. In general, we consider first order ODEs in the form

𝑀 (𝑥, 𝑦) 𝑑𝑥 + 𝑁 (𝑥, 𝑦) 𝑑𝑦 = 0.

This also covers the variables separable case since an ODE in the form 𝑔(𝑦)𝑦 0 = 𝑓 (𝑥)
can be rewritten as
−𝑓 (𝑥) 𝑑𝑥 + 𝑔(𝑦) 𝑑𝑦 = 0.
We can solve the general ODE
above provided we find that the expression on the
left is a differential 𝑑 𝑢 (𝑥, 𝑦) . In this case, we may integrate to obtain the general
solution as 𝑢 (𝑥, 𝑦) = 𝐶. So, the question is when can we get a function 𝑢 (𝑥, 𝑦) so
that

𝑑 𝑢 (𝑥, 𝑦) = 𝑀 𝑑𝑥 + 𝑁 𝑑𝑦
is true. First, we look for some necessary conditions. Suppose that there exists a
function 𝑢 (𝑥, 𝑦) such that
𝑑𝑢 = 𝑀 𝑑𝑥 + 𝑁 𝑑𝑦
From calculus we know that
𝜕𝑢 𝜕𝑢
𝑑𝑢 = 𝑑𝑥 + 𝑑𝑦.
𝜕𝑥 𝜕𝑥
First Order ODE 11
Hence, the necessary condition is that
𝜕𝑢 𝜕𝑢
𝑀= , 𝑁 = .
𝜕𝑥 𝜕𝑦

If we assume that the second derivatives of 𝑢 (𝑥, 𝑦) are continuous, then 𝑢𝑥𝑦 = 𝑢𝑦𝑥 .
The above condition would imply that
𝜕𝑀 𝜕𝑁
= 𝑢𝑥𝑦 = .
𝜕𝑦 𝜕𝑥

In fact, this condition is also sufficient as the following result shows.

(1.13) Theorem
Let 𝑀 (𝑥, 𝑦) and 𝑁 (𝑥, 𝑦) be real valued functions having continuous partial deriva-
tives on the rectangle 𝑅 : 𝑎 < 𝑥 < 𝑏, 𝑐 < 𝑦 < 𝑑. Then, the following are equivalent:
(1) There exists a function 𝑢 (𝑥, 𝑦) defined on 𝑅 such that 𝑑𝑢 = 𝑀 𝑑𝑥 + 𝑁 𝑑𝑦.
(2) 𝑀𝑦 = 𝑁𝑥 holds in 𝑅.
(3) There exists a function 𝑢 (𝑥, 𝑦) satisfying 𝑀 = 𝑢𝑥 and 𝑁 = 𝑢𝑦 .

Proof. (1) ⇒ (2): Suppose 𝑑𝑢 = 𝑀 𝑑𝑥 + 𝑁 𝑑𝑦 is true in 𝑅. Then, 𝑢𝑥𝑦 and 𝑢𝑦𝑥 exist
and are continuous. By the Chain rule, 𝑑𝑢 = 𝑢𝑥 𝑑𝑥 + 𝑢𝑦𝑑𝑦. Comparing these two
equations, we get 𝑀 = 𝑢𝑥 and 𝑁 = 𝑢𝑦 . Thus, 𝑀𝑦 = 𝑢𝑥𝑦 and 𝑁𝑥 = 𝑢𝑦𝑥 . Since 𝑢𝑥𝑦 and
𝑢𝑦𝑥 are continuous, they are equal. Hence, 𝑀𝑦 = 𝑁𝑥 holds in 𝑅.
(2) ⇒ (3): Suppose that 𝑀𝑦 = 𝑁𝑥 . Integrate with respect to 𝑥 to get
∫
𝑁 (𝑥, 𝑦) = 𝑀𝑦 𝑑𝑥 + 𝑔(𝑦).

Here, 𝑔(𝑦) is an arbitrary function of 𝑦 alone. Define

∫ ∫
𝑢 (𝑥, 𝑦) = 𝑀 (𝑥, 𝑦) 𝑑𝑥 + 𝑔(𝑦) 𝑑𝑦.

Then, ∫
𝜕
𝑢𝑥 = 𝑀 (𝑥, 𝑦) + 𝑔(𝑦) 𝑑𝑦 = 𝑀 (𝑥, 𝑦) + 0 = 𝑀 (𝑥, 𝑦).
𝜕𝑥
∫
𝑢𝑦 = 𝑀𝑦 𝑑𝑥 + 𝑔(𝑦) = 𝑁 (𝑥, 𝑦).

(3) ⇒ (1): Suppose that there exists 𝑢 (𝑥, 𝑦) such that 𝑀 = 𝑢𝑥 and 𝑁 = 𝑢𝑦 . Then
𝑀 𝑑𝑥 + 𝑁 𝑑𝑦 = 𝑢𝑥 𝑑𝑥 + 𝑢𝑦 𝑑𝑦 = 𝑑𝑢.
In view of this result, we say that
12 MA2020 Classnotes

the ODE 𝑀 (𝑥, 𝑦) 𝑑𝑥 + 𝑁 (𝑥, 𝑦) 𝑑𝑦 = 0 is an exact equation iff 𝑀𝑦 = 𝑁𝑥 .

The proof of (1.13) shows how to compute a function 𝑢 (𝑥, 𝑦) if the condition
𝑀𝑦 = 𝑁𝑥 is satisfied. It is:
∫ ∫
𝑢 (𝑥, 𝑦) = 𝑀 (𝑥, 𝑦) 𝑑𝑥 + 𝑔(𝑦) 𝑑𝑦
∫ ∫ ∫ ∫
= 𝑀 (𝑥, 𝑦) 𝑑𝑥 + 𝑁 (𝑥, 𝑦) 𝑑𝑦 − 𝑀𝑦 𝑑𝑥 𝑑𝑦.

Since this formula holds under the assumption 𝑀𝑦 = 𝑁𝑥 , we also have

∫ ∫ ∫ ∫
𝑢 (𝑥, 𝑦) = 𝑀 (𝑥, 𝑦) 𝑑𝑥 + 𝑁 (𝑥, 𝑦) 𝑑𝑦 − 𝑁𝑥 𝑑𝑦 𝑑𝑥 .

We will not memorize these formulas. Instead, we understand the method and
then use it in any particular problem. This understanding gives rise to three ways
of solving an exact equation. So, let the given exact equation be

𝑀 (𝑥, 𝑦) 𝑑𝑥 + 𝑁 (𝑥, 𝑦) 𝑑𝑦 = 0.

The exactness implies that 𝑀𝑦 = 𝑁𝑥 , which, due to (1.13) guarantees the existence
of a function 𝑢 (𝑥, 𝑦) such that 𝑀 = 𝑢𝑥 and 𝑁 = 𝑢𝑦 .
∫
First method: Since 𝑀 = 𝑢𝑥 , we have 𝑢 = 𝑀 (𝑥, 𝑦) 𝑑𝑥 + 𝑔(𝑦). Differentiating with
respect to 𝑦, we get
∫ ∫
0 𝜕𝑀
𝑔 (𝑦) = 𝑢𝑦 − 𝑀𝑦 𝑑𝑥 = 𝑁 (𝑥, 𝑦) − 𝑑𝑥 .
𝜕𝑦

Then, we determine 𝑔(𝑦) from this and substitute back to get 𝑢 (𝑥, 𝑦). Recall that
the solution curves are given by 𝑢 (𝑥, 𝑦) = 𝐶.
∫
Second method: As 𝑁 = 𝑢𝑦 , we have 𝑢 = 𝑁 (𝑥, 𝑦) 𝑑𝑦 + ℎ(𝑥). Differentiating with
respect to 𝑥, we obtain
∫ ∫
0 𝜕𝑁
ℎ (𝑥) = 𝑢𝑥 − 𝑁𝑥 𝑑𝑦 = 𝑀 (𝑥, 𝑦) − 𝑑𝑦.
𝜕𝑥

We determine ℎ(𝑥) from this and substitute back to obtain 𝑢 (𝑥, 𝑦).
Third method: Using both 𝑀 = 𝑢𝑥 and 𝑁 = 𝑢𝑦 , we get
∫ ∫
𝑢 (𝑥, 𝑦) = 𝑀 (𝑥, 𝑦) 𝑑𝑥 + 𝑔(𝑦), 𝑢 (𝑥, 𝑦) = 𝑁 (𝑥, 𝑦) 𝑑𝑦 + ℎ(𝑥).

Inspecting these two expression, we determine 𝑔(𝑦), ℎ(𝑥); and then 𝑢 (𝑥, 𝑦).
First Order ODE 13
(1.14) Example
Find the general solution of the ODE 3𝑦 + 𝑒 𝑥 + (3𝑥 + cos 𝑦)𝑦 0 = 0.
The ODE is 𝑀 𝑑𝑥 + 𝑁 𝑑𝑦 = 0 with 𝑀 = 3𝑦 + 𝑒 𝑥 and 𝑁 = 3𝑥 + cos 𝑦.
We find that 𝑀𝑦 = 3 and 𝑁𝑥 = 3. So, it is an exact equation. Hence, there exists a
function 𝑢 (𝑥, 𝑦) such that
(a) 𝑀 = 3𝑦 + 𝑒 𝑥 = 𝑢𝑥 (𝑥, 𝑦), (b) 𝑁 = 3𝑥 + cos 𝑦 = 𝑢𝑦 (𝑥, 𝑦).
We illustrate the three methods to determine 𝑢 (𝑥, 𝑦).
First method: Integrating (a) with respect to 𝑥, we get
∫
𝑢 = (3𝑦 + 𝑒 𝑥 ) 𝑑𝑥 = 3𝑥𝑦 + 𝑒 𝑥 + 𝑔(𝑦).

Differentiating with respect to 𝑦 and using (b), we have

𝑢𝑦 = 3𝑥 + 𝑔0 (𝑦) ⇒ 3𝑥 + cos 𝑦 = 3𝑥 + 𝑔0 (𝑦) ⇒ 𝑔0 (𝑦) = cos 𝑦 ⇒ 𝑔(𝑦) = sin 𝑦.
Here, we need not consider the constant of integration, since in the solution this
constant will re-appear as 𝑢 (𝑥, 𝑦) = 𝐶. Also, we need just one such 𝑔(𝑦).
Then, 𝑢 = 3𝑥𝑦 + 𝑒 𝑥 + 𝑔(𝑦) = 3𝑥𝑦 + 𝑒 𝑥 + sin 𝑦. The solution curves are given by
𝑢 (𝑥, 𝑦) = 𝐶 or, 3𝑥𝑦 + 𝑒 𝑥 + sin 𝑦 = 𝐶.
Second method: Integrate (b) with respect to 𝑦 to get
∫
𝑢 = (3𝑥 + cos 𝑦) 𝑑𝑦 = 3𝑥𝑦 + sin 𝑦 + ℎ(𝑥).

Differentiate with respect to 𝑥 and use (a) to get

𝑢𝑥 = 3𝑦 + ℎ0 (𝑥) ⇒ 3𝑦 + 𝑒 𝑥 = 3𝑦 + ℎ0 (𝑥) ⇒ ℎ0 (𝑥) = 𝑒 𝑥 ⇒ ℎ(𝑥) = 𝑒 𝑥 .
Again, we neglect the constant of integration. It says that
𝑢 (𝑥, 𝑦) = 3𝑥𝑦 + sin 𝑦 + ℎ(𝑥) = 3𝑥𝑦 + sin 𝑦 + 𝑒 𝑥 .
And, the solution curves are given by 𝑢 (𝑥, 𝑦) = 𝐶 or, 3𝑥𝑦 + sin 𝑦 + 𝑒 𝑥 = 𝐶.
Third method: We integrate (a) with respect to 𝑥 and also integrate (b) with respect
to 𝑦 to obtain
∫ ∫
𝑥 𝑥
𝑢 = (3𝑦 + 𝑒 )𝑑𝑥 = 3𝑥𝑦 + 𝑒 + 𝑔(𝑦), 𝑢 = (3𝑥 + cos 𝑦)𝑑𝑦 = 3𝑥𝑦 + sin 𝑦 + ℎ(𝑥).

Matching them we find that 𝑔(𝑦) = sin 𝑦 and ℎ(𝑥) = 𝑒 𝑥 . Then 𝑢 = 3𝑥𝑦 + 𝑒 𝑥 + sin 𝑦
gives the solution curves as 3𝑥𝑦 + 𝑒 𝑥 + sin 𝑦 = 𝐶.

Out of the three, the third method is the easiest provided one is able to guess
correctly. One should also get familiarized with other methods.
14 MA2020 Classnotes

(1.15) Example
Solve cos(𝑥 + 𝑦) 𝑑𝑥 + 3𝑦 2 + 2𝑦 + cos(𝑥 + 𝑦) 𝑑𝑦 = 0.

Here, 𝑀 = cos(𝑥 + 𝑦) and 𝑁 = 3𝑦 2 + 2𝑦 + cos(𝑥 + 𝑦). Then 𝑀𝑦 = − sin(𝑥 + 𝑦)

and 𝑁𝑥 = − sin(𝑥 + 𝑦) = 𝑀𝑦 . Hence, it is an exact equation. Thus, there exists a
function 𝑢 (𝑥, 𝑦) such that

(a) 𝑢𝑥 = 𝑀 = cos(𝑥 + 𝑦), (b) 𝑢𝑦 = 𝑁 = 3𝑦 2 + 2𝑦 + cos(𝑥 + 𝑦).

We determine 𝑢 (𝑥, 𝑦) by inspection (Third method) as follows.

Integrate (a) with respect to 𝑥 and integrate (b) with respect to 𝑦 to get

𝑢 = sin(𝑥 + 𝑦) + 𝑔(𝑦), 𝑢 = 𝑦 3 + 𝑦 2 + sin(𝑥 + 𝑦) + ℎ(𝑥).

Matching these, we find that 𝑔(𝑦) = 𝑦 3 + 𝑦 2 and ℎ(𝑥) = 0. Then the solution curves
are given by 𝑢 (𝑥, 𝑦) = 𝐶 or, sin(𝑥 + 𝑦) + 𝑦 3 + 𝑦 2 = 𝐶.

(1.16) Example
Solve the IVP (cos 𝑦 sinh 𝑥 + 1) 𝑑𝑥 − sin 𝑦 cosh 𝑥 𝑑𝑦 = 0, 𝑦 (1) = 2.
Here, 𝑀 = cos 𝑦 sinh 𝑥 + 1 and 𝑁 = − sin 𝑦 cosh 𝑥. It gives 𝑀𝑦 = − sin 𝑦 sinh 𝑥 and
𝑁𝑥 = − sin 𝑦 sinh 𝑥. As 𝑀𝑦 = 𝑁𝑥 , the ODE is exact. Then, there exists a function
𝑢 (𝑥, 𝑦) such that 𝑢𝑥 = 𝑀 and 𝑢𝑦 = 𝑁 . To determine 𝑢, we integrate 𝑢𝑥 = 𝑀 with
respect to 𝑥 to obtain
𝑢 = cos 𝑦 cosh 𝑥 + 𝑥 + 𝑔(𝑦).
Differentiating with respect to 𝑦 and using 𝑢𝑦 = 𝑁 , we have

𝑢𝑦 = − sin 𝑦 cosh 𝑥+𝑔0 (𝑦) ⇒ 𝑔0 (𝑦) = − sin 𝑦 cosh 𝑥+sin 𝑦 cosh 𝑥 = 0 ⇒ 𝑔(𝑦) = 𝐾 .

Since we need only one such 𝑢, we take 𝐾 = 0 so that

𝑢 (𝑥, 𝑦) = cos 𝑦 cosh 𝑥 + 𝑥 + 𝑔(𝑦) = cos 𝑦 cosh 𝑥 + 𝑥 .

A general solution is cos 𝑦 cosh 𝑥 +𝑥 = 𝐶. Using 𝑦 (1) = 2, we have cos 2 cosh 1+1 =
𝐶. Then the solution to the IVP is given by

cos 𝑦 cosh 𝑥 + 𝑥 = cos 2 cosh 1 + 1.

(1.17) Example
Solve the IVP 3𝑥 2𝑦 + 8𝑥𝑦 2 + (𝑥 3 + 8𝑥 2𝑦 + 12𝑦 2 )𝑦 0 = 0, 𝑦 (2) = 1.
Here, 𝑀 = 3𝑥 2𝑦 + 8𝑥𝑦 2 and 𝑁 = 𝑥 3 + 8𝑥 2𝑦 + 12𝑦 2 . Then 𝑀𝑦 = 3𝑥 2 + 16𝑥𝑦 and
𝑁𝑥 = 3𝑥 2 + 16𝑥𝑦 + 0 = 𝑀𝑦 . So, the ODE is exact. Then, there exists a function
𝑢 (𝑥, 𝑦) such that

(a) 𝑢𝑥 = 3𝑥 2𝑦 + 8𝑥𝑦 2, (b) 𝑢𝑦 = 𝑥 3 + 8𝑥 2𝑦 + 12𝑦 2 .

First Order ODE 15
Integrating (a) and (b) with respect to 𝑥 and 𝑦, respectively, we get

𝑢 = 𝑥 3𝑦 + 4𝑥 2𝑦 2 + 𝑔(𝑦), 𝑢 = 𝑥 3𝑦 + 4𝑥 2𝑦 2 + 4𝑦 3 + ℎ(𝑥).

Matching these we have 𝑔(𝑦) = 4𝑦 3 and ℎ(𝑥) = 0. Then the solution curves are
given by 𝑢 (𝑥, 𝑦) = 𝐶 or, 𝑥 3𝑦 +4𝑥 2𝑦 2 +4𝑦 3 = 𝐶. Using the initial condition: 𝑦 (2) = 1,
we see that
23 · 1 + 4 · 22 · 12 + 4 · 13 = 𝐶 ⇒ 𝐶 = 28.
Hence, the solution to the IVP is 𝑥 3𝑦 + 4𝑥 2𝑦 2 + 4𝑦 3 = 28.

1.5 Integrating factors

We must be cautious while using the three methods discussed in the last section.
Remember that the methods work for exact equations. If the ODE is not exact, then
the methods need not give solutions to the ODE, or even, they may fail towards
obtaining a solution. See the following example.

(1.18) Example
Solve the ODE −𝑦 𝑑𝑥 + 𝑥 𝑑𝑦 = 0.
here, 𝑀 = −𝑦 and 𝑁 = 𝑥. We find 𝑢 (𝑥, 𝑦) so that 𝑢𝑥 = 𝑀 = −𝑦 and 𝑢𝑦 = 𝑁 = 𝑥.
Integrating first with respect to 𝑥, we get 𝑢 = −𝑥𝑦 + 𝑔(𝑦). Differentiating with
respect to 𝑦, we have 𝑢𝑦 = −𝑥 + 𝑔0 (𝑦). Since 𝑢𝑦 = 𝑥, we have 𝑔0 (𝑦) = −2𝑥. There is
something wrong, since our method assumes that 𝑔(𝑦) is a function of 𝑦 alone.
We find that the ODE is not exact, because 𝑀𝑦 = −1 whereas 𝑁𝑥 = 1. Thus, none
of the three methods above are applicable.
However, we can separate the variables and solve it as follows:
∫ ∫
𝑑𝑦 𝑑𝑥
𝑥 𝑑𝑦 = 𝑦 𝑑𝑥 ⇒ = ⇒ log |𝑦| = log |𝑥 | + 𝐶 1 ⇒ 𝑦 = 𝐶𝑥 .
𝑦 𝑥
So, before applying any one of the three methods, one must check that the ODE
is exact.
Though the ODE in (1.18) is not exact, it can be made exact. Look at the solution
curves. They are 𝑦/𝑥 = 𝐶. So, our function of two variables is 𝑢 (𝑥, 𝑦) = 𝑦/𝑥. Now,
its differential is
𝑦 𝑑𝑦
𝑑𝑢 = 𝑢𝑥 𝑑𝑥 + 𝑢𝑦 𝑑𝑦 = − 2 𝑑𝑥 + .
𝑥 𝑥
The ODE is given as −𝑦 𝑑𝑥 + 𝑥 𝑑𝑦 = 0. Comparing these, we find that if we
multiply 1/𝑥 2 to the given ODE, we would obtain an exact equation. We explore
this possibility further.
16 MA2020 Classnotes

For the ODE 𝑀 𝑑𝑥 + 𝑁 𝑑𝑦 = 0, a function 𝜇 (𝑥, 𝑦) is called an integrating factor

iff 𝜇 (𝑥, 𝑦)𝑀 𝑑𝑥 + 𝜇 (𝑥, 𝑦)𝑁 𝑑𝑦 = 0 is an exact equation; that is, multiplying 𝜇 (𝑥, 𝑦)
the new ODE becomes exact.
Notice that 𝜇 (𝑥, 𝑦)𝑀 𝑑𝑥 + 𝜇 (𝑥, 𝑦)𝑁 𝑑𝑦 = 0 is exact when
   
𝜕 𝜇 (𝑥, 𝑦)𝑀 𝜕 𝜇 (𝑥, 𝑦)𝑁
= .
𝜕𝑦 𝜕𝑥

Using the Chain rule, it means that 𝜇 (𝑥, 𝑦) is an integrating factor of the ODE
𝑀 𝑑𝑥 + 𝑁 𝑑𝑦 = 0 iff
𝜕𝜇 𝜕𝑀 𝜕𝜇 𝜕𝑁
𝑀 +𝜇 =𝑁 +𝜇 .
𝜕𝑦 𝜕𝑦 𝜕𝑥 𝜕𝑥
However, solving such an equation for determining 𝜇 (𝑥, 𝑦) may be more difficult
than solving the original ODE. So, we look for some special cases.
We ask whether it is possible for the function 𝜇 (𝑥, 𝑦) to depend on 𝑥 alone?
What could be the conditions that yield this situation? When 𝜇 = 𝜇 (𝑥), its partial
derivative with respect to 𝑦 becomes 0 so that the above equation simplifies to

𝜕𝑀 𝑑𝜇 𝜕𝑁 𝑑𝜇 𝑀𝑦 − 𝑁𝑥
𝜇 =𝑁 +𝜇 ⇔ = 𝜇.
𝜕𝑦 𝑑𝑥 𝜕𝑥 𝑑𝑥 𝑁

𝑀𝑦 − 𝑁𝑥
Notice that this expression is meaningless unless is a function of 𝑥 alone.
𝑁
So, suppose
𝑀𝑦 − 𝑁𝑥
= 𝑓 (𝑥).
𝑁
Then 𝜇 is obtained by solving 𝜇 0 (𝑥) = 𝑓 (𝑥)𝜇. We need just one such 𝜇; so we ignore
the constants of integration. By separating the variables, we have
∫ ∫ ∫ ∫ 
𝑑𝜇
= 𝑓 (𝑥) 𝑑𝑥 ⇒ log 𝜇 = 𝑓 (𝑥) 𝑑𝑥 ⇒ 𝜇 (𝑥) = exp 𝑓 (𝑥) 𝑑𝑥 .
𝜇

Here, we do not bother about taking |𝜇| since our requirement is one such 𝜇. Our
method boils down to the following:
Integrating factor 1: If the ODE 𝑀 𝑑𝑥 + 𝑁 𝑑𝑦 = 0 is not exact and

𝑀𝑦 − 𝑁𝑥
= 𝑓 (𝑥),
𝑁
∫

a function of 𝑥 alone, then 𝜇 (𝑥) = exp 𝑓 (𝑥) 𝑑𝑥 is an integrating factor of the
ODE.
Similarly, we have the following method when an analogous expression is a
function of 𝑦 alone.
First Order ODE 17
Integrating factor 2: If the ODE 𝑀 𝑑𝑥 + 𝑁 𝑑𝑦 = 0 is not exact and
𝑀𝑦 − 𝑁𝑥
= 𝑔(𝑦),
𝑀
∫

a function of 𝑦 alone, then 𝜇 (𝑦) = exp − 𝑔(𝑦) 𝑑𝑦 is an integrating factor of the
ODE.
We illustrate these methods in the following examples.

(1.19) Example
𝑦2
Solve the ODE + 2𝑦𝑒 𝑥 + (𝑦 + 𝑒 𝑥 )𝑦 0 = 0.
2
Here, 𝑀 = 𝑦 2 /2 + 2𝑦𝑒 𝑥 , 𝑁 = 𝑦 + 𝑒 𝑥 ⇒ 𝑀𝑦 = 𝑦 + 2𝑒 𝑥 , 𝑁𝑥 = 𝑒 𝑥 ≠ 𝑀𝑦 . Hence, it is
not an exact equation. Now,
𝑀𝑦 − 𝑁𝑥 𝑦 + 𝑒 𝑥
= = 1 = 𝑓 (𝑥).
𝑁 𝑦 + 𝑒𝑥
∫
∫

It is a function of 𝑥 alone. Hence, 𝜇 = exp 𝑓 (𝑥) 𝑑𝑥 = exp 𝑑𝑥 = 𝑒 𝑥 is an
integrating factor. There exists a function 𝑢 (𝑥, 𝑦) such that
𝑒𝑥𝑦2
(a) 𝑢𝑥 = 𝜇𝑀 = + 2𝑦𝑒 2𝑥 , (b) 𝑢𝑦 = 𝜇𝑁 = 𝑦𝑒 𝑥 + 𝑒 2𝑥 .
2
Integrating (a) with respect to 𝑥 and (b) with respect to 𝑦, we have
𝑒𝑥𝑦2 𝑒𝑥𝑦2
𝑢= + 𝑦𝑒 2𝑥 + 𝑔(𝑦), 𝑢= + 𝑦𝑒 2𝑥 + ℎ(𝑥).
2 2
Matching these, we take 𝑔(𝑦) = ℎ(𝑥) = 0 to get the solution curve as 𝑢 (𝑥, 𝑦) = 𝐶
𝑒𝑥𝑦2
or, + 𝑦𝑒 2𝑥 = 𝐶.
2
(1.20) Example



Solve the IVP 𝑒 𝑥+𝑦 + 𝑦𝑒𝑦 𝑑𝑥 + 𝑥𝑒𝑦 − 1 𝑑𝑦 = 0, 𝑦 (0) = −1.
Here, 𝑀 = 𝑒 𝑥+𝑦 + 𝑦𝑒𝑦 , 𝑁 = 𝑥𝑒𝑦 − 1 ⇒ 𝑀𝑦 − 𝑁𝑥 = 𝑒 𝑥+𝑦 + 𝑦𝑒𝑦 . So, it is not an exact
equation. We find that
𝑀𝑦 − 𝑁𝑥 𝑒 𝑥+𝑦 + 𝑦𝑒𝑦
=
𝑁 𝑥𝑒𝑦 − 1
is not a function of 𝑥 alone. And,
𝑀𝑦 − 𝑁𝑥 𝑒 𝑥+𝑦 + 𝑦𝑒𝑦
= 𝑥+𝑦 = 1 = 𝑔(𝑦)
𝑀 𝑒 + 𝑦𝑒𝑦
is a function of 𝑦 alone. So, we take the integrating factor as
 ∫   ∫ 
𝜇 (𝑦) = exp − 𝑔(𝑦) 𝑑𝑦 = exp − 𝑑𝑦 = 𝑒 −𝑦 .
18 MA2020 Classnotes

Multiplying it with the ODE, we have

𝑒 𝑥 + 𝑦 𝑑𝑥 + 𝑥 − 𝑒 −𝑦 𝑑𝑦 = 0.

Since it is an exact equation, there exists a function 𝑢 (𝑥, 𝑦) such that

(a) 𝑢𝑥 = 𝑒 𝑥 + 𝑦, (b) 𝑢𝑦 = 𝑥 − 𝑒 −𝑦 .

Integrating (a) with respect to 𝑥 and (𝑏) with respect to 𝑦 we get

𝑢 (𝑥, 𝑦) = 𝑒 𝑥 + 𝑥𝑦 + 𝑔1 (𝑦), 𝑢 (𝑥, 𝑦) = 𝑥𝑦 + 𝑒 −𝑦 + ℎ 1 (𝑥).

Matching these we see that 𝑔1 (𝑦) = 𝑒 −𝑦 and ℎ 1 (𝑥) = 𝑒 𝑥 . Then the solution curves
are given by 𝑢 (𝑥, 𝑦) = 𝐶 or,

𝑒 𝑥 + 𝑥𝑦 + 𝑒 −𝑦 = 𝐶.

Since 𝑦 (0) = −1, we get 𝑒 0 + 0(−1) + 𝑒 1 = 𝐶 ⇒ 𝐶 = 1 + 𝑒. Then the solution of

the IVP is given by 𝑒 𝑥 + 𝑥𝑦 + 𝑒 −𝑦 = 1 + 𝑒.

1.6 Linear equations

A very special type of ODE that often comes up in applications is a linear equation.
A linear first order ODE is an ODE in the form

𝑦 0 + 𝑝 (𝑥) 𝑦 = 𝑟 (𝑥).

When the right hand side is 0, that is, 𝑟 (𝑥) = 0, the equation is called a linear first
order homogeneous ODE.
The linear ODE can be written in the differential form as


𝑝 (𝑥)𝑦 − 𝑟 (𝑥) 𝑑𝑥 + 𝑑𝑦 = 0.

Here, 𝑀 = 𝑝 (𝑥)𝑦 − 𝑟 (𝑥), 𝑁 = 1 so that 𝑀𝑦 − 𝑁𝑥 = 𝑝 (𝑥). Thus, the equation is exact

when 𝑝 ∫(𝑥) = 0. In that case, the equation is 𝑦 0 = 𝑟 (𝑥) whose solution can be written
as 𝑦 = 𝑟 (𝑥) 𝑑𝑥 + 𝐶. In case 𝑝 (𝑥) ≠ 0, we should seek an integrating factor. We
observe that
𝑀𝑦 − 𝑁𝑥 𝑝 (𝑥)
= = 𝑝 (𝑥)
𝑁 1
is a function of 𝑥 alone. Hence, an integrating factor is given by
∫ 
𝜇 (𝑥) = exp 𝑝 (𝑥) 𝑑𝑥 .
First Order ODE 19


Multiplying the ODE with 𝜇 (𝑥), we have 𝜇 (𝑥) 𝑝 (𝑥)𝑦 − 𝑟 (𝑥) 𝑑𝑥 + 𝜇 (𝑥) 𝑑𝑦 = 0, or,

𝜇 (𝑥)𝑝 (𝑥)𝑦 𝑑𝑥 + 𝜇 (𝑥) 𝑑𝑦 = 𝜇 (𝑥)𝑟 (𝑥) 𝑑𝑥 .

We see that
 ∫  ∫  𝑑 ∫ 
0 𝑑
𝜇 (𝑥) = exp 𝑝 (𝑥) 𝑑𝑥 = exp 𝑝 (𝑥) 𝑑𝑥 𝑝 (𝑥) 𝑑𝑥 = 𝜇 (𝑥)𝑝 (𝑥).
𝑑𝑥 𝑑𝑥
Thus, the ODE reduces to

𝜇 0 (𝑥)𝑦 𝑑𝑥 + 𝜇 (𝑥) 𝑑𝑦 = 𝜇 (𝑥)𝑟 (𝑥) 𝑑𝑥 ⇒ 𝑑 𝜇 (𝑥)𝑦 = 𝜇 (𝑥)𝑟 (𝑥) 𝑑𝑥 .

∫
Integrating, we obtain 𝜇 (𝑥)𝑦 = 𝜇 (𝑥)𝑟 (𝑥) 𝑑𝑥. Along with the constant of integra-
tion, we obtain
h∫ i ∫ 
−1
𝑦 = [𝜇 (𝑥)] 𝜇 (𝑥)𝑟 (𝑥) 𝑑𝑥 + 𝐶 , where 𝜇 (𝑥) = exp 𝑝 (𝑥) 𝑑𝑥 .

This is the general solution of the linear first order ODE. We need not remember
this formula, but use the method by multiplying the linear ODE with the integrating
factor 𝜇 (𝑥). It is enough remember that 𝜇 0 (𝑥) = 𝜇 (𝑥)𝑝 (𝑥).

(1.21) Example
Find the general solution of the ODE 𝑦 0 − 2𝑥𝑦 = 𝑥.
It is a linear first order ODE with 𝑝 (𝑥) = −2𝑥. Its integrating factor is
∫  ∫  2
𝜇 (𝑥) = exp 𝑝 (𝑥) 𝑑𝑥 = exp (−2𝑥) 𝑑𝑥 = 𝑒 −𝑥 .

Multiplying it with the equation, we get

2 2 2
 2 0 2
𝑒 −𝑥 𝑦 0 − 2𝑥𝑦𝑒 −𝑥 = 𝑥𝑒 −𝑥 ⇒ 𝑒 −𝑥 𝑦 = 𝑥𝑒 −𝑥 .

Integrating, we obtain
∫
−𝑥 2 2 1 2 2 1
𝑒 𝑦= 𝑥𝑒 −𝑥 𝑑𝑥 = − 𝑒 −𝑥 + 𝐶 ⇒ 𝑦 = 𝐶𝑒 𝑥 − .
2 2
(1.22) Example
Solve the IVP 𝑦 0 + 2𝑥𝑦 = 𝑥, 𝑦 (1) = 2.
It is a linear first order ODE with 𝑝 (𝑥) = 2𝑥 so that its integrating factor is
∫  ∫  2
𝜇 (𝑥) = exp 𝑝 (𝑥) 𝑑𝑥 = exp 2𝑥 𝑑𝑥 = 𝑒 𝑥 .
20 MA2020 Classnotes

Multiplying with the equation, we have

2 2 2 2
2
𝑒 𝑥 𝑦 0 + 𝑒 𝑥 2𝑥𝑦 = 𝑒 𝑥 𝑥 ⇒ 𝑒 𝑥 𝑦 = 𝑥𝑒 𝑥 .

Integrating, we obtain
∫
𝑥2 2 1 2
𝑒 𝑦= 𝑥𝑒 𝑥 𝑑𝑥 = 𝑒 𝑥 + 𝐶.
2

As 𝑦 (1) = 2, we get 𝑒 1 · 2 = 21 𝑒 1 + 𝐶 ⇒ 𝐶 = 3𝑒
2. Hence, the solution of the IVP is
2 2
𝑦 = 𝐶𝑒 −𝑥 + 12 = 32 𝑒 1−𝑥 + 12 .

(1.23) Example
Solve the IVP 𝑦 0 + 𝑦 tan 𝑥 = sin(2𝑥), 𝑦 (0) = 1.
It is a linear first order ODE with 𝑝 (𝑥) = tan 𝑥. Its integrating factor is
∫ 
𝜇 (𝑥) = exp tan 𝑥 𝑑𝑥 = exp(log sec 𝑥) = sec 𝑥 .

Multiplying it with the equation, we get

sec 𝑥𝑦 0 + sec 𝑥 tan 𝑥𝑦 = 2 sin 𝑥 ⇒ (sec 𝑥𝑦) 0 = 2 sin 𝑥 .

Integrating, we obtain

sec 𝑥𝑦 = −2 cos 𝑥 + 𝐶 ⇒ 𝑦 = −2 cos2 𝑥 + 𝐶 cos 𝑥 .

Now, 𝑦 (0) = 1 ⇒ −2 + 𝐶 = 1 ⇒ 𝐶 = 3. Then the solution to the IVP is

𝑦 = 3 cos 𝑥 − 2 cos2 𝑥.

There are many ODEs that can be reduced to linear ODEs by suitable substitutions.
One such is the Bernoulli equation:

𝑦 0 + 𝑝 (𝑥)𝑦 = 𝑔(𝑥)𝑦 𝛼 .

Notice that this first order ODE is linear for 𝛼 = 0, 1. So, suppose 𝛼 ≠ 0 and 𝛼 ≠ 1.
Substitute 𝑧 (𝑥) = [𝑦 (𝑥)] 1−𝛼 . Then

𝑧 0 (𝑥) = (1 − 𝛼)𝑦 −𝛼 𝑦 0 = (1 − 𝛼)𝑦 −𝛼 𝑔(𝑥)𝑦 𝛼 − 𝑝 (𝑥)𝑦

 

= (1 − 𝛼) 𝑔(𝑥) − 𝑝 (𝑥)𝑦 1−𝛼 = (1 − 𝛼) 𝑔(𝑥) − 𝑝 (𝑥)𝑧 (𝑥)

= −(1 − 𝛼)𝑝 (𝑥)𝑧 (𝑥) + (1 − 𝛼)𝑔(𝑥).

That is, we have the linear first order ODE

𝑧 0 (𝑥) + (1 − 𝛼)𝑝 (𝑥)𝑧 (𝑥) = (1 − 𝛼)𝑔(𝑥).

First Order ODE 21
(1.24) Example
Solve the Logistic equation 𝑦 0 = 𝐴𝑦 − 𝐵𝑦 2 .
Observe that it is a Bernoulli equation with 𝛼 = 2. We substitute 𝑧 (𝑥) = 𝑦 −1 . Then

𝑧 0 = −𝑦 −2𝑦 0 = −𝑦 −2 (𝐴𝑦 − 𝐵𝑦 2 ) = 𝐵 − 𝐴𝑦 −1 = 𝐵 − 𝐴𝑧 ⇒ 𝑧 0 + 𝐴𝑧 = 𝐵.
∫
For this linear first order ODE, the integrating factor is 𝜇 = exp( 𝐴 𝑑𝑥) = 𝑒 𝐴𝑥 .
Multiplying, we have

0
𝑒 𝐴𝑥 𝑧 0 + 𝐴𝑒 𝐴𝑥 𝑧 = 𝑒 𝐴𝑥 𝐵 ⇒ 𝑒 𝐴𝑥 𝑧 = 𝑒 𝐴𝑥 𝐵.

It gives ∫
𝐵 𝐴𝑥 𝐵
𝐴𝑥
𝑒 𝑧= 𝑒 𝐴𝑥 𝐵 𝑑𝑥 =𝑒 + 𝐶 ⇒ 𝑧 = + 𝐶𝑒 −𝐴𝑥 .
𝐴 𝐴
−1
Substituting 𝑧 = 𝑦 −1 , we get 𝑦 = 𝐵/𝐴 + 𝐶𝑒 −𝐴𝑥 .


Notice that this general solution does not include the solution 𝑦 (𝑥) = 0.
2
Second Order ODE

2.1 Introduction
As we have seen all first order equations could not be solved. We could only solve
exact equations and those which could be reduced to exact equations in two special
cases. The second order equations put more difficult challenges. In general, a
second order equation looks like

𝑓 (𝑥, 𝑦, 𝑦 0, 𝑦 00) = 0.

A special case is when we can solve such an equation for the second derivative. It
then looks like
𝑦 00 = 𝑔(𝑥, 𝑦, 𝑦 0).
Unfortunately, there is no method to solve even this special type. General methods
are available to solve a still special class, and that to partially. The special class is
the second order linear ODEs, which have the form

𝑦 00 + 𝑝 (𝑥)𝑦 0 + 𝑞(𝑥)𝑦 = 𝑟 (𝑥).

When 𝑟 (𝑥) = 0, such an ODE is called homogeneous, otherwise, non-homogeneous.

The initial value problems or IVPs involving second order equations come with
two conditions given at a point such as

Initial values : 𝑦 (𝑥 0 ) = 𝑦0, 𝑦 0 (𝑥 0 ) = 𝑦00 .

Thus, the initial, value problem with a homogeneous linear second order ODE looks
like
𝑦 00 + 𝑝 (𝑥)𝑦 0 + 𝑞(𝑥)𝑦 = 0, 𝑦 (𝑥 0 ) = 𝑦0, 𝑦 0 (𝑥 0 ) = 𝑦00 (2.1.1)
for some given real numbers 𝑥 0, 𝑦0, and 𝑦00 . We are interested in finding a solution of
the IVP in an open interval that contains the point 𝑥 0 . The existence and uniqueness
of a solution to such an initial value problem is guaranteed under certain mild
conditions.

22
Second Order ODE 23
(2.1) Theorem (Existence-Uniqueness)
Let the functions 𝑝 (𝑥) and 𝑞(𝑥) be continuous in the open interval 𝑎 < 𝑥 < 𝑏 and
let 𝑎 < 𝑥 0 < 𝑏. Then, there exists a unique function 𝑦 = 𝑦 (𝑥) defined on the interval
𝑎 < 𝑥 < 𝑏 satisfying the IVP (2.1.1).

We will not prove this theorem. Observe that, in particular, if the initial conditions
are zero conditions, that is, if 𝑦0 = 0 = 𝑦00 , then 𝑦 (𝑥) is the zero function. This
means, if 𝑦 (𝑥) satisfies the homogeneous linear ODE and for some 𝑥 0 in the open
interval, 𝑦 (𝑥 0 ) = 0 = 𝑦 0 (𝑥 0 ), then at all points 𝑥 in the same open interval 𝑦 (𝑥) = 0.

2.2 The Wronskian

Before actually solving the homogeneous linear second order ODE
𝑦 00 + 𝑝 (𝑥)𝑦 0 + 𝑞(𝑥)𝑦 = 0 (2.2.1)
we will discuss some important properties of the solutions, or rather, properties of
the set of all solutions. This will help us in solving the ODE. Due to the Existence-
uniqueness theorem, we assume that 𝑝 (𝑥) and 𝑞(𝑥) are continuous functions in a
nontrivial open interval.

(2.2) Theorem
Let 𝑦1 (𝑥) and 𝑦2 (𝑥) be two solutions of the ODE (2.2.1). Let 𝑐 1, 𝑐 2 be two constants.
Then 𝑦 (𝑥) = 𝑐 1𝑦1 (𝑥) + 𝑐 2𝑦2 (𝑥) is also a solution of (2.2.1).

Proof. Since 𝑦1 (𝑥) and 𝑦2 (𝑥) are solutions of (2.2.1), we have

𝑦100 + 𝑝 (𝑥)𝑦10 + 𝑞(𝑥)𝑦1 = 0 = 𝑦200 + 𝑝 (𝑥)𝑦20 + 𝑞(𝑥)𝑦2 .
Then multiplying the first equation with 𝑐 1 and the second with 𝑐 2 , and adding, we
obtain
(𝑐 1𝑦1 + 𝑐 2𝑦2 ) 00 + 𝑝 (𝑥)(𝑐 1𝑦1 + 𝑐 2𝑦2 ) 0 + 𝑞(𝑥)(𝑐 1𝑦1 + 𝑐 2𝑦2 ) = 0.
Of course, the above result does not hold for non-homogeneous ODEs.

(2.3) Example
The ODE 𝑦 00 + 𝑦 = 0 has solutions 𝑦1 (𝑥) = cos 𝑥 and 𝑦2 (𝑥) = sin 𝑥. From (2.2) it
follows that 𝑦 (𝑥) = 𝐴 cos 𝑥 + 𝐵 sin 𝑥 is also a solution of the ODE. Indeed,
𝑦 00 = (𝐴 cos 𝑥 + 𝐵 sin 𝑥) 00 = (−𝐴 sin 𝑥 + 𝐵 cos 𝑥) 0 = (−𝐴 cos 𝑥 − 𝐵 sin 𝑥) = −𝑦.
It verifies what the theorem states.

We will show that any solution of this ODE is in the form 𝑦 (𝑥) = 𝐴 cos 𝑥 + 𝐵 sin 𝑥.
24 MA2020 Classnotes

Instead of cos 𝑥 and sin 𝑥 suppose we take any two distinct functions, say, 𝑦1 (𝑥)
and 𝑦2 (𝑥), with 𝑦1 (𝑥) ≠ 𝑦2 (𝑥) for some 𝑥. For instance, take 𝑦1 = cos 𝑥 and
𝑦2 = 2 cos 𝑥. Then they are distinct functions but the solution sin 𝑥 cannot be
written as 𝑐 1 cos 𝑥 + 𝑐 2 (2 cos 𝑥). Thus, we need some condition on the functions 𝑦1
and 𝑦2 in order to write any solution as 𝑐 1𝑦1 + 𝑐 2𝑦2 .
Let 𝑦1 (𝑥) and 𝑦2 (𝑥) be two continuously differentiable functions defined on a
nontrivial open interval 𝐼 . The Wronskian of𝑦1 (𝑥) and𝑦2 (𝑥), written𝑊 [𝑦1, 𝑦2 ] (𝑥),
is defined by
𝑊 [𝑦1, 𝑦2 ] (𝑥) = 𝑦1 (𝑥)𝑦20 (𝑥) − 𝑦10 (𝑥)𝑦2 (𝑥).

Notice that the Wronskian is a function of 𝑥. The reason for defining this is the
following result.

(2.4) Theorem
Let 𝑦1 (𝑥) and 𝑦2 (𝑥) be two solutions of the ODE (2.2.1) on a nontrivial open interval
𝐼 with 𝑊 [𝑦1, 𝑦2 ] (𝑥) ≠ 0 for some 𝑥 ∈ 𝐼 . Then the general solution of the ODE
(2.2.1) is 𝑦 (𝑥) = 𝑐 1𝑦1 (𝑥) + 𝑐 2𝑦2 (𝑥), where 𝑐 1, 𝑐 2 are arbitrary constants.

Proof. Let 𝑦 (𝑥) be a solution of (2.2.1). We need to find two constants 𝑐 1, 𝑐 2 such
that 𝑦 (𝑥) = 𝑐 1𝑦1 (𝑥) + 𝑐 2𝑦2 (𝑥) for each 𝑥 ∈ 𝐼 . To this end, let 𝑥 0 ∈ 𝐼 be such that
𝑊 [𝑦1, 𝑦2 ] (𝑥 0 ) ≠ 0. Let 𝑦0 denote 𝑦 (𝑥 0 ) and let 𝑦00 denote 𝑦 0 (𝑥 0 ). If such constants
𝑐 1, 𝑐 2 exist, then evaluating at 𝑥 0 , we must have

𝑐 1𝑦1 (𝑥 0 ) + 𝑐 2𝑦2 (𝑥 0 ) = 𝑦0, 𝑐 1𝑦10 (𝑥 0 ) + 𝑐 2𝑦20 (𝑥 0 ) = 𝑦00 .

Since 𝑊 [𝑦1, 𝑦2 ] (𝑥 0 ) ≠ 0, we have 𝑦1 (𝑥 0 )𝑦20 (𝑥 0 ) − 𝑦10 (𝑥 0 )𝑦2 (𝑥 0 ) ≠ 0. It follows that

there exist unique constants 𝑐 1, 𝑐 2 satisfying the above two linear algebraic equations.
Now, define
𝑧 (𝑥) = 𝑐 1𝑦1 (𝑥) + 𝑐 2𝑦2 (𝑥) for 𝑎 < 𝑥 < 𝑏.

Due to (2.2), 𝑧 (𝑥) is a solution of (2.2.1). Further,

𝑧 (𝑥 0 ) = 𝑐 1𝑦1 (𝑥 0 ) + 𝑐 2𝑦2 (𝑥 0 ) = 𝑦 (𝑥 0 ), 𝑧 0 (𝑥 0 ) = 𝑐 1𝑦10 (𝑥 0 ) + 𝑐 2𝑦20 (𝑥 0 ) = 𝑦00 .

That is, 𝑧 (𝑥) is a solution to the IVP consisting of (2.2.1) and the initial conditions
𝑧 (𝑥 0 ) = 𝑦0 and 𝑧 0 (𝑥 0 ) = 𝑦00 . But 𝑦 (𝑥) is also a solution to the same IVP. Hence, by
the Existence-uniqueness theorem, 𝑦 (𝑥) = 𝑧 (𝑥). That is, 𝑦 (𝑥) = 𝑐 1𝑦1 (𝑥) + 𝑐 2𝑦2 (𝑥)
for 𝑎 < 𝑥 < 𝑏.
Look at the statement in (2.4). It looks that the same conclusion will hold
irrespective of whether the Wronskian is nonzero at 𝑥 0 or at 𝑥 1 as long as 𝑥 0, 𝑥 1 ∈ 𝐼 .
In fact, if the Wronskian of two solutions of (2.2.1) is nonzero at some point in 𝐼 ,
then it is nonzero at each point of 𝐼 . We show this fact below.
Second Order ODE 25
(2.5) Theorem
Let 𝑊 (𝑥) be the Wronskian of two solutions 𝑦1 (𝑥) and 𝑦2 (𝑥) of the ODE (2.2.1).
Then 𝑊 0 (𝑥) + 𝑝 (𝑥)𝑊 (𝑥) = 0.

Proof. Since 𝑦1 and 𝑦2 are solutions of (2.2.1), we have

𝑦100 = −𝑝 (𝑥)𝑦10 − 𝑞(𝑥)𝑦1, 𝑦200 = −𝑝 (𝑥)𝑦20 − 𝑞(𝑥)𝑦2 .

Using these, we obtain

𝑊 0 (𝑥) = (𝑦1𝑦20 − 𝑦10 𝑦2 ) 0 = 𝑦10 𝑦20 + 𝑦1𝑦200 − 𝑦100𝑦2 − 𝑦10 𝑦20

= 𝑦1𝑦200 − 𝑦100𝑦2 = 𝑦1 − 𝑝 (𝑥)𝑦20 − 𝑞(𝑥)𝑦2 + 𝑝 (𝑥)𝑦10 + 𝑞(𝑥)𝑦1 𝑦2

= −𝑝 (𝑥) 𝑦1𝑦20 − 𝑦10 𝑦2 = −𝑝 (𝑥)𝑊 (𝑥).

(2.6) Theorem
Let 𝑝 (𝑥) and 𝑞(𝑥) be continuous on a nontrivial open interval 𝐼 . Let 𝑦1 (𝑥) and
𝑦2 (𝑥) be two solutions of the ODE (2.2.1). Then, 𝑊 [𝑦1, 𝑦2 ] (𝑥) is either identically
zero, or is never zero for any 𝑥 ∈ 𝐼 .

Proof. Take any 𝑥 0 ∈ 𝐼 . Write𝑊 (𝑡) = 𝑊 [𝑦1, 𝑦2 ] (𝑡). By (2.5),𝑊 0 (𝑡) = −𝑝 (𝑡)𝑊 (𝑡)
for 𝑡 ∈ 𝐼 . Separating the variables and integrating from 𝑥 0 to any 𝑥 ∈ 𝐼 , we have
∫𝑥 0 ∫𝑥 ∫𝑥
𝑊 (𝑡)
𝑑𝑡 = − 𝑝 (𝑡) 𝑑𝑡 ⇒ log |𝑊 (𝑥)| − log |𝑊 (𝑥 0 )| = − 𝑝 (𝑡) 𝑑𝑡
𝑥 0 𝑊 (𝑡) 𝑥0 𝑥0
 ∫𝑥 
⇒ |𝑊 (𝑥)| = |𝑊 (𝑥 0 )| exp − 𝑝 (𝑡) 𝑑𝑡 .
𝑥0

The exponential term is never zero. Thus, 𝑊 (𝑥) = 0 iff 𝑊 (𝑥 0 ) = 0. That is, 𝑊 (𝑥)
is either identically zero or is never zero for any 𝑥 ∈ 𝐼 .
∫𝑥

The formula |𝑊 (𝑥)| = |𝑊 (𝑥 0 )| exp − 𝑥 𝑝 (𝑡) 𝑑𝑡 derived in the proof of (2.6) is
0
called Abel’s formula.
Caution: The Wronskian of any two arbitrary functions need not have the property
proved in (2.6). It so happens only for solutions 𝑦1 and 𝑦2 of a homogeneous linear
second order ODE. For instance, consider 𝑦1 (𝑥) = 𝑥 and 𝑦2 (𝑥) = sin 𝑥. We find
that
𝑊 (𝑥) = 𝑊 [𝑦1, 𝑦2 ] (𝑥) = 𝑥 (sin 𝑥) 0 − 𝑥 0 sin 𝑥 = 𝑥 cos 𝑥 − sin 𝑥 .
Then 𝑊 (0) = 0 but 𝑊 (𝜋/2) = −1. That is, 𝑊 (𝑥) is neither identically zero nor
that it is never zero. It means that the functions 𝑦1 (𝑥) = 𝑥 and 𝑦2 (𝑥) = sin 𝑥 cannot
both be solutions of the same homogeneous linear second order ODE.
Suppose 𝑦 ( 𝑥) is a solution of (2.2.1). Then for any constant 𝑐, the function
𝑦2 (𝑥) = 𝑐𝑦1 (𝑥) is also a solution. We see that their Wronskian
𝑊 [𝑦1, 𝑦2 ] (𝑥) = 𝑦1 (𝑐𝑦1 ) 0 − 𝑦10 (𝑐𝑦1 ) = 0.
26 MA2020 Classnotes

That is, when one solution is a constant multiple of the other, then their Wronskian
is zero. We show that the converse is also true.

(2.7) Theorem
Let 𝑦1 (𝑥) and 𝑦2 (𝑥) be solutions of the ODE (2.2.1) on a nontrivial open interval
𝐼 . Suppose 𝑊 [𝑦1, 𝑦2 ] (𝑥 0 ) = 0 for some 𝑥 0 ∈ 𝐼 . Then, one of these solutions is a
constant multiple of the other.

Proof. Since𝑊 [𝑦1, 𝑦2 ] (𝑥 0 ) = (𝑦1𝑦20 −𝑦10 𝑦2 )(𝑥 0 ) = 0, the linear algebraic equations

𝑐 1𝑦1 (𝑥 0 ) + 𝑐 2𝑦2 (𝑥 0 ) = 0, 𝑐 1𝑦10 (𝑥 0 ) + 𝑐 2𝑦20 (𝑥 0 ) = 0

have a nontrivial solution. That is, there exist constants 𝑐 1, 𝑐 2 not both zero such
that both the equations above are satisfied. With this choice of 𝑐 1, 𝑐 2 , write 𝑦 (𝑥) =
𝑐 1𝑦1 (𝑥) + 𝑐 2𝑦2 (𝑥). By (2.2), 𝑦 (𝑥) is a solution of (2.2.1). The above two equations
imply that 𝑦 (𝑥 0 ) = 0 and 𝑦 0 (𝑥 0 ) = 0. Thus, by the Existence-uniqueness theorem,
the IVP consisting of (2.2.1) and these two initial conditions has a unique solution.
However, the zero function is a solution of this IVP. Hence, 𝑦 (𝑥) = 0, the zero
function. That is,

𝑐 1𝑦1 (𝑥) + 𝑐 2𝑦2 (𝑥) = 0 for each 𝑥 ∈ 𝐼 .

Now, if 𝑐 1 ≠ 0, then 𝑦1 (𝑥) = −(𝑐 2 /𝑐 1 )𝑦2 (𝑥); and if 𝑐 2 ≠ 0, then 𝑦2 (𝑥) =

−(𝑐 1 /𝑐 2 )𝑦1 (𝑥). In either case, one is a multiple of the other.
Again, we must remember that the above result is true only for solutions𝑦1 (𝑥), 𝑦2 (𝑥)
of a homogeneous linear second order ODE. It need not be true for arbitrary func-
tions 𝑦1 (𝑥) and 𝑦2 (𝑥).
Let 𝑦1 (𝑥) and 𝑦2 (𝑥) be two functions defined on an open interval 𝑎 < 𝑥 < 𝑏. We
say that the functions 𝑦1, 𝑦2 are linearly dependent iff one of them is a constant
multiple of the other. We say that 𝑦1, 𝑦2 are linearly independent iff they are not
linearly dependent.
Further, two solutions 𝑦1 (𝑥) and 𝑦2 (𝑥) of (2.2.1) are said to form a fundamental
set of solutions iff any solution of the ODE is expressible in the form 𝑐 1𝑦1 + 𝑐 2𝑦2
for suitable constants 𝑐 1, 𝑐 2 .
Using these terminology, we can summarize our results as in the following.

(2.8) Theorem
Let 𝑦1 (𝑥) and 𝑦2 (𝑥) be solutions of the ODE (2.2.1) in a nontrivial open interval
𝐼 , where the functions 𝑝 (𝑥) and 𝑞(𝑥) are continuous. Then the following are
equivalent:
(1) 𝑦1 (𝑥) and 𝑦2 (𝑥) are linearly independent.
(2) 𝑊 [𝑦1, 𝑦2 ] (𝑥) ≠ 0 for some 𝑥 ∈ 𝐼 .
Second Order ODE 27
(3) 𝑊 [𝑦1, 𝑦2 ] (𝑥) ≠ 0 for every 𝑥 ∈ 𝐼 .
(4) 𝑦1 (𝑥) and 𝑦2 (𝑥) form a fundamental set of solutions for (2.2.1).
(2.9) Example
The ODE 𝑦 00 + 𝑦 = 0 has solutions 𝑦1 (𝑥) = cos 𝑥 and 𝑦2 (𝑥) = sin 𝑥. So,
𝑦 (𝑥) = 𝐴 cos 𝑥 + 𝐵 sin 𝑥 is also a solution of the ODE. We compute the Wronskian
of 𝑦1 (𝑥) and 𝑦2 (𝑥) for any 𝑥 in a nontrivial open interval 𝐼 :
𝑊 [𝑦1, 𝑦2 ] (𝑥) = 𝑦1𝑦20 − 𝑦10 𝑦2 = cos 𝑥 · cos 𝑥 − (− sin 𝑥) sin 𝑥 = 1 ≠ 0.
Therefore, these two functions form a fundamental set; that is, any solution of the
ODE 𝑦 00 + 𝑦 = 0 is in the form 𝑐 1 cos 𝑥 + 𝑐 2 sin 𝑥 for some constants 𝑐 1 and 𝑐 2 .

2.3 Constant coefficients

Consider the simpler case of the ODE (2.2.1), where both 𝑝 (𝑥) and 𝑞(𝑥) are
constants. We may rewrite the simpler case as
𝑎𝑦 00 + 𝑏𝑦 0 + 𝑐𝑦 = 0. (2.3.1)
Since the ODE is of second order, we implicitly assume that 𝑎 ≠ 0. The theorems
of the last section say that there are two linearly independent solutions which may
be used to express all solutions. Unfortunately, the results do not tell us how to
obtain a solution. We will have some sort of guess work. Observe that the functions
𝑦 (𝑥), 𝑦 0 (𝑥) and 𝑦 00 (𝑥) should be such that they cancel among themselves and give
us 0.
For example, if 𝑦 (𝑥) is 𝑥 9 , then 𝑦 0 (𝑥) is a constant times 𝑥 8 and 𝑦 00 is a constant
times 𝑦 7 . They cannot cancel to give us 0. If 𝑦 (𝑥) is cos 𝑥, then 𝑦 0 (𝑥) will be a
constant multiple of sin 𝑥 and 𝑦 00 a constant multiple of cos 𝑥. Again, this is not a
right candidate. If 𝑦 (𝑥) is an exponential, say 𝑒 𝜆𝑥 , then 𝑦 0 (𝑥) and 𝑦 00 (𝑥) are also
constant times 𝑒 𝜆𝑥 . It looks this is a possible choice. So, let us try 𝑦 (𝑥) = 𝑒 𝜆𝑥 . Then
𝑦 0 (𝑥) = 𝜆𝑒 𝜆𝑥 and 𝑦 00 = 𝜆 2𝑒 𝜆𝑥 . Substituting these in (2.3.1), we get
𝑎(𝑒 𝜆𝑥 ) 00 + 𝑏 (𝑒 𝜆𝑥 ) 0 + 𝑐𝑒 𝜆𝑥 = 0 ⇒ (𝑎𝜆 2 + 𝑏𝜆 + 𝑐)𝑒 𝜆𝑥 = 0.
Thus, 𝑦 (𝑥) = 𝑒 𝜆𝑥 is a solution of (2.3.1) iff
𝑎𝜆 2 + 𝑏𝜆 + 𝑐 = 0. (2.3.2)
This equation is called the characteristic equation of (2.3.1). It has two roots
𝜆1, 𝜆2 given by
√ √
−𝑏 + 𝑏 2 − 4𝑎𝑐 −𝑏 − 𝑏 2 − 4𝑎𝑐
𝜆1 = , 𝜆2 = .
2𝑎 2𝑎
28 MA2020 Classnotes

Depending on the sign of 𝑏 2 − 4𝑎𝑐 we have three different cases.

Case 1 (Distinct Real Roots): First, suppose that 𝑏 2 − 4𝑎𝑐 > 0.
Then 𝜆1, 𝜆2 ∈ R and 𝜆1 ≠ 𝜆2 . We know that 𝑦1 (𝑥) = 𝑒 𝜆1𝑥 and 𝑦2 (𝑥) = 𝑒 𝜆2𝑥 are two
distinct solutions of (2.3.1). Now,

0
0
𝑊 [𝑦1, 𝑦2 ] (𝑥) = 𝑒 𝜆1𝑥 𝑒 𝜆2𝑥 − 𝑒 𝜆1𝑥 𝑒 𝜆2𝑥 = (𝜆2 − 𝜆1 )𝑒 (𝜆1 +𝜆2 )𝑥 .
As 𝜆2 ≠ 𝜆1 , 𝑊 [𝑦1, 𝑦2 ] (𝑥) ≠ 0 for any 𝑥. By (2.8), these two solutions form a
fundamental set. That is, the general solution of the homogeneous linear second
order ODE (2.3.1) is given by
𝑦 (𝑥) = 𝑐 1𝑒 𝜆1𝑥 + 𝑐 2𝑒 𝜆2𝑥 .
Before discussing other cases, we solve some examples.

(2.10) Example
Find the general solution of 𝑦 00 + 5𝑦 0 + 4𝑦 = 0.
It is a homogeneous linear second order ODE with constant coefficients. Its char-
acteristic equation is
𝜆 2 + 5𝜆 + 4 = 0 ⇒ (𝜆 + 1)(𝜆 + 4) = 0.
So, the characteristic roots are 𝜆1 = −1 and 𝜆2 = −4; these are distinct and real.
Thus, 𝑦1 (𝑥) = 𝑒 −𝑥 and 𝑦2 (𝑥) = 𝑒 −4𝑥 form a fundamental set of solutions. That is,
the general solution is given by
𝑦 (𝑥) = 𝑐 1𝑦1 + 𝑐 2𝑦2 = 𝑐 1𝑒 −𝑥 + 𝑐 2𝑒 −4𝑥
where 𝑐 1, 𝑐 2 are arbitrary constants.

(2.11) Example
Solve the IVP: 𝑦 00 + 𝑦 0 − 2𝑦 = 0, 𝑦 (0) = 4, 𝑦 0 (0) = −5.
It is a homogeneous linear second order ODE with constant coefficients. Its char-
acteristic equation is 𝜆 2 + 𝜆 − 2 = 0 ⇒ (𝜆 − 1)(𝜆 + 2) = 0. The characteristic roots
are 𝜆1 = 1 and 𝜆2 = −2. So, the general solution is
𝑦 (𝑥) = 𝑐 1𝑒 𝑥 + 𝑐 2𝑒 −2𝑥 .
The initial conditions give
𝑦 (0) = 𝑐 1 + 𝑐 2 = 4, 𝑦 0 (0) = 𝑐 1 − 2𝑐 2 = −5.
Solving these equations, we have 𝑐 1 = 1, 𝑐 2 = 3. Thus, the solution to the IVP is
𝑦 (𝑥) = 𝑒 𝑥 + 3𝑒 −2𝑥 .

(2.12) Example
Find the general solution of the ODE 𝑦 00 + 4𝑦 0 − 2𝑦 = 0 and then solve the IVP
𝑦 00 + 4𝑦 0 − 2𝑦 = 0, 𝑦 (0) = 1, 𝑦 0 (0) = 2.
Second Order ODE 29
It is a homogeneous linear second order ODE with constant coefficients. Its char-
acteristic equation is
𝜆 2 + 4𝜆 − 2 = 0.
Since 42 − 4(1)(−2) > 0, there are two distinct real characteristic roots
√ √
−4 + 16 + 8 √ −4 − 16 + 8 √
𝜆1 = = −2 + 6, 𝜆2 = = −2 − 6.
2 2
The fundamental set of solutions comprise 𝑦1 (𝑥) = 𝑒 𝜆1𝑥 and 𝑦2 (𝑥) = 𝑒 𝜆2𝑥 . The
general solution is
√ √
𝑦 (𝑥) = 𝑐 1𝑦1 + 𝑐 2𝑦2 = 𝑐 1𝑒 (−2+ 6)𝑥
+ 𝑐 2𝑒 (−2− 6)𝑥

for arbitrary constants 𝑐 1, 𝑐 2 . Using the initial conditions, we have

√ √
𝑐 1 + 𝑐 2 = 1, (−2 + 6)𝑐 1 + (−2 − 6)𝑐 2 = 2.
√ √
Solving these equations, we get 𝑐 1 = 2/ 6 and 𝑐 2 = 1/2 − 2/ 6. Then the solution
of the IVP is 1 √ √
2  1 2 
𝑦 (𝑥) = + √ 𝑒 (−2+ 6)𝑥 − √ 𝑒 −(2+ 6)𝑥 .
2 6 2 6
Case 2 (Complex Conjugate Roots): Suppose that 𝑏 2 − 4𝑎𝑐 < 0.
Then the characteristic roots 𝜆1, 𝜆2 are given by
√
𝑏 4𝑎𝑐 − 𝑏 2
𝜆1 = 𝛼 + 𝛽𝑖, 𝜆2 = 𝛼 − 𝛽𝑖, 𝛼 = − ∈ R, 𝛽 = ∈ R \ {0}.
2𝑎 2𝑎
For the time being, pretend that we be satisfied with complex solutions. Then, as in
Case 1, the two solutions will be
𝑧 1 = 𝑒 (𝛼+𝑖𝛽)𝑥 , 𝑧 2 = 𝑒 (𝛼−𝑖𝛽)𝑥 .
Their linear combinations, that is, any expression of the form 𝑐 1𝑧 1 + 𝑐 2𝑧 2 is also a
solution. In particular,
𝑧1 + 𝑧2 𝑧1 − 𝑧2
𝑦1 = = 𝑒 𝛼𝑥 cos(𝛽𝑥), 𝑦2 = = 𝑒 𝛼𝑥 sin(𝛽𝑥)
2 2𝑖
are also solutions. Notice that 𝑦1, 𝑦2 are real solutions. This suggests we try to show
directly that 𝑦1, 𝑦2 are solutions of the ODE. For 𝑦1 we proceed as follows, using
the values of 𝛼, 𝛽 as obtained earlier:
𝑦1 = 𝑒 𝛼𝑥 cos(𝛽𝑥)
𝑦10 = 𝑒 𝛼𝑥 𝛼 cos(𝛽𝑥) − 𝛽 sin(𝛽𝑥)

𝑦100 = 𝛼𝑒 𝛼𝑥 𝛼 cos(𝛽𝑥) − 𝛽 sin(𝛽𝑥) + 𝑒 𝛼𝑥 − 𝛼𝛽 sin(𝛽𝑥) − 𝛽 2 cos(𝛽𝑥)

⇒ 𝑎𝑦100 + 𝑏𝑦10 + 𝑐𝑦1

 
= 𝑒 𝛼𝑥 cos(𝛽𝑥) 𝑎𝛼 2 − 𝑎𝛽 2 + 𝑏𝛼 + 𝑐 − (2𝑎𝛼 + 𝑏)𝛽 sin(𝛽𝑥)


  2
4𝑎𝑐 − 𝑏 2 

𝛼𝑥 𝑏 −𝑏
=𝑒 𝑎 − +𝑏 + 𝑐 cos(𝛽𝑥) = 0.
4𝑎 2 4𝑎 2 2𝑎
30 MA2020 Classnotes

That is, 𝑦1 (𝑥) is a solution to the ODE. Similarly, it is easily verified that 𝑦2 (𝑥) is
also a solution of the ODE. Clearly, these two solutions are linearly independent.
Hence, the general solution is given by


𝑦 (𝑥) = 𝑒 𝛼𝑥 𝑐 1 cos(𝛽𝑥) + 𝑐 2 sin(𝛽𝑥) .

Observe that 𝑒 (𝛼+𝑖𝛽)𝑥 = 𝑦1 (𝑥) + 𝑖𝑦2 (𝑥) and 𝑒 (𝛼−𝑖𝛽)𝑥 = 𝑦1 (𝑥) − 𝑖𝑦2 (𝑥). Thus, the
two linearly independent solutions are the real and imaginary parts of 𝑒 𝜆𝑥 where 𝜆
is a complex characteristic root.

(2.13) Example
Find the general solution of 4𝑦 00 + 4𝑦 0 + 5𝑦 = 0.
It is a homogeneous linear second order ODE with constant coefficients. Its char-
acteristic equation is 𝜆 2 + 4𝜆 + 5 = 0; its characteristic roots are
1 1
𝜆1 = − + 𝑖, 𝜆2 = − − 𝑖.
2 2
Hence, the two linearly independent solutions are

𝑦1 (𝑥) = 𝑒 −𝑥/2 cos 𝑥, 𝑦2 (𝑥) = 𝑒 −𝑥/2 sin 𝑥 .

Thus, the general solution is 𝑦 (𝑥) = 𝑒 −𝑥/2 (𝑐 1 cos 𝑥 + 𝑐 2 sin 𝑥).

(2.14) Example
Find the solution of the IVP: 𝑦 00 + 2𝑦 0 + 4𝑦 = 0, 𝑦 (0) = 1, 𝑦 0 (0) = 1.
√
The characteristic equation is 𝜆 2 + 2𝜆 + 4 = 0. The characteristic roots are −1 ± 3 𝑖.
Hence, the two linearly independent solutions are
√ √
𝑦1 (𝑥) = 𝑒 −𝑥 cos( 3 𝑥), 𝑦2 (𝑥) = 𝑒 −𝑥 sin( 3 𝑥).
√ √
The general solution is 𝑦 (𝑥) = 𝑒 −𝑥 𝑐 1 cos( 3 𝑥) + 𝑐 2 sin( 3 𝑥) .

The constants 𝑐 1, 𝑐 2 are determined from the initial conditions

√
1 = 𝑦 (0) = 𝑐 1, 1 = 𝑦 0 (0) = −𝑐 1 + 3 𝑐 2 .
√
They give 𝑐 1 = 1 and 𝑐 2 = 2/ 3. So, the solution to the IVP is
√ √ √
𝑦 (𝑥) = 𝑒 −𝑥 cos( 3 𝑥) + (2/ 3) sin( 3 𝑥) .
 

Case 3 (Equal Roots): Suppose 𝑏 2 − 4𝑎𝑐 = 0.

Then the characteristic roots are real and equal; that is, 𝜆1 = 𝜆2 = −𝑏/(2𝑎). We
have at least one solution, namely, 𝑦1 (𝑥) = 𝑒 𝜆1𝑥 = 𝑒 [−𝑏/(2𝑎)]𝑥 . The second solution,
namely, 𝑒 𝜆2𝑥 is same as 𝑦1 (𝑥); and we would not obtain a fundamental set. We use
this known solution to obtain the second one in a clever way.
Second Order ODE 31
If 𝑦2 (𝑥) is another solution so that 𝑦1 (𝑥) and 𝑦2 (𝑥) are linearly independent, then
𝑦2 (𝑥)/𝑦1 (𝑥) is not a constant function. So, we start with 𝑦2 (𝑥) = 𝑦1 (𝑥)𝑢 (𝑥) and
try to determine 𝑢 (𝑥) from the ODE 𝑎𝑦 00 + 𝑏𝑦 0 + 𝑐𝑦 = 0. With this substitution, we
have
𝑦2 = 𝑦1𝑢, 𝑦20 = 𝑦10 𝑢 + 𝑦1𝑢 0, 𝑦200 = 𝑦100𝑢 + 2𝑦10 𝑢 0 + 𝑦1𝑢 00 .
Since 𝑦2 satisfies the ODE, we get

0 = 𝑎𝑦200 + 𝑏𝑦20 + 𝑐𝑦2

= 𝑎(𝑦100𝑢 + 2𝑦10 𝑢 0 + 𝑦1𝑢 00) + 𝑏 (𝑦10 𝑢 + 𝑦1𝑢 0) + 𝑐𝑦1𝑢
= 𝑎(𝑦100 + 𝑏𝑦10 + 𝑐𝑦1 )𝑢 + 𝑎𝑦1𝑢 00 + (2𝑎𝑦10 + 𝑏𝑦1 )𝑢 0 .

As 𝑦1 also satisfies the ODE, we have 𝑎𝑦100 + 𝑏𝑦10 + 𝑐𝑦1 = 0. Further,

𝑏
0 𝑏 −𝑏 − 𝑏 𝑥 𝑏
2𝑎𝑦10 + 𝑏𝑦1 = 2𝑎 𝑒 − 2𝑎 𝑥 + 𝑏𝑒 − 2𝑎 𝑥 = 2𝑎 · 𝑒 2𝑎 + 𝑏𝑒 − 2𝑎 𝑥 = 0.
2𝑎
Then the above equation reduces to 𝑎𝑦1𝑢 00 = 0. Also, 𝑎𝑦1 ≠ 0. Hence 𝑢 00 = 0 of
which one solution is 𝑢 (𝑥) = 𝑥.
It follows that 𝑦2 (𝑥) = 𝑥𝑦1 (𝑥) is another solution of the same ODE. Clearly, 𝑦1 (𝑥)
and 𝑦2 (𝑥) are linearly independent. Therefore, the general solution of (2.3.1) is
given by
𝑏
𝑦 (𝑥) = (𝑐 1 + 𝑐 2𝑥)𝑦1 (𝑥) = (𝑐 1 + 𝑐 2 )𝑒 𝜆1𝑥 , 𝜆1 = − .
2𝑎
As a caution, we should remember that this 𝑦 (𝑥) is not a solution of the ODE if 𝜆1
is not a double root of the characteristic equation.
Observe that we could have tried this solution in the beginning and got it imme-
diately. However, it is good to familiarize with the method followed above. We will
see the use of this method later in a more general setting.

(2.15) Example
Solve the IVP 𝑦 00 + 4𝑦 0 + 4 = 0, 𝑦 (0) = 1, 𝑦 0 (0) = 3.
The characteristic equation is 𝜆 2 + 4𝜆 + 4 = 0. So, the characteristic roots are
𝜆1 = 𝜆2 = −2. Thus, the general solution is

𝑦 (𝑥) = (𝑐 1 + 𝑐 2𝑥)𝑒 −2𝑥 .

The initial conditions imply that

1 = 𝑦 (0) = 𝑐 1, 3 = 𝑦 0 (0) = −2𝑐 1 + 𝑐 2 ⇒ 𝑐 1 = 1, 𝑐 2 = 5.

So, the general solution of the IVP is 𝑦 (𝑥) = (1 + 5𝑥)𝑒 −2𝑥 .

In fact, the derivation similar to that for linear homogeneous second order ODEs
with constant coefficients may be pursued to obtain linearly independent solutions
32 MA2020 Classnotes

for linear homogeneous higher order ODEs with constant coefficients. We summa-
rize the result. Consider the ODE
𝑎𝑛𝑦 (𝑛) + 𝑎𝑛−1𝑦 (𝑛−1) + · · · + 𝑎 1𝑦 0 + 𝑎 0𝑦 = 0.
Its characteristic equation is
𝑎𝑛 𝜆𝑛 + 𝑎𝑛−1𝜆𝑛−1 + · · · + 𝑎 1𝜆 + 𝑎 0 = 0.

1. If 𝜆 is a simple real root of the characteristic equation, then corresponding to

it we take the solution as 𝑒 𝜆𝑥 .
2. If 𝛼 + 𝑖𝛽 and 𝛼 − 𝑖𝛽 are a pair of complex roots of the characteristic equation,
then corresponding to these two roots, we take the two linearly independent
solutions as 𝑎𝛼𝑥 cos(𝛽𝑥) and 𝑒 𝛼𝑥 sin(𝛽𝑥).
3. If 𝜆 is a root of the characteristic equation having multiplicity 𝑚 > 1, then
corresponding to this, we take the 𝑚 linearly independent solutions as 𝑒 𝜆𝑥 ,
𝑥𝑒 𝜆𝑥 , . . . and 𝑥 𝑚−1𝑒 𝜆𝑥 .
As earlier, the general solution is obtained by multiplying these solutions with
arbitrary constants and adding them together.

(2.16) Example
Find the general solution of the ODE
𝑦 (5) − 10𝑦 (4) + 54𝑦 (3) − 132𝑦 00 + 137𝑦 0 − 50𝑦 = 0.
The characteristic equation is
𝜆 5 − 10𝜆 4 + 54𝜆 3 − 132𝜆 2 + 137𝜆 − 50 = 0.
Trying 1 and 2 as possible values of 𝜆, we factor the left hand side as follows:
𝜆 5 − 10𝜆 4 + 54𝜆 3 − 132𝜆 2 + 137𝜆 − 50
= (𝜆 − 1)(𝜆 4 − 9𝜆 3 + 45𝜆 2 − 87𝜆 + 50)
= (𝜆 − 1)(𝜆 − 1)(𝜆 3 − 8𝜆 2 + 37𝜆 − 50)
= (𝜆 − 1) 2 (𝜆 − 2)(𝜆 2 − 6𝜆 + 25)
= (𝜆 − 1) 2 (𝜆 − 2) (𝜆 − 3) 2 + 42 .

Hence, the characteristic equation has a simple root as 𝜆1 = 2, a double root as

𝜆2 = 𝜆3 = 1, and a pair of complex conjugate roots 𝜆4 = 3 + 4𝑖 and 𝜆5 = 3 − 4𝑖. The
corresponding linearly independent solutions are
𝑦1 = 𝑒 2𝑥 , 𝑦2 = 𝑒 𝑥 , 𝑦3 = 𝑥𝑒 𝑥 , 𝑦4 = 𝑒 3𝑥 cos(4𝑥), 𝑦5 = 𝑒 3𝑥 sin(4𝑥).
Therefore, the general solution is
𝑦 (𝑥) = 𝑐 1𝑒 2𝑥 + (𝑐 2 + 𝑐 3𝑥)𝑒 𝑥 + 𝑒 3𝑥 𝑐 4 cos(4𝑥) + 𝑐 5 sin(4𝑥) .


Second Order ODE 33

2.4 Euler-Cauchy Equation

A particular type of ODE, called the Euler-Cauchy equations, do not have constant
coefficients but they can be solved by the methods suitable for constant coefficients.
The Euler-Cauchy equation is a linear second order ODE of the form

𝑥 2𝑦 00 + 𝑎𝑥𝑦 0 + 𝑏𝑦 = 0 for 𝑥 > 0 (2.4.1)

where 𝑎 and 𝑏 are constants. Notice that it is a linear second order ODE but not of
constant coefficients type.
Substitute 𝑡 = log 𝑥 so that 𝑥 = 𝑒 𝑡 . Using the chain rule, we get
𝑑𝑦 𝑑𝑦 𝑑𝑡 𝑑𝑦 1 𝑑𝑦 −𝑡
= = = 𝑒 .
𝑑𝑥 𝑑𝑡 𝑑𝑥 𝑑𝑡 𝑥 𝑑𝑡
𝑑 2𝑦 𝑑  𝑑𝑦 −𝑡  𝑑  𝑑𝑦 −𝑡  𝑑𝑡  𝑑𝑦
−𝑡 𝑑𝑦 −𝑡 1

= 𝑒 = 𝑒 = 𝑒 + (−𝑒 )
𝑑𝑥 2 𝑑𝑥 𝑑𝑡 𝑑𝑡 𝑑𝑡 𝑑𝑥 𝑑𝑡 2 𝑑𝑡 𝑥
 𝑑 2𝑦 𝑑𝑦  1
= − .
𝑑𝑡 2 𝑑𝑡 𝑥 2
Substituting these in (2.4.1), we obtain

𝑑 2𝑦 𝑑𝑦 𝑑𝑦 𝑑 2𝑦 𝑑𝑦
0 = 𝑥 2𝑦 00 + 𝑎𝑥𝑦 0 + 𝑏𝑦 = 2
− + 𝑎 + 𝑏𝑦 = 2
+ (𝑎 − 1) + 𝑏𝑦.
𝑑𝑡 𝑑𝑡 𝑑𝑡 𝑑𝑡 𝑑𝑡
Thus, we have got a linear homogeneous second order ODE with constant coeffi-
cients, whose characteristic equation is

𝜆 2 + (𝑎 − 1)𝜆 + 𝑏 = 0.

Notice that this equation can also be written as

𝜆(𝜆 − 1) + 𝑎𝜆 + 𝑏 = 0.

This equation being the characteristic equation for the ODE with the new variable
𝑡, is called the Auxiliary equation for the original ODE (2.4.1).
We solve the above ODE with constant coefficients having the independent vari-
able as 𝑡, and then substitute back to obtain a solution to (2.4.1) with the independent
variable as 𝑥. The three ensuing cases are as follows.
Case 1: Suppose 𝜆1 ≠ 𝜆2 are the two real roots of the auxiliary equation. Then the
general solution is given by

𝑦 (𝑥) = 𝑐 1𝑒 𝜆1𝑡 + 𝑐 2𝑒 𝜆2𝑡 = 𝑐 1𝑒 𝜆1 log 𝑥 + 𝑐 2𝑒 𝜆2 log 𝑥 = 𝑐 1𝑥 𝜆1 + 𝑐 2𝑥 𝜆2 .

Indeed, it is easily verified that 𝑦1 = 𝑥 𝜆1 and 𝑦2 = 𝑥 𝜆2 satisfy the ODE (2.4.1).

34 MA2020 Classnotes

Case 2: Suppose 𝜆2 = 𝛼 + 𝑖𝛽, 𝜆2 = 𝛼 − 𝑖𝛽 for 𝛼, 𝛽 ∈ R. Then the general solution

is given by (with 𝑡 = log 𝑥)
   
𝑦 (𝑥) = 𝑒 𝛼𝑡 𝑐 1 cos(𝛽𝑡) + 𝑐 2 sin(𝛽𝑡) = 𝑥 𝛼 𝑐 1 cos(𝛽 log 𝑥) + 𝑐 2 sin(𝛽 log 𝑥) .

Also, we can directly verify that 𝑦1 = 𝑥 𝛼 cos(𝛽 log 𝑥) and 𝑦2 = 𝑥 𝛼 sin(𝛽 log 𝑥) are
solutions of (2.4.1).
Case 3: Suppose 𝜆2 = 𝜆1 ∈ R. Then the general solution is given by (𝑡 = log 𝑥)

𝑦 (𝑥) = (𝑐 1 + 𝑐 2𝑡)𝑒 𝜆1𝑡 = (𝑐 1 + 𝑐 2 log 𝑥)𝑥 𝜆1 .

Again, this fact can be verified without going through the details of derivation.
Notice that finally, one solution is obtained in the form 𝑥 𝜆 instead of 𝑒 𝜆𝑥 as used
to be for the constant coefficients case. This is also easy to guess since the second
order derivative is multiplied with 𝑥 2 and the first order derivative is multiplied with
𝑥. If we try a solution in the form 𝑦 = 𝑥 𝜆 , then the equation (2.4.1) yields

0 = 𝑥 2 (𝑥 𝜆 ) 00 + 𝑎𝑥 (𝑥 𝜆 ) 0 + 𝑏 (𝑥 𝜆 ) = 𝑥 𝜆 𝜆(𝜆 − 1) + 𝑎𝜆 + 𝑏 .

Since 𝑥 𝜆 is not the zero function, we get the auxiliary equation

𝜆(𝜆 − 1) + 𝑎𝜆 + 𝑏 = 0.

This is another heuristic way to solve the Euler-Cauchy equation.

(2.17) Example
(1) The ODE 2𝑥 2𝑦 00 + 3𝑥𝑦 0 − 𝑦 = 0 is the Euler-Cauchy equation

𝑥 2𝑦 00 + (3/2)𝑥𝑦 0 − (1/2)𝑦 = 0.

Its auxiliary equation 𝜆(𝜆 − 1) + (3/2)𝜆 − (1/2) = 0 has roots 𝜆1 = 1/2 and 𝜆2 = −1.
As in Case 1, the general solution is
√
𝑦 = 𝑐 1 𝑥 + 𝑐 2 /𝑥 .

(2) The ODE 100𝑥 2𝑦 00 + 60𝑥𝑦 0 + 1604𝑦 = 0 is the Euler-Cauchy equation

𝑥 2𝑦 00 + 0.6𝑥𝑦 0 + 16.04𝑦 = 0.

Its auxiliary equation 𝜆(𝜆 − 1) + 0.6𝜆 + 16.04 = 0 has roots 𝜆1 = 0.2 + 4𝑖 and
𝜆2 = 0.2 − 4𝑖. As in Case 2, the general solution is

𝑦 = 𝑥 0.2 𝑐 1 cos(4 log 𝑥) + 𝑐 2 sin(4 log 𝑥) .

 

(3) The Euler-Cauchy equation

𝑥 2𝑦 00 − 5𝑥𝑦 + 9𝑦 = 0
Second Order ODE 35
has the auxiliary equation 𝜆(𝜆 − 1) − 5𝜆 + 9 = 0. The roots of the auxiliary equation
are 𝜆1 = 𝜆2 = 3. As in Case 3, the general solution is
𝑦 = 𝑥 3 (𝑐 1 + 𝑐 2 log 𝑥).
(2.18) Example
Solve the ODE 4𝑥 2𝑦 00 + 4𝛼𝑥𝑦 0 + (𝛼 − 1) 2𝑦 = 0 for 𝑥 > 0.
This is an Euler-Cauchy equation with 𝑎 = 𝛼 and 𝑏 = (𝛼 − 1) 2 /4. The auxiliary
equation is 𝜆(𝜆 − 1) + 𝛼𝜆 + (𝛼 − 1) 2 /4 = 0 or 𝜆 2 + (𝛼 − 1)𝜆 + (𝛼 − 1) 2 /4 = 0. Its
roots are 𝜆1 = 𝜆2 = (1 − 𝛼)/2. Hence, the general solution is
𝑦 = 𝑥 (1−𝛼)/2 (𝑐 1 + 𝑐 2 log 𝑥).

2.5 Reduction of order

Consider the homogeneous linear second order ODE
𝑦 00 + 𝑝 (𝑥)𝑦 0 + 𝑞(𝑥)𝑦 = 0. (2.5.1)
In this ODE, the coefficients of 𝑦 and 𝑦 0 are functions of 𝑥. The method of taking
characteristic equations will not apply to this case. Unfortunately, there is no simple
method to solve these “variable coefficients" type of ODEs. However, methods
exist to get a second solution if one solution is known so that the two would become
linearly independent. This method is a simple adaptation of the same for the
"constants coefficients" case.
So, suppose 𝑦1 (𝑥) is a solution to (2.5.1). That means
𝑦100 + 𝑝 (𝑥)𝑦10 + 𝑞(𝑥)𝑦1 = 0.
We wish to determine a second solution 𝑦2 (𝑥) so that 𝑦1 and 𝑦2 are linearly inde-
pendent. That means, if 𝑦2 (𝑥) = 𝑦1 (𝑥)𝑢 (𝑥), then the function 𝑢 (𝑥) should not be a
constant function. We thus assume that
𝑦2 (𝑥) = 𝑦1 (𝑥)𝑢 (𝑥)
for some function 𝑢 (𝑥) which we wish to find. Now,
𝑦20 = 𝑦10 𝑢 + 𝑦1𝑢 0, 𝑦200 = 𝑦100𝑢 + 2𝑦10 𝑢 0 + 𝑦1𝑢 00 .
Then the ODE (2.5.1) becomes
0 = 𝑦200 + 𝑝 (𝑥)𝑦20 + 𝑞(𝑥)𝑦2
= 𝑦100𝑢 + 2𝑦10 𝑢 0 + 𝑦1𝑢 00 + 𝑝 (𝑥)(𝑦10 𝑢 + 𝑦1𝑢 0) + 𝑞(𝑥)𝑦1𝑢
= 𝑦1𝑢 00 + 2𝑦10 + 𝑝 (𝑥)𝑦1 𝑢 0 + 𝑦100 + 𝑝 (𝑥)𝑦10 + 𝑞(𝑥)𝑦1 𝑢

= 𝑦1𝑢 00 + 2𝑦10 + 𝑝 (𝑥)𝑦1 𝑢 0

36 MA2020 Classnotes

We see that 𝑦2 (𝑥) = 𝑦1 (𝑥)𝑢 (𝑥) is a solution of (2.5.1) provided 𝑣 (𝑥) = 𝑢 0 (𝑥) satisfies

𝑦1𝑣 0 + 2𝑦10 + 𝑝 (𝑥)𝑦1 𝑣 = 0.

This is a linear first order equation. Its solution is

 ∫  0
𝑦1 (𝑥)  
𝑣 (𝑥) = 𝑐 exp − 2 + 𝑝 (𝑥) 𝑑𝑥
𝑦1 (𝑥)
∫ 0
 ∫   𝑦1 (𝑥) 
= 𝑐 exp − 𝑝 (𝑥) 𝑑𝑥 exp − 2 𝑑𝑥
𝑦1 (𝑥)
 ∫ 
𝑐 exp − 𝑝 (𝑥) 𝑑𝑥
= .
𝑦12 (𝑥)

Since we are interested in only one function 𝑢 (𝑥), we set the constant 𝑐 = 1 in the
above and obtain  ∫ 
exp − 𝑝 (𝑥) 𝑑𝑥
𝑢 0 (𝑥) = 𝑣 (𝑥) = .
𝑦12 (𝑥)
Integrating this and setting the arbitrary constant to 0, we obtain the function 𝑢 (𝑥).
Therefore, the second solution 𝑦2 (𝑥) is given by
 ∫ 
∫ exp − 𝑝 (𝑥) 𝑑𝑥
𝑦2 (𝑥) = 𝑦1 (𝑥)𝑢 (𝑥) = 𝑦1 (𝑥) 𝑣 (𝑥) 𝑑𝑥, 𝑣 (𝑥) = .
𝑦12 (𝑥)

In this method, we solve a second order equation by solving another first order
equations in 𝑣 (𝑥). This is the reason, the method is named as the method of
reduction of order. However, it applies when we have already got one solution of
the ODE.

(2.19) Example
Find the solution of the IVP

(1 − 𝑥 2 )𝑦 00 + 2𝑥𝑦 0 − 2𝑦 = 0, 𝑦 (0) = 3, 𝑦 0 (0) = −4 for − 1 < 𝑥 < 1.

We see that 𝑦1 (𝑥) = 𝑥 is a solution of the ODE. To get another solution, we use
the method of reduction of order. So, let 𝑦2 (𝑥) = 𝑦1 (𝑥)𝑢 (𝑥). Since in (2.5.1) the
coefficient of 𝑦 00 is taken as 1, we rewrite the ODE as

2𝑥 0 2
𝑦 00 + 𝑦 − 𝑦 = 0.
1 − 𝑥2 1 − 𝑥2
Second Order ODE 37
Now, 𝑝 (𝑥) = 2𝑥/(1 − 𝑥 2 ) and 𝑞(𝑥) = 1/(1 − 𝑥 2 ). Our formula for the second
solution gives
 ∫   ∫ 
2𝑥
exp − 𝑝 (𝑥) 𝑑𝑥 exp − 1−𝑥 2 𝑑𝑥 2
𝑒 log(1−𝑥 ) 1 − 𝑥 2
𝑣 (𝑥) = = = = .
𝑦12 (𝑥) 𝑥2 𝑥2 𝑥2
1 − 𝑥2
∫ ∫ 1 
𝑦2 (𝑥) = 𝑦1 (𝑥)𝑢 (𝑥) = 𝑥 𝑣 (𝑥) 𝑑𝑥 = 𝑥 2
𝑑𝑥 = −𝑥 + 𝑥 = −(1 + 𝑥 2 ).
𝑥 𝑥
Therefore, the general solution is

𝑦 (𝑥) = 𝑐 1𝑥 − 𝑐 2 (1 + 𝑥 2 ).

The initial conditions imply that

3 = 𝑦 (0) = −𝑐 2, −4 = 𝑦 0 (0) = 𝑐 1 .

Hence the solution to the IVP is 𝑦 (𝑥) = 3 − 4𝑥 + 3𝑥 2 .

(2.20) Example
Given that 𝑦1 (𝑥) = 𝑒 𝑎𝑥 is a solution of the ODE 𝑥𝑦 00 − (1 + 3𝑥)𝑦 0 + 3𝑦 = 0 for some
𝑎 ∈ R, find the general solution.
Substituting 𝑦 = 𝑒 𝑎𝑥 in the ODE, we get

𝑥𝑎 2𝑒 𝑎𝑥 − (1 + 3𝑥)𝑎𝑒 𝑥 + 3𝑒 𝑎𝑥 = 0 ⇒ (𝑎 − 3)(𝑥𝑎 − 1) = 0.

Since 𝑎 ∈ R, a constant, we have 𝑎 = 3. So, 𝑦1 (𝑥) = 𝑒 3𝑥 .

The ODE is re-written as
1  3
00 0
𝑦 − +3 𝑦 + 𝑦=0
𝑥 𝑥
so that 𝑝 (𝑥) = −(1/𝑥 + 3) and 𝑞(𝑥) = 3/𝑥.
For the second solution, we set 𝑦2 (𝑥) = 𝑦1 (𝑥)𝑢 (𝑥) and 𝑣 (𝑥) = 𝑢 0 (𝑥). By the
method of reduction of order, we have
 ∫   ∫ 
exp − 𝑝 (𝑥) 𝑑𝑥 exp − (−1/𝑥 + 3) 𝑑𝑥
𝑣 (𝑥) = =
𝑦12 (𝑥) 𝑒 3𝑥
= 𝑒 −6𝑥 exp(log 𝑥 − 3𝑥) = 𝑥𝑒 −9𝑥 .
∫ ∫  𝑥 1  −9𝑥
𝑦2 (𝑥) = 𝑦1 (𝑥) 𝑣 (𝑥) 𝑑𝑥 = 𝑒 3𝑥
(𝑥𝑒 −9𝑥 ) 𝑑𝑥 = 𝑒 3𝑥 − + 𝑒
9 81
1
= (1 − 9𝑥)𝑒 −6𝑥 .
81
Then, the general solution is 𝑦 (𝑥) = 𝑐 1𝑒 3𝑥 + 𝑐 2 (1 − 9𝑥)𝑒 −6𝑥 .
38 MA2020 Classnotes

2.6 Non-homogeneous second order linear ODE

Before proceeding to discuss any method to to solve the general non-homogeneous
linear second order ODE, we will discuss some properties of such equations. These
properties will help us in finding general solutions. We will consider the non-
homogeneous ODE

𝑦 00 + 𝑝 (𝑥)𝑦 0 + 𝑞(𝑥)𝑦 = 𝑟 (𝑥) for 𝑥 ∈ 𝐼 (2.6.1)

where 𝐼 is an open interval and the right hand side function 𝑟 (𝑥) is not the zero
function. Corresponding to this we will consider the homogeneous ODE on the
interval 𝐼 :
𝑦 00 + 𝑝 (𝑥)𝑦 0 + 𝑞(𝑥)𝑦 = 0 for 𝑥 ∈ 𝐼 . (2.6.2)
We assume that the functions 𝑝 (𝑥), 𝑞(𝑥) and 𝑟 (𝑥) are continuous on 𝐼 so that the
corresponding IVPs with initial conditions 𝑦 (𝑥 0 ) = 𝑦0 and 𝑦 0 (𝑥 0 ) = 𝑦00 have unique
solutions for a given 𝑥 0 ∈ 𝐼 .
Relations between solutions of the solutions of (2.6.1-2.6.2) is given by the
following results.

(2.21) Theorem
The difference of any two solutions of (2.6.1) is a solution of (2.6.2).

Proof. Let 𝑢 1 (𝑥) and 𝑢 2 (𝑥) be two solutions of (2.6.1). Then

𝑢 100 + 𝑝 (𝑥)𝑢 10 + 𝑞(𝑥)𝑢 1 = 𝑟 (𝑥), 𝑢 200 + 𝑝 (𝑥)𝑢 20 + 𝑞(𝑥)𝑢 2 = 𝑟 (𝑥).

Subtracting the second from the first, we get

(𝑢 1 − 𝑢 2 ) 00 + 𝑝 (𝑥)(𝑢 1 − 𝑢 2 ) 0 + 𝑞(𝑥)(𝑢 1 − 𝑢 2 ) = 0.

That is, 𝑢 1 − 𝑢 2 is a solution of (2.6.2).

(2.22) Theorem
Let 𝑦1 (𝑥) and 𝑦2 (𝑥) be two linearly independent solutions of the homogeneous equa-
tion (2.6.2) and let 𝜙 (𝑥) be any one (particular) solution of the non-homogeneous
equation (2.6.1). Then every solution of (2.6.1) is in the form

𝑦 (𝑥) = 𝑐 1𝑦1 (𝑥) + 𝑐 2𝑦2 (𝑥) + 𝜙 (𝑥)

for some constants 𝑐 1 and 𝑐 2 .

Proof. Let 𝑦 (𝑥) be any solution of (2.6.1). By (2.21), 𝑦 (𝑥) − 𝜙 (𝑥) is a solution of
(2.6.2). By (2.8), 𝑦1 (𝑥) and 𝑦2 (𝑥) form a fundamental set of solutions of (2.6.2).
Second Order ODE 39
So, 𝑦 (𝑥) − 𝜙 (𝑥) = 𝑐 1𝑦1 (𝑥) + 𝑐 2𝑦2 (𝑥) for some constants 𝑐 1 and 𝑐 2 . It then follows
that 𝑦 (𝑥) = 𝑐 1𝑦1 (𝑥) + 𝑐 2𝑦2 (𝑥) + 𝜙 (𝑥).
Notice that this theorem reduces the problem of finding a general solution of
the non-homogeneous problem to finding two linearly independent solutions of the
homogeneous problem and just one solution of the non-homogeneous problem.

(2.23) Example
Find the general solution of the ODE 𝑦 00 + 𝑦 = 𝑥.
The functions𝑦1 (𝑥) = cos 𝑥 and𝑦2 (𝑥) = sin 𝑥 are two linearly independent solutions
of 𝑦 00 + 𝑦 = 0. The function 𝜙 (𝑥) = 𝑥 satisfies the ODE 𝑦 00 + 𝑦 = 𝑥. Hence, the
general solution of the given ODE is

𝑦 (𝑥) = 𝑐 1 cos 𝑥 + 𝑐 2 sin 𝑥 + 𝑥 .

(2.24) Example
If 𝜙 1 (𝑥) = 𝑥, 𝜙 2 (𝑥) = 𝑥 + 𝑒 𝑥 and 𝜙 3 (𝑥) = 1 + 𝑥 + 𝑒 𝑥 are three solutions of a certain
non-homogeneous linear second order ODE, then find its general solution.
By (2.21), 𝜓 1 (𝑥) = 𝜙 2 − 𝜙 1 = 𝑒 𝑥 and 𝜓 2 (𝑥) = 𝜙 3 − 𝜙 2 = 1 satisfy the corresponding
homogeneous linear second order ODE. Also, 𝜓 1 (𝑥) and 𝜓 2 (𝑥) are linearly inde-
pendent. The function 𝜙 1 (𝑥) is a particular solution of the non-homogeneous ODE.
By (2.22), the general solution is given by

𝑦 (𝑥) = 𝑐 1𝜓 1 (𝑥) + 𝑐 2𝜓 2 (𝑥) + 𝜙 1 (𝑥) = 𝑐 1𝑒 𝑥 + 𝑐 2 + 𝑥

where 𝑐 1, 𝑐 2 are arbitrary constants.

To sum up, we know how to solve a homogeneous linear second order ODE with
constant coefficients. For the variable coefficients case, if we already know one
solution, then we can find another solution linearly independent with the known one
by the method of reduction of order. For the non-homogeneous case, we also need
a particular solution.

2.7 Variation of parameters

How do we find a particular solution of a non-homogeneous linear second order
ODE? We discuss a method that can compute such a particular solution from the
two linearly independent solutions of the corresponding homogeneous ODE.
We consider the non-homogeneous ODE

𝑦 00 + 𝑝 (𝑥)𝑦 0 + 𝑞(𝑥)𝑦 = 𝑟 (𝑥) for 𝑥 ∈ 𝐼 (2.7.1)

40 MA2020 Classnotes

where the functions 𝑝 (𝑥), 𝑞(𝑥) and 𝑟 (𝑥) are continuous on the open interval 𝐼 .
Let 𝑦1 (𝑥) and 𝑦2 (𝑥) be two linearly independent solutions of the corresponding
homogeneous ODE

𝑦 00 + 𝑝 (𝑥)𝑦 0 + 𝑞(𝑥)𝑦 = 0 for 𝑥 ∈ 𝐼 . (2.7.2)

We know that any function in the form 𝑐 1𝑦1 (𝑥) + 𝑐 2𝑦2 (𝑥) is a solution of (2.7.2).
Idea : if we treat the constants 𝑐 1, 𝑐 2 as functions, then probably we will be able to
satisfy (2.7.1). So, we try to determine two functions 𝑢 1 (𝑥) and 𝑢 2 (𝑥) so that

𝜙 (𝑥) = 𝑢 1 (𝑥)𝑦1 (𝑥) + 𝑢 2 (𝑥)𝑦2 (𝑥)

is a solution of (2.7.1). It looks that the idea is a bogus one, since for determining one
function 𝜙 (𝑥) we now need to determine two functions 𝑢 1 (𝑥) and 𝑢 2 (𝑥). However,
it also implicitly says that we have probably some freedom in choosing these two
functions. That is, if necessary we can impose some more conditions suitably so
that our work becomes simple. With 𝜙 (𝑥) in the above form, we see that

𝜙 0 (𝑥) = 𝑢 1𝑦1 + 𝑢 2𝑦2 ) 0 = 𝑢 1𝑦10 + 𝑢 2𝑦20 + 𝑢 10 𝑦1 + 𝑢 20 𝑦2 .

We will also require 𝜙 00 (𝑥). It will involve the second order derivatives of the
unknown functions 𝑢 1 and 𝑢 2 . In order to make our work simple, we impose the
condition that
𝑢 10 (𝑥)𝑦1 (𝑥) + 𝑢 20 (𝑥)𝑦2 (𝑥) = 0.
Then, 𝜙 0 (𝑥) = 𝑢 1𝑦10 + 𝑢 2𝑦20 . As 𝜙 (𝑥) satisfies (2.7.1),we have

𝑟 (𝑥) = 𝜙 00 (𝑥) + 𝑝 (𝑥)𝜙 0 (𝑥) + 𝑞(𝑥)𝜙 (𝑥)

0
= 𝑢 1𝑦10 + 𝑢 2𝑦20 + 𝑝 (𝑥) 𝑢 1𝑦10 + 𝑢 2𝑦20 + 𝑞(𝑥) 𝑢 1𝑦1 + 𝑢 2𝑦2

= 𝑢 10 𝑦10 + 𝑢 20 𝑦20 + 𝑢 1 𝑦100 + 𝑝𝑦10 + 𝑞𝑦1 + 𝑢 2 𝑦200 + 𝑝𝑦20 + 𝑞𝑦2

= 𝑢 10 𝑦10 + 𝑢 20 𝑦20

The last equality follows since both 𝑦1 (𝑥) and 𝑦2 (𝑥) are solutions of (2.7.2). To sum
up, we see that 𝜙 (𝑥) = 𝑢 1𝑦1 +𝑢 2𝑦2 is a solution of (2.7.1) provided that 𝑢 1 (𝑥), 𝑢 2 (𝑥)
satisfy
𝑦1𝑢 10 + 𝑦2𝑢 20 = 0, 𝑦10 𝑢 10 + 𝑦20 𝑢 20 = 𝑟 (𝑥).
We need to solve these linear equations in the unknowns 𝑢 10 and 𝑢 20 . So, multiply
the first equation by 𝑦20 and second by 𝑦2 , then subtract to get

𝑦1𝑦20 − 𝑦10 𝑦2 𝑢 10 = −𝑟 (𝑥)𝑦2 .

Similarly, multiplying the first by 𝑦10 and the second by 𝑦1 , then subtracting we get

𝑦1𝑦20 − 𝑦10 𝑦2 𝑢 20 = 𝑟 (𝑥)𝑦1 .

Second Order ODE 41
Recall that the Wronskian is 𝑊 [𝑦1, 𝑦2 ] (𝑥) = 𝑦1𝑦20 − 𝑦10 𝑦2 . Hence, we obtain
𝑟 (𝑥)𝑦2 (𝑥) 𝑟 (𝑥)𝑦1 (𝑥)
𝑢 10 (𝑥) = − , 𝑢 20 (𝑥) = − . (2.7.3)
𝑊 [𝑦1, 𝑦2 ] (𝑥) 𝑊 [𝑦1, 𝑦2 ] (𝑥)
Finally, we get 𝑢 1 (𝑥) and 𝑢 2 (𝑥) by integrating these. Of course, we can take
any suitable constant of integration to make our choices of 𝑢 1, 𝑢 2 simple. This
method of determining a particular solution for a non-homogeneous equation from
a fundamental set of solutions for the corresponding homogeneous equation is called
the method of variation of parameters due to Lagrange.

(2.25) Example
Solve the IVP 𝑦 00 + 𝑦 = tan 𝑡 for −𝜋/2 < 𝑡 < 𝜋/2; with 𝑦 (0) = 1 = 𝑦 0 (0).
The corresponding homogeneous equation 𝑦 00 + 𝑦 = 0 has two linearly independent
solutions as 𝑦1 = cos 𝑥 and 𝑦2 = sin 𝑥. To get a particular solution of the given
non-homogeneous equation, we first compute the Wronskian. Now,
𝑊 [𝑦1, 𝑦2 ] (𝑥) = 𝑦1𝑦20 − 𝑦10 𝑦2 = cos 𝑥 (cos 𝑥) − (− sin 𝑥) sin 𝑥 = 1.
Using variation of parameters, we seek a particular solution 𝜙 (𝑥) in the form
𝜙 (𝑥) = 𝑢 1 (𝑥)𝑦1 + 𝑢 2 (𝑥)𝑦2 = 𝑢 1 (𝑥) cos 𝑥 + 𝑢 2 (𝑥) sin 𝑥
where due to (2.7.3),
𝑢 10 (𝑥) = − tan 𝑥 sin 𝑥, 𝑢 20 (𝑥) = tan 𝑥 cos 𝑥 .
Integrating and ignoring the constants of integration, we have
cos2 𝑥 − 1
∫ ∫
𝑢 1 (𝑥) = tan 𝑥 sin 𝑥 𝑑𝑥 = 𝑑𝑥 = sin 𝑥 − log(sec 𝑥 + tan 𝑥).
∫ ∫ cos 𝑥
𝑢 2 (𝑥) = tan 𝑥 cos 𝑥 𝑑𝑥 = sin 𝑥 𝑑𝑥 = − cos 𝑥
 
𝜙 (𝑥) = sin 𝑥 − log(sec 𝑥 + tan 𝑥) cos 𝑥 − cos 𝑥 sin 𝑥
= − cos 𝑥 log(sec 𝑥 + tan 𝑥).
Then, the general solution of the ODE is given by
𝑦 (𝑥) = 𝑐 1𝑦1 + 𝑐 2𝑦2 + 𝜙 = 𝑐 1 cos 𝑥 + 𝑐 2 sin 𝑥 − cos 𝑥 log(sec 𝑥 + tan 𝑥).
The initial conditions give
1 = 𝑦 (0) = 𝑐 1, 1 = 𝑦 0 (0) = 𝑐 2 − 1 ⇒ 𝑐 1 = 1, 𝑐 2 = 2.
Thus, the solution of the IVP is
𝑦 (𝑥) = cos 𝑥 + 2 sin 𝑥 − cos 𝑥 log(sec 𝑥 + tan 𝑥).
42 MA2020 Classnotes

2.8 Method of undetermined coefficients

The variation of parameters is a very general method to determine a solution of
the non-homogeneous equation. If the coefficients of the unknown variable 𝑦 and
its derivatives are constants and the right hand side function involves exponentials,
polynomials, or trigonometric functions of certain particular forms, then a particular
solution can be determined without resorting to integration.
We consider the non-homogeneous linear second order ODE with constant coef-
ficients:
𝑎𝑦 00 + 𝑏𝑦 0 + 𝑐𝑦 = 𝑟 (𝑥), 𝑎 ≠ 0. (2.8.1)
Its characteristic equation is

𝑎𝜆 2 + 𝑏𝜆 + 𝑐 = 0.

The method of undetermined coefficients asserts that when 𝑟 (𝑥) is in certain form,
the particular solution 𝜙 (𝑥) of the ODE (2.8.1) is of certain form. These statements,
written as ‘Rules’ below follow from the method of variation of parameters. They
are as follows.
Rule 1: Suppose 𝑟 (𝑥) = 𝑝𝑛 (𝑥)𝑒 𝛼𝑥 , where 𝑝𝑛 (𝑥) is a polynomial of degree 𝑛. Then,
the particular solution 𝜙 (𝑥) of (2.8.1) is in the following form, where 𝑢𝑛 (𝑥) is some
polynomial of degree at most 𝑛:
(A) If 𝛼 is not a root of the characteristic equation, then 𝜙 (𝑥) = 𝑢𝑛 (𝑥)𝑒 𝛼𝑥 .
(B) If 𝛼 is a simple root of the characteristic equation, then 𝜙 (𝑥) = 𝑥𝑢𝑛 (𝑥)𝑒 𝛼𝑥 .
(C) If 𝛼 is a double root of the characteristic equation, then 𝜙 (𝑥) = 𝑥 2𝑢𝑛 (𝑥)𝑒 𝛼𝑥 .
A particular case of Rule 1 is worth mentioning. In Rule 1, if 𝛼 = 0, then we get
the following.
Rule 2: Suppose 𝑟 (𝑥) = 𝑝𝑛 (𝑥), a polynomial of degree 𝑛. Then, the particular
solution 𝜙 (𝑥) of (2.8.1) is in the following form:
(A) If 𝑐 ≠ 0, then 𝜙 (𝑥) = 𝑢𝑛 (𝑥).
(B) If 𝑐 = 0, 𝑏 ≠ 0, then 𝜙 (𝑥) = 𝑥𝑢𝑛 (𝑥).
(C) If 𝑐 = 0 = 𝑏, then 𝜙 (𝑥) = 𝑥 2𝑢𝑛 (𝑥).
As earlier, 𝑢𝑛 (𝑥) is a polynomial of degree at most 𝑛.
 
Rule 3: Suppose 𝑟 (𝑥) = 𝑒 𝛼𝑥 𝑝 (𝑥) cos(𝛽𝑥) + 𝑞(𝑥) sin(𝛽𝑥) , where 𝑝 (𝑥), 𝑞(𝑥) are
polynomials. Then, the particular solution 𝜙 (𝑥) of (2.8.1) is in the following form:
(A) If 𝛼 + 𝑖𝛽 is not a root of the characteristic polynomial, then
 
𝜙 (𝑥) = 𝑒 𝛼𝑥 𝑢 (𝑥) cos(𝛽𝑥) + 𝑣 (𝑥) sin(𝛽𝑥) .
Second Order ODE 43
(B) If 𝛼 + 𝑖𝛽 is a root of the characteristic polynomial, then
 
𝜙 (𝑥) = 𝑥𝑒 𝛼𝑥 𝑢 (𝑥) cos(𝛽𝑥) + 𝑣 (𝑥) sin(𝛽𝑥) .

Here, 𝑢 (𝑥) and 𝑣 (𝑥) are some polynomials whose degrees are at most the highest
degree of the polynomials 𝑝 (𝑥) and 𝑞(𝑥).
We emphasize that if one of the polynomials 𝑝 (𝑥) or 𝑞(𝑥) is equal to 0, then
𝑟 (𝑥) involves only one of the terms cos(𝛽𝑥) or sin(𝛽𝑥). In that case, 𝜙 (𝑥) may still
involve both thee terms cos(𝛽𝑥) and sin(𝛽𝑥).
This rule says that we must try to determine the coefficients in 𝑢𝑛 (𝑥) by plugging
in this 𝜙 (𝑥) in (2.8.1).
In Rule 3, if 𝛼 = 0 and the polynomials 𝑝 (𝑥) and 𝑞(𝑥) are constants, we get the
following important case.
Rule 4: Suppose 𝑟 (𝑥) = 𝑑 1 cos(𝛽𝑥) +𝑑 2 sin(𝛽𝑥) for some constants 𝑑 1 and 𝑑 2 . Then
𝜙 (𝑥) is in the following form:
(A) If 𝛽𝑖 is not a root of the characteristic equation, then

𝜙 (𝑥) = 𝐴 cos(𝛽𝑥) + 𝐵 sin(𝛽𝑥).

(B) If 𝛽𝑖 is a root of the characteristic equation, then

 
𝜙 (𝑥) = 𝑥 𝐴 cos(𝛽𝑥) + 𝐵 sin(𝛽𝑥) .

We remark that if 𝑟 (𝑥) is a sum of functions, then their corresponding 𝜙 (𝑥) are
to be added.

(2.26) Example
Find a particular solution of the ODE 𝑦 00 + 𝑦 0 + 𝑦 = 𝑥 2 .
By Rule 2, a particular solution may be tried in the form 𝜙 (𝑥) = 𝐴 + 𝐵𝑥 + 𝐶𝑥 2 . As
𝜙 satisfies the ODE, we obtain

𝑥 2 = 𝜙 00 + 𝜙 0 + 𝜙 = 2𝐶 + (𝐵 + 2𝐶𝑥) + 𝐴 + 𝐵𝑥 + 𝐶𝑥 2
= (𝐴 + 𝐵 + 2𝐶) + (𝐵 + 2𝐶)𝑥 + 𝐶𝑥 2
⇒ 𝐴 + 𝐵 + 2𝐶 = 0, 𝐵 + 2𝐶 = 0, 𝐶 = 1 ⇒ 𝐴 = 0, 𝐵 = −2, 𝐶 = 1

Hence, 𝜙 (𝑥) = −2𝑥 + 𝑥 2 is a particular solution.

(2.27) Example
Find a particular solution of the ODE 𝑦 00 − 3𝑦 0 + 2𝑦 = (1 + 𝑥)𝑒 3𝑥 .
To use Rule 1, we should check whether 3 is a root of the characteristic equation.
44 MA2020 Classnotes

The characteristic equation is 𝜆 2 − 3𝜆 + 2 = 0, and 3 is not its root. So, by Rule

1(A), 𝜙 (𝑥) = (𝐴 + 𝐵𝑥)𝑒 3𝑥 . Then
(1 + 𝑥)𝑒 3𝑥 = 𝜙 00 − 3𝜙 0 + 2𝜙
= 𝑒 3𝑥 (9𝐴 + 6𝐵 + 9𝐵𝑥) − 3(3𝐴 + 𝐵 + 3𝐵𝑥) + 2(𝐴 + 𝐵𝑥)

= 𝑒 3𝑥 (2𝐴 + 3𝐵) + 2𝐵𝑥

⇒ 2𝐵 = 1, 2𝐴 + 3𝐵 = 1 ⇒ 𝐴 = −1/4, 𝐵 = 1/2.
Hence, a particular solution is 𝜙 (𝑥) = (−1/4 + 𝑥/2)𝑒 3𝑥 .

(2.28) Example
Solve the ODE 𝑦 00 − 7𝑦 0 + 6𝑦 = (𝑥 − 2)𝑒 𝑥 .
The characteristic equation is 𝜆 2 − 7𝜆 + 6 = 0 whose roots are 6 and 1. Here, the
right hand side is in the form 𝑝 1 (𝑥)𝑒 𝑥 and 𝛼 = 1 is a simple root of the characteristic
equation. So, we seek a particular solution in the form 𝜙 (𝑥) = 𝑥 (𝐴 + 𝐵𝑥)𝑒 𝑥 .
Plugging in the equation, we obtain
(𝑥 − 2)𝑒 𝑥 = (𝐴𝑥 + 𝐵𝑥 2 ) + (2𝐴 + 4𝐵𝑥) + 2𝐵 − 7(𝐴𝑥 + 𝐵𝑥 2 )
−7(𝐴 + 2𝐵𝑥) + 6(𝐴𝑥 + 𝐵𝑥 2 ) 𝑒 𝑥

= (−5𝐴 + 2𝐵 − 10𝐵𝑥)𝑒 𝑥
⇒ −5𝐴 + 2𝐵 = −2, −10𝐵 = 1 ⇒ 𝐴 = 9/25, 𝐵 = −1/10.
Hence, 𝜙 (𝑥) = 𝑥 (9/25 − 𝑥/10)𝑒 𝑥 is a particular solution. The general solution of
the ODE is 𝑦 (𝑥) = 𝑐 1𝑒 6𝑥 + 𝑐 2𝑒 𝑥 + 𝑥 (9/25 − 𝑥/10)𝑒 𝑥 .

(2.29) Example
Find a particular solution of the ODE 𝑦 00 + 4𝑦 = sin(2𝑥).
The characteristic equation 𝜆 2 + 4 = 0 has roots ±2𝑖. By Rule 4(B), a particular
solution is in the form 𝜙 (𝑥) = 𝑥 𝐴 cos(3𝑥) +𝐵 sin(3𝑥) . Plugging it in the equation,
we get
sin(2𝑥) = 𝜙 00 + 4𝜙
   
= 𝑥 − 4𝐴 cos(2𝑥) − 4𝐵 sin(2𝑥) + − 2𝐴 sin(2𝑥) + 2𝐵 cos(2𝑥)
   
+ − 2𝐴 sin(2𝑥) + 2𝐵 cos(2𝑥) + 4𝑥 𝐴 cos(2𝑥) + 𝐵 sin(2𝑥)
= −4𝐴 sin(2𝑥) + 4𝐵 cos(2𝑥)
Comparing the left and the right hand sides, we get 𝐴 = −1/4 and 𝐵 = 0. Then, a
particular solution is given by 𝜙 (𝑥) = −(𝑥/4) cos(2𝑥).

(2.30) Example
Solve the IVP 𝑦 00 + 2𝑦 0 + 0.75𝑦 = 2 cos 𝑥 − 0.25 sin 𝑥 + 0.09𝑥,
𝑦 (0) = 2.78, 𝑦 0 (0) = −0.43.
Second Order ODE 45
The characteristic equation is 𝜆 2 + 2𝜆 + 0.75 = 0 having roots as 𝜆1 = −1/2 and
𝜆2 = −3/2. Hence, two linearly independent solutions of the homogeneous equation
are 𝑦1 = 𝑒 −𝑥/2 and 𝑦2 = 𝑒 −3𝑥/2 .
The non-homogeneous term is 𝑟 (𝑥) = (2 cos 𝑥 − 0.25 sin 𝑥) + 0.09𝑥. We first find
a particular solution 𝜙 (𝑥) for

𝑦 00 + 2𝑦 0 + 0.75𝑦 = 2 cos 𝑥 − 0.25 sin 𝑥 .

Since 1 is not a root of the characteristic equation, by Rule 4(A), we try a particular
solution in the form 𝜙 (𝑥) = 𝐴 cos 𝑥 + 𝐵 sin 𝑥. Plugging it in the ODE, we get

2 cos 𝑥 − 0.25 sin 𝑥 = 𝜙 00 + 2 0 + 0.75𝜙

= (−𝐴 cos 𝑥 − 𝐵 sin 𝑥) + 2(−𝐴 sin 𝑥 + 𝐵 cos 𝑥) + 0.75(𝐴 cos 𝑥 + 𝐵 sin 𝑥)
= (−𝐴 + 2𝐵 + 0.75) cos 𝑥 + (−𝐵 − 2𝐴 + 0.75) sin 𝑥
⇒ −𝐴 + 2𝐵 + 0.75 = 2, −𝐵 − 2𝐴 + 0.75 = −0.25
⇒ 𝐴 = 0, 𝐵 = 1.

So, 𝜙 (𝑥) = sin 𝑥. Next, we find a particular solution 𝜓 (𝑥) for

𝑦 00 + 2𝑦 0 + 0.75𝑦 = 0.99𝑥 .

By Rule 2, we try 𝜓 (𝑥) = 𝐶 + 𝐷𝑥. As 𝜓 0 = 𝐷 and 𝜓 00 = 0, we get

0.09𝑥 = 𝜓 00 + 2𝜓 0 + 0.75𝜓 = 2𝐷 + 0.75(𝐶 + 𝐷𝑥) ⇒ 𝐶 = −0.32, 𝐷 = 0.12.

Hence, 𝜓 (𝑥) = 0.12𝑥 − 0.32. Therefore, a particular solution of the ODE is

𝜙 + 𝜓 = sin 𝑥 + 0.12𝑥 − 0.32.

That is, the general solution of the ODE is

𝑦 (𝑥) = 𝑐 1𝑒 −𝑥/2 + 𝑐 2𝑒 −3/2 + sin 𝑥 + 0.12𝑥 − 0.32.

The initial conditions imply

1 3
2.78 = 𝑦 (0) = 𝑐 1 + 𝑐 2 − 0.32, −0.4 = 𝑦 0 (0) = − 𝑐 1 − 𝑐 2 + 1 + 0.12.
2 2
Solving it, we obtain 𝑐 1 = 3.1 and 𝑐 2 = 0. So, the solution of the IVP is

𝑦 = 3.1𝑒 −𝑥/2 + sin 𝑥 + 0.12𝑥 − 0.32.

(2.31) Example
Find the general solution of 𝑦 00 − 4𝑦 0 + 4𝑦 = (1 + 𝑥 + 𝑥 2 · · · + 𝑥 25 )𝑒 2𝑥 .
46 MA2020 Classnotes

The characteristic equation is 𝜆 2 − 4𝜆 + 4 = 0 of which the roots are 𝜆1 = 𝜆2 = 2.

Hence, 𝑦1 (𝑥) = 𝑒 2𝑥 and 𝑦2 (𝑥) = 𝑥𝑒 2𝑥 are two linearly independent solutions of
the corresponding homogeneous equation. A particular solution 𝜙 (𝑥) is in the form

𝜙 (𝑥) = 𝑥 2 𝐴0 + 𝐴1𝑥 + · · · + 𝐴25𝑥 25 𝑒 2𝑥

It is of course sheer waste of time to plug in such a 𝜙 (𝑥) in the ODE and try to
evaluate 𝐴0, 𝐴1, . . . , 𝐴25 . Following the method of variation of parameters, we rather
write
𝜙 (𝑥) = 𝑢 (𝑥)𝑒 2𝑥
and plug it in the ODE. It gives

𝜙 0 (𝑥) = 𝑢 0 (𝑥) + 2𝑢 (𝑥) 𝑒 2𝑥 , 𝜙 00 (𝑥) = 𝑢 00 (𝑥) + 4𝑢 0 (𝑥) + 4𝑢 (𝑥) 𝑒 2𝑥 .

As 𝜙 (𝑥) satisfies the ODE, we get

𝜙 00 − 4𝜙 0 + 4𝜙 = 𝑢 00 (𝑥)𝑒 2𝑥 = 1 + 𝑥 + 𝑥 2 · · · + 𝑥 25 𝑒 2𝑥 .

That is, 𝑢 00 (𝑥) = 1 + 𝑥 + 𝑥 2 · · · + 𝑥 25 . Integrating twice and setting the constants of

integration to 0, we have

𝑥2 𝑥3 𝑥 27
𝑢 (𝑥) = + +···+ .
1·2 2·3 26 · 27
Hence, the general solution is
 𝑥2 𝑥3 𝑥 27  2𝑥
𝑦 (𝑥) = 𝑐 1 + 𝑐 2𝑥 + + +···+ 𝑒 .
1·2 2·3 26 · 27
3
Series Solutions

3.1 Introduction
To recall, we could solve a linear homogeneous second order ODE with constant
coefficients some what satisfactorily. For such an ODE with variable coefficients,
we could only get a second solution provided a first solution is already known. How
do we get this first solution? We relied on guess work. The main aim of this chapter
is to obtain a first solution by using power series. We recall some facts about power
series.
A power series about 𝑥 = 𝑥 0 is in the form
∞
Õ
2
𝑎 0 + 𝑎 1 (𝑥 − 𝑥 0 ) + 𝑎 2 (𝑥 − 𝑥 0 ) + · · · = 𝑎𝑛 (𝑥 − 𝑥 0 ) (3.1.1)
𝑛=0

where 𝑎 0, 𝑎 1, . . . are constants.

Each power series has an interval of convergence. That is, there exists 𝑟 ≥ 0 such
that the power series (3.1.1) converges for all 𝑥 with |𝑥 − 𝑥 0 | < 𝑟 and diverges for
all 𝑥 with |𝑥 − 𝑥 0 | > 𝑟 . This number 𝑟 is called the radius of convergence of the
power series (3.1.1).
|𝑎𝑛 |
If lim exists, then the limit is equal to the radius of convergence of (3.1.1).
𝑛→∞ |𝑎𝑛+1 |
  −1
Also, the radius of convergence of (3.1.1) is equal to lim |𝑎𝑛 | 1/𝑛 provided this
𝑛→∞
limit exists in R ∪ {∞}.
Í∞ Í∞
Two power series 𝑛=0 𝑎 0 (𝑥 − 𝑥 0 )𝑛 and 𝑛=0 𝑏𝑛 𝑥 𝑛 are equal iff 𝑎𝑛 = 𝑏𝑛 for
Í∞
each 𝑛 = 0, 1, 2, . . .. In particular, 𝑛=0 𝑎𝑛 (𝑥 − 𝑥 0 ) = 0 iff 𝑎𝑛 = 0 for each
𝑛

𝑛 = 0, 1, 2, . . ..
Two power series can be added and multiplied the following way:
∞
Õ ∞
Õ ∞
Õ
𝑛 𝑛
𝑎𝑛 (𝑥 − 𝑥 0 ) + 𝑏𝑛 (𝑥 − 𝑥 0 ) = (𝑎𝑛 + 𝑏𝑛 )(𝑥 − 𝑥 0 )𝑛 .
𝑛=0 𝑛=0 𝑛=0
∞
Õ ∞
Õ  ∞
Õ
𝑛 𝑛
𝑎𝑛 (𝑥 − 𝑥 0 ) 𝑏𝑛 (𝑥 − 𝑥 0 ) = 𝑐𝑛 (𝑥 − 𝑥 0 )𝑛
𝑛=0 𝑛=0 𝑛=0
where 𝑐𝑛 = 𝑎 0𝑏𝑛 + 𝑎 1𝑏𝑛−1 + · · · + 𝑎𝑛𝑏 0 .

47
48 MA2020 Classnotes
𝑎 0 + 𝑎 1𝑥 + · · ·
If 𝑏 0 ≠ 0, then the quotient of two power series is a power series.
𝑏 0 + 𝑏 1𝑥 + · · ·
The power series (3.1.1) can be differentiated and integrated term by term and the
resultant series has the same radius of convergence. In particular,
∞
Õ 0 ∞
Õ
𝑛 2
0
𝑎𝑛 𝑥 = 𝑎 0 + 𝑎 1𝑥 + 𝑎 2𝑥 + · · · = 𝑎 1 + 2𝑎 2𝑥 + · · · = 𝑛𝑎𝑛 𝑥 𝑛−1 .
𝑛=0 𝑛=0

A function 𝑓 (𝑥) is said to be analytic at 𝑥 = 𝑥 0 iff there exist constants 𝑎 0, 𝑎 1, 𝑎 2, . . .

such that for all 𝑥 in a neighborhood of 𝑥 0 ,
∞
Õ
2
𝑓 (𝑥) = 𝑎 0 + 𝑎 1 (𝑥 − 𝑥 0 ) + 𝑎 2 (𝑥 − 𝑥 0 ) + · · · = 𝑎𝑛 (𝑥 − 𝑥 0 )𝑛 .
𝑛=0

This series is called the Taylor series of the function 𝑓 (𝑥) at 𝑥 = 𝑥 0 and the
coefficients satisfy
𝑎𝑛 (𝑛!) = 𝑓 (𝑛) (𝑥 0 ).
When 𝑥 0 = 0, the Taylor series is called the Maclaurin series. We are familiar with
the following Maclaurin series:
∞
Õ
−1
(1 − 𝑥) = 𝑥𝑛 = 1 + 𝑥 + 𝑥 2 + 𝑥 3 + · · · for |𝑥 | < 1.
𝑛=0
∞
Õ 𝑥𝑛 𝑥2 𝑥3
𝑒𝑥 = = 1+𝑥 + + +··· for 𝑥 ∈ R.
𝑛=0
𝑛! 2 3!
∞
Õ (−1)𝑛 𝑥 2𝑛 𝑥2 𝑥4
cos 𝑥 = =1− + −··· for 𝑥 ∈ R
𝑛=0
(2𝑛)! 2! 4!
∞
Õ (−1)𝑛 𝑥 2𝑛+1 𝑥3 𝑥5
sin 𝑥 = =𝑥− + −··· for 𝑥 ∈ R.
𝑛=0
(2𝑛 + 1)! 3! 5!

3.2 Regular and singular points

Our plan is to plug in a power series in place of 𝑦 (𝑥) in a linear second order
homogeneous ODE and try to evaluate the coefficients 𝑎𝑛 . We hope that if a
solution of the ODE has an analytic solution at a point, then we should be able to
find out the coefficients. Before discussing what will be the general case, let us
consider an example, and try to execute our ideas.

(3.1) Example
Find a power series solution of the first order ODE 𝑦 0 − 𝑦 = 0.
Series Solutions 49
Í∞
Assume that 𝑦 (𝑥) = 𝑛=0 𝑎𝑛 𝑥 𝑛 is a solution of the ODE. Plugging it in the ODE
and using term by term differentiation, we get
∞
Õ ∞
Õ ∞
Õ ∞
Õ ∞
Õ
𝑛−1 𝑛 𝑛−1 𝑛−1
0 = 𝑛𝑥 − 𝑎𝑛 𝑥 = 𝑛𝑥 − 𝑎𝑛−1𝑥 = (𝑛𝑎𝑛 − 𝑎𝑛−1 )𝑥 𝑛−1
𝑛=1 𝑛=0 𝑛=1 𝑛=1 𝑛=1
𝑎𝑛−1
⇒ 𝑎𝑛 = for 𝑛 ≥ 1.
𝑛
We obtain a recurrence relation between the coefficients. It gives
𝑎0 𝑎1 𝑎0 𝑎2 𝑎0 𝑎0
𝑎1 = , 𝑎2 = = , 𝑎3 = = , . . . , 𝑎𝑛 = .
1 2 2! 3 3! 𝑛!
Notice that the constant 𝑎 0 remains arbitrary. Then
∞ ∞ ∞
Õ Õ 𝑎0 Õ 𝑥𝑛
𝑦 (𝑥) = 𝑎𝑛 𝑥 𝑛 = 𝑥 𝑛 = 𝑎0 = 𝑎 0𝑒 𝑥 .
𝑛=0 𝑛=0
𝑛! 𝑛=0
𝑛!

We see that we have obtained the general solution of the ODE.

We wish to apply the power series method to linear second order ODEs. For this
purpose, we consider the following linear second order homogeneous ODE:

𝑦 00 + 𝑝 (𝑥)𝑦 0 + 𝑞(𝑥)𝑦 = 0. (3.2.1)

When the coefficient of 𝑦 00 is 1 as in (3.2.1), we say that the ODE is in its standard
form. The central fact about such equations is that the nature of solutions depend
on the nature of the coefficient functions 𝑝 (𝑥) and 𝑞(𝑥). The ease of obtaining a
solution depends on how smooth are the coefficient functions. To demarcate the
cases, we will need some terminology.
A point 𝑥 0 is said to be an ordinary point of the ODE (3.2.1) iff both the functions
𝑝 (𝑥) and 𝑞(𝑥) are analytic at 𝑥 = 𝑥 0 . An ordinary point is sometimes called a regular
point.
A point 𝑥 0 is called a singular point of the ODE iff it is not an ordinary point of
the ODE. At a singular point at least one of 𝑝 (𝑥) or 𝑞(𝑥) fails to be analytic.
A singular point 𝑥 0 of the ODE (3.2.1) is called a regular singular point iff both
the functions (𝑥 − 𝑥 0 )𝑝 (𝑥) and (𝑥 − 𝑥 0 ) 2𝑞(𝑥) are analytic at 𝑥 0 .
A singular point 𝑥 0 of the ODE (3.2.1) is called an irregular singular point iff
at least one of the functions (𝑥 − 𝑥 0 )𝑝 (𝑥) or (𝑥 − 𝑥 0 ) 2𝑞(𝑥) fails to be analytic at 𝑥 0 .
Roughly speaking at a regular singular point 𝑥 0 , 𝑝 (𝑥) is not worse than (𝑥 − 𝑥 0 ) −1
and 𝑞(𝑥) is not worse than (𝑥 −𝑥 0 ) −2 . The reason for defining regular singular point
is that we can still obtain a solution to the ODE which involves a power series at 𝑥 0 .
We will soon see this in the guise of a theorem.
We should take care to bring a given equation to the form of (3.2.1), which is
called the standard form while deciding about a point being ordinary or singular.
50 MA2020 Classnotes

Further, in some ODEs we will require the behavior of a solution as 𝑥 approaches

∞. Thus, we need to determine whether 𝑥 0 = ∞ is an ordinary point or a regular
singular point or neither. In such a case, we transform the ODE to one by taking
𝑡 = 1/𝑥 and then find what kind of a point 𝑡 = 0 is. So, write 𝑌 (𝑡) = 𝑦 (𝑥) = 𝑦 (1/𝑡).
Then
𝑑𝑌 /𝑑𝑡 𝑑𝑌
𝑦 0 (𝑥) = = −𝑡 2
𝑑𝑥/𝑑𝑡 𝑑𝑡
𝑑  𝑑𝑌  𝑑 2𝑌 𝑑𝑌
𝑦 00 (𝑥) = −𝑡 2 − 𝑡2 = 𝑡 4 2 + 2𝑡 3 .
𝑑𝑡 𝑑𝑡 𝑑𝑡 𝑑𝑡
Then the ODE (3.2.1) reduces to
𝑑 2𝑌
𝑑𝑌
𝑡4 2
+ 2𝑡 3 − 𝑡 2𝑝 (1/𝑡) + 𝑞(1/𝑡)𝑌 (𝑡) = 0.
𝑑𝑡 𝑑𝑡
Next, we say that 𝑥 = ∞ is an ordinary, a regular singular , or an irregular singular
point of the original ODE according as 𝑡 = 0 is a respective point of the above ODE.

(3.2) Example
1. Consider the ODE 𝑥𝑦 00 − 𝑦 = 0. In the standard form of (3.2.1), 𝑝 (𝑥) = 0 and
𝑞(𝑥) = −1/𝑥. Thus 𝑥 0 = 1 is an ordinary point. But 𝑥 0 = 0 is a singular point.
Further, 𝑥𝑝 (𝑥) = 0 and 𝑥𝑞(𝑥) = −1 are analytic at 𝑥 0 = 0. Hence 𝑥 0 = 0 is a
regular singular point.
2. The Legendre’s equation (1 − 𝑥 2 )𝑦 00 − 2𝑥𝑦 0 + 6𝑦 = 0, in standard form, is
2𝑥 0 6
𝑦 00 − 𝑦 + = 0.
1 − 𝑥2 1 − 𝑥2
Here, 𝑝 (𝑥) = −2𝑥/(1 − 𝑥 2 ) and 𝑞(𝑥) = 6/(1 − 𝑥 2 ). The point 𝑥 = 0 is
an ordinary point of the ODE. In fact every point other than 𝑥 = ±1 is an
ordinary point of the ODE. The points 𝑥 = ±1 are its singular points. Further,
(𝑥 −1)𝑝 (𝑥) = 2𝑥/(1+𝑥) and (𝑥 −1) 2𝑞(𝑥) = 6(1−𝑥)/(𝑥 +1) are analytic at 𝑥 0 = 1.
Hence, 𝑥 0 = 1 is a regular singular point. Similarly, (𝑥 + 1)𝑝 (𝑥) = 2𝑥/(𝑥 − 1)
and (𝑥 + 1) 2𝑞(𝑥) = 6(𝑥 + 1)/(1 − 𝑥) are analytic at 𝑥 = −1. So, 𝑥 0 = −1 is a
regular singular point.
3. Consider the ODE (𝑥 + 1) 2𝑦 00 + 𝑦 0 − 𝑦 = 0. Here, 𝑝 (𝑥) = (𝑥 + 1) −2 and
𝑞(𝑥) = −(𝑥 + 1) −2 . Any point 𝑥 0 ≠ −1 is an ordinary point. The point 𝑥 0 = −1
is a singular point. Now, (𝑥 + 1)𝑝 (𝑥) = (𝑥 + 1) −1 is not analytic at 𝑥 0 = −1.
Hence, 𝑥 0 = −1 is an irregular singular point.
4. Consider Airy’s equation 𝑦 00 − 𝑥𝑦 = 0. To classify the point at infinity, we put
𝑡 = 1/𝑥. Here 𝑝 (𝑥) = 0 and 𝑞(𝑥) = −1. The ODE is reduced to
𝑑 2𝑌
𝑑𝑌 𝑑 2𝑌 2 𝑑𝑌 1
𝑡4 + 2𝑡 3
− 𝑡 2
(0) + (−1)𝑌 (𝑡) = 0 or + − 𝑌 (𝑡) = 0.
𝑑𝑡 2 𝑑𝑡 𝑑𝑡 2 𝑡 𝑑𝑡 𝑡 3
Series Solutions 51
Here, 𝑝 (𝑡) = 2/𝑡 and 𝑞(𝑡) = −1/𝑡 3 . Since 𝑡 2𝑞(𝑡) = −1/𝑡 is not analytic at 𝑡 0 = 0,
we conclude that 𝑡 0 = 0 is an irregular singular point. Therefore, 𝑥 0 = ∞ is an
irregular singular point of Airy’s equation.

In the following two sections, we discuss how to obtain a series solution of (3.2.1)
about an ordinary point, and also about a regular singular point. Unfortunately, we
do not have any general method for finding a series solution of a linear homogeneous
second order ODE with variable coefficients when the concerned point is an irregular
singular point.

3.3 Power series solution at an ordinary point

We assume that the point 𝑥 0 is an ordinary point of the ODE (3.2.1). That is, the
coefficient functions 𝑝 (𝑥) and 𝑞(𝑥) have power series expansions at 𝑥 0 . Due to
(3.3), we assume that 𝑦 (𝑥) = 𝑎 0 + 𝑎 1 (𝑥 − 𝑥 0 ) + 𝑎 2 (𝑥 − 𝑥 0 ) 2 + · · · is a solution of the
ODE. Using term by term differentiation, we get the series expansions of 𝑦 0 (𝑥) and
𝑦 00 (𝑥). Substituting these expressions in to (3.2.1) and comparing the coefficients
of powers of 𝑥, we determine the coefficients 𝑎𝑛 except possibly two. These two
constants will remain arbitrary and we would obtain a general solution of (3.2.1).
The following result guarantees that the above method works.
We will use the following result without proof.

(3.3) Theorem
Let 𝑥 0 be a regular point of the ODE (3.2.1). Then the following are true:
(1) There exists a solution 𝑦 (𝑥) of (3.2.1) which is analytic at 𝑥 0 .
(2) The IVP consisting of the ODE (3.2.1) and the initial conditions 𝑦 (𝑥 0 ) = 𝑦0,
𝑦 0 (𝑥 0 ) = 𝑦00 for 𝑦0, 𝑦00 ∈ R has a unique solution 𝑦 (𝑥) which is analytic at 𝑥 0 .
(3) If 𝑝 (𝑥) and 𝑞(𝑥) have Taylor series expansions about 𝑥 = 𝑥 0 convergent for
all 𝑥 with |𝑥 − 𝑥 0 | < 𝜌 for some 𝜌 > 0, then in both (1)-(2), the radius of
convergence of the Taylor series for 𝑦 (𝑥) is at least 𝜌.

(3.4) Example
Solve Legendre’s equation (1 − 𝑥 2 )𝑦 00 − 2𝑥𝑦 0 + 2𝑦 = 0 by power series method.
Here 𝑝 (𝑥) = −2𝑥/(1 − 𝑥 2 ) and 𝑞(𝑥) = 2/(1 − 𝑥 2 ) are analytic at 𝑥 = 0. By (3.3),
there exists a power series solution to the ODE about 𝑥 = 0. So, we assume
∞
Õ
𝑦 (𝑥) = 𝑎𝑛 𝑥 𝑛 .
𝑛=0
52 MA2020 Classnotes

Plugging it in the ODE, we obtain

∞
Õ ∞
Õ ∞
Õ
0 = (1 − 𝑥 2 ) 𝑛(𝑛 − 1)𝑎𝑛 𝑥 𝑛−2 − 2𝑥 𝑛𝑎𝑛 𝑥 𝑛−1 + 2 𝑎𝑛 𝑥 𝑛
𝑛=0 𝑛=0 𝑛=0
∞
Õ ∞
Õ
= 𝑛(𝑛 − 1)𝑎𝑛 𝑥 𝑛−2 + [−𝑛(𝑛 − 1) − 2𝑛 + 2]𝑎𝑛 𝑥 𝑛
𝑛=0 𝑛=0
∞
Õ ∞
Õ
𝑛
= (𝑛 + 2)(𝑛 + 1)𝑎𝑛+2𝑥 − (𝑛 − 1)(𝑛 + 2)𝑎𝑛 𝑥 𝑛
𝑛=0 𝑛=0

𝑛−1
So, coefficient of each power of 𝑥 is 0. It gives 𝑎𝑛+2 = 𝑎𝑛 for 𝑛 ≥ 0.
𝑛+1
This recurrence relation gives 𝑎 3 = 0, 𝑎 5 = 24 𝑎 3 = 0, . . .. That is, all odd
coefficients except 𝑎 1 are 0. And,

1 1 3 13 1
𝑎 2 = −𝑎 0, 𝑎 4 = 𝑎 2 = − 𝑎 0, 𝑎 6 = 𝑎 4 = − 𝑎 0 = 𝑎 0, . . .
3 3 5 35 5

1
The even coefficients are given by 𝑎 2𝑛 = − 𝑎 0 . Hence,
2𝑛 − 1

1 1
𝑦 (𝑥) = 𝑎 1𝑥 + 𝑎 0 (1 − 𝑥 2 − 𝑥 4 − 𝑥 6 − · · · ).
3 5

(3.5) Example
Determine two linearly independent solutions of

3𝑥 0 1
𝑦 00 + 𝑦 + 𝑦 = 0.
1 + 𝑥2 1 + 𝑥2

Then, find the solution 𝑦 (𝑥) of the ODE that satisfies the initial conditions 𝑦 (0) = 2
and 𝑦 0 (0) = 3.
We will use the power series method for solving the IVP. The functions 3𝑥/(1 + 𝑥 2 )
and 1/(1 + 𝑥 2 ) are analytic at 𝑥 = 0. Due to (3.3), we try a solution in the form

∞
Õ
𝑦 (𝑥) = 𝑎𝑛 𝑥 𝑛 = 𝑎 0 + 𝑎 1 𝑥 + 𝑎 2 𝑥 2 + · · ·
𝑛=0

Instead of plugging in the expression for 𝑦 in the ODE, we multiply the ODE with
(1 + 𝑥 2 ) and then put the series for 𝑦 (𝑥). This will make our computations simpler.
Then
Series Solutions 53

0 = (1 + 𝑥 2 )𝑦 00 + 3𝑥𝑦 0 + 𝑦
Õ∞ ∞
Õ ∞
Õ
= (1 + 𝑥 2 ) 𝑛(𝑛 − 1)𝑎𝑛 𝑥 𝑛−2 + 3𝑥 𝑛𝑎𝑛 𝑥 𝑛−1 + 𝑎𝑛 𝑥 𝑛
𝑛=0 𝑛=0 𝑛=0
∞
Õ ∞
Õ
= 𝑛(𝑛 − 1)𝑎𝑛 𝑥 𝑛−2 + [𝑛(𝑛 − 1) + 3𝑛 + 1]𝑎𝑛 𝑥 𝑛
𝑛=0 𝑛=0
∞
Õ ∞
Õ
= (𝑛 + 2)(𝑛 + 1)𝑎𝑛+2𝑥 + 𝑛
(𝑛 + 1) 2𝑎𝑛 𝑥 𝑛 .
𝑛=0 𝑛=0

So, the coefficient of like powers of 𝑥 is 0; it gives (𝑛 +2)(𝑛 +1)𝑎𝑛+2 + (𝑛 +1) 2𝑎𝑛 = 0.
Hence
(𝑛 + 1) 2𝑎𝑛 (𝑛 + 1)𝑎𝑛
𝑎𝑛+2 = − =− for 𝑛 ≥ 0.
(𝑛 + 2)(𝑛 + 1) 𝑛+2
This is a recurrence relation to express 𝑎 2, 𝑎 3, . . . in terms of 𝑎 0 and 𝑎 1 . We choose
two simplest cases: (i) 𝑎 0 = 1, 𝑎 1 = 0; (ii) 𝑎 0 = 0, 𝑎 1 = 1 to obtain two linearly
independent solutions.
(i) 𝑎 0 = 1, 𝑎 1 = 0. Now, 𝑎 3 = 0, 𝑎 5 = 0; in fact, all odd coefficients are 0. The even
coefficients are determined from
𝑎0 1 3𝑎 2 1 · 3
𝑎2 = − = − , 𝑎4 = − = , ....
2 2 4 2·4
Proceeding inductively, we find that

1 · 3 · 5 · · · (2𝑛 − 1) 1 · 3 · · · (2𝑛 − 1)
𝑎 2𝑛 = (−1)𝑛 = (−1)𝑛 .
2 · 4 · 6 · · · (2𝑛) 2𝑛𝑛!

Thus,
∞
Õ 1 · 3 · · · (2𝑛 − 1) 𝑡2 1 · 3 4
𝑦1 (𝑥) = (−1)𝑛 = 1 − + 𝑡 +··· .
𝑛=0
2𝑛𝑛! 2 2·4

(ii) 𝑎 0 = 0, 𝑎 1 = 1. In this case, all even coefficients are 0, and the odd coefficients
are determined from
2𝑎 1 2 4𝑎 3 2 · 4
𝑎3 = − = − , 𝑎5 = − = , ....
3 3 5 3·5
Proceeding inductively, we find that

2 · 4 · · · (2𝑛) (−1)𝑛 2𝑛𝑛!

𝑎 2𝑛+1 = (−1)𝑛 = .
3 · 5 · · · (2𝑛 + 1) 3 · 5 · · · (2𝑛 + 1)
54 MA2020 Classnotes

Thus,
∞
Õ (−1)𝑛 2𝑛𝑛! 2 2·4 5
𝑦2 (𝑥) = 𝑥 2𝑛+1 = 𝑥 − 𝑥 3 + 𝑥 −··· .
𝑛=0
3 · 5 · · · (2𝑛 + 1) 3 3·5

It is easily verified that both the power series for 𝑦1 (𝑥) and 𝑦2 (𝑥) converge for
|𝑥 | < 1 and diverge for |𝑥 | > 1.
Further, observe that by construction, the solutions 𝑦1 (𝑥) and 𝑦2 (𝑥) satisfy

𝑦1 (0) = 1, 𝑦10 (0) = 0, 𝑦2 (0) = 0, 𝑦20 (0) = 1.

Therefore, the initial conditions 𝑦 (0) = 2 and 𝑦 0 (0) = 3 are satisfied by the solution
𝑦 (𝑥) = 2𝑦1 (𝑥) + 3𝑦2 (𝑥).

(3.6) Example
Solve the IVP: (𝑥 2 − 2𝑥)𝑦 00 + 5(𝑥 − 1)𝑦 0 + 3𝑦 = 0, 𝑦 (1) = 7, 𝑦 0 (1) = 3.
The initial conditions are given at 𝑥 = 1. So, we try a solution as a power series
Í∞
about 𝑥 = 1. Set 𝑦 (𝑥) = 𝑎𝑛 (𝑥 − 1)𝑛 . Plugging it in the ODE, we obtain
𝑛=0

0 = (𝑥 2 − 2𝑥)𝑦 00 + 5(𝑥 − 1)𝑦 0 + 3𝑦

∞ ∞ ∞
2
Õ 𝑛−2
Õ
𝑛−1
Õ
= (𝑥 − 1) − 1 𝑛(𝑛 − 1)𝑎𝑛 (𝑥 − 1) + 5(𝑥 − 1) 𝑛𝑎𝑛 (𝑥 − 1) +3 𝑎𝑛 (𝑥 − 1)𝑛
𝑛=0 𝑛=0 𝑛=0
∞
Õ ∞
Õ ∞
Õ
=− 𝑛(𝑛 − 1)𝑎𝑛 (𝑥 − 1)𝑛−2 + 𝑛(𝑛 − 1)𝑎𝑛 (𝑥 − 1)𝑛 + (5𝑛 + 3)𝑎𝑛 (𝑥 − 1)𝑛
𝑛=0 𝑛=0 𝑛=0
∞
Õ ∞
Õ
=− (𝑛 + 2)(𝑛 + 1)𝑎𝑛+2 (𝑥 − 1)𝑛 + (𝑛 2 + 4𝑛 + 3)𝑎𝑛 (𝑥 − 1)𝑛 .
𝑛=0 𝑛=0

So, the coefficient of all powers of (𝑥 − 1) are 0. It gives

𝑛 2 + 4𝑛 + 3 𝑛+3
𝑎𝑛+2 = 𝑎𝑛 = 𝑎𝑛 for 𝑛 ≥ 0.
(𝑛 + 2)(𝑛 + 1) 𝑛+2
Now, 𝑎 0 = 𝑦 (1) = 7 and 𝑎 1 = 𝑦 0 (1)/1! = 3. Using the above recurrence relation,
3 3 5 5·3 7 7·5·3
𝑎 2 = 𝑎 0 = · 7, 𝑎 4 = 𝑎 2 = · 7, 𝑎 6 = 𝑎 4 = · 7, . . .
2 2 4 4·2 6 6·4·2
4 4 6 6·4 8 8·6·4
𝑎 3 = 𝑎 1 = · 3, 𝑎 5 = 𝑎 3 = · 3, 𝑎 7 = 𝑎 5 = · 3, . . .
3 3 5 5·3 7 7·5·3
Proceeding inductively, we find that
3 · 5 · · · (2𝑛 + 1) 4 · 6 · · · (2𝑛 + 2)
𝑎 0 = 7, 𝑎 2𝑛 = · 7, 𝑎 1 = 3, 𝑎 2𝑛+1 = ·3 for 𝑛 ≥ 1.
2 · 4 · · · (2𝑛) 3 · 5 · · · (2𝑛 + 1)
Series Solutions 55
∞
Õ
And, 𝑦 (𝑥) = 𝑎𝑛 (𝑥 − 1)𝑛 , where 𝑎𝑛 s are as shown above.
𝑛=0

We remark that in the recurrence relation for the coefficients, it can very well
happen that 𝑎𝑛+2 depends on 𝑎𝑛−1 , 𝑎𝑛 and 𝑎𝑛+1 . In such a case, we may not be able
to write 𝑎𝑛+2 as an expression in 𝑛.

3.4 Series solution about a regular singular point

Suppose 𝑥 = 0 is a regular singular point of the ODE (3.2.1). Then 𝑥𝑝 (𝑥) and
𝑥 2𝑞(𝑥) have Maclaurin series expansion. This means
𝑝0
𝑝 (𝑥) = + 𝑝 1 + 𝑝 2𝑥 + 𝑝 3𝑥 2 + · · ·
𝑥
𝑞0 𝑞1
𝑞(𝑥) = 2 + + 𝑞 2 + 𝑞 3𝑥 + 𝑞 4𝑥 2 + · · ·
𝑥 𝑥
Moreover, 𝑝 0, 𝑞 0 and 𝑞 1 are nonzero, so that 𝑝 (𝑥) and 𝑞(𝑥) are not analytic at 𝑥 = 0.
In this case, we cannot apply (3.3). In fact, for such equations we do not have a
power series solution at 𝑥 = 0. The following result shows that in such a case a
solution can be obtained in the form of 𝑥 𝑟 times a power series for some real number
𝑟 . But this also is guaranteed only under some more restrictions.

(3.7) Theorem (Frobenius)

Let 𝑥 = 0 be a regular singular point of the ODE (3.2.1) so that the functions 𝑥𝑝 (𝑥)
and 𝑥 2𝑞(𝑥) are analytic at 𝑥 = 0 with power series expansions

𝑥𝑝 (𝑥) = 𝑝 0 + 𝑝 1𝑥 + 𝑝 2𝑥 2 + · · · , 𝑥 2𝑞(𝑥) = 𝑞 0 + 𝑞 1𝑥 + 𝑞 2𝑥 2 + · · ·

which converge for |𝑥 | < 𝜌 for some 𝜌 > 0. Let 𝑟 1 and 𝑟 2 be the roots of the equation
(called the indical equation)

𝑟 (𝑟 − 1) + 𝑝 0𝑟 + 𝑞 0 = 0.

Then the ODE (3.2.1) has two linearly independent solutions in the following form
on the interval 0 < 𝑥 < 𝜌:
(a) If 𝑟 1, 𝑟 2 ∈ R, 𝑟 1 > 𝑟 2 and 𝑟 1 − 𝑟 2 is neither 0 nor a positive integer, then
∞
Õ ∞
Õ
𝑟1 𝑛 𝑟2
𝑦1 (𝑥) = 𝑥 𝑎𝑛 𝑥 , 𝑦2 (𝑥) = 𝑥 𝑏𝑛 𝑥 𝑛 , 𝑎 0 ≠ 0, 𝑏 0 ≠ 0.
𝑛=0 𝑛=0
56 MA2020 Classnotes

(b) If 𝑟 1, 𝑟 2 ∈ R and 𝑟 1 − 𝑟 2 is a positive integer, then

∞
Õ ∞
Õ
𝑦1 (𝑥) = 𝑥 𝑟 1 𝑎𝑛 𝑥 𝑛 , 𝑦2 (𝑥) = 𝑎𝑦1 (𝑥) log 𝑥 + 𝑥 𝑟 2 𝑏𝑛 𝑥 𝑛 , 𝑎 0 ≠ 0.
𝑛=0 𝑛=1

Here, the constant 𝑎 may turn out to be 0.

(c) If 𝑟 1, 𝑟 2 ∈ R and 𝑟 1 = 𝑟 2 , then
∞
Õ ∞
Õ
𝑟1 𝑛 𝑟1
𝑦1 (𝑥) = 𝑥 𝑎𝑛 𝑥 , 𝑦2 (𝑥) = 𝑦1 (𝑥) log 𝑥 + 𝑥 𝑏𝑛 𝑥 𝑛 , 𝑎 0 ≠ 0.
𝑛=0 𝑛=1

(d) If 𝑟 1 = 𝛼 + 𝑖𝛽 and 𝑟 2 = 𝛼 − 𝑖𝛽 with 𝛽 ≠ 0, then

∞
Õ
an xn .



𝑦1 (𝑥) = Re z(x) , y2 (x) = Im z(x) , z(x) = x𝛼+i𝛽
n=0

The indical equation referred to in the above result comes from trying a solution
of the ODE in the form 𝑥 𝑟 times a power series. In fact, we will use the above
theorem to determine the form of the series which could be a solution of the given
ODE. Next, we plug in this series in the ODE and setting the coefficients of all
powers to 0, we determine the coefficients.

(3.8) Example
Find two linearly independent solutions of 2𝑥𝑦 00 + 𝑦 0 + 𝑥𝑦 = 0 for 𝑥 > 0.
Here, 𝑝 (𝑥) = (2𝑥) −1 and 𝑞(𝑥) = 1/2. At 𝑥 = 0, 𝑞(𝑥) is analytic, but 𝑝 (𝑥) is not.
However, 𝑥𝑝 (𝑥) = 1/2 and 𝑥 2𝑞(𝑥) = 𝑥 2 /2 are analytic at 𝑥 = 0. So, 𝑥 = 0 is a
regular singular point of the ODE. We use Frobenius method to get a solution of
the ODE. Since 𝑝 0 = 1/2 and 𝑞 0 = 0, the indical equation gives
𝑟 𝑟 1
𝑟 (𝑟 − 1) + 𝑝 0𝑥 + 𝑞 0 = 𝑟 2 − 𝑟 + = 𝑟 2 + = 0 ⇒ 𝑟 1 = , 𝑟 2 = 0.
2 2 2
Since 𝑟 2 − 𝑟 1 is not an integer, by (3.7)(a) the two linearly independent solutions are
in the form
∞
Õ ∞
Õ
𝑛+1/2
𝑦1 (𝑥) = 𝑎𝑛 𝑥 , 𝑦2 (𝑥) = 𝑏𝑛 𝑥 𝑛 , 𝑎 0 ≠ 0, 𝑏 0 ≠ 0.
𝑛=0 𝑛=0

Instead of determining 𝑎𝑛 s and 𝑏𝑛 s separately, we take any solution 𝑦 (𝑥) as

∞
Õ ∞
Õ
𝑟 𝑛
𝑦 (𝑥) = 𝑥 𝑎𝑛 𝑥 = 𝑎𝑛 𝑥 𝑛+𝑟 where 𝑎 0 ≠ 0
𝑛=0 𝑛=0
Series Solutions 57
and then try to determine the coefficients 𝑎𝑛 by considering two cases 𝑟 = 0 or
𝑟 = 1/2 at an appropriate stage. So, plugging it in the ODE, we obtain

0 = 2𝑥𝑦 00 + 𝑦 0 + 𝑥𝑦
Õ∞ ∞
Õ ∞
Õ
= 2𝑥 (𝑛 + 𝑟 )(𝑛 + 𝑟 − 1)𝑎𝑛 𝑡 𝑛+𝑟 −1 + (𝑛 + 𝑟 )𝑎𝑛 𝑡 𝑛+𝑟 −1 + 𝑥 𝑎𝑛 𝑥 𝑛+𝑟 +1
𝑛=0 𝑛=0 𝑛=0
𝑟 −1
    𝑟
= 2𝑟 (𝑟 − 1)𝑎 0 + 𝑟𝑎 0 𝑥 + 2(1 + 𝑟 )𝑟𝑎 1 + (1 + 𝑟 )𝑎 1 𝑥
∞
Õ
2(𝑛 + 𝑟 )(𝑛 + 𝑟 − 1)𝑎𝑛 + (𝑛 + 𝑟 )𝑎𝑛 + 𝑎𝑛−2 𝑥 𝑛+𝑟 −1 .
 
+
𝑛=2

Setting the coefficient of each power of 𝑥 to 0, we get

1. 2𝑟 (𝑟 − 1)𝑎 0 + 𝑟𝑎 0 = 𝑟 (2𝑟 − 1)𝑎 0 = 0 ⇒ 𝑟 = 0 or 𝑟 = 1/2, as we had got
earlier. In fact, this gives the indical equation.
2. 2(𝑟 + 1)𝑟𝑎 1 + (𝑟 + 1)𝑎 1 = (𝑟 + 1)(2𝑟 + 1)𝑎 1 = 0.
 
3. 2(𝑛 + 𝑟 )(𝑛 + 𝑟 − 1)𝑎𝑛 + (𝑛 + 𝑟 )𝑎𝑛 = (𝑛 + 𝑟 ) 2(𝑛 + 𝑟 ) − 1 𝑎𝑛 = −𝑎𝑛−2 for 𝑛 ≥ 2.
−𝑎𝑛−2
(a) 𝑟 = 0. The recurrence formula (3) gives 𝑎𝑛 = for 𝑛 ≥ 2.
𝑛(2𝑛 − 1)
Since 𝑎 1 = 0, all odd coefficients are 0. The even coefficients are determined from
(3) and they are:
−𝑎 0 −𝑎 2 𝑎0 −𝑎 4 −𝑎 0
𝑎2 = , 𝑎4 = = , 𝑎6 = = , ....
2·3 4·7 2·4·3·7 6 · 11 2 · 4 · 6 · 3 · 7 · 11
Since we will account for constants later, set 𝑎 0 = 1 to get one solution as
∞
𝑥3 𝑥4 Õ (−1)𝑛 𝑥 2𝑛
𝑦1 (𝑥) = 1 − + +··· = 1+ .
2·3 2·4·3·7 𝑛=1
2𝑛𝑛!3 · 7 · · · (4𝑛 − 1)

It is easily verified that this series, a power series, converges for all 𝑥 > 0.
(b) 𝑟 = 1/2. The recurrence formula (3) gives
−𝑎𝑛−2 −𝑎𝑛−2
𝑎𝑛 =   = for 𝑛 ≥ 2.
(𝑛 + 1/2) 2(𝑛 + 1/2) − 1 𝑛(2𝑛 + 1)

All odd coefficients are 0; and the even coefficients are given by
−𝑎 0 −𝑎 2 𝑎0 −𝑎 4 −𝑎 0
𝑎2 = , 𝑎4 = = , 𝑎6 = = , ....
2·5 4·9 2·4·5·9 6 · 13 2 · 4 · 6 · 5 · 9 · 13
Again, setting 𝑎 0 = 1, we have
∞
1/2
 Õ (−1)𝑛 𝑡 2𝑛 
𝑦2 (𝑥) = 𝑥 1+ .
𝑛=1
2𝑛𝑛!5 · 9 · · · (4𝑛 + 1)
58 MA2020 Classnotes

It is easily verified that the series here converges for all 𝑥 > 0. Clearly, 𝑦1 (𝑥)
and 𝑦2 (𝑥) are linearly independent. Then the general solution of the ODE is
𝑦 (𝑥) = 𝑐 1𝑦1 (𝑥) + 𝑐 2𝑦2 (𝑥) for 𝑥 > 0.

In the indical equation 𝑟 (𝑟 − 1) + 𝑝 0𝑟 + 𝑞 0 = 0, the constants 𝑝 0 and 𝑞 0 are the

constant terms in the Maclaurin series expansions of 𝑥𝑝 (𝑥) and 𝑥 2𝑞(𝑥), respectively.
Thus,
𝑝 0 = lim 𝑥𝑝 (𝑥) , 𝑞 0 = lim 𝑥 2𝑞(𝑥) .
   
𝑥→0 𝑥→0
Alternatively, the indical equation is obtained from the ODE by substituting the
Í∞
series 𝑦 (𝑥) = 𝑛=0 𝑎𝑛 𝑥 𝑛+𝑟 in the ODE and then setting the coefficient of the least
power in 𝑥 to 0. In practice, we obtain the indical equation this way.
When the indical equation has a double root or the roots differ by an integer, it
is usually extremely difficult to determine the second solution 𝑦2 (𝑥). In fact, 𝑦2 (𝑥)
there has been obtained by using reduction of order. Sometimes, it is easier to
get 𝑦2 (𝑥) by using the method of reduction of order directly once 𝑦1 (𝑥) is already
available. If that is also difficult, which is often the case, then one only finds a few
terms in the series expansion of 𝑦2 (𝑥).

(3.9) Example
Solve the ODE (𝑥 2 − 𝑥)𝑦 00 + (3𝑥 − 1)𝑦 0 + 𝑦 = 0 for 𝑥 > 0 using Frobenius method.
Here, 𝑝 (𝑥) = (3𝑥 − 1)/(𝑥 2 − 𝑥) is not analytic at 𝑥 = 0. However, 𝑥𝑝 (𝑥) is analytic
at 𝑥 = 0 and 𝑥 2𝑞(𝑥) = 𝑥/(𝑥 − 1) is also analytic at 𝑥 = 0. Hence 𝑥 = 0 is a regular
singular point of the ODE. We try a solution in the form
∞
Õ
𝑦 (𝑥) = 𝑎𝑛 𝑥 𝑛+𝑟 , 𝑎 0 ≠ 0.
𝑛=0

Substituting this in the ODE, we obtain

∞
Õ ∞
Õ
0= (𝑛 + 𝑟 )(𝑛 + 𝑟 − 1)𝑎𝑛 𝑥 𝑛+𝑟 − (𝑛 + 𝑟 )(𝑛 + 𝑟 − 1)𝑎𝑛 𝑥 𝑛+𝑟 −1
𝑛=0 𝑛=0
∞
Õ ∞
Õ ∞
Õ
+3 (𝑛 + 𝑟 )𝑎𝑛 𝑥 𝑛+𝑟 − (𝑛 + 𝑟 )𝑎𝑛 𝑥 𝑛+𝑟 −1 + 𝑎𝑛 𝑥 𝑛+𝑟 .
𝑛=0 𝑛=0 𝑛=0

Equating the coefficient of the least power of 𝑥 to 0, we obtain the indical equation
as
0 = − 𝑟 (𝑟 − 1) − 𝑟 𝑎 0 ⇒ 𝑟 2 = 0.
 

Since 𝑟 = 0 is a double root, by (3.7)(b), the two linearly independent solutions are
of the form:
∞
Õ Õ∞
𝑛
𝑦1 (𝑥) = 𝑎𝑛 𝑥 , 𝑦2 (𝑥) = 𝑦1 (𝑥) log 𝑥 + 𝑏𝑛 𝑥 𝑛 , 𝑎 0 ≠ 0.
𝑛=0 𝑛=0
Series Solutions 59
To find 𝑦1 (𝑥), we take 𝑟 = 0 and equate the coefficient of 𝑥 𝑛 to 0 in the above to
obtain the following recurrence relation:

𝑛(𝑛 − 1)𝑎𝑛 − (𝑛 + 1)𝑛𝑎𝑛+1 + 3𝑛𝑎𝑛 − (𝑛 + 1)𝑎𝑛+1 + 𝑎𝑛 = 0.

It gives 𝑎𝑛+1 = 𝑎𝑛 . By choosing 𝑎 0 = 1, we get one solution as

1
𝑦1 (𝑥) = 1 + 𝑥 + 𝑥 2 + · · · = for |𝑥 | < 1.
1−𝑥

For the second solution, we use reduction of order. From § 2.5, we have
 ∫ 
∫ exp − 𝑝 (𝑥) 𝑑𝑥
𝑦2 (𝑥) = 𝑦1 (𝑥) 𝑣 (𝑥) 𝑑𝑥, 𝑣 (𝑥) = ,
𝑦12 (𝑥)

where the ODE is in standard form, that is, when the coefficient of 𝑦 00 is 1. For our
ODE, (with 𝑥 > 0)
∫ ∫ ∫ 
3𝑥 − 1 2 1
− 𝑝 (𝑥) 𝑑𝑥 = − 𝑑𝑥 = − + 𝑑𝑥 = −2 log |𝑥 − 1| − log 𝑥 .
𝑥 (𝑥 − 1) 𝑥 −1 𝑥
1 1
𝑣 (𝑥) = 2 exp(−2 log |𝑥 − 1| − log 𝑥) = (1 − 𝑥) 2 (𝑥 (𝑥 − 1) 2 ) −1 = .
𝑦1 (𝑥) 𝑥
∫ ∫
1 𝑑𝑥 log 𝑥
𝑦2 (𝑥) = 𝑦1 (𝑥) 𝑣 (𝑥) 𝑑𝑥 = = .
1−𝑥 𝑥 1−𝑥

Hence, the general solution of the ODE is 𝑦 (𝑥) = (1 − 𝑥) −1 (𝑐 1 + 𝑐 2 log 𝑥).

(3.10) Example
Solve the ODE 𝑥 2𝑦 00 + 3𝑥𝑦 0 + (1 − 𝑥)𝑦 = 0 by Frobenius method.
Here, 𝑝 (𝑥) = 3/𝑥 and 𝑞(𝑥) = (1 − 𝑥)/𝑥 2 which are not analytic at 𝑥 = 0 but
𝑥𝑝 (𝑥) = 3 and 𝑥 2𝑞(𝑥) = 1 − 𝑥 are analytic at 𝑥 = 0. Hence, 𝑥 = 0 is a regular
singular point of the ODE. Here, 𝑝 0 = 3 and 𝑞 0 = 1; so the indical equation is

𝑟 (𝑟 − 1) + 3𝑟 + 1 = 𝑟 2 − 𝑟 + 2𝑟 + 1 = (𝑟 + 1) 2 = 0 ⇒ 𝑟 1 = −1, 𝑟 2 = −1.

Since 𝑟 = −1 is a double root, by (3.7)(b), the two lienarly independent solutions of

the ODE are in the form
∞
Õ ∞
Õ
𝑛−1
𝑦1 (𝑥) = 𝑎𝑛 𝑥 , 𝑦2 (𝑥) = 𝑦1 (𝑥) log 𝑥 + 𝑏𝑛 𝑥 𝑛−1, 𝑎 0 ≠ 0.
𝑛=0 𝑛=0
60 MA2020 Classnotes
∞
Í
So, let 𝑦 (𝑥) = 𝑎𝑛 𝑥 𝑛+𝑟 with 𝑎 0 ≠ 0. Putting it in the ODE gives
𝑛=0

∞
Õ ∞
Õ ∞
Õ ∞
Õ
0 = 𝑥2 (𝑛 + 𝑟 )(𝑛 + 𝑟 − 1)𝑎𝑛 𝑥 𝑛+𝑟 −2 + 3𝑥 (𝑛 + 𝑟 )𝑎𝑛 𝑥 𝑛+𝑟 −1 + 𝑎𝑛 𝑥 𝑛+𝑟 − 𝑎𝑛 𝑥 𝑛+𝑟 +1
𝑛=0 𝑛=0 𝑛=0 𝑛=0
∞
Õ Õ∞
= [(𝑛 + 𝑟 )(𝑛 + 𝑟 − 1 + 3) + 1]𝑎𝑛 𝑥 𝑛+𝑟 − 𝑎𝑛 𝑥 𝑛+𝑟 +1
𝑛=0 𝑛=0
Õ∞ ∞
Õ
= (𝑛 + 𝑟 + 1) 2𝑎𝑛 𝑥 𝑛+𝑟 − 𝑎𝑛−1𝑥 𝑛+𝑟
𝑛=1 𝑛=1
∞
Õ
= (𝑟 + 1) 2𝑎 0𝑥 𝑟 + (𝑛 + 𝑟 + 1) 2𝑎𝑛 − 𝑎𝑛−1 𝑥 𝑛+𝑟 .
 
𝑛=1

Setting the coefficients of all powers of 𝑥 to 0, we obtain

𝑎𝑛−1
(𝑟 + 1) 2 = 0, 𝑎𝑛 = for 𝑛 ≥ 1.
(𝑛 + 𝑟 + 1) 2

Since 𝑟 = −1, we have 𝑎𝑛 = 𝑎𝑛−1 /𝑛 2 . Then

𝑎0 𝑎1 𝑎0 𝑎2 𝑎0
𝑎1 = 2
= 𝑎 0, 𝑎 2 = 2 = 2 , 𝑎 3 = 2 = 2 2 , . . .
1 2 2 3 2 3
𝑎0
Proceeding inductively, we obtain 𝑎𝑛 = . Setting 𝑎 0 = 1 we have a solution of
(𝑛!) 2
the ODE as
∞
Õ 𝑥 𝑛−1
𝑦1 (𝑥) = .
𝑛=0
(𝑛!) 2

We do not compute the second solution, but remark that after some cumbersome
computation, the second solution is found to be
∞ ∞
Õ 𝑥 𝑛−1 2 Õ 𝐻𝑛 𝑥 𝑛  1
𝑦2 (𝑥) = 2
log 𝑥 − 1 + 2
where 𝐻𝑛 = 1 + 2 + · · · + 𝑛1 .
𝑛=0
(𝑛!) 𝑥 𝑛=1
(𝑛!)

(3.11) Example
Solve the ODE (𝑥 2 − 𝑥)𝑦 00 − 𝑥𝑦 0 + 𝑦 = 0 for 𝑥 > 1 by Frobenius method.
Í∞
Here, 𝑥 = 0 is a regular singular point. Write 𝑦 (𝑥) = 𝑛=0 𝑎𝑛 𝑥 𝑛+𝑟 and substitute in
the ODE to get
∞
Õ ∞
Õ ∞
Õ
2 𝑛+𝑟 −2 𝑛+𝑟 −1
(𝑥 − 𝑥) (𝑛 + 𝑟 )(𝑛 + 𝑟 − 1)𝑎𝑛 𝑥 −𝑥 (𝑛 + 𝑟 )𝑎𝑛 𝑥 + 𝑎𝑛 𝑥 𝑛+𝑟 = 0.
𝑛=0 𝑛=0 𝑛=0
Series Solutions 61
Simplifying we get
∞
Õ ∞
Õ
2
(𝑛 + 𝑟 − 1) 𝑎𝑛 𝑥 𝑛+𝑟
− (𝑛 + 𝑟 )(𝑛 + 𝑟 − 1)𝑎𝑛 𝑥 𝑛+𝑟 −1 = 0.
𝑛=0 𝑛=0

The lowest power of 𝑥 is 𝑥 𝑟 −1 . Equating its coefficient to 0, we get the indical

equation. It gives
𝑟 (𝑟 − 1) = 0 ⇒ 𝑟 1 = 1, 𝑟 2 = 0.
Since the roots differ by an integer, using (3.7)(3), we compute the first solution
𝑦1 (𝑥) as follows.
Taking the 𝑟 = 𝑟 1 = 1 and setting the coefficient of 𝑥 𝑛+1 to 0, we get
∞
Õ  2 
𝑛 𝑎𝑛 − (𝑛 + 2)(𝑛 + 1)𝑎𝑛+1 𝑥 𝑛+1 = 0.
𝑛=0

It implies the recurrence relation

𝑛2
𝑎𝑛+1 = 𝑎𝑛 for 𝑛 ≥ 0.
(𝑛 + 1)(𝑛 + 2)

For 𝑛 = 0, we have 𝑎 1 = 0. Consequently, 𝑎𝑛 = 0 for all 𝑛 ≥ 1. Choosing 𝑎 0 = 1 we

get the first solution as 𝑦1 (𝑥) = 𝑎 0𝑥 𝑟 1 = 𝑥.
For the second solution, we use the method of reduction of order. Here, 𝑝 (𝑥) =
1/(1 − 𝑥). Then
∫ ∫
𝑑𝑥
− 𝑝 (𝑥) 𝑑𝑥 = = log |𝑥 − 1| = log(𝑥 − 1) as 𝑥 > 1
𝑥 −1
 ∫ 
exp − 𝑝 (𝑥) 𝑑𝑥 𝑥 −1
⇒ 𝑣 (𝑥) = 2
= 2
𝑦1 (𝑥) 𝑥
∫ ∫
𝑥 −1  1
⇒ 𝑦2 (𝑥) = 𝑦1 (𝑥) 𝑣 (𝑥) 𝑑𝑥 = 𝑥 𝑑𝑥 = 𝑥 log 𝑥 + = 𝑥 log 𝑥 + 1.
𝑥2 𝑥

Hence, the general solution of the ODE is 𝑦 (𝑥) = 𝑐 1𝑥 + 𝑐 2 (log 𝑥 + 1).

(3.12) Example
Find a series solution of Euler-Cauchy equation 𝑥 2𝑦 00 − 𝑥𝑦 0 + 10𝑦 = 0 for 𝑥 > 0.
Here, 𝑝 (𝑥) = −1/𝑥 and 𝑞(𝑥) = 10/𝑥 2 . Thus, 𝑥 = 0 is a regular singular point. We
have 𝑝 0 = lim 𝑥𝑝 (𝑥) = −1 and 𝑞 0 = lim 𝑥 2𝑞(𝑥) = 10. So, the indical equation is
𝑥→0 𝑥→0

𝑟 (𝑟 − 1) + (−1)𝑟 + 10 = 𝑟 2 − 2𝑟 + 10 = 0.
62 MA2020 Classnotes

It has complex roots 1 ± 3𝑖. We require now complex solutions of the ODE. We
proceed as earlier for the recurrence relations. Substituting
∞
Õ
𝑧 (𝑥) = 𝑦 (𝑥) = 𝑎𝑛 𝑥 𝑛+𝑟
𝑛=0

into the ODE and setting powers of 𝑥 to 0, we obtain

∞
Õ ∞
Õ ∞
Õ
𝑛+𝑟 𝑛+𝑟
0= (𝑛 + 𝑟 )(𝑛 + 𝑟 − 1)𝑎𝑛 𝑥 − (𝑛 + 𝑟 )𝑎𝑛 𝑥 + 10𝑎𝑛 𝑥 𝑛+𝑟
𝑛=0 𝑛=0 𝑛=0
∞
Õ  
= (𝑛 + 𝑟 )(𝑛 + 𝑟 − 2) + 10 𝑎𝑛 𝑥 𝑛+𝑟
𝑛=0
2
= (𝑟 − 2𝑟 + 10)𝑎 0 = 0, (𝑛 + 𝑟 )(𝑛 + 𝑟 + 10)𝑎𝑛 = 0 for 𝑛 ≥ 1.

The first one gives the indical equation. In the second one, with 𝑟 = 1 ± 3𝑖, the
factor (𝑛 + 𝑟 )(𝑛 + 𝑟 + 10) ≠ 0. Hence 𝑎𝑛 = 0 for each 𝑛 ≥ 1. Thus the complex
solutions are given by

𝑧 (𝑥) = 𝑎 0𝑥 1+3𝑖 = 𝑎 0𝑥 exp log(𝑥 3𝑖 ) = 𝑎 0𝑥 cos(3 log 𝑥) + 𝑖 sin(3 log 𝑥) .

Setting the constant 𝑎 0 = 1, and using (3.7)(4), the two linearly independent solutions
are
𝑦1 (𝑥) = 𝑥 cos(3 log 𝑥), 𝑦2 (𝑥) = 𝑥 sin(3 log 𝑥).
Thus, the series solution of the ODE is given by

𝑦 (𝑥) = 𝑐 1𝑦1 + 𝑐 2𝑦2 = 𝑐 1𝑥 cos(3 log 𝑥) + 𝑐 2𝑥 sin(3 log 𝑥)

where 𝑐 1 and 𝑐 2 are arbitrary constants.

4
Special Functions

4.1 Legendre polynomials

In this chapter we discuss some special types of ODEs whose series solutions give
rise to the special functions. First, we consider the Legendre equation in its general
form. It is
(1 − 𝑥 2 )𝑦 00 − 2𝑥𝑦 0 + 𝑝 (𝑝 + 1)𝑦 = 0 for |𝑥 | < 1. (4.1.1)
where 𝑝 is a constant, often called a parameter. So, this equation is actually a family
of ODEs. We should not be surprised if the nature of solutions differs for various
values of 𝑝.
The ODE in (4.1.1) has the standard form
2𝑥 𝑝 (𝑝 + 1)
𝑦 00 − 2
𝑦0 + 𝑦 = 0.
1−𝑥 1 − 𝑥2
The coefficient functions −2𝑥/(1 − 𝑥 2 ) and 𝑝 (𝑝 + 1)/(1 − 𝑥 2 ) are analytic at 𝑥 = 0.
That is, 𝑥 = 0 is an ordinary point of the ODE. Thus, the ODE has a power series
solution in the form
Õ∞
𝑦 (𝑥) = 𝑎𝑛 𝑥 𝑛 .
𝑛=0
Substituting it in the ODE and setting the coefficients of 𝑥 𝑛 to 0, we obtain
∞
Õ ∞
Õ
2 𝑛−2
(1 − 𝑥 ) 𝑛(𝑛 − 1)𝑎𝑛 𝑥 − 2𝑥 𝑛𝑎𝑛 𝑥 𝑛−1 + 𝑝 (𝑝 + 1)𝑎𝑛 𝑥 𝑛 = 0
𝑛=0 𝑛=0
⇒ (𝑛 + 2)(𝑛 + 1)𝑎𝑛+2 − 𝑛(𝑛 − 1)𝑎𝑛 − 2𝑛𝑎𝑛 + 𝑝 (𝑝 + 1)𝑎𝑛 = 0
⇒ (𝑛 + 2)(𝑛 + 1)𝑎𝑛+2 = (𝑛 2 − 𝑛 + 2𝑛 − 𝑝 2 − 𝑝)𝑎𝑛 = (𝑛 2 − 𝑝 2 + 𝑛 − 𝑝)𝑎𝑛
(𝑛 − 𝑝)(𝑝 + 𝑛 + 1)
⇒ 𝑎𝑛+2 = − 𝑎𝑛 .
(𝑛 + 1)(𝑛 + 2)
The recurrence relation is used to compute the coefficients of 𝑎 2, 𝑎 3, . . . in terms of
𝑎 0 and 𝑎 1 , which are left arbitrary. To compute a few,
𝑝 (𝑝 + 1) (𝑝 − 2)(𝑝 + 3) 𝑝 (𝑝 − 2)(𝑝 + 1)(𝑝 + 3)
𝑎2 = − 𝑎 0, 𝑎 4 = − = ,...
1·2 3·4 4!
(𝑝 − 1)(𝑝 + 2) (𝑝 − 3)(𝑝 + 4) (𝑝 − 1)(𝑝 − 3)(𝑝 + 2)(𝑝 + 4)
𝑎3 = − 𝑎 1, 𝑎 5 = − 𝑎3 = 𝑎 1, . . .
2·3 4·5 5!

63
64 MA2020 Classnotes

We thus get a formal solution 𝑦 (𝑥) = 𝑎 0𝑦1 (𝑥) + 𝑎 1𝑦2 (𝑥) , where
𝑝 (𝑝 + 1) 2 𝑝 (𝑝 − 2)(𝑝 + 1)(𝑝 + 3) 4
𝑦1 (𝑥) = 1 − 𝑥 + 𝑥 −···
2! 4!
(𝑝 − 1)(𝑝 + 2) 3 (𝑝 − 1)(𝑝 − 3)(𝑝 + 2)(𝑝 + 4) 5
𝑦2 (𝑥) = 𝑥 − 𝑥 + 𝑥 −···
3! 5!
When 𝑝 is not an integer, the numerators in the coefficients of powers of 𝑥 do not
vanish. In the series for 𝑦1 (𝑥), taking the absolute value of ratio of a term and its
preceding term, we find that

 2𝑛+2𝑥 2𝑛+2   (𝑝 − 2𝑛)(𝑝 + 2𝑛 + 1) 

𝑎   
2𝑛
=  → |𝑥 | 2 as 𝑛 → ∞.
(2𝑛 + 1)(2𝑛 + 2)
  
𝑎 2𝑛 𝑥
Hence, the radius of convergence of the series for 𝑦1 (𝑥) is 1. Similarly, it is easy to
show that the radius of convergence for the series for 𝑦2 (𝑥) is also 1 in case 𝑝 is not
an integer. That is, the formal solution given above is a solution for −1 < 𝑥 < 1.
Notice that this is the best we can expect since the coefficient functions −2𝑥/(1−𝑥 2 )
and 𝑝 (𝑝 + 1)/(1 − 𝑥 2 ) are not analytic at 𝑥 = 1.
Next, we consider the interesting case when 𝑝 is a non-negative integer. We
consider the cases 𝑝 = 0, 𝑝 is nonzero even, and 𝑝 is nonzero odd separately.
Case 1: Suppose 𝑝 = 0. Then 𝑦1 (𝑥) = 1 and
(−1)(2) 3 (−1)(−3)(2)(4) 5
𝑦2 (𝑥) = 𝑥 − 𝑥 + 𝑥 −···
3! 5!
Here, 𝑦1 (𝑥) is a constant and 𝑦2 (𝑥) is a power series.
Case 2: Suppose 𝑝 is nonzero and even, say, 𝑝 = 2𝑘 for some 𝑘 ≥ 1. Then
2𝑘 (2𝑘 + 1) 2 2𝑘 (2𝑘 − 2) · · · (2)(2𝑘 + 1)(2𝑘 + 3) · · · (2𝑘 + 2𝑘 − 1) 2𝑘
𝑦1 (𝑥) = 1− 𝑥 +· · ·+(−1)𝑘 𝑥 .
2! (2𝑘)!
The next term in the series for 𝑦1 (𝑥) has in the numerator the factor (𝑝 − 2𝑘) = 0.
All succeeding terms are then 0. Therefore, 𝑦1 (𝑥) terminates there, and it is a
polynomial. In this case, 𝑦2 (𝑥) is a power series.
Case 3: Suppose 𝑝 is odd, say, 𝑝 = 2𝑘 + 1 for some 𝑘 ≥ 0. Then
(2𝑘)(2𝑘 + 3) 3 (2𝑘)(2𝑘 − 2) · · · (2)(2𝑘 + 3)(2𝑘 + 5) · · · (2𝑘 + 2𝑘 + 1) 2𝑘+1
𝑦2 (𝑥) = 𝑥− 𝑥 +· · ·+(−1)𝑘 𝑥 .
3! (2𝑘 + 1)!
The next term in the series for 𝑦2 (𝑥) is 0 since the numerator has a factor
(𝑝 − (2𝑘 + 1)) = 0. All succeeding terms are then 0. Therefore, 𝑦2 (𝑥) termi-
nates there, and it is a polynomial. In this case, 𝑦1 (𝑥) is a power series.
We thus find that if 𝑝 is an integer, then exactly one of 𝑦1 (𝑥) or 𝑦2 (𝑥) is a
polynomial.
Special Functions 65
When 𝑝 = 0, the ODE is (1 − 𝑥 2 )𝑦 00 − 2𝑥𝑦 0 = 0. Since 𝑝 = 0, the polynomial
solution of this ODE is 𝑦1 (𝑥) = 1. This polynomial 𝑦1 (𝑥) is of degree 0 with
𝑦1 (1) = 1.
When 𝑝 = 2, the ODE is (1 − 𝑥 2 )𝑦 00 − 2𝑥𝑦 0 + 6𝑦 = 0. Since 𝑝 is even, the
polynomial solution of this ODE is (with 𝑝 = 2𝑘, 𝑘 = 1)
2(3) 2
𝑦1 (𝑥) = 1 − 𝑥 = 1 − 3𝑥 2 .
2!
This polynomial 𝑦1 (𝑥) is of degree 2 with 𝑦1 (1) = 1 − 3 = −2.
It continues this way for even 𝑝. Let us look at a few cases when 𝑝 is odd.
When 𝑝 = 1, the ODE is (1−𝑥 2 )𝑦 00 −2𝑥𝑦 0 +2𝑦 = 0. Since 𝑝 is odd, the polynomial
solution is (with 𝑝 = 2𝑘 + 1, 𝑘 = 0)

𝑦2 (𝑥) = 𝑥 .

This polynomial 𝑦2 (𝑥) is of degree 1 with 𝑦2 (1) = 1.

When 𝑝 = 3, the ODE is (1 − 𝑥 2 )𝑦 00 − 2𝑥𝑦 0 + 12𝑦 = 0. The polynomial solution
is (with 𝑝 = 2𝑘 + 1, 𝑘 = 1)
(2)(2 + 3) 3 5
𝑦2 (𝑥) = 𝑥 − 𝑥 = 𝑥 − 𝑥 3.
3! 3
This polynomial 𝑦2 (𝑥) is of degree 3 with 𝑦2 (1) = 1 − 5/3 = −2/3.
As we see from the above cases, the polynomials when evaluated at 𝑥 = 1 give
the values as follows:
Parameter 𝑝: 0 1 2 3
Degree of polynomial: 0 1 2 3
Which solution: 𝑦1 𝑦2 𝑦1 𝑦2
Its value at 1: 1 1 −2 −2/3
Notice that since 𝑦1 (𝑥) is a solution of an appropriate Legendre equation, any
constant multiple of 𝑦1 (𝑥) is also a solution of the same Legendre equation. The
same is also true for 𝑦2 (𝑥). In particular, the polynomials and there constant
multiples are also solutions of suitable Legendre equations. Thus, we can choose
to multiply an appropriate constant in each case so that the resulting polynomial
when evaluated at 1 will give the value 1. Such polynomials are called Legendre
polynomials.
Thus, the Legendre polynomial of degree 𝑛, denoted by 𝑃𝑛 (𝑥), is the polynomial
of degree 𝑛 that satisfies the Legendre equation

(1 − 𝑥 2 )𝑦 00 − 2𝑥𝑦 0 + 𝑛(𝑛 + 1)𝑦 = 0 with 𝑦 (1) = 1.

We find that if 𝑛 is even, then 𝑃𝑛 (𝑥) = 𝑦1 (𝑥) and it does not have any odd power
of 𝑥; and if 𝑛 is odd, then 𝑃𝑛 (𝑥) = 𝑦2 (𝑥) and it does not have any even power of
66 MA2020 Classnotes

𝑥. Further, these polynomials satisfy 𝑃𝑛 (1) = 1. Using the above computations, we

obtain the following:

𝑃0 (𝑥) = 𝑦1 (𝑥) ( with 𝑝 = 0) = 1.

𝑃1 (𝑥) = 𝑦2 (𝑥) ( with 𝑝 = 1) = 𝑥 .
1
𝑃2 (𝑥) = 𝑦1 (𝑥) ( with 𝑝 = 2) = −2 (1 − 3𝑥 2 ) = 12 (3𝑥 2 − 1).
5

𝑃3 (𝑥) = 𝑦2 (𝑥) ( with 𝑝 = 3) = − 32 𝑥 − 𝑥 3 = 12 (5𝑥 3 − 3𝑥).
3
There is another way to choose these constants so that the condition 𝑃𝑛 (1) = 1
is satisfied. This way we may be able to express Legendre’s polynomials in close
form. The idea is to assume certain nice form of the coefficient of highest power of
𝑥 in 𝑃𝑛 (𝑥). So, suppose 𝑎𝑛 is the coefficient of 𝑥 𝑛 in 𝑃𝑛 (𝑥). We choose the constants
in such a way that

(2𝑛)! 1 · 3 · 5 · · · (2𝑛 − 1)
𝑎𝑛 = = for 𝑛 ≥ 0.
2𝑛 (𝑛!) 2 𝑛!
Using our recurrence relation for the coefficients derived earlier, we have

𝑛(𝑛 − 1) 𝑛(𝑛 − 1)(2𝑛)! (2𝑛 − 2)!

𝑎𝑛−2 = − 𝑎𝑛 = − 2
=− 𝑛 .
2(2𝑛 − 1) 2(2𝑛 − 1)(𝑛!) 2 (𝑛 − 1)!(𝑛 − 2)!
(𝑛 − 2)(𝑛 − 3) (2𝑛 − 4)!
𝑎𝑛−4 = − 𝑎𝑛−2 = 𝑛 .
4(2𝑛 − 3) 2 2!(𝑛 − 2)!(𝑛 − 4)!
(2𝑛 − 2𝑘)!
𝑎𝑛−2𝑘 = (−1)𝑘 𝑛 for 𝑛 ≥ 2𝑘.
2 𝑘!(𝑛 − 𝑘)!(𝑛 − 2𝑘)!
Using this, Legendre polynomial of degree 𝑛 may be written as
𝑚
Õ (2𝑛 − 2𝑘)!
𝑃𝑛 (𝑥) = (−1)𝑘 𝑥 𝑛−2𝑘 where 𝑚 = [𝑛/2]
2𝑛𝑘!(𝑛 − 𝑘)!(𝑛 − 2𝑘)!
𝑘=0
(2𝑛)! 𝑛 (2𝑛 − 2)!
= 2
𝑥 − 𝑛 𝑥 𝑛−2 + · · · (4.1.2)
2 (𝑛!)
𝑛 2 1!(𝑛 − 1)!(𝑛 − 2)!

To see that it is the same 𝑃𝑛 (𝑥) we have defined earlier we need only to check that
𝑃𝑛 (1) = 1 for each 𝑛. We will show it later in (4.1.4).
Though 𝑃𝑛 (𝑥) is a polynomial, it is treated as a special function because it has
some nice properties and it comes in various disguises. One of its useful form is
the following:

1 𝑑𝑛 2
Rodrigue’s formula : 𝑃𝑛 (𝑥) = (𝑥 − 1)𝑛 . (4.1.3)
2𝑛𝑛! 𝑑𝑥 𝑛
Special Functions 67
To see that the formula is correct, notice that
𝑑 𝑛 2𝑛−2𝑘 (2𝑛 − 2𝑘)! 𝑛−2𝑘
𝑥 = 𝑥 for 0 ≤ 𝑘 ≤ 𝑚 = [𝑛/2].
𝑑𝑥 𝑛 (𝑛 − 2𝑘)!
Thus, 𝑃𝑛 (𝑥) is rewritten as
𝑚
Õ (2𝑛 − 2𝑘)!
𝑃𝑛 (𝑥) = (−1)𝑘 𝑥 𝑛−2𝑘
2𝑛𝑘!(𝑛 − 𝑘)!(𝑛 − 2𝑘)!
𝑘=0
𝑚
1 𝑑𝑛 Õ 𝑛!
= 𝑛 (−1)𝑘 𝑥 2𝑛−2𝑘 .
2 𝑛! 𝑑𝑥 𝑛 𝑘!(𝑛 − 𝑘)!
𝑘=0

When 𝑘 > 𝑚 = [𝑛/2], any term in the sum above is a polynomial of degree less
than 𝑛 so that its 𝑛th derivative is 0. Hence, the sum above can be extended from
𝑚 + 1 to 𝑛 without changing the value on the left hand side. So,
𝑛
1 𝑑𝑛 Õ 𝑘 𝑛! 2𝑛−2𝑘 1 𝑑𝑛 2
𝑃𝑛 (𝑥) = 𝑛 (−1) 𝑥 = (𝑥 − 1)𝑛 .
2 𝑛! 𝑑𝑥 𝑛 𝑘!(𝑛 − 𝑘)! 2𝑛𝑛! 𝑑𝑥 𝑛
𝑘=0

The last equality follows from the Binomial expansion of (𝑥 2 − 1)𝑛 .

Various useful properties of Legendre polynomials follow from Rodrigue’s for-
mula with the help of Leibniz rule for computing the 𝑛th derivative of a product of
two functions. Leibniz’s rule says that
𝑛
𝑑 𝑛 (𝑓 𝑔) Õ 𝑛! 𝑑 𝑘 𝑓 𝑑 𝑛−𝑘 𝑔
= ,
𝑑𝑥 𝑛 𝑘!(𝑛 − 𝑘)! 𝑑𝑥 𝑘 𝑑𝑥 𝑛−𝑘
𝑘=0

where the 0th derivative of a function is taken as the function itself.

In Rodrigue’ formula writing (𝑥 2 − 1)𝑛 = (𝑥 + 1)𝑛 (𝑥 − 1)𝑛 and applying Leibniz
rule we obtain
𝑛
1 Õ 𝑛! 𝑑 𝑘 [(𝑥 + 1)𝑛 ] 𝑑 𝑛−𝑘 [(𝑥 − 1)𝑛 ]
𝑃𝑛 (𝑥) = 𝑛 .
2 𝑛! 𝑘!(𝑛 − 𝑘)! 𝑑𝑥 𝑘 𝑑𝑥 𝑛−𝑘
𝑘=0

The first term in the above sum is

𝑑 0 [(𝑥 + 1)𝑛 ] 𝑑 𝑛 [(𝑥 − 1)𝑛 ]
0
· 𝑛
= (𝑥 + 1)𝑛𝑛!.
𝑑𝑥 𝑑𝑥
Each of the remaining terms contains the factor (𝑥 − 1). Thus, when evaluated at
𝑥 = 1, each term except the first in the sum becomes 0. Thus,
1
𝑃𝑛 (1) = (1 + 1)𝑛𝑛! = 1. (4.1.4)
2𝑛𝑛!
68 MA2020 Classnotes

It is often helpful to get the generating function for the Legendre polynomials.
We will show that the generating function is (1 − 2𝑥𝑡 + 𝑡 2 ) −1 . That is,
∞
Õ

−1/2
1 − 2𝑥𝑡 + 𝑡 2 = 𝑃𝑛 (𝑥)𝑡 𝑛 . (4.1.5)
𝑛=0

To see this, we apply the Binomial theorem on the left hand side expression.
Recall that the Binomial theorem asserts that
∞
𝑟
Õ 𝑟 (𝑟 − 1) · · · (𝑟 − 𝑛 + 1) 𝑛
(1 + 𝑧) = 𝑧 for |𝑧| < 1.
𝑛=0
𝑛!

Taking 𝑧 = 𝑡 2 − 2𝑥𝑡 = 𝑡 (𝑡 − 2𝑥) and assuming that |𝑡 2 − 2𝑥𝑡 | < 1, we obtain

∞
Õ (− 12 )(− 23 ) · · · (− 12 − 𝑛 + 1) 𝑛
(1 − 2𝑥𝑡 + 𝑡 2 ) −1/2 = 𝑡 (𝑡 − 2𝑥)𝑛
𝑛=0
𝑛!
∞
Õ (−1)𝑛 (2𝑛)! 𝑛
= 2𝑛 [𝑛!] 2
𝑡 (𝑡 − 2𝑥)𝑛
𝑛=0
2
∞ 𝑛
Õ (−1)𝑛 (2𝑛)! 𝑛  Õ 𝑛! 𝑘 𝑛−𝑘

= 𝑡 𝑡 (−2𝑥)
𝑛=0
22𝑛 (𝑛!) 2 𝑘!(𝑛 − 𝑘)!
𝑘=0
∞ Õ
𝑛
Õ (−1)𝑘 (2𝑛)! 𝑛+𝑘
= 2𝑛
𝑡 (2𝑥)𝑛−𝑘 .
𝑛=0 𝑘=0
2 𝑛!𝑘!(𝑛 − 𝑘)!

In general, if 𝐶𝑘,𝑛 is any expression depending on 𝑘 and 𝑛, we have

∞ Õ
Õ 𝑛 ∞ [𝑛/2]
Õ Õ
𝑛+𝑘
𝐶𝑘,𝑛 𝑡 = 𝐶𝑘,𝑛−𝑘 𝑡 𝑛 .
𝑛=0 𝑘=0 𝑛=0 𝑘=0

Using this on the above sum, we obtain

∞ [𝑛/2] ∞
2 −1/2
Õ Õ (−1)𝑘 (2𝑛 − 2𝑘)! 𝑛 𝑛−2𝑘
Õ
(1 − 2𝑥𝑡 + 𝑡 ) = 𝑡 𝑥 = 𝑃𝑛 (𝑥)𝑡 𝑛 .
𝑛=0 𝑘=0
2 𝑘!(𝑛 − 𝑘)!(𝑛 − 2𝑘)!
𝑛
𝑛=0

An important property of the Legendre polynomials is that they are orthogonal

to each other. It means
∫1
𝑃𝑚 (𝑥)𝑃𝑛 (𝑥) 𝑑𝑥 = 0 for 𝑚 ≠ 𝑛. (4.1.6)
−1

To see this we use the fact that Legendre polynomials are solutions of the Legendre
ODE, which can be rewritten as
[(1 − 𝑥 2 )𝑦 0] 0 + 𝑝 (𝑝 + 1)𝑦 = 0.
Special Functions 69
Therefore,

[(1 − 𝑥 2 )𝑃𝑚0 (𝑥)] 0 + 𝑚(𝑚 + 1) = 0, [(1 − 𝑥 2 )𝑃𝑛0 (𝑥)] 0 + 𝑛(𝑛 + 1) = 0.

Multiply the first with 𝑃𝑛 and the second with 𝑃𝑚 , subtract, and integrate to get

∫1 ∫1
2
)𝑃𝑚0 ] 0 −𝑃𝑚 [(1 −𝑥 2 )𝑃𝑛0 ] 0 𝑑𝑥 −

 
𝑃𝑛 [(1 −𝑥 𝑚(𝑚 + 1) −𝑛(𝑛 + 1) 𝑃𝑚 𝑃𝑛 𝑑𝑥 = 0.
−1 −1

Evaluate the first integral by using integration by parts. It gives

∫1
𝑃𝑛 [(1 − 𝑥 2 )𝑃𝑚0 ] 0 − 𝑃𝑚 [(1 − 𝑥 2 )𝑃𝑛0 ] 0 𝑑𝑥


−1
h i1 h i1
2 0 2 0
= 𝑃𝑛 (1 − 𝑥 )𝑃𝑚 − 𝑃𝑚 (1 − 𝑥 )𝑃𝑛
−1 −1
∫1
 0
𝑃𝑛 (1 − 𝑥 2 )𝑃𝑚0 − 𝑃𝑚0 (1 − 𝑥 2 )𝑃𝑛0 𝑑𝑥 = 0.

−
−1

∫1
Hence, If 𝑚 ≠ 𝑛, then −1 𝑃𝑚 (𝑥)𝑃𝑛 (𝑥) 𝑑𝑥 = 0.
What happens when 𝑚 = 𝑛? We use Rodrigue’s formula and integration by parts
as follows:
∫1 ∫1
 2 𝑑𝑛 2
𝑃𝑛 (𝑥) 𝑑𝑥 = 𝑃𝑛 (𝑥) 𝑛
(𝑥 − 1)𝑛 𝑑𝑥
−1 −1 𝑑𝑥
i1 ∫1
1 h 𝑑 𝑛−1 2 1 𝑑 𝑛−1
= 𝑛 𝑃𝑛 (𝑥) 𝑛−1 (𝑥 − 1) 𝑛
− 𝑛 𝑃𝑛0 (𝑥) 𝑛−1 (𝑥 2 − 1)𝑛 𝑑𝑥
2 𝑛! 𝑑𝑥 −1 2 𝑛! −1 𝑑𝑥
∫1
1 𝑑 𝑛−1
=0− 𝑛 𝑃𝑛0 (𝑥) 𝑛−1 (𝑥 2 − 1)𝑛 𝑑𝑥
2 𝑛! −1 𝑑𝑥
..
.
(−1)𝑛 1
∫1
𝑑0
∫
(𝑛)
= 𝑛 [𝑃𝑛 (𝑥)] 𝑃𝑛0 (𝑥) 0 (𝑥 2 − 1)𝑛 𝑑𝑥
2 𝑛! −1 −1 𝑑𝑥
∫1
(−1) 𝑛 (2𝑛)!
= 𝑛 (1 − 𝑥 2 )𝑛 𝑑𝑥
2 𝑛! −1 2𝑛𝑛!
2(2𝑛)! 1
∫
= 2𝑛 (1 − 𝑥 2 )𝑛 𝑑𝑥 (put 𝑥 = sin 𝜃 )
2 (𝑛!) 2 0
∫
2(2𝑛)! 𝜋/2
= 2𝑛 2
cos2𝑛+1 𝜃 𝑑𝜃
2 (𝑛!) 0
70 MA2020 Classnotes
∫ 𝜋/2
2(2𝑛)! 2𝑛
= cos2𝑛−1 𝜃 𝑑𝜃
22𝑛 (𝑛!) 2 2𝑛 + 1 0
..
.
∫
2(2𝑛)! 2𝑛 2𝑛 − 2 2 𝜋/2
= 2𝑛 ··· cos 𝜃 𝑑𝜃
2 (𝑛!) 2 2𝑛 + 1 2𝑛 − 1 3 0
2(2𝑛)! 2𝑛 2𝑛 − 2 2 2
= 2𝑛 ··· = .
2 (𝑛!) 2 2𝑛 + 1 2𝑛 − 1 3 2𝑛 + 1
Hence, ∫1
 22
𝑃𝑛 (𝑥) 𝑑𝑥 =
. (4.1.7)
−1 2𝑛 + 1
Many problems in engineering depend on the possibility of expanding a given
function in a series of Legendre polynomials. It is easy to see that a polynomial
can always be expanded this way. For example, consider a polynomial of degree at
most 3, say
𝑝 (𝑥) = 𝑏 0 + 𝑏 1𝑥 + 𝑏 2𝑥 2 + 𝑏 3𝑥 3 .
With 𝑃0 (𝑥) = 1, 𝑃1 (𝑥) = 𝑥, 𝑃 2 (𝑥) = 21 (3𝑥 2 − 1), 𝑃3 (𝑥) = 21 (5𝑥 3 − 3𝑥), we see that

1 2 3 2
1 = 𝑃0 (𝑥), 𝑥 = 𝑃1 (𝑥), 𝑥 2 = 𝑃0 (𝑥) + 𝑃 2 (𝑥), 𝑥 3 = 𝑃1 (𝑥) + 𝑃3 (𝑥).
3 3 5 5
Hence,
 𝑏2   3𝑏 3  2𝑏 2 2𝑏 3
𝑝 (𝑥) = 𝑏 0 + 𝑃0 (𝑥) + 𝑏 1 + 𝑃1 (𝑥) + 𝑃2 (𝑥) + 𝑃 3 (𝑥).
3 5 3 5
Í
Similarly, 𝑥 𝑛 can be expanded as 𝑛𝑘=0 𝑎𝑘 𝑃𝑘 (𝑥) for some constants 𝑎𝑘 . It looks that
if a function has a power series expansion, then it can also be expanded in terms
of Legendre polynomials 𝑃𝑛 (𝑥). However, some conditions my be required so that
the obtained series is convergent. We rather concentrate on how to compute the
coefficients in such a series expansion if it exists.
When a function 𝑓 (𝑥) for −1 < 𝑥 < 1, can be written in the form
∞
Õ
𝑓 (𝑥) = 𝑎𝑛 𝑃𝑛 (𝑥)
𝑛=0

we say that 𝑓 (𝑥) has a Legendre series expansion. Our question is, if 𝑓 (𝑥) has a
Legendre series expansion, then how do we compute the coefficients 𝑎𝑛 ?
We multiply the above with 𝑃𝑚 (𝑥), integrate term by term (assuming that this is
permissible), and use (4.1.6-4.1.7) to obtain
∫1 ∞ ∫1
Õ 2𝑎𝑚
𝑓 (𝑥)𝑃𝑚 (𝑥) 𝑑𝑥 = 𝑎𝑛 𝑃𝑚 (𝑥)𝑃𝑛 (𝑥) 𝑑𝑥 = .
−1 𝑛=0 −1 2𝑚 + 1
Special Functions 71
Therefore,
1 1
 ∫
𝑎𝑛 = 𝑛 + 𝑓 (𝑥)𝑃𝑛 (𝑥) 𝑑𝑥 .
2 −1
Many other properties of Legendre polynomials are included in the exercises. As
a convention, when 𝑃𝑛 (𝑥) is treated as a function, we assume that −1 ≤ 𝑥 ≤ 1.

4.2 Bessel Functions

The linear homogeneous second order ordinary differential equation

𝑥 2𝑦 00 + 𝑥𝑦 0 + (𝑥 2 − 𝜈 2 )𝑦 = 0 (4.2.1)

is called the Bessel’s equation with non-negative parameter 𝜈. (It is nu not vee.) It
arises many where in applications. In standard form, the equation is

𝑦0  𝜈2 
𝑦 00 + + 1 − 2 𝑦 = 0.
𝑥 𝑥
The point 𝑥 = 0 is a regular singular point of the ODE. Hence the ODE has a
solution in the form
∞
Õ
𝑦 (𝑥) = 𝑎𝑘 𝑥 𝑘+𝑟 with 𝑎 0 ≠ 0.
𝑘=0
Substituting it in (4.2.1), we obtain
∞
Õ ∞
Õ
(𝑘 + 𝑟 )(𝑘 + 𝑟 − 1)𝑎𝑘 𝑥 𝑘+𝑟 + (𝑘 + 𝑟 )𝑎𝑘 𝑥 𝑘+𝑟
𝑘=0 𝑘=0
∞
Õ ∞
Õ
+ 𝑎𝑘 𝑥 𝑘+𝑟 +2 − 𝜈 2 𝑎𝑘 𝑥 𝑘+𝑟 = 0.
𝑘=0 𝑘=0

Thus coefficients of 𝑥 𝑟 , 𝑥 𝑟 +1 and 𝑥 𝑘+𝑟 for 𝑘 ≥ 2, are 0. It follows that

1. 𝑟 (𝑟 − 1)𝑎 0 + 𝑟𝑎 0 − 𝜈 2𝑎 0 = 0.
2. (𝑟 + 1)𝑟𝑎 1 + (𝑟 + 1)𝑎 1 − 𝜈 2𝑎 1 = 0.
3. (𝑘 + 𝑟 )(𝑘 + 𝑟 − 1)𝑎𝑘 + (𝑘 + 𝑟 )𝑎𝑘 + 𝑎𝑘−2 − 𝜈 2𝑎𝑘 = 0 for 𝑘 ≥ 2.
The first one gives the indical equation as (𝑟 + 𝜈)(𝑟 − 𝜈) = 0. The roots are 𝑟 1 = 𝜈
and 𝑟 2 = −𝜈. Corresponding to 𝑟 = 𝜈, the first solution of the ODE is
∞
Õ
𝑦1 (𝑥) = 𝑎𝑘 𝑥 𝑘+𝜈 .
𝑘=0
72 MA2020 Classnotes

We must find the coefficients 𝑎𝑘 . For 𝑟 = 𝜈, the second equation above implies

(𝜈 2 + 𝜈 + 𝜈 + 1 − 𝜈 2 )𝑎 1 = (2𝜈 + 1)𝑎 1 = 0 ⇒ 𝑎 1 = 0.

Substituting 𝑟 = 𝜈 in the third equation, we obtain

(𝑘 + 𝜈)(𝑘 + 𝜈 − 1)𝑎𝑘 + (𝑘 + 𝜈)𝑎𝑘 + 𝑎𝑘−2 − 𝜈 2𝑎𝑘 = 0

⇒ (𝑘 + 𝜈)(𝑘 + 𝜈 − 1 + 1) − 𝜈 2 𝑎𝑘 + 𝑎𝑘−2 = 0
 

⇒ 𝑘 (𝑘 + 2𝜈)𝑎𝑘 + 𝑎𝑘−2 = 0
𝑎𝑘−2
⇒ 𝑎𝑘 = −
𝑘 (𝑘 + 2𝜈)

Since 𝑎 1 = 0 it follows that all odd coefficients are 0. For even coefficients, say,
𝑘 = 2𝑚, the above recurrence looks like
𝑎 2𝑚−2
𝑎 2𝑚 = − 2
for 𝑚 = 1, 2, 3, . . .
2 𝑚(𝜈 + 𝑚)

It implies that
𝑎0 𝑎2 𝑎0
𝑎2 = − , 𝑎4 = − = ,...
22 (𝜈 + 1) 22 2(𝜈 + 2) 24 2!(𝜈 + 1)(𝜈 + 2)

Proceeding inductively, we get

(−1)𝑚 𝑎 0
𝑎 2𝑚 = for 𝑚 = 1, 2, 3, . . .
22𝑚𝑚!(𝜈 + 1)(𝜈 + 2) · · · (𝜈 + 𝑚)

By choosing the constant 𝑎 0 , all even coefficients are evaluated. It is customary to

choose
1
𝑎0 = 𝜈 .
2 Γ(𝜈 + 1)
Here, ∫∞
Γ(𝑥) = 𝑒 −𝑡 𝑡 𝑥−1 𝑑𝑡 for 𝑥 ≥ 0.
0

Notice that Γ(𝜈 + 1) is well defined since 𝜈 is non-negative. Some useful properties
of the gamma function are as follows:
√
Γ(𝑥 + 1) = 𝑥 Γ(𝑥), Γ(1/2) = 𝜋, Γ(𝑛 + 1) = 𝑛! for 𝑛 = 0, 1, 2, . . . .

It then follows that

(𝑥 + 1)(𝑥 + 2) · · · (𝑥 + 𝑚)Γ(𝑥 + 1) = Γ(𝑥 + 𝑚 + 1) for 𝑚 ∈ N ∪ {0}.

Special Functions 73
With the above choice of 𝑎 0 , we obtain
(−1)𝑚 𝑎 0
𝑎 2𝑚 =
22𝑚𝑚!(𝜈 + 1)(𝜈 + 2) · · · (𝜈 + 𝑚)
(−1)𝑚
= 2𝑚
2 𝑚!2𝜈 Γ(𝜈 + 1)(𝜈 + 1)(𝜈 + 2) · · · (𝜈 + 𝑚)
(−1)𝑚
= 𝜈+2𝑚 for 𝑚 = 1, 2, 3, . . . .
2 𝑚!Γ(𝜈 + 𝑚 + 1)
Í∞
With these coefficients, the solution 𝑦1 (𝑥) = 𝑘=0 𝑎𝑘 𝑥 𝑘+𝜈 is written as 𝐽𝜈 (𝑥), and is
called the Bessel function of first kind with order 𝜈. Thus,
∞
𝜈
Õ (−1)𝑚 𝑥 2𝑚
𝐽𝜈 (𝑥) = 𝑥 . (4.2.2)
𝑚=0
2𝜈+2𝑚𝑚!Γ(𝜈 + 𝑚 + 1)
The absolute value of the ratio of a term to its succeeding term in the seris for
𝐽𝜈 (𝑥) is given by
 2𝑚−2   22𝑚(𝜈 + 𝑚) 
𝑎   
=  → ∞ for any nonzero 𝑥 .
𝑥2

𝑎 2𝑚
The ratio test implies that the series in 𝐽𝜈 (𝑥) is convergent. Notice that the con-
vergence of the series is fast since factorials are in the denominator. The series
obviously converges for 𝑥 = 0. Hence, 𝐽𝜈 (𝑥) is well defined for all 𝑥.
In particular, when 𝜈 = 𝑛 ∈ N ∪ {0}, we have Γ(𝜈 + 1) = Γ(𝑛 + 1) = 𝑛!. Thus,
1
𝑎0 =
2𝑛𝑛!
(−1)𝑚 (−1)𝑚
𝑎 2𝑚 = = for 𝑚 = 1, 2, , 3, . . . .
2𝜈+2𝑚𝑚!Γ(𝜈 + 𝑚 + 1) 2𝑛+2𝑚𝑚!(𝑛 + 𝑚)!
The odd coefficients are 0 as earlier. Therefore, the first solution 𝑦1 (𝑥) of Bessel’s
equation
𝑥 2𝑦 00 + 𝑥𝑦 0 + (𝑥 2 − 𝑛 2 )𝑦 = 0, 𝑛 ∈ N ∪ {0}
is given by
∞
𝑛
Õ (−1)𝑚 𝑥 2𝑚
𝑦1 (𝑥) = 𝐽𝑛 (𝑥) = 𝑥 for 𝑛 ∈ N ∪ {0}. (4.2.3)
𝑚=0
2𝑛+2𝑚 𝑚!(𝑛 + 𝑚)!
Of course, this expression is directly obtained from (4.2.2) by taking 𝜈 = 𝑛. For
instance, the Bessel functions of first kind and order 0 and 1 are as follows.
∞
Õ (−1)𝑚 𝑥 2𝑚 𝑥2 𝑥4 𝑥6
𝐽0 (𝑥) = = 1 − + − +···
𝑚=0
22𝑚 (𝑚!) 2 22 (1!) 2 24 (2!) 2 26 (3!) 2
∞
Õ (−1)𝑚 𝑥 2𝑚+1 𝑥 𝑥3 𝑥5 𝑥7
𝐽1 (𝑥) = = − + − +···
𝑚=0
22𝑚+1𝑚!(𝑚 + 1)! 2 23 1!2! 25 2!3! 27 3!4!
74 MA2020 Classnotes

Notice that 𝐽𝑛 (0) = 0 for 𝑛 ≥ 1. It can be shown that

r
2  𝑛𝜋 𝜋 
𝐽𝑛 (𝑥) ≈ cos 𝑥 − − for large 𝑥 .
𝜋𝑥 2 4
For a general solution of Bessel’s equation (4.2.1), we consider two cases.
Case 1: Suppose the non-negative parameter 𝜈 is not an integer. Then the second
solution 𝑦2 (𝑥) of Bessel’s equation (4.2.1) is given by
∞
Õ (−1)𝑚 𝑥 2𝑚−𝜈
𝑦2 (𝑥) = 𝐽−𝜈 (𝑥) = 2𝑚−𝜈 𝑚!Γ(𝑚 − 𝜈 + 1)
. (4.2.4)
𝑚=0
2

This follows from a derivation similar to that of 𝐽𝜈 (𝑥). Also, by substituting 𝜈 with
−𝜈 in (4.2.2), we obtain this expression for 𝐽−𝜈 (𝑥).
Observe that any power of 𝑥 in 𝐽𝜈 (𝑥) is 𝑥 2𝑚+𝜈 and any power of 𝑥 in 𝐽−𝜈 (𝑥) is
𝑥 2𝑚−𝜈 . Since 𝜈 is not an integer, no power of 𝑥 in 𝐽𝜈 (𝑥) matches with any power
of 𝑥 in 𝐽−𝜈 (𝑥). Hence 𝐽𝜈 (𝑥) and 𝐽−𝜈 (𝑥) are linearly independent. Therefore, any
solution 𝑦 (𝑥) of Bessel’s equation with non-integral parameter 𝜈 is given by

𝑦 (𝑥) = 𝑐 1 𝐽𝜈 (𝑥) + 𝑐 2 𝐽−𝜈 (𝑥) for 𝜈 ∉ Z.

Case 2: Suppose 𝜈 = 𝑛 is an integer. We know the first solution as 𝐽𝑛 (𝑥) for 𝑛 ≥ 0.

For the second solution, let us look at 𝐽−𝑛 (𝑥). From (4.2.4) we have
∞
Õ (−1)𝑚 𝑥 2𝑚−𝑛
𝐽−𝑛 (𝑥) = . (4.2.5)
𝑚=0
22𝑚−𝑛𝑚!(𝑚 − 𝑛)!

We can also get 𝐽−𝑛 (𝑥) from (4.2.4) another way. In (4.2.4), let 𝜈 approach a positive
integer 𝑛. Then the Gamma function in the first 𝑛 terms approach ∞ so that the
coefficients in the first 𝑛 terms approach 0. The summation starts from 𝑚 = 𝑛 as
the Gamma function there is equal to Γ(𝑚 − 𝑛 + 1) = (𝑚 − 𝑛)! for 𝑚 ≥ 𝑛. Then,
shifting the index with 𝑘 = 𝑚 − 𝑛, we obtain
∞ ∞
Õ (−1)𝑚 𝑥 2𝑚−𝑛 Õ (−1)𝑛+𝑘 𝑥 2𝑘+𝑛
𝐽−𝑛 (𝑥) = = .
𝑚=𝑛
22𝑚−𝑛𝑚!(𝑚 − 𝑛)! 𝑘=0 22𝑘+𝑛𝑘!(𝑘 + 𝑛)!
Special Functions 75
Comparing the last expression with (4.2.5) we find that it is (−1)𝑛 𝐽𝑛 (𝑥). Therefore,
𝐽−𝑛 (𝑥) = (−1)𝑛 𝐽𝑛 (𝑥) for 𝑛 ∈ Z. (4.2.6)
It implies that 𝐽𝑛 (𝑥) and 𝐽−𝑛 (𝑥) are linearly dependent. Thus, we cannot take the
second solution 𝑦2 (𝑥) as 𝐽−𝑛 (𝑥). The second solution, denoted by 𝑌𝑛 (𝑥) can be
obtained by using reduction of order; it is fairly complicated. We only mention the
final result:
∞
2  𝑥  𝑥𝑛 Õ (−1)𝑚−1 (ℎ𝑚 + ℎ𝑚+𝑛 𝑥 2𝑚
𝑌𝑛 (𝑥) = 𝐽𝑛 (𝑥) log + 𝛾 +
𝜋 2 𝜋 𝑚=0 22𝑚+𝑛𝑚!(𝑚 + 𝑛)!

𝑥 −𝑛 Õ (𝑛 − 𝑚 − 1)!𝑥 2𝑚
𝑛−1
− for 𝑥 > 0 (4.2.7)
𝜋 𝑚=0 22𝑚−𝑛𝑚!
1
where 𝑛 = 0, 1, 2, . . ., ℎ 0 = 0, ℎ 1 = 1, ℎ𝑘 = 1 + 2 + · · · + 𝑘1 , and 𝛾 = lim (ℎ𝑘 − log 𝑘)
𝑘→∞
is Euler constant. In particular,
∞
2h
Õ (−1)𝑚−1ℎ𝑚 𝑥 2𝑚 i
𝑌0 (𝑥) = 𝐽0 (𝑥) log(𝑥/2) + 𝛾 + .
𝜋 𝑚=1
22𝑚 (𝑚!) 2
It can be seen that 𝑌0 (𝑥) behaves like log 𝑥 for small 𝑥 and 𝑌0 (𝑥) → −∞ when
𝑥 → 0.
In fact, both the cases above can be unified to obtain a function 𝑌𝜈 (𝑥) which is a
second solution of Bessel’s equation. It is as follows:
 
𝑌𝜈 (𝑥) = cosec(𝜈𝜋) 𝐽𝜈 (𝑥) cos(𝜈𝜋) − 𝐽−𝜈 (𝑥) .
With this definition, it can be seen that
lim 𝑌𝜈 (𝑥) = 𝑌𝑛 (𝑥).
𝜈→𝑛

But remember that when 𝜈 is not an integer, it does not say that 𝐽−𝜈 (𝑥) is equal to
𝑌−𝜈 (𝑥). In fact for 𝜈 ∉ Z, 𝑌−𝜈 (𝑥) = 𝑎𝐽𝜈 (𝑥) + 𝑏 𝐽−𝜈 (𝑥) for some nonzero 𝑎 and 𝑏.
Nonetheless, 𝐽𝜈 (𝑥) and 𝑌𝜈 (𝑥) are linearly independent and 𝑌𝜈 (𝑥) is also a solution of
Bessel’s equation (4.2.1). This function 𝑌𝜈 (𝑥) is called Bessel function of second
kind of order 𝜈. With the help of this function we thus say that the general solution
of Bessel’s equation (4.2.1) is given by
𝑦 (𝑥) = 𝑐 1 𝐽𝜈 (𝑥) + 𝑐 2𝑌𝜈 (𝑥)
for all values of 𝜈 and for 𝑥 > 0.
The complex solutions of Bessel’s equation may be given by
𝐻𝜈(1) (𝑥) = 𝐽𝜈 (𝑥) + 𝑖𝑌𝜈 (𝑥), 𝐻𝜈(2) (𝑥) = 𝐽𝜈 (𝑥) − 𝑖𝑌𝜈 (𝑥).
These two linearly independent complex solutions of Bessel’s equation are called
Bessel functions of third kind of order 𝜈.
76 MA2020 Classnotes

4.3 Properties of 𝐽𝜈 and 𝐽𝑛

In what follows we write 𝐽𝑛 to indicate that the parameter 𝜈 in 𝐽𝜈 is an integer 𝑛. In
this section we discuss some well known properties of 𝐽𝜈 (𝑥) and of 𝐽𝑛 (𝑥).
Multiply (4.2.2) by 𝑥 𝜈 to get
∞
𝜈
Õ (−1)𝑚 𝑥 2𝜈+2𝑚
𝑥 𝐽𝜈 (𝑥) = .
𝑚=0
2𝜈+2𝑚𝑚!Γ(𝜈 + 𝑚 + 1)

Differentiate with respect to 𝑥, cancel 2, pull out 𝑥 2𝜈−1 , and use the relation
(𝜈 + 𝑚)Γ(𝜈 + 𝑚) = Γ(𝜈 + 𝑚 + 1) to obtain
∞ ∞
𝜈
0 Õ (−1)𝑚 2(𝜈 + 𝑚)𝑥 2𝜈+2𝑚−1 𝜈 𝜈−1
Õ (−1)𝑚 𝑥 2𝑚
𝑥 𝐽𝜈 (𝑥) = = 𝑥 𝑥 .
𝑚=0
2𝜈+2𝑚𝑚!Γ(𝜈 + 𝑚 + 1) 𝑚=0
2𝜈+2𝑚−1𝑚!Γ(𝜈 + 𝑚)

Comparing the last expression with (4.2.2), we find that

0
𝑥 𝜈 𝐽𝜈 (𝑥) = 𝑥 𝜈 𝐽𝜈−1 (𝑥). (4.3.1)
Multiply (4.2.2) by 𝑥 −𝜈 , differentiate with respect to 𝑥, cancel 2𝑚, and shift the
index by taking 𝑚 = 𝑘 + 1, to obtain
∞ ∞
−𝜈
0 Õ (−1)𝑚 𝑥 2𝑚−1 Õ (−1)𝑘+1𝑥 2𝑘+1
𝑥 𝐽𝜈 (𝑥) = = .
𝑚=1
2𝜈+2𝑚−1 (𝑚 − 1)!Γ(𝜈 + 𝑚 + 1) 𝑘=0 2𝜈+2𝑘+1𝑘!Γ(𝜈 + 𝑘 + 2)

Now, in (4.2.2) take 𝜈 as 𝜈 + 1 and 𝑚 as 𝑘 so that you get the last expression as
−𝑥 −𝜈 𝐽𝜈+1 (𝑥). Therefore,

0
𝑥 −𝜈 𝐽𝜈 (𝑥) = −𝑥 −𝜈 𝐽𝜈+1 (𝑥). (4.3.2)
From (4.3.1)-(4.3.2), we get

0
𝐽𝜈−1 (𝑥) = 𝑥 −𝜈 𝑥 𝜈 𝐽𝜈 (𝑥) = 𝑥 −𝜈 𝑥 𝜈 𝐽𝜈0 (𝑥) + 𝜈𝑥 𝜈−1 𝐽𝜈 (𝑥) = 𝐽𝜈0 (𝑥) + 𝜈𝑥 −1 𝐽𝜈 (𝑥).
 

0
𝐽𝜈+1 (𝑥) = −𝑥 𝜈 𝑥 −𝜈 𝐽𝜈 (𝑥) = −𝑥 𝜈 𝑥 −𝜈 𝐽𝜈0 (𝑥) − 𝜈𝑥 −𝜈−1 𝐽𝜈 (𝑥) = −𝐽𝜈0 (𝑥) + 𝜈𝑥 −1 𝐽𝜈 (𝑥).
 

Subtracting the second from the first, we obtain

𝐽𝜈−1 (𝑥) − 𝐽𝜈+1 (𝑥) = 2𝐽𝜈0 (𝑥). (4.3.3)

And, adding those two equalities, we get
2𝜈
𝐽𝜈−1 (𝑥) + 𝐽𝜈+1 (𝑥) = 𝐽𝜈 (𝑥). (4.3.4)
𝑥
Special Functions 77
This identity can be rewritten as
2𝜈
𝐽𝜈+1 (𝑥) = 𝐽𝜈 (𝑥) − 𝐽𝜈−1 (𝑥). (4.3.5)
𝑥
Now, we can use it to compute
√ Bessel functions of higher order from lower ones.
Recall that Γ(1/2) = 𝜋. Then,
∞
r ∞
√ Õ (−1)𝑚 𝑥 2𝑚 2Õ (−1)𝑚 𝑥 2𝑚+1
𝐽1/2 (𝑥) = 𝑥 2𝑚+1/2𝑚!Γ(𝑚 + 3/2)
= 2𝑚+1𝑚!Γ(𝑚 + 1/2)
.
𝑚=0
2 𝑥 𝑚=0
2
However,
2𝑚𝑚! = 2𝑚(2𝑚 − 2) · · · 4 · 2.
2𝑚+1 Γ(𝑚 + 1/2) = 2𝑚+1 (𝑚 + 1/2)(𝑚 − 1/2) · · · (3/2) · (1/2)Γ(1/2)
√
= (2𝑚 + 1)(2𝑚 − 1) · · · 3 · 1 · 𝜋 .
√
22𝑚+1𝑚!Γ(𝑚 + 1/2) = [2𝑚𝑚!] [2𝑚+1 Γ(𝑚 + 1/2)] = (2𝑚 + 1)! 𝜋 .
Hence,
∞
r r
2 Õ (−1)𝑚 𝑥 2𝑚+1 2
𝐽1/2 (𝑥) = √ = sin 𝑥 . (4.3.6)
𝑥 𝑚=0 (2𝑚 + 1)! 𝜋 𝜋𝑥
√
Multiply by 𝑥, differentiate with respect to 𝑥, and use (4.3.1) with 𝜈 = 1/2 to
obtain r
√
0 2 √
0 √
𝑥 𝐽1/2 (𝑥) = cos 𝑥, 𝑥 𝐽1/2 (𝑥) = 𝑥 𝐽1/2−1 (𝑥).
𝜋
Therefore, r
2
𝐽−1/2 (𝑥) = cos 𝑥 . (4.3.7)
𝜋𝑥
Due to (4.3.5) 𝐽𝑘/2 (𝑥) for any integer 𝑘, can be expressed as a product of some
rational function and a trigonometric function.
To find a generating
 function for 𝐽𝑛 (𝑥) and 𝐽−𝑛 (𝑥), let us expand the function
exp 𝑡𝑥/2 − 𝑥/(2𝑡) . We find that
∞
! ∞ !
 𝑡𝑥 𝑥 Õ 1  𝑡𝑥  𝑟 Õ 1  𝑥  𝑠
exp − =
2 2𝑡 𝑟 =0
𝑟! 2 𝑠=0
𝑠! 2𝑡
∞
! ∞
!
Õ 1  𝑥  𝑟 𝑟 Õ (−1)𝑠  𝑥  𝑠 −𝑠
= 𝑡 𝑡
𝑟 =0
𝑟 ! 2 𝑠=0
𝑠! 2
∞ Õ ∞
Õ (−1) 𝑥 𝑟 +𝑠 𝑟 −𝑠
𝑠  
= 𝑡
𝑟 =0 𝑠=0
𝑟 !𝑠! 2
∞  ∞ 
Õ Õ (−1)𝑠  𝑥  𝑛+2𝑠  𝑛
= 
𝑡

𝑛=−∞ 𝑠=max{0,−𝑛}
𝑠!(𝑛 + 𝑠)! 2 
 
78 MA2020 Classnotes

For 𝑛 ≥ 0, the coefficient of 𝑡 𝑛 in the above expression is

∞ ∞
Õ (−1)𝑠 𝑥 2𝑠
Õ (−1)𝑠  𝑥  𝑛+2𝑠
= 𝑥𝑛 = 𝐽𝑛 (𝑥).
𝑠=0
(𝑠!(𝑛 + 𝑠)! 2 𝑠=0
2𝑛+2𝑠 𝑠!(𝑛 + 𝑠)!

And, for 𝑛 ≥ 0, the coefficient of 𝑡 −𝑛 is (shifting the index with 𝑘 = 𝑠 − 𝑛):

∞ ∞
Õ (−1)𝑠  𝑥  −𝑛+2𝑠 Õ (−1)𝑘+𝑛  𝑥  𝑛+2𝑠
= = (−1)𝑛 𝐽𝑛 (𝑥) = 𝐽−𝑛 (𝑥).
𝑠=𝑛
𝑠!(𝑛 − 𝑠)! 2 (𝑛 + 𝑘)!𝑘! 2
𝑘=0

We thus conclude that the generating function for 𝐽𝑛 (𝑥) for 𝑛 ∈ Z is

 𝑡𝑥 𝑥
exp − .
2 2𝑡
It means
 𝑡𝑥 ∞
𝑥 Õ
exp − = 𝐽𝑛 (𝑥)𝑡 𝑛 . (4.3.8)
2 2𝑡 𝑛=−∞
Some more properties of Bessel functions of first kind are to be found in the
exercises.
The zeros of Bessel functions of first kind play an important role in modeling
of vibrations. It is known that there are infinite number of positive zeros of 𝐽𝑛 (𝑥).
It is also known that between any two zeros of 𝐽𝑛 (𝑥) there exists a unique zero of
𝐽𝑛+1 (𝑥).

4.4 Sturm-Liouville problems

We have seen that the Legendre polynomials are orthogonal in the sense that
∫1
𝑃 (𝑥)𝑃𝑛 (𝑥) 𝑑𝑥 = 0 when 𝑚 ≠ 𝑛. A similar relation holds for Bessel func-
−1 𝑚
tions. There is a generalization of all these types of functions that are defined by a
second order ODE. We will discuss this generalization here. Later we will conclude
many useful properties about Bessel functions using this generalized problem.
Any ODE in the following form is called a Sturm-Liouville equation:
0 
𝑝 (𝑥)𝑦 0 + 𝑞(𝑥) + 𝜆𝑟 (𝑥) 𝑦 = 0 for 𝑎 < 𝑥 < 𝑏
 
(4.4.1)

Here, 𝜆 ∈ R is a parameter.

(4.1) Example
1. The simple harmonic motion equation 𝑦 00 +𝑛 2𝑦 = 0 is a Sturm-Liouville equation
with 𝑝 (𝑥) = 1, 𝑞(𝑥) = 0, 𝑟 (𝑥) = 1 and 𝜆 = 𝑛 2 .
Special Functions 79
2. The Legendre equation (1 − 𝑥 2 )𝑦 00 − 2𝑥𝑦 0 + 𝑝 (𝑝 + 1)𝑦 = 0 is a Sturm-Liouville
equation with 𝑝 (𝑥) = 1 − 𝑥 2 , 𝑞(𝑥) = 0, 𝑟 (𝑥) = 1 and 𝜆 = 𝑝 (𝑝 + 1).
3. The Bessel’s equation

𝑑 2𝑦 𝑑𝑦
𝑡2 2
+𝑡 + (𝑡 2 − 𝜈 2 )𝑦 = 0 for 𝑡 > 0
𝑑𝑡 𝑑𝑡
is a Sturm-Liouville equation. To see this, put 𝑡 = 𝑘𝑥 for 𝑘 > 0. We have

𝑑𝑦 𝑑𝑦 𝑑 2𝑦 2
2𝑑 𝑦
=𝑘 , = 𝑘 .
𝑑𝑥 𝑑𝑡 𝑑𝑥 2 𝑑𝑡 2
Then the above Bessel equation is reduced to

𝑑 2𝑦 𝑑𝑦
𝑘 2𝑥 2 2
+ 𝑘𝑥 + (𝑘 2𝑥 2 − 𝜈 2 )𝑦 = 0 or,
𝑑𝑡 𝑑𝑡

𝑥 2𝑦 00 + 𝑥𝑦 0 + (𝑘 2𝑥 2 − 𝜈 2 )𝑦 = 0.
However, 𝑥 (𝑥𝑦 0) 0 = 𝑥 (𝑥𝑦 00 +𝑦 0) = 𝑥 2𝑦 00 +𝑥𝑦 0. Hence, the above ODE is rewritten
as
 𝜈2 
0 0
(𝑥𝑦 ) + − + 𝜆𝑥 𝑦 = 0 where 𝜆 = 𝑘 2 .
𝑥
This is a Sturm-Liouville equation with 𝑝 (𝑥) = 𝑥, 𝑞(𝑥) = −𝜈 2 /𝑥, 𝑟 (𝑥) = 𝑥 for
𝑥 > 0.
Notice that 𝐽𝑛 (𝜆𝑥) satisfies this ODE.

With the Sturm-Liouville equation, we associate one of the following conditions:

𝑘 1𝑦 (𝑎) + 𝑘 2𝑦 0 (𝑎) = 0, ℓ1𝑦 (𝑏) + ℓ2𝑦 0 (𝑏) = 0 (4.4.2)

𝑝 (𝑎) = 𝑝 (𝑏), 𝑦 (𝑎) = 𝑦 (𝑏), 𝑦 0 (𝑎) = 𝑦 0 (𝑏). (4.4.3)

𝑝 (𝑎) = 0, ℓ1𝑦 (𝑏) + ℓ2𝑦 0 (𝑏) = 0, 𝑦 (𝑥) remains bounded. (4.4.4)
𝑘 1𝑦 (𝑎) + 𝑘 2𝑦 0 (𝑎) = 0, 𝑝 (𝑏) = 0, 𝑦 (𝑥) remains bounded. (4.4.5)
Here, 𝑘 1, 𝑘 2, ℓ1, ℓ2 are constants where at least one 𝑘 is nonzero and at least one ℓ is
nonzero, 𝑝 (𝑥), 𝑝 0 (𝑥), 𝑞(𝑥), 𝑟 (𝑥) are continuous on 𝑎 ≤ 𝑥 ≤ 𝑏, and 𝑝 (𝑥) is a non-zero
function. We also assume that either 𝑟 (𝑥) > 0 for all 𝑥 ∈ [𝑎, 𝑏] or 𝑟 (𝑥) < 0 for all
𝑥 ∈ [𝑎, 𝑏].
The conditions in (4.4.2)-(4.4.5) are prescribed at two points instead of at one
single point; thus these conditions are called boundary conditions. Accordingly,
the Sturm-Liouville equation (4.4.1) along with one of these boundary conditions
is called a Sturm-Liouville problem. The names associated with these problems
are as follows:
80 MA2020 Classnotes

regular Sturm-Liouville problem: (4.4.1) and (4.4.2)

periodic Sturm-Liouville problem: (4.4.1) and (4.4.3)
singular Sturm-Liouville problems: (4.4.1) with any one of (4.4.4) or (4.4.5)
We must remember that if a solution of the BVP exists, then it must be well defined
over the interval [𝑎, 𝑏].
If the zero function is a solution of a Sturm-Liouville problem, then it is called
the trivial solution. We are interested in getting non-trivial solutions.
Suppose a Sturm-Liouville problem is given. Corresponding to each value of
the parameter 𝜆, there may or may not exist a nontrivial solution of the problem.
Those values of 𝜆 for which the problem has a non-trivial solution are called
eigenvalues. Corresponding to an eigenvalue 𝜆, the non-trivial solutions 𝑦 (𝑥) are
called eigenfunction.

(4.2) Example
Find the eigenvalues and eigenfunctions of the Sturm-Liouville problem
𝑦 00 + 𝜆𝑦 = 0, 𝑦 (0) = 0, 𝑦 (𝜋) = 0.
This is a regular Sturm-Liouville problem with 𝑎 = 0, 𝑏 = 𝜋, 𝑝 (𝑥) = 1, 𝑞(𝑥) = 0,
𝑟 (𝑥) = 1, and 𝑘 1 = ℓ1 = 1, 𝑘 2 = ℓ2 = 0. Since the boundary conditions are given
at 𝑥 = 0 and at 𝑥 = 𝜋, the eigenfunctions if exist, must be defined over the interval
[0, 𝜋].
For 𝜆 = 0, the equation is 𝑦 00 = 0 giving the general solution as 𝑦 (𝑥) = 𝑐 1 + 𝑐 2𝑥.
Now, 𝑦 (0) = 0 ⇒ 𝑐 1 = 0. So, 𝑦 (𝑥) = 𝑐 1𝑥. Then, 𝑦 (𝜋) = 0 ⇒ 𝑐 2 = 0.
So, 𝑦 (𝑥) = 0, the zero function. Thus, 𝜆 = 0 is not an eigenvalue; it means that
corresponding to 𝜆 = 0, there does not exist any eigenfunction (non-trivial solution).
Let 𝜆 < 0. Write 𝜆 = −𝛼 2 for nonzero 𝛼 ∈ R. The ODE is 𝑦 00 = 𝛼 2𝑦. Its general
solution is 𝑦 = 𝑐 1𝑒 𝛼𝑥 + 𝑐 2𝑒 −𝛼𝑥 . The boundary conditions imply that 𝑐 1 + 𝑐 2 = 0 and
𝑐 1𝑒 𝜋 + 𝑐 2𝑒 −𝜋 = 0. The solution of these two linear equations in 𝑐 1, 𝑐 2 is unique and
it is 𝑐 1 = 0 = 𝑐 2 . Consequently, 𝑦 (𝑥) = 0, the zero function. Hence, no negative
number is an eigenvalue√of this Sturm-Liouville problem.
Let 𝜆 > 0. Write 𝛽 = 𝜆. The ODE is 𝑦 00 + 𝛽 2𝑦 = 0. Its general solution is 𝑦 (𝑥) =
𝑐 1 cos(𝛽𝑥) +𝑐 2 sin(𝛽𝑥). Now, 𝑦 (0) = 0 implies 𝑐 1 = 0. So, 𝑦 (𝑥) = 𝑐 2 sin(𝛽𝑥). Then
𝑦 (𝜋) = 𝑐 2 sin(𝛽𝜋). If 𝑐 2 = 0, then 𝑦 (𝑥) = 0, the zero function. Thus, in order that
𝑦 (𝑥) be non-trivial, we must have 𝑐 2 ≠ 0. Then, sin(𝛽𝑥) = 0 ⇒ 𝛽 ∈ Z.
Write 𝛽 = 𝑛 ∈ Z. Then 𝜆 = 𝑛 2 for 𝑛 ∈ Z, are the eigenvalues. That is, the
eigenvalues are 𝜆 = 𝑛 2 for 𝑛 = 0, 1, 2, 3, . . . and the corresponding eigenfunctions
are 𝑦 (𝑥) = sin(𝑛𝑥) defined over the interval [0, 𝜋].

(4.3) Example
Find the eigenvalues and eigenfunctions of the Sturm-Liouville problem
𝑦 00 + 𝜆𝑦 = 0, 𝑦 (0) = 0, 𝑦 0 (𝜋) = 0.
Special Functions 81
This is again a regular Sturm-Liouville problem. As earlier, we consider three cases.
If 𝜆 = 0, then the general solution is 𝑦 (𝑥) = 𝑐 1 + 𝑐 2𝑥. Now, 𝑦 (0) = 0 ⇒ 𝑐 1 = 0.
Then 𝑦 (𝑥) = 𝑐 2𝑥 ⇒ 𝑦 0 (𝑥) = 𝑐 2 . Then, 𝑦 0 (𝜋) = 0 ⇒ 𝑐 2 = 0. So, 𝑦 (𝑥) = 0.
Therefore, 0 is not an eigenvalue.
If 𝜆 < 0, then write 𝜆 = −𝛼 2 for 𝛼 > 0. The general solution is 𝑦 (𝑥) =
𝑐 1𝑒 𝛼𝑥 + 𝑐 2𝑒 −𝛼𝑥 so that 𝑦 0 (𝑥) = 𝑐 1𝛼𝑒 𝛼𝑥 − 𝑐 2𝛼𝑒 −𝛼𝑥 . The boundary conditions imply
𝑐 1 + 𝑐 2 = 1 and 𝑐 1𝛼𝑒 𝛼𝜋 − 𝑐 2𝛼𝑒 −𝛼𝜋 = 0. It gives 𝑐 1 = 𝑐 2 = 0 so that 𝑦 (𝑥) = 0, the zero
function. Hence, negative numbers are not eigenvalues.
So, let 𝜆 = 𝛽 2 for 𝛽 > 0. The general solution is 𝑦 (𝑥) = 𝑐 1 cos(𝛽𝑥) + 𝑐 2 sin(𝛽𝑥).
Now, 𝑦 (0) = 0 ⇒ 𝑐 1 = 0 so that 𝑦 (𝑥) = 𝑐 2 sin(𝛽𝑥). Then 𝑦 0 (𝑥) = 𝑐 2 𝛽 cos(𝛽𝑥). The
boundary condition 𝑦 0 (𝜋) = 0 implies 𝑐 2 𝛽 cos(𝛽𝜋) = 0. Now, 𝑐 2 = 0 would give
only trivial solution. Otherwise, 𝛽 cos(𝛽𝜋) = 0 ⇒ cos(𝛽𝜋) = 0 ⇒ 𝛽 = 𝑛 + 1/2
for 𝑛 ∈ Z. Thus, the eigenvalues are

(2𝑛 + 1) 2
𝜆𝑛 = 𝛽 2 = for 𝑛 = 0, 1, 2, 3, . . .
4
Notice that negative values of 𝑛 give rise to already listed eigenvalues. The corre-
sponding eigenfunctions are

𝑦𝑛 (𝑥) = sin(𝛽𝑥) = sin(𝑛 + 1/2)𝑥 for 𝑛 = 0, 1, 2, 3, . . . .

(4.4) Example
Find the eigenvalues and eigenfunctions of the periodic Sturm-Liouville problem

𝑦 00 + 𝜆𝑦 = 0, 𝑦 (0) = 𝑦 (ℓ), 𝑦 0 (0) = 𝑦 0 (ℓ)

where ℓ > 0 is given.

If 𝜆 = 0, then the general solution is 𝑦 (𝑥) = 𝑐 1 + 𝑐 2𝑥. Now, 𝑦 (0) = 𝑦 (ℓ) ⇒ 𝑐 1 =
𝑐 1 + 𝑐 2 ℓ ⇒ 𝑐 2 = 0 ⇒ 𝑦 (𝑥) = 𝑐 1 , which is a nonzero function for 𝑐 1 ≠ 0. Thus,
𝜆 = 0 is an eigenvalue and 𝑦 (𝑥) = 1 is a corresponding eigenfunction.
If 𝜆 < 0, say, 𝜆 = −𝛼 2 for 𝛼 > 0, then the general solution is 𝑦 (𝑥) = 𝑐 1𝑒 𝛼𝑥 +𝑐 2𝑒 −𝛼𝑥 .
The boundary conditions imply

𝑐 1 (1 − 𝑒 𝛼ℓ ) = 𝑐 2 (𝑒 −𝛼ℓ − 1), 𝑐 1 (1 − 𝑒 𝛼ℓ ) = −𝑐 2 (𝑒 −𝛼ℓ − 1).

Solving these, we get 𝑐 2 = 0 = 𝑐 1 . This leads to the trivial solution. So, no negative
number can be an eigenvalue.
If 𝜆 > 0, say, 𝜆 = 𝛽 2 for 𝛽 > 0, then the general solution is 𝑦 = 𝑐 1 cos(𝛽𝑥) +
𝑐 2 sin(𝛽𝑥). The boundary conditions give



𝑐 1 1 − cos(𝛽ℓ) = 𝑐 2 sin(𝛽ℓ), 𝑐 1 1 − cos(𝛽ℓ) = −𝑐 1 sin(𝛽ℓ).
82 MA2020 Classnotes


Eliminating 𝑐 2 , we obtain 2𝑐 1 1 − cos(𝛽ℓ) = 0. It implies either 𝑐 1 = 0 or
cos(𝛽ℓ) = 1.
If 𝑐 1 = 0, then 𝑐 2 = 0 so that 𝑦 (𝑥) is the trivial solution. This does not give any
eigenvalue. So, let cos(𝛽ℓ) = 1. Then, 𝛽ℓ = 2𝑛𝜋 for 𝑛 ∈ Z. Then,

𝜆𝑛 = 𝛽 2 = 4𝑛 2 𝜋 2 /ℓ 2 for 𝑛 = 0, 1, 2, 3, . . .

are the eigenvalues. The corresponding solutions are

𝑦𝑛 (𝑥) = 𝑐 1 cos(𝛽𝑛 𝑥) + 𝑐 2 sin(𝛽𝑛 𝑥), 𝛽𝑛 = 2𝑛𝜋/ℓ for 𝑛 = 0, 1, 2, 3, . . .

Thus, both the functions cos(𝛽𝑛 𝑥) and sin(𝛽𝑛 𝑥) are eigenfunctions associated with
the eigenvalue 𝛽𝑛2 . That is, the eigenvalues and the corresponding eigenfunctions
are
4𝑛 2 𝜋 2 1
 2𝑛𝜋𝑥 
2
 2𝑛𝜋𝑥 
𝜆𝑛 = , 𝑦 𝑛 (𝑥) = cos , 𝑦𝑛 (𝑥) = sin
ℓ2 ℓ ℓ
for 𝑛 = 0, 1, 2, 3, . . ., defined over [0, ℓ].

(4.5) Example
Find the eigenvalues and eigenfunctions of the regular Sturm-Liouville problem

𝑦 00 + 𝜆𝑦 = 0, 𝑦 (0) = 𝑦 0 (0), 𝑦 (1) + 𝑦 0 (1) = 0.

Notice that the eigenfunctions must be well defined over [0, 1]. As earlier we
consider the three cases.
If 𝜆 = 0, then the general solution is 𝑦 (𝑥) = 𝑐 1 + 𝑐 2𝑥. The boundary conditions
give 𝑐 1 = 𝑐 2, 2𝑐 1 + 𝑐 2 = 0 ⇒ 𝑐 1 = 0 = 𝑐 2 . So, 𝑦 (𝑥) = 0. Hence, 𝜆 = 0 is not an
eigenvalue.
Let 𝜆 < 0. Write 𝜆 = −𝛼 2 for 𝛼 > 0. The general solution is 𝑦 (𝑥) = 𝑐 1𝑒 𝛼𝑥 +𝑐 2𝑒 −𝛼𝑥 .
The boundary conditions give

𝑐 1 (1 − 𝛼) = −𝑐 2 (1 + 𝛼), 𝑐 1 (1 + 𝛼)𝑒 𝛼 + 𝑐 2 (1 − 𝛼)𝑒 −𝛼 = 0.

 

If 𝛼 = 1, then the first equation gives 𝑐 2 = 0; then the second equation gives 𝑐 1 = 0.
It leads to the trivial solution. So, let 𝛼 ≠ 1. Eliminating 𝑐 2 from the above
equations, we get
𝑐 1 (1 + 𝛼) 2𝑒 𝛼 − (1 − 𝛼) 2𝑒 −𝛼 = 0.
 

Since the bracketed term is nonzero, 𝑐 1 = 0. It then follows that 𝑐 2 = 0 so that there
is no non-trivial solution. In any case, no non-trivial solution exists. So, a negative
number cannot be an eigenvalue.
Then, consider 𝜆 > 0. Write 𝜆 = 𝛽 2 for 𝛽 > 0. The general solution is
𝑦 (𝑥) = 𝑐 1 cos(𝛽𝑥) + 𝑐 2 sin(𝛽𝑥). The boundary conditions give

𝑐 1 = 𝛽𝑐 2, 𝑐 1 cos 𝛽 + 𝑐 2 sin 𝛽 − 𝛽𝑐 1 sin 𝛽 + 𝛽𝑐 2 cos 𝛽 = 0.

Special Functions 83
Eliminating 𝑐 1 , we have 𝑐 2 2𝛽 cos 𝛽 + (1 − 𝛽 2 ) sin 𝛽 = 0. If 𝑐 2 = 0, then 𝑐 1 =
 

0 and it leads to the trivial solution. For a non-trivial solution, we must have
2𝛽 cos 𝛽 + (1 − 𝛽 2 ) sin 𝛽 = 0. It gives
2𝛽
tan 𝛽 = .
𝛽2 −1
That is, the eigenvalues are 𝛽 2 , where 𝛽 satisfies the above equation. The corre-
sponding eigenfunctions are 𝑦 1 (𝑥) = cos(𝛽𝑥) and 𝑦 2 (𝑥) = sin(𝛽𝑥).

(4.6) Example
Find the eigenvalues and eigenfunctions of Bessel’s equation
2𝑦
2𝑑𝑑𝑦
𝑡 2
+ (𝑡 2 − 𝜈 2 )𝑦 = 0 for 𝑡 > 0
+𝑡
𝑑𝑡 𝑑𝑡
with the condition that the solution remains bounded on the interval [0, 𝑎] and
𝑦 (𝑎) = 0.
See Example 4.1(3); taking 𝑡 = 𝑘𝑥 for 𝑘 > 0, the ODE is transformed to the
Sturm-Liouville equation
 𝜈2 
(𝑥𝑦 0) 0 + − + 𝜆𝑥 𝑦 = 0 where 𝜆 = 𝑘 2 .
𝑥
Notice that 𝑝 (0) = 0 so that this is a singular Sturm-Lioville problem, where
𝑦 (𝑎) = 0.
The linearly independent solutions of the above Bessel equation are 𝐽𝑛 (𝑡) and
𝑌𝑛 (𝑡). Hence, the general solution of the Sturm-Liouville equation is
𝑦 (𝑥) = 𝑐 1 𝐽𝑛 (𝑘𝑥) + 𝑐 2𝑌𝑛 (𝑘𝑥).
Recall that 𝑌𝑛 (𝑘𝑥) → ∞ as 𝑥 → 0. Since we need only bounded solutions, we must
set 𝑐 2 = 0. Thus, the required non-trivial solution is 𝑦 (𝑥) = 𝑐 1 𝐽𝑛 (𝑘𝑥).
Write 𝑅 = 𝑎/𝑘. When 𝑡 = 𝑎, we have 𝑘𝑥 = 𝑎 ⇒ 𝑥 = 𝑅. The boundary condition
says that 𝐽𝑛 (𝑎) = 0 or
𝐽𝑛 (𝑘𝑅) = 0.
This condition is satisfied when 𝑘𝑅 is a zero of 𝐽𝑛 (𝑥). Denote the zeros of 𝐽𝑛 (𝑥) by
𝑧𝑛,𝑟 with 𝑟 = 1, 2, 3, . . .. [It is known that there are infinite number of zeros of 𝐽𝑛 (𝑥)]
Then, the values of 𝑘 are
𝑧𝑛,𝑟
𝑘= for 𝑟 = 1, 2, 3, . . .
𝑅
As 𝜆 = 𝑘 2 , the eigenvalues and the corresponding eigenfunctions are
𝑧 2 𝑧 𝑥 
𝑛,𝑟 𝑛,𝑟
𝜆𝑟 = , 𝑦𝑟 (𝑥) = 𝐽𝑛 for 𝑟 = 1, 2, 3, . . .
𝑅 𝑅
where 𝑧𝑛,𝑟 is the 𝑟 th positive zero of 𝐽𝑛 (𝑥).
84 MA2020 Classnotes

4.5 Orthogonality
The most important property of eigenfunctions of a Sturm-Liouville problem is that
the eigenfunctions are orthogonal. For the plane vectors, orthogonality is obtained
via the dot product. To generalize this notion, we introduce the so called inner
products of funtions.
Let 𝑦1 (𝑥), 𝑦2 (𝑥), . . . be functions defined on an interval [𝑎, 𝑏]. Let 𝑟 (𝑥) be a
positive function defined on [𝑎, 𝑏], that is, 𝑟 (𝑥) > 0 for each 𝑥 ∈ [𝑎, 𝑏]. Let
𝑚, 𝑛 ∈ N. The inner product with weight 𝑟 (𝑥) of 𝑦𝑚 (𝑥) and 𝑦𝑛 (𝑥) is denoted by
h𝑦𝑚 , 𝑦𝑛 i and is defined as
∫𝑏
h𝑦𝑚 , 𝑦𝑛 i := 𝑟 (𝑥) 𝑦𝑚 (𝑥) 𝑦𝑛 (𝑥) 𝑑𝑥 .
𝑎
It follows that when 𝑚 = 𝑛, h𝑦𝑚 , 𝑦𝑚 i ≥ 0. The norm of a function 𝑦𝑚 (𝑥) is denoted
by k𝑦𝑚 k and is defined as
s
p ∫𝑏
k𝑦𝑚 k = h𝑦𝑚 , 𝑦𝑚 i = 𝑟 (𝑥) [𝑦𝑚 (𝑥)] 2 𝑑𝑥 .
𝑎

We say that 𝑦𝑚 and 𝑦𝑛 are orthogonal to each other with weight 𝑟 (𝑥) iff h𝑦𝑚 , 𝑦𝑛 i = 0.
The functions 𝑦1 (𝑥), 𝑦2 (𝑥), . . . are called orthogonal with weight 𝑟 (𝑥) iff 𝑦𝑚 (𝑥) is
orthogonal to 𝑦𝑛 (𝑥) for all 𝑚, 𝑛 ∈ N, 𝑚 ≠ 𝑛. The functions 𝑦1 (𝑥), 𝑦2 (𝑥), . . . are
called orthonormal with weight 𝑟 (𝑥) iff they are orthogonal and the norm of each
𝑦 𝑗 is 1. This happens when for all 𝑚, 𝑛 ∈ N, we find that
∫𝑏 (
0 if 𝑚 ≠ 𝑛
h𝑦𝑚 , 𝑦𝑛 i = 𝑟 (𝑥) 𝑦𝑚 (𝑥) 𝑦𝑛 (𝑥) 𝑑𝑥 = 𝛿𝑚,𝑛 =
𝑎 1 if 𝑚 = 𝑛.
(4.7) Example
The functions 𝑦 𝑗 (𝑥) = sin( 𝑗𝑥), 𝑗 = 1, 2, . . . are orthogonal on the interval [−𝜋, 𝜋]
with the weight function 𝑟 (𝑥) = 1. Indeed, if 𝑚 ≠ 𝑛, then
∫𝜋 ∫ ∫
1 𝜋 1 𝜋
h𝑦𝑚 , 𝑦𝑛 i = sin(𝑚𝑥) sin(𝑛𝑥) 𝑑𝑥 = cos(𝑚−𝑛)𝑥 𝑑𝑥− cos(𝑚+𝑛)𝑥 𝑑𝑥 = 0.
−𝜋 2 −𝜋 2 −𝜋
∫𝜋
2
Also, we find that k𝑦𝑚 k = h𝑦𝑚 , 𝑦𝑚 i = sin2 (𝑚𝑥) 𝑑𝑥 = 𝜋 .
−𝜋
sin 𝑥 sin(2𝑥) sin(3𝑥)
Hence, the functions √ , √ , √ , . . . are orthonormal.
𝜋 𝜋 𝜋
In general, if 𝑦1 (𝑥), 𝑦2 (𝑥), 𝑦3 (𝑥), . . . are orthogonal, then the normalized functions
𝑦1 (𝑥) 𝑦2 (𝑥) 𝑦3 (𝑥)
, , , ...
k𝑦1 k k𝑦2 k k𝑦3 k
Special Functions 85
are orthonormal. Similar to the last example, if 𝑚 ≠ 𝑛, then
∫𝜋 ∫𝜋 ∫𝜋
1 · cos(𝑚𝑥) 𝑑𝑥 = 0, 1 · sin(𝑚𝑥) 𝑑𝑥 = 0, cos(𝑚𝑥) cos(𝑛𝑥) 𝑑𝑥 = 0,
−𝜋 −𝜋 −𝜋
∫𝜋 ∫𝜋
cos(𝑚𝑥) sin(𝑛𝑥) 𝑑𝑥 = 0, sin(𝑚𝑥) sin(𝑛𝑥) 𝑑𝑥 = 0, and
−𝜋 −𝜋
∫𝜋 ∫𝜋 ∫𝜋
2
1 𝑑𝑥 = 2𝜋, cos (𝑚𝑥) 𝑑𝑥 = 𝜋, sin2 (𝑚𝑥) 𝑑𝑥 = 𝜋 .
−𝜋 −𝜋 −𝜋

1 cos(𝑚𝑥) sin(𝑚𝑥)
Hence, √ , √ , √ for 𝑚 = 1, 2, 3, . . . are orthonormal.
2𝜋 𝜋 𝜋
We mention an important fact about Sturm-Liouville problems.

(4.8) Theorem
Consider the Sturm-Liouville problem (4.4.1) either with 𝑝 (𝑎) = 𝑝 (𝑏) = 0 or with
one of the boundary conditions in (4.4.2)-(4.4.5). Let 𝑝 (𝑥), 𝑞(𝑥), 𝑟 (𝑥), 𝑝 0 (𝑥) be
continuous and 𝑟 (𝑥) > 0 on 𝑎 ≤ 𝑥 ≤ 𝑏. Then all eigenvalues are real, and they
may be arranged in order as 𝜆1 < 𝜆2 < 𝜆3 < · · · , where lim 𝜆𝑛 = ∞. Further, if
𝑛→∞
𝑦𝑚 (𝑥) and 𝑦𝑛 (𝑥) are eigenfunctions corresponding to distinct eigenvalues 𝜆𝑚 and
𝜆𝑛 , respectively, then 𝑦𝑚 and 𝑦𝑛 are orthogonal with weight function 𝑟 (𝑥). That is,
∫𝑏
𝑟 (𝑥) 𝑦𝑚 (𝑥) 𝑦𝑛 (𝑥) 𝑑𝑥 = 0 for 𝑚 ≠ 𝑛.
𝑎

Proof. We prove only the orthogonality of the eigenfunctions corresponding to

distinct eigenvalues. So, let 𝜆𝑚 ≠ 𝜆𝑛 be eigenvalues with corresponding eigenfunc-
tions as 𝑦𝑚 (𝑥) and 𝑦𝑛 (𝑥). These eigenfunctions satisfy the Sturm-Lioville equation.
That is,
0 0
(𝑝𝑦𝑚 ) + (𝑞 + 𝜆𝑟 )𝑦𝑚 = 0, (𝑝𝑦𝑛0 ) 0 + (𝑞 + 𝜆𝑟 )𝑦𝑛 = 0.
Multiply the first with 𝑦𝑛 , the second with −𝑦𝑚 , and add to get

(𝜆𝑚 − 𝜆𝑛 )𝑟𝑦𝑚𝑦𝑛 = 𝑦𝑚 (𝑝𝑦𝑛0 ) 0 − 𝑦𝑛 (𝑝𝑦𝑚

0 0
).

However,
 0 0
0  0  0  0
𝑝 (𝑦𝑛𝑦𝑚 − 𝑦𝑚 𝑦𝑛 ) = 𝑦𝑚 (𝑝𝑦𝑛0 ) − 𝑦𝑛 (𝑝𝑦𝑚
0
) = 𝑦𝑚 (𝑝𝑦𝑛0 ) − 𝑦𝑛 (𝑝𝑦𝑚
0
)
0
= 𝑦𝑚 (𝑝𝑦𝑛0 ) + 𝑦𝑚 (𝑝𝑦𝑛0 ) 0 − 𝑦𝑛0 (𝑝𝑦𝑚
0
) − 𝑦𝑛 (𝑝𝑦𝑚 0 0
) = 𝑦𝑚 (𝑝𝑦𝑛0 ) 0 − 𝑦𝑛 (𝑝𝑦𝑚
0 0
).

Hence, (𝜆𝑚 − 𝜆𝑛 )𝑟𝑦𝑚𝑦𝑛 = 𝑝 (𝑦𝑛0 𝑦𝑚 − 𝑦𝑚 0 𝑦 ) 0 . Integrating from 𝑎 to 𝑏, we obtain

 
𝑛
∫𝑏 h i𝑏
𝜙 := (𝜆𝑚 − 𝜆𝑛 ) 𝑟𝑦𝑚𝑦𝑛 𝑑𝑥 = 𝑝 (𝑦𝑛0 𝑦𝑚 − 𝑦𝑚0
𝑦𝑛 )
𝑎 𝑎
 0 0
  0 0

= 𝑝 (𝑏) 𝑦𝑛 (𝑏)𝑦𝑚 (𝑏) − 𝑦𝑚 (𝑏)𝑦𝑛 (𝑏) − 𝑝 (𝑎) 𝑦𝑛 (𝑎)𝑦𝑚 (𝑎) − 𝑦𝑚 (𝑎)𝑦𝑛 (𝑎) .
86 MA2020 Classnotes

The orthogonality of the eigenfunction is proved if 𝜙 evaluates to 0. We show that

this is the case by breaking into following cases:
Case 1: 𝑝 (𝑎) = 0 = 𝑝 (𝑏). Then, 𝜙 = 0.
Case 2: 𝑝 (𝑎) ≠ 0, 𝑝 (𝑏) = 0. Then 𝜙 = −𝑝 (𝑎) 𝑦𝑛0 (𝑎)𝑦𝑚 (𝑎) − 𝑦𝑚 0 (𝑎)𝑦 (𝑎) . The
 
𝑛
boundary conditions in (4.4.5) are applicable. Since 𝑦𝑚 and 𝑦𝑛 are solutions of the
BVP, we have
0
𝑘 1𝑦𝑚 (𝑎) + 𝑘 2𝑦𝑚 (𝑎) = 0 = 𝑘 1𝑦𝑛 (𝑎) + 𝑘 2𝑦𝑛0 (𝑎).
At least one of 𝑘 1, 𝑘 2 is nonzero. Suppose 𝑘 2 ≠ 0. Multiply the first equation by
𝑦𝑚 (𝑎), the second by −𝑦𝑛 (𝑎) and add to get

𝑘 2 𝑦𝑛0 (𝑎)𝑦𝑚 (𝑎) − 𝑦𝑚

0
 
(𝑎)𝑦𝑛 (𝑎) = 0.

As 𝑘 2 ≠ 0, we get 𝑦𝑛0 (𝑎)𝑦𝑚 (𝑎) − 𝑦𝑚 0 (𝑎)𝑦 (𝑎) = 0 so that 𝜙 = 0. A similar proof is

 
𝑛
given when 𝑘 1 ≠ 0.
Case 3: 𝑝 (𝑎) = 0, 𝑝 (𝑏) ≠ 0. This case is similar to Case 2.
Case 4: 𝑝 (𝑎) ≠ 0, 𝑝 (𝑏) ≠ 0, 𝑝 (𝑎) ≠ 𝑝 (𝑏). We use both the conditions in (4.4.2)
and proceed as in Case 2.
Case 5: 𝑝 (𝑎) = 𝑝 (𝑏). The condition (4.4.3) says that 𝑦 (𝑎) = 𝑦 (𝑏) and 𝑦 0 (𝑎) = 𝑦 0 (𝑏).
These are satisfied for both 𝑦 = 𝑦𝑚 and 𝑦 = 𝑦𝑛 . Then, 𝜙 evaluates to 0.

(4.9) Example
Consider the Sturm-Liouville problem of (4.3):

𝑦 00 + 𝜆𝑦 = 0, 𝑦 (0) = 0, 𝑦 0 (𝜋) = 0.

Here, 𝑝 (𝑥) = 1, 𝑞(𝑥) = 0 and 𝑟 (𝑥) = 1. We found the eigenvalues and eigenfunc-
tions as
(2𝑛 + 1) 2
𝑦𝑛 (𝑥) = sin 𝑛 + 21 𝑥 ,


𝜆𝑛 = , 𝑛 = 0, 1, 2, 3, . . .
4
By (4.8), we conclude that
∫𝜋
sin 𝑚 + 12 𝑥 sin 𝑛 + 12 𝑥 𝑑𝑥 = 0 for 𝑚 ≠ 𝑛.

 

0

Of course, it is easy to verify this directly.

(4.10) Example
Legendre’e equation (1−𝑥 2 )𝑦 00 −2𝑥𝑦 0 +𝜌 (𝜌 +1)𝑦 = 0 is a Sturm-Liouville equation
with 𝑝 (𝑥) = 1 − 𝑥 2 , 𝑞(𝑥) = 0, 𝑟 (𝑥) = 1 and 𝜆 = 𝜌 (𝜌 + 1). Here, 𝑝 (−1) = 𝑝 (1) = 0.
Hence, this is a singular Sturm-Liouville problem on the interval −1 ≤ 𝑥 ≤ 1. We
know that 𝑃𝑛 (𝑥) is a solution of this equation for 𝜆 = 𝑛(𝑛 + 1), where 𝑛 = 0, 1, 2, . . ..
Special Functions 87
That is, corresponding to the eigenvalue 𝜆𝑛 = 𝑛(𝑛 + 1), the eigenfunction is 𝑃𝑛 (𝑥).
By (4.8), these eigenfunctions are orthogonal with weight 𝑟 (𝑥) = 1. It means that
∫1
𝑃𝑚 (𝑥)𝑃𝑛 (𝑥) 𝑑𝑥 = 0 for 𝑚 ≠ 𝑛.
−1

We have seen that this is the case.

(4.11) Example
As we have seen in (4.6), the Bessel’s equation

𝑑 2𝑦 𝑑𝑦
𝑡2 2
+𝑡 + (𝑡 2 − 𝜈 2 )𝑦 = 0 for 𝑡 > 0
𝑑𝑡 𝑑𝑡
with the condition that the solution remains bounded on [0, 𝑎] and 𝑦 (𝑎) = 0 is the
Sturm-Liouville problem (Take 𝑡 = 𝑘𝑥.)
 𝜈2 
(𝑥𝑦 0) 0 + − + 𝜆𝑥 𝑦 = 0 where 𝜆 = 𝑘 2 .
𝑥
Here, 𝑝 (0) = 0 so that this is a singular Sturm-Lioville problem, where 𝑦 (𝑅) = 0
with 𝑅 = 𝑎/𝑘. Its eigenvalues and eigenfunctions have been found to be
𝑧 2 𝑧 𝑥 
𝑛,𝑟 𝑛,𝑟
𝜆𝑟 = , 𝑦𝑟 (𝑥) = 𝐽𝑛 for 𝑟 = 1, 2, 3, . . .
𝑅 𝑅
where 𝑧𝑛,𝑟 is the 𝑟 th positive zero of 𝐽𝑛 (𝑥).
By (4.8), the eigenfunctions are orthogonal with weight 𝑟 (𝑥) = 𝑥 on the interval
[0, 𝑅]. That is,
∫𝑅 𝑧 𝑥  𝑧 𝑥 
𝑛,𝑚 𝑛,𝑗
𝑥 𝐽𝑛 𝐽𝑛 𝑑𝑥 = 0 for 𝑚 ≠ 𝑗 .
0 𝑅 𝑅
We see that the permissible values of 𝑘 in the transformation 𝑡 = 𝑘𝑥 are 𝑧𝑛,𝑟 /𝑅.
Notice that for fixed
 𝑛 and a fixed 𝑅 > 0, we have infinitely many orthogonal
𝑧𝑛,𝑚 𝑥
functions 𝐽𝑛 𝑅 . The 𝑅 in this orthogonality can be chosen according to our
convenience, but it is to be fixed.

The above example shows that there are infinitely many orthogonal sets of Bessel
functions, one for each of 𝐽0, 𝐽1, 𝐽2, . . . on an interval 0 ≤ 𝑥 ≤ 𝑅 with a fixed
positive 𝑅 of our choice and with the weight function 𝑟 (𝑥) = 𝑥.
We have only proved the orthogonality of the Bessel functions. In fact, the norms
of those can also be computed from the following result, which is left as an exercise.
 𝑧 𝑥  2 ∫ 𝑅 h  𝑧 𝑥 i 2
𝑛,𝑟 𝑛,𝑟 𝑅2 h  i 2
= 𝑑𝑥 = (4.5.1)

𝐽𝑛 𝑥 𝐽𝑛 𝐽𝑛+1 𝑧𝑛,𝑟 .
𝑅 0 𝑅 2
88 MA2020 Classnotes

Orthogonality helps in expanding functions as series of eigenfunctions just like

Fourier series. We have seen in § 4.1 how to express a function defined on [−1, 1]
as a series involving the Legendre polynomials. By using orthogonality of Bessel
functions, similar series expansion can be obtained.
Fix 𝑛 ∈ N ∪ {0}. Let 𝑓 (𝑥) be a real valued peicewise smooth function defined on
an interval 0 ≤ 𝑥 ≤ 𝑅. A Fourier-Bessel series of 𝑓 (𝑥) using the Bessel function
𝐽𝑛 may be written as
∞ 𝑧  𝑧 𝑥  𝑧 𝑥 
𝑛,𝑚 𝑥
Õ 𝑛,1 𝑛,2
𝑓 (𝑥) = 𝑎𝑚 𝐽𝑛 = 𝑎 1 𝐽𝑛 + 𝑎 2 𝐽𝑛 +···
𝑚=1
𝑅 𝑅 𝑅
 
𝑧 𝑥
Fix ℓ ∈ N. Multiply the above equation with 𝑥 𝐽𝑛 𝑛,ℓ𝑅 and integrate from 0 to 𝑅 to
get
∫𝑅 𝑧 𝑥  Õ∞ ∫𝑅 𝑧 𝑥  𝑧 𝑥 
𝑛,ℓ 𝑛,𝑚 𝑛,ℓ
𝑥 𝑓 (𝑥)𝐽𝑛 𝑑𝑥 = 𝑎𝑚 𝑥 𝐽𝑛 𝐽𝑛 𝑑𝑥 .
0 𝑅 𝑚=1 0 𝑅 𝑅
Due to orthogonality, the integral in the above summand is 0 when 𝑚 ≠ ℓ. So, we
obtain

𝑅2 2
∫𝑅 ∫𝑅 h 
𝑧 𝑥 
𝑛,ℓ 𝑧𝑛,ℓ 𝑥 i 2

𝑥 𝑓 (𝑥) 𝑗𝑛 𝑑𝑥 = 𝑎 ℓ 𝑥 𝐽𝑛 𝑑𝑥 = 𝐽𝑛+1 𝑧𝑛,ℓ .
0 𝑅 0 𝑅 2

The last equality follows from (4.5.1). This gives the coefficient 𝑎 ℓ for ℓ ∈ N. Thus,
the Fourier-Bessel series for 𝑓 (𝑥) on an interval [0, 𝑅] is given as follows:
∞ ∫𝑅
Õ 𝑧
𝑛,𝑚 𝑥
 2 𝑧
𝑛,𝑚 𝑥

𝑓 (𝑥) = 𝑎𝑚 𝐽𝑛 , where 𝑎𝑚 = 2 2 𝑥 𝑓 (𝑥) 𝐽𝑛 𝑑𝑥 .
𝑚=1
𝑅 𝑅 𝐽𝑛+1 (𝑧𝑛,𝑚 ) 0 𝑅
(4.5.2)
Notice that we have written 𝑓 (𝑥) is equal to its Fourier-Bessel series for deriving
the coefficients. However, the series so obtained may or may not converge to the
function 𝑓 (𝑥). This question of convergence is answered by the following result,
which we mention without proof.

(4.12) Theorem (Convergence of Fourier-Bessel series)

Let 𝑓 (𝑥) be a piecewise smooth function defined on the interval 0 < 𝑥 < 𝑅. Then
the Fourier-Bessel series (4.5.2) of 𝑓 (𝑥) converges to 𝑔(𝑥), where
(
𝑓 (𝑥) if 𝑓 is continuous at x
𝑔(𝑥) = 1  
2 𝑓 (𝑥+) + 𝑓 (𝑥−) if 𝑓 is discontinuous at x.

It thus follows that if 𝑓 (𝑥) is continuous on 0 < 𝑥 < 𝑅, then its Fourier-Bessel
series converges to 𝑓 (𝑥).
Special Functions 89
(4.13) Example
Find the Fourier-Bessel series for the function 𝑓 (𝑥) = 1 on 0 < 𝑥 < 1.
Here, 𝑅 = 1; we choose 𝑛 = 0. By (4.5.2), the Fourier-Bessel series of 𝑓 (𝑥) = 1 is
given by (Write 𝑧 0,𝑚 as 𝑧𝑚 .)
Õ∞
𝑎𝑚 𝐽0 (𝑧𝑚 𝑥)
𝑚=1
where 𝑧𝑚 for 𝑚 = 1, 2, 3, . . . are the positive zeros of 𝐽0 (𝑥) and the coefficients 𝑎𝑚
are given by
∫1
2
𝑎𝑚 = 2 𝑥 𝐽0 (𝑧𝑚 𝑥) 𝑑𝑥 .
𝐽1 (𝑧𝑚 ) 0
 0
We use the identity 𝑥 𝐽1 (𝑥) = 𝑥 𝐽0 (𝑥) given in (4.3.1) to evaluate the above integral.
Substitute 𝑡 = 𝑧𝑚 𝑥. Then, 𝑑𝑡 = 𝑧𝑚 𝑑𝑥, and when 𝑥 varies from 0 to 1, 𝑡 varies from
0 to 𝑧𝑚 . Hence,
∫ 𝑧𝑚 i 𝑧𝑚 2𝑧 𝐽 (𝑧 )
2 2 𝑚 1 𝑚 2
𝑎𝑚 = 2 2 𝑡 𝐽0 (𝑡) 𝑑𝑡 = 2 2 𝑡 𝐽1 (𝑡) = 2 2 = .
𝑧𝑚 𝐽1 (𝑧𝑚 ) 0 𝑧𝑚 𝐽1 (𝑧𝑚 ) 0 𝑧𝑚 𝐽1 (𝑧𝑚 ) 𝑧𝑚 𝐽1 (𝑧𝑚 )

Since 𝑓 (𝑥) is continuous everywhere on (0, 1), by the convergence theorem,

∞
Õ 2𝐽0 (𝑧𝑚 𝑥)
1= .
𝑧 𝐽 (𝑧 )
𝑚=1 𝑚 1 𝑚
5
Partial Differential Equations

5.1 Introduction
Suppose 𝑢 (𝑥, 𝑦) is a function of two independent variables. Instead of derivatives
𝜕𝑢 𝜕𝑢
we now think of its partial derivatives 𝑢𝑥 = and 𝑢𝑦 = . We may also have
𝜕𝑥 𝜕𝑦
higher order partial derivatives such as 𝑢𝑥𝑥 , 𝑢𝑥𝑦 , 𝑢𝑦𝑥 , 𝑢𝑦𝑦 , 𝑢𝑥𝑥𝑥 , etc.
An equation involving 𝑥, 𝑦, 𝑢 and some of its partial derivatives is called a partial
differential equation, or PDE for short. The order of the highest order derivative
of 𝑢 is called the order of the PDE.
Usually, we will be concerned with first and second order PDEs. Of course, there
can be more than two independent variables. We will be generally taking two or
three independent variables and one dependent variable.
The general form of a first order PDE with dependent variable 𝑢 and two inde-
pendent variables 𝑥, 𝑦 is
𝐹 (𝑥, 𝑦, 𝑢, 𝑢𝑥 , 𝑢𝑦 ) = 0
where 𝐹 is an expression (also a function) involving 𝑥, 𝑦, 𝑢, 𝑢𝑥 and 𝑢𝑦 . Similarly,
a general first order PDE with one dependent variable 𝑢 and three independent
variables 𝑥, 𝑦, 𝑧 may be written as
𝐹 (𝑥, 𝑦, 𝑧, 𝑢, 𝑢𝑥 , 𝑢𝑦 , 𝑢𝑧 ) = 0.
If such a function 𝐹 is linear in the dependent variable and its derivatives, then it is
called a linear PDE. Notice that in a linear PDE, the coefficients of the dependent
variable and its derivatives must be functions of 𝑥, 𝑦 only. The general first order
linear PDE with two independent variables looks like
𝑎(𝑥, 𝑦)𝑢𝑥 + 𝑏 (𝑥, 𝑦)𝑢𝑦 + 𝑐 (𝑥, 𝑦)𝑢 = 𝑑 (𝑥, 𝑦).
When 𝑑 (𝑥, 𝑦) = 0, the linear PDE is called homogeneous, else, it is called a
non-homogeneous PDE. For example, the following are first order linear PDEs:
𝑥𝑢𝑥 + 𝑦𝑢𝑦 − 𝑢 = 0.
𝑢𝑥 + (𝑥 + 𝑦)𝑢𝑦 − 5𝑢 = 𝑒 𝑥 .
𝑦𝑢𝑥 + 𝑥𝑦𝑢 = 𝑥𝑦.
(𝑦 − 𝑧)𝑢𝑥 + (𝑧 − 𝑥)𝑢𝑦 + (𝑥 − 𝑦)𝑢𝑧 = 0.

90
Partial Differential Equations 91
The first and the fourth are homogeneous, whereas the second and the third are
non-homogeneous.
A PDE which is not linear is called nonlinear. Among the nonlinear PDEs there
are some easier classes of problems. A first order PDE is called semilinear iff
the coefficients of the derivatives of the dependent variable are functions of the
independent variables only. A general form of a semilinear first order PDE with
two independent varaibles is

𝑎(𝑥, 𝑦)𝑢𝑥 + 𝑏 (𝑥, 𝑦)𝑢𝑦 = 𝑐 (𝑥, 𝑦, 𝑢).

A first order PDE is called quasi-linear iff the expression 𝐹 (· · · ) is linear in the
derivatives of the dependent variable. It means, the coefficients of the derivatives are
now allowed to involve the dependent variable. The general first order quasi-linear
PDE with two independent variables looks like

𝑎(𝑥, 𝑦, 𝑢)𝑢𝑥 + 𝑏 (𝑥, 𝑦, 𝑢)𝑢𝑦 = 𝑐 (𝑥, 𝑦, 𝑢).

Some examples of quasi-linear PDEs are

𝑥 (𝑦 2 + 𝑢)𝑢𝑥 − 𝑦 (𝑥 2 + 𝑢)𝑢𝑦 = (𝑥 2 − 𝑦 2 )𝑢.

𝑢𝑢𝑥 + 𝑢𝑦 + 𝑢 2 = 0.
(𝑦 2 − 𝑢 2 )𝑦𝑥 − 𝑥𝑦𝑢𝑦 = 𝑥𝑢.

Sometimes it is possible to use the methods of ordinary differential equations to

solve a PDE. This method is used when all the derivatives can be integrated with
respect to some independent variable, or when by substituting a derivative as a new
variable an ODE results. Usually, the general solution of an 𝑛th order PDE would
involve 𝑛 number of arbitrary functions. See the following examples.

(5.1) Example
Solve 𝑢𝑥 (𝑥, 𝑦) = 𝑥 + 𝑦.
Integrating with respect to 𝑥, where 𝑦 is kept constant, we get

𝑥2
∫
𝑢 (𝑥, 𝑦) = (𝑥 + 𝑦) 𝑑𝑥 = + 𝑥𝑦 + 𝑓 (𝑦).
2
Here, the constant of integration must not depend on 𝑥, but it can depend on 𝑦. So,
we had taken it as 𝑓 (𝑦), an arbitrary function of the variable 𝑦.

(5.2) Example
Solve 𝑢𝑥𝑦 (𝑥, 𝑦) = 0.
Integrating with respect to 𝑦, we get (𝑥 is kept constant)

𝑢𝑥 (𝑥, 𝑦) = 𝑓 (𝑥).
92 MA2020 Classnotes

Integrating with respect to 𝑥, we obtain

∫
𝑢 (𝑥, 𝑦) = 𝑓 (𝑥) 𝑑𝑥 + 𝑔(𝑦).

Since 𝑓 (𝑥) is an arbitrary function, we may write its integral as ℎ(𝑥), where this
ℎ(𝑥) is also an arbitrary function. Hence, the general solution of the PDE is
𝑢 (𝑥, 𝑦) = ℎ(𝑥) + 𝑔(𝑦) for arbitrary functions ℎ(𝑥) of 𝑥 and 𝑔(𝑦) of 𝑦.

There can be initial and boundary conditions along with a PDE, and they are taken
care while solving the PDE.

(5.3) Example
Find 𝑢 (𝑥, 𝑦) that satisfies the PDE 𝑢𝑥𝑥 = 𝑦 2 cos2 𝑥 and 𝑢 (0, 𝑦) = 0 = 𝑢 (𝜋/2, 𝑦).
Integrating the given equation with respect to 𝑥, we get
 𝑥 sin(2𝑥) 
𝑢𝑥 = 𝑦 2 + + 𝑓 (𝑦).
2 4
Here, 𝑓 (𝑦) is an arbitrary function of 𝑦 alone. Integrating once more with respect
to 𝑥, we obtain
 𝑥 2 cos(2𝑥) 
𝑢 = 𝑦2 − + 𝑓 (𝑦)𝑥 + 𝑔(𝑦).
4 8
The condition 𝑢 (0, 𝑦) = 0 implies
 1 𝑦2
0 = 𝑦2 − + 𝑔(𝑦) ⇒ 𝑔(𝑦) = .
8 8
Using this expression for 𝑔(𝑦) and using the condition 𝑢 (𝜋/2, 𝑦) = 0 we get
𝜋2 1 𝜋 𝑦2 𝜋 1  2
0 = 𝑦2 + + 𝑓 (𝑦) + ⇒ 𝑓 (𝑦) = − + 𝑦 .
16 8 2 8 8 2𝜋
Hence, the solution is
 𝑥 2 cos(2𝑥)   𝜋 1  2 𝑦2
𝑢 (𝑥, 𝑦) = 𝑦 2 − − + 𝑥𝑦 + .
4 8 8 2𝜋 8
Solutions of PDEs with a dependent variable and two independent variables
are also called integral surfaces. In such a case, e usually write the independent
variables as 𝑥, 𝑦 and the dependent variable as 𝑧 to rhyme with the geometrical
language.

5.2 Lagrange method

We will consider the method of characteristics by Lagrange for solving the quasi-
linear first order PDE. We assume that the coefficient functions are continuous in
Partial Differential Equations 93
the domain of consideration. Also, we assume that they are not simultaneously 0.
(Otherwise, the PDE is no more a PDE.) Lagrange’s method is encapsulated in the
following theorem.

(5.4) Theorem (Lagrange’s method of characteristics)

Suppose 𝑎(𝑥, 𝑦, 𝑢), 𝑏 (𝑥, 𝑦, 𝑢) and 𝑐 (𝑥, 𝑦, 𝑢) are continuous and they are not simul-
taneously 0 at any point in a domain. Then, the general solution of the first order
quasi-linear PDE
𝑎(𝑥, 𝑦, 𝑢)𝑢𝑥 + 𝑏 (𝑥, 𝑦, 𝑢)𝑢𝑦 = 𝑐 (𝑥, 𝑦, 𝑢) (5.2.1)
is given by 𝑓 (𝜙,𝜓 ) = 0, where 𝑓 is an arbitrary function of two variables, and
𝜙 (𝑥, 𝑦, 𝑢) = 𝑐 1, 𝜓 (𝑥, 𝑦, 𝑢) = 𝑐 2, for arbitrary constants 𝑐 1, 𝑐 2 , are solutions of

𝑑𝑥 𝑑𝑦 𝑑𝑢
= = . (5.2.2)
𝑎(𝑥, 𝑦, 𝑢) 𝑏 (𝑥, 𝑦, 𝑢) 𝑐 (𝑥, 𝑦, 𝑢)

Equations in (5.2.2) are called the characteristic equations of the PDE (5.2.1).
Their solutions 𝜙 (𝑥, 𝑦, 𝑢) = 𝑐 1 and 𝜓 (𝑥, 𝑦, 𝑢) = 𝑐 2 are called the characteristic
curves. Lagrange’s method of characteristic reduces the problem of solving the
quasi-linear first order PDE to solving two ODEs.
Proof. Suppose 𝜙 (𝑥, 𝑦, 𝑢) = 𝑐 1 and 𝜓 (𝑥, 𝑦, 𝑢) = 𝑐 2 are solutions of (5.2.2). Since
the PDE (5.2.1) has order 1, and 𝑓 (·, ·) is an arbitrary function of two variables,
𝑓 (𝜙,𝜓 ) = 0 is the general solution provided it is at all a solution.
Now, 𝑓 (𝜙,𝜓 ) = 0
is a solution means that if 𝑢 (𝑥, 𝑦) satisfies 𝑓 𝜙 (𝑥, 𝑦, 𝑢),𝜓 (𝑥, 𝑦, 𝑢) = 0, then 𝑢 (𝑥, 𝑦)
also satisfies the PDE (5.2.1). We show that this is the case.
So, suppose 𝑢 (𝑥, 𝑦) satisfies 𝑓 (𝜙,𝜓 ) = 0. Computing the differentials of 𝜙 and 𝜓
we get

𝑑𝜙 = 𝜙𝑥 𝑑𝑥 + 𝜙𝑦𝑑𝑦 + 𝜙𝑢 𝑑𝑢 = 0, 𝑑𝜓 = 𝜓𝑥 𝑑𝑥 + 𝜓𝑦𝑑𝑦 + 𝜓𝑢 𝑑𝑢 = 0.

However, 𝜙 (𝑥, 𝑦, 𝑢) = 𝑐 1 and 𝜓 (𝑥, 𝑦, 𝑢) = 𝑐 2 are solutions of (5.2.2). So,

𝑎𝜙𝑥 + 𝑏𝜙𝑦 + 𝑐𝜙𝑢 = 0, 𝑎𝜓𝑥 + 𝑏𝜓𝑦 + 𝑐𝜓𝑢 = 0.

Eliminating 𝑎 from these two equations, we get 𝑏 (𝜙𝑥𝜓𝑦 − 𝜙𝑦𝜓𝑥 ) = 𝑐 (𝜙𝑢𝜓𝑥 − 𝜙𝑥𝜓𝑢 ).
Eliminating 𝑏 we get 𝑎(𝜙𝑥𝜓𝑦 − 𝜙𝑦𝜓𝑥 ) = 𝑐 (𝜙𝑦𝜓𝑢 − 𝜙𝑢𝜓𝑦 ). Hence,

𝑎 𝑏 𝑐
= = . (5.2.3)
𝜙𝑦𝜓𝑢 − 𝜙𝑢𝜓𝑦 𝜙𝑢𝜓𝑥 − 𝜙𝑥𝜓𝑢 𝜙𝑥𝜓𝑦 − 𝜙𝑦𝜓𝑥

Since 𝑓 (𝜙,𝜓 ) = 0, differentiating with respect to 𝑥 and also 𝑦, and using the Chain
rule, we have
𝑓𝜙 (𝜙𝑥 + 𝜙𝑢𝑢𝑥 ) + 𝑓𝜓 (𝜓𝑥 + 𝜓𝑢𝑢𝑥 ) = 0.
94 MA2020 Classnotes

𝑓𝜙 (𝜙𝑦 + 𝜙𝑢𝑢𝑦 ) + 𝑓𝜓 (𝜓𝑦 + 𝜓𝑢𝑢𝑦 ) = 0.

Since 𝑓 (𝜙,𝜓 ) is an arbitrary function, 𝑓𝜙 and 𝑓𝜓 are not necessarily the zero functions.
Then, the above two linear equations have a non-trivial solution. So, the determinant
of the system is 0. That is,
(𝜙𝑥 + 𝜙𝑢𝑢𝑥 )(𝜓𝑦 + 𝜓𝑢𝑢𝑦 ) = (𝜙𝑦 + 𝜙𝑢𝑢𝑦 )(𝜓𝑥 + 𝜓𝑢𝑢𝑥 ).
It simplifies to
(𝜙𝑦𝜓𝑢 − 𝜙𝑢𝜓𝑦 )𝑢𝑥 + (𝜙𝑢𝜓𝑥 − 𝜙𝑥𝜓𝑢 )𝑦𝑦 = 𝜙𝑥𝜓𝑦 − 𝜙𝑦𝜓𝑥 .
By (5.2.3), 𝑎𝑢𝑥 + 𝑏𝑢𝑦 = 𝑐.
We remark that for more than two independent variables, the statement in (5.4)
also holds so that Lagrange’s method is still applicable. That is, to solve the
quasi-linear PDE
𝑎 1𝑢 1 + · · · + 𝑎𝑛𝑢𝑛 = 𝑐
where 𝑢 = 𝑢 (𝑥 1, 𝑥 2, . . . , 𝑥𝑛 ), 𝑎𝑖 = 𝑎𝑖 (𝑥 1, 𝑥 2, . . . , 𝑥𝑛 , 𝑢), 𝑢𝑖 = 𝑢𝑥𝑖 (𝑥 1, 𝑥 2, . . . , 𝑥𝑛 ) and
𝑐 = 𝑐 (𝑥 1, 𝑥 2, . . . , 𝑥𝑛 , 𝑢), we form the characteristic equations
𝑑𝑥 1 𝑑𝑥 2 𝑑𝑥𝑛 𝑑𝑢
= = ··· = = .
𝑎1 𝑎2 𝑎𝑛 𝑐
We get its solution as 𝜙 𝑗 (𝑥 1, 𝑥 2, . . . , 𝑥𝑛 ) = 𝑐 𝑗 for 𝑗 = 1, 2, . . . , 𝑛. Then, the general
solution of the PDE is given implicitly by 𝑓 (𝜙 1, 𝜙 2, . . . , 𝜙𝑛 ) = 0 for an arbitrary
function 𝑓 of 𝑛 arguments.
If 𝜙 (𝑥, 𝑦, 𝑢) = 𝑐 1 and 𝜓 (𝑥, 𝑦, 𝑢) = 𝑐 2 are solutions of the characteristic equations
in (5.2.2), then the general solution may also be written by assuming certain de-
pendence of these two constants. That is, we may write the general solution as
𝜙 (𝑥, 𝑦, 𝑢) = 𝑔(𝜓 (𝑥, 𝑦, 𝑢)) for an arbitrary function 𝑔(·). Notice that this is an explicit
way of writing the same general solution 𝑓 (𝜙,𝜓 ) = 0. The implicit way of writing
is more general than the explicit way. However, if one of the characteristic curves is
𝑢 = 𝑐 1 , then the explicit way of writing is as general as the implicit way of writing.

(5.5) Example
Find the general solution of the PDE 𝑢𝑥 + 𝑢𝑦 = 1.
The characteristic equations are 𝑑𝑥 = 𝑑𝑦 = 𝑑𝑢. Taking them in pairs and integrating,
we have
𝑑𝑥 − 𝑑𝑦 = 0 ⇒ 𝑥 − 𝑦 = 𝑐 1 .
𝑑𝑦 − 𝑑𝑢 = 0 ⇒ 𝑦 − 𝑢 = 𝑐 2 .
Thus, the general solution is 𝑓 (𝑥 − 𝑦, 𝑦 − 𝑢) = 0 for an arbitrary function 𝑓 of
two arguments. We may also write the general solution as 𝑦 − 𝑢 = 𝑔(𝑥 − 𝑦) or
𝑢 = 𝑦 − 𝑔(𝑥 − 𝑦) for an arbitrary function 𝑔 of one variable.
Partial Differential Equations 95
(5.6) Example
Find the general solution of the PDE 𝑥𝑢𝑥 + 𝑦𝑢𝑦 = 𝑢.
The characteristic equations are
𝑑𝑥 𝑑𝑦 𝑑𝑢
= = .
𝑥 𝑦 𝑢
Taking in pairs and integrating we obtain
𝑦 𝑢
= 𝑐 1, = 𝑐2.
𝑥 𝑥
Thus, the general solution is 𝑓 (𝑦/𝑥, 𝑢/𝑥) = 0 for an arbitrary function 𝑓 (·, ·). We
may also write the general solution as 𝑢/𝑥 = 𝑔(𝑦/𝑥) or 𝑢 = 𝑥𝑔(𝑦/𝑥).

(5.7) Example
Find the general solution of the PDE 𝑥 2𝑢𝑥 + 𝑦 2𝑢𝑦 = (𝑥 + 𝑦)𝑢.
The characteristic equations are
𝑑𝑥 𝑑𝑦 𝑑𝑢
= = .
𝑥 2 𝑦 2 (𝑥 + 𝑦)𝑢
First two equations give 𝑥 −1 − 𝑦 −1 = 𝑐 1 . To get another solution, we subtract the
first two and find that
𝑑𝑥 − 𝑑𝑦 𝑑𝑢 𝑑 (𝑥 − 𝑦) 𝑑𝑢
= ⇒ = .
𝑥 2 − 𝑦2 (𝑥 + 𝑦)𝑢 𝑥 −𝑦 𝑢
Integrating, we get (𝑥
− 𝑦)/𝑢 = 𝑐 2 . Thus, the general solution is given by
𝑓 𝑥 −1 − 𝑦 −1, (𝑥 − 𝑦)/𝑢 = 0. Since 𝑥 −1 − 𝑦 −1 is a constant and (𝑥 − 𝑦)/𝑢 is a
constant, it follows that 𝑥𝑦/𝑢 is a constant. Thus, we can also write the general
solution as 𝑔(𝑥𝑦/𝑢, (𝑥 − 𝑦)/𝑢) = 0.

(5.8) Example
Find the general solution of (𝑦 − 𝑧)𝑢𝑥 + (𝑧 − 𝑥)𝑢𝑦 + (𝑥 − 𝑦)𝑢𝑧 = 0.
The characteristic equations are
𝑑𝑥 𝑑𝑦 𝑑𝑧 𝑑𝑢
= = =
𝑦 −𝑧 𝑧 −𝑥 𝑥 −𝑦 0
For ease in integration, instead of pairs of equations, we consider the following
equivalent ones:
𝑑𝑢 = 0, 𝑑𝑥 + 𝑑𝑦 + 𝑑𝑧 = 0, 𝑥𝑑𝑥 + 𝑦𝑑𝑦 + 𝑧𝑑𝑧 = 0.
The solutions are 𝑢 = 𝑐 1, 𝑥 + 𝑦 + 𝑧 = 𝑐 2, 𝑥 2 + 𝑦 2 + 𝑧 2 = 𝑐 3 . The general solution can
be written as 𝑢 = 𝑔(𝑥 + 𝑦 + 𝑧, 𝑥 2 + 𝑦 2 + 𝑧 2 ) for an arbitrary function 𝑔(·, ·).
96 MA2020 Classnotes

(5.9) Example
Find a function 𝑢 (𝑥, 𝑦) that satisfies 𝑥𝑢𝑦 = 𝑦𝑢𝑥 and 𝑢 (0, 𝑦) = 𝑦 2 .
The characteristic equations are
𝑑𝑥 𝑑𝑦 𝑑𝑢
= = .
𝑦 −𝑥 0
Taking in pairs and integrating, we get

𝑥𝑑𝑥 + 𝑦𝑑𝑦 = 0 = 𝑑𝑢 ⇒ 𝑥 2 + 𝑦 2 = 𝑐 1, 𝑢 = 𝑐 2 .

So, the general solution of the PDE is 𝑓 (𝑥 2 + 𝑦 2, 𝑢) = 0. Since 𝑢 (0, 𝑦) = 𝑦 2 , we

have 𝑓 (02 + 𝑦 2, 𝑦 2 ) = 0. One such 𝑓 is 𝑓 (𝑣, 𝑤) = 𝑣 − 𝑤. Thus, 𝑢 = 𝑥 2 + 𝑦 2 is a
general solution satisfying the given condition.
If we write the general solution in an explicit way, it is given by 𝑐 2 = 𝑔(𝑐 1 ) or,
𝑢 = 𝑔(𝑥 2 + 𝑦 2 ) for an arbitrary function 𝑔(·). The associated condition 𝑢 (0, 𝑦) = 𝑦 2
implies that 𝑔(02 + 𝑦 2 ) = 𝑦 2 . One such 𝑔 is 𝑔(𝑢) = 𝑢. Then a general solution
satisfying the given condition is 𝑢 = 𝑥 2 + 𝑦 2 as earlier.

(5.10) Example
Find a solution of the PDE 𝑥𝑢𝑦 = 𝑦𝑢𝑥 which contains the circle 𝑢 = 1, 𝑥 2 + 𝑦 2 = 4.
From the last example, we see that the general solution of the PDE is 𝑓 (𝑥 2 +𝑦 2, 𝑢) = 0.
Since it contains the given curve, we have 𝑓 (4, 1) = 0.
One such 𝑓 is 𝑓 (𝑣, 𝑤) = 𝑣 − 4𝑤 in which case, a solution is given by 𝑥 2 + 𝑦 2 = 4𝑢.
Another 𝑓 is 𝑓 (𝑣, 𝑤) = 𝑣 − 𝑤 − 3, in which case a solution is 𝑥 2 + 𝑦 2 − 3 = 𝑢. One
more is 𝑓 (𝑣, 𝑤) = 𝑣 + 𝑤 2 − 5 in which case a solution is 𝑥 2 + 𝑦 2 + 𝑢 2 = 5. In fact,
there are infinitely many such solutions.

(5.11) Example
Find a solution of 𝑢 (𝑥 + 𝑦)𝑢𝑥 + 𝑢 (𝑥 − 𝑦)𝑢𝑦 = 𝑥 2 + 𝑦 2, where 𝑢 = 0 on the line
𝑦 = 2𝑥.
The characteristic equations are
𝑑𝑥 𝑑𝑦 𝑑𝑢
= = 2 .
𝑢 (𝑥 + 𝑦) 𝑢 (𝑥 − 𝑦) 𝑥 + 𝑦 2
The equations imply (We require two equations.)

𝑦𝑑𝑥 + 𝑥𝑑𝑦 − 𝑢𝑑𝑢 = 0, 𝑥𝑑𝑥 − 𝑦𝑑𝑦 − 𝑢𝑑𝑢 = 0.

Writing as differentials, these are

 𝑢2   𝑥 2 − 𝑦2 − 𝑢2 
𝑑 𝑥𝑦 − = 0, 𝑑 = 0.
2 2
Partial Differential Equations 97
Integrating we get 2𝑥𝑦 − 𝑢 2 = 𝑐 1 and 𝑥 2 − 𝑦 2 − 𝑢 2 = 𝑐 2 . We write the general
solution in the form 𝑐 2 = 𝑓 (𝑐 1 ), that is,

𝑥 2 − 𝑦 2 − 𝑢 2 = 𝑓 (2𝑥𝑦 − 𝑢 2 )

for an arbitrary function 𝑓 (·). Since 𝑢 (𝑥, 𝑦) also satisfies the given condition, we
substitute 𝑦 = 2𝑥 and 𝑢 = 0 simultaneously to get

𝑥 2 − 4𝑥 2 = 𝑓 (4𝑥 2 ) ⇒ 𝑓 (4𝑥 2 ) = −3𝑥 2 .

We may take 𝑓 (𝑢) = − 43 𝑢 which satisfies this condition. So, one solution is
given by 𝑥 2 − 𝑦 2 − 𝑢 2 = − 34 (2𝑥𝑦 − 𝑢 2 ) or 4(𝑥 2 − 𝑦 2 − 𝑢 2 ) + 3(2𝑥𝑦 − 𝑢 2 ) = 0 or,
7𝑢 2 = 6𝑥𝑦 + 4(𝑥 2 − 𝑦 2 ).

For nonlinear PDEs of first order, there does not exist any such general method as
Lagrange’s. However, numerical techniques exist to solve nonlinear PDEs, which
you will learn elsewhere.

5.3 Second order linear PDEs

A general second order linear PDE with two independent variables is given by

𝑎𝑢𝑥𝑥 + 𝑏𝑢𝑥𝑦 + 𝑐𝑢𝑦𝑦 + 𝑑𝑢𝑥 + 𝑒𝑢𝑦 + 𝑓 𝑢 = 𝑔 (5.3.1)

where 𝑎, . . . , 𝑔 are functions of 𝑥 and 𝑦 that do not vanish simultaneously at any

point of the domain of definition of 𝑢 (𝑥, 𝑦). We also assume that these functions
and the function 𝑢 have continuous second order partial derivatives on this domain.
Some examples are:
𝑢𝑡𝑡 = 𝑘 2𝑢𝑥𝑥 One-dimensional wave equation
𝑢𝑡 = 𝑘 2𝑢𝑥𝑥 One-dimensional heat equation
𝑢𝑥𝑥 + 𝑢𝑦𝑦 = 0 Two-dimensional Laplace equation
𝑢𝑥𝑥 + 𝑢𝑦𝑦 = 𝑓 (𝑥, 𝑦) Two-dimensional Poisson equation
𝑢𝑡𝑡 = 𝑘 2 (𝑢𝑥𝑥 + 𝑢𝑦𝑦 ) Two-dimensional wave equation
The linear second order PDE (5.3.1) is called homogeneous iff 𝑔(𝑥, 𝑦) is the
zero function; else it is called non-homogeneous. Just like ODEs, if 𝑢 1 (𝑥, 𝑦) and
𝑢 2 (𝑥, 𝑦) are two solutions of a homogeneous linear second order PDE, then their
linear combination 𝑢 = 𝑐 1𝑢 1 + 𝑐 2𝑢 2 , for constants 𝑐 1, 𝑐 2 , is also a solution of the
same homogeneous PDE. Sometimes we can use the method of ODEs to solve these
PDEs if it is so possible.
98 MA2020 Classnotes

(5.12) Example
1. Solve the PDE 𝑢𝑥𝑥 (𝑥, 𝑦) − 𝑢 (𝑥, 𝑦) = 0.
Since derivatives are taken with respect to 𝑥 only, we can use ODE methods.
Integrating with respect to 𝑥, we get 𝑢 (𝑥, 𝑦) = 𝜙 (𝑦)𝑒 𝑥 + 𝜓 (𝑦)𝑒 −𝑥 . Observe that
the constants of integration are now functions of 𝑦. The functions 𝜙 (𝑦) and
𝜓 (𝑦) are arbitrary. In a second order PDE, it is expected that there will be two
arbitrary functions.
2. Solve 𝑢𝑥𝑦 + 𝑢𝑥 = 0.
We assume that 𝑢 is a function of 𝑥 and 𝑦. Let 𝑢𝑥 = 𝑣. Then, the equation is
𝑣𝑦 = −𝑣 whose solution is 𝑣 = 𝜙 1 (𝑥)𝑒 −𝑦 . Observe that since integration is with
respect to 𝑦, the constant of integration can be a function of 𝑥, in general. Now,
𝑢𝑥 = 𝑣 = 𝜙 1 (𝑥)𝑒 −𝑦 gives
∫
−𝑦
𝑢 =𝑒 𝜙 1 (𝑥) 𝑑𝑥 + 𝜓 (𝑦).

Since 𝜙 1 (𝑥) is an arbitrary function, so is its integral, which we then write as

𝜙 (𝑥). Hence, the general solution of the PDE is 𝑢 (𝑥, 𝑦) = 𝑒 −𝑦 𝜙 (𝑥) +𝜓 (𝑦), where
𝜙 (𝑥) and 𝜓 (𝑦) are arbitrary functions of 𝑥, 𝑦, respectively.

The ODE methods suggest that we try to determine certain transformations so

that a linear second order PDE may take one of the following forms:

𝑢𝑥𝑥 = 𝜙 (𝑥, 𝑦, 𝑢, 𝑢𝑥 , 𝑢𝑦 ), 𝑢𝑥𝑦 = 𝜙 (𝑥, 𝑦, 𝑢, 𝑢𝑥 , 𝑢𝑢 ), 𝑢𝑦𝑦 = 𝜙 (𝑥, 𝑦, 𝑢, 𝑢𝑥 , 𝑢𝑦 ).

Here, 𝜙 (𝑥, 𝑦, 𝑢, 𝑢𝑥 , 𝑢𝑦 ) is an expression which is linear in 𝑢, 𝑢𝑥 and 𝑢𝑦 . That is,

𝜙 (𝑥, 𝑦, 𝑢, 𝑢𝑥 , 𝑢𝑦 ) = 𝑓1 (𝑥, 𝑦) + 𝑓2 (𝑥, 𝑦)𝑢 + 𝑓3 (𝑥, 𝑦)𝑢𝑥 + 𝑓4 (𝑥, 𝑦)𝑢𝑦 for some functions
𝑓1, 𝑓2, 𝑓3 and 𝑓4 .
However, all linear second order PDEs cannot be transformed to these two forms.
The ones which cannot be transformed to one of the above two forms can be
transformed to the forms

𝑢𝑥𝑥 + 𝑢𝑦𝑦 = 𝜙 (𝑥, 𝑦, 𝑢, 𝑢𝑥 , 𝑢𝑦 ), 𝑢𝑥𝑥 − 𝑢𝑦𝑦 = 𝜙 (𝑥, 𝑦, 𝑢, 𝑢𝑥 , 𝑢𝑦 ).

Further, we can show that any PDE in the form 𝑢𝑥𝑥 −𝑢𝑦𝑦 = 𝜙 (𝑥, 𝑦, 𝑢, 𝑢𝑥 , 𝑢𝑦 ) can also
be transformed to the form

𝑢𝑥𝑦 = 𝜙 (𝑥, 𝑦, 𝑢, 𝑢𝑥 , 𝑢𝑦 ).

These forms of linear second order PDEs are called standard forms or canonical
forms.
To find out which types of PDEs can be transformed to which form, we look
at the discriminant 𝑏 2 − 4𝑎𝑐 of the PDE (5.3.1). Depending on the sign of the
Partial Differential Equations 99
discriminant, we classify the linear PDEs. We say that the linear second order PDE
(5.3.1) is
hyperbolic iff 𝑏 2 − 4𝑎𝑐 > 0,
parabolic iff 𝑏 2 − 4𝑎𝑐 = 0, and
elliptic iff 𝑏 2 − 4𝑎𝑐 < 0.
Notice that 𝑏 2 − 4𝑎𝑐 is a function of 𝑥 and 𝑦. Its sign is required to be same
thorough out the domain of interest. It is quite possible that a linear second order
PDE is of one type in some domain and of another type in another domain.
We see that the discriminant concerns the coefficients of 𝑢𝑥𝑥 , 𝑢𝑥,𝑦 and 𝑢𝑦𝑦 only.
Let us look at the signs of the discriminants of the PDEs in standard form. They
are as follows:
Canonical form 𝑎, 𝑏, 𝑐 𝑏 2 − 4𝑎𝑐 Type
𝑢𝑥𝑦 = 𝜙 𝑎 = 0, 𝑏 = 1, 𝑐 =0 >0 Hyperbolic
𝑢𝑥𝑥 − 𝑢𝑦𝑦 = 𝜙 𝑎 = 1, 𝑏 = 0, 𝑐 = −1 >0 Hyperbolic
𝑢𝑥𝑥 = 𝜙 𝑎 = 1, 𝑏 = 0, 𝑐 =0 =0 Parabolic
𝑢𝑥𝑥 + 𝑢𝑦𝑦 = 𝜙 𝑎 = 1, 𝑏 = 0, 𝑐 =1 <0 Elliptic
As you may be surmising the type of the PDE should remain the same while
transforming one to its standard form. It means that the sign of the discriminant
will not change when we change the independent variables. We show this key fact
below.
Reduction to Standard Form: To transform (5.3.1) to its standard form, we change
the independent variables, say,

𝜉 = 𝜉 (𝑥, 𝑦), 𝜂 = 𝜂 (𝑥, 𝑦).

We assume that the functions 𝜉 and 𝜂 have continuous second order partial deriva-
tives and the Jacobian
𝐽 = 𝜉𝑥 𝜂𝑦 − 𝜉𝑦 𝜂𝑥 ≠ 0
in the concerned region. This assumption 𝐽 ≠ 0 guarantees that 𝑥 and 𝑦 can be
determined from given 𝜉 and 𝜂. To change the variables, we compute the derivatives
as follows:

𝑢 𝑥 = 𝑢 𝜉 𝜉 𝑥 + 𝑢𝜂 𝜂 𝑥
𝑢𝑦 = 𝑢𝜉 𝜉𝑦 + 𝑢𝜂 𝜂𝑦
𝑢𝑥𝑥 = 𝑢𝜉𝜉 𝜉𝑥2 + 2𝑢𝜉𝜂 𝜉𝑥 𝜂𝑥 + 𝑢𝜂𝜂 𝜂𝑥2 + 𝑢𝜉 𝜉𝑥𝑥 + 𝑢𝜂 𝜂𝑥𝑥
𝑢𝑥𝑦 = 𝑢𝜉𝜉 𝜉𝑥 𝜉𝑦 + 𝑢𝜉𝜂 (𝜉𝑥 𝜂𝑦 + 𝜉𝑦 𝜂𝑥 ) + 𝑢𝜂𝜂 𝜂𝑥 𝜂𝑦 + 𝑢𝜉 𝜉𝑥𝑦 + 𝑢𝜂 𝜂𝑥𝑦
𝑢𝑦𝑦 = 𝑢𝜉𝜉 𝜉𝑦2 + 2𝑢𝜉𝜂 𝜉𝑦 𝜂𝑦 + 𝑢𝜂𝜂 𝜂𝑦2 + 𝑢𝜉 𝜉𝑦𝑦 + 𝑢𝜂 𝜂𝑦𝑦 .

Substituting these in (5.3.1) and grouping together terms, we obtain

𝐴𝑢𝜉𝜉 + 𝐵𝑢𝜉𝜂 + 𝐶𝑢𝜂𝜂 + 𝐷𝑢𝜉 + 𝐸𝑢𝜂 + 𝐹𝑢 = 𝐺 (5.3.2)

100 MA2020 Classnotes

where 𝐴, . . . , 𝐺 are functions of 𝜉 and 𝜂 and they are given by

𝐴 = 𝑎𝜉𝑥2 + 𝑏𝜉𝑥 𝜉𝑦 + 𝑐𝜉𝑦2

𝐵 = 2𝑎𝜉𝑥 𝜂𝑥 + 𝑏 (𝜉𝑥 𝜂𝑦 + 𝜉𝑦 𝜂𝑥 ) + 𝑐𝜉𝑦 𝜂𝑦
𝐶 = 𝑎𝜂𝑥2 + 𝑏𝜂𝑥 𝜂𝑦 + 𝑐𝜂𝑦2
𝐷 = 𝑎𝜉𝑥𝑥 + 𝑏𝜉𝑥𝑦 + 𝑐𝜉𝑦𝑦 + 𝑑𝜉𝑥 + 𝑒𝜉𝑦 (5.3.3)
𝐸 = 𝑎𝜂𝑥𝑥 + 𝑏𝜂𝑥𝑦 + 𝑐𝜂𝑦𝑦 + 𝑑𝜂𝑥 + 𝑒𝜂𝑦
𝐹 = 𝑓 (𝑥 (𝜉, 𝜂), 𝑦 (𝜉, 𝜂))
𝐺 = 𝑔(𝑥 (𝜉, 𝜂), 𝑦 (𝜉, 𝜂)).

Notice that on the right side, 𝑎, . . . , 𝑒 should first be expressed in terms of 𝜉 and 𝜂
so that 𝐴, . . . , 𝐸 are also expressed in terms of 𝜉 and 𝜂. And, there is no change in 𝐹
and 𝐺; they are now expressed in terms of 𝜉 and 𝜂.
Computing the discriminant 𝐵 2 − 4𝐴𝐶 for the new equations, we find that

2
𝐵 2 − 4𝐴𝐶 = 𝜉𝑥 𝜂𝑦 − 𝜉𝑦 𝜂𝑥 (𝑏 2 − 4𝑎𝑐).

Since 𝐽 = 𝜉𝑥 𝜂𝑦 − 𝜉𝑦 𝜂𝑥 ≠ 0, the sign of the discriminant remains invariant. Hence,

the type of the PDE remains same under such a general transformation. We thus
need to choose particular 𝜉 and 𝜂 for reducing a PDE to its standard form. Our
choice will depend on the type of the problem. Observe that if 𝑎 ≠ 0, then 𝐴 and 𝐶
in (5.3.3) can be factored as follows:
p p
𝐴 = (4𝑎) −1 2𝑎𝜉𝑥 + 𝑏 + 𝑏 2 − 4𝑎𝑐 𝜉𝑦 2𝑎𝜉𝑥 + 𝑏 − 𝑏 2 − 4𝑎𝑐 𝜉𝑦



p p
𝐶 = (4𝑎) −1 2𝑎𝜂𝑥 + 𝑏 + 𝑏 2 − 4𝑎𝑐 𝜂𝑦 2𝑎𝜂𝑥 + 𝑏 − 𝑏 2 − 4𝑎𝑐 𝜂𝑦 .(5.3.4)




Hyperbolic type: Suppose the PDE (5.3.1) is hyperbolic; that is, 𝑏 2 − 4𝑎𝑐 > 0 in
the region of interest. If both 𝑎 = 0 = 𝑐, then the PDE is already in its standard
form. Else, assume that 𝑎 is nonzero. To bring the PDE to its standard form, we
put 𝐴 = 𝐶 = 0. To obtain two different solutions, we take different factors in the
factorizations of 𝐴 and 𝐶 in (5.3.4). That is, we set
p
p

2𝑎𝜉𝑥 + 𝑏 + 𝑏 2 − 4𝑎𝑐 𝜉𝑦 = 0, 2𝑎𝜂𝑥 + 𝑏 − 𝑏 2 − 4𝑎𝑐 𝜂𝑦 = 0.

Solving these first order PDEs by Lagrange’s method, we have the characteristic
equations as
𝑑𝑥 𝑑𝑦 𝑑𝑥 𝑑𝑦
= √ , = √
. (5.3.5)
2𝑎 𝑏 + 𝑏 2 − 4𝑎𝑐 2𝑎 𝑏 − 𝑏 2 − 4𝑎𝑐
If the solutions of the characteristics are respectively 𝜙 (𝑥, 𝑦) = 𝑐 1 and 𝜓 (𝑥, 𝑦) = 𝑐 2 ,
then we take the transformation as

𝜉 = 𝜙 (𝑥, 𝑦), 𝜂 = 𝜓 (𝑥, 𝑦).

Partial Differential Equations 101
As we see, this will make 𝐴 = 0 = 𝐶, 𝐵 ≠ 0 in (5.3.2-5.3.3) so that the PDE (5.3.1)
is transformed to its standard form

1

𝑢𝜉𝜂 = 𝐺 − 𝐷𝑢𝜉 − 𝐸𝑢𝜂 − 𝐹𝑢 . (5.3.6)
𝐵

This is called the first standard form of a hyperbolic PDE. Notice that by this choice
of 𝜉 and 𝜂, their Jacobian remains nonzero.
By taking new independent variables as 𝛼 = 𝜉 +𝜂 and 𝛽 = 𝜉 −𝜂, the above standard
form is again transformed to

1

𝑢𝛼𝛼 − 𝑢 𝛽𝛽 = 𝐺 − (𝐷 + 𝐸)𝑢𝛼 − (𝐷 − 𝐸)𝑢 𝛽 − 𝐹𝑢 (5.3.7)
𝐵

where 𝐵, . . . , 𝐺 are expressed in terms of 𝛼 and 𝛽. That is, we replace 𝜉 = (𝛼 + 𝛽)/2

and 𝜂 = (𝛼 − 𝛽)/2 in the earlier expressions of 𝐵, . . . , 𝐺 to express those in terms
of 𝛼 and 𝛽, and use the resulting expressions here. This standard form is called the
second standard form of a hyperbolic PDE.
If 𝑎 = 0, then 𝑐 is nonzero, and we switch the roles of 𝑥 and 𝑦. That is, we
interchange 𝑥 and 𝑦, proceed as above. Notice that the standard form will involve
𝜉 and 𝜂. Since 𝑢𝜉𝜂 = 𝑢𝜂𝜉 , interchanging 𝑥 and 𝑦 there will have no effect. But this
interchange will affect the transformations 𝜉 and 𝜂. See (5.15) below.

(5.13) Example
Reduce the PDE 𝑢𝑥𝑥 + 8𝑢𝑥𝑦 + 7𝑢𝑦𝑦 + 𝑢𝑥 + 2𝑢𝑦 + 3𝑢 + 𝑦 = 0 to its standard form.
As per the notation in (5.3.1), 𝑎 = 1, 𝑏 = 8, 𝑐 = 7, 𝑑 = 1, 𝑒 = 2, 𝑓 = 3 and 𝑔 = 𝑦 so
that the discriminant 𝑏 2 − 2
√ 4𝑎𝑐 = 8 − 28 = 36 > 0. The PDE is hyperbolic on the
whole of R2 . Now, 𝑏 ± 𝑏 2 − 4𝑎𝑐 = 8 ± 6 = 14, 2. By (5.3.5), the characteristic
equations are

𝑑𝑥 𝑑𝑦 𝑑𝑥 𝑑𝑦
= , = ⇒ 𝑑𝑦 − 7𝑑𝑥 = 0, 𝑑𝑦 − 𝑑𝑥 = 0.
2 14 2 2

Its solutions are 𝑦 − 7𝑥 = 𝑐 1 and 𝑦 − 𝑥 = 𝑐 2 . Thus, we take

𝜉 (𝑥, 𝑦) = 𝑦 − 7𝑥, 𝜂 (𝑥, 𝑦) = 𝑦 − 𝑥 .

One can proceed directly from this place to get the derivatives and substitute in the
PDE to get one in standard form. We use the formula given in (5.3.6) as in the
following.
102 MA2020 Classnotes

𝜉𝑥 = −7, 𝜉𝑦 = 1, 𝜂𝑥 = −1, 𝜂𝑦 = 1, 𝑥 = (𝜂 − 𝜉)/6, 𝑦 = (7𝜂 − 𝜉)/6.

𝐵 = 2𝑎𝜉𝑥 𝜂𝑥 + 𝑏 (𝜉𝑥 𝜂𝑦 + 𝜉𝑦 𝜂𝑥 ) + 2𝑐𝜉𝑦 𝜂𝑦


= 2(−7)(−1) + 8 (−7)(1) + (1)(−1) + 2(7)(1)(1) = −36.
𝐷 = 𝑎𝜉𝑥𝑥 + 𝑏𝜉𝑥𝑦 + 𝑐𝜉𝑦𝑦 + 𝑑𝜉𝑥 + 𝑒𝜉𝑦 = −7 + 2 = −5.
𝐸 = 𝑎𝜂𝑥𝑥 + 𝑏𝜂𝑥𝑦 + 𝑐𝜂𝑦𝑦 + 𝑑𝜂𝑥 + 𝑒𝜂𝑦 = −1 + 2 = 1.
𝐹 = 3, 𝐺 = −𝑦 = (𝜉 − 7𝜂)/6.

Then the PDE is transformed to its first standard form:

1

𝑢𝜉𝜂 = 𝐺 − 𝐷𝑢𝜉 − 𝐸𝑢𝜂 − 𝐹𝑢 .
𝐵
1  𝜉 − 7𝜂 
= − (−5)𝑢𝜉 − (1)𝑢𝜂 − 3𝑢
−36 6
1  1 
= − 5𝑢𝜉 + 𝑢𝜂 + 3𝑢 + (7𝜂 − 𝜉) .
36 6
For the second standard form, we take 𝛼 = 𝜉 +𝜂 and 𝛽 = 𝜉 −𝜂. Thus, 𝜉 = (𝛼 + 𝛽)/2
and 𝜂 = (𝛼 − 𝛽)/2. Except 𝐺, all other coefficients in (5.3.7) are constants. Now,
𝜉 − 7𝜂 1  𝛼 + 𝛽 𝛼 − 𝛽  4𝛽 − 3𝛼
𝐺= = −7 = .
6 6 2 2 6
By (5.3.7), the transformed PDE with independent variables 𝛼, 𝛽 is,
1

𝑢𝛼𝛼 − 𝑢 𝛽𝛽 = 𝐺 − (𝐷 + 𝐸)𝑢𝛼 − (𝐷 − 𝐸)𝑢 𝛽 − 𝐹𝑢
𝐵
1  4𝛽 − 3𝛼 
= − (−5 + 1)𝑢𝛼 − (−5 − 1)𝑢 𝛽 − 3𝑢
−36 6
1  4𝛽 − 3𝛼 
= + 4𝑢𝛼 + 6𝑢 𝛽 − 3𝑢 .
−36 6
(5.14) Example
Transform the PDE 𝑦 2𝑢𝑥𝑥 − 𝑥 2𝑢𝑦𝑦 = 0 for 𝑥𝑦 ≠ 0, to its standard form.
Here, 𝑎 = 𝑦 2, 𝑏 = 0, 𝑐 = −𝑥 2, 𝑑 = 2
√ 0, 𝑒 = 0, 𝑓 = 0 and 𝑔 = 0. Now, 𝑏 − 4𝑎𝑐 =
2 2
4𝑥 𝑦 > 0 since 𝑥𝑦 ≠ 0. And, 𝑏 ± 𝑏 2 − 4𝑎𝑐 = ±2𝑥𝑦. The characteristic equations
are
𝑑𝑥 𝑑𝑦 𝑑𝑥 𝑑𝑦
2
= , 2
= ⇒ 𝑦𝑑𝑦 − 𝑥𝑑𝑥 = 0, 𝑦𝑑𝑦 + 𝑥𝑑𝑥 = 0.
2𝑦 2𝑥𝑦 2𝑦 −4𝑥 2𝑦 2
Their general solutions are (𝑦 2 − 𝑥 2 )/2 = 𝑐 1 and (𝑦 2 + 𝑥 2 )/2 = 𝑐 2 , respectively. We
use the transformation
𝑦2 − 𝑥 2 𝑦2 + 𝑥 2
𝜉= , 𝜂= .
2 2
Partial Differential Equations 103
Instead of using the formula, let us compute the derivative directly. We have

𝑢𝑥 = 𝑢𝜉 𝜉𝑥 + 𝑢𝜂 𝜂𝑥 = −𝑥𝑢𝜉 + 𝑥𝑢𝜂 .
𝑢𝑦 = 𝑢𝜉 𝜉𝑦 + 𝑢𝜂 𝜂𝑦 = 𝑦𝑢𝜉 + 𝑦𝑢𝜂 .
𝑢𝑥𝑥 = 𝑢𝜉𝜉 𝜉𝑥2 + 2𝑢𝜉𝜂 𝜉𝑥 𝜂𝑥 + 𝑢𝜂𝜂 𝜂𝑥2 + 𝑢𝜉 𝜉𝑥𝑥 + 𝑢𝜂 𝜂𝑥𝑥
= 𝑥 2𝑢𝜉𝜉 − 2𝑥 2𝑢𝜉𝜂 + 𝑥 2𝑢𝜂𝜂 − 𝑢𝜉 + 𝑢𝜂 .
𝑢𝑦𝑦 = 𝑢𝜉𝜉 𝜉𝑦2 + 2𝑢𝜉𝜂 𝜉𝑦 𝜂𝑦 + 𝑢𝜂𝜂 𝜂𝑦2 + 𝑢𝜉 𝜉𝑦𝑦 + 𝑢𝜂 𝜂𝑦𝑦
= 𝑦 2𝑢𝜉𝜉 + 2𝑦 2𝑢𝜉𝜂 + 𝑦 2𝑢𝜂𝜂 + 𝑢𝜉 + 𝑢𝜂 .

Substituting these in the given PDE and simplifying we obtain the standard form:
𝜂 𝜉
𝑢𝜉𝜂 = 𝑢 𝜉 − 𝑢𝜂 .
2(𝜉 2 − 𝜂 2 ) 2(𝜉 2 − 𝜂 2 )
(5.15) Example
Reduce the PDE 4𝑢𝑥𝑦 + 𝑢𝑦𝑦 + 𝑢𝑦 = 0 to its standard form.
This is a hyperbolic PDE with the coefficient of 𝑢𝑥𝑥 as 0. We interchange the
variables 𝑥 and 𝑦 to get
𝑢𝑥𝑥 + 4𝑢𝑥𝑦 + 𝑢𝑥 = 0.
√
Here, 𝑎 = 1, 𝑏 = 4, 𝑐 = 0, 𝑑 = 1 and 𝑒 = 𝑓 = 𝑔 = 0. Now, 𝑏 ± 𝑏 2 − 4𝑎𝑐 = 4 ± 4 =
8, 0. By (5.3.5) the characteristics are
𝑑𝑥 𝑑𝑦 𝑑𝑥 𝑑𝑦
= , = ⇒ 𝑑𝑦 − 4𝑑𝑥 = 0, 𝑑𝑦 = 0.
2 8 2 0
The solutions are 𝑦 − 4𝑥 = 𝑐 1 and 𝑦 = 𝑐 2 . We take the transformations as 𝜉 = 𝑦 − 4𝑥
and 𝜂 = 𝑦. Then 𝜉𝑥 = −4, 𝜉𝑦 = 1, 𝜂𝑥 = 0 and 𝜂𝑦 = 1. By (5.3.3),

𝐵 = 2𝑎𝜉𝑥 𝜂𝑥 + 𝑏 (𝜉𝑥 𝜂𝑦 + 𝜉𝑦 𝜂𝑥 ) + 𝑐𝜉𝑦 𝜂𝑦 = 4(−4) = −16.

𝐷 = 𝑎𝜉𝑥𝑥 + 𝑏𝜉𝑥𝑦 + 𝑐𝜉𝑦𝑦 + 𝑑𝜉𝑥 + 𝑒𝜉𝑦 = −4
𝐸 = 𝑎𝜂𝑥𝑥 + 𝑏𝜂𝑥𝑦 + 𝑐𝜂𝑦𝑦 + 𝑑𝜂𝑥 + 𝑒𝜂𝑦 = 0
𝐹 = 𝑓 (𝑥 (𝜉, 𝜂), 𝑦 (𝜉, 𝜂)) = 0
𝐺 = 𝑔(𝑥 (𝜉, 𝜂), 𝑦 (𝜉, 𝜂)) = 0.

The first standard form is

𝐺 − 𝐷𝑢𝜉 − 𝐸𝑢𝜂 − 𝐹𝑢 4𝑢𝜉 1
𝑢𝜉𝜂 = = ⇒ 𝑢𝜉𝜂 + 𝑢𝜉 = 0.
𝐵 −16 4
Interchanging 𝑥 and 𝑦 retains the above standard form. But the transformations
change to 𝜉 = 𝑥 − 4𝑦 and 𝜂 = 𝑥. You can verify that if we take this transformation
directly, then the given PDE reduces to 𝑢𝜉𝜂 + 41 𝑢𝜉 = 0 as earlier.
104 MA2020 Classnotes

Parabolic type: Suppose the PDE (5.3.1) is parabolic; that is, 𝑏 2 − 4𝑎𝑐 = 0 in the
region of interest. From (5.3.4), we obtain
1
𝐴 = 𝑎𝜉𝑥2 + 𝑏𝜉𝑥 𝜉𝑦 + 𝑐𝜉𝑦2 = 4𝑎 2𝜉𝑥2 + 4𝑎𝑏𝜉𝑥 𝜉𝑦 + 4𝑎𝑐𝜉𝑦2


4𝑎
1 1
4𝑎 2𝜉𝑥2 + 4𝑎𝑏𝜉𝑥 𝜉𝑦 + 𝑏 2𝜉𝑦2 = (2𝑎𝜉𝑥 + 𝑏𝜉𝑦 ) 2 .


=
4𝑎 4𝑎
Computing similarly for 𝐶, we find that
1
𝐶= (2𝑎𝜂𝑥 + 𝑏𝜂𝑦 ) 2 .
4𝑎
𝑑𝑥 𝑑𝑦
Now, both 2𝑎𝜉𝑥 + 𝑏𝜉𝑦 = 0 and 2𝑎𝜂𝑥 + 𝑏𝜂𝑦 give the same characteristic = or,
2𝑎 𝑏
𝑏 𝑑𝑥 − 2𝑎 𝑑𝑦 = 0. (5.3.8)

It says that parabolic equations have only one characteristic curve. Suppose the
general solution of this characteristic is 𝜙 (𝑥, 𝑦) = 𝑐 1 . We choose 𝜂 = 𝜙 (𝑥, 𝑦). This
will make 𝐶 = 0. Since 𝐵 2 − 4𝐴𝐶 = 0, it will force 𝐵 = 0. The only nonzero term is
the remaining 𝑢𝜉𝜉 so that the reduced PDE will be in the standard form. Recall that
this computation assumes that the Jacobian is nonzero. Hence, after choosing 𝜂 we
choose 𝜉 in such a manner that the Jacobian

𝐽 = 𝜉𝑥 𝜂𝑦 − 𝜉𝑦 𝜂𝑥 ≠ 0.

We thus have 𝑥 = 𝑥 (𝜉, 𝜂) and 𝑦 = 𝑦 (𝜉, 𝜂) and the reduced PDE is

𝐴𝑢𝜉𝜉 = 𝐺 − 𝐷𝑢𝜉 − 𝐸𝑢𝜂 − 𝐹𝑢. (5.3.9)

Here again, 𝐴, . . . , 𝐺 in (5.3.7) are expressed in terms of 𝜉 and 𝜂 by using 𝑥 = 𝑥 (𝜉, 𝜂)

and 𝑦 = 𝑦 (𝜉, 𝜂).

(5.16) Example
Reduce the PDE 𝑢𝑥𝑥 + 4𝑢𝑥𝑦 + 4𝑢𝑦𝑦 + 𝑢𝑥 + 3𝑥 = 0 to its standard form.
Here, 𝑎 = 1, 𝑏 = 4, 𝑐 = 4, 𝑑 = 1, 𝑒 = 0, 𝑓 = 0 and 𝑔 = −3𝑥. The discriminant
𝑏 2 − 4𝑎𝑐 = 0. So, it is a parabolic PDE with 𝑎 ≠ 0 and 𝑐 ≠ 0. The characteristic
curve is, by (5.3.8),

𝑏 𝑑𝑥 − 2𝑎 𝑑𝑦 = 0 ⇒ 4 𝑑𝑥 − 2 𝑑𝑦 = 0 ⇒ 𝑦 − 2𝑥 = 𝑐 1 .

Thus, we take 𝜂 = 𝑦 − 2𝑥. Here, 𝜂𝑥 = −2 and 𝜂𝑦 = 1. We choose 𝜉 = 𝑥 so that

𝜉𝑥 = 1 and 𝜉𝑦 = 0. This makes Jacobian

𝐽 = 𝜉𝑥 𝜂𝑦 − 𝜉𝑦 𝜂𝑥 = 1 · 1 − 0 · (−2) = 1 ≠ 0.
Partial Differential Equations 105
From (5.3.3) we get

𝐴 = 𝑎𝜉𝑥2 + 𝑏𝜉𝑥 𝜉𝑦 + 𝑐𝜉𝑦2 = 1.

𝐷 = 𝑎𝜉𝑥𝑥 + 𝑏𝜉𝑥𝑦 + 𝑐𝜉𝑦𝑦 + 𝑑𝜉𝑥 + 𝐸𝜉𝑦 = 1.
𝐸 = 𝑎𝜂𝑥𝑥 + 𝑏𝜂𝑥𝑦 + 𝑐𝜂𝑦𝑦 + 𝑑𝜂𝑥 + 𝑒𝜂𝑦 = −2.
𝐹 = 𝑓 (𝑥 (𝜉, 𝜂), 𝑦 (𝜉, 𝜂)) = 0.
𝐺 = 𝑔(𝑥 (𝜉, 𝜂), 𝑦 (𝜉, 𝜂)) = −3𝜉.

By (5.3.9), the PDE is transformed to the standard form

𝐴𝑢𝜉𝜉 = 𝐺 − 𝐷𝑢𝜉 − 𝐸𝑢𝜂 − 𝐹𝑢 ⇒ 𝑢𝜉𝜉 = −3𝜉 − 𝑢𝜉 + 2𝑢𝜂 .

(5.17) Example
Reduce the PDE 𝑥 2𝑢𝑥𝑥 − 2𝑥𝑦𝑢𝑥𝑦 + 𝑦 2𝑢𝑦𝑦 + 𝑥𝑢𝑥 + 𝑦𝑢𝑦 = 0 for 𝑥 > 0 to its standard
form.
Here, 𝑎 = 𝑥 2 , 𝑏 = −2𝑥𝑦, 𝑐 = 𝑦 2 , 𝑑 = 𝑥, 𝑒 = 𝑦, 𝑓 = 𝑔 = 0 so that 𝑏 2 − 4𝑎𝑐 =
4𝑥 2𝑦 2 − 4𝑥 2𝑦 2 = 0. It is a parabolic PDE. By (5.3.8), the characteristic is

𝑏 𝑑𝑥 − 2𝑎 𝑑𝑦 = 0 ⇒ −2𝑥𝑦 𝑑𝑥 − 2𝑥 2 𝑑𝑦 = 0 ⇒ 𝑦𝑑𝑥 + 𝑥𝑑𝑦 = 0 ⇒ 𝑥𝑦 = 𝑐 1 .

Thus, 𝜂 = 𝑥𝑦. Then 𝜂𝑥 = 𝑦 and 𝜂𝑦 = 𝑥. We choose 𝜉 = 𝑥 so that 𝜉𝑥 = 1, 𝜉𝑦 = 0 and

the Jacobian 𝐽 = 𝜉𝑥 𝜂𝑦 − 𝜉𝑦 𝜂𝑥 = 𝑥 is nonzero. Also, 𝑥 = 𝜉 and 𝑦 = 𝜂/𝑥 = 𝜂/𝜉. By
(5.3.3),

𝐴 = 𝑎𝜉𝑥2 + 𝑏𝜉𝑥 𝜉𝑦 + 𝑐𝜉𝑦2 = 𝑎 = 𝑥 2 = 𝜉 2 .

𝐷 = 𝑎𝜉𝑥𝑥 + 𝑏𝜉𝑥𝑦 + 𝑐𝜉𝑦𝑦 + 𝑑𝜉𝑥 + 𝑒𝜉𝑦 = 𝑑 = 𝑥 = 𝜉.
𝐸 = 𝑎𝜂𝑥𝑥 + 𝑏𝜂𝑥𝑦 + 𝑐𝜂𝑦𝑦 + 𝑑𝜂𝑥 + 𝑒𝜂𝑦 = 𝑏 + 𝑑𝑦 + 𝑒𝑥 = −2𝑥𝑦 + 𝑥𝑦 + 𝑥𝑦 = 0.
𝐹 = 𝑓 (𝑥 (𝜉, 𝜂), 𝑦 (𝜉, 𝜂)) = 0.
𝐺 = 𝑔(𝑥 (𝜉, 𝜂), 𝑦 (𝜉, 𝜂)) = 0.

By (5.3.9), the PDE is transformed to the standard form

𝐴𝑢𝜉𝜉 = 𝐺 − 𝐷𝑢𝜉 − 𝐸𝑢𝜂 − 𝐹𝑢 ⇒ 𝜉 2𝑢𝜉𝜉 = −𝜉𝑢𝜉 ⇒ 𝑢𝜉𝜉 + 1𝜉 𝑢𝜉 = 0.

Elliptic type: Suppose that the PDE (5.3.1) is elliptic; that is, 𝑏 2 − 4𝑎𝑐 < 0 in
a region of interest. The factors of 𝐴 and 𝐶 in (5.3.4) are now complex. Thus,
elliptic PDEs have no characteristics. The standard form of an elliptic PDE have
the coefficient of 𝑢𝜉𝜂 as 0 and the coefficients of 𝑢𝜉𝜉 and 𝑢𝜂𝜂 are equal. It means, in
(5.3.2), we must have 𝐴 − 𝐶 = 𝐵 = 0. That is, using (5.3.3), we have

𝐴 − 𝐶 = 𝑎(𝜉𝑥2 − 𝜂𝑥2 ) + 𝑏 (𝜉𝑥 𝜉𝑦 − 𝜂𝑥 𝜂𝑦 ) + 𝑐 (𝜉𝑦2 − 𝜂𝑦2 ) = 0

𝐵 = 2𝑎𝜉𝑥 𝜂𝑥 + 𝑏 (𝜉𝑥 𝜂𝑦 + 𝜉𝑦 𝜂𝑥 ) + 𝑐𝜉𝑦 𝜂𝑦 = 0.
106 MA2020 Classnotes

Multiply the second with 𝑖, add to the first, and write 𝜙 = 𝜉 + 𝑖𝜂 to obtain

0 = (𝐴 − 𝐶) + 𝑖𝐵
= 𝑎(𝜉𝑥2 − 𝜂𝑥2 ) + 𝑖 2𝑎𝜉𝑥 𝜂𝑥 + 𝑏 (𝜉𝑥 𝜉𝑦 − 𝜂𝑥 𝜂𝑦 ) + 𝑖𝑏 (𝜉𝑥 𝜂𝑦 + 𝜉𝑦 𝜂𝑥 ) + 𝑐 (𝜉𝑦2 − 𝜂𝑦2 ) + 𝑖𝑐𝜉𝑦 𝜂𝑦

= 𝑎(𝜉𝑥 + 𝑖𝜂𝑥 ) 2 + 𝑏 (𝜉𝑥 + 𝑖𝜂𝑥 )(𝜉𝑦 + 𝑖𝜂𝑦 ) + 𝑐 (𝜉𝑦 + 𝑖𝜂𝑦 ) 2

= 𝑎𝜙𝑥2 + 𝑏𝜙𝑥 𝜙𝑦 + 𝑐𝜙𝑦2 .

Since 𝑎 ≠ 0, we can factor the last equation as

1  p
 p

2𝑎𝜙𝑥 + 𝑏 + 𝑖 4𝑎𝑐 − 𝑏 2 𝜙𝑦 2𝑎𝜙𝑥 + 𝑏 − 𝑖 4𝑎𝑐 − 𝑏 2 𝜙𝑦 = 0.
4𝑎
We are interested in real solutions, and each of these factors will give rise to same
pair of real solutions as their real and imaginary parts. So, we consider the first
factor: p

2𝑎𝜙𝑥 + 𝑏 + 𝑖 4𝑎𝑐 − 𝑏 2 𝜙𝑦 = 0.
Using Lagrange’s method, we set its corresponding ODE:

𝑑𝑥 𝑑𝑦
= √ .
2𝑎 𝑏 + 𝑖 4𝑎𝑐 − 𝑏 2

We rewrite it as follows and refer to it by telling the complex characteristic :

p

𝑏 + 𝑖 4𝑎𝑐 − 𝑏 2 𝑑𝑥 − 2𝑎 𝑑𝑦 = 0. (5.3.10)

Suppose 𝜙 (𝑥, 𝑦) = 𝑐 1 is the general solution of (5.3.11). Then, we use the change
of variables as 𝜉 = Re(𝜙) and 𝜂 = Im(𝜙). In this case, it can be shown that the
Jacobian is nonzero so that we will be able to uniquely determine 𝑥 = 𝑥 (𝜉, 𝜂) and
𝑦 = 𝑦 (𝜉, 𝜂). This change of variables will make 𝐴 = 𝐶 and 𝐵 = 0. Hence, the given
elliptic PDE (5.3.1) is reduced to

𝐴𝑢𝜉𝜉 + 𝐴𝑢𝜂𝜂 + 𝐷𝑢𝜉 + 𝐸𝑢𝜂 + 𝐹𝑢 = 𝐺 (5.3.11)

where the coefficients 𝐴, 𝐷, 𝐸, 𝐹, 𝐺 are as in (5.3.3) expressed in terms of 𝜉 and 𝜂.

(5.18) Example
Reduce the PDE 5𝑢𝑥𝑥 − 2𝑢𝑥𝑦 + 2𝑢𝑦𝑦 + 2𝑢𝑦 + 4𝑦 = 0 to its standard form.
As per the notation in (5.3.1), 𝑎 = 5, 𝑏 = −2, 𝑐 = 2, 𝑑 = 0, 𝑒 = 2, 𝑓 = 0 and 𝑔 = −4𝑦.
The discriminant 𝑏 2 − 4𝑎𝑐 = 4 − 40 = −36 < 0; so the PDE is elliptic on the whole
R2 . By (5.3.10), the complex characteristic is
p

𝑏 + 𝑖 4𝑎𝑐 − 𝑏 2 𝑑𝑥 − 2𝑎 𝑑𝑦 = 0 ⇒ (−2 + 6𝑖) 𝑑𝑥 − 10 𝑑𝑦 = 0.
Partial Differential Equations 107
Its general solution is (−2 + 6𝑖)𝑥 − 10𝑦 = 𝑐 1 or 𝜙 (𝑥, 𝑦) = (𝑥 + 5𝑦) − 𝑖 (3𝑥) = 𝑐 2 .
Thus, the change of variable is
𝜉 = Re(𝜙) = 𝑥 + 5𝑦, 𝜂 = Im(𝜙) = 3𝑥 .
Then, we find that 𝑥 = 𝜂/3, 𝑦 = (3𝜉 − 𝜂)/15, 𝜉𝑥 = 1, 𝜉𝑦 = 5, 𝜂𝑥 = 3 and 𝜂𝑦 = 0. By
(5.3.3), we have
𝐴 = 𝑎𝜉𝑥2 + 𝑏𝜉𝑥 𝜉𝑦 + 𝑐𝜉𝑦2 = 45.
𝐷 = 𝑎𝜉𝑥𝑥 + 𝑏𝜉𝑥𝑦 + 𝑐𝜉𝑦𝑦 + 𝑑𝜉𝑥 + 𝑒𝜉𝑦 = 10.
𝐸 = 𝑎𝜂𝑥𝑥 + 𝑏𝜂𝑥𝑦 + 𝑐𝜂𝑦𝑦 + 𝑑𝜂𝑥 + 𝑒𝜂𝑦 = 0.
𝐹 = 𝑓 (𝑥 (𝜉, 𝜂), 𝑦 (𝜉, 𝜂)) = 0.
𝐺 = 𝑔(𝑥 (𝜉, 𝜂), 𝑦 (𝜉, 𝜂)) = −4(3𝜉 − 𝜂)/15.
By (5.3.11), the PDE has the standard form 𝐴(𝑢𝜉𝜉 +𝑢𝜂𝜂 ) + 𝐷𝑢𝜉 + 𝐸𝑢𝜂 + 𝐹𝑢 = 𝐺 which
gives
2 4 4
𝑢𝜉𝜉 + 𝑢𝜂𝜂 + 𝑢𝜉 + 𝜉− 𝜂 = 0.
9 225 675
(5.19) Example
Reduce the PDE 𝑢𝑥𝑥 + 𝑥𝑢𝑦𝑦 = 0 for 𝑥 > 0, to its standard form.
Here, 𝑎 = 1, 𝑏 = 0, 𝑐 = 𝑥, 𝑑 = 𝑒 = 𝑓 = 𝑔 = 0 so that 𝑏 2 − 4𝑎𝑐 = −4𝑥 < 0 for
𝑥 > 0. Hence it is an elliptic PDE on the given region. By (5.3.10), the complex
characteristic is p

𝑏 + 𝑖 4𝑎𝑐 − 𝑏 2 𝑑𝑥 − 2𝑎 𝑑𝑦 = 0.
√
It gives 𝑖 (2 𝑥) 𝑑𝑥 − 2 𝑑𝑦 = 0 ⇒ 𝑖 34 𝑥 3/2 − 2𝑦 = 𝑐 1 or, 𝑥 3/2 + 𝑖 23 𝑦 = 𝑐 2 With
𝜙 = 𝑥 3/2 + 𝑖 23 𝑦, the transformation is given by
3
𝜉 = Re(𝜙) = 𝑥 3/2, 𝜂 = Im(𝜙) = 𝑦.
2
Then, 𝑥 = 𝜉 2/3 , 𝑦 = 23 𝜂, 𝜉𝑥 = 32 𝑥 1/2 , 𝜂𝑦 = 23 , and by (5.3.3),
9 9
𝐴 = 𝑎𝜉𝑥2 + 𝑏𝜉𝑥 𝜉𝑦 + 𝑐𝜉𝑦2 = 𝑥 = 𝜉 2/3 .
4 4
3 3
𝐷 = 𝑎𝜉𝑥𝑥 + 𝑏𝜉𝑥𝑦 + 𝑐𝜉𝑦𝑦 + 𝑑𝜉𝑥 + 𝑒𝜉𝑦 = 𝑥 −1/2 = 𝜉 −1/3 .
4 4
𝐸 = 𝑎𝜂𝑥𝑥 + 𝑏𝜂𝑥𝑦 + 𝑐𝜂𝑦𝑦 + 𝑑𝜂𝑥 + 𝑒𝜂𝑦 = 0.
𝐹 = 𝑓 (𝑥 (𝜉, 𝜂), 𝑦 (𝜉, 𝜂)) = 0.
𝐺 = 𝑔(𝑥 (𝜉, 𝜂), 𝑦 (𝜉, 𝜂)) = 0.
By (5.3.11), the PDE has the standard form 𝐴(𝑢𝜉𝜉 +𝑢𝜂𝜂 ) + 𝐷𝑢𝜉 + 𝐸𝑢𝜂 + 𝐹𝑢 = 𝐺 which
gives
9 2/3
3 1
𝜉 𝑢𝜉𝜉 + 𝑢𝜂𝜂 + 𝜉 −1/3𝑢𝜉 = 0 ⇒ 𝑢𝜉𝜉 + 𝑢𝜂𝜂 + 𝑢𝜉 = 0.
4 4 3𝜉
108 MA2020 Classnotes

Reduction of linear second order PDEs to standard forms helps in solving the PDE,
at least in hyperbolic and parabolic cases. We illustrate this idea in the following
examples.

(5.20) Example
Obtain the general solution of the PDE 3𝑢𝑥𝑥 + 10𝑢𝑥𝑦 + 3𝑢𝑦𝑦 = 0.
𝑔 = 0 so that 𝑏 2 − 4𝑎𝑐 = 64 > 0 implies that
Here, 𝑎 = 3, 𝑏 = 10, 𝑐 = 3, 𝑑 = 𝑒 = 𝑓 = √
the PDE is hyperbolic on R2 . Now, 𝑏 ± 𝑏 2 − 4𝑎𝑐 = 10 ± 8 = 18, 2. By (5.3.5), the
characteristics are given by
𝑑𝑥 𝑑𝑦 𝑑𝑥 𝑑𝑦 𝑑𝑥
= , = ⇒ 𝑑𝑦 − 3𝑥 = 0, 𝑑𝑦 − = 0.
6 18 6 2 3
Their solutions are 𝑦 − 3𝑥 = 𝑐 1 and 𝑦 − 𝑥/3 = 𝑐 2 . Thus, the transformation is
𝑥
𝜉 = 𝑦 − 3𝑥, 𝜂 =𝑦− .
3
We have 𝜉𝑥 = −3, 𝜉𝑦 = 1, 𝜂𝑥 = −1/3, 𝜂𝑦 = 1, and from (5.3.3),

𝐵 = 2𝑎𝜉𝑥 𝜂𝑥 + 𝑏 (𝜉𝑥 𝜂𝑦 + 𝜉𝑦 𝜂𝑥 ) + 𝑐𝜉𝑦 𝜂𝑦 = −73/3.

𝐷 = 𝑎𝜉𝑥𝑥 + 𝑏𝜉𝑥𝑦 + 𝑐𝜉𝑦𝑦 + 𝑑𝜉𝑥 + 𝑒𝜉𝑦 = 0.
𝐸 = 𝑎𝜂𝑥𝑥 + 𝑏𝜂𝑥𝑦 + 𝑐𝜂𝑦𝑦 + 𝑑𝜂𝑥 + 𝑒𝜂𝑦 = 0.
𝐹 = 𝑓 (𝑥 (𝜉, 𝜂), 𝑦 (𝜉, 𝜂)) = 0.
𝐺 = 𝑔(𝑥 (𝜉, 𝜂), 𝑦 (𝜉, 𝜂)) = 0.

By (5.3.9), the first standard form is

𝐺 − 𝐷𝑢𝜉 − 𝐸𝑢𝜂 − 𝐹𝑢 3
𝑢𝜉𝜂 = = − × 0 = 0.
𝐵 73
Its general solution is 𝑢 (𝜉, 𝜂) = ℎ 1 (𝜉) + ℎ 2 (𝜂). In terms of the original variables,
the general solution may be given by
 𝑥
𝑢 (𝑥, 𝑦) = ℎ 1 (𝑦 − 3𝑥) + ℎ 2 𝑦 −
3
where ℎ 1 and ℎ 2 are arbitrary functions of one argument each.

(5.21) Example
Reduce the PDE 𝑥 2𝑢𝑥𝑥 + 2𝑥𝑦𝑢𝑥𝑦 + 𝑦 2𝑢𝑦𝑦 = 0 for 𝑦 ≠ 0, to its standard form and
then find its general solution.
Here, 𝑎 = 𝑥 2 , 𝑏 = 2𝑥𝑦, 𝑐 = 𝑦 2 , 𝑑 = 𝑒 = 𝑓 = 𝑔 = 0 so that 𝑏 2 − 4𝑎𝑐 = 0. So, it is a
parabolic PDE on the whole plane. By (5.3.8), the characteristic is given by
𝑥
𝑏 𝑑𝑥 − 2𝑎 𝑑𝑦 = 0 ⇒ 2𝑥𝑦 𝑑𝑥 − 2𝑥 2 𝑑𝑦 = 0 ⇒ 𝑦 𝑑𝑥 − 𝑥 𝑑𝑦 = 0 ⇒ = 𝑐1.
𝑦
Partial Differential Equations 109
We thus take 𝜂 = 𝑥/𝑦. Now, 𝜂𝑥 = 1/𝑦 and 𝜂𝑦 = −𝑥/𝑦 2 . Choose 𝜉 = 𝑦 so that 𝜉𝑥 = 0
and 𝜉𝑦 = 1. Then the Jacobian

1
𝐽 = 𝜉𝑥 𝜂𝑦 − 𝜉𝑦 𝜂𝑥 = − ≠ 0.
𝑦

With this choice of the change of variables 𝜉 = 𝑦 and 𝜂 = 𝑥/𝑦, we have 𝑥 = 𝜉𝜂,
𝑦 = 𝜉, and

𝐴 = 𝑎𝜉𝑥2 + 𝑏𝜉𝑥 𝜉𝑦 + 𝑐𝜉𝑦2 = 𝑦 2 = 𝜉 2 .

𝐷 = 𝑎𝜉𝑥𝑥 + 𝑏𝜉𝑥𝑦 + 𝑐𝜉𝑦𝑦 + 𝑑𝜉𝑥 + 𝑒𝜉𝑦 = 0.
𝐸 = 𝑎𝜂𝑥𝑥 + 𝑏𝜂𝑥𝑦 + 𝑐𝜂𝑦𝑦 + 𝑑𝜂𝑥 + 𝑒𝜂𝑦 = 0.
𝐹 = 𝑓 (𝑥 (𝜉, 𝜂), 𝑦 (𝜉, 𝜂)) = 0.
𝐺 = 𝑔(𝑥 (𝜉, 𝜂), 𝑦 (𝜉, 𝜂)) = 0.

By (5.3.9), the PDE is transformed to the standard form

𝐴𝑢𝜉𝜉 = 𝐺 − 𝐷𝑢𝜉 − 𝐸𝑢𝜂 − 𝐹𝑢 ⇒ 𝜉 2𝑢𝜉𝜉 = 0.

The domain is 𝑦 > 0, that is, 𝜉 > 0. Hence, the reduced PDE is 𝑢𝜉𝜉 = 0. Integrating
the equation with respect to 𝜉, we have

𝑢𝜉 = ℎ 1 (𝜂) ⇒ 𝑢 (𝜉, 𝜂) = ℎ 1 (𝜂)𝜉 + ℎ 2 (𝜂),

where ℎ 1 (𝜂) and ℎ 2 (𝜂) are arbitrary functions of 𝜂. Substituting

the expressions

for
𝜉 and 𝜂 the general solution is written as 𝑢 (𝑥, 𝑦) = ℎ 1 𝑥/𝑦 𝑦 + ℎ 2 𝑥/𝑦 .
6
Separation of Variables

6.1 Modeling wave

In most engineering problems, we need to model and solve wave propagation and
heat distribution. We start with a very brief introduction to modeling wave in a
vibrating string. An elastic string is fixed at two ends, say at 𝑥 = 0 and 𝑥 = 𝐿. It is
distorted at some instant of time, say 𝑡 = 0 and is released to vibrate. The problem
is to determine its deflection 𝑢 (𝑥, 𝑡) at any point 𝑥 ∈ [0, 𝐿] and time 𝑡 > 0.

For a simple model we assume the following:

1. The string is perfectly elastic; it does not resist to bend.
2. It is homogeneous, i.e., mass of the string per unit length is constant, denote
it by 𝜌.
3. The string has been fastened by stretching it and the tension due to the
stretching is so high that the action of gravitation on it is negligible.
4. Every particle of the string moves strictly vertically so that the deflection and
the slope st every point on it remains small in absolute value.
We consider the forces acting on a small portion Δ𝑥 of the string. Due to the
above assumptions, the tension on the string is tangential to the initial shape (we
distorted it) of the string at each point. Let 𝑇1 and 𝑇2 be the tension at the points 𝑃
(point 𝑥) and 𝑄 (point 𝑥 + Δ𝑥) of that portion. There is no horizontal motion, i.e.,
the horizontal components of tension is constant. See the figure. It means

𝑇1 cos 𝛼 = 𝑇2 cos 𝛽 = 𝑇 = constant.

The vertical component at 𝑃 is downward and at 𝑄 is upward; so they are −𝑇1 sin 𝛼
and 𝑇2 sin 𝛽. By Newton’s second law, the resultant of these forces is equal to the

110
Separation of Variables 111
mass 𝜌Δ𝑥 times the acceleration 𝑢𝑡𝑡 evaluated at some point 𝑥 = 𝑥 ∗ between 𝑥 and
𝑥 + Δ𝑥. Hence,
𝑇2 sin 𝛽 − 𝑇1 sin 𝛼 = 𝜌 Δ𝑥 𝑢𝑡𝑡 (𝑥 ∗, 𝑡).
Dividing by 𝑇 and using the previous equation, we get
𝑇2 sin 𝛽 𝑇1 sin 𝛼 𝜌Δ𝑥 𝜌Δ𝑥
− = 𝑢𝑡𝑡 (𝑥 ∗, 𝑡) ⇒ 𝑢𝑡𝑡 (𝑥 ∗, 𝑡) = tan 𝛽 − tan 𝛼 .
𝑇2 cos 𝛽 𝑇1 cos 𝛼 𝑇 𝑇
However, tan 𝛼 is the slope of the (distorted) string at the point 𝑥. Similarly, tan 𝛽
is the slope at the point 𝑥 + Δ𝑥. That is,
tan 𝛼 = 𝑢𝑥 (𝑥, 𝑡), tan 𝛽 = 𝑢𝑥 (𝑥 + Δ𝑥, 𝑡).
Hence,
𝑇 𝑢𝑥 (𝑥 + Δ𝑥) − 𝑢𝑥 (𝑥)
𝑢𝑡𝑡 (𝑥 ∗, 𝑡) = .
𝜌 Δ𝑥
Write 𝑇 /𝜌 = 𝑐 2 since it is positive. Take limit of both sides as Δ𝑥 → 0. Then,
𝑥 + Δ𝑥 → 𝑥 and 𝑥 ∗ → 𝑥 so that we obtain
𝑢𝑡𝑡 = 𝑐 2𝑢𝑥𝑥 where 𝑐 > 0. (6.1.1)
This is called the one-dimensional wave equation. It is a linear homogeneous
second order PDE.

6.2 D’ Alembert’s solution of wave equation

We consider solving the wave equation in (6.1.1):
𝑢𝑡𝑡 − 𝑐 2𝑢𝑥𝑥 = 0.
Notice that 𝑢 = 𝑢 (𝑥, 𝑡), a function of 𝑥 and 𝑡. As a linear second order PDE,
comparing it with (5.3.1) with 𝑦 there as 𝑡 here, we find that 𝑎 = −𝑐 2 , 𝑏 = 0,
𝑐 (𝑥, 𝑡) = 1, 𝑑 = 𝑒 = 𝑓 = 𝑔 = 0. The discriminant is 𝑏 2 − 4𝑎𝑐 = 4𝑐 2 > 0. So, it is a
hyperbolic PDE. By (5.3.5), the characteristics are
𝑑𝑥 𝑑𝑡 𝑑𝑥 𝑑𝑡
= √ ⇒ 2
= ⇒ 𝑥 − 𝑐𝑡 = 𝑐 1 .
2𝑎 𝑏 + 𝑏 2 − 4𝑎𝑐 −𝑐 −2𝑐
𝑑𝑥 𝑑𝑡 𝑑𝑥 𝑑𝑡
= √ ⇒ 2
= ⇒ 𝑥 + 𝑐𝑡 = 𝑐 2 .
2𝑎 −2𝑐 2𝑐


𝑏 − 𝑏 2 − 4𝑎𝑐
Thus, the transformation is given by
𝜉 = 𝑥 + 𝑐𝑡, 𝜂 = 𝑥 − 𝑐𝑡 .
112 MA2020 Classnotes

We find that 𝑥 = (𝜉 + 𝜂)/2, 𝑡 = (𝜉 − 𝜂)/(2𝑐), 𝜉𝑥 = 1, 𝜉𝑡 = 𝑐, 𝜂𝑥 = 1 and 𝜂𝑡 = −𝑐. By

(5.3.3) with the variable 𝑦 as 𝑡, the new coefficients are given by

𝐵 = 2𝑎𝜉𝑥 𝜂𝑥 + 𝑏 (𝜉𝑥 𝜂𝑡 + 𝜉𝑡 𝜂𝑥 ) + 𝑐 (𝑥, 1)𝜉𝑡 𝜂𝑡 = −3𝑐 2

𝐷 = 𝑎𝜉𝑥𝑥 + 𝑏𝜉𝑥𝑡 + 𝑐𝜉𝑡𝑡 + 𝑑𝜉𝑥 + 𝑒𝜉𝑡 = 0
𝐸 = 𝑎𝜂𝑥𝑥 + 𝑏𝜂𝑥𝑡 + 𝑐𝜂𝑡𝑡 + 𝑑𝜂𝑥 + 𝑒𝜂𝑡 = 0
𝐹 = 𝑓 (𝑥 (𝜉, 𝜂), 𝑡 (𝜉, 𝜂)) = 0
𝐺 = 𝑔(𝑥 (𝜉, 𝜂), 𝑡 (𝜉, 𝜂)) = 0.

By (5.3.6), the PDE is reduced to its standard form

1
𝑢𝜉𝜂 = (𝐺 − 𝐷𝑢𝜉 − 𝐸𝑢𝜂 − 𝐹𝑢) = 0.
𝐵
You can also directly compute 𝑢𝑡𝑡 and 𝑢𝑥𝑥 using the Chain rule and substitute to get
the same equation 𝑢𝜉𝜂 = 0.
Integrating the above equation with respect to 𝜂, we get

𝑢𝜉 = 𝑓1 (𝜉)

for an arbitrary function 𝑓1 of 𝜉. Integrating this equation with respect to 𝜉, we get

∫
𝑢 (𝜉, 𝜂) = 𝑓1 (𝜉) 𝑑𝜉 + 𝑓2 (𝜂).
∫
Since 𝑓1 (𝜉) ia an arbitrary function, we may write 𝑓1 (𝜉) 𝑑𝜉 as another arbitrary
function, say 𝑓3 (𝜉). Hence, the general solution of the above equation is 𝑢 (𝜉, 𝜂) =
𝑓3 (𝜉) + 𝑓2 (𝜂). Going back to the variables 𝑥 and 𝑡, we obtain the general solution of
the wave equation (6.1.1) as

𝑢 (𝑥, 𝑡) = 𝜙 (𝑥 + 𝑐𝑡) + 𝜓 (𝑥 − 𝑐𝑡) (6.2.1)

where 𝜙 and 𝜓 are arbitrary functions of 𝑥 and 𝑡. This solution is known as the
D’ Alembert’s solution of the wave equation.
Suppose the initial distortion of the string is given as a function of 𝑥, say, 𝑓 (𝑥),
and the initial velocity, when we leave the string to vibrate is given by a function of
𝑥, say, 𝑔(𝑥). In our notation, the wave equation (6.1.1) now comes with two initial
conditions
𝑢 (𝑥, 0) = 𝑓 (𝑥), 𝑢𝑡 (𝑥, 0) = 𝑔(𝑥). (6.2.2)
To get a solution of the initial value problem (6.1.1) and (6.2.2), we start with
D’ Alembert’s solution and try to determine the arbitrary functions 𝜙 and 𝜓 . From
(6.2.1), we get
𝜕(𝑥 + 𝑐𝑡) 𝜕(𝑥 − 𝑐𝑡)
𝑢𝑡 (𝑥, 𝑡) = 𝜙 0 (𝑥 + 𝑐𝑡) − 𝜓 0 (𝑥 − 𝑐𝑡) = 𝑐𝜙 0 (𝑥 + 𝑐𝑡) − 𝑐𝜙 0 (𝑥 − 𝑐𝑡).
𝜕𝑡 𝜕𝑡
Separation of Variables 113
The initial condition imply that

𝑢 (𝑥, 0) = 𝜙 (𝑥) + 𝜓 (𝑥) = 𝑓 (𝑥), 𝑢𝑡 (𝑥, 0) = 𝑐𝜙 0 (𝑥) − 𝑐𝜓 0 (𝑥) = 𝑔(𝑥).

Taking the definite integral of the second equation with respect to 𝑥 varying from
any fixed 𝑥 0 to any 𝑦 in the range of values of the variable 𝑥, we obtain
∫𝑦
   
𝑐 𝜙 (𝑦) − 𝜓 (𝑦) − 𝑐 𝜙 (𝑥 0 ) − 𝜓 (𝑥 0 ) = 𝑔(𝑠) 𝑑𝑠.
𝑥0

So, we have now 𝜙 (𝑦) + 𝜓 (𝑦) = 𝑓 (𝑦) and 𝜙 (𝑦) − 𝜓 (𝑦) from the above. Then,
∫
1 1 𝑦 1 
𝜙 (𝑦) = 𝑓 (𝑦) + 𝑔(𝑠) 𝑑𝑠 + 𝜙 (𝑥 0 ) − 𝜓 (𝑥 0 ) .
2 2𝑐 𝑥 0 2
∫𝑦
1 1 1 
𝜓 (𝑦) = 𝑓 (𝑦) − 𝑔(𝑠) 𝑑𝑠 − 𝜙 (𝑥 0 ) − 𝜓 (𝑥 0 ) .
2 2𝑐 𝑥 0 2

Replacing 𝑦 by 𝑥 + 𝑐𝑡 in the first and 𝑥 − 𝑐𝑡 in the second, we obtain

𝑢 (𝑥, 𝑡) = 𝜙 (𝑥 + 𝑐𝑡) + 𝜓 (𝑥 − 𝑐𝑡)

∫ ∫
1  1 𝑥+𝑐𝑡 1 𝑥−𝑐𝑡
= 𝑓 (𝑥 + 𝑐𝑡) + 𝑓 (𝑥 − 𝑐𝑡) + 𝑔(𝑠) 𝑑𝑠 − 𝑔(𝑠) 𝑑𝑠
2 2𝑐 𝑥 0 2𝑐 𝑥 0
∫ ∫
1  1 𝑥+𝑐𝑡 1 𝑥0
= 𝑓 (𝑥 + 𝑐𝑡) + 𝑓 (𝑥 − 𝑐𝑡) + 𝑔(𝑠) 𝑑𝑠 + 𝑔(𝑠) 𝑑𝑠
2 2𝑐 𝑥 0 2𝑐 𝑥−𝑐𝑡
∫
1  1 𝑥+𝑐𝑡
= 𝑓 (𝑥 + 𝑐𝑡) + 𝑓 (𝑥 − 𝑐𝑡) + 𝑔(𝑠) 𝑑𝑠.
2 2𝑐 𝑥−𝑐𝑡

We observe that two initial conditions as given in (6.2.2) determine the solution of
the wave equation (6.1.1) uniquely.
In particular, when the initial velocity is 0, the function 𝑔(𝑥) is the zero function.
We see that the solution is 𝑢 (𝑥, 𝑡) = [𝑓 (𝑥 + 𝑐𝑡) + 𝑓 (𝑥 − 𝑐𝑡)]/2.

6.3 Series solution of the wave equation

Physically, the string has two fixed end-points, which we have not considered while
discussing D’ Alembert’s solution. The end-points are fixed at 𝑥 = 0 and 𝑥 = 𝐿; it
means that the deflection is 0 for all time to come. This translates to the boundary
conditions
𝑢 (0, 𝑡) = 0, 𝑢 (𝐿, 𝑡) = 0, for 𝑡 ≥ 0. (6.3.1)
114 MA2020 Classnotes

We still have the same initial conditions that initial deflection is 𝑓 (𝑥) and initial
velocity is 𝑔(𝑥), but now, it is valid only for 0 ≤ 𝑥 ≤ 𝐿. That is,

𝑢 (𝑥, 0) = 𝑓 (𝑥), 𝑢𝑡 (𝑥, 0) = 𝑔(𝑥), for 0 ≤ 𝑥 ≤ 𝐿. (6.3.2)

Notice that D’ Alembert’s solution is Uniquely determined when the wave equa-
tion is given only the initial conditions. So, that may not satisfy the boundary
conditions (6.3.1). Indeed, D’ Alembert’s solution now involves 𝑓 (−𝑐𝑡) which does
not mean anything physically. This solution is valid for all 𝑥 and not only for
0 ≤ 𝑥 ≤ 𝐿. Potentially, this solution applies to a string that is elongated from −∞
to ∞. Thus, it leaves open the case that when 𝑥 is restricted to the interval [0, 𝐿],
there may or may not exist solutions which will also satisfy the initial conditions.
We will describe the simple and powerful method of separating the variables for
obtaining such a solution. In this method, we use the heuristic that possibly there is
a solution of the wave equation in the form
∞
Õ
𝑢 (𝑥, 𝑡) = 𝐹𝑛 (𝑥) 𝐺𝑛 (𝑡)
𝑛=1

which also satisfies the initial conditions and the boundary conditions. However, we
do not directly plug it in the wave equation so as to satisfy the initial and boundary
conditions. We rather think of 𝑢𝑛 (𝑥, 𝑡) = 𝐹𝑛 (𝑥) 𝐺𝑛 (𝑡) to satisfy the wave equation
and the boundary conditions only. The series would then be required when we try
to satisfy the initial conditions.
So, we start with 𝑢 (𝑥, 𝑡) = 𝐹 (𝑥) 𝐺 (𝑡) initially. We plug it in the wave equation to
obtain two ODEs, one for 𝐹 (𝑥) and the other for 𝐺 (𝑡). This constitutes Step 1 of
the method. In Step 2, we determine (nonzero) solutions of these ODEs that satisfy
the boundary conditions in (6.3.1) thereby obtaining possible 𝑢𝑛 (𝑥, 𝑡). In Step 3, we
Í
use a series 𝑎𝑛𝑢𝑛 (𝑥, 𝑡) to compose the solutions found in Step 2 so that the series
solution satisfies the initial conditions. We execute the plan as in the following.
Step 1: Suppose 𝑢 (𝑥, 𝑡) = 𝐹 (𝑥) 𝐺 (𝑡). Differentiating, we get
••
𝑢𝑡𝑡 = 𝐹 𝐺, 𝑢𝑥𝑥 = 𝐹 00𝐺 .

Here, the dot denotes derivative with respect to 𝑡 and prime denotes derivative with
respect to 𝑥. Then the wave equation 𝑢𝑡𝑡 = 𝑐 2𝑢𝑥𝑥 in (6.1.1) takes the form
••
••
2 00 𝐺 𝐹 00
𝐹𝐺 = 𝑐 𝐹 𝐺 ⇒ 2 = .
𝑐 𝐺 𝐹
The left side is independent of 𝑥 and the right side is independent of 𝑡. So, both are
independent of 𝑥 and 𝑡, that is, it is a constant, say, 𝑘. Of course, the constant 𝑘 is
yet unknown. We then have
••
𝐹 00 − 𝑘𝐹 = 0, 𝐺 − 𝑐 2𝑘𝐺 = 0. (6.3.3)
Separation of Variables 115
Step 2: We are interested in nonzero solutions. The boundary conditions in (6.3.1)
take the form

𝑢 (0, 𝑡) = 𝐹 (0)𝐺 (𝑡) = 0 ⇒ 𝐹 (0) = 0, 𝑢 (𝐿, 𝑡) = 𝐹 (𝐿)𝐺 (𝑡) = 0 ⇒ 𝐹 (𝐿) = 0.

These are conditions on 𝐹 (𝑥) only. Again, 𝐹 (𝑥) = 0 satisfies these conditions but
we require nonzero solutions. The ODE 𝐹 00 − 𝑘𝐹 = 0 for 𝐹 in (6.3.3) involves an
unknown constant 𝑘.
If 𝑘 = 0, then 𝐹 = 𝑎𝑥 + 𝑏 for constants 𝑎 and 𝑏. Now, 𝐹 (0) = 0 ⇒ 𝑏 = 0. So,
𝐹 (𝑥) = 𝑎𝑥. And, 𝐹 (𝐿) = 0 ⇒ 𝑎𝐿 = 0 ⇒ 𝑎 = 0. So, 𝐹 (𝑥) is the zero function,
which we do not require. √ √
If 𝑘 > 0, then 𝐹 (𝑥) = 𝑎𝑒 𝑘𝑥 + 𝑏𝑒 − 𝑘𝑥 . The conditions 𝐹 (0) = 0 = 𝐹 (𝐿) imply that
√ √
𝑎 + 𝑏 = 0, 𝑎𝑒 𝑘𝐿
+ 𝑏𝑒 − 𝑘𝐿
= 0 ⇒ 𝑎 = 0 = 𝑏.

So, 𝐹 (𝑥) is the zero function, which we do not require.

So, 𝑘 < 0; and we write 𝑘 = −𝑝 2 for 𝑝 > 0. Notice that 𝑝 is yet to be determined.
Now, the equation of 𝐹 (𝑥) is 𝐹 00 + 𝑝 2 𝐹 = 0. Its general solution is

𝐹 (𝑥) = 𝑎 cos(𝑝𝑥) + 𝑏 sin(𝑝𝑥).

Now, 𝐹 (0) = 0 ⇒ 𝑎 = 0. So, 𝐹 (𝑥) = 𝑏 sin(𝑝𝑥). Then, 𝐹 (𝐿) = 0 ⇒ 𝑏 sin(𝑝𝐿) = 0.

By taking 𝑏 = 0, we get only trivial solution. So, we take the other alternative
sin(𝑝𝐿) = 0. It gives
𝑛𝜋
𝑝𝐿 = 𝑛𝜋 ⇒ 𝑝 = for 𝑛 = 1, 2, 3, . . . .
𝐿
Corresponding to each value of 𝑝, we obtain a solution. These are:
 𝑛𝜋𝑥 
𝐹𝑛 (𝑥) = 𝑏𝑛 sin for 𝑛 = 1, 2, 3, . . . . (6.3.4)
𝐿
Now that the possible values for 𝑝 has been obtained, we use these values to solve the

2
equation for 𝐺 (𝑡) in (6.3.3). Notice that 𝑘 = −𝑝 2 = − 𝑛𝜋/𝐿 ⇒ 𝑐 2𝑘 = − 𝑐𝑛𝜋/𝐿) 2 .
••
The equation 𝐺 − 𝑐 2𝑘𝐺 = 0 for 𝐺 (𝑡) now reads as
•• 𝑐𝑛𝜋
𝐺 + 𝜆𝑛2𝐺 = 0, 𝜆𝑛 = for 𝑛 = 1, 2, 3, . . . .
𝐿
Its general solution is

𝐺𝑛 (𝑡) = 𝑐𝑛 cos(𝜆𝑛 𝑡) + 𝑑𝑛 sin(𝜆𝑛 𝑡) for 𝑛 = 1, 2, 3, . . . .

 
Then 𝑢𝑛 (𝑥, 𝑡) = 𝐹𝑛𝐺𝑛 = 𝑏𝑛 sin(𝑛𝜋𝑥/𝐿) 𝑐𝑛 cos(𝜆𝑛 𝑡) + 𝑑𝑛 sin(𝜆𝑛 𝑡) . However, we do
not expect any of these 𝑢𝑛 s to satisfy the initial conditions, in general. So, we will
Í∞
be taking a series 𝑢 (𝑥, 𝑡) = 𝑛=1 𝑎𝑛𝑢𝑛 (𝑥, 𝑡) and try to satisfy the initial conditions.
116 MA2020 Classnotes

In that case, notice that we do not require so many constants like 𝑎𝑛 , 𝑏𝑛 , 𝑐𝑛 and 𝑑𝑛 .
Í∞
Only 𝑐𝑛 and 𝑑𝑛 will suffice. It is enough to consider 𝑢 (𝑥, 𝑡) = 𝑛=1 𝑢𝑛 (𝑥, 𝑡), where
   𝑛𝜋𝑥 
𝑢𝑛 (𝑥, 𝑡) = 𝑐𝑛 cos(𝜆𝑛 𝑡) + 𝑑𝑛 sin(𝜆𝑛 𝑡) sin for 𝑛 = 1, 2, 3, . . . . (6.3.5)
𝐿
The numbers 𝜆𝑛 = 𝑐𝑛𝜋/𝐿 are called the eigenvalues and the corresponding func-
tions 𝑢𝑛 (𝑥, 𝑡) above are called the eigenfunctions of the vibrating string. The set
{𝜆1, 𝜆2, . . .} of eigenvalues is called the spectrum.
Observe that each 𝑢𝑛 represents a harmonic motion with frequency as 𝜆𝑛 /(2𝜋)
cycles per unit time. This motion is called the normal mode of the string. The first
mode, corresponding to 𝑛 = 1, is called the fundamental mode and the others are
called the overtones. Since sin(𝑛𝜋𝑥/𝐿) = 0 for 𝑥 = 𝐿/𝑛, 2𝐿/𝑛, . . . , (𝑛 − 1)𝐿/𝑛,
the 𝑛th normal mode has 𝑛 − 1 nodes. Like the end-points, the string does not move
at the nodes. This is expected due to the wave-like movement of the string, from
which the name for the equation in (6.1.1) comes.
Step 3: We have seen that the eigenfunctions in (6.3.5) satisfy the wave equation
and the boundary conditions. We do not expect a single 𝑢𝑛 (𝑥, 𝑡) to satisfy the initial
conditions. As discussed earlier, we set
∞
Õ Õ∞  𝑛𝜋𝑥 
 
𝑢 (𝑥, 𝑡) = 𝑢𝑛 (𝑥, 𝑡) = 𝑐𝑛 cos(𝜆𝑛 𝑡) + 𝑑𝑛 sin(𝜆𝑛 𝑡) sin . (6.3.6)
𝑛=1 𝑛=1
𝐿

With this 𝑢 (𝑥, 𝑡), the first initial condition in (6.3.2) gives
∞
Õ  𝑛𝜋𝑥 
𝑢 (𝑥, 0) = 𝑐𝑛 sin = 𝑓 (𝑥) for 0 ≤ 𝑥 ≤ 𝐿.
𝑛=1
𝐿

It says that 𝑓 (𝑥) has been expanded in its Fourier sine series. Thus,
∫
2 𝐿  𝑛𝜋𝑥 
𝑐𝑛 = 𝑓 (𝑥) sin 𝑑𝑥 for 𝑛 = 1, 2, 3, . . . . (6.3.7)
𝐿 0 𝐿
For the second initial condition, we first differentiate 𝑢 (𝑥, 𝑡) in (6.3.6), evaluate it at
𝑡 = 0 to get
"∞ #
Õ   𝑛𝜋𝑥 
𝑢𝑡 (𝑥, 0) = − 𝑐𝑛 𝜆𝑛 sin(𝜆𝑛 𝑡) + 𝑑𝑛 𝜆𝑛 cos(𝜆𝑛 𝑡) sin
𝑛=1
𝐿
𝑡=0
∞
Õ  𝑛𝜋𝑥 
= 𝑑𝑛 𝜆𝑛 sin = 𝑔(𝑥).
𝑛=1
𝐿

Hence, 𝑔(𝑥) is expanded in its Fourier sine series. Thus,

∫
2 𝐿  𝑛𝜋𝑥 
𝑑𝑛 𝜆𝑛 = 𝑔(𝑥) sin 𝑑𝑥 .
𝐿 0 𝐿
Separation of Variables 117
Putting back the value of 𝜆𝑛 = 𝑐𝑛𝜋/𝐿 we get
∫𝐿
2  𝑛𝜋𝑥 
𝑑𝑛 = 𝑔(𝑥) sin 𝑑𝑥 for 𝑛 = 1, 2, 3, . . . . (6.3.8)
𝑐𝑛𝜋 0 𝐿
To summarize, the solution of the wave equation (6.1.1) with boundary conditions
in (6.3.1) and initial conditions in (6.3.2) is given by (6.3.6) with 𝜆𝑛 = 𝑐𝑛𝜋/𝐿, where
𝑐𝑛 and 𝑑𝑛 are as in (6.3.7-6.3.8).
It can be shown that the series in (6.3.6) is convergent for 0 ≤ 𝑥 ≤ 𝐿 and all
𝑡 ≥ 0. Further, the solution 𝑢 (𝑥, 𝑡) in the above series form is a solution of the wave
equation with the initial and boundary conditions if 𝑓 (𝑥) is twice differentiable on
0 < 𝑥 < 𝐿, and it has one-sided second derivatives at the end-points 𝑥 = 0 and
𝑥 = 𝐿, which are equal to 0.

(6.1) Example
Find the solution of the wave equation 𝑢𝑡𝑡 = 𝑐 2𝑢𝑥𝑥 satisfying 𝑢 (0, 𝑡) = 0 = 𝑢 (𝐿, 𝑡),
𝑢𝑡 (𝑥, 0) = 0 and 𝑢 (𝑥, 0) = 2𝑘𝑥/𝐿 for 0 ≤ 𝑥 ≤ 𝐿/2, 𝑢 (𝑥, 0) = 2𝑘 (𝐿 − 𝑥)/𝐿 for
𝐿/2 < 𝑥 ≤ 𝐿.
According to (6.3.6), the solution is given by
Õ∞  𝑛𝜋𝑥 
 
𝑢 (𝑥, 𝑡) = 𝑐𝑛 cos(𝜆𝑛 𝑡) + 𝑑𝑛 sin(𝜆𝑛 𝑡) sin
𝑛=1
𝐿

where 𝜆𝑛 = 𝑐𝑛𝜋/𝐿 and by (6.3.7-6.3.8),

∫
2 𝐿  𝑛𝜋𝑥 
𝑐𝑛 = 𝑓 (𝑥) sin 𝑑𝑥
𝐿 0 𝐿
∫ ∫
2 𝐿/2 2𝑘  𝑛𝜋𝑥  2 𝐿 2𝑘  𝑛𝜋𝑥 
= 𝑥 sin 𝑑𝑥 + (𝐿 − 𝑥)𝑠𝑖𝑛 𝑑𝑥
𝐿 0 𝐿 𝐿 𝐿 0 𝐿 𝐿
8𝑘  𝑛𝜋 
= 2 2 sin
𝑛 𝜋 2
∫𝐿
2  𝑛𝜋𝑥 
𝑑𝑛 = 𝑔(𝑥) sin 𝑑𝑥 = 0.
𝑐𝑛𝜋 0 𝐿
Since sin(𝑛𝜋/2) is 0 for even 𝑛, 1 for 𝑛 = 4𝑚 + 1, and −1 for 𝑛 = 4𝑚 + 3, we find
that
8𝑘 h 1 𝜋𝑥 𝜋𝑐𝑡 1 3𝜋𝑥 3𝜋𝑐𝑡 i
𝑢 (𝑥, 𝑡) = 2 2 sin cos − 2 sin cos +··· .
𝜋 1 𝐿 𝐿 3 𝐿 𝐿
(6.2) Example
Suppose the vibration of a stretched string of length 1 unit is clamped at each end
and starts from rest with the initial shape 𝑢 (𝑥, 0) = 𝑘𝑥 (1 − 𝑥). Here, 𝑘 > 0 is such
that the maximum transverse displacement is small. Find the vibration 𝑢 (𝑥, 𝑡).
118 MA2020 Classnotes

The function 𝑢 (𝑥, 𝑡) satisfies 𝑢𝑡𝑡 = 𝑐 2𝑢𝑥𝑥 for some constant 𝑐 depending on the
material of the string, the boundary conditions are 𝑢 (0, 𝑡) = 0 = 𝑢 (1, 𝑡), and
the initial conditions are 𝑢 (𝑥, 0) = 𝑘𝑥 (1 − 𝑥) and 𝑢𝑡 (𝑥, 0) = 0. Here, 𝐿 = 1,
𝑓 (𝑥) = 𝑘𝑥 (1 − 𝑥) and 𝑔(𝑥) = 0. By (6.3.6),
Õ∞  𝑛𝜋𝑥 
 
𝑢 (𝑥, 𝑡) = 𝑐𝑛 cos(𝜆𝑛 𝑡) + 𝑑𝑛 sin(𝜆𝑛 𝑡) sin
𝑛=1
𝐿

where 𝜆𝑛 = 𝑐𝑛𝜋/𝐿 = 𝑐𝑛𝜋, and

∫𝐿 ∫𝐿
2  𝑛𝜋𝑥  2  𝑛𝜋𝑥 
𝑐𝑛 = 𝑓 (𝑥) sin 𝑑𝑥, 𝑑𝑛 = 𝑔(𝑥) sin 𝑑𝑥 .
𝐿 0 𝐿 𝑐𝑛𝜋 0 𝐿

Since 𝑔(𝑥) = 0, we have 𝑑𝑛 = 0. And,

∫1 (
2𝑘
0 if 𝑛 even
𝑐𝑛 = 2 𝑘𝑥 (1 −𝑥) sin(𝑛𝜋𝑥) 𝑑𝑥 = 2 3 3 1 − cos(𝑛𝜋) =
0 𝑛 𝜋 8𝑘/(𝑛 3 𝜋 3 ) if 𝑛 odd.

Since only odd terms remain, we write 𝑛 = 2𝑚 + 1 for 𝑚 = 0, 1, 2, 3, . . .. Then

∞
Õ
𝑢 (𝑥, 𝑡) = 8𝑘𝜋 −3 (2𝑚 + 1) −3 sin (2𝑚 + 1)𝜋𝑥 cos (2𝑚 + 1)𝑐𝜋𝑡 .



𝑚=0

In this section we have discussed how to use the method of separation of variables
for solving the wave equation. The same method can be used to solve first order
PDEs. You can work out the details by solving the exercises.

6.4 One-dimensional heat flow

Consider the temperature in a long thin metal wire of constant cross sectional area.
Assume that it is perfectly insulated so that heat flows in one direction only. Call
the direction of flow as the 𝑥-axis. Write the temperature as 𝑢 (𝑥, 𝑡), where 𝑡 is time.
Write 𝐾 for the thermal conductivity, 𝑐 for the thermal diffusivity, 𝜎 for specific
heat, and 𝜌 for the density of the wire. Then 𝑐 2 = 𝐾/𝜌𝜎 and the heat flow is
governed by the heat equation
𝑢𝑡 = 𝑐 2𝑢𝑥𝑥 . (6.4.1)
Suppose that the wire is of length 𝐿 and its ends prescribed by 𝑥 = 0 and 𝑥 = 𝐿 are
kept at zero temperature. This gives the boundary conditions

𝑢 (0, 𝑡) = 0, 𝑢 (𝐿, 0) = 0 for 𝑡 ≥ 0. (6.4.2)

Separation of Variables 119
In particular, 𝑢 (0, 0) = 𝑢 (𝐿, 0) = 0. Further, assume that the initial temperature on
the wire at time 𝑡 = 0 is given as a function of 𝑥; say, 𝑓 (𝑥). Then

𝑢 (𝑥, 0) = 𝑓 (𝑥) for 0 ≤ 𝑥 ≤ 𝐿. (6.4.3)

Notice that due to the boundary conditions, the function 𝑓 (𝑥) cannot be arbitrary,
but it must satisfy 𝑓 (0) = 𝑓 (𝐿) = 0.
We will use the method of separation of variables to get a series solution of (6.4.1)
satisfying (6.4.2) and (6.4.3).
Step 1: Let 𝑢 (𝑥, 𝑡) = 𝐹 (𝑥)𝐺 (𝑡). Substitute in (6.4.1) to get
•
𝐺 𝐹 00
= .
𝑐 2𝐺 𝐹
The left side is independent of 𝑥 and the right side is independent of 𝑡. So, each is
equal to a constant. As in the case of wave equation, if this constant is 0 or positive,
we would get only the trivial solution 𝑢 (𝑥, 𝑡) = 0. So, suppose that each ratio in the
above equation is negative, that is, it is equal to −𝑝 2 for 𝑝 > 0. Then, we get two
ODEs
•
𝐹 00 + 𝑝 2 𝐹 = 0, 𝐺 + 𝑐 2𝑝 2𝐺 = 0.
Step 2: Solving the equation for 𝐹 we get

𝐹 (𝑥) = 𝑎 cos(𝑝𝑥) + 𝑏 sin(𝑝𝑥).

From the boundary condition (6.4.2), we have

𝑢 (0, 𝑡) = 𝐹 (0)𝐺 (𝑡) = 0, 𝑢 (𝐿, 𝑡) = 𝐹 (𝐿)𝐺 (𝑡) = 0.

We do not take 𝐺 (𝑡) = 0 since it leads to the trivial solution 𝑢 (𝑥, 𝑡) = 0. So,
𝐹 (0) = 0 and 𝐹 (𝐿) = 0. Now, 𝐹 (0) = 0 ⇒ 𝑎 = 0 ⇒ 𝐹 (𝑥) = 𝑏 sin(𝑝𝑥). Then,
𝐹 (𝐿) = 0 ⇒ 𝑏 sin(𝑝𝐿) = 0. Again, 𝑏 = 0 ⇒ 𝐹 (𝑥) = 0 which leads to the trivial
solution. So, sin(𝑝𝐿) = 0. Since 𝑝 > 0, it gives
𝑛𝜋
𝑝= for 𝑛 = 1, 2, 3, . . . .
𝐿
The corresponding solutions for 𝐹 (𝑥) are given by
 𝑛𝜋𝑥 
𝐹𝑛 (𝑥) = sin for 𝑛 = 1, 2, 3, . . . .
𝐿
For 𝑝 = 𝑛𝜋/𝐿, the equation for 𝐺 (𝑡) becomes
• 𝑐𝑛𝜋
𝐺 + 𝜆𝑛2𝐺 = 0 where 𝜆𝑛 = .
𝐿
120 MA2020 Classnotes

Corresponding to each 𝑛, the general solution is 𝑏𝑛 exp(−𝜆𝑛2𝑡). Since constants will

be accommodated later, we set 𝑏𝑛 = 1 to obtain

𝐺𝑛 (𝑡) = exp(−𝜆𝑛2𝑡)

as possible non-trivial solution for 𝐺 (𝑡) corresponding to the value 𝑛𝜋/𝐿 of 𝑝. Then,
 𝑛𝜋𝑥 
𝑢𝑛 (𝑥, 𝑡) = 𝐹𝑛 (𝑥)𝐺𝑛 (𝑡) = sin exp(−𝜆𝑛2𝑡) for 𝑛 = 1, 2, 3, . . .
𝐿
is a possible solution corresponding to the value 𝑛𝜋/𝐿 of 𝑝. This function 𝑢𝑛 (𝑥, 𝑡)
is called an eigenfunction with respect to the eigenvalue 𝜆𝑛 = 𝑐𝑛𝜋/𝐿, as earlier.
Step 3: None of the 𝑢𝑛 s may satisfy the initial condition. So, we propose to have
our solution as a series of eigenfunctions. So, let
∞ ∞  𝑛𝜋𝑥 
Õ Õ 𝑐𝑛𝜋
𝑢 (𝑥, 𝑡) = 𝑎𝑛𝑢𝑛 (𝑥, 𝑡) = 𝑎𝑛 sin exp(−𝜆𝑛2𝑡), where 𝜆𝑛 = .
𝑛=1 𝑛=1
𝐿 𝐿
(6.4.4)
The initial condition (6.4.3) now gives
∞
Õ  𝑛𝜋𝑥 
𝑢 (𝑥, 0) = 𝑎𝑛 sin = 𝑓 (𝑥).
𝑛=1
𝐿

So, 𝑎𝑛 s are the Fourier coefficients of the Fourier sine series for 𝑓 (𝑥). Thus,
∫𝐿
2  𝑛𝜋𝑥 
𝑎𝑛 = 𝑓 (𝑥) sin 𝑑𝑥 for 𝑛 = 1, 2, 3, . . . . (6.4.5)
𝐿 0 𝐿

It can be verified that 𝑢 (𝑥, 𝑡) of (6.4.4) in series form is a solution of the heat
equation (6.4.1) satisfying the conditions in (6.4.2)-(6.4.3) if 𝑓 (𝑥) is piecewise
continuous on 0 ≤ 𝑥 ≤ 𝐿, and has one-sided derivatives at all points of discontinuity.

(6.3) Example
Find the temperature 𝑢 (𝑥, 𝑡) in a laterally insulated copper bar 80 cm long if the
initial temperature is 100 sin(𝜋𝑥/80) ◦ C. Assume the following physical data for the
bar: density is 8.92 g/cm3 , specific heat is 0.992 cal/(g◦ C), thermal conductivity
is 0.95 cal/(cm sec◦ C). How long it will take for the maximum temperature in the
bar to drop to 50◦ C?
Here, 𝐿 = 80, 𝑓 (𝑥) = 100 sin(𝜋𝑥/80), 𝑐 2 = 𝐾/(𝜌𝜎 ) = 0.95/(0.092 × 8.92) =
1.158 cm2 /sec◦ C. Computing the coefficients from (6.4.5), we find that
∫ 80
2  𝑛𝜋𝑥 
𝑎1 = 100 sin2 𝑑𝑥 = 100, 𝑎𝑛 = 0 for 𝑛 ≥ 1.
80 0 80
Separation of Variables 121
Thus, we need only 𝜆12 which equals 1.158 × 9.870/802 = 0.001785[sec−1 ].
Hence, the solution is given by
 𝜋𝑥 
𝑢 (𝑥, 𝑡) = 100 sin 𝑒 −0.001785 𝑡 .
80
The maximum temperature in the bar is achieved when sin(𝜋𝑥/80) = 1. It drops to
50 implies 100𝑒 −0.001785 𝑡 = 50 ⇒ 𝑡 = log(0.5)/(−0.001785) = 388 [sec].

(6.4) Example
Find the temperature in a laterally insulated bar of length 𝐿 whose ends are kept at
temperature 0 assuming that the initial temperature is 𝑓 (𝑥) = 𝑥 for 0 ≤ 𝑥 ≤ 𝐿/2
and 𝑓 (𝑥) = 𝐿 − 𝑥 for 𝐿/2 < 𝑥 ≤ 𝐿.
We compute the coefficients from (6.4.5) as follows:

∫ 𝐿/2 ∫𝐿  
 0 if 𝑛 even
2

𝑛𝜋𝑥 𝑛𝜋𝑥 

𝑎𝑛 = 𝑠 sin 𝑑𝑥 + (𝐿 − 𝑥) sin 𝑑𝑥 = 𝑛4𝐿
2𝜋 2 if 𝑛 = 4𝑚 + 1
𝐿 0 𝐿 𝐿/2 𝐿 
 − 4𝐿

if 𝑛 = 4𝑚 + 3.
 𝑛2 𝜋 2
Hence, the solution is
4𝐿 h 𝜋𝑥 
2  1 3𝜋𝑥  2
 i
𝑢 (𝑥, 𝑡) = 2 sin exp − 𝑐𝜋/𝐿 𝑡 − sin exp − (3𝑐𝜋/𝐿) 𝑡 + · · · .
𝜋 𝐿 9 𝐿
Notice that this is a decreasing function of 𝑡. Physically this happens because the
ends are kept in zero temperature.

(6.5) Example
Find the solution 𝑢 (𝑥, 𝑡) of the heat equation 𝑢𝑡 = 𝑐 2𝑢𝑥𝑥 satisfying the conditions
𝑢𝑥 (0, 𝑡) = 𝑢𝑥 (𝐿, 𝑡) = 0 for all 𝑡, and 𝑢 (𝑥, 0) = 𝑓 (𝑥) for 0 ≤ 𝑥 ≤ 𝐿.
We set 𝑢 (𝑥, 𝑡) = 𝐹 (𝑥)𝐺 (𝑡). As earlier we reach at
•
𝐹 (𝑥) = 𝑎 cos(𝑝𝑥) + 𝐵 sin(𝑝𝑥), 𝐺 + 𝑐 2𝑝 2𝐺 = 0.

Then

𝐹 0 (𝑥) = −𝑎𝑝 sin(𝑝𝑥) + 𝑏𝑝 cos(𝑝𝑥) ⇒ 𝐹 0 (0) = 𝑏𝑝, 𝐹 0 (𝐿) = −𝑎𝑝 sin(𝑝𝐿).

The boundary conditions give

𝑢𝑥 (0, 𝑡) = 𝐹 0 (0)𝐺 (𝑡) = 𝑏𝑝 = 0, 𝑢𝑥 (𝐿, 𝑡) = 𝐹 0 (𝐿)𝐺 (𝑡) = −𝑎𝑝 sin(𝑝𝐿) = 0.

Since we need a non-zero solution, we assume that 𝐺 (𝑡) ≠ 0 and at least one of 𝑎
or 𝑏 is equal to 0. To obtain a series solution, we take 𝑏 = 0 and 𝑎 ≠ 0. Further,
122 MA2020 Classnotes

constants will get accommodated in a series. So, we take 𝑎 = 1. Then we have

𝑝 = 0 or sin(𝑝𝐿) = 0. It implies the possibilities for 𝑝 as
𝑛𝜋
𝑝 = 𝑝𝑛 = for 𝑛 = 0, 1, 2, 3, . . . .
𝐿
Neglecting the coefficients, we get 𝐹𝑛 (𝑥) = cos(𝑛𝜋𝑥/𝐿). This does not disturb 𝐺𝑛 s.
That is, as earlier, 𝐺𝑛 (𝑡) = exp(−𝜆𝑛2𝑡), where 𝜆𝑛 = 𝑐𝑛𝜋/𝐿. Hence, the eigenfunctions
are
𝑛𝜋𝑥
𝑢𝑛 (𝑥, 𝑡) = 𝐹𝑛 (𝑥)𝐺𝑛 (𝑡) = cos exp(𝜆𝑛2 𝑡) for 𝑛 = 0, 1, 2, 3, . . . .
𝐿
Notice that comparing these eigenfunctions with those in (6.4.4), we have an ex-
tra eigenvalue, namely 𝜆0 = 0 and corresponding to it the extra eigenfunction
𝑢 0 = constant. Notice that this is also a solution of the problem when 𝑓 (𝑥) is a
constant function.
As earlier, we have the solution as
∞ ∞
Õ Õ 𝑛𝜋𝑥 −𝜆𝑛2 𝑡 𝑐𝑛𝜋
𝑢 (𝑥, 𝑡) = 𝑎𝑛𝑢𝑛 (𝑥, 𝑡) = 𝑎𝑛 cos 𝑒 , where 𝜆𝑛 = .
𝑛=0 𝑛=0
𝐿 𝐿

The coefficients are obtained from the initial condition 𝑢 (𝑥, 0) = 𝑓 (𝑥). However,
Í∞
𝑢 (𝑥, 0) = 𝑛=0 𝑎𝑛 cos(𝑛𝜋𝑥/𝐿). Thus, 𝑎𝑛 s are the Fourier coefficients of the Fourier
cosine series of 𝑓 (𝑥). That is,
∫
1 𝐿 2 𝑛𝜋𝑥
𝑎0 = 𝑓 (𝑥) 𝑑𝑥, 𝑎𝑛 = 𝑓 (𝑥) cos 𝑑𝑥 𝑛 = 1, 2, 3, . . ..
𝐿 0 𝐿 𝐿
When the two ends of the wire are kept in constant temperatures, we get the
boundary conditions as 𝑢 (0, 𝑡) = 𝐴 and 𝑢 (𝐿, 𝑡) = 𝐵. We try a solution in the form
𝐵 −𝐴
𝑢 (𝑥, 𝑡) = 𝐴 + 𝑥 + 𝑣 (𝑥, 𝑡). Then, 𝑣 (𝑥, 𝑡) will satisfy the heat equation with
𝐿
homogeneous boundary conditions. We use the method of separation of variables
for determining 𝑣 (𝑥, 𝑡). You may need this trick to solve some problems from the
exercises.

6.5 Laplace equation

Instead of a metal rod, consider heat distribution on a metal plate. We may approach
the problem of modeling in a way similar to the derivation of one-dimensional wave
and heat equations. We would arrive at the two-dimensional heat equation

𝑢𝑡 = 𝑐 2 (𝑢𝑥𝑥 + 𝑢𝑦𝑦 ).
Separation of Variables 123
When the steady state is achieved, we find that 𝑢𝑡 = 0 and it yields the Laplace
equation
𝑢𝑥𝑥 + 𝑢𝑦𝑦 = 0.
When the metal plate is rectangular, the Cartesian coordinates system is suitable.
Similarly, if the plate is circular, it may be easier to use the polar coordinates. We
need to express the Laplacian 𝑢𝑥𝑥 + 𝑢𝑦𝑦 in polar coordinates.
The relation between Cartesian and the polar coordinates is expressed by
q
𝑦
𝑥 = 𝑟 cos 𝜃, 𝑦 = 𝑟 sin 𝜃, 𝑟 = 𝑥 2 + 𝑦 2, tan 𝜃 = .
𝑥
Suppose 𝑢 = 𝑢 (𝑥, 𝑦, 𝑡) is a function of 𝑥, 𝑦 and 𝑡. We are interested in computing
𝑢𝑥𝑥 + 𝑢𝑦𝑦 in 𝑟, 𝜃 form. By the chain rule,

𝑢 𝑥 = 𝑢𝑟 𝑟 𝑥 + 𝑢𝜃 𝜃 𝑥 .

Differentiating again, we obtain

𝑢𝑥𝑥 = (𝑢𝑟 𝑟 𝑥 )𝑥 + (𝑢𝜃 𝜃 𝑥 )𝑥

= (𝑢𝑟 )𝑥 𝑟 𝑥 + 𝑢𝑟 𝑟 𝑥𝑥 + (𝑢𝜃 )𝑥 𝜃 𝑥 + 𝑢𝜃 𝜃 𝑥𝑥
= (𝑢𝑟𝑟 𝑟 𝑥 + 𝑢𝑟𝜃 𝜃 𝑥 ) 𝑟 𝑥 + 𝑢𝑟 𝑟 𝑥𝑥 + (𝑢𝜃𝑟 𝑟 𝑥 + 𝑢𝜃𝜃 𝜃 𝑥 ) 𝜃 𝑥 + 𝑢𝜃 𝜃 𝑥𝑥 .

Using the expressions for 𝑟 and 𝜃 in terms of 𝑥, 𝑦, we obtain

𝑥 𝑥 1 −𝑦 𝑦
𝑟𝑥 = p = , 𝜃𝑥 = 2
× 2 = − 2.
𝑥 2 + 𝑦2 𝑟 1 + (𝑦/𝑥) 𝑥 𝑟

𝑟 − 𝑥 𝑟𝑥 1 𝑥 2 𝑦 2 −2 2𝑥𝑦
𝑟 𝑥𝑥 = = − = , 𝜃 𝑥𝑥 = −𝑦 × 𝑟 𝑥 = .
𝑟2 𝑟 𝑟3 𝑟3 𝑟3 𝑟4
Assuming that 𝑢 is two times continuously differentiable with respect to 𝑟 and 𝜃 ,
we get 𝑢𝑟𝜃 = 𝑢𝜃𝑟 . Substituting the expressions above into that of 𝑢𝑥𝑥 leads to

𝑥2 2𝑥𝑦 𝑦2 𝑦2 2𝑥𝑦
𝑢𝑥𝑥 = 2
𝑢 𝑟𝑟 − 3
𝑢 𝑟𝜃 + 4
𝑢 𝜃𝜃 + 3
𝑢𝑟 + 4 𝑢𝜃 .
𝑟 𝑟 𝑟 𝑟 𝑟
Similarly,
𝑦2 2𝑥𝑦 𝑥2 𝑥2 2𝑥𝑦
𝑢𝑦𝑦 = 2 𝑢𝑟𝑟 + 3 𝑢𝑟𝜃 + 4 𝑢𝜃𝜃 + 3 𝑢𝑟 − 4 𝑢𝜃 .
𝑟 𝑟 𝑟 𝑟 𝑟
Adding the two above and suing the fact that 𝑥 2 + 𝑦 2 = 𝑟 2 , we obtain the expression
for the Laplacian in polar coordinates as follows:
𝑢𝑟 𝑢𝜃𝜃
𝑢𝑥𝑥 + 𝑢𝑦𝑦 = 𝑢𝑟𝑟 + + 2.
𝑟 𝑟
124 MA2020 Classnotes

Using these and the method of separation of variables, we solve some problems
on heat distribution on rectangular and circular plates.

(6.6) Example
Find the steady state temperature distribution 𝑢 (𝑥, 𝑦) on the rectangular region
0 ≤ 𝑥 ≤ 𝜋, 0 ≤ 𝑦 ≤ 2, given that on the side 𝑦 = 0, 0 ≤ 𝑥 ≤ 𝜋, 𝑢 (𝑥, 0) = 𝑥 sin 𝑥,
and the temperature on the other three sides are maintained at 𝑢 = 0.
The steady state temperature 𝑢 (𝑥, 𝑦) satisfies the Laplacian

𝑢𝑥𝑥 + 𝑢𝑦𝑦 = 0.

We try 𝑢 (𝑥, 𝑦) = 𝐹 (𝑥)𝐺 (𝑦). Substituting it in the equation and simplifying, we get

1 𝑑 2𝐹 1 𝑑 2𝐺
= − .
𝐹 𝑑𝑥 2 𝐺 𝑑𝑦 2

The left side is independent of 𝑦 and the right side is independent of 𝑥; so each is a
constant, say, 𝑐. It then follows that

𝑑 2𝐹 𝑑 2𝐺
= 𝑐𝐹, + 𝑐𝐺 = 0.
𝑑𝑥 2 𝑑𝑦 2

The boundary conditions 𝑢 (0, 𝑦) = 0 = 𝑢 (𝜋, 𝑦) imply that 𝐹 (0) = 𝐹 (𝜋) = 0. When
𝑐 = 0, 𝐹 = 𝑎 + 𝑏𝑥. These condition on 𝐹 imply that 𝐹 (𝑥) = 0, leading√to the trivial
√
solution 𝑢 (𝑥, 𝑦) = 0 which is not the case. If 𝑐 > 0, then 𝐹 (𝑥) = 𝑎𝑒 𝑐𝑥 + 𝑏𝑒 − 𝑐𝑥 .
Again, the conditions 𝐹 (0) = 𝐹 (𝜋) = 0 lead to 𝐹 (𝑥) = 0. So, 𝑐 < 0; then let 𝑐 = −𝜆 2
for 𝜆 > 0. We now have the equations as

𝑑 2𝐹 𝑑 2𝐺
2
+ 𝜆 2 𝐹 = 0, 2
− 𝜆 2𝐺 = 0.
𝑑𝑥 𝑑𝑦

Then 𝐹 (𝑥) = 𝑎 cos(𝜆𝑥) + 𝑏 sin(𝜆𝑥). 𝐹 (0) = 0 ⇒ 𝑎 = 0. So, 𝐹 (𝑥) = 𝑏 sin(𝜆𝑥).

𝑋 (𝜋) = 0 ⇒ sin(𝜆𝜋) = 0 as 𝑏 = 0 leads to the trivial solution. Hence, the
eigenvalues are
𝜆𝑛 = 𝑛 for 𝑛 = 1, 2, 3, . . . .
The corresponding eigenfunctions are (we take 𝑏𝑛 = 1.)

𝐹𝑛 (𝑥) = sin(𝑛𝑥) for 𝑛 = 1, 2, 3, . . . .

Now the equation for 𝐺 reads as

𝑑 2𝐺
2
− 𝑛 2𝐺 = 0.
𝑑𝑦
Separation of Variables 125
To show dependence of 𝐺 on the parameter 𝑛, we write its solution as 𝐺𝑛 (𝑦).
Then, 𝐺𝑛 (𝑦) = 𝑐 cosh(𝑛𝑦) + 𝑑 sinh(𝑛𝑦). The boundary condition 𝑢 (𝑥, 2) = 0 gives
𝐹 (𝑥)𝐺 (2) = 0 ⇒ 𝐺 (2) = 0. Or,

𝑐 cosh(2𝑛) + 𝑑 sinh(2𝑛) = 0 ⇒ 𝑑 = −𝑐 coth(2𝑛).

Using this in the expression for 𝐺𝑛 (𝑦) and setting 𝑐 = 1, we obtain

𝐺𝑛 (𝑦) = cosh(𝑛𝑦) − coth(2𝑛) sinh(𝑛𝑦) = cosech(2𝑛) sinh(𝑛𝑦 − 2𝑛).

Since constants will be determined later from a series, we choose 𝑐𝑛 = sinh(2𝑛) in

obtaining 𝑢𝑛 . Then,

𝑢𝑛 (𝑥, 𝑦) = sin(𝑛𝑥) sinh(𝑛𝑦 − 2𝑛) for 𝑛 = 1, 2, 3, . . . .

To satisfy the other conditions, we set

∞
Õ ∞
Õ
𝑢 (𝑥, 𝑦) = 𝑎𝑛 𝑢𝑛 = 𝑎𝑛 sin(𝑛𝑥) sinh(𝑛𝑦 − 2𝑛).
𝑛=1 𝑛=1

Now, 𝑢 (𝑥, 0) = 𝑥 sin 𝑥 implies

∞
Õ
𝑥 sin 𝑥 = 𝑎𝑛 sin(𝑛𝑥) sinh(−2𝑛).
𝑛=1

Multiply sin(𝑚𝑥) and integrate over 0 ≤ 𝑥 ≤ 𝜋 to obtain

∫𝜋 ∫𝜋
𝑥 sin 𝑥 sin(𝑚𝑥) 𝑑𝑥 = −𝑎𝑚 sinh(2𝑛) sin(𝑛𝑥) sin(𝑚𝑥) 𝑑𝑥 .
0 0

Using the orthogonality property of {sin(𝑛𝑥)} as in evaluating the Fourier coeffi-

cients, we get


𝜋 4𝑛 1 + (−1)𝑛
𝑎1 = − , 𝑎𝑛 = 2 for 𝑛 = 2, 3, 4, . . . .
2 sinh 2 (𝑛 − 1) 2 𝜋 sinh(2𝑛)

Substituting these values of 𝑎𝑛 in the series, we obtain

∞

𝜋 sin 𝑥 sinh(𝑦 − 2) Õ 4𝑛 1 + (−1)𝑛
𝑢 (𝑥, 𝑦) = − + sin(𝑛𝑥) sinh(𝑛𝑦 − 2𝑛).
2 sinh 2 𝑛=2
(𝑛 2 − 1) 2 𝜋 sinh(2𝑛)
126 MA2020 Classnotes

(6.7) Example
Find the temperature distribution 𝑢 (𝑟, 𝜃, 𝑡) in a thin (negligible thickness) semicir-
cular metal plate 0 ≤ 𝑟 ≤ 1, 0 ≤ 𝜃 ≤ 𝜋 given that its plane faces are insulated to
prevent heat loss through them, the straight edge of the plate formed by the diameter
0 ≤ 𝑟 ≤ 1, 𝜃 = 0 and 𝜃 = 𝜋 is insulated, the semicircular boundary is maintained at
zero temperature, and the initial temperature distribution is 𝑢 (𝑟, 𝜃, 0) = (1−𝑟 ) cos 𝜃 .
Using the Laplacian in polar coordinates, the heat equation on the plate is
𝑢𝑟 𝑢𝜃𝜃

𝑢𝑡 = 𝑐 2 𝑢𝑟𝑟 + + 2 .
𝑟 𝑟
The bounding diameter is insulated and semicircular boundary is kept at zero
temperature. This means that

𝑢 (𝑟, 0, 𝑡) = 0, 𝑢𝜃 (𝑟, 𝜋, 𝑡) = 0, 𝑢 (1, 𝜃, 𝑡) = 0.

To use the separation of variables, we take 𝑢 (𝑟, 𝜃, 𝑡) = 𝐸 (𝑟 )𝐹 (𝜃 )𝐺 (𝑡). Substituting

in the heat equation above we get
•
𝐺  00 1 𝐸 0 1 𝐹 (2) 
2 𝐸
=𝑐 + + .
𝐺 𝐸 𝑟 𝐸 𝑟2 𝐹
Here 𝐸 0 = 𝑑𝐸/𝑑𝑟, 𝐸 00 = 𝑑 2 𝐸/𝑑𝑟 2 and 𝐹 (𝑛) means 𝑑 𝑛 𝐹 /𝑑𝜃 𝑛 . The left side is inde-
pendent of 𝑟 and 𝜃 , and the right side is independent of 𝑡. Hence, all of them are
independent of 𝑟, 𝜃, 𝑡 so that they are equal to a constant. Further, the temperature
decreases with time; so the constant must be negative. We may also consider three
cases of this constant, and verify that non-negative values of this constant lead to
the trivial solution.
Now that each side is equal to some negative constant, say, −𝜆 2 with 𝜆 > 0, we
obtain two equations:

• 𝐸 00 𝐸0 𝐹 (2)
𝐺 + 𝑐 2𝜆 2𝐺 = 0, 𝑟2 + 𝑟 + 𝜆 2𝑟 2 = − .
𝐸 𝐸 𝐹
Again, the second equation has a left side independent of 𝜃 and the right side
independent of 𝑟 . Hence, each is a constant. We may verify that for negative values
of this constant, we get only the trivial solution. So, we assume that this constant is
non-negative. We write it as 𝑞 ≥ 0. Then, the second equation gives two equations:

𝐹 (2) + 𝑞𝐹 = 0, 𝑟 2 𝐸 00 + 𝑟𝐸 0 + (𝜆 2𝑟 2 − 𝑞)𝐸 = 0.

The general solution for 𝐹 (𝜃 ) is

√ √
𝐹 (𝜃 ) = 𝐴 cos( 𝑞 𝜃 ) + 𝐵 sin( 𝑞 𝜃 ).
Separation of Variables 127
The boundary condition 𝑢𝜃 (𝑟, 0, 𝑡) = 0 implies 𝐹 (1) (0) = 0 and the condition
𝑢𝜃 (𝑟, 𝜋, 𝑡) = 0 implies 𝐹 (1) (𝜋) = 0. The first condition yields 𝐵 = 0 and the second
√ √
leads to sin( 𝑞 𝜋) = 0. Hence, 𝑞 = 𝑚 for 𝑚 = 0, 1, 2, 3, . . .. Setting the arbitrary
constant 𝐴 to 1, we get

𝐹 (𝜃 ) = cos(𝑛𝜃 ) for 𝑚 = 0, 1, 2, 3, . . .

The equation for 𝐸 (𝑟 ) now becomes

𝑟 2 𝐸 00 + 𝑟𝐸 0 + (𝜆 2𝑟 − 𝑚 2 )𝐸 = 0.

We recognize this as the Bessel’s equation, whose general solution is

𝐸𝑚 (𝑟 ) = 𝑎𝐽𝑚 (𝜆𝑟 ) + 𝑏𝑌𝑚 (𝜆𝑟 ).

Recall that 𝑌𝑚 (𝜆𝑟 ) → ∞ as 𝑟 → 0. However, the temperature on the plate remains

finite. Hence, 𝑏 = 0. Further, we will be getting a series solution finally; so, we set
𝑎 = 1 and continue with the solutions

𝐸𝑚 (𝑟 ) = 𝐽𝑚 (𝜆𝑟 ) for 𝑚 = 0, 1, 2, 3, . . .

For the boundary condition 𝑢 (1, 𝜃, 𝑡) = 0, we must have 𝐸 (1) = 0. It means,

𝐽𝑚 (𝜆) = 0. Hence, the eigenvalues 𝜆s are the positive zeros of 𝐽𝑚 , the Bessel
function. So, we take 𝜆𝑛 = 𝑧𝑚,𝑛 , the 𝑛th positive zero of 𝐽𝑚 . •
Using these 𝜆s in the equation for 𝐺 (𝑡), which was 𝐺 + 𝑐 2𝜆 2𝐺 = 0, we have
2
𝐺𝑚,𝑛 (𝑡) = 𝑏𝑚,𝑛 exp(−𝑧𝑚,𝑛 𝑐 2𝑡).

Combining the results for 𝐸 (𝑟 ), 𝐹 (𝜃 ) and 𝐺 (𝑡), we obtain

∞ Õ
Õ ∞
2
𝑢 (𝑟, 𝜃, 𝑡) = 𝑏𝑚,𝑛 𝐽𝑚 (𝑧𝑚,𝑛𝑟 ) cos(𝑚𝜃 ) exp(−𝑧𝑚,𝑛 𝑐 2𝑡).
𝑚=0 𝑛=1

When 𝑡 = 0, the initial condition 𝑢 (𝑟, 𝜃, 0) = (1 − 𝑟 ) cos 𝜃 gives

∞ Õ
Õ ∞
(1 − 𝑟 ) cos 𝜃 = 𝑏𝑚,𝑛 𝐽𝑚 (𝑧𝑚,𝑛𝑟 ) cos(𝑚𝜃 ).
𝑚=0 𝑛=1

This is the Fourier-Bessel series of the left side function. We multiply cos 𝜃 and
integrate over 0 ≤ 𝜃 ≤ 𝜋. Every term on the right hand side vanishes except those
corresponding to 𝑚 = 1. Thus,
∞
Õ ∞
Õ
(1 − 𝑟 ) cos 𝜃 = 𝑏 1,𝑛 𝐽1 (𝑧 1,𝑛𝑟 ) cos 𝜃 ⇒ 1 − 𝑟 = 𝑏 1,𝑛 𝐽1 (𝑧 1,𝑛𝑟 ).
𝑛=1 𝑛=1
128 MA2020 Classnotes

Multiply the last expression by 𝑟 𝐽1 (𝑧 1,𝑠 𝑟 ) and integrate over 0 ≤ 𝑟 ≤ 1. Using the
orthogonality of the Bessel functions, we get
∫1
1 2
𝑟 (1 − 𝑟 ) 𝐽1 (𝑧 1,𝑠 𝑟 ) 𝑑𝑟 = 𝑏 1,𝑠 𝐽 (𝑧 1,𝑠 ).
0 2 2
This gives 𝑏 1,𝑠 . We write in terms of 𝑛:
∫1
−2
(𝑟 − 𝑟 2 )𝐽1 (𝑧 1,𝑛𝑟 ) 𝑑𝑟

𝑏 1,𝑛 = 2 𝐽2 (𝑧 1,𝑛 )] for 𝑛 = 1, 2, 3, . . . .
0

Then the required solution is

∞
Õ
2 2
𝑢 (𝑟, 𝜃, 𝑡) = 𝑏 1,𝑛 𝐽1 (𝑧 1,𝑛𝑟 ) cos 𝜃 exp(−𝑧 1,𝑛 𝑐 𝑡).
𝑛=1

To obtain numerical values for specific tuples (𝑟, 𝜃, 𝑡) one must use the tables for
the zeros of Bessel functions.
Index

Abel’s formula, 25 Legendre polynomials, 65

Analytic function, 48 Legendre series, 70
Auxiliary equation, 33 Linearly independent, 26
Linear forst order ODE, 18
Bernoulli equation, 20 Linear PDE, 90
Bessel’s equation, 71
Bessel function, 73, 75 Nodes, 116
Boundary conditions, 79 Non-linear PDE, 91
Norm, 84
Canonical form, 98 Normal mode, 116
Characteristic curves, 93
Characteristic equation, 27, 93 ODE, 1
One-dimensional wave equation, 111
D’Ȧlembert’s solution, 112 Order of ODE, 1
Differential equation, 1 Order of PDE, 90
Discriminant, 98 Ordinary point, 49
Eigenfunctions, 80, 116, 120 Orthogonal, 68, 84
Eigenvalues, 80, 116, 120 Orthonormal, 84
Elliptic, 99 Overtones, 116
Exact equation, 12 Parabolic, 99
Fourier-Bessel series, 88 Partial differential equation, 90
Fundamental mode, 116 PDE, 90
Fundamental solutions, 26 Periodic Sturm-Liouville, 80

Generating function, 68, 78 Quasi-linear PDE, 91

Homogeneous ODE, 18, 22 Reduction to standard form, 99

Homogeneous PDE, 90, 97 Regular point, 49
Hyperbolic, 99 Regular singular point, 49
Regular Sturm-Liouville, 80
Initial condition, 2 Rodrigue’s formula, 66
Initial value problem, 2
Inner product, 84 Second order linear ODE, 22
Integrating factor, 16 Semilinear PDE, 91
Irregular singular point, 49 Singular point, 49
IVP, 2, 22 Singular Sturm-Liouville, 80
Solution of ODE, 1
Laplace equation, 123 Spectrum, 116
Laplacian, 123 Standard form, 49, 98

129
130 Index

Sturm-Liouville equation, 78 Trivial solution, 80

Sturm-Liouville problem, 79 Wronskian, 24